Datasets:

longevity-genie

/

tacutu_papers

like 0

Follow

Longevity Genie 1

2022-03-22T16:26:07Z

CCBY

https://academic.oup.com/nar/article-pdf/43/D1/D873/7330380/gku843.pdf

GOLD

1ef366e034d08317e9305bb390aed643459baba4

journals/nar/CraigSTWWSJSMAM15

10.1093/nar/gku843

2169010206

25232097

4384002

The Digital Ageing Atlas: integrating the diversity of age-related changes into a unified resource 2015 Thomas Craig Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool LiverpoolUK Chris Smelick University of North Carolina at Chapel Hill NCUSA Robi Tacutu Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool LiverpoolUK Daniel Wuttke Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool LiverpoolUK Shona H Wood Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool LiverpoolUK Henry Stanley Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool LiverpoolUK Georges Janssens Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool LiverpoolUK Ekaterina Savitskaya Skolkovo Institute of Science and Technology Moscow regionRussia Alexey Moskalev Institute of Biology of Komi Science Center of RAS SyktyvkarRussia Moscow Institute of Physics and Technology DolgoprudnyRussia Robert Arking Department of Biological Sciences Wayne State University DetroitMIUSA João Pedro De Magalhães Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool LiverpoolUK The Digital Ageing Atlas: integrating the diversity of age-related changes into a unified resource Nucleic Acids Research 43201510.1093/nar/gku843Received June 24, 2014; Revised August 25, 2014; Accepted September 03, 2014 Multiple studies characterizing the human ageing phenotype have been conducted for decades. However, there is no centralized resource in which data on multiple age-related changes are collated. Currently, researchers must consult several sources, including primary publications, in order to obtain age-related data at various levels. To address this and facilitate integrative, system-level studies of ageing we developed the Digital Ageing Atlas (DAA). The DAA is a one-stop collection of human age-related data covering different biological levels (molecular, cellular, physiological, psychological and pathological) that is freely available online (http://ageing-map.org/). Each of the >3000 age-related changes is associated with a specific tissue and has its own page displaying a variety of information, including at least one reference. Age-related changes can also be linked to each other in hierarchical trees to represent different types of relationships. In addition, we developed an intuitive and user-friendly interface that allows searching, browsing and retrieving information in an integrated and interactive fashion. Overall, the DAA offers a new approach to systemizing ageing resources, providing a manually-curated and readily accessible source of age-related changes.DIGITAL AGEING ATLAS CONTENT, INTERFACE AND STRUCTUREConceptually, an age-related change represents an observed difference of a molecule, parameter or process between young and old, and various and diverse types of properties can be represented in a quantitative and/or qualitative way. INTRODUCTION Ageing can be defined as a progressive functional decline, or a gradual deterioration of physiological function with age, often including a decrease in fecundity (1). Human ageing is characterized by multiple changes at different levels of bi-ological organization (2,3). It is still not clear which (if any) molecular, cellular or physiological changes are more important drivers of the process of ageing or how they influence each other. One difficulty in understanding how different processes at different scales relate to ageing as a whole is the lack of integrative, holistic views of ageing. This hinders studies of how different molecular, cellular and physiological components interact with each other, in spite of the recognized importance of such approaches (4,5). Particularly now in the post-genome era, efforts to obtain a more comprehensive and detailed characterization of molecular changes with ageing, such as those using -omics approaches (6)(7)(8), have been widespread. Use of this quantitative data, including its meta-and re-analysis, allows the application of systems biology approaches to ageing research. Consequently, there is now a drive to link these molecular level changes to cellular and physiological processes. The ultimate aim is to elucidate how molecular changes with age, for example, may influence or are influenced by changes in the wider organism, e.g. hormonal changes, and ultimately how these interactions contribute to pathology. Nonetheless, collating and converting raw data into information that is usable and can be cross compared is time consuming and difficult. In this context, we developed the Digital Ageing Atlas (DAA; http://ageing-map. org/), the first portal encompassing age-related changes at different biological levels, including a large amount ofomics data, already processed, categorized and filtered for statistical significance. To catalogue and organize age-related changes, in the DAA they fall into four broad categories: molecular, physiological, psychological and pathological changes ( Table 1). The DAA contains: more than 3000 molecular ageing changes, which include gene expression, epigenetic and proteomic changes; over 300 physiological changes, which include cellular, hormonal and changes at various scales (including organs and the whole organism); and psychological or cognitive changes. Also included are pathological changes, listing epidemiological data on the incidence and/or mortality of major age-related diseases. Our focus is on changes occurring during normal ageing across populations, though e.g. gender-specific changes are indicated. As detailed below, data was manually-curated from the literature, such as textbooks and papers, and retrieved from public databases like GEO (9). All changes are fully referenced making it possible to access the raw data. In total, the DAA currently details 3526 biological changes in humans and 713 changes in mice. The DAA focuses on human data, however mouse data has been included, in particular gene expression data, and cross-linked to relevant human entries (e.g., homologous genes), to enhance and expand the information on human ageing. We anticipate the addition of data from other model organisms in the future. Presenting information in an easy-to-understand visual form is a powerful means of fostering the analysis and interpretation of large datasets and of allowing researchers to identify gaps in knowledge and develop new research directions (10,11). Without it the comprehension of large-scale or diverse datasets is impeded. Therefore, not only does the DAA merge different types of data into a single repository, but we developed an intuitive and user-friendly web resource that allows accessing, searching, browsing and retrieving the datasets in an integrated and interactive fashion. Specifically, we developed an anatomical diagram to allow users to browse and select their organ of interest (Figure 1). The use of keyword term searching (e.g. 'heart' will show both tissues and changes associated with the heart while 'p53' will show changes related to any gene with p53 in its name or alias) and more general anatomical selection offers a great deal of flexibility to users, ensuring that users of any level of technical skill can access the resources, including non-researchers, opening up the field of ageing to a wider audience. Each age-related change in the DAA has its own page displaying a variety of information. Typically, entries include a description of the change with age, a quantification (if available) of the change with age (e.g. a percentage gene expression change between two ages), at least one reference and relevant links ( Figure 2). The way in which the changes are stored in the database is best described in an objectorientated way. The key objects in the DAA are change, tis- sue, gene, property and data. The change object stores the basic information on a change including type, age of occurrence, gender (if available) and organism. The gene object contains basic information on a gene, e.g. symbol and name, mapping to other information such as homologues in other organisms, Gene Ontology (GO) terms and links to external resources, for instance cross-linking to the GenAge database of ageing-related genes (12) (Figure 3). Gene information can then be associated with multiple changes to prevent repetition and ensure ease of updating when elements such as the gene symbol change. It also allows for the DAA to display all changes associated with a gene making it easier to find information. The tissue object contains details on a tissue such as a name and description. The tissue objects (currently 284 different tissues are represented) are arranged into a simple hierarchical structure, based upon the ontology created by eVOContology (13), supplemented by descriptive data from both Brenda (14) and Wikipedia Nucleic Acids Research, 2015, Vol. 43, Database issue D875 Figure 2. A labelled diagram of the entry for IGF1 age-related changes in the plasma: (1) Each change is colour coded for easy identification of type. (2) As all changes are assigned to a tissue it is easy to see the different changes occurring on an organ level. (3) Each change is fully referenced allowing for additional details into the methodology and access to the original data. (4) Clear identification of the amount and direction of change with age (if applicable) is provided, along with how it was derived. (5) Changes can be stored persistently between sessions as well as compared on-site using the graphing functionality. (6) Descriptions provide more details, including greater clarification regarding the context in which the change was observed and/or measured. (7) Linking changes to genes allows, much like linking tissues, the ability to see all the changes associated with a particular gene. (http://en.wikipedia.org) and further expanded in our lab. Each tissue has a parent and zero or more children. The root parent represents the whole organism and the tissue hierarchy can be navigated on our interface. Each change is associated with one or more tissues, allowing for exploration of the number and types of changes occurring in each tissue or organ. The property object allows for non-duplicate properties to be defined and associated with changes (e.g. the location property may take values like synapse, mitochondria, cellular, etc). These properties are defined by the curators and can encompass any value which may be relevant to the change but is not recorded elsewhere. This allows for great flexibility in recording type-specific information (e.g. subcellular location) and can be filtered against in the search interface. The data object allows the association of specific types of data with a change. It is divided into sub-objects that cover a class of data, such as percentage, equation or dataset. These store specific sets of data with some fields such as 'change measured' common between all. Percentage objects store a simple percentage value; equation objects store the components of an equation describing the change in a quantitative fashion; dataset objects store an arbitrary number of data points to plot more fragmentary data such as mortality rates. With this the database can cover most types of data that can be associated with a change during the life course, and more can be easily added, if required, with very little effort. These data are presented on the details page and used for the display of increase/decrease icons on the search page, among others. The relationship object stands alone as it is not directly related to a change. Instead, similar to the tissue object, it stores a hierarchical representation of information. Each relationship object is linked to a single change and optionally another relationship object. These are then chained together to create a tree. Multiple trees can be associated with a change and can describe different types of associations from causal relationships to similar processes that are occurring together. Linking the relationship objects to changes allows the construction of complex hierarchies often encompassing different biological levels while still permitting a given change to appear in multiple hierarchies. A change can appear in multiple trees as the change may be a part of multiple processes, some of which may not be closely related. A good example of this is DAA982 (which refers to changes in CD16 expression in the elderly) in which there are two trees describing how the gene (a molecular change) reacts during two distinct physiological ageing changes. Relationships also link pathologies to physiological and molecular changes associated with them, like for Alzheimer's disease (DAA615) which is associated with neuritic plaques (DAA723) and beta-amyloid deposits (DAA1996). Interpretation and visualization of the data is facilitated by tools built into the DAA. Any numerical change within the DAA can be compared against others by adding them to a list which can then be analysed in a graph form within the D876 Nucleic Acids Research, 2015, Vol. 43, Database issue website. For example, a comparison can be made of molecular changes presented in graph form allowing the comparison of the gene expression levels recorded by those changes (Figure 4). More complex analyses can be performed using external tools as the DAA permits downloading both through its export tool and through the availability of the complete DAA dataset for download. The export tool itself takes advantage of the search and filtering supported and allows for specific subsets of data to be extracted and saved as a tab-delimited text file. The DAA and all its data is made available under a permissive licence. DATA SELECTION AND CURATION Data in the DAA is manually curated and each age-related change has been selected based upon clearly defined criteria. First, only age-related changes for which there is direct, empirical evidence supported by one or more references are included. Second, only ageing changes occurring in vivo are incorporated into the DAA. Since our goal is to define typical age-related changes, we focused on those observed during healthy ageing, with the obvious exception of pathological age-related changes that describe mortality and incidence rates of specific diseases of the aged. Although the goal is to make the DAA as complete as possible, the focus is on what are likely the most important age-related changes, which in many cases are the changes that are also involved in determining age-related pathologies. Negative results can be included if these are deemed relevant to understand ageing, e.g. DAA711 refers to measurements of heart physiology that are unchanged with age. Our general policy regarding conflicting reports is to cite all conflicting reports and let users make their own decisions on how to interpret them. Molecular changes (e.g. gene expression, protein levels and methylation) from high-throughput approaches are usually selected based on criteria for statistical significance that the authors have used in the sourced data, though data and methods (e.g. correction for multiple hypothesis testing) are examined as part of our QC procedures which are described in de Magalhães et al. (6). A P-value cut-off of 0.001 or lower is normally used for genome-wide approaches. This value was reached based on standard practice and observation of effects of the data and ensures that the changes added are above the noise threshold. Ours is an inclusive policy, however, and effect sizes and P-values are included in the DAA to allow users to make their own decisions about which data is relevant. The primary sources of data for the molecular section have been the meta-analysis by de Magalhães et al. (6), public databases like GEO (9) and primary publications. Therefore, quantitative data can be taken directly from publications or recomputed as in (6) with the specific type of equation used indicated in the details page for a given entry. At the time of writing the DAA includes 24 datasets from high throughput screens (mainly microarrays) that cover 22 different tissues. Physiological changes were sourced from books (2,3,15), reviews and primary publications. Major changes in cell populations are likely to contribute to age-related physiological and pathological changes, therefore studies of cellular alterations with age are included, but results from in vitro cellular models of ageing are not included in the DAA. Pathological and some physiological changes were sourced from the Centers for Disease Control and Prevention (CDC) (http://www.cdc.gov/nchs/hdi.htm), selected based on their relevance to chronic ageing conditions and the ages which they cover. Psychological changes were sourced from the same locations as the physiological changes with the condition that they must indicate a change in behaviour or cognition as the organism ages. Ageing changes vary between individuals and populations. It is not the goal of the DAA to capture the individual diversity of age-related changes and thus the relative dependence on large datasets. The objective of the DAA is to provide an overview of major age-related changes and so typical values are featured, though outliers are in-dicated in notes-specific to each data type. For example, gender-specific changes are featured and properly annotated. An attempt is made to obtain data from consistent sources. In mice, age-related changes and even lifespan can vary between strains, therefore the C57BL/6 strain is used as the 'gold standard' in the case of conflicting findings, however discrepancies are highlighted when they occur. The C57BL/6 strain was selected because, currently, it is the most commonly used mouse strain for ageing studies. This strategy is consistent with other similar projects like AGEMAP (16) and the Allen Brain Atlas (17) that also focus on the C57BL/6 strain. If age-related changes are suspected of being population-specific, then this is indicated in the DAA through a specific property. The site itself uses the Python-based Django framework and is served by an Nginx web server. It uses PostgreSQL 9.1 as a database backend, implementing a number of constraints to ensure entries are not duplicated or left as orphans when a change instance is deleted. A web-based curation application was also created that allows for easy addition and updating of data without requiring knowledge of the technical operations of the portal. We encourage contributions by the wider research community. By having an intuitive and easily usable curation interface this provides the ability to both quickly correct and add relevant information as well as allowing specialists to directly contribute D878 Nucleic Acids Research, 2015, Vol. 43, Database issue to its improvement, thus ensuring that it stays at the forefront of ageing research. AVAILABILITY The DAA is available at http://ageing-map.org with the data made available under the permissive Creative Commons licence, allowing data to be used in other analyses. There are options to either download the entire database or to download more focused data using the export tool. Feedback via email is welcome, as are submissions of new data, for which a submission form is provided to ensure that the relevant information is included. CONCLUSION The DAA is an integrated web resource for studying and visualizing human age-associated changes at various biological levels. It can aid researchers to perform integrative, system-level analysis of ageing. While target users are primarily fundamental researchers, it is anticipated that the DAA will also be useful to clinicians, students and the public in general. Other existing ageing-related resources such as GenAge (12), AgeFactDB (18) and SAGEWEB (http: //sageweb.org/) focus on genes and factors that alter lifespan and/or ageing. By providing a manually-curated and readily accessible source of age-related changes during the normal life course, the DAA is thus complementary to existing resources and offers a new approach to systemizing ageing resources. This brings numerous benefits, limiting duplication of efforts and maintaining the accuracy of data which is essential given the rapid pace at which the field of ageing is progressing. In conclusion, the DAA aims to become the most comprehensive source for data related to ageing changes, consistently providing high-quality data, covering a wide variety of different biological levels. Figure 1 . 1The DAA anatomical model. Moving the mouse over a given organ reveals the number of age-related changes in the DAA, along with a breakdown of the number of each specific type of change. Colours indicate the number of changes for each change type (orange: physiological, red: pathological, blue: molecular, green: psychological). Figure 3 . 3The details page for the gene GH1. This shows the two ageing changes associated with it and the links to external resources including GO terms, orthologs and various other databases. Figure 4 . 4Storing changes for later analysis. A combination of two screenshots showing how changes can be added to the saved list and then compared against each other using the graphing capabilities of the Digital Ageing Atlas. Filters allow for a narrowing of results based on the properties of each change; Multiple filters can be applied. The actions column provides the ability to add and remove changes to the stored list. Table 1 . 1The number of human age-related changes for each category in the Digital Ageing AtlasType of change Description ACKNOWLEDGEMENTWe would like to thank Ana Fonseca, Gerald Keil and Krishna Madireddy for useful suggestions, testing and contributions. Messages from mortality: the evolution of death rates in the old. L Partridge, M Mangel, Trends Ecol. Evol. 14Partridge,L. and Mangel,M. (1999) Messages from mortality: the evolution of death rates in the old. Trends Ecol. Evol., 14, 438-442. B Arking, Biology of Aging: Observations and Principles. USAOxford University PressArking,B. (2006) Biology of Aging: Observations and Principles. Oxford University Press, USA. Physiological Basis of Aging and Geriatrics. P S Timiras, CRC PressBoca Raton, FLTimiras,P. S. (1994) Physiological Basis of Aging and Geriatrics. CRC Press, Boca Raton, FL. Systems biology of ageing and longevity. T B Kirkwood, Philos. Trans. R. Soc. Lond. B. Biol. Sci. 366Kirkwood,T. B. (2011) Systems biology of ageing and longevity. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 366, 64-70. Systems biology and longevity: an emerging approach to identify innovative anti-aging targets and strategies. E Cevenini, E Bellavista, P Tieri, G Castellani, F Lescai, M Francesconi, M Mishto, A Santoro, S Valensin, S Salvioli, Curr. Pharm. Des. 16Cevenini,E., Bellavista,E., Tieri,P., Castellani,G., Lescai,F., Francesconi,M., Mishto,M., Santoro,A., Valensin,S., Salvioli,S. et al. (2010) Systems biology and longevity: an emerging approach to identify innovative anti-aging targets and strategies. Curr. Pharm. Des. 16, 802-813. . J P De Magalhães, J Curado, G M Church, de Magalhães,J. P., Curado,J. and Church,G. M. (2009) Meta-analysis of age-related gene expression profiles identifies common signatures of aging. Bioinformatics. 25Meta-analysis of age-related gene expression profiles identifies common signatures of aging. Bioinformatics 25, 875-881. Gene expression profile of aging in human muscle. S Welle, A I Brooks, J M Delehanty, N Needler, C A Thornton, Physiol. Genomics. 14Welle,S., Brooks,A. I., Delehanty,J. M., Needler,N. and Thornton,C. A. (2003) Gene expression profile of aging in human muscle. Physiol. Genomics 14, 149-159. Gene regulation and DNA damage in the ageing human brain. T Lu, Y Pan, S Y Kao, C Li, I Kohane, J Chan, B A Yankner, Nature. 429Lu,T., Pan,Y., Kao,S. Y., Li,C., Kohane,I., Chan,J. and Yankner,B. A. (2004) Gene regulation and DNA damage in the ageing human brain. Nature 429, 883-891. NCBI GEO: archive for functional genomics data sets-update. T Barrett, S E Wilhite, P Ledoux, C Evangelista, I F Kim, M Tomashevsky, K A Marshall, K H Phillippy, P M Sherman, M Holko, Nucleic Acids Res. 41Barrett,T., Wilhite,S. E., Ledoux,P., Evangelista,C., Kim,I. F., Tomashevsky,M., Marshall,K. A., Phillippy,K. H., Sherman,P. M., Holko,M. et al. (2013) NCBI GEO: archive for functional genomics data sets-update. Nucleic Acids Res. 41, D991-D995. Finding the right questions: exploratory pathway analysis to enhance biological discovery in large datasets. T Kelder, B R Conklin, C T Evelo, A R Pico, PLoS Biol. 81000472Kelder,T., Conklin,B. R., Evelo,C. T. and Pico,A. R. (2010) Finding the right questions: exploratory pathway analysis to enhance biological discovery in large datasets. PLoS Biol. 8, e1000472. Visualizing information across multidimensional post-genomic structured and textual databases. Y Tao, C Friedman, Y A Lussier, Bioinformatics. 21Tao,Y., Friedman,C. and Lussier,Y. A. (2005) Visualizing information across multidimensional post-genomic structured and textual databases. Bioinformatics. 21, 1659-1667. Human ageing genomic resources: integrated databases and tools for the biology and genetics of ageing. R Tacutu, T Craig, A Budovsky, D Wuttke, G Lehmann, D Taranukha, J Costa, V E Fraifeld, De Magalhães, Nucleic Acids Res. 41J. P.Tacutu,R., Craig,T., Budovsky,A., Wuttke,D., Lehmann,G., Taranukha,D., Costa,J., Fraifeld,V. E. and de Magalhães,J. P. (2013) Human ageing genomic resources: integrated databases and tools for the biology and genetics of ageing. Nucleic Acids Res. 41, D1027-D1033. Simplified ontologies allowing comparison of developmental mammalian gene expression. A Kruger, O Hofmann, P Carninci, Y Hayashizaki, W Hide, Genome Biol. 8229Kruger,A., Hofmann,O., Carninci,P., Hayashizaki,Y. and Hide,W. (2007) Simplified ontologies allowing comparison of developmental mammalian gene expression. Genome Biol. 8, R229. The BRENDA Tissue Ontology (BTO): the first all-integrating ontology of all organisms for enzyme sources. M Gremse, A Chang, I Schomburg, A Grote, M Scheer, C Ebeling, D Schomburg, Nucleic Acids Res. 39Gremse,M., Chang,A., Schomburg,I., Grote,A., Scheer,M., Ebeling,C. and Schomburg,D. (2011) The BRENDA Tissue Ontology (BTO): the first all-integrating ontology of all organisms for enzyme sources. Nucleic Acids Res., 39, D507-D513. The Aging Body: Physiological Changes and Psychological Consequences. S K Whitbourne, Springer-VerlagNYWhitbourne,S. K. (1985) The Aging Body: Physiological Changes and Psychological Consequences. Springer-Verlag, NY. AGEMAP: a gene expression database for aging in mice. J M Zahn, S Poosala, A B Owen, D K Ingram, A Lustig, A Carter, A T Weeraratna, D D Taub, M Gorospe, K Mazan-Mamczarz, PLoS Genet. 3201Zahn,J. M., Poosala,S., Owen,A. B., Ingram,D. K., Lustig,A., Carter,A., Weeraratna,A. T., Taub,D. D., Gorospe,M., Mazan-Mamczarz,K. et al. (2007) AGEMAP: a gene expression database for aging in mice. PLoS Genet. 3, e201. Allen Brain Atlas: an integrated spatio-temporal portal for exploring the central nervous system. S M Sunkin, L Ng, C Lau, T Dolbeare, T L Gilbert, C L Thompson, M Hawrylycz, C Dang, Nucleic Acids Res. 41Sunkin,S. M., Ng,L., Lau,C., Dolbeare,T., Gilbert,T. L., Thompson,C. L., Hawrylycz,M. and Dang,C. (2012) Allen Brain Atlas: an integrated spatio-temporal portal for exploring the central nervous system. Nucleic Acids Res. 41, D996-D1008. AgeFactDB-the JenAge ageing factor database-towards data integration in ageing research. R Hühne, T Thalheim, J Sühnel, Nucleic Acids Res. 42Hühne,R., Thalheim,T. and Sühnel,J. (2014) AgeFactDB-the JenAge ageing factor database-towards data integration in ageing research. Nucleic Acids Res. 42, D892-D896.

2022-01-15T10:33:33Z

CCBY

https://www.mdpi.com/1422-0067/22/3/1073/pdf

GOLD

8e7ed1b584fe0542b030798e61721097dbea002d

10.3390/ijms22031073

33499037

7865694

Machine Learning Analysis of Longevity-Associated Gene Expression Landscapes in Mammals 2021 Anton Y Kulaga Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania International Longevity Alliance 92330SceauxFrance CellFabrik SRL 060512BucharestRomania Eugen Ursu Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania Dmitri Toren Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania The Shraga Segal Department of Microbiology, Immunology and Genetics Faculty of Health Sciences Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 8410501Beer-ShevaIsrael Vladyslava Tyshchenko SoftServe Inc 49044DniproUkraine Rodrigo Guinea Escuela de Postgrado, Pontificia Universidad Católica del Perú 15023San MiguelPeru Malvina Pushkova Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania Vadim E Fraifeld The Shraga Segal Department of Microbiology, Immunology and Genetics Faculty of Health Sciences Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 8410501Beer-ShevaIsrael Robi Tacutu Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania Machine Learning Analysis of Longevity-Associated Gene Expression Landscapes in Mammals International Journal of Molecular Sciences Article Int. J. Mol. Sci 221073202110.3390/ijms22031073Received: 15 December 2020 Accepted: 20 January 2021 Citation: Kulaga, A.Y.; Ursu, E.; Toren, D.; Tyshchenko, V.; Guinea, R.; Pushkova, M.; Fraifeld, V.E.; Tacutu, R. Machine Learning Analysis of Longevity-Associated Gene Expression Landscapes in Mammals. Introduction Numerous studies have showed that the average lifespan, and in some cases even maximum lifespan (MLS), could be modified by genetic interventions. Hundreds of genes have been shown to be involved in the control of longevity in model organisms or in the etiopathogenesis of aging-related diseases, with many being highly conserved and interacting in a cooperative manner [1][2][3]. Still, until now, only a~1.5-fold lifespan increase has been achieved through genetic interventions in mammals [4], and even less with pharmacological interventions [5,6]. In contrast, MLS varies in at least a 100-fold range across the Mammalia class [3], hinting that the comparative biology of aging has not been exhausted yet and novel genetic interventions might still be discovered by looking at the differences between various species. Studying the variations in MLS and transcription across multiple species is an informative method for investigating the evolution of longevity. Recent studies demonstrated that differences in gene expression between long-and short-living mammals exist [7][8][9][10]. Results and Discussion Data Collection and Processing of Gene Expression across Mammalian Species Using publicly available RNA-Seq data, we build a cross-species dataset of gene expression levels. The dataset consists of 408 samples from 41 mammalian species and covers five organs: liver, kidney, lung, brain, or heart (the full list of species and sequencing run IDs is available in Supplementary Table S1). The dataset was normalized, processed, and further augmented with species data for studying the associations among gene expression levels and systemic species variables: MLS, body mass, temperature, metabolic rate, gestation period and GC content of mitochondrial DNA, all of which have been suggested to be determinants of MLS [15][16][17]. Linear, LightGBM-SHAP, and Bayesian network models were employed to identify and describe the associations among gene expression levels and MLS (as described in detail in Figure 1). Independent results from three approaches were integrated to investigate which genes will appear as the top MLS predictors, regardless of the methodology differences. Linear Correlations between Gene Expression and Maximum Lifespan To investigate to what extent the expression level of evolutionarily conserved genes correlates with MLS across mammals, linear models were first constructed for 11,831 orthologous genes that are found between 33 mammalian species (Figure 2a). For each of these genes, the coefficient of determination (R 2 ), which indicates how well the trained linear models explain the MLS variability, was computed. The numbers of genes that were significantly associated with MLS in every organ under analysis are as follows: brain-381, liver-390, kidney-154, heart-535, and lung-756. The median R 2 was similar across organs: for brain-0.36, liver-0.36, kidney-0.35, heart-0.38, and lung-0. 38. The analysis of the linear models identified that only three genes (CRYGS, TCFL5, SPATA20) have significant positive correlations with MLS (FDR < 0.05, R 2 > 0. 3), in a consistent manner among all five studied organs. It should be noted that the sample size for the heart and lung is relatively lower than that for the other organs, which is mainly due to the generally lower availability for these samples. As such, this bias could be responsible for the small number of genes found to associate in all of the five organs. Consequently, we also looked at the significant correlations that are observable only in the organs with a high sample size: brain, liver, and kidney. The results led to a slightly extended list of 12 genes (SPATA20, TCFL5, TIMP1, HSPB1, RASSF4, SLC25A23, NASP, CCDC14, A2M, NOXA1, C20orf96, CRYGS) whose expression correlates with MLS (FDR < 0.05, R 2 > 0. 3) in the brain, liver, and kidney. For a full list of genes associated with MLS and other species' features, see Supplementary Table S2. Genes that are predictive for MLS might also correlate with at least one other lifehistory trait. For example, it is known that MLS correlates with body mass, body temperature, metabolic rate, gestation age, and mitochondrial GC%, which raises the possibility that the associations with MLS are in fact found due to indirect causes. In the brain and kidney, we identified no genes that correlate with MLS uniquely (i.e., genes whose expression correlates with MLS, but not with other variables). In the liver, only one gene (CERS4) correlates with MLS, but not with the other investigated traits. In the heart and lung, we identified 4 and 131 unique associations, respectively, but conclusions drawn from these two organs might be biased because of the lower sample size (we had access to 28 lung samples from 16 species, compared to 121 liver samples from 30 species). Linear Correlations between Gene Expression and Maximum Lifespan To investigate to what extent the expression level of evolutionarily conserved genes correlates with MLS across mammals, linear models were first constructed for 11,831 orthologous genes that are found between 33 mammalian species (Figure 2a). For each of these genes, the coefficient of determination (R 2 ), which indicates how well the trained linear models explain the MLS variability, was computed. The numbers of genes that were significantly associated with MLS in every organ under analysis are as follows: brain-381, liver-390, kidney-154, heart-535, and lung-756. The median R 2 was similar across organs: for brain-0.36, liver-0.36, kidney-0.35, heart-0.38, and lung-0.38. The analysis of the linear models identified that only three genes (CRYGS, TCFL5, SPATA20) have significant positive correlations with MLS (FDR < 0.05, R 2 > 0.3), in a consistent manner among all five studied organs. It should be noted that the sample size for the heart and lung is relatively lower than that for the other organs, which is mainly due to the generally lower availability for these samples. As such, this bias could be responsible for the small number of genes found to associate in all of the five organs. Consequently, we also looked at the significant correlations that are observable only in the organs with a high sample size: brain, liver, and kidney. The results led to a slightly extended list of 12 genes (SPATA20, TCFL5, TIMP1, HSPB1, RASSF4, SLC25A23, NASP, CCDC14, A2M, NOXA1, C20orf96, CRYGS) whose expression correlates with MLS (FDR < 0.05, R 2 > 0.3) in the brain, liver, and kidney. For a full list of genes associated with MLS and other species' features, see Supplementary Table S2. Genes that are predictive for MLS might also correlate with at least one other lifehistory trait. For example, it is known that MLS correlates with body mass, body temperature, metabolic rate, gestation age, and mitochondrial GC%, which raises the possibility that the associations with MLS are in fact found due to indirect causes. In the brain and kidney, we identified no genes that correlate with MLS uniquely (i.e., genes whose expression correlates with MLS, but not with other variables). In the liver, only one gene (CERS4) correlates with MLS, but not with the other investigated traits. In the heart and lung, we identified 4 and 131 unique associations, respectively, but conclusions drawn from these two organs might be biased because of the lower sample size (we had access to 28 lung samples from 16 species, compared to 121 liver samples from 30 species). Supplementary Table S2). (b) Top linear correlations for MLS and pathway enrichment scores. The heatmap represents the significant associations between MLS and the computed enrichment score (ES) for pathways, obtained using the signature projection approach, which takes into account the expression of all the expressed genes belonging to each pathway. The presented pathways are grouped by categories and include the following: 1) preselected pathways that had been previously shown to be associated with longevity (by independent studies), and 2) additional pathways identified by the current analysis to be associated significantly with MLS in at least four organs. Quantitative details on pathway score associations with other species features can be found in Supplementary Table S3. (a,b) Red represents significant positive associations for genes or pathways; blue represents significant negative associations. Colorless represents no significant associations found in this study. Linear Relationships between Maximum Lifespan and Pathway Enrichment Scores In order to expand our perspective from individual genes to biological pathways, we used the signature projection approach (ssGSEA) to assess the association between pathway activity (estimated from gene expression) in various organs and MLS (Figure 2b). The ssGSEA method allows for transforming the gene expression space into the biological pathways' activity space using prior knowledge in the form of gene-pathway association sets. For this, we first selected a list of pathways, such as the mTOR signaling pathway (hsa04150), Insulin signaling pathway (hsa04910), DNA repair pathways (e.g., base exci- Supplementary Table S2). (b) Top linear correlations for MLS and pathway enrichment scores. The heatmap represents the significant associations between MLS and the computed enrichment score (ES) for pathways, obtained using the signature projection approach, which takes into account the expression of all the expressed genes belonging to each pathway. The presented pathways are grouped by categories and include the following: 1) preselected pathways that had been previously shown to be associated with longevity (by independent studies), and 2) additional pathways identified by the current analysis to be associated significantly with MLS in at least four organs. Quantitative details on pathway score associations with other species features can be found in Supplementary Table S3. (a,b) Red represents significant positive associations for genes or pathways; blue represents significant negative associations. Colorless represents no significant associations found in this study. Linear Relationships between Maximum Lifespan and Pathway Enrichment Scores In order to expand our perspective from individual genes to biological pathways, we used the signature projection approach (ssGSEA) to assess the association between pathway activity (estimated from gene expression) in various organs and MLS (Figure 2b). The ssGSEA method allows for transforming the gene expression space into the biological pathways' activity space using prior knowledge in the form of gene-pathway association sets. For this, we first selected a list of pathways, such as the mTOR signaling pathway (hsa04150), Insulin signaling pathway (hsa04910), DNA repair pathways (e.g., base excision repair (hsa03410), homologous recombination (hsa03440)), ubiquitin-mediated proteolysis (hsa04120), focal adhesion (hsa04510), etc., which have already been linked to aging or longevity [2,18,19], and then we applied ssGSEA to them. Interestingly, no significant association was identified for the mTOR pathway, suggesting perhaps that the MLS modulation by mTOR is working amongst the individuals of one species rather than optimizing MLS at the inter-species level, or through other mechanisms such as post-translational modification (e.g., phosphorylation). Contrastingly, for the genes involved in the insulin pathway, we did identify a positive association for the kidney (p = 0.005). We also identified multiple strong positive associations of the enrichment score (ES) for several of the DNA repair pathways in the brain: mismatch repair (MMR) (p = 3.27 × 10 −5 ), nucleotide excision repair (NER) (p = 8.35 × 10 −5 ), base excision repair (BER) (p = 5.46 × 10 −11 ), homologous recombination (HR) (p = 6.64 × 10 −7 ), non-homologous end-joining (NHEJ) (p = 4.60 × 10 −8 ). Besides the brain, the BER pathway also shows a significant positive association in both liver (p = 1.27 × 10 −6 ) and kidney (p = 0.001), while for the HR pathway the ES is significantly associated only in the liver (p = 0.003). For the ubiquitin-mediated proteolysis, we identified a small negative association in liver (p = 2.88 × 10 −5 ) and brain (p = 6 × 10 −4 ), but also for the proteasome pathway in liver (p = 0.01), whereas for the ubiquinone and other terpenoid-quinone biosynthesis pathways, we found strong negative associations in the liver (p = 1.64 × 10 −9 ), kidney (p = 0.001), brain (p = 6.82 × 10 −9 ) and heart (p = 1.52 × 10 −5 ). Focusing on cell adhesion, the focal adhesion pathway displays a small but highly significant positive association, which was detected in the liver (p = 1.39 × 10 −5 ) and kidney (p = 7 × 10 −4 ), and for cell adhesion molecules associations were found in the liver (p = 9.71 × 10 −7 ), brain (p = 0.02) and kidney (p = 2.70 × 10 −6 ). Besides exploring the pathways known to be associated with longevity, we also investigated the pathways that were detected to have significant correlations consistently, i.e., with the same direction of the association, in at least four out of five organs. A set of KEGG pathways involved in infection, inflammation, and the immune response were found to be associated with MLS in multiple organs (Figure 2b), including allograft rejection, asthma, autoimmune thyroid disease, complement and coagulation cascades, influenza A, intestinal immune network for IgA production, measles, systemic lupus erythematosus, and viral myocarditis. To our knowledge, these pathways have not been established as longevity pathways; however, upregulation of the immune response and inflammation with MLS has been shown in both cross-species transcriptomics [9], as well as in studies of long-living animals such as naked mole-rats [10,20]. Enrichment for pathways that are specific to humans (e.g., measles) can be found because of genes from the corresponding gene sets that are involved more generally in inflammation and immunity (e.g., IL-2, IFNα/β). Three pathways involved in metabolism were found to be negatively associated with MLS: fatty acid metabolism, glutathione metabolism, and glycerolipid metabolism. Previous studies have shown a positive association of membrane fatty acid composition [21], and a negative association of glutathione levels in the liver, with lifespan in vertebrates [22]. A negative association of glycerolipid metabolism with lifespan was also found in a cross-species lipidomics study [23]. We also found a negative association for the PPAR signaling pathway. PPARs play an important role in regulating metabolism, and several PPARs display a decreased level with aging, PPAR activity being associated with rising levels of inflammatory mediators during aging [24,25]. Other pathways positively associated with MLS include apoptosis, cell adhesion molecules, dorso-ventral axis formation, ErbB signaling pathway and phototransduction. Some of these have been indirectly linked to health previously; for example, a decreased expression of ErbB signaling in humans is associated with neurodegenerative diseases [26]. This is in agreement with our results, which indicate that increased ErbB signaling is associated with increasing mammalian MLS. Overall, the linear models in our study emphasize the transcriptomic differences that correlate most with mammalian MLS. The results presented above are generally in line with the findings presented by other cross-species studies, both based on gene expression, but also on metabolomics and lipidomics. In addition to this, we suggest that several pathways, which have been relatively less studied with regard to cross-species lifespan variation, might also play important roles in longevity: PPAR signaling, glutathione metabolism and ErbB signaling. It is important to note, however, that those linear models detect statistical associations that are not necessarily causal for increased longevity. First of all, known or unknown confounding variables could be responsible for the association of expression with MLS. Even if the expression is directly associated to lifespan (without any confounding elements), the results do not allow us to detect whether a pathway is associated with longevity because their components contribute biologically to the increased lifespan (making them "pro-longevity expression traits"), or whether it is associated with MLS due to a deteriorating activity that occurs with aging, and for which longer MLS means longer deterioration (making them "deterioration markers of extended lifespans"). Nevertheless, the detected pathways highlight biological processes whose activities change significantly across the spectrum of mammalian MLS, and which warrant further study in relation to longevity. SHAP Explanations for Universal Gene Expression Patterns To also investigate the potentially non-linear patterns of association existing between gene expression and lifespan, we further used an interpretable machine learning approach. Briefly, we applied a gradient boosting decision tree algorithm, using LightGBM [27], to select genes associated with MLS. We then applied SHAP (Shapley additive explanations), a game-theory approach that can be used to explain the output of machine learning models [28], to the LightGBM model, both as part of the selection process and as the main interpretation method. Briefly, the key difference between LightGBM and its predecessors is mainly the higher accuracy and computational efficiency. LightGBM has proven (i) to be highly effective for tabular data; (ii) to be a highly explainable model, especially when combined with the SHAP framework; (iii) that it is not sensitive to correlations in the features of the dataset, which is expected for gene expression data; and (iv) that, in comparison to other models (such as neural networks and SVR), it is less prone to overfitting on wide (many features) and small (few samples) datasets. LightGBM is a non-linear regression model that, once trained, can be used to derive the importance of the employed features (i.e., genes), in predicting the target variable (i.e., MLS), therefore providing means for feature ranking and feature selection. In order to minimize the possibility for selection results to occur by chance, a fold stratified crossvalidation procedure is applied and repeated multiple times. Only features selected in all the repeated models were considered further. For a particular single prediction, i.e., a sample, a SHAP value for a particular feature (i.e., a gene) represents the impact (i.e., the contribution) of the feature on the model's predicted MLS within that sample. The SHAP value results from the difference between the prediction when the feature takes a certain value (i.e., the expression in a sample) and the prediction that would be made if the feature took a random value, the latter being an estimation of the average model prediction. One can get a measure of the global feature importance of a gene by aggregating the Shapley values for the gene across all the predictions (i.e., samples). Initially, we generated the SHAP explanations for life-history traits without including gene-expression data, which showed the high influence of the mitochondrial DNA GC content and of the gestation period on determining MLS (see Supplementary Figure S1). This LightGBM-SHAP model reached a Huber loss = 2.19, MAE = 2.86, MSE = 28.28 and R 2 = 0.96 when predicting MLS, with other life-history traits being considered (maximum lifespan, body mass, temperature, metabolic rate, gestation period, and mitochondrial GC%). Of these features, mitochondrial DNA GC content and gestation period were the most impactful traits (Supplementary Figure S1a). This is consistent with previous findings: the gestation period has been linked to senescence [29] and mitochondrial DNA content has been associated with determination of MLS [16]. In addition, several SHAP interaction effects were observed when predicting MLS: between mitochondrial GC content and body mass, between mitochondrial GC content and gestation period, and between body mass and temperature (Supplementary Figure S1b), meaning that these traits conjointly influence MLS variation across species. Next, we investigated the effects of various genes whose expression level has an impact on MLS and other life-history traits. For this purpose, we used a two-stage backward feature selection strategy (explained in detail in the methods section). When predicting MLS, the stage II LightGBM-SHAP model used the genes selected by the stage I models as input (155 genes To prioritize genes from the stage II model, we relied on SHAP feature importance (mean absolute contribution) as a way to evaluate the global importance of an individual gene in predicting MLS. That is, for each gene, we have computed the absolute value of SHAP for every gene expression sample, and subsequently, these values have been averaged to obtain the SHAP feature importance. In total, 57 genes that have a mean absolute contribution of more than 0.1 years on MLS prediction were identified. A SHAP summary plot for the top genes which are most predictive for MLS is provided in Figure 3. For many genes, the distribution of SHAP values appears to be skewed to the right (Figure 3a), i.e., the expression of these genes strongly impacts lifespan prediction, positively, in many samples, while most of the genes that impact it negatively are only slightly below zero. In particular, the expressions of the most impactful genes, DYRK4 and NFKBIL1, make a very strong positive contribution (SHAP > +20 in added years) to the MLS prediction for the human samples, as seen in Figure 3b (right-most red trajectories) and Figure 3d (leftmost columns). Out of the top 15 genes, 5 genes (DYRK4, NFKBIL1, TRAPPC2L, ETV2, CHCHD3) had a significant mean absolute impact (more than 1 year) on the model's MLS prediction. Moreover, two of these five genes have been previously linked to aging: NFKBIL1 is involved in multiple methylation aging clocks [30], and has been associated with accelerated aging and cellular senescence in studies with genetically engineered mice [31], while CHCHD3 participates in the cross-talk between mitochondrial fusion and the hippo pathway in controlling cell proliferation (apoptosis) [32], which were both found to be involved in longevity in C. elegans [33]. STAG3 is essential for maintaining centromere chromatid cohesion and is required for DNA repair and synapsis between homologous chromosomes [34]. It is also known that DYRK4 has different roles in short-and long-living animals. In particular, mouse isoforms of DYRK4 are shorter and expressed mostly in the testis, while human isoforms are longer, expressed in many organs, and differ in localization and substrate specificity [35]. Despite the associations of many DYRKs family genes with neuronal development, Down syndrome, and age-related neurodegenerative diseases [36], DYRK4 still remains understudied in relation to longevity. Besides correlating with MLS, our analysis has shown that these five genes are also predictive for other life-history traits, in particular, DYRK4 and NFKBIL1 are predictive for body mass, DYRK4 and ETV2 for gestation days, TRAPPC2L for mtGC, and NFKBIL1 with CHCHD3 for metabolic rate. To our knowledge, there is no information about the links between TRAPPC2L, ETV2, and aging; however, based on the SHAP analysis, it could be suggested that they might be novel candidates as longevity regulators and should be further investigated. Out of the top 15 genes, 5 genes (DYRK4, NFKBIL1, TRAPPC2L, ETV2, CHCHD3) had a significant mean absolute impact (more than 1 year) on the model's MLS prediction. Calculating the correlation between SHAP values corresponding to gene expression levels, relative to the baseline value (21.6 years; the estimated value obtained if all genes are used), allows for estimating the correlation of the target (i.e., MLS) with the feature (i.e., gene expression), by isolating the effects of other features (genes) on the target. By computing the Kendall tau-b ranked correlation coefficient (further denoted as Kendall tau), we classified the selected genes with regard to their positive or negative impact on the model prediction (further referred to as pro-MLS and anti-MLS, respectively). NEIL1, NOXRED1, CALCOCO2, CEL, C1orf56 and LRR1 are the strongest pro-MLS genes (Kendall tau ≥ 0.6) and C6orf89, PPP1CA, DNAJC15, DPP9 and VARS2 are the strongest anti-MLS genes (Kendall tau ≤ −0.6). Of note, the impact direction of the pro-and anti-MLS genes resulting from the LightGBM-SHAP models generally coincides with the ones computed by organspecific linear models. From the 57 genes selected by the stage II LightGBM-SHAP model, 37 (65%) are found to be significantly associated with MLS (FDR < 0.05, R 2 > 0.3) by the linear models in at least one organ. The pro-/anti-longevity direction computed as the sign of Kendall tau in LightGBM-SHAP coincides for all 37 genes with the sign of the regression slope in the linear models (additionally, the direction is in concordance for significant results in more than one organ, as well). The strongest pro-MLS gene, CALCOCO2, had a positive Kendall tau of 0.72 in the LightGBM-SHAP model and was also selected as a pro-MLS gene by the linear models in the lung (R 2 = 0.56), brain (R 2 = 0.54), liver (R 2 = 0.53) and heart (R 2 = 0.39), while the C6orf89 gene that had a negative Kendall tau of -0.79 was selected as anti-MLS by linear models in the heart (R 2 = 0.61) and liver (R 2 = 0.40). The C6orf89 gene is linked to the NF-κB system [37], whose overactivation is harmful to humans, through a series of age-related processes such as a chronic inflammatory response, increases in apoptotic resistance, a decline in autophagic cleansing, and tissue atrophy [38]. The negative role of PPP1CA in aging has also been recorded, as for instance, it has been found in mice that it plays a role in cognitive aging and its overexpression in cardiac cells resulted in premature heart failure [39]. Interactions between MLS-Associated Genes A gene may impact longevity not only by itself, but also by cooperation with other genes. Based on the cumulative knowledge of interactions between longevity-associated genes (LAGs), it has been previously shown that genes that have a role in determining average or maximum lifespan (as LAGs for example) may have a variety of combined effects-synergistic, additive, dependent, or antagonistic [40]. Thus, the total lifespan changes as a result of genetic interventions targeting two genes are usually not the simple sum of their impacts. To estimate the two-way interactions of the genes selected as potential MLS-associated genes, we also computed SHAP interaction values. The SHAP interaction value of two genes on a sample is the contribution to the prediction of the combined genes after accounting for the contributions of the individual two genes. The matrix shows the strength of interactions, by plotting the difference between the combined SHAP value of a gene pair and the sum of their individual SHAP values effects (depicted by the color intensity in Figure 3c). The results showed that each pair of genes has different effects on MLS prediction. As is shown in Figure 3c, the following gene pairs might have a strong interaction: DYRK4 and NFKBIL1, DYRK4 and RNH1, STAG3 and RNH1, and also TRAPPC2L and ETV2. We explored the gene pairs with the highest magnitude of interaction (Supplementary Figure S2). DYRK4 is a top gene in terms of overall positive MLS impact; however, when NFKBIL1 is highly expressed, DYRK4 increases MLS to a lower extent (Supplementary Figure S2a). RNH1 (ribonuclease/angiogenin inhibitor 1) is a gene with a high number of interactions. RNH1 is a known regulator of vascularization and a mediator of oxidative stress, which has antioxidant [41] and redox homeostatic [41] effects. As can be seen in the SHAP feature dependency plots, RNH1 has very similar nonlinear interactions with both CAPN3 (Supplementary Figure S2b), DYRK4 (Supplementary Figure S2c), and STAG3 (Supplementary Figure S2d). In all three pairs, the high expression of CAPN3, DYRK4, and STAG3 increases the magnitude of RNH1's MLS impact in both positive and negative directions, while their low expression values keep the RNH1 impact close to zero. Bayesian Networks In this section, a Bayesian networks approach was used to identify genes that have the potential to be causally associated with MLS while accounting for redundancy and spuriousness. In this context, the potentially causal association of any pair of variables (i.e., gene expression values) was defined as being asserted whenever the correlation between the two holds, regardless of the values others could take (i.e., whenever the two variables do not display conditional independence) [42]. For this, first we constructed a Bayesian network, using the genes included in this study and MLS. We employed the notion of a Markov blanket to identify genes that might be causally linked to MLS-the Markov blanket of MLS is the set of genes that are parents, children, or parents of children of the MLS node in the Bayesian network. By definition, the variability in the genes from the Markov blanket of MLS will contain all the useful information about MLS. Subsequently, "potentially causal" gene signatures were defined as gene subsets of the Market blanket of MLS. Implementation-wise, gene signatures were found with SES, a constrained-based variable selection algorithm [43]. While the above-mentioned approach does not fully guarantee causality, it allows a data-driven exploratory analysis to identify valid causal inferences under strict assumptions. In order to give a causal interpretation in an absolute sense, three assumptions would have to hold: the causal Markov assumption, faithfulness, and causal sufficiency [44]. Unfortunately, with only transcriptomics data, which do not include all possible confounders, completely proving causality is not practically achievable, and the employed algorithm deals with these assumptions as best as possible (by construction, the causal Markov condition holds, and using several stratified partitions, the dataset's underlying conditional independence structure could be approximated, thus attaining faithfulness). Even if causal sufficiency cannot be proven, the results obtained with this method still provide important information about the conditional independence structure between genes and MLS [45]. The Bayesian network analysis included 50 iterations, each corresponding to one stratified train-validation partition, resulting in a set of 50 gene signatures (see Supplementary Table S4 and Figure S3; for methodological details, please see Material and Methods). Supplementary Table S5 shows the relative frequency associated with each gene, representing the proportion of times it was included as part of a signature (i.e., when its p-value < 0.01) from the set of all signatures (50). The relative frequencies can be used to rank and therefore prioritize genes with respect to potential causal relations with MLS. Considering all 50 signatures, the most robustly/frequently included genes were the following: NOXA1, C6orf89, NEU2, NDUFA6, RBM46, KCNMB3, and CEL, with relative frequencies of 1.00, 0.94, 0.94, 0.90, 0.82, 0.72, and 0.60, respectively (Supplementary Table S5). From a biological point of view, the obtained results are in line with those from the previous section, as NOXA1, C6orf89, and CEL were also identified to be important for MLS determination by LightGBM-SHAP. Additionally, it has been shown that NEU2 upregulation triggers myoblast differentiation in C2C12 cells, and since it is involved in the growth and differentiation of satellite cells, it might implicate the regenerative capabilities of organs such as muscle or heart [46]. NDUFA6 is a key component in Complex I [47,48], which was found to be a biomarker of aging in mice [49] and is downregulated with age in humans [47]. RBM46 might be indirectly linked to aging through its role in the degradation of β-Catenin mRNA in mice [50], and thus through proteasomal degradation in the Wnt/β-signaling pathways. Finally, KCNMB3 has been found to be important for insulin signaling and β-cell function [51]. Taking these genes together and performing a network-based enrichment analysis, it appears that their most enriched biological functions are mitochondria-related, such as the NADH dehydrogenase complex, mitochondrial respiratory chain, mitochondrial organization, and mitochondrial metabolism disease. Integration of Linear, LightGBM-SHAP, and Bayesian Networks Models To find the overlaps between the genes predicted by the models employed in this study, we compared the most predictive genes from SHAP explanations, Bayesian networks, and organ-based linear models. Joint Predictions in Linear and LightGBM-SHAP Models In most cases, the positive and negative predictive impact of the genes was shared between LightGBM-SHAP (pro-and anti-MLS genes) and linear organ-based models (positively and negatively correlated genes). The pro-MLS genes NOXA1, KCNMB3, CEL, CALCOCO2, LRR1, CAPN3, HRH4, C1orf56 and FIGNL1, which were selected by LightGBM-SHAP, were also selected for different organs by linear models (Supplementary Table S6). Of note, CALCOCO2 and LRR1 have been selected by linear models in all of the organs, hinting that they might be universal determinants. CALCOCO2 (also known as NDP52) is involved in innate immunity and autophagy, which declines with age [52], while Leucine-rich repeat protein 1 (LRR-1) is known to be a determinant of genome stability [53,54]. It is known that FIGNL1 is related to homologous DNA repair [55], HRH4 encodes a histamine receptor that is predominantly expressed in hematopoietic cells and is known to be associated with age-related macular degeneration [56], and KCNMB3 is important for insulin signaling and β-cell function, but its association with glucose-related traits is still unclear [51]. The CAPN3 gene is a major intracellular protease, and some in silico associations with aging exist [57]. NOXA1 is an enhancer for NADPH, which is one of the major reactive oxygen species sources [58], while C1orf56 and CEL are potentially novel LAGs. Several anti-MLS genes in the SHAP analysis were also selected in the linear models, including C6orf89 and DPP9 (Supplementary Table S6). With regards to their functions, it has been shown that C6orf89 exhibits histone deacetylase (HDAC) enhancer properties [59], while DPP9 regulates mitochondrial protein levels and localization [60], and is linked to a variety of age-related pathologies including type 2 diabetes, obesity and cancer. Joint Predictions in Bayesian and LightGBM-SHAP Models Genes selected by the Bayesian networks model (the relative frequencies of genes that were part of gene-signatures are provided in the Supplementary Table S5) had high absolute Kendall tau values and were also selected by linear models in some of the organs. At the same time, these genes (NOXA1, C6orf89, NEU2, NDUFA6, RBM46, KCNMB3, and CEL) were not the most impactful in terms of mean absolute SHAP values. However, out of them CEL and KCNMB3 belong to the top 10 most impactful genes in terms of SHAP values. Joint Predictions in All Three Models Among the genes selected by all models, some strongly correlated with life-history traits and mitochondrial GC content. In particular, the SHAP values of NOXA1 (mitochondriaassociated gene) and KCNMB3 strongly and positively correlate with both mitochondrial GC content (NOXA1 Kendall tau = 0.65; KCNMB3 Kendall tau = 0.62) and gestation period (NOXA1 Kendall tau = 0.62; KCNMB3 Kendall tau = 0.64). We also observed a positive association between C1orf56 and mitochondrial GC content (Kendall tau = 0.6189), and between FIGNL1 and gestation period (Kendall tau = 0.41), as well as a negative association between C6orf89 and metabolic rate (Kendall tau = −0.57) (Supplementary Table S6). Composite Ranking To evaluate the genes selected by different models, the lists of the most predictive genes for linear, tree-based, and Bayesian networks models were combined, and a composite ranking was implemented (Supplementary Table S6). For this, each gene was assigned with a particular rank within each performance metric, and multiple ranks were aggregated (for detailed criteria, please see Methods). The composite ranking was then used to determine core gene signatures (as parsimonious as possible in terms of size) with as high an impact on the MLS as possible. Two approaches were used for this-one linear and one non-linear. For the linear approach, to study the final selection of genes and their ability to explain the variability in MLS, an organ-wise Bayesian multilevel linear model with random coefficients was used. The top 11 genes were selected based on the penalized deviance criterion. This can be interpreted as "sharing" information across regressions fitted for each organ-a reasonable assumption since we only have so many samples and we can safely assume that information about a particular organ can help us model the regressions done on other organs. As can be seen in Table 1, the genes considered to fit the regression were able to explain more than 70% of the MLS variability for each organ, the brain being the one with the greatest R 2 = 0.79 (liver displayed the smallest R 2 = 0.71). In the non-linear approach, to explain how the genes with the highest composite ranking impact MLS, multiple LightGBM-SHAP models were built with different numbers of top-ranked genes. According to the threshold criteria described in the methods, the top-six genes model was considered most significant, having the lowest number of genes, while having the highest increase in model performance. Upon the training of this stage III LightGBM-SHAP model, average Huber loss = 6.41, MAE = 7.45, MSE = 233.6 and R 2 = 0.67 were achieved in a five-fold cross-validation with genes CEL, SPATA20, C6orf89, NOXA1, CALCOCO2 and PPP1CA. Unlike the top genes of the stage II model (Figure 3a), the SHAP distribution of the six genes was more balanced, all genes having both positive and negative SHAP impacts, while better separated clustered samples for pro-MLS (CEL, SPATA20, NOXA1, CALCOCO2) and anti-MLS (C6orf89, PPP1CA) genes could be visually observed (Figure 4a,b). As shown in Figure 4c, the top six genes from the multi-model analysis correlate with systemic features. Remarkably, all six genes, in all five organs, correlate with body temperature, which was previously shown to be an independent determinant of mammalian longevity [17]. the SHAP distribution of the six genes was more balanced, all genes having both positive and negative SHAP impacts, while better separated clustered samples for pro-MLS (CEL, SPATA20, NOXA1, CALCOCO2) and anti-MLS (C6orf89, PPP1CA) genes could be visually observed (Figure 4a,b). As shown in Figure 4c, the top six genes from the multi-model analysis correlate with systemic features. Remarkably, all six genes, in all five organs, correlate with body temperature, which was previously shown to be an independent determinant of mammalian longevity [17]. Materials and Methods Bioinformatic Workflow and Analysis Design The comparison of expression levels for different species involves tackling many degrees of uncertainty and potential errors due to technical issues (discussed in detail by Toren et al., 2020). With this in mind, we approached the problem as a feature reduction problem, i.e., finding a small set of genes that is highly predictive for MLS variation between species in different organs. The designed bioinformatic pipeline includes processing expression data from multiple species and selecting potential LAGs based on three distinct approaches: linear based models to investigate organ-specific patterns, LightGBM-SHAP explanations models to research the impacts of individual genes and their interactions, and Bayesian networks models to identify potential causality relationships with MLS. Samples Selection and Data Quality All mammalian species with transcriptome annotations in the 99th release of the Ensembl Compara Database [61], for which RNA-Seq samples of healthy liver, kidney, lung, brain, or heart organs were available in the NCBI Sequence Read Archive, were selected in this study. The full list of species and sequencing run IDs can be found in Supplementary Table S1. For each species, samples from juvenile, diseased, and very old animals were discarded in order to control for signals pertaining to the developmental or potentially pathological state of the included organisms. Only species that had at least two RNA-Seq samples were kept. In addition, to account for the heterogeneity in expression values, each dataset has been checked for anomalous distributions by exploratory analysis; extreme outliers were manually analyzed and removed. Overall, 408 samples from 38 species and five organs have been selected and processed using the same RNA-Seq quantification pipeline to avoid heterogeneity in the data processing. Linear models were fitted on a reduced dataset in which 5 species were removed: 4 species that contributed with samples to a single organ and 1 species with outlier sample distribution (Canis lupus familiaris). Orthology Orthology relationships and transcriptome annotations have been obtained from Ensembl, release 99th, preprocessed, and imported to GraphDB, which was further used for the analysis of orthology and for expression value extraction. The preprocessing scripts have been developed in-house and are available as notebooks in our group's repository at https://github.com/antonkulaga/species-notebooks. For the linear models, expression values for genes in species with more than one paralog were considered as missing values to reduce ambiguity. The final dataset for linear models included 11,831 genes. For the LightGBM models, 12,323 orthologous coding genes were selected by combining one-to-one and high-confidence one-to-many orthologs. For this analysis, genes for which orthologs were not present in more than 90% of the species were excluded. Species Life-History Data Maximum lifespan, body mass, temperature, metabolic rate, and gestation period data have been obtained from the AnAge database, build 14 [3], and mitochondrial GC has been obtained from the MitoAge database, version 1.0 [62]. RNA-Seq Pipeline For quality control, adapter cutting and trimming, Fastp (version 0.20.1) [63] was used. Transcript quantification was done with salmon (version 1.4.0) [64]. Transcript expressions were aggregated at the gene level with tximport (version 3.12). Raw read counts were normalized on a per-sample basis, using the transcript per million (TPM) normalization. For performing the comparative analysis, the gene expression levels were normalized with TPM, which accounts for potential gene length differences across species. Linear Models The linear models' analysis was performed in Python (version 3.8), using several libraries, including statsmodels (version 0.11.1) [65], pandas (version 1.1.1), seaborn (version 0.10.1), pyUpsetPlots (version 0.4.0), Venn (version 0.1.3), etc. The code necessary to reproduce the analysis can be found at https://github.com/ursueugen/cross-specieslinear-models. To investigate the association between gene expression and MLS, organ-specific linear models were built, allowing for the selection of genes highly associated with MLS, in each organ. Single-variable linear models of the form GeneExpression~β 0 + β 1 x SpeciesTrait were fitted independently for every gene and every life-history trait. For this analysis, the variables were log 2 -transformed and normalized to z-scores, prior to fitting. For each model, missing data (species missing one-to-one ortholog, species with no samples in one of the organs, missing life-history trait) were ignored. Since the number of human samples/data points might bias the analysis and result in much higher leverage for the human species samples (outliers in terms of MLS), we decided to perform the linear analysis while excluding the human samples, since linear regression is sensitive to high leverage points. The p-values for β 1 were used as the statistical significance of the associations and the signs for the direction of the association, and R 2 as the goodness-of-fit metric. All obtained p-values were adjusted using the Benjamini-Hochberg multiple testing correction. Associations with adjusted p-values < 0.05 and R 2 > 0.3 were considered significant. For all results, please see Supplementary Table S2. The expression activity across pathways was evaluated by using the single-sample gene set enrichment analysis (ssGSEA), also known as the signature projection method, as was first described by Barbie et al. [66]. For pathways data, we used the KEGG pathway database [67]. The signature projection method was applied using the ssgsea function from the gseapy Python package (version 0.9.18) [68][69][70]. After obtaining the enrichment score for each pathway, linear models of the form EnrichmentScore~β 0 + β 1 x SpeciesTrait were fitted and interpreted following the same approach as described for single genes. For the full pathway results, please see Supplementary Table S3. Light GBM Models with SHAP Explanations To investigate the non-linear patterns of gene expression, we applied a two-stage backward selection with LightGBM and SHAP additive explanations. In the first stage, six separate models were trained, each predicting one of the life-history traits (lifespan, body mass, metabolic rate, temperature, gestation period, and mtGC). For each of the models, the other five species life-history variables were not used in the analysis, such that only genes would provide the prediction power for the target. The union of genes selected by all six models was used as an input for the second stage that made the final selection of the MLS associated genes. For both stages, the selection procedure involved applying 5-fold cross-validation (CV) with sorted stratification ten times [71]. Sorted stratification was used for achieving similar distributions of MLS in every fold. On each fold, SHAP values were calculated for each gene. The genes that had non-zero SHAP values across all folds were selected (i.e., considered significant), ensuring therefore that the selected genes are resilient to different ways of sample selection and splitting. For each of those genes, we calculated the Kendall tau-b rank correlation coefficient between their expressions and the SHAP values across all folds, as a measure of the magnitude and direction of the association between a gene's expression and the target variable. It is important to avoid possible bias when the prediction is done solely by the model identifying the species from gene expression. For this reason, data were split into training and validation sets, in such a manner that, on each fold, the validation set contained samples of two species not found in the corresponding training set (unique to every fold). For each model, we repeated the cross-validation selection ten times, each time with a different random seed. Overall, 10 × 5 = 50 non-unique species pairs were used in the validation across different folds. For each of the 5 models of the first stage, we selected the genes which have non-zero SHAP values in at least two out of the ten cross-validation repeats (i.e., at least 5 × 2 = 10 folds). In the second stage, the selection procedure was made more stringent by selecting genes that have non-zero SHAP values in all the crossvalidation iterations/repeats (i.e., 10 × 5 = 50 folds). Through an empirical procedure, the number of top genes to be further characterized was thresholded to 15, corresponding to the elbow of the monotonic graph of SHAP feature importance vs. gene rank (Supplementary Figure S4). All models were hyper-parametrically optimized in a multi-objective study with the Optuna Framework [72], using maximization of R 2 , absolute mean Kendall tau-b rank correlation coefficient (between selected gene expressions and their SHAP values), and minimization of Huber loss as optimization targets. The parameter set with the best R 2 from the Paretto front was selected for the stage I model, and that with the smallest Huber loss for the Stage II model. For the stage II optimization, Huber loss was prioritized over R 2 due to a higher Kendall tau-b, combined with a smaller Huber loss (which might be caused by Huber loss' resistance to outlier predictions resulting in the selection of genes with better Kendall tau-b). A multi-objective tree-structured parzen estimator in Optuna implementation was used to traverse the feature space [73]. The hyper-parametric optimization process was performed independently from the gene selection process. The cross-validation configuration in the processes was similar: 5-fold stratified cross-validation was used in both, except that one fold was excluded in the hyper-parametric optimization (kept as a hold-out and used for evaluating the entire hyperparameter optimization process). Optuna sqlite databases with trials are provided with a source-code repository. To investigate the effects of gene expressions on MLS and other life-history traits, we used 6 separate regression LightGBM models for predicting from gene expression each of the species' features: maximum lifespan, body mass, gestation days, temperature, metabolic rate, and mtDNA GC%. For each of the models, the other five species variables were not used in the analysis, such that only genes would provide the prediction power for the target. Bayesian Networks A Bayesian network encodes the conditional independence structure among a set of variables, in our case among the input genes and MLS. A potential causal relationship between a gene and MLS is identified when the gene belongs to the Market blanket of MLS in the Bayesian network. This is used as a basis for a variable selection approach that can detect causal relationships under strict assumptions. Implementation-wise, as part of the variable selection methodology, SES, a constraint-based variable selection algorithm [43], was used. This algorithm has proved to be appropriate in the context of high dimensional datasets [43], since the selection is not based on the optimization of an objective function (i.e., loss function), so it is not prone to overfitting. In this study, the MXM R package was used. It generates multiple statistically equivalent parent-children sets (i.e., subsets of the target's variable Markov blanket). Two of such sets are said to be statistically equivalent if some of their features can be swapped without affecting the inference or the conclusions [43]. Missing values constitute 2.38% in the expression matrix, resulting in 8116 genes and 358 samples to contain at least one missing value. This effect is due to the following: (1) genes being included in the analysis even if they do not have orthologs in 100% of the included species, resulting in missing values for the species without orthologs (see the Orthology section); (2) genes not being expressed at a detectable level in certain organs and thus being flagged by different technologies as missing. In order not to discard a significant amount of samples or genes, a model agnostic algorithm called missForest [74] was used for imputation. This led to missing data being replaced with their estimated values. The aggregated Out of Bag (OOB) NRMSE was 0.33, comparable to errors achieved in other studies [74]. As with the LightGBM-SHAP models, 10 rounds of 5-fold cross-validation with sorted stratification were used, resulting in 50 pairs of training validation sets. For each pair of sets, the SES algorithm used the imputed training set to find at least one gene signature of MLS. For evaluating the performance of each signature, a LightGBM (1800 trees) [75] was fit to the corresponding non-imputed training set and the RMSE between the predictions made using the non-imputed validation set, and the corresponding MLS real values were computed. The signature with the smallest RMSE was saved. Finally, the frequency of the appearance of each gene across all the identified gene signatures was calculated as a measurement of the association between a gene and MLS. The performance of the signature-selection algorithm was measured using the signature RMSE distribution (median = 12.89 years, RMSE range = 6.59-21.58). See Supplementary Figure S3 for the distribution of RMSEs, saved after each iteration of the algorithm. The fact that some of the signatures from the Bayesian network's selection (see Supplementary Table S4 and Figure S3) have a big RMSE (i.e., >15 years) should be considered with care, as it does not mean that all the genes within those signatures are unimportant for MLS determination. It is usually the case that some test partitions are "harder" to learn than others. Integration of Predicted Genes in the Linear, LightGBM-SHAP, and Bayesian Networks Models Each of the model types (linear, tree-based, and Bayesian networks) resulted in an ordered list of genes, ranked based on various specific metrics. For the tree-based LightGBM model, the metrics were the number of repeats (number of rounds when gene mean absolute SHAP value was non-zero), mean Kendall's tau (correlation between gene SHAP values and expression values), and mean absolute SHAP value (mean absolute SHAP value of each selected gene across all rounds). For the linear models, selected genes were assigned as a score the maximal linear R 2 (maximum R 2 across all organ-specific linear models or zero if the gene had no contribution to any linear model). For the Bayesian networks model, the relative frequency (i.e., frequency of appearance of a gene across all identified gene signatures) was considered. Constructing the composite ranking included joining all the above-mentioned metrics (number of repeats, mean Kendall's tau, mean absolute SHAP value, maximum linear R 2 , and relative frequency), and additionally, GenAge mentions were also accounted for for each selected gene (boolean metric: 0 or 1, depending on whether the gene was reported in the GenAge database) [3]. To form a list of genes considered most predictive by all models along with GenAge mentions, the composite rank was computed, as follows: each selected gene was assigned with 6 different ranks, whereby each rank indicated the rank of a gene in a value space of each of the 6 metrics mentioned above. Ranks were calculated with Pandas rank function with the method's parameter set to dense [76]. Given 6 ranks for each gene, we computed the composite rank of each gene as a sum of its 6 ranks (Supplementary Table S6). An Explanatory Multilevel Linear Model for Composite Integration The linear model used for the explanatory analysis of the final selection of genes was a Bayesian organ-wise multilevel model with random coefficients. A normal likelihood, uninformative normal prior for each coefficient, and a gamma uninformative prior for the precision parameter, were used. The regression parameters were fitted using the Gibbs sampling algorithm implemented in the R Package RJags [77]. The ORQ transformation method [78] was used to do a "transform-both-sides" regression with the imputed dataset since the variables (i.e., gene expressions) and the target variable (i.e., MLS) were nonnormal. The penalized deviance (2e4 iterations) for the multilevel models with the top 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 and 13 genes was 770.5, 661.7, 639.3, 563.4, 556.3, 545.9, 554.6, 523.8, 491.6, 497.2 and 497.9, respectively. Based on this, the threshold was set to the top 11 genes for the multilevel model. For the single-level model, the penalized deviance was 452.4 for that same selection of genes. The Gibbs sampling algorithm was run using 3 chains, 20e3 iterations, and a burnout period of 1e3 iterations. Finally, for explanatory purposes, we included a column in Table 1 to show the genes that were considered to be significantly involved in the multilevel regression for each organ (i.e., only those whose coefficients' probability of being greater or less than zero is greater than the threshold 0.70), except for the all-organs regression wherein we include all 11 genes. Explanatory LightGBM-SHAP Model for Composite Integration Using the composite ranking and an elbow plot to select the threshold (Supplementary Figure S4), the 15 genes with the highest impact for all models were selected. Multiple LightGBM-SHAP models were then fitted with different numbers of top-ranked genes as features, from top-5 to top-15 genes. For those models, we used the same methodology and species partitioning as in the stage II model. The sequential comparison of the accuracy of those models showed that a significant decrease in Huber loss (loss decrease = 0.8) was present during the transition from top-5 to top-6 genes models opposite to a flat decrease after adding more genes to the model (max loss decrease = 0.29). Thus, the top-6 genes model was considered most significant, having the lowest number of genes while having the highest increase in model performance. Concluding Remarks In this work, transcriptomic data from 41 mammalian species were analyzed, using both linear and non-linear organ-specific models. Overall, more than 1800 genes were found to correlate linearly with MLS in at least one of the studied organs. Remarkably, some of these relationships are universal in multiple organs; however, for many other genes, the mechanisms seem to be limited to only some organs. Many of the genes that correlate both with metabolic variables and MLS seem to be expressed in the brain, whereas the liver has the largest number of genes that are associated with MLS independently of other confounders. Pathway enrichment shows that some of the genes found in the analysis are involved in longevity-related biological pathways; however, other pathways, less described or studied so far in relation to aging, also surface and might be of interest. Using the LightGBM-SHAP and Bayesian networks models, we found gene signatures formed by only a few genes that are highly predictive towards MLS. Interestingly, the genes that we found in signatures are not directly related to each other and belong to different pathways. Remarkably, not many genes that correlate with MLS across mammals are known LAGs, though some LAGs identified in interventional studies do overlap with our results. This is somewhat expected as LAGs are usually found experimentally by knockout or overexpression, and do not necessarily impact species MLS through their expression level. Even so, many of the genes found in our analysis, whose expressions potentially determine mammalian MLS, seem to be directly or indirectly involved in longevity-associated processes. Through a combination of linear, non-linear and Bayesian networks, our analysis highlights novel potential longevity regulators in mammals. This approach could have particular significance for predicting new longevity regulators, analyzing the links between determinants of longevity and associated processes, and studying the mechanisms of aging and longevity. Figure 1 . 1Schematic representation of the analysis workflow used in this study. Figure 1 . 1Schematic representation of the analysis workflow used in this study. Figure 2 . 2(a) Top linear correlations for species features and gene expression. The heatmap represents significant statistical associations (FDR < 0.05, R 2 > 0.3) between the expression levels of specific genes with MLS and other life-history traits, in different organs. Included are the 50 genes with the highest maximum coefficient of determination (maximum R 2 across all organs) for predicting MLS (full list of genes associated with lifespan and other species features in Figure 2 . 2(a) Top linear correlations for species features and gene expression. The heatmap represents significant statistical associations (FDR < 0.05, R 2 > 0.3) between the expression levels of specific genes with MLS and other life-history traits, in different organs. Included are the 50 genes with the highest maximum coefficient of determination (maximum R 2 across all organs) for predicting MLS (full list of genes associated with lifespan and other species features in ), improving the metrics for MLS prediction from Huber loss = 3.92, MAE = 4.733, MSE = 64.87 and R 2 = 0.90 in the stage I model to Huber loss = 2.4, MAE = 3.04, MSE = 36.8 and R 2 = 0.95 in the stage II model (outputting 57 genes). Interestingly, several genes included in the stage II model (17 out of 155, slightly enriched, non-significant, p = 0.17) are also orthologous to known LAGs recorded in the GenAge database [3]: GNAS, FXN, TERT, MSRA, XRCC6, UQCRB, MEMO1, NEIL1, RPS8, COX7C, RXRB, EIF4EBP1, RCL1, PCBP2, EIF3K, PKN3, CLHC1. Figure 3 . 3SHAP values and interactions of the top 15 genes with the highest predicted contribution to MLS. (a) SHAP summary plot of the impact values for the topmost predictive genes of MLS, based on their gene expression levels. Each dot represents an individual sample in the model, with the horizontal position showing the impact (in years) on the MLS prediction. Colors show the expression of a particular gene in comparison to its baseline expression level across all samples: from blue (lower) to red (higher). Genes are sorted in decreasing order based on the feature importance (mean of the absolute values of impact). (b) SHAP decision plot. Each sample's MLS prediction is represented by a colored line. At the top of the plot, each line intersects the X-axis at its corresponding predicted MLS. The color of the fragmented line is defined by the position on the X-axis, the predicted MLS (from blue-short-lived species, to red-long-lived species). At the bottom of the plot, the lines will converge at the baseline value (21.6 years), which is an estimation of the average model prediction obtained if all genes used by the stage II model are added. Moving from the bottom of the plot to the top, SHAP values for each feature are added to the model's baseline value. This shows how each gene contributes to the MLS prediction in a layer-wise, propagating manner. (c) The interaction effect (in years) showing stronger or weaker values between pairs of the topmost interacting genes. Each cell depicts the added or subtracted predicted impact (measured in years) that a combination of two genes has, compared to the sum of their individual effects: from blue (lower interaction effect) to red (higher interaction effect). The intensity of the effect (given by the color) does not take into account the direction of the interaction (positive/negative). (d) SHAP heatmap plot with the model's predictions for all samples. Each value on the X-axis represents a sample. Genes are displayed on the Y-axis, with SHAP values for genes in samples being encoded from blue (lower) to red (higher). The model's predictions are shown on the top panel (MLS = f(x)), sorted by the MLS value in descending order. The global importance of the genes is shown as black bars on the right side of the figure. Figure 3 . 3SHAP values and interactions of the top 15 genes with the highest predicted contribution to MLS. (a) SHAP summary plot of the impact values for the topmost predictive genes of MLS, based on their gene expression levels. Each dot represents an individual sample in the model, with the horizontal position showing the impact (in years) on the MLS prediction. Colors show the expression of a particular gene in comparison to its baseline expression level across all samples: from blue (lower) to red (higher). Genes are sorted in decreasing order based on the feature importance (mean of the absolute values of impact). (b) SHAP decision plot. Each sample's MLS prediction is represented by a colored line. At the top of the plot, each line intersects the X-axis at its corresponding predicted MLS. The color of the fragmented line is defined by the position on the X-axis, the predicted MLS (from blue-short-lived species, to red-long-lived species). At the bottom of the plot, the lines will converge at the baseline value (21.6 years), which is an estimation of the average model prediction obtained if all genes used by the stage II model are added. Moving from the bottom of the plot to the top, SHAP values for each feature are added to the model's baseline value. This shows how each gene contributes to the MLS prediction in a layer-wise, propagating manner. (c) The interaction effect (in years) showing stronger or weaker values between pairs of the topmost interacting genes. Each cell depicts the added or subtracted predicted impact (measured in years) that a combination of two genes has, compared to the sum of their individual effects: from blue (lower interaction effect) to red (higher interaction effect). The intensity of the effect (given by the color) does not take into account the direction of the interaction (positive/negative). (d) SHAP heatmap plot with the model's predictions for all samples. Each value on the X-axis represents a sample. Genes are displayed on the Y-axis, with SHAP values for genes in samples being encoded from blue (lower) to red (higher). The model's predictions are shown on the top panel (MLS = f(x)), sorted by the MLS value in descending order. The global importance of the genes is shown as black bars on the right side of the figure. Figure 4 . 4Expression signature of the top six genes in a multi-model analysis. (a) SHAP summary plot of the impact of the six most predictive genes of the MLS, based on their expression levels. Each dot represents an individual sample in the model, where its X-axis position is an impact (in years) on the MLS prediction of the model in the sample. Colors show Figure 4 . 4Expression signature of the top six genes in a multi-model analysis. (a) SHAP summary plot of the impact of the six most predictive genes of the MLS, based on their expression levels. Each dot represents an individual sample in the model, where its X-axis position is an impact (in years) on the MLS prediction of the model in the sample. Colors show the expression of a particular gene in comparison to its baseline expression level across all samples: from blue (lower) to red (higher). The genes are sorted in decreasing order based on the feature importance (average of the absolute value of impacts). (b) SHAP decision plot with interactions. Each sample's MLS prediction is represented by a colored line. At the top of the plot, each line strikes the X-axis at its corresponding predicted MLS value, which also defines the color of the line on a spectrum. Unlike in regular decision plot interactions, effects between pairs of genes are included. At the bottom of the plot, the lines approach the base value of 21.3. Moving from the bottom of the plot to the top, SHAP values for each feature are added to the model's base value of 21.3 years. This shows how each gene contributes to the MLS prediction in a layer-wise propagation manner. (c) Linear correlations for species features and gene expressions of the six most predictive genes. The heatmap represents significant statistical associations (FDR < 0.05, R 2 > 0.3) between the expression levels of the six genes with MLS and other life-history traits, in different organs. Table 1 . 1Goodness of fit statistics for the multilevel Bayesian linear model.Organ Number of Samples RMSE (Years) R 2 MAE Significant Genes for the Linear Regression Brain 132 15.47 0.79 8.72 C1orf56, LRR1, C6orf89, CALCOCO2, CEL, DCTD, DNAJC15, PPP1CA, SPATA20, DPP9 Heart 39 12.43 0.78 7.12 C1orf56, C6orf89, CALCOCO2, CEL, DCTD, DNAJC15, SPATA20, DPP9 Kidney 65 11.88 0.73 8.07 C1orf56, C6orf89, CALCOCO2, CEL, DCTD, NOXA1, PPP1CA, SPATA20 Liver 139 8.92 0.71 4.32 C1orf56, LRR1, C6orf89, CALCOCO2, CEL, DCTD, DNAJC15, NOXA1, PPP1CA, SPATA20, DPP9 Lung 33 21.04 0.73 13.40 C1orf56, CALCOCO2, DCTD, NOXA1, SPATA20, DPP9 All organs 408 15.00 0.69 7.90 NOXA1, CEL, CALCOCO2, C6orf89, PPP1CA, SPATA20, DPP9, DCTD, LRR1, DNAJC15, C1orf56 Acknowledgments:We are very grateful for the help and useful comments/suggestions received from Eliza Martin, Ivan Shcheklein, Dmitry Petrov, Robert Irimia, Laurence Ion, Ioan Valentin Matei and Gabriela Bunu.Conflicts of Interest:The authors declare no conflict of interest. Int. J. Mol. Sci. 2021,22,1073Data Availability Statement:The preprocessing scripts are available at: https://github.com/ antonkulaga/species-notebooks; The bioinformatic WDL pipeline that was used for downloading, quality control and qualification is available at: https://github.com/antonkulaga/rna-seq/tree/ master/pipelines/quantification; Code for linear models analysis for genes and pathways is available at: https://github.com/ursueugen/cross-species-linear-models; Code for LightGBM+Shap analysis, the intersection of the results from multiple models and ranking is available at: https://github.com/ antonkulaga/yspecies; Code for Bayesian networks analysis and multilevel Bayesian linear modeling available at: https://github.com/rodguinea/bayesian_networks_and_bayesian_linear_modeling. Longevity network: Construction and implications. A Budovsky, A Abramovich, R Cohen, V Chalifa-Caspi, V Fraifeld, 10.1016/j.mad.2006.11.018Mech. Age. Dev. 128Budovsky, A.; Abramovich, A.; Cohen, R.; Chalifa-Caspi, V.; Fraifeld, V. Longevity network: Construction and implications. Mech. Age. Dev. 2007, 128, 117-124. [CrossRef] Wide-scale comparative analysis of longevity genes and interventions. H Yanai, A Budovsky, T Barzilay, R Tacutu, V E Fraifeld, 10.1111/acel.12659Aging Cell. 16Yanai, H.; Budovsky, A.; Barzilay, T.; Tacutu, R.; Fraifeld, V.E. Wide-scale comparative analysis of longevity genes and interven- tions. Aging Cell 2017, 16, 1267-1275. [CrossRef] Human ageing genomic resources: New and updated databases. R Tacutu, D Thornton, E Johnson, A Budovsky, D Barardo, T Craig, E Diana, G Lehmann, D Toren, J Wang, 10.1093/nar/gkx1042Nucl. Acids Res. 46Tacutu, R.; Thornton, D.; Johnson, E.; Budovsky, A.; Barardo, D.; Craig, T.; Diana, E.; Lehmann, G.; Toren, D.; Wang, J.; et al. Human ageing genomic resources: New and updated databases. Nucl. Acids Res. 2018, 46, D1083-D1090. [CrossRef] Growth hormone-releasing hormone disruption extends lifespan and regulates response to caloric restriction in mice. L Y Sun, A Spong, W R Swindell, Y Fang, C Hill, J A Huber, J D Boehm, R Westbrook, R Salvatori, A Bartke, 10.7554/eLife.01098eLife 2013, 2, e01098. [CrossRefSun, L.Y.; Spong, A.; Swindell, W.R.; Fang, Y.; Hill, C.; Huber, J.A.; Boehm, J.D.; Westbrook, R.; Salvatori, R.; Bartke, A. Growth hormone-releasing hormone disruption extends lifespan and regulates response to caloric restriction in mice. eLife 2013, 2, e01098. [CrossRef] org: A new, structured and curated database of current therapeutic interventions in aging and age-related disease. A Moskalev, E Chernyagina, J P De Magalhães, D Barardo, H Thoppil, M Shaposhnikov, A Budovsky, V E Fraifeld, A Garazha, V Tsvetkov, 10.18632/aging.100799Aging. 7PubMedMoskalev, A.; Chernyagina, E.; de Magalhães, J.P.; Barardo, D.; Thoppil, H.; Shaposhnikov, M.; Budovsky, A.; Fraifeld, V.E.; Garazha, A.; Tsvetkov, V.; et al. Geroprotectors.org: A new, structured and curated database of current therapeutic interventions in aging and age-related disease. Aging 2015, 7, 616-628. [CrossRef] [PubMed] The DrugAge database of aging-related drugs. D Barardo, D Thornton, H Thoppil, M Walsh, S Sharifi, S Ferreira, A Anžič, M Fernandes, P Monteiro, T Grum, 10.1111/acel.12585Aging Cell. 16PubMedBarardo, D.; Thornton, D.; Thoppil, H.; Walsh, M.; Sharifi, S.; Ferreira, S.; Anžič, A.; Fernandes, M.; Monteiro, P.; Grum, T.; et al. The DrugAge database of aging-related drugs. Aging Cell 2017, 16, 594-597. [CrossRef] [PubMed] Insights into the evolution of longevity from the bowhead whale genome. M Keane, J Semeiks, A E Webb, Y I Li, V Quesada, T Craig, L B Madsen, S Van Dam, D Brawand, P I Marques, 10.1016/j.celrep.2014.12.008Cell Rep. 10PubMedKeane, M.; Semeiks, J.; Webb, A.E.; Li, Y.I.; Quesada, V.; Craig, T.; Madsen, L.B.; van Dam, S.; Brawand, D.; Marques, P.I.; et al. Insights into the evolution of longevity from the bowhead whale genome. Cell Rep. 2015, 10, 112-122. [CrossRef] [PubMed] Comparative genetics of longevity and cancer: Insights from long-lived rodents. V Gorbunova, A Seluanov, Z Zhang, V N Gladyshev, J Vijg, 10.1038/nrg3728Nat. Rev. Genet. 15PubMedGorbunova, V.; Seluanov, A.; Zhang, Z.; Gladyshev, V.N.; Vijg, J. Comparative genetics of longevity and cancer: Insights from long-lived rodents. Nat. Rev. Genet. 2014, 15, 531-540. [CrossRef] [PubMed] Gene expression defines natural changes in mammalian lifespan. A A Fushan, A A Turanov, S.-G Lee, E B Kim, A V Lobanov, S H Yim, R Buffenstein, S.-R Lee, K.-T Chang, H Rhee, 10.1111/acel.12283Aging Cell. 14Fushan, A.A.; Turanov, A.A.; Lee, S.-G.; Kim, E.B.; Lobanov, A.V.; Yim, S.H.; Buffenstein, R.; Lee, S.-R.; Chang, K.-T.; Rhee, H.; et al. Gene expression defines natural changes in mammalian lifespan. Aging Cell 2015, 14, 352-365. [CrossRef] Molecular signatures of longevity: Insights from cross-species comparative studies. S Ma, V N Gladyshev, 10.1016/j.semcdb.2017.08.007Semin. Cell Dev. Biol. 70Ma, S.; Gladyshev, V.N. Molecular signatures of longevity: Insights from cross-species comparative studies. Semin. Cell Dev. Biol. 2017, 70, 190-203. [CrossRef] Gray whale transcriptome reveals longevity adaptations associated with DNA repair and ubiquitination. D Toren, A Kulaga, M Jethva, E Rubin, A V Snezhkina, A V Kudryavtseva, D Nowicki, R Tacutu, A A Moskalev, V E Fraifeld, 10.1111/acel.13158Aging Cell. 19PubMedToren, D.; Kulaga, A.; Jethva, M.; Rubin, E.; Snezhkina, A.V.; Kudryavtseva, A.V.; Nowicki, D.; Tacutu, R.; Moskalev, A.A.; Fraifeld, V.E. Gray whale transcriptome reveals longevity adaptations associated with DNA repair and ubiquitination. Aging Cell 2020, 19, e13158. [CrossRef] [PubMed] Cell culture-based profiling across mammals reveals DNA repair and metabolism as determinants of species longevity. S Ma, A Upneja, A Galecki, Y.-M Tsai, C F Burant, S Raskind, Q Zhang, Z D Zhang, A Seluanov, V Gorbunova, 10.7554/eLife.19130eLife 2016, 5, e19130. [CrossRef. PubMedMa, S.; Upneja, A.; Galecki, A.; Tsai, Y.-M.; Burant, C.F.; Raskind, S.; Zhang, Q.; Zhang, Z.D.; Seluanov, A.; Gorbunova, V.; et al. Cell culture-based profiling across mammals reveals DNA repair and metabolism as determinants of species longevity. eLife 2016, 5, e19130. [CrossRef] [PubMed] Longitudinal comparative transcriptomics reveals unique mechanisms underlying extended healthspan in bats. Z Huang, C V Whelan, N M Foley, D Jebb, F Touzalin, E J Petit, S J Puechmaille, E C Teeling, 10.1038/s41559-019-0913-3Nat. Ecol. Evol. 3PubMedHuang, Z.; Whelan, C.V.; Foley, N.M.; Jebb, D.; Touzalin, F.; Petit, E.J.; Puechmaille, S.J.; Teeling, E.C. Longitudinal comparative transcriptomics reveals unique mechanisms underlying extended healthspan in bats. Nat. Ecol. Evol. 2019, 3, 1110-1120. [CrossRef] [PubMed] Single-cell transcriptomics of the naked mole-rat reveals unexpected features of mammalian immunity. H G Hilton, N D Rubinstein, P Janki, A T Ireland, N Bernstein, N L Fong, K M Wright, M Smith, D Finkle, B Martin-Mcnulty, 10.1371/journal.pbio.3000528PLoS Biol. 17e3000528. [CrossRef. PubMedHilton, H.G.; Rubinstein, N.D.; Janki, P.; Ireland, A.T.; Bernstein, N.; Fong, N.L.; Wright, K.M.; Smith, M.; Finkle, D.; Martin-McNulty, B.; et al. Single-cell transcriptomics of the naked mole-rat reveals unexpected features of mammalian immunity. PLoS Biol. 2019, 17, e3000528. [CrossRef] [PubMed] Pair-wise linear and 3D nonlinear relationships between the liver antioxidant enzyme activities and the rate of body oxygen consumption in mice. Free Radic. K K Muradian, N A Utko, T G Mozzhukhina, A Y Litoshenko, I N Pishel, V V Bezrukov, V E Fraifield, 10.1016/S0891-5849(02)01196-6Biol. Med. 33Muradian, K.K.; Utko, N.A.; Mozzhukhina, T.G.; Litoshenko, A.Y.; Pishel, I.N.; Bezrukov, V.V.; Fraifield, V.E. Pair-wise linear and 3D nonlinear relationships between the liver antioxidant enzyme activities and the rate of body oxygen consumption in mice. Free Radic. Biol. Med. 2002, 33, 1736-1739. [CrossRef] Do mitochondrial DNA and metabolic rate complement each other in determination of the mammalian maximum longevity?. G Lehmann, E Segal, K K Muradian, V E Fraifeld, 10.1089/rej.2008.0676Rejuvenat. Res. 11PubMedLehmann, G.; Segal, E.; Muradian, K.K.; Fraifeld, V.E. Do mitochondrial DNA and metabolic rate complement each other in determination of the mammalian maximum longevity? Rejuvenat. Res. 2008, 11, 409-417. [CrossRef] [PubMed] Telomere length and body temperature-independent determinants of mammalian longevity? Front. G Lehmann, K K Muradian, V E Fraifeld, 10.3389/fgene.2013.00111Genet. 2013, 4, 111. [CrossRefLehmann, G.; Muradian, K.K.; Fraifeld, V.E. Telomere length and body temperature-independent determinants of mammalian longevity? Front. Genet. 2013, 4, 111. [CrossRef] Molecular links between cellular senescence, longevity and age-related diseases-A systems biology perspective. R Tacutu, A Budovsky, H Yanai, V E Fraifeld, 10.18632/aging.100413Aging. 3Tacutu, R.; Budovsky, A.; Yanai, H.; Fraifeld, V.E. Molecular links between cellular senescence, longevity and age-related diseases-A systems biology perspective. Aging 2011, 3, 1178-1191. [CrossRef] The signaling hubs at the crossroad of longevity and age-related disease networks. M Wolfson, A Budovsky, R Tacutu, V Fraifeld, 10.1016/j.biocel.2008.08.026Int. J. Biochem. Cell Biol. 41Wolfson, M.; Budovsky, A.; Tacutu, R.; Fraifeld, V. The signaling hubs at the crossroad of longevity and age-related disease networks. Int. J. Biochem. Cell Biol. 2009, 41, 516-520. [CrossRef] Genome sequencing reveals insights into physiology and longevity of the naked mole rat. E B Kim, X Fang, A A Fushan, Z Huang, A V Lobanov, L Han, S M Marino, X Sun, A A Turanov, P Yang, 10.1038/nature10533Nature. 479Kim, E.B.; Fang, X.; Fushan, A.A.; Huang, Z.; Lobanov, A.V.; Han, L.; Marino, S.M.; Sun, X.; Turanov, A.A.; Yang, P.; et al. Genome sequencing reveals insights into physiology and longevity of the naked mole rat. Nature 2011, 479, 223-227. [CrossRef] Metabolism and longevity: Is there a role for membrane fatty acids?. A J Hulbert, 10.1093/icb/icq007Integr. Comp. Biol. 50Hulbert, A.J. Metabolism and longevity: Is there a role for membrane fatty acids? Integr. Comp. Biol. 2010, 50, 808-817. [CrossRef] Maximum life span in vertebrates: Relationship with liver antioxidant enzymes, glutathione system, ascorbate, urate, sensitivity to peroxidation, true malondialdehyde, in vivo H2O2, and basal and maximum aerobic capacity. M Lopez-Torres, R Perez-Campo, C Rojas, S Cadenas, G Barja, 10.1016/0047-6374(93)90047-UMech. Ageing Dev. 70Lopez-Torres, M.; Perez-Campo, R.; Rojas, C.; Cadenas, S.; Barja, G. Maximum life span in vertebrates: Relationship with liver antioxidant enzymes, glutathione system, ascorbate, urate, sensitivity to peroxidation, true malondialdehyde, in vivo H2O2, and basal and maximum aerobic capacity. Mech. Ageing Dev. 1993, 70, 177-199. [CrossRef] Lipidome determinants of maximal lifespan in mammals. K Bozek, E E Khrameeva, J Reznick, D Omerbašić, N C Bennett, G R Lewin, J Azpurua, V Gorbunova, A Seluanov, P Regnard, 10.1038/s41598-017-00037-7Sci. Rep. 2017, 7, 5. [CrossRef. PubMedBozek, K.; Khrameeva, E.E.; Reznick, J.; Omerbašić, D.; Bennett, N.C.; Lewin, G.R.; Azpurua, J.; Gorbunova, V.; Seluanov, A.; Regnard, P.; et al. Lipidome determinants of maximal lifespan in mammals. Sci. Rep. 2017, 7, 5. [CrossRef] [PubMed] Molecular mechanism of PPAR in the regulation of age-related inflammation. J H Chung, A Y Seo, S W Chung, M K Kim, C Leeuwenburgh, B P Yu, H Y Chung, 10.1016/j.arr.2008.01.001Age. Res. Rev. 7PubMedChung, J.H.; Seo, A.Y.; Chung, S.W.; Kim, M.K.; Leeuwenburgh, C.; Yu, B.P.; Chung, H.Y. Molecular mechanism of PPAR in the regulation of age-related inflammation. Age. Res. Rev. 2008, 7, 126-136. [CrossRef] [PubMed] Age-associated changes in long-chain fatty acid profile during healthy aging promote pro-inflammatory monocyte polarization via PPARγ. C Pararasa, J Ikwuobe, S Shigdar, A Boukouvalas, I T Nabney, J E Brown, A Devitt, C J Bailey, S J Bennett, H R Griffiths, 10.1111/acel.12416Aging Cell. 15Pararasa, C.; Ikwuobe, J.; Shigdar, S.; Boukouvalas, A.; Nabney, I.T.; Brown, J.E.; Devitt, A.; Bailey, C.J.; Bennett, S.J.; Griffiths, H.R. Age-associated changes in long-chain fatty acid profile during healthy aging promote pro-inflammatory monocyte polarization via PPARγ. Aging Cell 2016, 15, 128-139. [CrossRef] . Int. J. Mol. Sci. 22Int. J. Mol. Sci. 2021, 22, 1073 20 of 21 A strain-independent postnatal neurodegeneration in mice lacking the EGF receptor. M Sibilia, J P Steinbach, L Stingl, A Aguzzi, E F Wagner, 10.1093/emboj/17.3.719EMBO J. 17Sibilia, M.; Steinbach, J.P.; Stingl, L.; Aguzzi, A.; Wagner, E.F. A strain-independent postnatal neurodegeneration in mice lacking the EGF receptor. EMBO J. 1998, 17, 719-731. [CrossRef] LightGBM: A highly efficient gradient boosting decision tree. G Ke, Q Meng, T Finley, T Wang, W Chen, W Ma, Q Ye, T.-Y Liu, Proceedings of the Advances in Neural Information Processing Systems. the Advances in Neural Information Processing SystemsLong Beach, CA, USAKe, G.; Meng, Q.; Finley, T.; Wang, T.; Chen, W.; Ma, W.; Ye, Q.; Liu, T.-Y. LightGBM: A highly efficient gradient boosting decision tree. In Proceedings of the Advances in Neural Information Processing Systems, Long Beach, CA, USA, 4-9 December 2017. A Unified Approach to Interpreting Model Predictions. S M Lundberg, S.-I Lee, Proceedings of the Advances in Neural Information Processing Systems. the Advances in Neural Information Processing SystemsLong Beach, CA, USALundberg, S.M.; Lee, S.-I. A Unified Approach to Interpreting Model Predictions. In Proceedings of the Advances in Neural Information Processing Systems, Long Beach, CA, USA, 4-9 December 2017. Life-history connections to rates of aging in terrestrial vertebrates. R E Ricklefs, 10.1073/pnas.1005862107Proc. Natl. Acad. Sci. Natl. Acad. SciUSA107Ricklefs, R.E. Life-history connections to rates of aging in terrestrial vertebrates. Proc. Natl. Acad. Sci. USA 2010, 107, 10314-10319. [CrossRef] Methylomic predictors demonstrate the role of NF-κB in old-age mortality and are unrelated to the aging-associated epigenetic drift. J Jylhävä, L Kananen, J Raitanen, S Marttila, T Nevalainen, A Hervonen, M Jylhä, M Hurme, 10.18632/oncotarget.8278Oncotarget. 7Jylhävä, J.; Kananen, L.; Raitanen, J.; Marttila, S.; Nevalainen, T.; Hervonen, A.; Jylhä, M.; Hurme, M. Methylomic predictors demonstrate the role of NF-κB in old-age mortality and are unrelated to the aging-associated epigenetic drift. Oncotarget 2016, 7, 19228-19241. [CrossRef] Nuclear lamina defects cause ATM-dependent NF-κB activation and link accelerated aging to a systemic inflammatory response. F G Osorio, C Bárcena, C Soria-Valles, A J Ramsay, F De Carlos, J Cobo, A Fueyo, J M P Freije, C López-Otín, 10.1101/gad.197954.112Genes Dev. 26Osorio, F.G.; Bárcena, C.; Soria-Valles, C.; Ramsay, A.J.; de Carlos, F.; Cobo, J.; Fueyo, A.; Freije, J.M.P.; López-Otín, C. Nuclear lamina defects cause ATM-dependent NF-κB activation and link accelerated aging to a systemic inflammatory response. Genes Dev. 2012, 26, 2311-2324. [CrossRef] Cross-talk between mitochondrial fusion and the hippo pathway in controlling cell proliferation during drosophila development. Q Deng, T Guo, X Zhou, Y Xi, X Yang, W Ge, 10.1534/genetics.115.186445Genetics. 203Deng, Q.; Guo, T.; Zhou, X.; Xi, Y.; Yang, X.; Ge, W. Cross-talk between mitochondrial fusion and the hippo pathway in controlling cell proliferation during drosophila development. Genetics 2016, 203, 1777-1788. [CrossRef] Increased mitochondrial fusion allows the survival of older animals in diverse C. elegans longevity pathways. S N Chaudhari, E Kipreos, 10.1038/s41467-017-00274-4Nat. Commun. 8182PubMedChaudhari, S.N.; Kipreos, E.T. Increased mitochondrial fusion allows the survival of older animals in diverse C. elegans longevity pathways. Nat. Commun. 2017, 8, 182. [CrossRef] [PubMed] Meiosis-specific cohesin component, Stag3 is essential for maintaining centromere chromatid cohesion, and required for DNA repair and synapsis between homologous chromosomes. J Hopkins, G Hwang, J Jacob, N Sapp, R Bedigian, K Oka, P Overbeek, S Murray, P W Jordan, 10.1371/journal.pgen.1004413PLoS Genet. 10PubMedHopkins, J.; Hwang, G.; Jacob, J.; Sapp, N.; Bedigian, R.; Oka, K.; Overbeek, P.; Murray, S.; Jordan, P.W. Meiosis-specific cohesin component, Stag3 is essential for maintaining centromere chromatid cohesion, and required for DNA repair and synapsis between homologous chromosomes. PLoS Genet. 2014, 10, e1004413. [CrossRef] [PubMed] Splice variants of the dual specificity tyrosine phosphorylation-regulated kinase 4 (DYRK4) differ in their subcellular localization and catalytic activity. C Papadopoulos, K Arato, E Lilienthal, J Zerweck, M Schutkowski, N Chatain, G Müller-Newen, W Becker, S De La Luna, 10.1074/jbc.M110.157909J. Biol. Chem. 286PubMedPapadopoulos, C.; Arato, K.; Lilienthal, E.; Zerweck, J.; Schutkowski, M.; Chatain, N.; Müller-Newen, G.; Becker, W.; de la Luna, S. Splice variants of the dual specificity tyrosine phosphorylation-regulated kinase 4 (DYRK4) differ in their subcellular localization and catalytic activity. J. Biol. Chem. 2011, 286, 5494-5505. [CrossRef] [PubMed] Selective DYRK1A inhibitor for the treatment of neurodegenerative diseases: Alzheimer, parkinson, huntington, and down syndrome. R B Kargbo, 10.1021/acsmedchemlett.0c00346ACS Med. Chem. Lett. 11Kargbo, R.B. Selective DYRK1A inhibitor for the treatment of neurodegenerative diseases: Alzheimer, parkinson, huntington, and down syndrome. ACS Med. Chem. Lett. 2020, 11, 1795-1796. [CrossRef] Bombesin Receptor-Activated Protein (BRAP) Modulates NF-κB Activation in Bronchial Epithelial Cells by Enhancing HDAC Activity. Y Liu, X.-Q Qin, H C Weber, Y Xiang, C Liu, H.-J Liu, H Yang, J Jiang, X Qu, 10.1002/jcb.25406J. Cell. Biochem. 117Liu, Y.; Qin, X.-Q.; Weber, H.C.; Xiang, Y.; Liu, C.; Liu, H.-J.; Yang, H.; Jiang, J.; Qu, X. Bombesin Receptor-Activated Protein (BRAP) Modulates NF-κB Activation in Bronchial Epithelial Cells by Enhancing HDAC Activity. J. Cell. Biochem. 2016, 117, 1069-1077. [CrossRef] Genetics vs. entropy: Longevity factors suppress the NF-kappaB-driven entropic aging process. A Salminen, K Kaarniranta, 10.1016/j.arr.2009.11.001Age. Res. Rev. 9Salminen, A.; Kaarniranta, K. Genetics vs. entropy: Longevity factors suppress the NF-kappaB-driven entropic aging process. Age. Res. Rev. 2010, 9, 298-314. [CrossRef] The PPP1CA Gene and Its Putative Association with Human Ageing. 11The PPP1CA Gene and Its Putative Association with Human Ageing. Available online: https://genomics.senescence.info/genes/ entry.php?hgnc=PPP1CA (accessed on 11 December 2020). SynergyAge: A curated database for synergistic and antagonistic interactions of longevity-associated genes. G Bunu, D Toren, C F Ion, D Barardo, L Sarghie, L G Grigore, J P De Magalhaes, V Fraifeld, R Tacutu, 10.1038/s41597-020-00710-zSci. Data 2020, 7, 366. [CrossRefBunu, G.; Toren, D.; Ion, C.F.; Barardo, D.; Sarghie, L.; Grigore, L.G.; de Magalhaes, J.P.; Fraifeld, V.; Tacutu, R. SynergyAge: A curated database for synergistic and antagonistic interactions of longevity-associated genes. Sci. Data 2020, 7, 366. [CrossRef] The antioxidant effects of ribonuclease inhibitor. X Y Cui, P F Fu, D N Pan, Y Zhao, J Zhao, B C Zhao, 10.1080/10715760310001600408Free Radic. Res. 37Cui, X.Y.; Fu, P.F.; Pan, D.N.; Zhao, Y.; Zhao, J.; Zhao, B.C. The antioxidant effects of ribonuclease inhibitor. Free Radic. Res. 2003, 37, 1079-1085. [CrossRef] Practical approaches to causal relationship exploration. J Li, L Liu, T D Le, In Springerbriefs in Electrical and Computer Engineering. Li, J.; Liu, L.; Le, T.D. Practical approaches to causal relationship exploration. In Springerbriefs in Electrical and Computer Engineering; Feature Selection with the R PackageMXM: Discovering Statistically Equivalent Feature Subsets. V Lagani, G Athineou, A Farcomeni, M Tsagris, I Tsamardinos, 10.18637/jss.v080.i07J. Stat. Softw. 80Lagani, V.; Athineou, G.; Farcomeni, A.; Tsagris, M.; Tsamardinos, I. Feature Selection with the R PackageMXM: Discovering Sta- tistically Equivalent Feature Subsets. J. Stat. Softw. 2017, 80. [CrossRef] Unable to find information for 10168452. Unable to find information for 10168452. P Spirtes, C Glymour, Scheines, R. Causation, Prediction, and Search. Cambridge, MA, USAMIT Press97802625279272nd ed.Spirtes, P.; Glymour, C.; Scheines, R. Causation, Prediction, and Search, 2nd ed.; MIT Press: Cambridge, MA, USA, 2001; ISBN 9780262527927. Overexpression of cytosolic sialidase Neu2 induces myoblast differentiation in C2C12 cells. A Fanzani, R Giuliani, F Colombo, D Zizioli, M Presta, A Preti, S Marchesini, 10.1016/S0014-5793(03)00709-9FEBS Lett. 547Fanzani, A.; Giuliani, R.; Colombo, F.; Zizioli, D.; Presta, M.; Preti, A.; Marchesini, S. Overexpression of cytosolic sialidase Neu2 induces myoblast differentiation in C2C12 cells. FEBS Lett. 2003, 547, 183-188. [CrossRef] Effect of aging on mitochondrial energetics in the human atria. L Emelyanova, C Preston, A Gupta, M Viqar, U Negmadjanov, S Edwards, K Kraft, K Devana, E Holmuhamedov, D O&apos;hair, 10.1093/gerona/glx160J. Gerontol. A Biol. Sci. Med. Sci. 73PubMedEmelyanova, L.; Preston, C.; Gupta, A.; Viqar, M.; Negmadjanov, U.; Edwards, S.; Kraft, K.; Devana, K.; Holmuhamedov, E.; O'Hair, D.; et al. Effect of aging on mitochondrial energetics in the human atria. J. Gerontol. A Biol. Sci. Med. Sci. 2018, 73, 608-616. [CrossRef] [PubMed] Bi-allelic mutations in NDUFA6 establish its role in early-onset isolated mitochondrial complex I deficiency. C L Alston, J Heidler, M G Dibley, L S Kremer, L S Taylor, C Fratter, C E French, R I C Glasgow, R G Feichtinger, I Delon, 10.1016/j.ajhg.2018.08.013Am. J. Hum. Genet. 103PubMedAlston, C.L.; Heidler, J.; Dibley, M.G.; Kremer, L.S.; Taylor, L.S.; Fratter, C.; French, C.E.; Glasgow, R.I.C.; Feichtinger, R.G.; Delon, I.; et al. Bi-allelic mutations in NDUFA6 establish its role in early-onset isolated mitochondrial complex I deficiency. Am. J. Hum. Genet. 2018, 103, 592-601. [CrossRef] [PubMed] Low abundance of the matrix arm of complex I in mitochondria predicts longevity in mice. S Miwa, H Jow, K Baty, A Johnson, R Czapiewski, G Saretzki, A Treumann, T Von Zglinicki, 10.1038/ncomms4837Nat. Commun. 5PubMedMiwa, S.; Jow, H.; Baty, K.; Johnson, A.; Czapiewski, R.; Saretzki, G.; Treumann, A.; von Zglinicki, T. Low abundance of the matrix arm of complex I in mitochondria predicts longevity in mice. Nat. Commun. 2014, 5, 3837. [CrossRef] [PubMed] Rbm46 regulates mouse embryonic stem cell differentiation by targeting β-Catenin mRNA for degradation. L Zhai, C Wang, Y Chen, S Zhou, L Li, 10.1371/journal.pone.0172420PLoS ONE. 12Zhai, L.; Wang, C.; Chen, Y.; Zhou, S.; Li, L. Rbm46 regulates mouse embryonic stem cell differentiation by targeting β-Catenin mRNA for degradation. PLoS ONE 2017, 12, e0172420. [CrossRef] Polyunsaturated fatty acids modulate the association between PIK3CA-KCNMB3 genetic variants and insulin resistance. J.-S Zheng, D K Arnett, L D Parnell, Y.-C Lee, Y Ma, C E Smith, K Richardson, D Li, I B Borecki, K L Tucker, 10.1371/journal.pone.0067394PLoS ONE. 8e67394. [CrossRefZheng, J.-S.; Arnett, D.K.; Parnell, L.D.; Lee, Y.-C.; Ma, Y.; Smith, C.E.; Richardson, K.; Li, D.; Borecki, I.B.; Tucker, K.L.; et al. Polyunsaturated fatty acids modulate the association between PIK3CA-KCNMB3 genetic variants and insulin resistance. PLoS ONE 2013, 8, e67394. [CrossRef] . Int. J. Mol. Sci. 22Int. J. Mol. Sci. 2021, 22, 1073 21 of 21 Mitophagy: An emerging role in aging and age-associated diseases. G Chen, G Kroemer, O Kepp, 10.3389/fcell.2020.00200Front. Cell Dev. Biol. Chen, G.; Kroemer, G.; Kepp, O. Mitophagy: An emerging role in aging and age-associated diseases. Front. Cell Dev. Biol. 2020, 8, 200. [CrossRef] Atlas on substrate recognition subunits of CRL2 E3 ligases. S Wang, W Xia, M Qiu, X Wang, F Jiang, R Yin, L Xu, 10.18632/oncotarget.8732Oncotarget. 7Wang, S.; Xia, W.; Qiu, M.; Wang, X.; Jiang, F.; Yin, R.; Xu, L. Atlas on substrate recognition subunits of CRL2 E3 ligases. Oncotarget 2016, 7, 46707-46716. [CrossRef] LRRpredictor-A New LRR Motif Detection Method for Irregular Motifs of Plant NLR Proteins Using an Ensemble of Classifiers. E C Martin, O C A Sukarta, L Spiridon, L G Grigore, V Constantinescu, R Tacutu, A Goverse, A.-J Petrescu, 10.3390/genes11030286Genes. 2020Martin, E.C.; Sukarta, O.C.A.; Spiridon, L.; Grigore, L.G.; Constantinescu, V.; Tacutu, R.; Goverse, A.; Petrescu, A.-J. LRRpredictor-A New LRR Motif Detection Method for Irregular Motifs of Plant NLR Proteins Using an Ensemble of Classifiers. Genes 2020, 11, 286. [CrossRef] FIGNL1-containing protein complex is required for efficient homologous recombination repair. J Yuan, J Chen, 10.1073/pnas.1220662110Proc. Natl. Acad. Sci. Natl. Acad. SciUSA110Yuan, J.; Chen, J. FIGNL1-containing protein complex is required for efficient homologous recombination repair. Proc. Natl. Acad. Sci. USA 2013, 110, 10640-10645. [CrossRef] Histamime receptor H4 as a new therapeutic target for age-related macular degeneration. H Kaneko, Nippon Ganka Gakkai Zasshi. 120Kaneko, H. Histamime receptor H4 as a new therapeutic target for age-related macular degeneration. Nippon Ganka Gakkai Zasshi 2016, 120, 747-753. Improving supervised prediction of aging-related genes via dynamic network analysis. Q Li, T Milenkovic, arXiv:2005.036592020Li, Q.; Milenkovic, T. Improving supervised prediction of aging-related genes via dynamic network analysis. arXiv 2020, arXiv:2005.03659. Key modulators in aging and age-related cardiovascular diseases?. S Sahoo, D N Meijles, P J Pagano, Oxidases, 10.1042/CS20150087Clin. Sci. 130PubMedSahoo, S.; Meijles, D.N.; Pagano, P.J. NADPH oxidases: Key modulators in aging and age-related cardiovascular diseases? Clin. Sci. 2016, 130, 317-335. [CrossRef] [PubMed] C6orf89 encodes three distinct HDAC enhancers that function in the nucleolus, the golgi and the midbody. V S Lalioti, S Vergarajauregui, A Villasante, D Pulido, I V Sandoval, 10.1002/jcp.24355J. Cell. Physiol. 228PubMedLalioti, V.S.; Vergarajauregui, S.; Villasante, A.; Pulido, D.; Sandoval, I.V. C6orf89 encodes three distinct HDAC enhancers that function in the nucleolus, the golgi and the midbody. J. Cell. Physiol. 2013, 228, 1907-1921. [CrossRef] [PubMed] Proteasomal degradation induced by DPP9-mediated processing competes with mitochondrial protein import. Y Finger, M Habich, S Gerlich, S Urbanczyk, E Van De Logt, J Koch, L Schu, K J Lapacz, M Ali, C Petrungaro, 10.15252/embj.2019103889EMBO J. 2020, 39, e103889. [CrossRefFinger, Y.; Habich, M.; Gerlich, S.; Urbanczyk, S.; van de Logt, E.; Koch, J.; Schu, L.; Lapacz, K.J.; Ali, M.; Petrungaro, C.; et al. Proteasomal degradation induced by DPP9-mediated processing competes with mitochondrial protein import. EMBO J. 2020, 39, e103889. [CrossRef] J Herrero, M Muffato, K Beal, S Fitzgerald, L Gordon, M Pignatelli, A J Vilella, S M J Searle, R Amode, S Brent, 10.1093/database/bav096Ensembl comparative genomics resources. bav096. [CrossRefHerrero, J.; Muffato, M.; Beal, K.; Fitzgerald, S.; Gordon, L.; Pignatelli, M.; Vilella, A.J.; Searle, S.M.J.; Amode, R.; Brent, S.; et al. Ensembl comparative genomics resources. Database 2016, 2016, bav096. [CrossRef] A database for comparative analysis of mitochondrial DNA, with a special focus on animal longevity. D Toren, T Barzilay, R Tacutu, G Lehmann, K K Muradian, V E Fraifeld, Mitoage, 10.1093/nar/gkv1187Nucl. Acids Res. 44Toren, D.; Barzilay, T.; Tacutu, R.; Lehmann, G.; Muradian, K.K.; Fraifeld, V.E. MitoAge: A database for comparative analysis of mitochondrial DNA, with a special focus on animal longevity. Nucl. Acids Res. 2016, 44, D1262-D1265. [CrossRef] An ultra-fast all-in-one FASTQ preprocessor. S Chen, Y Zhou, Y Chen, J Gu, Fastp, 10.1093/bioinformatics/bty560Bioinformatics. 34Chen, S.; Zhou, Y.; Chen, Y.; Gu, J. fastp: An ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 2018, 34, i884-i890. [CrossRef] Salmon provides fast and bias-aware quantification of transcript expression. R Patro, G Duggal, M I Love, R A Irizarry, C Kingsford, 10.1038/nmeth.4197Nat. Methods. 14Patro, R.; Duggal, G.; Love, M.I.; Irizarry, R.A.; Kingsford, C. Salmon provides fast and bias-aware quantification of transcript expression. Nat. Methods 2017, 14, 417-419. [CrossRef] Econometric and statistical modeling with python. S Seabold, J Perktold, Statsmodels, Proceedings of the 9th Python in Science Conference. the 9th Python in Science ConferenceAustin, TX, USASeabold, S.; Perktold, J. Statsmodels: Econometric and statistical modeling with python. In Proceedings of the 9th Python in Science Conference, Austin, TX, USA, 28 June-3 July 2010; pp. 92-96. Systematic RNA interference reveals that oncogenic KRAS-driven cancers require TBK1. D A Barbie, P Tamayo, J S Boehm, S Y Kim, S E Moody, I F Dunn, A C Schinzel, P Sandy, E Meylan, C Scholl, 10.1038/nature08460Nature. 462Barbie, D.A.; Tamayo, P.; Boehm, J.S.; Kim, S.Y.; Moody, S.E.; Dunn, I.F.; Schinzel, A.C.; Sandy, P.; Meylan, E.; Scholl, C.; et al. Systematic RNA interference reveals that oncogenic KRAS-driven cancers require TBK1. Nature 2009, 462, 108-112. [CrossRef] KEGG for integration and interpretation of large-scale molecular data sets. M Kanehisa, S Goto, Y Sato, M Furumichi, M Tanabe, 10.1093/nar/gkr988Nucl. Acids Res. 40Kanehisa, M.; Goto, S.; Sato, Y.; Furumichi, M.; Tanabe, M. KEGG for integration and interpretation of large-scale molecular data sets. Nucl. Acids Res. 2012, 40, D109-D114. [CrossRef] Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. A Subramanian, P Tamayo, V K Mootha, S Mukherjee, B L Ebert, M A Gillette, A Paulovich, S L Pomeroy, T R Golub, E S Lander, 10.1073/pnas.0506580102Proc. Natl. Acad. Sci. Natl. Acad. SciUSA102PubMedSubramanian, A.; Tamayo, P.; Mootha, V.K.; Mukherjee, S.; Ebert, B.L.; Gillette, M.A.; Paulovich, A.; Pomeroy, S.L.; Golub, T.R.; Lander, E.S.; et al. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA 2005, 102, 15545-15550. [CrossRef] [PubMed] Enrichr: Interactive and collaborative HTML5 gene list enrichment analysis tool. E Y Chen, C M Tan, Y Kou, Q Duan, Z Wang, G V Meirelles, N R Clark, A Ma&apos;ayan, 10.1186/1471-2105-14-128BMC Bioinformatics. 14PubMedChen, E.Y.; Tan, C.M.; Kou, Y.; Duan, Q.; Wang, Z.; Meirelles, G.V.; Clark, N.R.; Ma'ayan, A. Enrichr: Interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinformatics 2013, 14, 128. [CrossRef] [PubMed] Enrichr: A comprehensive gene set enrichment analysis web server 2016 update. M V Kuleshov, M R Jones, A D Rouillard, N F Fernandez, Q Duan, Z Wang, S Koplev, S L Jenkins, K M Jagodnik, A Lachmann, 10.1093/nar/gkw377Nucl. Acids Res. 44PubMedKuleshov, M.V.; Jones, M.R.; Rouillard, A.D.; Fernandez, N.F.; Duan, Q.; Wang, Z.; Koplev, S.; Jenkins, S.L.; Jagodnik, K.M.; Lachmann, A.; et al. Enrichr: A comprehensive gene set enrichment analysis web server 2016 update. Nucl. Acids Res. 2016, 44, W90-W97. [CrossRef] [PubMed] Stratified Validation Splits for Regression Problems. S C Lowe, 11Lowe, S.C. Stratified Validation Splits for Regression Problems. Available online: https://scottclowe.com/2016-03-19-stratified- regression-partitions/ (accessed on 11 December 2020). Optuna: A next-generation hyperparameter optimization framework. T Akiba, S Sano, T Yanase, T Ohta, M Koyama, Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining-KDD '19. the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining-KDD '19Anchorage, AK, USA; New York, NY, USAACM PressAkiba, T.; Sano, S.; Yanase, T.; Ohta, T.; Koyama, M. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining-KDD '19, Anchorage, AK, USA, 3-7 August 2019; ACM Press: New York, NY, USA, 2019; pp. 2623-2631. Multiobjective tree-structured parzen estimator for computationally expensive optimization problems. Y Ozaki, Y Tanigaki, S Watanabe, M Onishi, Proceedings of the 2020 Genetic and Evolutionary Computation Conference. the 2020 Genetic and Evolutionary Computation ConferenceCancun, Mexico; New York, NY, USAACMOzaki, Y.; Tanigaki, Y.; Watanabe, S.; Onishi, M. Multiobjective tree-structured parzen estimator for computationally expensive optimization problems. In Proceedings of the 2020 Genetic and Evolutionary Computation Conference, Cancun, Mexico, 8-12 July 2020; ACM: New York, NY, USA, 2020; pp. 533-541. MissForest-non-parametric missing value imputation for mixed-type data. D J Stekhoven, P Bühlmann, 10.1093/bioinformatics/btr597Bioinformatics. 28Stekhoven, D.J.; Bühlmann, P. MissForest-non-parametric missing value imputation for mixed-type data. Bioinformatics 2012, 28, 112-118. [CrossRef] Research Welcome to LightGBM's Documentation!-LightGBM 3.1.1.99 Documentation. K Guolin, Microsoft, 11Available online: githubGuolin, K. Microsoft Research Welcome to LightGBM's Documentation!-LightGBM 3.1.1.99 Documentation. 2016. Available online: github.com (accessed on 11 December 2020). Available online: github.com (accessed on. W Mckinney, Pandas, 11McKinney, W. Pandas. 2008. Available online: github.com (accessed on 11 December 2020). . Jags-Just Another Gibbs Sampler, 11JAGS-Just Another Gibbs Sampler. Available online: http://mcmc-jags.sourceforge.net/ (accessed on 11 December 2020). Ordered quantile normalization: A semiparametric transformation built for the cross-validation era. R A Peterson, J E Cavanaugh, 10.1080/02664763.2019.1630372J. Appl. Stat. 2020Peterson, R.A.; Cavanaugh, J.E. Ordered quantile normalization: A semiparametric transformation built for the cross-validation era. J. Appl. Stat. 2020, 47, 2312-2327. [CrossRef]

2022-08-31T06:02:33Z

CCBY

https://academic.oup.com/nar/article-pdf/44/D1/D1262/9482914/gkv1187.pdf

GOLD

9a929618736bf8a69f64ef9087381ee99394fe74

journals/nar/TorenBTLMF16

10.1093/nar/gkv1187

2174493298

26590258

4702847

MitoAge: a database for comparative analysis of mitochondrial DNA, with a special focus on animal longevity 2016 Dmitri Toren The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Thomer Barzilay The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Robi Tacutu The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Gilad Lehmann The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael The Rappaport Faculty of Medicine and Research Institute Tumor and Vascular Biology Research Center Technion -Israel Institute of Technology HaifaIsrael Khachik K Muradian Institute of Gerontology Life Span Prolongation Group KievUkraine Vadim E Fraifeld The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael MitoAge: a database for comparative analysis of mitochondrial DNA, with a special focus on animal longevity Nucleic Acids Research 44201610.1093/nar/gkv1187Received August 20, 2015; Revised October 21, 2015; Accepted October 23, 2015 Mitochondria are the only organelles in the animal cells that have their own genome. Due to a key role in energy production, generation of damaging factors (ROS, heat), and apoptosis, mitochondria and mtDNA in particular have long been considered one of the major players in the mechanisms of aging, longevity and age-related diseases. The rapidly increasing number of species with fully sequenced mtDNA, together with accumulated data on longevity records, provides a new fascinating basis for comparative analysis of the links between mtDNA features and animal longevity. To facilitate such analyses and to support the scientific community in carrying these out, we developed the MitoAge database containing calculated mtDNA compositional features of the entire mitochondrial genome, mtDNA coding (tRNA, rRNA, protein-coding genes) and non-coding (D-loop) regions, and codon usage/amino acids frequency for each protein-coding gene. MitoAge includes 922 species with fully sequenced mtDNA and maximum lifespan records. The database is available through the MitoAge website (www.mitoage.org or www.mitoage.info), which provides the necessary tools for searching, browsing, comparing and downloading the data sets of interest for selected taxonomic groups across the Kingdom Animalia. The Mi-toAge website assists in statistical analysis of different features of the mtDNA and their correlative links to longevity. INTRODUCTION Mitochondria are the only organelles in the animal cells that have their own genome. The stability of the mitochondrial DNA (mtDNA) is vital for mitochondrial proper functioning; therefore, changes in mtDNA may have far-reaching consequences for the cell fate and, ultimately, for the whole organism. Not surprisingly, due to a key role in energy production, generation of damaging factors (ROS, heat), and regulation of apoptosis, mitochondria and mtDNA in particular have long been considered one of the major players in the mechanisms of aging, longevity and age-related diseases (1)(2)(3)(4)(5)(6). Mitochondrial DNA exists in multiple copies and typically contains genes encoding for 13 key subunits of the respiratory chain enzymes, a set of 22 tRNA genes, and 2 genes for the large (16S) and small (12S) rRNA subunits. In contrast to the nuclear DNA, mtDNA is a circular, intronless, extremely compact molecule, with asymmetric distribution of nucleotides between the heavy (G-rich) and light (C-rich) strands (7,8). With very few exceptions, such structure of mtDNA is typical for the vast majority of animal species. Longevity (generally estimated by maximum lifespan, MLS) varies greatly among animal species (9) (genomics.senescence.info/species). Species also differ in their mtDNA compositional features (10), which to a great extent may determine the mtDNA stability and mutability. A few lines of evidence point towards a putative significance of mtDNA in aging and longevity. Firstly, mtDNA mutations accumulate with advanced age (5,11,12). Secondly, strong correlative links between mammalian MLS and mtDNA compositional features have been found (4,6,(13)(14)(15)(16)(17)(18). The rapidly increasing number of species with fully sequenced mtDNA genomes, together with accumulated data on longevity records, provide now a strong basis for * To whom correspondence should be addressed. Tel: +972 8 6477292; Fax: +972 8 6477626; Email: [email protected] † These authors contributed equally to the paper as first authors. comprehensive comparative analysis of the links between mtDNA features and animal longevity. Yet, efficient processing of such amount of data is computationally demanding. In turn, this generates a need for appropriate databases and bioinformatics tools. With the creation of MitoAge, we aim to encourage the modeling of mtDNA-longevity relationships, providing the scientific community with one single place to access, compare and analyze the data, based on most updated resources. MitoAge (www.mitoage.org, www.mitoage.info) is a curated, publicly available database, which contains an extensive repository of mtDNA data integrated with longevity records and the results of the statistical and correlative analysis of the links between them. DATABASE CONTENT AND INTERFACE To date, 5337 entries with complete mitochondrial genomes and 4237 entries with longevity records are available at NCBI RefSeq database and the AnAge database, respectively (see the Data Sources section). The overlap of these two data sets after curation, encompassing 922 animal species covering 304 families, 106 orders and 13 classes from the Kingdom Animalia, was included in MitoAge, Build 1.0. As seen in Table 1, the vast majority of species are vertebrates, with only few representatives of invertebrates. MitoAge contains compositional features (base content, GC%, AT%, sequence length) of the entire mitochondrial genome, mtDNA coding (tRNA, rRNA, protein-coding genes) and non-coding (D-loop) regions for each species and taxonomic group. For protein-coding genes of a given species, codon usage with distribution of codons by base position (e.g. codons with first base G, C, A or T) and amino acids frequencies are included. Along with mtDNA data, longevity records (MLS) are presented. Additionally, the MitoAge database tools provide the user with a number of options for (i) computation of basic statistics (range, median, mean ± standard deviation, coefficient of variation, Pearson's coefficient of correlation with log-transformed MLS); (ii) comparison of stats between selected taxonomic groups (two or more); (iii) data export for a data set of interest (in a CSV format), without downloading the entire database. If a user needs a more complex analysis, the website allows downloading the entire database, which can be done from the Download page in versioned releases (numbered database builds). MitoAge has a user-friendly website interface with simple and intuitive navigation tools (Figures 1 and 2). Searching can be done either by species common or scientific name, or by taxonomy groups (i.e. by classes, orders or families). Alternatively, the data can be reached by Browsing in three different ways: (i) Browsing Taxonomy (classes, orders, families or species); (ii) Browsing Stats, which calculates on-thefly statistical information for the total mtDNA or specific genes/regions for the selected taxonomic group; (iii) Brows- ing Genes, an option similar to that of browsing stats, but providing data restricted to a gene of interest. DATA SOURCES AND DATA CURATION The MitoAge database was constructed using publicly available data, is being constantly updated through automatic tools and is manually curated for problematic issues. Complete mtDNA sequences were taken from NCBI RefSeq database (www.ncbi.nlm.nih.gov/refseq) (19); longevity records were retrieved from the HAGR: Human Ageing Genomic Resources--AnAge database (genomics.senescence.info/species) (9), and full taxonomy data were retrieved from the Integrated Taxonomic Information System (ITIS) on-line database [31-Jul-2015 Build] (www.itis.gov). Most of the data included in the MitoAge database were computed offline, using a series of automated scripts and programs developed in our lab, with a number of parameters computed on-the-fly through the website. Together with a series of administration tools, this ensures frequent updates of the database. Data were computed and analyzed as follows: (i) base composition and size were generated for total mtDNA and its specific regions/genes (D-loop, protein-coding genes, rRNA-coding genes and tRNA-coding genes); (ii) for each protein-coding gene and for the total protein-coding sequence, both base composition and codon usage/amino acids frequency were computed. All data available at MitoAge have undergone a threestep validation: (i) the collected raw data from NCBI Ref-Seq was processed through our software and log files were generated; (ii) reports of errors and inconsistencies from log files were then manually curated; and finally (iii) additional tests for database consistency (e.g. duplicates, faulty values, etc.) were performed by the administrative tools in the MitoAge website. The list of errors and inconsistencies found in the mtDNA sequence annotations were reported to the NCBI admin and subsequently corrected by the NCBI team. Upon successful validation, the corrected data were uploaded into the MitoAge database and various statistical metrics were computed. DATA CALCULATION The compositional features of the mtDNA heavy strand (H-strand) were mostly considered (unless indicated otherwise), since the H-strand represents a primary target for the directional mutation pressure (20). For computing codon usage, data were taken from the complement of the coding strand and Thymine (T) was replaced with Uracil (U) (i.e. transforming it into the mRNA sequence). When combining multiple genes (e.g. when computing the total proteincoding genes), we append one gene to another and use the entire sequence for the analysis. In case of overlap between genes, we count the overlapping sequences only once for the computation of base composition, and we count them twice for the computation of codon usage and amino acids frequencies for each gene. Sequences are always analyzed from 5 to 3 (both for DNA and RNA). AVAILABILITY The MitoAge database is available at www.mitoage.org and www.mitoage.info, with the data made available under the permissive Creative Commons license, allowing data to be used in other analyses. There are options to either download the entire database or its parts, or to export specific results of the website. Feedback via email is welcome. CONCLUSION The MitoAge database is an integrated web resource for comparative analysis of mtDNA, with a special focus on animal longevity. Thus, it fills the gap in one of the 'hot spots' at the crossroad of aging research, evolutionary and mitochondrial biology. MitoAge novelties, together with a userfriendly interface, provide unique capabilities for in-depth investigation of: (i) the abundance of mtDNA bases and its variability across animal taxa with different MLS; (ii) the links between longevity and the mtDNA compositional features (total, region or gene-specific), including codon usage and amino acids frequency. Using the MitoAge database, we revealed more than 3780 statistically significant correlations between mtDNA features and animal longevity in different taxonomic groups. MitoAge could be also a useful supporting tool for building predictive models of animal longevity, involving mtDNA features. CFigure 1 . 1The Author(s) 2015. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Nucleic Acids Research, 2016, Vol. 44, Database issue D1263 A labeled diagram of the MitoAge interface. Figure 2 . 2Example: viewing data for Arctica islandica. FUNDING Fund in Memory of Dr Amir Abramovich. Funding for open access charge: Ben-Gurion University of the Negev [to V.E.F.]. Conflict of interest statement. None declared. Table 1 . 1Number of species per taxon in MitoAgeTaxa (scientific name) A mitochondrial paradigm of metabolic and degenerative diseases, aging, and cancer: a dawn for evolutionary medicine. D C Wallace, Annu. Rev. Genet. 39Wallace,D.C. (2005) A mitochondrial paradigm of metabolic and degenerative diseases, aging, and cancer: a dawn for evolutionary medicine. Annu. Rev. Genet., 39, 359-407. The hallmarks of aging. C Lopez-Otin, M A Blasco, L Partridge, M Serrano, G Kroemer, Cell. 153Lopez-Otin,C., Blasco,M.A., Partridge,L., Serrano,M. and Kroemer,G. (2013) The hallmarks of aging. Cell, 153, 1194-1217. Mitochondria sentencing about cellular life and death: a matter of oxidative stress. N Apostolova, A Blas-Garcia, J V Esplugues, Curr. Pharm. Des. 17Apostolova,N., Blas-Garcia,A. and Esplugues,J.V. (2011) Mitochondria sentencing about cellular life and death: a matter of oxidative stress. Curr. Pharm. Des., 17, 4047-4060. Do mitochondrial DNA and metabolic rate complement each other in determination of the mammalian maximum longevity?. G Lehmann, E Segal, K K Muradian, V E Fraifeld, Rejuvenation Res. 11Lehmann,G., Segal,E., Muradian,K.K. and Fraifeld,V.E. (2008) Do mitochondrial DNA and metabolic rate complement each other in determination of the mammalian maximum longevity? Rejuvenation Res., 11, 409-417. ) mtDNA mutations in human aging and longevity: controversies and new perspectives opened by high-throughput technologies. F Sevini, C Giuliani, D Vianello, E Giampieri, A Santoro, F Biondi, P Garagnani, G Passarino, D Luiselli, M Capri, Exp. Gerontol. 56Sevini,F., Giuliani,C., Vianello,D., Giampieri,E., Santoro,A., Biondi,F., Garagnani,P., Passarino,G., Luiselli,D., Capri,M. et al. (2014) mtDNA mutations in human aging and longevity: controversies and new perspectives opened by high-throughput technologies. Exp. Gerontol., 56, 234-244. Mitochondrial DNA mutations in aging. K Khrapko, D Turnbull, Prog. Mol. Biol. Transl. Sci. 127Khrapko,K. and Turnbull,D. (2014) Mitochondrial DNA mutations in aging. Prog. Mol. Biol. Transl. Sci., 127, 29-62. Transcription and replication of mitochondrial DNA. D A Clayton, Hum. Reprod. 15Suppl. 2Clayton,D.A. (2000) Transcription and replication of mitochondrial DNA. Hum. Reprod., 15(Suppl. 2), 11-17. Replication and transcription of mammalian mitochondrial DNA. P Fernandez-Silva, J A Enriquez, J Montoya, Exp. Physiol. 88Fernandez-Silva,P., Enriquez,J.A. and Montoya,J. (2003) Replication and transcription of mammalian mitochondrial DNA. Exp. Physiol., 88, 41-56. Human Ageing Genomic Resources: integrated databases and tools for the biology and genetics of ageing. R Tacutu, T Craig, A Budovsky, D Wuttke, G Lehmann, D Taranukha, J Costa, V E Fraifeld, De Magalhaes, Nucleic Acids Res. 41J.P.Tacutu,R., Craig,T., Budovsky,A., Wuttke,D., Lehmann,G., Taranukha,D., Costa,J., Fraifeld,V.E. and de Magalhaes,J.P. (2013) Human Ageing Genomic Resources: integrated databases and tools for the biology and genetics of ageing. Nucleic Acids Res., 41, D1027-D1033. Mitochondrial DNA in metazoa: degree of freedom in a frozen event. C Saccone, C Gissi, A Reyes, A Larizza, E Sbisa, G Pesole, Gene. 286Saccone,C., Gissi,C., Reyes,A., Larizza,A., Sbisa,E. and Pesole,G. (2002) Mitochondrial DNA in metazoa: degree of freedom in a frozen event. Gene, 286, 3-12. Reconsidering the Role of Mitochondria in Aging. M Gonzalez-Freire, R De Cabo, M Bernier, S J Sollott, E Fabbri, P Navas, L Ferrucci, J. Gerontol. A Biol. Sci. Med. Sci. 70Gonzalez-Freire,M., de Cabo,R., Bernier,M., Sollott,S.J., Fabbri,E., Navas,P. and Ferrucci,L. (2015) Reconsidering the Role of Mitochondria in Aging. J. Gerontol. A Biol. Sci. Med. Sci., 70, 1334-1342. Do mtDNA deletions drive premature aging in mtDNA mutator mice?. Y Kraytsberg, D K Simon, D M Turnbull, K Khrapko, Aging Cell. 8Kraytsberg,Y., Simon,D.K., Turnbull,D.M. and Khrapko,K. (2009) Do mtDNA deletions drive premature aging in mtDNA mutator mice? Aging Cell, 8, 502-506. Telomere length and body temperature-independent determinants of mammalian longevity?. G Lehmann, K K Muradian, V E Fraifeld, Front. Genet. 4111Lehmann,G., Muradian,K.K. and Fraifeld,V.E. (2013) Telomere length and body temperature-independent determinants of mammalian longevity? Front. Genet., 4, 111. Life span is related to the free energy of mitochondrial DNA. D C Samuels, Mech. Ageing Dev. 126Samuels,D.C. (2005) Life span is related to the free energy of mitochondrial DNA. Mech. Ageing Dev., 126, 1123-1129. Mitochondrial DNA repeats constrain the life span of mammals. D C Samuels, Trends Genet. 20Samuels,D.C. (2004) Mitochondrial DNA repeats constrain the life span of mammals. Trends Genet., 20, 226-229. Inverse relationship between longevity and evolutionary rate of mitochondrial proteins in mammals and birds. N Galtier, P U Blier, B Nabholz, Mitochondrion. 9Galtier,N., Blier,P.U. and Nabholz,B. (2009) Inverse relationship between longevity and evolutionary rate of mitochondrial proteins in mammals and birds. Mitochondrion, 9, 51-57. Mitochondrial genome anatomy and species-specific lifespan. G Lehmann, A Budovsky, K K Muradian, V E Fraifeld, Rejuvenation Res. 9Lehmann,G., Budovsky,A., Muradian,K.K. and Fraifeld,V.E. (2006) Mitochondrial genome anatomy and species-specific lifespan. Rejuvenation Res., 9, 223-226. Mitochondrial inverted repeats strongly correlate with lifespan: mtDNA inversions and aging. J N Yang, A Seluanov, V Gorbunova, PLoS One. 873318Yang,J.N., Seluanov,A. and Gorbunova,V. (2013) Mitochondrial inverted repeats strongly correlate with lifespan: mtDNA inversions and aging. PLoS One, 8, e73318. Update on RefSeq microbial genomes resources. T Tatusova, S Ciufo, S Federhen, B Fedorov, R Mcveigh, K O&apos;neill, I Tolstoy, L Zaslavsky, Nucleic Acids Res. 43Tatusova,T., Ciufo,S., Federhen,S., Fedorov,B., McVeigh,R., O'Neill,K., Tolstoy,I. and Zaslavsky,L. (2015) Update on RefSeq microbial genomes resources. Nucleic Acids Res., 43, D599-D605. Asymmetrical directional mutation pressure in the mitochondrial genome of mammals. A Reyes, C Gissi, G Pesole, C Saccone, Mol. Biol. Evol. 15Reyes,A., Gissi,C., Pesole,G. and Saccone,C. (1998) Asymmetrical directional mutation pressure in the mitochondrial genome of mammals. Mol. Biol. Evol., 15, 957-966.

2022-08-27T11:09:50Z

CCBY

https://www.mdpi.com/2073-4425/11/3/286/pdf

GOLD

b34684ec3cbc3c9e40621ece2d4af67510153c76

10.3390/genes11030286

3009570400

32182725

7140858

LRRpredictor-A New LRR Motif Detection Method for Irregular Motifs of Plant NLR Proteins Using an Ensemble of Classifiers Eliza C Martin [email protected]. Department of Bioinformatics and Structural Biochemistry Institute of Biochemistry of the Romanian Academy Splaiul Independentei 296060031BucharestRomania C A Octavina Sukarta Laboratory of Nematology Wageningen University and Research 6700ESWageningenThe Netherlands Laurentiu Spiridon [email protected]. Department of Bioinformatics and Structural Biochemistry Institute of Biochemistry of the Romanian Academy Splaiul Independentei 296060031BucharestRomania Laurentiu G Grigore Space Comp SRL 041512BucharestRomania Vlad Constantinescu [email protected]. Department of Bioinformatics and Structural Biochemistry Institute of Biochemistry of the Romanian Academy Splaiul Independentei 296060031BucharestRomania Robi Tacutu [email protected]. Department of Bioinformatics and Structural Biochemistry Institute of Biochemistry of the Romanian Academy Splaiul Independentei 296060031BucharestRomania Aska Goverse Laboratory of Nematology Wageningen University and Research 6700ESWageningenThe Netherlands Andrei-Jose Petrescu Department of Bioinformatics and Structural Biochemistry Institute of Biochemistry of the Romanian Academy Splaiul Independentei 296060031BucharestRomania LRRpredictor-A New LRR Motif Detection Method for Irregular Motifs of Plant NLR Proteins Using an Ensemble of Classifiers 10.3390/genes11030286Received: 7 February 2020; Accepted: 4 March 2020; Published: 8 March 2020genes G C A T T A C G G C A T Article * Correspondence: [email protected] (A.G.); [email protected] (A.-J.P.);leucine-rich repeat predictionsupervised learningLRR motifLRR structureNOD-like receptorsR proteins Leucine-rich-repeats (LRRs) belong to an archaic procaryal protein architecture that is widely involved in protein-protein interactions. In eukaryotes, LRR domains developed into key recognition modules in many innate immune receptor classes. Due to the high sequence variability imposed by recognition specificity, precise repeat delineation is often difficult especially in plant NOD-like Receptors (NLRs) notorious for showing far larger irregularities. To address this problem, we introduce here LRRpredictor, a method based on an ensemble of estimators designed to better identify LRR motifs in general but particularly adapted for handling more irregular LRR environments, thus allowing to compensate for the scarcity of structural data on NLR proteins. The extrapolation capacity tested on a set of annotated LRR domains from six immune receptor classes shows the ability of LRRpredictor to recover all previously defined specific motif consensuses and to extend the LRR motif coverage over annotated LRR domains. This analysis confirms the increased variability of LRR motifs in plant and vertebrate NLRs when compared to extracellular receptors, consistent with previous studies. Hence, LRRpredictor is able to provide novel insights into the diversification of LRR domains and a robust support for structure-informed analyses of LRRs in immune receptor functioning.Genes 2020, 11, 286 2 of 26 LRRK2 kinase enzyme, lead to Parkinson's disease and other associated inflammatory diseases[5,6], whereas mutations in leucine-rich proteoglycans have been previously shown to be involved in osteoarthritis[7], and last but not least PRELP mutations might have a role in Hutchinson-Gilford, an accelerated progeroid syndrome characterized by premature aging[8]. Hence, understanding the structural aspects of binding properties and specificities of LRR domains opens wide possibilities for receptor engineering with vast implications not only for improved crop resistance to plant diseases, but also for a wide range of medical applications.In innate immunity, LRR modules are found in various domain organizations in many receptor classes such as plant receptor-like kinases (RLK), receptor-like proteins (RLP), NOD-like receptors (NLR), or metazoan NLR and Toll-like receptors (TLR). In plant basal immunity, LRR N-terminal domains face the extracellular environment and are found in either receptor-like kinases (RLK) or receptor-like proteins (RLPs) depending on the presence or absence of a C-terminal kinase domain on the cytosolic side of the receptor. By contrast, LRRs constitute the C-terminal domains of intracellular NOD-like receptors (NLR), also known as resistance (R) proteins, and face the cytosolic environment to mediate resistance against specific pathogens. Depending on their N-terminal domain, which is either a coiled-coil (CC) or a toll-like receptor domain (TIR), R proteins fall into two main NLR classes: the CNL and TNL receptors, respectively[9]. Both these classes contain however a central nucleotide binding domain (NBS) which acts as a 'switch' that changes its conformation upon ADP/ATP binding [9,10].Metazoan NLRs show a similar organization with plant NLRs. They encode a variety of N-terminal 'sensors' (caspase activation and recruitment domains-CARD, baculovirus inhibitor of apoptosis repeat-BIR, etc.), the central 'switch' STAND domain (signal transduction ATPases with numerous domains) -NBS/NACHT domain (NAIP (neuronal apoptosis inhibitory protein), CIITA (MHC class II transcription activator), HET-E (incompatibility locus protein from Podospora anserina) and TP1(telomerase-associated protein)) and the LRR domain at the C-terminal end. Last but not least, we mention here the metazoan toll-like receptors (TLRs) that have an extracellular LRR sensor domain as seen in the RLK/RLP case and a TIR domain on the cytosolic side involved in signal transduction[11].From a structural point of view LRR domains have a solenoidal 'horseshoe' like 3D architecture composed of a variable number of repeats varying each from ≈15 to ≈30 amino acids in length. Repeats are held together through a network of hydrogen bonds which forms a beta sheet located on the ventral side of the 'horseshoe'. This is generated by a conserved sequence pattern named the LRR motif that in its minimal form is of the type 'LxxLxL' where L is generally leucine and to a lesser degree other hydrophobic amino acids[12]. Comprehensive sequence analysis of LRR immune receptors resulted in several classifications of LRR domains showing preferred amino acid conservation outside the minimal motif such as the two type classification proposed by Matsushima et al. [13] for TLR receptors or the seven type classification proposed by Kobe and Kajava [14] for all known LRR domains across all Kingdoms. However, exceptions to such rules are frequent as revealed by the Hidden Markov Model approach carried out by Ng et al. [15]. This highlighted the fact that most of the analyzed classes of human proteins containing LRR domains also display many irregular motifs alongside repeats showing the well-defined class specific motif[15].While the above mentioned receptor classes were shown to present LRR irregularities[15], studies on plant NLR proteins such as Lr10 and Pm3 from wheat, Rx1 and Gpa2 from potato, or ZAR1 from Arabidopsis show that their LRR domains have a far more variable and irregular structure than their extracellular counterparts[16][17][18][19][20][21][22]. These factors combined contribute to the challenge for the accurate prediction of LRR motifs in plant NLRs.A proper annotation of each LRR motif in a given LRR domain is instrumental in generating an accurate 3D model[12,23]and by this in properly defining the domain surface and identifying potential protein-protein interaction interfaces. An illustrative example is the conservation mapping performed by Helft et al. in 2011, which was used to identify new interaction partners of plant RLPs and RLKs by studying conserved 3D relationships among amino acids inferred from annotation of LRR repeats[24]. Introduction The leucine-rich-repeat (LRR) domains are present in all of the tree of life branches. As they are involved in protein-protein interactions, LRR domains are found in receptors having a vast number of functions such as pathogen detection, immune response propagation, hormone perception, enzyme inhibition, or cell adhesion [1]. In both plants and mammals, a number of studies have detailed adverse effects associated with mutations in the LRR domains such as that reported for various immune-related receptors, resulting in compromised functions and enhanced disease progression [2]. For example, mutating a single residue in the LRR domain of the rice Pita receptor results in complete loss of recognition against the fungus Magnaporthe grisea [3] while mutations in the metazoan NLRC4-LRR contributes to autoinflammatory disease phenotypes [4]. Additionally, mutations in the Based on our previous work, identifying the individual true motifs in a LRR domain is hindered by the following: (a) in its minimal form, a 'LxxLxL' pattern is trivial and frequently occurs randomly in any protein; (b) in many cases several 'LxxLxL' patterns do overlap in less than 15 aa range in NLR-LRRs making the precise delineation difficult; (c) the number of 3D experimental structures from which to learn is low; and (d) this small 3D learning set is class and phyla biased-as around half of the structures are of mammalian origin while plant NLRs only have one recently documented structure [21,22]. Thus, given the above described indeterminacies the precise LRR motif identification becomes the most problematic step in the correct repeat delineation within a LRR domain. This also explains why LRR domains and their individual repeats are poorly annotated in genomes or protein databases in contrast to the better annotated, relatively more conserved NBS domain, which has therefore been used in phylogenetic analyses [10,25]. Hence, these major limitations hamper the study of NLRs at various levels such as in the context of plant innate immunity. To address these challenges, in this paper we propose a new LRR motif detection method: LRRpredictor, designed to be more sensitive to motif irregularities than the existing methods like LRRfinder [26] or LRRsearch [27] and to detect irregular and short LRR signatures as are often found in plant NLRs, but not limited to this class. We assessed how LRRpredictor behaves within different classes of immune-related receptors that contain LRR domains, such as plant NLRs, RLPs, and RLKs and vertebrate NLRs and TLRs with the aim to provide novel insights into the diversification of LRR domains and their role in the functioning of immune receptors. Materials and Methods Assembly and Analysis of the LRR Structural Dataset Various protein domain databases, such as CATH [28], Pfam [29], and Interpro collection [30] were used to obtain a dataset of 611 structure files of proteins annotated to contain LRR domains. These files were processed and filtered out to extract a clean set of LRR chains sharing less than 90% sequence identity using Pisces server [31]. This set containing 178 LRR chains were visually inspected and subjected to LRR repeat delineation based on the distinctive LRR ventral beta-sheet secondary structure pattern. Annotated LRR domains consisting in less than five LRR repeats, as well as incomplete repeats not covering at least five amino acids upstream and downstream of the "LxxLxL" minimal motif were further eliminated. Using this procedure, we generated the 90% identity data set, ID90, consisting of 172 N-ter LRR 'entry' repeats (N), 1792 LRR 'core' repeats (L), and 154 C-ter LRR 'exit' repeats (C) (File S1). To avoid redundancy in the training data the level of identity has to be further significantly reduced. However, given the small size of ID90 (<180 chains), a trade-off between increase in entropy and loss of data had to be reached. As seen from Figure A1a, a proper inflection point shapes up at around 50% identity and was considered the best compromise in generating a nonredundant set of repeats. In practical terms, the nonredundant ID50 set was generated from ID90 by selecting repeats showing less than eight identical amino acids on a 16 amino acid window centered on the 'LxxLxL' minimal LRR motif, i.e., the window comprising five amino acids upstream and downstream 'LxxLxL'. This nonredundant ID50 set was comprised of 106 N-ter 'entry' repeats (N), 659 'core' repeats (L), and 88 C-ter 'exit' repeats (C), i.e., ≈40% of the 90ID set ( Figure 1, File S1). Jensen-Shannon divergence (JSD) scores ( Figure 1e) were computed using Capra et al. implementation [32], using the BLOSUM62 matrix for background probabilities and a window parameter 0. The phyla distribution shown in Figure 1c was computed using the Environment for Tree Exploration (ETE3) library v3.1.1 [33]. Training and Testing Datasets Construction In order to provide a representative collection of non-LRR examples, we selected a representative example of each CATH [28] domains' topology (except LRR) from a nonredundant dataset provided by CATH where all proteins share less than 20% identity or have a less than 60% overlap (cath-dataset-nonredundant-S20 set-09.12.2019). Given potential synchronization problems between various databases used to build the overall learning set comprising (a) the nonredundant 50ID LRRs, containing the 'entry'-, 'core'-, 'exit'-repeats and the flanking nonLRR domains when present and (b) the CATH nonLRR domains-the data was subjected to a third redundancy filter performed with a similar CATH methodology, aimed at eliminating sequences that fail one of the below bounds: • the length of the alignment is over 100 and the identity is over 20%. • length of the alignment is between 40 and 100 with an identity over 20% and the overlap with respect to both sequences is more than 60%. • LRR repeats with alignments lengths ≥16 aa and ≥50% identical (equivalent of at most 8/16 aa constraint imposed initially on the motifs). The final dataset built as above and used herein for training and testing classifiers, contains 648 LRR core repeats, 100 N-ter entry, and 67 C-ter exit nonredundant repeats (including the LRR domain flanking regions) and 875 non-LRR domains from CATH. From this set, 1/5th was used to generate the test dataset, while the remaining 4/5 were used to build the training datasets, preserving the class ratio between the sets. The test dataset contains 40,241 amino acid samples of which only 150, i.e., less than 0.4%, are initiating LRR motifs. Similarly, over the training set less than 0.5% of the samples are LRR initiators. The training set was further split into four cross-validation sets that were used for parameter optimization. All these sets are provided in File S2. Feature Selection and Data Pre-Processing In developing LRRpredictor we tested sequence-based (SeqB) features: solely or combined with structural based (StrB) features. The SeqB features comprise position-specific scoring matrices PSSM over the above discussed 16 amino acids interval summing up to 320 features corresponding to 20 amino acid types over the 16 positions. The StrB features comprise: (a) the three state (H-helix, E-extended, C-coil) secondary structure probabilities, (b) the three class (B-buried, M-medium and E-exposed) residue relative solvent accessibility, RSA probabilities and (c) intrinsic disorder probability-summing up to seven extra structural features per residue, resulting in a total of 432 features per 16 aa window. The structural based predictions were performed with RaptorX-Property software [34][35][36][37]. Sequence PSSMs were computed on Uniprot20 protein sequence database, using HHblits [38,39] that is based on HMM-HMM alignments shown to improve accuracy of alignments at low sequence homology levels. In the pre-processing stage, feature variables were normalized, centered, and rescaled, as standard procedure involves. Data whitening using principal component analysis (PCA) decomposition was not used as it did not provide better performance on the tested classifiers. Machine Learning Model Selection Several classifiers such as support vector classification (SVC) [40], multi-layer perceptron (MLP) [41,42], and AdaBoost [43] as well as several oversampling techniques such as Adasyn [44] and SMOTE-based varieties [45][46][47], or over-and under-sampling combined approaches SmoteTomek [48] and SmoteEEN [49], were tested and parameter optimized via cross-validation using Scikit-learn library v.0.22.1 [50]. Multiclass estimators for N-entry (N), core (L), and C-exit (C) motif types that use either one-vs.-one or one-vs.-rest approaches were also investigated, but they performed worse than when treating all LRR motifs as a single class. The best performing classifiers with tuned parameters were further studied in the context of a soft voter (that averages predicted probabilities of the ensemble constituents), and a final predictor, Genes 2020, 11, 286 5 of 26 further referred to as LRRpredictor, was chosen based on its out-of-sample performance on test set and overfitting behavior on the training data. LRRpredictor is composed of a set of eight classifiers (C1-C8) that use different strategies and consider all N, L, C motif types as a single class, aggregated within an ensemble based on the soft voting scheme, as shown in Figure 2d. Classifiers C1-C4 use solely sequence-based features while C5-C8 use both sequence and structural-based features. Classifiers C1 and C5 use the support vector classification (SVC) algorithm [40], with a radial basis function (RBF) kernel, one-vs.-rest ('ovr') decision function. The margin penalty and the RBF scale (gamma) parameters were optimized through grid search to 1 and 0.01 for C1 and 1 and 0.001 for C4, respectively. Class imbalance was treated by adjusting the SVM weights inversely proportional to class frequency and class probabilities were inferred using sigmoid probability calibration. Classifiers C2, C3, C6, C7 use multi-layer perceptron (MLP) [41,42]. A depth of three hidden layers was sufficient to describe the system, as adding additional hidden layers provided little to no difference in out-of-sample performance. The number of hidden nodes for each hidden layer was selected via grid search as follows: C2 (300-250-100), C3 (250-150-100), C5 (250-150-100), C6 (125-100-10). Classifiers C2, C3, C7 use the Limited-Memory BFGS [51] solver, while C6 uses Adam [52] optimizer for stochastic gradient descent [53] with early-stopping over a validation fraction of 0.2. All four classifiers use rectified linear unit (ReLU) activation function [54]. Classifiers C3 and C7 approach the imbalance problem through synthetic resampling using the combined over-and under-sampling method SMOTETomek [48], as implemented in imbalanced-learn library v 0.6.1 [55]. Classifiers C4 and C8 use a ensemble boosting approach-AdaBoost [43]-using tree classifiers of depth 1, as base estimators, SAMME.R real boosting algorithm, and sigmoid probability calibration. A maximum number of 50 base estimators was selected to maximize performance while avoiding overfitting. Assembly of Protein Family Sets Containing LRR Domains In order to investigate LRRpredictor behavior on previously annotated LRR domains from various functional protein groups, we generated a collection of randomly selected 500 representatives from Uniprot50 database (i.e., below 50% identity between themselves at a given minimum overlap-version available at 20.11.2019-release-2019_10) which were annotated by Interpro to contain a LRR domain (IPR032675 and Interpro v77.0 protein2ipr database). A total of six groups were generated: four groups of sequences of CNLs, TNLs, RLKs, and RLPs protein classes from flowering plants and two groups of TLRs and NLRs from vertebrates. Given the high conservation of vertebrate TLRs this set gathered only ≈350 sequences (File S3). Within the CNL group, there were included only proteins annotated by Interpro to contain a single coiled-coil (CC) domain, a single NBS domain, and a LRR domain in this order, and sequences that contained a different domain organization, such as two annotated NBS domains or a different domain order were not included in the analysis. Similarly, for the TNL group we selected only sequences that contain a TIR-NBS-LRR domain organization. The RLK group was built with sequences displaying a "LRR-TM predicted region-kinase" domain organization, while the RLP group contained sequences with "LRR-TM" organization and did not contain other annotated domains by Interpro. In generating the vertebrate NLR group we included any annotated NACHT or NBS domains followed by a LRR domain annotation without discriminating on the N-terminal domain, as animal NLRs can have upstream of the NACHT/NBS domain a multitude of N-terminal domain types, while vertebrate TLRs group contains sequences with a "LRR-TM-TIR" configuration. Transmembrane predictions were performed using Phobius [56]. In analyzing the length of the LRR domains covered by individual repeat annotations, we used all Interpro annotation codes associated with LRR repeat types. We considered as having the status of 'annotated as domain' LRRs with the IPR032675 label and 'annotated as repeats' any amino acids Genes 2020, 11, 286 6 of 26 that had attached by at least one predictor part of Interpro collection one of the following tags: leucine-rich repeat (IPR001611), leucine-rich repeat, typical subtype (IPR003591), leucine-rich repeat, cysteine-containing subtype (IPR006553), leucine-rich repeat 2 (IPR013101), leucine-rich repeat 3 (IPR011713), leucine rich repeat 4 (IPR025875), BspA type leucine rich repeat region (IPR026906), CD180 leucine-rich repeat (IPR041281), DUF4458 domain-containing protein, leucine-rich repeat (IPR041403). Annotations referring to the N-ter cap of the LRR domain (IPR000372, IPR041302) were not considered as these are not LRR repeats. Assessment of LRR Motif Conservation Across Protein Groups Intra-and inter-group sequence variability was also analyzed using a subset of 1000 predicted 16 aa extended motifs from each group. In order to avoid a potential bias induced by false 'entry' (N) or 'exit' (C) repeats, only 'core' (L) repeats were used in this analysis. The similarity measure used here is the distance mapping defined by Halperin et al. [57]. This consists of the inner product of BLOSUM scores between each pair of amino acids summed up over the motif span, as this function can be used as a metric distance for several BLOSUM matrices. Considering d to be the distance between a pair of amino acids i and j, that have the s (i, j) BLOSUM score: d(i, j) = s(i, i) + s( j, j) − 2·s(i, j)(1) The distance between two sequences a and b of equal length l, would be the sum of distances of each pair of amino acids a i and b i across the length of the sequence: D a,b = l i=0 d(a i , b i ),(2) This definition of distance is expected to reflect amino acids compatibilities, as BLOSUM scores are inferred from amino acid mutation probabilities observed on large datasets. As a BLOSUM matrix we selected an updated version of the original BLOSUM matrix, which was recently recalculated on a large dataset and satisfies the triangle inequality. (RBLOSUM59_14.3) [58,59]. Starting from the above described distance function, we calculated Silhouette coefficients [60] between each pair of groups, and precomputed distances were used for manifold learning using metric multi-dimensional scaling (MDS) [61] as implemented in Scikit-learn library [50]. Sequences logos were generated using Weblogo [62], figures showing protein structures were obtained using PyMol [63], while other plots were generated in Microsoft Office or by using Matplotlib library [64]. Results Available LRR Domains in Structural Data A collection of 611 PDB structures previously annotated by several protein domain databases, such as CATH [28], Pfam [29], and Interpro-collection [30] to contain LRR horseshoe architectures was obtained. This collection was used to derive a clean set, ID90, of 178 LRR chains displaying 90% identity that was structurally analyzed in order to structurally delineate the LRR repeats based on the beta-sheet network. By this, a dataset of ≈2100 LRR motifs was obtained, as shown in Figure 1a. It is interesting to note here that less than 20% of these are annotated as LRR motifs in Pfam even though the 178 sequences were derived from known 3D structures. The LRR motif annotation of each repeat was performed starting with the first position (L 0 ) of the minimal motif 'L 0 XXL 3 XL 5 ', position that marks the beginning of the ventral side of the horseshoe domain ( Figure 1b). Superposition of the 2100 repeats indicates that the structural similarity extends in most of the cases over five positions upstream and downstream of the minimal motif defining a 16 positions region which is referred herein as the 'extended' motif ( Figure A1d). Due to this, the structural Genes 2020, 11, 286 7 of 26 LRR diversity concentrates mainly onto the dorsal side of the horseshoe which imposes onto the curvature and the overall geometry of the domain ( Figure 1b). As duplications of highly similar LRR repeats within the same LRR domain is abundant in the ID90 set, we opted to perform a second redundancy filter at the level of LRR repeats as described (M&M). This results in the ID50 nonredundant set consisting of ≈850 LRR repeats, that approximates well the ID90 distribution of lengths ( Figure A1c), phyla (Figure 1c), and the ratio between marginal N-terminal (N) and C-terminal (C) versus interior motifs (L) ( Figure A1b). The 'entry' N-ter LRR motifs are less regular than the 'core' motifs, especially at the first hydrophobic position (L 0 ) that is often found solvent exposed, as this position marks the end of the inter-domain linker and the beginning of the LRR domain. By contrast, the 'exit' C-ter LRR motifs better resemble the 'core' motifs (L) amino acid composition and the conventional LRR motif 'L 0 xxL 3 xL 5 xx(N/C) 8 xL 10 ' (Figure 1d). Interestingly, the stringency for leucine occurrence sequentially decreases from L 0 to L 3 and L 5 in core repeats, allowing other amino acids to be present in L 3 and L 5 more frequently (Figure 1d). This structurally correlates with a larger accessible space of the protein core structure around L 3 and L 5 positions, as can be seen from Figures 1b and A1d. It is also worth noting that the third L -3 position upstream of LxxLxL has a significant hydrophobic propensity presumably allowing the solenoid to form (Figure 1d). Another important facet that has to be carefully pondered is the high phyla bias of the structural data when compared to the baseline phyla distribution of the UniRef50 database. As can be seen from Figure 1c, around 50% of the repeats in ID50 are of mammalian origin while the UniRef50 baseline is of less than 3% in both annotated LRR proteins or any protein. Moreover, the ≈20% plant LRR motifs present in ID50 originate overwhelmingly from RLP and RLK proteins while plant NLRs are poorly represented in this set, with only a single 3D structure recently reported for the ZAR1 NLR protein from Arabidopsis thaliana [21,22]. well the ID90 distribution of lengths ( Figure A1c), phyla (Figure 1c), and the ratio between marginal N-terminal (N) and C-terminal (C) versus interior motifs (L) ( Figure A1b). The 'entry' N-ter LRR motifs are less regular than the 'core' motifs, especially at the first hydrophobic position (L0) that is often found solvent exposed, as this position marks the end of the inter-domain linker and the beginning of the LRR domain. By contrast, the 'exit' C-ter LRR motifs better resemble the 'core' motifs (L) amino acid composition and the conventional LRR motif 'L0xxL3xL5xx(N/C)8xL10' (Figure 1d). Interestingly, the stringency for leucine occurrence sequentially decreases from L0 to L3 and L5 in core repeats, allowing other amino acids to be present in L3 and L5 more frequently (Figure 1d). This structurally correlates with a larger accessible space of the protein core structure around L3 and L5 positions, as can be seen from Figures 1b and A1d. It is also worth noting that the third L-3 position upstream of LxxLxL has a significant hydrophobic propensity presumably allowing the solenoid to form (Figure 1d). Another important facet that has to be carefully pondered is the high phyla bias of the structural data when compared to the baseline phyla distribution of the UniRef50 database. As can be seen from Figure 1c, around 50% of the repeats in ID50 are of mammalian origin while the UniRef50 baseline is of less than 3% in both annotated LRR proteins or any protein. Moreover, the ≈20% plant LRR motifs present in ID50 originate overwhelmingly from RLP and RLK proteins while plant NLRs are poorly represented in this set, with only a single 3D structure recently reported for the ZAR1 NLR protein from Arabidopsis thaliana [21,22]. (PDB: 6J5W). The hydrophobic positions in the minimal 'L 0 xxL 3 xL 5 ' motif are shown in orange. The first N-entry repeat (blue) and the last C-exit repeat (red) are also mapped on the structure. (c) Phyla distribution of the initial LRR motif set ID90, the 50% identity trimmed LRR motifs set (ID50), annotated LRR proteins and all proteins from the UniRef50 database (from left to right). Percent values corresponding to the mammals group are shown in red. (d) Frequency plot of amino acid composition of the N-entry, core and C-exit motifs on the 50% identity trimmed set. Amino acids are colored according to their properties as follows: hydrophobic (yellow), acidic (red), basic (blue), asparagine and glutamine (purple), proline and glycine (green), others (black). (e) Jensen-Shannon divergence (JSD) score for each position of the LRR motif at different identity thresholds. Higher values show increased conservation. Development of the LRRpredictor Method In order to train a machine learning (ML) estimator for detecting LRR motifs we used an overall dataset comprising the filtered LRR ID50 dataset and a collection of 875 non-LRR domains composed of one representative of each CATH topology ( Figure 2a). As discussed in Section 1, the sequence patterns corresponding to the ≈850 actual true structural LRR motifs identified in ID50 are quite common in any protein. We will name here such sequence patterns as potential motifs. As expected, Table 1 shows that potential motifs occur with more or less equal probability in both the LRR and non-LRR domains of the overall dataset. Moreover, even when taking into account only LRR domains the number of potential motifs is larger than the number of true structural LRR motifs (Table 1). This allows the ML estimators to learn to detect true motifs from the far larger set of potential motifs by taking into account the larger 16 amino acid sequence context in which the true motifs are embedded. In this way the method developed herein can be used not only to delineate repeats in a given LRR domain but also to discriminate between protein products that do not have LRR domains from those hosting such domains. In developing LRRpredictor we tested 'sequence-based' features based on position-specific scoring matrices-PSSMs either solely or combined with 'structural-based' features as described in Section 2 ( Figure 2c). PSSM profiles are expected to provide context information on the overall sequence, to highlight the key amino acids position that are conserved, as the amino acids scores are derived from amino acid substitution probabilities conditioned by the homologues family they belong to. Therefore, it is expected that irregular LRR motifs would be more detectable when using sequence profiles, rather than amino acid sequence alone. scoring matrices-PSSMs either solely or combined with 'structural-based' features as described in Section 2 ( Figure 2c). PSSM profiles are expected to provide context information on the overall sequence, to highlight the key amino acids position that are conserved, as the amino acids scores are derived from amino acid substitution probabilities conditioned by the homologues family they belong to. Therefore, it is expected that irregular LRR motifs would be more detectable when using sequence profiles, rather than amino acid sequence alone. The dataset was split into five parts: one part was initially separated as test set and the other four were used as a training set in parameter tuning using a four-fold cross-validation (CV) approach, where models were iteratively trained on three of the CV sets and tested on the remaining fourth ( Figure 2b). A pool of estimators (representing algorithms for classification) that used either (1) sequence-based or (2) both sequence and structural features and (3) various imbalance class treatments were optimized via cross-validation. Finally, best performing estimators were studied in the context of an ensemble estimator. The selected ensemble classifier, further referred to as LRRpredictor is a soft voter aggregating eight classifiers C1-C8 (Figure 2d) which were trained to detect the LRR motif starting position-i.e., L 0 position from the minimalistic LRR motif 'L 0 xxL 3 xL 5 '. Finally, LRRpredictor was trained on the entire training set (all four CV sets) and tested on the test set which had been set aside. Assessment of LRRpredictor Performance The precision of LRRpredictor given by the fraction of true-positives (TP) predicted results over the sum of true-positives (TP) and false-positives (FP) varies between 89% and 97% on the test set and within cross-validation sets (Figure 3a). Similarly, the recall (also known as sensitivity), given by the fraction of TP over TP + false-negatives (FN) varies between 85% and 93%, while the F1-score (representing the harmonic mean between precision and recall) varies between 87% and 95% on the test set and cross-validation sets (Figure 3a,b). [26] and LRRsearch [27] (computed on their webservers using default parameters). As seen in Figure 4a, repeat lengths are rarely found outside the 19-35aa range, cases in which prediction becomes ambiguous. Too short repeats are improbable due to structural constraints and might indicate false positive predictions. Similarly, too long repeats-over 40 amino acids-could indicate either the presence of undetected repeats (false negatives) or cases in which an insertion or 'island' shapes up protruding the horseshoe structure (Figure 4a). Very large gaps between LRR motifs (more than 100 aa) were not included in computing the length distribution as these are rather indicating the presence of an inserted domain flanked by two LRR domains. We further analyzed the percent of the annotated LRR domain span that is covered by LRRpredictor and compared the predicted LRR motifs to LRRfinder [26] and LRRsearch [27] predictions and to the existing motif annotations from Interpro collection. In doing so we defined as predicted repeats motifs separated by 15-35 amino acids. Predicted motifs that superpose or cluster within 15 amino acids were counted only once, while when the distance between two motifs was higher than 35, the repeat was considered to be a potential terminal repeat or contain a domain break and the first 24 aa of such a stretch was assigned as a predicted repeat, given that this is the most frequent repeat length over the structural data. [26] and LRRsearch [27] (computed on their webservers using default parameters). LRRpredictor Behavior on Protein Families Containing LRR Domains As the available structural data is scarce, we further evaluated the extrapolation capabilities of LRRpredictor on a set of LRR domains annotated in Interpro collection. Groups of the most representative protein functional classes containing LRR domains were generated as follows: four groups from flowering plants-resistance proteins (CNL plants and TNL plants ) and extracellular receptors (RLK plants and RLP plants ) and two groups from vertebrates-NLR vert and TLR vert as described in Section 2. Selected sequences from each group were subjected to LRRpredictor motif detection. The repeat length distribution of the predicted LRR repeats (Figure 4a), is consistent with previously reported lengths within all protein groups of the seven type Kobe-Kajava (KK) classification [14,65]. The repeat length distribution of extracellular LRR domains (RLK plants , RLP plants , and TLR vert ) show a sharp peak at 24 amino acids, in agreement with the most frequent repeat length within plant-specific (PS) from KK classification [14,65]. As they often contain large helices over the dorsal side of the LRR horseshoe, vertebrate NLRs repeats have longer lengths (25-30 aa) as previously shown by the same classification, while plant NLRs (CNL plants and TNL plants ) have a larger distribution with a lower peak shaping up toward lower value side (20-24 aa) of repeat lengths range. (Figure 4b). For all six receptor classes analyzed herein, both LRRfinder and LRRsearch slightly increase the LRR coverage as compared to Interpro annotations especially in the case of extracellular receptors, with LRRsearch surpassing LRRfinder in the case of plant NLRs (Figure 4b). As also can be seen from Figure 4b, in comparison to Interpro and the two predictors mentioned above, LRRpredictor covers far larger regions of LRR domains with coverage percentages (CP) exceeding 60% and almost complete coverage in over 50% in all six groups (Figure 4b). It is interesting to note that Interpro annotation of extracellular LRR domains also include the N-terminal cap region, that is not formally a LRR repeat. This results in the fact that LRRpredictor covers in most cases only ≈90% of this domain, instead of 100% as in NLR groups (Figure 4b). Predicted Repeats Consensus in Each Class Further, the amino acid composition of the predicted LRR motifs was investigated solely on the 'core' predicted LRR repeats, i.e., repeats that are flanked by other predicted repeats within a 15-35 As seen in Figure 4a, repeat lengths are rarely found outside the 19-35aa range, cases in which prediction becomes ambiguous. Too short repeats are improbable due to structural constraints and might indicate false positive predictions. Similarly, too long repeats-over 40 amino acids-could indicate either the presence of undetected repeats (false negatives) or cases in which an insertion or 'island' shapes up protruding the horseshoe structure (Figure 4a). Very large gaps between LRR motifs (more than 100 aa) were not included in computing the length distribution as these are rather indicating the presence of an inserted domain flanked by two LRR domains. We further analyzed the percent of the annotated LRR domain span that is covered by LRRpredictor and compared the predicted LRR motifs to LRRfinder [26] and LRRsearch [27] predictions and to the existing motif annotations from Interpro collection. In doing so we defined as predicted repeats motifs separated by 15-35 amino acids. Predicted motifs that superpose or cluster within 15 amino acids were counted only once, while when the distance between two motifs was higher than 35, the repeat was considered to be a potential terminal repeat or contain a domain break and the first 24 aa of such a stretch was assigned as a predicted repeat, given that this is the most frequent repeat length over the structural data. The repeat coverage of analyzed LRR domains predicted by LRRpredictor, LRRfinder and LRRsearch were compared to the extent Interpro repeat annotations using a coverage percentage (CP) defined as the ratio between the sum of predicted/annotated repeat length vs the overall LRR domain length (Figure 4b). Plant NLRs from flowering plants show the lowest level of repeat annotation as 75% and 50% of CNL and TNL LRRs in Interpro lack any repeat annotation resulting in CP = 0% (Figure 4b). In comparison, repeats in vertebrate NLRs are better annotated in a CP ranging within 20-60% of the LRR domain. Even higher Interpro repeat annotations are shown by the extracellular plant and vertebrate receptors with CP ranging most frequently between 30% and 80% of the LRR domain size (Figure 4b). For all six receptor classes analyzed herein, both LRRfinder and LRRsearch slightly increase the LRR coverage as compared to Interpro annotations especially in the case of extracellular receptors, with LRRsearch surpassing LRRfinder in the case of plant NLRs (Figure 4b). As also can be seen from Figure 4b, in comparison to Interpro and the two predictors mentioned above, LRRpredictor covers far larger regions of LRR domains with coverage percentages (CP) exceeding 60% and almost complete coverage in over 50% in all six groups (Figure 4b). It is interesting to note that Interpro annotation of extracellular LRR domains also include the N-terminal cap region, that is not formally a LRR repeat. This results in the fact that LRRpredictor covers in most cases only ≈90% of this domain, instead of 100% as in NLR groups (Figure 4b). Predicted Repeats Consensus in Each Class Further, the amino acid composition of the predicted LRR motifs was investigated solely on the 'core' predicted LRR repeats, i.e., repeats that are flanked by other predicted repeats within a 15-35 aa range. In short, the results presented below clearly indicate that LRRpredictor is able to detect and reproduce all the consensus motifs previously defined for well-studied classes of RLKs, NLRs, and TLRs ( Figure 5). This is especially the case for vertebrate NLRs. The consensus follows the ribosomal inhibitor (RI) type -'x -3 xxL 0 xxL 3 xL 5 xx(N/C) 8 xL 10 xxxgoxxLxxoLxx' [14,65], with position '-3' being less relevant for this class of repeats. Additionally, the vertebrate TLRs predicted motif consensus matches the "T" type motif: L -3 xxL 0 xxL 3 xL 5 xxN 8 xL 10 xxL 13 xxxx(F/L) 18 xxL 21 xx defined in Matushima et al. classification [13] rather than the less encountered "S" type motif: L -3 xxL 0 xxL 3 xL 5 xxN 8 xL 10 xxL 13 Px(x)LPxx. In the case of plant extracellular receptors, the predicted motifs from RLK plant and RLP plant groups show a prolonged pattern that is in perfect agreement with the plant-specific (PS) type from Kobe and Kajava classification [14,65]-L -3 xxL 0 xxL 3 xL 5 xxN 8 xL 10 (S/T) 11 GxIPxxLxxLGx. Interestingly, the kinase containing receptors (RLK) have a more prominent consensus ( Figure 5). On the other hand, the predicted motifs in plant NLRs comprising the CNL plant and TNL plant groups display a remote similarity with the cysteine-containing (CC) type as defined by Kobe and Kajava classification [14,65]: "(C/L) -3 xxL 0 xxL 3 xL 5 xxC 8 xxITDxxOxxL(A/G) xx"-where O is any nonpolar residue. While the extended motif is satisfied (16 aa), a difference worth noting is that in both CNL and TNL groups cysteine is rare in position '-3' and outside this region any similarity with CC-type ends. Both plant NLR groups mainly confine their consensus to only the minimal L 0 xxL 3 xL 5 motif, with TNL extending it a little bit with C 8 position. By contrast, in plant extracellular receptors the consensus expands beyond the 16 amino acids of the 'extended' region covering all the four sides of the LRR solenoid. Despite being analogous in composition, the TNL plant group consensus is more pronounced, especially at positions C 8 and L 11 ( Figure 5). In all six classes L 0 , L 3 , and L 5 of minimal motif are as expected overwhelmingly hydrophobic, with all three positions occupied by leucine in around 50% of the cases, except CNLs where leucine occurrence seems less stringent ( Figure 6). When compared to the Kobe and Kajava classification, the majority of the motifs fall under the expected class and very few cross terms are seen between them ( Figure 6). While the extended motif is satisfied (16 aa), a difference worth noting is that in both CNL and TNL groups cysteine is rare in position '-3' and outside this region any similarity with CC-type ends. Both plant NLR groups mainly confine their consensus to only the minimal L0xxL3xL5 motif, with TNL extending it a little bit with C8 position. By contrast, in plant extracellular receptors the consensus expands beyond the 16 amino acids of the 'extended' region covering all the four sides of the LRR solenoid. Despite being analogous in composition, the TNLplant group consensus is more pronounced, especially at positions C8 and L11 ( Figure 5). In all six classes L0, L3, and L5 of minimal motif are as expected overwhelmingly hydrophobic, with all three positions occupied by leucine in around 50% of the cases, except CNLs where leucine occurrence seems less stringent ( Figure 6). When compared to the Kobe and Kajava classification, the majority of the motifs fall under the expected class and very few cross terms are seen between them ( Figure 6). CNLs and TNLs seem more dispersed even on a shorter 11 amino acid window consensus (W11), while the extracellular receptors obey in over 60% of the cases the corresponding W11 pattern that is shared simultaneous by all three classes ( Figure 6). [14] predicted with LRRpredictor across the six receptor classes. As the motif consensuses from KK classification were very strict, we adapted these consensuses to different sequences windows (W6, W11, W16, or more) centered around the minimal motif as shown in the table. Percentages of the predicted motifs compatible with each consensus are shown with grey bars. LRR Motifs Variability Across Classes Sequence variability is of critical importance for LRR domain function and in contrast to their common structural pattern, a wide spread in the sequence space is expected. To assess this, we analyzed the extended motifs, predicted by LRRpredictor, both the intra-and inter-group sequence similarity. This was performed over subsets of randomly selected 1000 examples of 'core' (L) motifs from each group. We selected as similarity measure a metric distance function [57] derived from BLOSUM scores which reflect the structural compatibility between amino acids, as described in Figure 6. Distribution of LRR motif types defined by Kobe and Kajava (KK) [14] predicted with LRRpredictor across the six receptor classes. As the motif consensuses from KK classification were very strict, we adapted these consensuses to different sequences windows (W6, W11, W16, or more) centered around the minimal motif as shown in the table. Percentages of the predicted motifs compatible with each consensus are shown with grey bars. CNLs and TNLs seem more dispersed even on a shorter 11 amino acid window consensus (W11), while the extracellular receptors obey in over 60% of the cases the corresponding W11 pattern that is shared simultaneous by all three classes ( Figure 6). LRR Motifs Variability Across Classes Sequence variability is of critical importance for LRR domain function and in contrast to their common structural pattern, a wide spread in the sequence space is expected. To assess this, we analyzed the extended motifs, predicted by LRRpredictor, both the intra-and inter-group sequence similarity. This was performed over subsets of randomly selected 1000 examples of 'core' (L) motifs from each group. We selected as similarity measure a metric distance function [57] derived from BLOSUM scores which reflect the structural compatibility between amino acids, as described in Section 2. Using this metric, we calculated the distance between each predicted LRR motif from all groups and analyzed how these distances behave intra-and inter-groups. Intra-group all-vs.-all distances distribution shows that the extracellular groups RLK, RLP from plants and TLRs from vertebrates form a denser group in terms of conservation, than plant and vertebrate NLRs (Figure 7a left). Figure 7b shows the silhouette coefficients. These scores show how separated two given clusters are, based on the distance between samples from each group, the maximal value of 1 corresponding to perfectly separated clusters, value 0 corresponds to clusters that coincide, while negative values with a minimum of -1 correspond to the case where samples from one group actually cluster better with the opposite group that is being compared. Silhouette coefficient of all versus all analyzed groups indicate that the NLR groups form a rather overlapping cluster, that has an increased variability among its sample motifs (i.e., expanded cluster) (Figure 7b left). Extracellular plant receptors RLK and RLP clusters are overlapping and have a more reduced span in terms of variability (i.e., more conserved motifs), while vertebrate TLR overlap plant RLK and RLP receptors have a slightly increased variability (Figure 7a,b left). Interestingly, within the minimal LRR motif region 'L 0 XXL 3 XL 5 ' there are no significant differences between groups (Figure 7a,b right). To have an overall view on the sequence dispersion in each protein class containing LRR domains Figure 7c shows the 2D embedding of the high dimensional sequence space of both the extended and minimal motifs of each class. Nonetheless such a reduction gives only a rough representation of distance relations between clusters in the original space as the normalized stress parameter (stress-1) of this 2D embedding is 0.25 and 0.21 for the extended and minimal motif space, respectively [66]. LRRpredictor Specificity Tested on Solenoid Architectures From a structural point of view the LRR protein architecture belongs to the larger class of solenoidal architectures which are defined by specific repeated structural patterns. Given the repetitiveness of such structures we asked if LRRpredictor is able to discriminate between LRR motifs and other repetitive sequence patterns. The main candidates considered for possible misclassifications are two classes of beta sheet repeat proteins-which are the closest structural To have an overall view on the sequence dispersion in each protein class containing LRR domains Figure 7c shows the 2D embedding of the high dimensional sequence space of both the extended and minimal motifs of each class. Nonetheless such a reduction gives only a rough representation of distance relations between clusters in the original space as the normalized stress parameter (stress-1) of this 2D embedding is 0.25 and 0.21 for the extended and minimal motif space, respectively [66]. LRRpredictor Specificity Tested on Solenoid Architectures From a structural point of view the LRR protein architecture belongs to the larger class of solenoidal architectures which are defined by specific repeated structural patterns. Given the repetitiveness of such structures we asked if LRRpredictor is able to discriminate between LRR motifs and other repetitive sequence patterns. The main candidates considered for possible misclassifications are two classes of beta sheet repeat proteins-which are the closest structural relatives of LRR domains: pectate lyases (PeLs) and trimeric LpxA architectures and two helical repetitive classes: armadillo and ankiryin architecture ( Figure A2b). To this end, 50 sequences from each of the above four classes annotated as such by Interpro were randomly selected from UniRef50. Figure A2a shows the probabilities returned by LRRpredictor that the potential motifs occurring in the 200 sequences are true LRR structural motifs. As can be seen in all four classes taken into account, the vast majority of potential motifs have a probability lower than 10% to be true motifs. Only 0.1% of such sites show a probability between 10% and 20% to be true motifs and none of these sites reaches a threshold of 40% for being a true LRR motif ( Figure A2a). From a technical point of view this result shows that LRRpredictor is highly specific for LRR domains. On the other hand this result is even more interesting from a structural and biological point of view indicating that even if LRRs and PeLs were considered to be members of the same LRR superfamily [67] the structural principles upon which they are built are different and presumably the two classes have diverged very early in evolution. Discussion Given the high number of indeterminacies generated by sequence variability, a proper annotation of LRR motifs and the correct delineation of repeats is critical in identifying potential protein-protein interaction sites of LRR domains. Here, we show that LRRpredictor is able to address this problem and by this, can be of use as a new tool in the analysis of especially plant NLR sequences that display a larger variability and irregularity as compared to other LRR domains [9,68]. This often results in the superposition or presence in less than a minimal repeat distance of potential alternative LRR motifs, as can be seen from Figure 8 illustrating such indeterminacies found on a 150 amino acid stretch from the potato CNL Gpa2 LRR domain. were considered to be members of the same LRR superfamily [67] the structural principles upon which they are built are different and presumably the two classes have diverged very early in evolution. Discussion Given the high number of indeterminacies generated by sequence variability, a proper annotation of LRR motifs and the correct delineation of repeats is critical in identifying potential protein-protein interaction sites of LRR domains. Here, we show that LRRpredictor is able to address this problem and by this, can be of use as a new tool in the analysis of especially plant NLR sequences that display a larger variability and irregularity as compared to other LRR domains [9,68]. This often results in the superposition or presence in less than a minimal repeat distance of potential alternative LRR motifs, as can be seen from Figure 8 illustrating such indeterminacies found on a 150 amino acid stretch from the potato CNL Gpa2 LRR domain. Given the scarcity of structural learning data consisting of less than 180 LRR structures with lower than 90% identity and only ≈850 motifs at hand in ID50 (<50% identity), in order to maximize LRRpredictor extrapolation abilities, the method was set to rely on aggregating a collection of eight classifiers based on different strategies, two of them designed to perform a massive oversampling of the real data ( Figure 2). In this context, LRRpredictor shows to perform well, with overall precision, recall, and F1 scores ranging between 85% and 97% on both test and cross validation sets (Figure 3a). In addition, LRRpredictor increases its performances when taking into account only the 'core' repeats (L), as the main prediction problems relate only to the N-'entry' repeats (N)-i.e., the first repeat of the LRR Figure 8. LRR motif and repeat indeterminacies onto a 150 aa stretch in Gpa2 potato NLR. Potential motifs that follow the minimal 'LxxLxL' pattern (where L is any hydrophobic amino acid) are illustrated above the sequence with black bars and yellow highlight, while LRRpredictor results are shown above with blue bars. Given the scarcity of structural learning data consisting of less than 180 LRR structures with lower than 90% identity and only ≈850 motifs at hand in ID50 (<50% identity), in order to maximize LRRpredictor extrapolation abilities, the method was set to rely on aggregating a collection of eight classifiers based on different strategies, two of them designed to perform a massive oversampling of the real data ( Figure 2). In this context, LRRpredictor shows to perform well, with overall precision, recall, and F1 scores ranging between 85% and 97% on both test and cross validation sets (Figure 3a). In addition, LRRpredictor increases its performances when taking into account only the 'core' repeats (L), as the main prediction problems relate only to the N-'entry' repeats (N)-i.e., the first repeat of the LRR domain (Figure 3a, Table A1). This can be explained in part by the increased irregularity of the sequence in this region, but also by the small sample size of the N-'entry' (N) motifs when compared to the 'core' (L) motifs. It is also important to note here the fact that false positives are almost never found in nonLRR domains but always in proteins containing LRR domains (Table A1). Here, such false positives shape up in close vicinity to the marginal repeats-where the LRR motif characteristics are more diffuse, or in linkers or different domains neighboring the LRR, but found in a 'one repeat range' to the N-entry motif. Other false predictions are caused by alignment artefacts. These yield to an offset of 1-3 amino acids in the predicted LRR motif starting position. Alignment artefacts are also frequently seen in regions with high beta structure propensity of insertion loops or 'islands' protruding from the LRR domain structures. This is mainly due to the fact that the multiple alignment on which PSSM relies forces the protruding loop in the queried sequence to align to regular repeats in the template LRRs of the database. Unfortunately, the number of such insertion loops or 'islands' is so small in ID90/ID50 that estimators cannot learn from the existing data to discriminate such false positives. Thus, only careful structural analysis performed in later modelling stages can handle such cases. Results on both cross-validation and test sets show that estimators using structural features in addition to the sequence based features (C5-C8) perform on average only slightly better compared to those sequence based only (C1-C4), with some interesting improvements on F1 scores (Figure 3b). This only marginal improvement may indicate that RaptorX-Property [34] training on the overall structural database that might have marginally overlapped with our testing dataset did not affect the results. Nevertheless, C5-C8 are expected to be better extrapolators (Table A1), while the structural predictions on which (C5-C8) are based, and that are present in the output file can prove instrumental in further dealing with ambiguous cases where two LRR motif signatures partially superpose or are within the limit of a repeat. Figures 3c and 4b compare predictions of three existing engines. LRRpredictor outperforms LRRsearch [27] and LRRfinder [26]. This is expected as the two previous methods were designed to focus mainly on specific LRR classes such as vertebrate TLRs or NLRs, respectively, while LRRpredictor relies on a newer larger dataset and was designed to identify LRR motifs in general. However, despite focusing on specific protein classes, both LRRsearch and LRRfinder show comparable efficiency in covering annotated LRR domains in plant extracellular receptors but decreased capabilities on plant NLRs (CNLs and TNLs) (Figure 4b). Furthermore, both LRRsearch and LRRfinder were intended for fast computation and they use a predefined PSSM matrix computed on a curated collection of LRR domains, instead of performing case by case basis sequence profiles as our method does. However, the increased performance of LRRpredictor comes with an attached computational cost and is not easily scalable for scanning large protein sequences databases such UniprotKB. The main reason for this is that generating case by case sequence profiles and performing predictions for each estimator aggregated in LRRpredictor is more computationally demanding than LRRfinder and LRRsearch workflow. Another matter of concern was related to the phyla bias of the database on which LRRpredictor relies, as ≈50% of ID90 have mammalian origin while the share of mammalian-from total-annotated LRRs in UniRef50 is only 2% (Figure 1c). Moreover, groups such as plant NLRs are extremely poorly represented, as only very recently the first plant NLR structure was reported [21,22]. In this context, in order to investigate the extrapolation capabilities of LRRpredictor we used a set of LRR domains annotated in Interpro collection from the six most representative immune receptor classes: R-proteins and extracellular receptors from flowering plants (CNL plants , TNL plants , RLK plants , RLP plants ) and their vertebrate counterparts (NLR vert , TLR vert ). The LRR motifs predicted by LRRpredictor show a good coverage of the LRR domains annotated by Interpro and follow the expected repeat length distribution for all these six classes [14,65] (Figure 4a,b). Moreover, the predicted motifs reproduce the expected LRR motif consensus of each protein class ( Figure 5) from Kobe and Kajava classification [14,65]. Combined, these indicate that LRRpredictor is able to extrapolate well in different LRR motif classes which is especially important for plant NLRs. Analysis of LRRpredictor detected motifs showed clear differences between the six classes within the extended 16 aa motif. Whether variation in these extended motifs directly relate to the functional diversification of the different receptor classes still remains to be addressed. By contrast, within the minimal 6 aa LRR motif region-L 0 XXL 3 XL 5 -there are no significant differences between the six groups ( Figure 7). This might suggest a common root of minimal structural criteria imposed by the solenoidal architecture from which the six classes have diverged to fulfil specific tasks in specific environments. For receptor function, such a solenoidal domain organization in which only three positions over a ≈25 repeat length are loosely conserved has two-fold evolutionary advantages: first the solenoid architecture ensures a large solvent exposed surface area [10] and second a high sequence variability can be achieved without disturbing the tertiary structure. The increased conservation seen at the level of the extended motif among all three extracellular LRR classes-plant RLK, RLP and vertebrate TLRs-when compared to plant and vertebrate NLRs could be related to N-glycosylation and the constraints imposed by the extracellular environment. On the one hand, plant NLRs recognize directly or indirectly a suite of pathogen effectors or (perturbations) of their host targets conferring host specific immunity. Single amino acid changes in the effector can already be detected or are sufficient to evade recognition by a NLR, resulting in a co-evolutionary arms race between pathogen effectors and host immune receptor [69,70]. In contrast, extracellular LRRs recognize often conserved microbial patterns to confer basal immunity thus lacking such a strong driver for diversification [71]. Vertebrate NLRs act more like basal immune receptors in innate immunity, recognizing conserved microbe-associated molecular patterns (MAMPs). The greater diversifying selection imposed by fast-evolving effectors may therefore account for the co-evolution of structurally highly variable LRR motifs in plant NLRs. In the future, it will be interesting to relate our LRR structural annotations to specific functional NLR sub-classes. This is relevant, for instance, as some plant NLRs-types are described to have a downstream 'helper' function rather than a role as a canonical 'sensor'. Another aspect is that LRR domains in plant NLRs have a dual role. They not only contribute to pathogen recognition, but also negatively regulate the switch function [72]. Hence, it will be interesting to link LRR structural annotations to specific intramolecular domain interactions between LRRs and other NLR subdomains to better understand the co-evolution of protein domains in NLRs. It is shown that subtle mutations in the interface between LRR and NB-ARC can have a major effect on NLR functioning, often resulting in constitutive immune activation or a complete loss-of-function [17,72]. This shows the tight link between structural and functional constraints underlying the shaping of NLRs in plants. Additionally, the link between LRR structural annotations and complex formation with other host proteins will be interesting to assess. LRR domains are known to interact with other components like chaperones (e.g., SGT1) which are required for proper NLR folding and functioning [73], or kinases (e.g., ZED1, RKS1) [21,22,74] but also NLR hetero-and homodimers are often formed [9,75,76] which could impose additional structural constraints on the shape and irregularity of LRR domains in plant NLRs. Conclusions The results presented herein indicate that LRRpredictor shows a good performance on the available 3D data and good extrapolation capabilities on plant NLRs (CNL/TNL), which are poorly represented in the training dataset. Predicted LRR repeats using LRRpredictor significantly increase the coverage of Interpro annotated LRR domains from main immune receptors groups. In addition, these predicted repeats are consistent with previously defined motif consensuses from all studied groups and also follow the repeat length range specific to each class. In conclusion, LRRpredictor is a tool worth using in research topics related to understanding immune receptors functions and structure-informed strategies for pathogen control technologies. loss of samples (left) and increase in entropy (right) at different identity thresholds. Displayed is the Shannon entropy averaged over the 16 amino acid extended motif. (b) Composition of LRR motif types: N-entry (N), core (L), and C-exit (C) LRR motifs in the initial set (ID90) and in the 50% identity trimmed dataset (ID50). (c) LRR repeat length distribution at different identity thresholds. C-exit motifs were not used. (d) Structural superposition of the LRR repeats from a plant NLR, vertebrate NLR, plant RLK and vertebrate TLR structures [22,[77][78][79] (from left to right). Hydrophobic positions of the minimal 'L0xxL3xL5' motif are shown in orange and position N8 in purple. Table A1. Detailed performance analysis on LRRpredictor and its classifiers. Precision recall and F1 scores are shown for in-sample (i.e., training data) and out-of-sample data (i.e., test data that was not used in training), for both cross-validation and test phase. In the cross-validation stage, classifiers were trained on three of the cross-validation (CV) sets and tested on the fourth set in an iterative manner, while in testing stage, classifiers were trained on all four CV sets and evaluated on the test set left aside from the beginning. The counts for true-negatives (TN), false-positives (FP), false-negative (FN), and true-positive (TP) within each set are also shown. As marginal repeats (N-entry and C-exit types) have a lower detection rate, also included is the recall calculated only with respect to 'core' repeats (L), indicated with blue font. Performance scores shading is according to a value based colormap from yellow (0.75) to blue (1.00). Dataset Classifier In Figure 1 . 1Available leucine-rich-repeat (LRR) domains in structural data. (a) LRR structural dataset construction. (b) LRR domain horseshoe architecture illustrated on the only plant NLR cryo-EM structure available-ZAR1-from Arabidopsis thaliana (left) and zoom-in view of a LRR repeat (right) (PDB: 6J5W). The hydrophobic positions in the minimal 'L0xxL3xL5' motif are shown in orange. The first N-entry repeat (blue) and the last C-exit repeat (red) are also mapped on the structure. (c) Phyla distribution of the initial LRR motif set ID90, the 50% identity trimmed LRR motifs set (ID50), annotated LRR proteins and all proteins from the UniRef50 database (from left to right). Percent values corresponding to the mammals group are shown in red. (d) Frequency plot of amino acid composition of the N-entry, core and C-exit motifs on the 50% identity trimmed set. Amino acids are colored according to their properties as follows: hydrophobic (yellow), acidic (red), basic (blue), asparagine and glutamine (purple), proline and glycine (green), others (black). (e) Jensen-Shannon Figure 1 . 1Available leucine-rich-repeat (LRR) domains in structural data. (a) LRR structural dataset construction. (b) LRR domain horseshoe architecture illustrated on the only plant NLR cryo-EM structure available-ZAR1-from Arabidopsis thaliana (left) and zoom-in view of a LRR repeat (right) L -strictly leucine; Ł-hydrophobic without leucine (I, V, M, F, W, Y, C, A); L-hydrophobic (L, I, V, M, F, W, Y, C, A); x-any amino acid. Figure 2 . 2LRRpredictor training and testing workflow: (a) training and testing dataset construction. (b) schematic representation of the training and testing procedure, (c) selected features, and (d) selected classifiers aggregated into LRRpredictor. Figure 2 . 2LRRpredictor training and testing workflow: (a) training and testing dataset construction. (b) schematic representation of the training and testing procedure, (c) selected features, and (d) selected classifiers aggregated into LRRpredictor. Figure 3 . 3LRRpredictor performance analysis: (a) LRRpredictor performance across datasets: precision, recall, and F1 scores are shown either considering all the LRR motif types (N-entry, core, and C-exit types), either solely core motifs (L); also shown are the true negative (TN), false positive (FP), false negative (FN) and true positive (TP) counts. (b) F1 scores of LRRpredictor and its individual classifiers. (c) Comparison between LRRpredictor and other LRR motif predictors: LRRfinder Figure 3 . 3LRRpredictor performance analysis: (a) LRRpredictor performance across datasets: precision, recall, and F1 scores are shown either considering all the LRR motif types (N-entry, core, and C-exit types), either solely core motifs (L); also shown are the true negative (TN), false positive (FP), false negative (FN) and true positive (TP) counts. (b) F1 scores of LRRpredictor and its individual classifiers. (c) Comparison between LRRpredictor and other LRR motif predictors: LRRfinder Figure 4 . 4LRRpredictor behavior on Interpro annotated LRR domains from different classes. (a) Length distribution of the predicted repeats using LRRpredictor within each protein group. Cterminal motifs were not used in computing the distribution. Repeat lengths size prone to ambiguity-i.e., either too short (potential FP) or too long (potential FN)-are shaded in red. (b) Distributions of the Interpro annotated LRR domain length that is covered by Interpro LRR repeat annotations (grey) or by predicted repeats using LRRpredictor (blue), LRRfinder (green), and LRRsearch (purple). Coverage percent distributions are shown within each protein group. Figure 4 . 4LRRpredictor behavior on Interpro annotated LRR domains from different classes. (a) Length distribution of the predicted repeats using LRRpredictor within each protein group. C-terminal motifs were not used in computing the distribution. Repeat lengths size prone to ambiguity-i.e., either too short (potential FP) or too long (potential FN)-are shaded in red. (b) Distributions of the Interpro annotated LRR domain length that is covered by Interpro LRR repeat annotations (grey) or by predicted repeats using LRRpredictor (blue), LRRfinder (green), and LRRsearch (purple). Coverage percent distributions are shown within each protein group. Figure 5 . 5Consensuses of the LRR motifs predicted by LRRpredictor across different classes. Logo heights correspond to amino acid relative entropy (in bits), higher heights implying higher conservation. A consensus for each class is displayed bellow each logo, highly conserved positions being shown in black boxes, while less conserved in gray. Minimal motif 'L0xxL3xL5' (green line) and the extended motif (black line) are indicated below each logo. Amino acids are colored according to their properties as inFigure 1d. Figure 5 . 5Consensuses of the LRR motifs predicted by LRRpredictor across different classes. Logo heights correspond to amino acid relative entropy (in bits), higher heights implying higher conservation. A consensus for each class is displayed bellow each logo, highly conserved positions being shown in black boxes, while less conserved in gray. Minimal motif 'L 0 xxL 3 xL 5 ' (green line) and the extended motif (black line) are indicated below each logo. Amino acids are colored according to their properties as inFigure 1d.Genes 2019, 10, x FOR PEER REVIEW 13 of 24 Figure 6 . 6Distribution of LRR motif types defined byKobe and Kajava (KK) Genes 2019 , 201910, x FOR PEER REVIEW 14 of 24 Figure 7 . 7LRR motifs variability in different protein classes. (a) Intra-group all-vs.-all distances on the extended (left) and minimal (right) motif (b) Silhouette coefficients inter-groups extended (left) and minimal (right) motif. (c) Histogram of a 2D embedding approximating the true distances between points for the extended (left) and minimal (right) motif. Histograms were computed using a 20 × 20 bins grid. Extended and minimal motif histograms cannot be compared as they refer to different sequence spaces. Figure 7 . 7LRR motifs variability in different protein classes. (a) Intra-group all-vs.-all distances on the extended (left) and minimal (right) motif (b) Silhouette coefficients inter-groups extended (left) and minimal (right) motif. (c) Histogram of a 2D embedding approximating the true distances between points for the extended (left) and minimal (right) motif. Histograms were computed using a 20 × 20 bins grid. Extended and minimal motif histograms cannot be compared as they refer to different sequence spaces. Genes 2019 , 201910, x FOR PEER REVIEW 15 of 24 Figure 8 . 8LRR motif and repeat indeterminacies onto a 150 aa stretch in Gpa2 potato NLR. Potential motifs that follow the minimal 'LxxLxL' pattern (where L is any hydrophobic amino acid) are illustrated above the sequence with black bars and yellow highlight, while LRRpredictor results are shown above with blue bars. Figure A1 . A1Available LRR domains structural data. (a) Identity cut-off versus loss of data: plots of the loss of samples (left) and increase in entropy (right) at different identity thresholds. Displayed is the Shannon entropy averaged over the 16 amino acid extended motif. (b) Composition of LRR motif types: N-entry (N), core (L), and C-exit (C) LRR motifs in the initial set (ID90) and in the 50% identity trimmed dataset (ID50). (c) LRR repeat length distribution at different identity thresholds. C-exit motifs were not used. (d) Structural superposition of the LRR repeats from a plant NLR, vertebrate NLR, plant RLK and vertebrate TLR structures[22,[77][78][79] (from left to right). Hydrophobic positions of the minimal 'L0xxL3xL5' motif are shown in orange and position N8 in purple. Figure A1 . A1Available LRR domains structural data. (a) Identity cut-off versus loss of data: plots of the loss of samples (left) and increase in entropy (right) at different identity thresholds. Displayed is the Shannon entropy averaged over the 16 amino acid extended motif. (b) Composition of LRR motif types: N-entry (N), core (L), and C-exit (C) LRR motifs in the initial set (ID90) and in the 50% identity trimmed dataset (ID50). (c) LRR repeat length distribution at different identity thresholds. C-exit motifs were not used. (d) Structural superposition of the LRR repeats from a plant NLR, vertebrate NLR, plant RLK and vertebrate TLR structures[22,[77][78][79] (from left to right). Hydrophobic positions of the minimal 'L 0 xxL 3 xL 5 ' motif are shown in orange and position N 8 in purple. Figure A2 . A2(a) LRRpredictor behavior on other solenoidal architectures. Shown are occurrence counts of LRR-like patterns versus LRRpredictor probabilities counts histogram. (b) Overall 3D structure of the four examined classes of solenoidal proteins[80][81][82][83]. Figure A2 . A2(a) LRRpredictor behavior on other solenoidal architectures. Shown are occurrence counts of LRR-like patterns versus LRRpredictor probabilities counts histogram. (b) Overall 3D structure of the four examined classes of solenoidal proteins[80][81][82][83]. Table 1 . 1Occurrence of LRR sequence patterns in the overall dataset used to train the machine learning (ML) estimators. Training & Testing Dataset (CV 1-4 and Test Sets)Full NonLRR Proteins LRR Proteins Table A1 . A1Cont.Dataset Classifier In-Sample Out-Of Sample Precision Recall F1-Score Precision Recall F1-Score TN FP FN TP Recall on Core(L) Only - Non-LRR Proteins LRR Proteins N-Entry (N) Core (L) C-Exit (C) N-Entry (N) Core (L) C-Exit (C) Test C1 0.900 0.997 0.946 0.874 0.880 0.877 35107 0 19 6 10 2 13 109 10 0.916 C2 0.962 0.956 0.959 0.941 0.847 0.891 35118 0 8 8 12 3 11 107 9 0.899 C3 0.882 0.896 0.889 0.852 0.847 0.850 35104 1 21 8 13 2 11 106 10 0.891 C4 0.940 0.874 0.906 0.907 0.840 0.872 35113 1 12 8 14 2 11 105 10 0.882 C5 0.895 0.934 0.914 0.862 0.873 0.868 35105 1 20 7 10 2 12 109 10 0.916 C6 0.940 0.901 0.920 0.928 0.860 0.893 35116 0 10 7 12 2 12 107 10 0.899 C7 0.990 1.000 0.995 0.827 0.827 0.827 35100 4 22 7 16 3 12 103 9 0.866 C8 0.942 0.874 0.906 0.920 0.840 0.878 35115 0 11 9 13 2 10 106 10 0.891 LRRpredictor 0.943 0.928 0.936 0.928 0.860 0.893 35116 0 10 7 12 2 12 107 10 0.899 Structural Principles of Leucine-Rich Repeat (LRR) Proteins. P Enkhbayar, M Kamiya, M Osaki, T Matsumoto, N Matsushima, 10.1002/prot.10605Proteins Struct. Funct. Bioinform. 54PubMedEnkhbayar, P.; Kamiya, M.; Osaki, M.; Matsumoto, T.; Matsushima, N. Structural Principles of Leucine-Rich Repeat (LRR) Proteins. Proteins Struct. Funct. Bioinform. 2004, 54, 394-403. [CrossRef] [PubMed] A mutation within the leucine-rich repeat domain of the arabidopsis disease resistance gene RPS5 partially suppresses multiple bacterial and downy mildew resistance genes. R F Warren, A Henk, P Mowery, E Holub, R W Innes, 10.1105/tpc.10.9.1439Plant Cell. 10PubMedWarren, R.F.; Henk, A.; Mowery, P.; Holub, E.; Innes, R.W. A mutation within the leucine-rich repeat domain of the arabidopsis disease resistance gene RPS5 partially suppresses multiple bacterial and downy mildew resistance genes. Plant Cell 1998, 10, 1439-1452. [CrossRef] [PubMed] Direct interaction of resistance gene and avirulence gene products confers rice blast resistance. Y Jia, S A Mcadams, G T Bryan, H P Hershey, B Valent, 10.1093/emboj/19.15.4004EMBO J. 19PubMedJia, Y.; McAdams, S.A.; Bryan, G.T.; Hershey, H.P.; Valent, B. Direct interaction of resistance gene and avirulence gene products confers rice blast resistance. EMBO J. 2000, 19, 4004-4014. [CrossRef] [PubMed] Autoinflammatory mutation in NLRC4 reveals a leucine-rich repeat (LRR)-LRR oligomerization interface. F Moghaddas, P Zeng, Y Zhang, H Schützle, S Brenner, S R Hofmann, R Berner, Y Zhao, B Lu, X Chen, 10.1016/j.jaci.2018.04.033J. Allergy Clin. Immunol. 142PubMedMoghaddas, F.; Zeng, P.; Zhang, Y.; Schützle, H.; Brenner, S.; Hofmann, S.R.; Berner, R.; Zhao, Y.; Lu, B.; Chen, X.; et al. Autoinflammatory mutation in NLRC4 reveals a leucine-rich repeat (LRR)-LRR oligomerization interface. J. Allergy Clin. Immunol. 2018, 142, 1956-1967.e6. [CrossRef] [PubMed] Cloning of the gene containing mutations that cause PARK8-linked Parkinson's disease. C Paisán-Ruíz, S Jain, E W Evans, W P Gilks, J Simón, M Van Der Brug, A L De Munain, S Aparicio, A M Gil, N Khan, 10.1016/j.neuron.2004.10.023Neuron. 44PubMedPaisán-Ruíz, C.; Jain, S.; Evans, E.W.; Gilks, W.P.; Simón, J.; Van Der Brug, M.; De Munain, A.L.; Aparicio, S.; Gil, A.M.; Khan, N.; et al. Cloning of the gene containing mutations that cause PARK8-linked Parkinson's disease. Neuron 2004, 44, 595-600. [CrossRef] [PubMed] Mutations in LRRK2 cause autosomal-dominant parkinsonism with pleomorphic pathology. A Zimprich, S Biskup, P Leitner, P Lichtner, M Farrer, S Lincoln, J Kachergus, M Hulihan, R J Uitti, D B Calne, 10.1016/j.neuron.2004.11.005Neuron. 44PubMedZimprich, A.; Biskup, S.; Leitner, P.; Lichtner, P.; Farrer, M.; Lincoln, S.; Kachergus, J.; Hulihan, M.; Uitti, R.J.; Calne, D.B.; et al. Mutations in LRRK2 cause autosomal-dominant parkinsonism with pleomorphic pathology. Neuron 2004, 44, 601-607. [CrossRef] [PubMed] The role of small leucine-rich proteoglycans in osteoarthritis pathogenesis. G X Ni, Z Li, Y Z Zhou, 10.1016/j.joca.2014.04.026Osteoarthritis Cartilage. 22Ni, G.X.; Li, Z.; Zhou, Y.Z. The role of small leucine-rich proteoglycans in osteoarthritis pathogenesis. Osteoarthritis Cartilage. 2014, 22, 896-903. [CrossRef] . M Lewis, Collagen Prelp, Hutchinson-Gilford Progeria, 10.1016/S1568-1637(02)00044-2Ageing Res. Rev. 2Lewis, M. PRELP, collagen, and a theory of Hutchinson-Gilford progeria. Ageing Res. Rev. 2003, 2, 95-105. [CrossRef] Structure-informed insights for NLR functioning in plant immunity. O C A Sukarta, E J Slootweg, A Goverse, 10.1016/j.semcdb.2016.05.012Semin. Cell Dev. Biol. 56Sukarta, O.C.A.; Slootweg, E.J.; Goverse, A. Structure-informed insights for NLR functioning in plant immunity. Semin. Cell Dev. Biol. 2016, 56, 134-149. [CrossRef] The NBS-LRR architectures of plant R-proteins and metazoan NLRs evolved in independent events. J M Urbach, F M Ausubel, 10.1073/pnas.1619730114Proc. Natl. Acad. Sci. Natl. Acad. SciUSA114Urbach, J.M.; Ausubel, F.M. The NBS-LRR architectures of plant R-proteins and metazoan NLRs evolved in independent events. Proc. Natl. Acad. Sci. USA 2017, 114, 1063-1068. [CrossRef] Comparative Geometrical Analysis of Leucine-Rich Repeat Structures in the Nod-Like and Toll-Like Receptors in Vertebrate Innate Immunity. N Matsushima, H Miyashita, P Enkhbayar, R H Kretsinger, 10.3390/biom5031955Biomolecules. 5PubMedMatsushima, N.; Miyashita, H.; Enkhbayar, P.; Kretsinger, R.H. Comparative Geometrical Analysis of Leucine-Rich Repeat Structures in the Nod-Like and Toll-Like Receptors in Vertebrate Innate Immunity. Biomolecules 2015, 5, 1955-1978. [CrossRef] [PubMed] Assessment of the ability to model proteins with leucine-rich repeats in light of the latest structural information. A V Kajava, B Kobe, 10.1110/ps.4010102Protein Sci. 11Kajava, A.V.; Kobe, B. Assessment of the ability to model proteins with leucine-rich repeats in light of the latest structural information. Protein Sci. 2002, 11, 1082-1090. [CrossRef] Comparative sequence analysis of leucine-rich repeats (LRRs) within vertebrate toll-like receptors. N Matsushima, T Tanaka, P Enkhbayar, T Mikami, M Taga, K Yamada, Y Kuroki, 10.1186/1471-2164-8-124BMC Genom. 8PubMedMatsushima, N.; Tanaka, T.; Enkhbayar, P.; Mikami, T.; Taga, M.; Yamada, K.; Kuroki, Y. Comparative sequence analysis of leucine-rich repeats (LRRs) within vertebrate toll-like receptors. BMC Genom. 2007, 8, 124. [CrossRef] [PubMed] The leucine-rich repeat as a protein recognition motif. B Kobe, A V Kajava, 10.1016/S0959-440X(01)00266-4Curr. Opin. Struct. Biol. 11Kobe, B.; Kajava, A.V. The leucine-rich repeat as a protein recognition motif. Curr. Opin. Struct. Biol. 2001, 11, 725-732. [CrossRef] Human leucine-rich repeat proteins: A genome-wide bioinformatic categorization and functional analysis in innate immunity. A C Y Ng, J M Eisenberg, R J W Heath, A Huett, C M Robinson, G J Nau, R J Xavier, 10.1073/pnas.1000093107Proc. Natl. Acad. Sci. Natl. Acad. SciUSA108PubMedNg, A.C.Y.; Eisenberg, J.M.; Heath, R.J.W.; Huett, A.; Robinson, C.M.; Nau, G.J.; Xavier, R.J. Human leucine-rich repeat proteins: A genome-wide bioinformatic categorization and functional analysis in innate immunity. Proc. Natl. Acad. Sci. USA 2011, 108, 4631-4638. [CrossRef] [PubMed] Ancient diversity of splicing motifs and protein surfaces in the wild emmer wheat (Triticum dicoccoides) LR10 coiled coil (CC) and leucine-rich repeat (LRR) domains. H Sela, L N Spiridon, A J Petrescu, M Akerman, Y Mandel-Gutfreund, E Nevo, C Loutre, B Keller, A H Schulman, T Fahima, 10.1111/j.1364-3703.2011.00744.xMol. Plant Pathol. 13Sela, H.; Spiridon, L.N.; Petrescu, A.J.; Akerman, M.; Mandel-Gutfreund, Y.; Nevo, E.; Loutre, C.; Keller, B.; Schulman, A.H.; Fahima, T. Ancient diversity of splicing motifs and protein surfaces in the wild emmer wheat (Triticum dicoccoides) LR10 coiled coil (CC) and leucine-rich repeat (LRR) domains. Mol. Plant Pathol. 2012, 13, 276-287. [CrossRef] Structural determinants at the interface of the ARC2 and leucine-rich repeat domains control the activation of the plant immune receptors Rx1 and Gpa2. E J Slootweg, L N Spiridon, J Roosien, P Butterbach, R Pomp, L Westerhof, R Wilbers, E Bakker, J Bakker, A J Petrescu, 10.1104/pp.113.218842Plant Physiol. 162Slootweg, E.J.; Spiridon, L.N.; Roosien, J.; Butterbach, P.; Pomp, R.; Westerhof, L.; Wilbers, R.; Bakker, E.; Bakker, J.; Petrescu, A.J.; et al. Structural determinants at the interface of the ARC2 and leucine-rich repeat domains control the activation of the plant immune receptors Rx1 and Gpa2. Plant Physiol. 2013, 162, 1510-1528. [CrossRef] Three-Dimensional Modeling and Diversity Analysis Reveals Distinct AVR Recognition Sites and Evolutionary Pathways in Wild and Domesticated Wheat Pm3 R Genes. H Sela, L N Spiridon, H Ashkenazi, N K Bhullar, S Brunner, A.-J Petrescu, T Fahima, B Keller, T Jordan, 10.1094/MPMI-01-14-0009-RMol. Plant-Microbe Interact. 27Sela, H.; Spiridon, L.N.; Ashkenazi, H.; Bhullar, N.K.; Brunner, S.; Petrescu, A.-J.; Fahima, T.; Keller, B.; Jordan, T. Three-Dimensional Modeling and Diversity Analysis Reveals Distinct AVR Recognition Sites and Evolutionary Pathways in Wild and Domesticated Wheat Pm3 R Genes. Mol. Plant-Microbe Interact. 2014, 27, 835-845. [CrossRef] An LRR/Malectin receptor-like kinase mediates resistance to non-adapted and adapted powdery mildew fungi in barley and wheat. J Rajaraman, D Douchkov, G Hensel, F L Stefanato, A Gordon, N Ereful, O F Caldararu, A J Petrescu, J Kumlehn, L A Boyd, 10.3389/fpls.2016.01836Front. Plant Sci. 7Rajaraman, J.; Douchkov, D.; Hensel, G.; Stefanato, F.L.; Gordon, A.; Ereful, N.; Caldararu, O.F.; Petrescu, A.J.; Kumlehn, J.; Boyd, L.A.; et al. An LRR/Malectin receptor-like kinase mediates resistance to non-adapted and adapted powdery mildew fungi in barley and wheat. Front. Plant Sci. 2016, 7, 1836. [CrossRef] Structure-function analysis of ZAR1 immune receptor reveals key molecular interactions for activity. M Baudin, K J Schreiber, E C Martin, A J Petrescu, J D Lewis, 10.1111/tpj.14547Plant J. 101Baudin, M.; Schreiber, K.J.; Martin, E.C.; Petrescu, A.J.; Lewis, J.D. Structure-function analysis of ZAR1 immune receptor reveals key molecular interactions for activity. Plant J 2020, 101, 352-370. [CrossRef] Reconstitution and structure of a plant NLR resistosome conferring immunity. J Wang, M Hu, J Wang, J Qi, Z Han, G Wang, Y Qi, H.-W Wang, J.-M Zhou, J Chai, 10.1126/science.aav5870Science. 364Wang, J.; Hu, M.; Wang, J.; Qi, J.; Han, Z.; Wang, G.; Qi, Y.; Wang, H.-W.; Zhou, J.-M.; Chai, J. Reconstitution and structure of a plant NLR resistosome conferring immunity. Science 2019, 364, eaav5870. [CrossRef] Ligand-triggered allosteric ADP release primes a plant NLR complex. J Wang, J Wang, M Hu, S Wu, J Qi, G Wang, Z Han, Y Qi, N Gao, H W Wang, 10.1126/science.aav5868Science. 364eaav5868. [CrossRefWang, J.; Wang, J.; Hu, M.; Wu, S.; Qi, J.; Wang, G.; Han, Z.; Qi, Y.; Gao, N.; Wang, H.W.; et al. Ligand-triggered allosteric ADP release primes a plant NLR complex. Science 2019, 364, eaav5868. [CrossRef] Modeling of the three-dimensional structure of proteins with the typical leucine-rich repeats. A V Kajava, G Vassart, S J Wodak, 10.1016/S0969-2126(01)00222-2Structure. 3Kajava, A.V.; Vassart, G.; Wodak, S.J. Modeling of the three-dimensional structure of proteins with the typical leucine-rich repeats. Structure 1995, 3, 867-877. [CrossRef] LRR conservation mapping to predict functional sites within protein leucine-rich repeat domains. L Helft, V Reddy, X Chen, T Koller, L Federici, J Fernández-Recio, R Gupta, A Bent, 10.1371/journal.pone.0021614PLoS ONE. 6Helft, L.; Reddy, V.; Chen, X.; Koller, T.; Federici, L.; Fernández-Recio, J.; Gupta, R.; Bent, A. LRR conservation mapping to predict functional sites within protein leucine-rich repeat domains. PLoS ONE 2011, 6, e21614. [CrossRef] Out of Water: The Origin and Early Diversification of Plant R-Genes. Y Gao, W Wang, T Zhang, Z Gong, H Zhao, G Z Han, 10.1104/pp.18.00185Plant Physiol. 177Gao, Y.; Wang, W.; Zhang, T.; Gong, Z.; Zhao, H.; Han, G.Z. Out of Water: The Origin and Early Diversification of Plant R-Genes. Plant Physiol. 2018, 177, 82-89. [CrossRef] 0: A webserver for the prediction of leucine-rich repeats. V Offord, D Werling, Lrrfinder2, 10.1177/1753425912465661Innate Immun. 19Offord, V.; Werling, D. LRRfinder2.0: A webserver for the prediction of leucine-rich repeats. Innate Immun. 2013, 19, 398-402. [CrossRef] LRRsearch: An asynchronous server-based application for the prediction of leucine-rich repeat motifs and an integrative database of NOD-like receptors. A Bej, B R Sahoo, B Swain, M Basu, P Jayasankar, M Samanta, 10.1016/j.compbiomed.2014.07.016Comput. Biol. Med. 53Bej, A.; Sahoo, B.R.; Swain, B.; Basu, M.; Jayasankar, P.; Samanta, M. LRRsearch: An asynchronous server-based application for the prediction of leucine-rich repeat motifs and an integrative database of NOD-like receptors. Comput. Biol. Med. 2014, 53, 164-170. [CrossRef] CATH: An expanded resource to predict protein function through structure and sequence. N L Dawson, T E Lewis, S Das, J G Lees, D Lee, P Ashford, C A Orengo, I Sillitoe, 10.1093/nar/gkw1098Nucleic Acids Res. 45Dawson, N.L.; Lewis, T.E.; Das, S.; Lees, J.G.; Lee, D.; Ashford, P.; Orengo, C.A.; Sillitoe, I. CATH: An expanded resource to predict protein function through structure and sequence. Nucleic Acids Res. 2017, 45, D289-D295. [CrossRef] The Pfam protein families database in 2019. S El-Gebali, J Mistry, A Bateman, S R Eddy, A Luciani, S C Potter, M Qureshi, L J Richardson, G A Salazar, A Smart, 10.1093/nar/gky995Nucleic Acids Res. 47El-Gebali, S.; Mistry, J.; Bateman, A.; Eddy, S.R.; Luciani, A.; Potter, S.C.; Qureshi, M.; Richardson, L.J.; Salazar, G.A.; Smart, A.; et al. The Pfam protein families database in 2019. Nucleic Acids Res. 2019, 47, D427-D432. [CrossRef] InterPro in 2019: Improving coverage, classification and access to protein sequence annotations. A L Mitchell, T K Attwood, P C Babbitt, M Blum, P Bork, A Bridge, S D Brown, H Y Chang, S El-Gebali, M I Fraser, 10.1093/nar/gky1100Nucleic Acids Res. 47Mitchell, A.L.; Attwood, T.K.; Babbitt, P.C.; Blum, M.; Bork, P.; Bridge, A.; Brown, S.D.; Chang, H.Y.; El-Gebali, S.; Fraser, M.I.; et al. InterPro in 2019: Improving coverage, classification and access to protein sequence annotations. Nucleic Acids Res. 2019, 47, D351-D360. [CrossRef] PISCES: A protein sequence culling server. G Wang, R L Dunbrack, 10.1093/bioinformatics/btg224Bioinformatics. 19Wang, G.; Dunbrack, R.L. PISCES: A protein sequence culling server. Bioinformatics 2003, 19, 1589-1591. [CrossRef] Predicting functionally important residues from sequence conservation. J A Capra, M Singh, 10.1093/bioinformatics/btm270Bioinformatics. 23Capra, J.A.; Singh, M. Predicting functionally important residues from sequence conservation. Bioinformatics 2007, 23, 1875-1882. [CrossRef] ETE 3: Reconstruction, Analysis, and Visualization of Phylogenomic Data. J Huerta-Cepas, F Serra, P Bork, 10.1093/molbev/msw046Mol. Biol. Evol. 33Huerta-Cepas, J.; Serra, F.; Bork, P. ETE 3: Reconstruction, Analysis, and Visualization of Phylogenomic Data. Mol. Biol. Evol. 2016, 33, 1635-1638. [CrossRef] RaptorX-Property: A web server for protein structure property prediction. S Wang, W Li, S Liu, J Xu, 10.1093/nar/gkw306Nucleic Acids Res. 44Wang, S.; Li, W.; Liu, S.; Xu, J. RaptorX-Property: A web server for protein structure property prediction. Nucleic Acids Res. 2016, 44, W430-W435. [CrossRef] AUC-maximized deep convolutional neural fields for protein sequence labeling. S Wang, S Sun, J Xu, Proceedings of the Lecture Notes in Computer Science. the Lecture Notes in Computer ScienceWang, S.; Sun, S.; Xu, J. AUC-maximized deep convolutional neural fields for protein sequence labeling. In Proceedings of the Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); . Springer, 9852Berlin, GermanySpringer: Berlin, Germany, 2016; Volume 9852 LNAI, pp. 1-16. Proteome-level protein disorder prediction by AUC-maximized deep convolutional neural fields. S Wang, J Ma, J Xu, Aucpred, Proceedings of the Bioinformatics. the BioinformaticsOxford, UKOxford University Press32Wang, S.; Ma, J.; Xu, J. AUCpreD: Proteome-level protein disorder prediction by AUC-maximized deep convolutional neural fields. In Proceedings of the Bioinformatics; Oxford University Press: Oxford, UK, 2016; Volume 32, pp. i672-i679. Protein Secondary Structure Prediction Using Deep Convolutional Neural Fields. S Wang, J Peng, J Ma, J Xu, 10.1038/srep18962Sci. Rep. Wang, S.; Peng, J.; Ma, J.; Xu, J. Protein Secondary Structure Prediction Using Deep Convolutional Neural Fields. Sci. Rep. 2016, 6, 18962. [CrossRef] Lightning-fast iterative protein sequence searching by HMM-HMM alignment. M Remmert, A Biegert, A Hauser, J Söding, Hhblits, 10.1038/nmeth.1818Nat. Methods. 9Remmert, M.; Biegert, A.; Hauser, A.; Söding, J. HHblits: Lightning-fast iterative protein sequence searching by HMM-HMM alignment. Nat. Methods 2012, 9, 173-175. [CrossRef] HH-suite3 for fast remote homology detection and deep protein annotation. M Steinegger, M Meier, M Mirdita, H Vöhringer, S J Haunsberger, J Söding, 10.1186/s12859-019-3019-7BMC Bioinform. 20Steinegger, M.; Meier, M.; Mirdita, M.; Vöhringer, H.; Haunsberger, S.J.; Söding, J. HH-suite3 for fast remote homology detection and deep protein annotation. BMC Bioinform. 2019, 20, 473. [CrossRef] Support-vector networks. C Cortes, V Vapnik, 10.1007/BF00994018Mach. Learn. 20Cortes, C.; Vapnik, V. Support-vector networks. Mach. Learn. 1995, 20, 273-297. [CrossRef] Fuzzy Sets, and Classification. S K Pal, S Mitra, Multilayer Perceptron, 10.1109/72.159058IEEE Trans. Neural Netw. 3Pal, S.K.; Mitra, S. Multilayer Perceptron, Fuzzy Sets, and Classification. IEEE Trans. Neural Netw. 1992, 3, 683-697. [CrossRef] The perceptron: A probabilistic model for information storage and organization in the brain. F Rosenblatt, 10.1037/h0042519Psychol. Rev. 65Rosenblatt, F. The perceptron: A probabilistic model for information storage and organization in the brain. Psychol. Rev. 1958, 65, 386-408. [CrossRef] A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting. Y Freund, R E Schapire, 10.1006/jcss.1997.1504J. Comput. Syst. Sci. 55Freund, Y.; Schapire, R.E. A Decision-Theoretic Generalization of On-Line Learning and an Application to Boosting. J. Comput. Syst. Sci. 1997, 55, 119-139. [CrossRef] Adaptive synthetic sampling approach for imbalanced learning. H He, Y Bai, E A Garcia, S Li, Adasyn, Proceedings of the 2008 IEEE International Joint Conference on Neural Networks. the 2008 IEEE International Joint Conference on Neural NetworksPiscataway, NJ, USAIEEEHe, H.; Bai, Y.; Garcia, E.A.; Li, S. ADASYN: Adaptive synthetic sampling approach for imbalanced learning. In Proceedings of the 2008 IEEE International Joint Conference on Neural Networks; IEEE: Piscataway, NJ, USA, 2008; pp. 1322-1328. Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning. H Han, W.-Y Wang, B.-H Mao, Springer3644Berlin/Heidelberg, GermanyHan, H.; Wang, W.-Y.; Mao, B.-H. Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning; Springer: Berlin/Heidelberg, Germany, 2005; Volume 3644. Synthetic minority over-sampling technique. N V Chawla, K W Bowyer, L O Hall, W P Kegelmeyer, Smote, 10.1613/jair.953J. Artif. Intell. Res. 16Chawla, N.V.; Bowyer, K.W.; Hall, L.O.; Kegelmeyer, W.P. SMOTE: Synthetic minority over-sampling technique. J. Artif. Intell. Res. 2002, 16, 321-357. [CrossRef] Borderline Over-sampling for Imbalanced Data Classification. H M Nguyen, E W Cooper, K Kamei, Proceedings: Fifth International Workshop on Computational Intelligence & Applications. Fifth International Workshop on Computational Intelligence & ApplicationsPiscataway, NJ, USAIEEENguyen, H.M.; Cooper, E.W.; Kamei, K. Borderline Over-sampling for Imbalanced Data Classification. In Proceedings: Fifth International Workshop on Computational Intelligence & Applications; IEEE: Piscataway, NJ, USA, 2009; Volume 2009, pp. 24-29. Balancing Training Data for Automated Annotation of Keywords: A Case Study. G E Batista, A L C Bazzan, M C Monard, WOB. Porto Alegre, BrazilUFRGSBatista, G.E.; Bazzan, A.L.C.; Monard, M.C. Balancing Training Data for Automated Annotation of Keywords: A Case Study. In WOB; UFRGS: Porto Alegre, Brazil, 2003; pp. 10-18. A study of the behavior of several methods for balancing machine learning training data. G E Batista, R C Prati, M C Monard, 10.1145/1007730.1007735ACM SIGKDD Explor. Newsl. 6Batista, G.E.; Prati, R.C.; Monard, M.C. A study of the behavior of several methods for balancing machine learning training data. ACM SIGKDD Explor. Newsl. 2004, 6, 20-29. [CrossRef] Scikit-learn: Machine Learning in Python. F Pedregosa, G Varoquaux, A Gramfort, V Michel, B Thirion, O Grisel, M Blondel, J. Mach. Learn. Res. 12Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 2011, 12, 2825-2830. On the limited memory BFGS method for large scale optimization. D C Liu, J Nocedal, 10.1007/BF01589116Math. Program. 45Liu, D.C.; Nocedal, J. On the limited memory BFGS method for large scale optimization. Math. Program. 1989, 45, 503-528. [CrossRef] A method for stochastic optimization. D P Kingma, J L Ba, Adam, Proceedings of the 3rd International Conference on Learning Representations, ICLR 2015-Conference Track Proceedings; International Conference on Learning Representations. the 3rd International Conference on Learning Representations, ICLR 2015-Conference Track Proceedings; International Conference on Learning RepresentationsSan Diego, CA, USAICLRKingma, D.P.; Ba, J.L. Adam: A method for stochastic optimization. In Proceedings of the 3rd International Conference on Learning Representations, ICLR 2015-Conference Track Proceedings; International Conference on Learning Representations; ICLR: San Diego, CA, USA, 2015. . H Robbins, S Monro, 10.1214/aoms/1177729586A Stochastic Approximation Method. Ann. Math. Stat. 22Robbins, H.; Monro, S. A Stochastic Approximation Method. Ann. Math. Stat. 1951, 22, 400-407. [CrossRef] Digital selection and analogue amplification coexist in a cortex-inspired silicon circuit. R H R Hahnioser, R Sarpeshkar, M A Mahowald, R J Douglas, H S Seung, 10.1038/35016072Nature. 405Hahnioser, R.H.R.; Sarpeshkar, R.; Mahowald, M.A.; Douglas, R.J.; Seung, H.S. Digital selection and analogue amplification coexist in a cortex-inspired silicon circuit. Nature 2000, 405, 947-951. [CrossRef] Imbalanced-learn: A Python Toolbox to Tackle the Curse of Imbalanced Datasets in Machine Learning. G Lemaitre, F Nogueira, C K Aridas, J. Mach. Learn. Res. 18Lemaitre, G.; Nogueira, F.; Aridas, C.K. Imbalanced-learn: A Python Toolbox to Tackle the Curse of Imbalanced Datasets in Machine Learning. J. Mach. Learn. Res. 2017, 18, 1-5. A combined transmembrane topology and signal peptide prediction method. L Käll, A Krogh, E L L Sonnhammer, 10.1016/j.jmb.2004.03.016J. Mol. Biol. 338Käll, L.; Krogh, A.; Sonnhammer, E.L.L. A combined transmembrane topology and signal peptide prediction method. J. Mol. Biol. 2004, 338, 1027-1036. [CrossRef] Detecting protein sequence conservation via metric embeddings. E Halperin, J Buhler, R Karp, R Krauthgamer, B Westover, 10.1093/bioinformatics/btg1016Bioinformatics. 19Halperin, E.; Buhler, J.; Karp, R.; Krauthgamer, R.; Westover, B. Detecting protein sequence conservation via metric embeddings. Bioinformatics 2003, 19, 122-129. [CrossRef] RBLOSUM performs better than CorBLOSUM with lesser error per query. R Govindarajan, B C Leela, A S Nair, 10.1186/s13104-018-3415-5BMC Res. Notes. 11PubMedGovindarajan, R.; Leela, B.C.; Nair, A.S. RBLOSUM performs better than CorBLOSUM with lesser error per query. BMC Res. Notes 2018, 11, 328. [CrossRef] [PubMed] BLOSUM62 miscalculations improve search performance. M P Styczynski, K L Jensen, I Rigoutsos, G Stephanopoulos, 10.1038/nbt0308-274Nat. Biotechnol. 26PubMedStyczynski, M.P.; Jensen, K.L.; Rigoutsos, I.; Stephanopoulos, G. BLOSUM62 miscalculations improve search performance. Nat. Biotechnol. 2008, 26, 274-275. [CrossRef] [PubMed] Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. P J Rousseeuw, 10.1016/0377-0427(87)90125-7J. Comput. Appl. Math. 20Rousseeuw, P.J. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis. J. Comput. Appl. Math. 1987, 20, 53-65. [CrossRef] Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. J B Kruskal, 10.1007/BF02289565Psychometrika. 29Kruskal, J.B. Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis. Psychometrika 1964, 29, 1-27. [CrossRef] A Sequence Logo Generator. G E Crooks, G Hon, J.-M Chandonia, S E Brenner, Weblogo, 10.1101/gr.849004Genome Res. 14Crooks, G.E.; Hon, G.; Chandonia, J.-M.; Brenner, S.E. WebLogo: A Sequence Logo Generator. Genome Res. 2004, 14, 1188-1190. [CrossRef] The PyMOL Molecular Graphics System, Version 2.2.3. L L C Schrödinger, 10Schrödinger, L.L.C. The PyMOL Molecular Graphics System, Version 2.2.3. Available online: https: //pymol.org/2/ (accessed on 10 November 2019). Matplotlib: A 2D graphics environment. J D Hunter, 10.1109/MCSE.2007.55Comput. Sci. Eng. 9Hunter, J.D. Matplotlib: A 2D graphics environment. Comput. Sci. Eng. 2007, 9, 99-104. [CrossRef] Structural diversity of leucine-rich repeat proteins. A V Kajava, 10.1006/jmbi.1998.1643J. Mol. Biol. 277Kajava, A.V. Structural diversity of leucine-rich repeat proteins. J. Mol. Biol. 1998, 277, 519-527. [CrossRef] Modern Multidimensional Scaling. I Borg, P J Groenen, SpringerBerlin, GermanyMDS Models and Measures of FitBorg, I.; Groenen, P.J.F. Modern Multidimensional Scaling, 2nd ed.; Chapter: MDS Models and Measures of Fit; Springer: Berlin, Germany, 2005; pp. 37-61. The relationship between the L1 and L2 domains of the insulin and epidermal growth factor receptors and leucine-rich repeat modules. C W Ward, T P Garrett, 10.1186/1471-2105-2-4BMC Bioinform. 2001, 2, 4. [CrossRefWard, C.W.; Garrett, T.P.J. The relationship between the L1 and L2 domains of the insulin and epidermal growth factor receptors and leucine-rich repeat modules. BMC Bioinform. 2001, 2, 4. [CrossRef] The leucine-rich repeat domain in plant innate immunity: A wealth of possibilities. M Padmanabhan, P Cournoyer, S P Dinesh-Kumar, 10.1111/j.1462-5822.2008.01260.xCell. Microbiol. 11Padmanabhan, M.; Cournoyer, P.; Dinesh-Kumar, S.P. The leucine-rich repeat domain in plant innate immunity: A wealth of possibilities. Cell. Microbiol. 2009, 11, 191-198. [CrossRef] Plant immunity: Towards an integrated view of plantĝ€ pathogen interactions. P N Dodds, J P Rathjen, 10.1038/nrg2812Nat. Rev. Genet. 11Dodds, P.N.; Rathjen, J.P. Plant immunity: Towards an integrated view of plantĝ€ pathogen interactions. Nat. Rev. Genet. 2010, 11, 539-548. [CrossRef] Co-evolutionary interactions between host resistance and pathogen effector genes in flax rust disease. M Ravensdale, A Nemri, P H Thrall, J G Ellis, P N Dodds, 10.1111/j.1364-3703.2010.00657.xMol. Plant Pathol. 12Ravensdale, M.; Nemri, A.; Thrall, P.H.; Ellis, J.G.; Dodds, P.N. Co-evolutionary interactions between host resistance and pathogen effector genes in flax rust disease. Mol. Plant Pathol. 2011, 12, 93-102. [CrossRef] Function of Nod-like receptors in microbial recognition and host defense. L Franchi, N Warner, K Viani, G Nuñez, 10.1111/j.1600-065X.2008.00734.xImmunol. Rev. 227Franchi, L.; Warner, N.; Viani, K.; Nuñez, G. Function of Nod-like receptors in microbial recognition and host defense. Immunol. Rev. 2009, 227, 106-128. [CrossRef] Regulation and evolution of NLR genes: A close interconnection for plant immunity. G M Borrelli, E Mazzucotelli, D Marone, C Crosatti, V Michelotti, G Valè, A M Mastrangelo, 10.3390/ijms19061662Int. J. Mol. Sci. Borrelli, G.M.; Mazzucotelli, E.; Marone, D.; Crosatti, C.; Michelotti, V.; Valè, G.; Mastrangelo, A.M. Regulation and evolution of NLR genes: A close interconnection for plant immunity. Int. J. Mol. Sci. 2018, 19, 1662. [CrossRef] Molecular genetic evidence for the role of SGT1 in the intramolecular complementation of Bs2 protein activity in Nicotiana benthamiana. R T Leister, D Dahlbeck, B Day, Y Li, O Chesnokova, B J Staskawicz, 10.1105/tpc.104.029637Plant Cell. 17Leister, R.T.; Dahlbeck, D.; Day, B.; Li, Y.; Chesnokova, O.; Staskawicz, B.J. Molecular genetic evidence for the role of SGT1 in the intramolecular complementation of Bs2 protein activity in Nicotiana benthamiana. Plant Cell 2005, 17, 1268-1278. [CrossRef] The Arabidopsis ZED1 pseudokinase is required for ZAR1-mediated immunity induced by the Pseudomonas syringae type III effector HopZ1a. J D Lewis, A H Lee, .-Y Hassan, J A Wan, J Hurley, B Jhingree, J R Wang, P W Lo, T Youn, J.-Y Guttman, D S , 10.1073/pnas.1315520110Proc. Natl. Acad. Sci. Natl. Acad. SciUSA110Lewis, J.D.; Lee, A.H.-Y.; Hassan, J.A.; Wan, J.; Hurley, B.; Jhingree, J.R.; Wang, P.W.; Lo, T.; Youn, J.-Y.; Guttman, D.S.; et al. The Arabidopsis ZED1 pseudokinase is required for ZAR1-mediated immunity induced by the Pseudomonas syringae type III effector HopZ1a. Proc. Natl. Acad. Sci. USA 2013, 110, 18722-18727. [CrossRef] Elicitor-mediated oligomerization of the tobacco N disease resistance protein. P Mestre, D C Baulcombe, 10.1105/tpc.105.037234Plant Cell. 18Mestre, P.; Baulcombe, D.C. Elicitor-mediated oligomerization of the tobacco N disease resistance protein. Plant Cell 2006, 18, 491-501. [CrossRef] Protein-protein interactions in the RPS4/RRS1 immune receptor complex. S U Huh, V Cevik, P Ding, Z Duxbury, Y Ma, L Tomlinson, P F Sarris, J D G Jones, 10.1371/journal.ppat.1006376PLoS Pathog. 13Huh, S.U.; Cevik, V.; Ding, P.; Duxbury, Z.; Ma, Y.; Tomlinson, L.; Sarris, P.F.; Jones, J.D.G. Protein-protein interactions in the RPS4/RRS1 immune receptor complex. PLoS Pathog. 2017, 13, e1006376. [CrossRef] Crystal structure of NLRC4 reveals its autoinhibition mechanism. Z Hu, C Yan, P Liu, Z Huang, R Ma, C Zhang, R Wang, Y Zhang, F Martinon, D Miao, 10.1126/science.1236381Science. 341Hu, Z.; Yan, C.; Liu, P.; Huang, Z.; Ma, R.; Zhang, C.; Wang, R.; Zhang, Y.; Martinon, F.; Miao, D.; et al. Crystal structure of NLRC4 reveals its autoinhibition mechanism. Science 2013, 341, 172-175. [CrossRef] Structural basis for flg22-induced activation of the Arabidopsis FLS2-BAK1 immune complex. Y Sun, L Li, A P Macho, Z Han, Z Hu, C Zipfel, J.-M Zhou, J Chai, 10.1126/science.1243825Science. 342Sun, Y.; Li, L.; Macho, A.P.; Han, Z.; Hu, Z.; Zipfel, C.; Zhou, J.-M.; Chai, J. Structural basis for flg22-induced activation of the Arabidopsis FLS2-BAK1 immune complex. Science 2013, 342, 624-628. [CrossRef] Crystal structure of human toll-like receptor 3. J Choe, M S Kelker, I A Wilson, 10.1126/science.1115253Science. 22Choe, J.; Kelker, M.S.; Wilson, I.A. Crystal structure of human toll-like receptor 3. Science 2005, 22, 581-585. [CrossRef] The structure of Bacillus subtilis pectate lyase in complex with calcium. R Pickersgill, J Jenkins, G Harris, W Nasser, J Robert-Baudouy, 10.1038/nsb1094-717Nat. Struct. Biol. 1Pickersgill, R.; Jenkins, J.; Harris, G.; Nasser, W.; Robert-Baudouy, J. The structure of Bacillus subtilis pectate lyase in complex with calcium. Nat. Struct. Biol. 1994, 1, 717-723. [CrossRef] Crystal structure and catalytic mechanism of PglD from Campylobacter jejuni. N B Olivier, B Imperiali, 10.1074/jbc.M801207200J. Biol. Chem. 283Olivier, N.B.; Imperiali, B. Crystal structure and catalytic mechanism of PglD from Campylobacter jejuni. J. Biol. Chem. 2008, 283, 27937-27946. [CrossRef] Three-dimensional structure of the armadillo repeat region of β-catenin. A H Huber, W J Nelson, W I Weis, 10.1016/S0092-8674(00)80352-9Cell. 90Huber, A.H.; Nelson, W.J.; Weis, W.I. Three-dimensional structure of the armadillo repeat region of β-catenin. Cell 1997, 90, 871-882. [CrossRef] . N Takahashi, S Hamada-Nakahara, Y Itoh, K Takemura, A Shimada, Y Ueda, M Kitamata, R Matsuoka, K Hanawa-Suetsugu, Y Senju, 10.1038/ncomms5994Nat. Commun. 5et al. TRPV4 channel activity is modulated by direct interaction of the ankyrin domain to PI(4,5)P 2Takahashi, N.; Hamada-Nakahara, S.; Itoh, Y.; Takemura, K.; Shimada, A.; Ueda, Y.; Kitamata, M.; Matsuoka, R.; Hanawa-Suetsugu, K.; Senju, Y.; et al. TRPV4 channel activity is modulated by direct interaction of the ankyrin domain to PI(4,5)P 2 . Nat. Commun. 2014, 5, 4994. [CrossRef] This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. © 2020 by the authors. Licensee MDPI. Basel, Switzerland© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

2022-01-10T05:35:55Z

CCBY

https://doi.org/10.18632/aging.203518

GOLD

1c8fde34e660e157d40048091edfdda3d5b367eb

10.18632/aging.203518

34506301

8457566

Knock-down of odr-3 and ife-2 additively extends lifespan and healthspan in C. elegans Published: September 9, 2021 Ioan Valentin Matei Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Vimbai Netsai Charity Samukange Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Gabriela Bunu Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Dmitri Toren Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Simona Ghenea [email protected] Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Robi Tacutu [email protected] Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Knock-down of odr-3 and ife-2 additively extends lifespan and healthspan in C. elegans 1317Published: September 9, 2021Received: June 11, 2021 Accepted: August 24, 202121040 AGING Research Paper *Equal contribution Correspondence to: Simona Ghenea, Robi Tacutu;lifespan extensiongenetic interventionssynergismife-2odr-3 Genetic manipulations can ameliorate the aging process and extend the lifespan of model organisms. The aim of this research was to identify novel genetic interventions that promote both lifespan and healthspan, by combining the effects of multiple longevity-associated gene inactivations in C. elegans. For this, the individual and combined effects of the odr-3 mutation and of ife-2 and cku-70 knock-downs were studied, both in the wild type and daf-16 mutant backgrounds. We found that besides increasing the lifespan of wild type animals, the knock-down of ife-2 (starting at L4) also extends the lifespan and healthspan of long-lived odr-3 mutants. In the daf-16 background, ife-2 and odr-3 impairment exert opposing effects individually, while the daf-16; odr-3; ife-2 deficient animals show a similar lifespan and healthspan as daf-16, suggesting that the odr-3 and ife-2 effector outcomes converge downstream of DAF-16. By contrast, cku-70 knock-down did not extend the lifespan of single or double odr-3; ife-2 inactivated animals, and was slightly deleterious to healthspan. In conclusion, we report that impairment of odr-3 and ife-2 increases lifespan and healthspan in an additive and synergistic manner, respectively, and that this result is not improved by further knocking-down cku-70.www.aging-us.com 21041 AGING expression levels in long-lived species[11,12], overall suggesting that some of the reported longevityassociated interventions could have therapeutic implications even in humans. INTRODUCTION The aging process might be defined by the progressive loss of viability and by an increase in fragility and vulnerability [1,2]. This in turn, results in a huge health-related cost for the elderly and a dramatic growth in the mortality rate. Understanding the mechanisms underlying aging is one of the major biological and biomedical challenges of our society, and could result in high dividends if the society would gain the capacity to extend lifespan, and more importantly healthspan (i.e. the interval of healthy, productive life years) [3][4][5]. Although there is still much debate about the molecular causes of aging, the general consensus in the field is that aging is malleable, and studies in model organisms have already shown that aging can be manipulated by both genetic and environmental factors [6][7][8]. Up until now, more than 2,200 single-gene interventions have been reported to modulate lifespan in model organisms [9]. Most of these genes have been found through genetic interventions, including partial or full loss-offunction mutations, RNA-induced gene silencing, gene over-expression, and genetic polymorphisms, which were reported to promote longevity or cause a premature aging phenotype [9]. More importantly, it has been shown that a large part of these genes play a conserved role as longevity regulators across diverse taxa [10], and some of them even share similar gene The effect on the mean and/or maximum lifespan of the modified organisms ranges from very modest values (5-10% change) up to very high values, for well-established longevity-associated genes -for example, two-fold for daf-2 in worms [13], six-fold for SIR2 in yeast [14], and even ten-fold for age-1 in worms [15]. Genetic modifications have been identified even in mammals, albeit the observed effects so far seem to be smaller (up to a maximum of 50%) [9]. These works have significantly increased our knowledge about the genetics of aging and longevity in model organisms, and they should be followed by investigations into the effect of epistatic, or more precisely synergistic gene combinations on lifespan. This aspect, however, has been unfortunately less popular, mainly because the epistasis between longevity-associated genes, and between the pathways they are involved in, is complex and most often non-linear [16,17], thus requiring much time and resources to be studied. In a recent paper, describing the SynergyAge database, we have defined three types of synergism, applied to the general case of N genetic interventions: 1) full synergism, in which lifespan values are known for all intermediary strains that contain any combination of the N interventions and the lifespan change for the n-mutant is greater than the sum of lifespan changes for any two intermediary k-mutant and (N-k)-mutant, 2) simple synergism, in which lifespan values are known for the final strain (N interventions) and for all single gene interventions, but not for all intermediary k-mutants, and in which the lifespan effect of the N-gene combination is greater than the sum of all the individual effects, and 3) partially known synergism, in which values are available only for an incrementally built model and for all genetic interventions in an Nsequence an increase in lifespan is observed [18]. The few seminal discoveries regarding longevity synergism generally include the well known IIF/FOXO pathway and the daf-2/daf-16 genes, and have been started in C. elegans [6,19,20]. The SynergyAge database reports 62 synergistic combinations of prolongevity interventions that include daf-2. Interestingly, based on SynergyAge data, we did not observe a general correlation between the strength of the longevity effect in WT with those in the long-lived daf-2 mutant. For example, RNAi of let-363 did not extend the lifespan of the daf-2(mu150) mutant [21], even though the two genes have the 2nd and 3rd largest increase of lifespan in WT (according to GenAge). sod-2, another important longevity-associated gene, whose deletion leads to a lifespan increase in WT, does not further extend the lifespan of daf-2 mutants [22]. Moreover, three of the top daf-2 enhancers have only a small effect in WT, when kept under same conditions as in the daf-2 background: clk-1 increases lifespan by only 1.18% compared to WT at 25° C [23] even though at this temperature extends daf-2 lifespan by 205%; rsks-1 increases lifespan of daf-2 by 106%, but only by 20% in the WT [24]; drp-1, which potentiates the effect of daf-2 by 73%, increases lifespan of WT by only 2% [25]; clk-2 increases daf-2 effect by +113% while in the WT the effect is limited to 68% [26]. In our study, the genes to be tested were selected based on several bioinformatic criteria (potential of being longevity enhancers for the daf-2 knock-down, genes being part of individual clusters in a cross-database interactome, number of shared KEGG pathways, chromosome positions, etc.), followed by manual curation and evaluation (of scientific literature) for the short-listed gene combinations. In mammals, the homologues of daf-2 and daf-16 are components of the mammalian insulin and insulin growth factor (IGF) signal transduction cascade (IIS) [27][28][29]. DAF-2 regulates endocrine responses to food availability, including longevity, dauer formation, and fat metabolism [13,30,31]. Mutations that reduce the function of DAF-2 extend lifespan through a mechanism that greatly depends on the activity of DAF-16 [32,33]. In addition to the central role in integrating signals from insulin/insulin-like pathways, DAF-16 integrates signals from multiple upstream pathways to regulate various biological processes [34]. Due to the increased amount of data on daf-2 and daf-16 mutants, it is extremely appealing to search for genetic interventions that act synergistically amongst themselves, but also with the daf-2 long-lived background. In this study, three such genes have been considered: odr-3, ife-2 and cku-70. Several sensory neurons are responsible for chemotaxis to volatile attractants found in food, pheromones or noxious odors [27,35,36], the nutrient perception by olfactory neurons being partially mediated by the DAF-2 pathway [36]. ODR-3, a G alpha protein with similarities to the members of Gi/Go protein family, is expressed in the sensory cilia of olfactory neurons, providing the main stimulatory signals for AWA and AWC sensory neurons [37,38]. Ablation of AWA and AWC sensory neurons, as well as loss-of-function mutations in odr-3, extend lifespan through a pathway that depends partially or completely on signaling via DAF-16 [27,36,39]. Food restriction can promote an adaptive metabolic response such as mobilization of fat stores through activation of AWC neurons [40], and decreased DAF-2 signaling is known to affect cellular metabolism by promoting the accumulation of lipids in the intestine and hypodermis [30]. All these suggest a link between food sensing, metabolic adaptation and longevity. On the other hand, the daf-2(e1370); odr-3(n1605) double mutant shows a greater lifespan extension than either of the single mutants and even than their cumulative effects, thus odr-3 and daf-2 could also function through complementary pathways [39]. While the relationship between ROS and longevity is still not completely understood and ROS can have both beneficial or detrimental effects on lifespan, most of the genetic manipulations that decrease ROS lead to an increased lifespan [41]. Like its mammalian orthologue, eIF4E, the C. elegans IFE-2 plays an important role in protein synthesis and its inactivation protects against oxidative stress and extends lifespan [42]. Since ife-2 impairment was found to extend the lifespan of longlived mutants such as daf-2, clk-1, eat-2 and let-363, it was suggested that down-regulation of protein synthesis induced by ife-2 deficiency might represent a distinct mechanism by which lifespan is regulated [21,42]. However, ife-2 inactivation might extend lifespan not only by decreasing the rate of protein synthesis, but also by regulating mitochondrial and peroxisomal metabolism, which in turn, could stabilize the homeostasis of reactive oxygen species and increase cellular accumulation of trehalose [43]. Lastly, CKU-70 is the C. elegans orthologue of KU70, which in mammals participates with KU80 to the DNA repair of double-strand breaks [44]. Downregulation of CKU-70 activity was found to increase sensitivity to genotoxic stress and thermotolerance, thus indicating a conserved role in both DNA repair and stress response [45,46]. Although RNA interference (RNAi) of cku-70 increases the lifespan of wild type (WT) animals only in an RNAi sensitized background, the fact that cku-70 knock-down extends the lifespan of daf-2 mutants as well [46] suggests that cku-70 might have an important role in aging. Since odr-3, ife-2 and cku-70 deficiencies all potentiate the lifespan-extending effects of daf-2 mutants, it is also interesting to find if their mechanisms involve downstream pathways that converge toward common effectors. In this work, we analyzed the effect of combined interventions in odr-3, ife-2 and cku-70, on lifespan and healthspan, starting with L4 age. Since our lifespan and healthspan assays were carried out for all the combinations of the above-mentioned interventions, the use of synergism in the remainder of the paper refers to the "full synergism" definition. Our results show that simultaneous suppression of odr-3 and ife-2 functions additively extends lifespan and synergistically improves healthspan in a daf-16 dependent manner. Knock-down of cku-70 did not confer further benefits to lifespan or motility of odr-3; ife-2 mutants. RESULTS RNA interference of ife-2 but not cku-70 increases lifespan of the long-lived odr-3 mutants To find new potential genetic interactions that could extend lifespan, we assessed the effect of a simultaneous depletion of ODR-3, IFE-2 and CKU-70. For this, we used the odr-3(n1605) putative null allele [37] and we knocked-down ife-2 and cku-70 by RNAi. The odr-3(n1605) animals exhibited at 20° C an increased mean (26.2%) and maximum (13.8%) lifespan compared with WT control animals. A significant mean lifespan extension was previously reported for odr-3(n1605) at 25° C, however the increase was very modest at 20° C [39]. Silencing of ife-2 by RNAi showed an 18.0% and 20.7% extension for the mean and maximum lifespan of WT, respectively ( Figure 1A and Supplementary Table 1), which are in agreement with previously reported data for both ife-2(ok306) mutants and ife-2 downregulated animals [21,42]. In our experiments, the RNAi knockdown of cku-70 in the WT worms produced only a marginal 4.4% increase for mean lifespan and was even slightly detrimental to maximum lifespan reducing it by 6.9% ( Figure 1B and Supplementary Table 1), which is in agreement with previously reported data [46]. The odr-3(n1605); ife-2(RNAi) mutants exhibited more than 11% and 18% increase in mean lifespan compared with the odr-3(n1605) and ife-2(RNAi) single gene interventions, respectively (Table 1). Similarly, the maximum lifespan was also increased by more than 18% and 11% (Supplementary Table 1). Overall, compared with the WT controls, the combined odr-3 and ife-2 interventions extended mean and maximum lifespan by 40.3% and 34.5%, respectively ( Figure 1A and Table 1 and Supplementary Table 1). This effect demonstrates an almost additive impact on mean lifespan, i.e. 40.3% increase compared to 44.2%, the sum of the two individual effects (Table 1). Similarly, the lifespan extension for odr-3; ife-2; EV (46.1%), i.e. worms exposed to a 1:1 mixture of HT115 (Empty Vector -EV) bacteria and of ife-2 RNAi clone, was greater than the sum of individual effects (26.2% + 11.2%), supporting the existence of additive/synergistic mechanisms (Table 1). Next, we assessed the effect of cku-70 silencing in both odr-3(n1605) mutant animals and ife-2 knock-down animals. We observed that cku-70 knock-down dramatically decreased the extension of mean lifespan conferred by the odr-3(n1605) mutation, from 26.2% to only 13.6% increase comparative with WT ( Figure 1B and Supplementary Table 1). The simultaneous knock-down of ife-2; cku-70 by RNAi was performed by co-feeding worms with a mixture of the two RNAi bacterial clones. As such, for an appropriate comparison, the survival curves of double knock-down worms (which are presumably exposed to about half dsRNA for each gene) have been compared with those of single knock-down worms exposed to the same concentration of dsRNA for each of the corresponding genes (concentrations obtained by co-feeding the worms with the target RNAi clone and the control RNAi(EV) in a 1:1 ratio). In general, we obtained very small differences between the lifespan of worms fed only with the RNAi clone and worms fed with the mixture of RNAi clone / RNAi(EV) (Supplementary Figure 1A-1C), with the only notable difference being for odr-3; cku-70 for which the mix (and hence lower concentration of cku-70 RNAi bacteria) did not show a pronounced lifespan reduction (Supplementary Figure 1D). AGING AGING In our assays, the lifespan of the double knock-down worms ife-2; cku-70 was 15% longer than that of WT animals (p = 3.0E-16), with a small increase compared to each of the single knock-down worms (3.5% and 8.22% longer lived than the ife-2; EV and the cku-70; EV animals, respectively) ( Figure 1C and Supplementary Table 1). It should however be noted that these changes are very modest and that a slightly larger lifespan increase was obtained when worms were exposed to ife-2 RNAi alone, without EV mixing (18% compared to WT). Moreover, cku-70 knock-down had a negative effect on the lifespan of odr-3(n1605) mutants treated with ife-2 RNAi, decreasing the mean lifespan extension from 46.1% to 35.4% ( Figure 1D and Supplementary Table 1). The extended longevity of odr-3; ife-2 double intervention might be independent of DAF-16 The FOXO family transcription factor DAF-16 is a transducer of many pro-longevity signaling pathways [47], thus it was natural to inquire to what extent the longevity of odr-3; ife-2 double inactivated animals require DAF-16. To answer this, we used the null daf-16(mu86) allele [32,48] that affects coding of all DAF-16 isoforms, to generate daf-16(mu86); odr-3(n1605) double mutants, and carried out RNAi silencing assays for ife-2 and cku-70 in this strain. We observed that the lifespan extension induced by the odr-3(n1605) mutation was not only suppressed by the daf-16(mu86) mutation, but also that the lifespan of the daf-16(mu86); odr-3(n1605) double mutants were even shorter than the lifespan of the daf-16 mutants alone ( Figure 1E; mean and maximum lifespan decreased by 14.6% and 9.1% compared to daf-16; mean and maximum lifespan decreased by 26.3% and 16.7% compared with WT). A slight decrease of daf-16(mu86) lifespan induced by the odr-3(n1605) mutation was also previously reported [39]. Silencing of ife-2 on the other hand, extended the lifespan of daf-16 mutants ( Figure 1E; mean lifespan 9.8% greater), although the mean and maximum lifespan were not completely reverted to the lifespan values of the WT (mean and maximum lifespan 5.3% and 4.2% lower than WT, respectively). The lifespan of the triple daf-16; odr-3; ife-2 inactivated animals did not significantly differ from daf-16 single mutants. This could be partly explained by the fact that DAF-16 is one of the main transducers of signaling pathways modulated by ODR-3 and IFE-2 activity. To clarify this aspect we examined the nuclear translocation of DAF-16::GFP in odr-3(n1605), ife-2(RNAi) and odr-3(n1605); ife-2(RNAi) animals, respectively (Supplementary Figure 2). Whereas odr-3(n1605) animals showed weak DAF-16::GFP nuclear accumulation in posterior intestinal cells, suggesting that ODR-3 could affect longevity partially through DAF-16 pathway, we did not observe consistent nuclear accumulation of DAF-16::GFP in ife-2(RNAi) and odr-3(n1605); ife-2(RNAi) animals. Therefore, ODR-3 and IFE-2 could AGING affect lifespan by mediating parallel signaling pathways, which in the daf-16 background have antagonistic effects -odr-3 further decreasing lifespan and ife-2 partially increasing the daf- 16(mu86) lifespan. In contrast to a previous study that reported a slight decrease of daf-16(m26) lifespan by cku-70 knockdown at 25° C [46], in our experiments the lifespan of daf-16; cku-70 at 20° C was similar to that of daf-16 single mutants, and cku-70 knock-down did not significantly influence the lifespan of daf-16; odr-3, nor of daf-16; ife-2 mutants ( Figure 1F, 1G). The quadruple daf-16; odr-3; ife-2; cku-70 mutants exhibited a lifespan similar to that of daf-16 single mutants ( Figure 1H). Overall, our results show that simultaneous inactivation of odr-3 and ife-2 produce an additive lifespan effect, while the additional cku-70 knock-down does not extend lifespan further ( Figure 2). Since the lifespan effects observed are different in the daf-16 background (Figure 2), it is possible that the mechanisms through which odr-3; ife-2 animals achieve lifespan extension overlap with the pleiotropic mechanisms determined by daf-16 ( Figure 2). AGING The ife-2(ok306) mutation also extends the lifespan of odr-3(n1605) mutant animals One question is whether the ife-2(ok306) deletion mutation will produce similar effects as the ife-2(RNAi) in the odr- 3(n1605) background. To answer this, we also generated the double odr-3(n1605); ife-2(ok306) and triple daf-16(mu86); odr-3(n1605); ife-2(ok306) mutants and conducted further lifespan assays. The mutant worms were cultured at the same temperature (20° C) as in the RNAi experiments and were fed OP50 bacteria. We observed similar trends ( Figure 3A), i.e. a 13.6% and 35.7% increase in mean and maximum lifespan for the odr-3; ife-2 double mutant, compared to wild type, and an additive effect of the single mutations (the 13.6% increase in mean lifespan was comparable with the sum of the individual genetic effects of odr-3 and ife-2 mutations: 7.5% and 4.5%). It was however noticeable that in this experiment the impact of the double mutation on lifespan was smaller than in the case of ife-2(RNAi), which used an HT115 diet. Since FUdR could cause an artefactual effect on the longevity of some mutants [49][50][51], we also conducted longevity experiments in the absence of FUdR. Similar to the previous results, we also observed an increased average lifespan for all strains even in the absence of FUdR ( Figure 3B, odr-3:+8.2%, ife-2:+8.6% and odr-3;ife-2:+14.3%). Next, we investigated the effect of the odr-3(n1605) and ife-2(ok306) mutations in the daf- 16 and triple daf-16(mu86); odr-3(n1605); ife-2(ok306) mutants showed very similar lifespans as that of the daf-16 ( Figure 3C). The odr-3; ife-2 impaired animals display increased motility and pharyngeal pumping Our finding that RNAi impairment of ife-2 in odr-3(n1605) animals increased lifespan prompted us to investigate the effect of their inactivation on healthspan. To assess healthspan, we focused on the evaluation of pharyngeal pumping and body movement. In C. elegans, these two physiological processes decrease with ageing, correlate between themselves and with other age-related declining properties, and ultimately can predict lifespan and healthy life [52]. In our experiments, up to late adulthood, individual or joint interventions in odr-3 and ife-2 did not produce obvious pathological changes in the phenotype, indicating that the effect on locomotion was caused by physiological age-related changes, rather than a specific disease. As such, the observed motility status, carried out along the lifespan assay, can be viewed as a measure of the healthspan of the population. To quantify this, we classified individuals in three motion stages based on their ability to move. Stage A (healthy, fully mobile worms) included animals in a physiological state that could move without any impediment, stage B (impaired worms) included animals with diminished locomotion, whereas stage C (frail worms) included animals found in a frailty state. Animals were scored daily and associated with one of the motion stages. The distribution of stages for each strain is presented graphically in Figure 4A-4D (for Considering the average healthspan of worms in stage A, we observed a similar trend to lifespan. Simultaneous inactivation of odr-3 and ife-2 produced a synergistic healthspan effect, while the additional cku-70 knock-down does not extend healthspan further (Supplementary Figure 4). Using the Kaplan-Meier method to estimate the fraction of mobile worms at each observation point, plotted against time, the average number of days the worms spent in each state was computed ( Figure 4E and Supplementary Table 2). This allowed us to model the transitions from the healthy to impaired or frail (see Materials and methods), thus, determining if a change in the locomotion status was induced by the genetic interventions and whether in addition to lifespan, the ratio between healthspan and lifespan was also changed. The WT animals spent on average 17.3 days in the mobile stage, which represent 84% of their mean lifetime, 1.5 days (7.3%) in an impaired stage and 1.8 days (8.7%) in the frailty stage. The odr-3 mutants remained mobile longer than WT animals (on average 20.8 days), however relative to their mean lifespan, they were fully mobile only for 80.0% of their lifetime, thus exhibiting a proportional (or greater) lifespan fraction in which they were impaired or frail (2.6 days, 10.0%, for both). ife-2 RNAi treated animals remained mobile on average more days (21.0 days, 86.4%) and exhibited a similar number of impaired (1.6 days, 6.6%) and frail days (1.7 days, 7.0%) (Supplementary Table 2). For the odr-3; ife-2 animals, the longest-lived strain in our study, a corresponding increase of days with full (25.0 days, 86.5%) and impaired motility (2.2 days, 7.6%) was observed, while the number of days of frailty remained similar as for WT animals (1.7 days, 5.9%). Next, we compared each fraction of being fully mobile with the corresponding fraction in the WT animals to find if the genetic interventions indeed conferred significant benefits for the quality of life, i.e. increasing the fraction of time spent in a mobile state and decreasing the impaired and frailty fractions. By doing this, we found that although the odr-3 mutants had extended longevity, this was not associated with a motility-based health benefit since the lifespan fraction in which worms lived as fully mobile actually decreased slightly by 4.8% (Supplementary Table 2), whereas the fraction of time spent in both the impaired and frailty stages increased by almost 40%. By contrast, ife-2 RNAi treated animals stayed mobile for a similar lifespan fraction as WT animals (3% longer), but they spent much less time as impaired or frail (these fractions were 9.6% and 19.5%, respectively, smaller than those for WT). Compared to WT control animals, silencing of ife-2 in odr-3 worms did not affect the mean lifespan fraction spent in the mobile and impaired stages, but decreased the lifespan fraction for the frailty period by more than 30%. To assess the significance of the changes in motility status, the Kaplan-Meier curves modeling the transitions from mobile to impaired and frailed were used and the relevant comparisons are included in the Supplementary Figure 5C To further determine whether healthspan is affected, we analyzed the decline of pharyngeal pumping with age, a process controlled by the cardiac-like pharyngeal muscle. As seen in Figure 4F, the number of pharyngeal movements strongly decreases with age in all cohorts. The changes for both single mutants, odr-3(n1605) and ife-2(ok306), are similar to those in the WT (no statistically significant difference observed when comparing to WT, at each day; one-way ANOVA). The odr-3(n1605); ife-2(ok306) double mutant showed a slower decline in pharyngeal contractions and a higher rate of pumping compared to WT. The improved healthspan was observed starting from day 5, when the average number of contractions was 17.8% higher than for WT (p = 0.002), and slightly increased, at day 10 being 21.4% higher (p = 0.14; ns). At day 15, the odr-3;ife-2 animals still showed approximately 42 contractions per minute (compared to 15 contractions per minute in the WT), a 187% increase (p <E-04). On day 15, the odr-3(n1605) and ife-2(ok306) mutations showed a synergistic effect on the pharyngeal pumping, the increase observed in the double mutant being higher than the sum of the individual effects (187% increase compared to 66% + 55% increase for single mutants), corresponding to a fully synergic effect [18]. The odr-3; ife-2 double mutant displays an increased resistance to stress Resistance to stress declines with age after early adulthood [53], and the loss of protein homeostasis together with the failure to activate cellular stress responses are among the earliest aging marks [54]. Elevated temperatures perturb the protein homeostasis due to accumulation of defective proteins, whereas production of reactive oxygen species (ROS) and H2O2 increase global cellular damage. Both insults activate the cellular stress responses, aiming to improve cellular fitness and organismal recovery. To find if the double intervention in odr-3 and ife-2 perturbs the stress response mechanisms, we monitored survival upon exposure to oxidative stress and acute heat stress. We induced oxidative stress by treatment with either paraquat, which generates reactive oxygen species, or NaN3, which generates H2O2. Both treatments dramatically reduce survival of WT animals by more than 50% (Figure 5A, 5B). In contrast, mutation in the odr-3 increases survival of the animals treated with both paraquat and NaN3 compared with treated WT, although not statistically significant for NaN3. ife-2 mutants exhibited an increased survival upon both treatments, as previously reported [42]. The odr-3; ife-2 double mutants were less sensitive to ROS and H2O2 toxicity than WT animals, but they did not show an additive effect when compared with single mutants ( Figure 5A, 5B). Exposure to 35° C for 4h decreases the survival rate of WT animals by more than 50% ( Figure 5C). Whereas the heat shock slightly increased the survival of odr-3 mutants, compared with WT (not statistically significant), it did not also affect the survival of ife-2 mutants or odr-3; ife-2 double mutants ( Figure 5C), overall indicating that impairment of odr-3 and/or ife-2 do not affect the heat stress response. Some genetic or non-genetic interventions that extend lifespan also reduce fecundity, implying a trade-off between longevity and reproduction [55][56][57]. We did not observe such an effect for odr-3 and ife-2 single and double mutants ( Figure 5D). Thus, the reproductive period and the age of the peak egg-laying rate of single and double mutants coincided with that of WT animals ( Figure 5D). Moreover, the brood size of the mutants is similar to that of WT animals ( Figure 5E). In conclusion, simultaneous depletion of odr-3 and ife-2 extends lifespan without affecting fecundity, promotes muscle activity and maintains activation of stress response mechanisms, consistent with increased health. DISCUSSION Longevity is regulated by a combination of genetic and non-genetic factors, such as environmental interactions and lifestyle. The identification of genetic mutations that extend lifespan in model organisms has shown that longevity is mainly regulated by a complex interplay between many signaling pathways that affect cellular functions as diverse as nutrient sensing, genome stability, mitochondria fitness, organelle proteostasis, intercellular communication, transcription, proliferation and cellular regeneration [58,59]. Previously, we have successfully used network-based approaches, building upon the list of known longevity-associated genes hosted in the GenAge database [9], to predict novel genetic or drug interventions that extend lifespan [60,61]. However, due to the existence of complex and intricate interactions between hundreds of longevityassociated genes [62][63][64], the lifespan modulation obtained with these methods was limited by how much a single gene can influence longevity. While combined genetic interventions that modulate longevity via parallel pathways or drugs targeting multiple evolutionary conserved aging pathways have been shown in many instances to extend lifespan [65], the number of gene combinations tested so far has not been very high [18]. Here, we assessed the effects on both lifespan and healthspan, given by the simultaneous inactivation of three genes: odr-3, ife-2 and cku-70. By reporting on the synergy of the pro-longevity effects of IFE-2 and ODR-3, and at the same time on the lack of synergy between CKU-70 and the above two genes, we hope that the current work will add to the accumulating data on longevity-related gene combinations, which could be used in future predictions of complex, multigene interventions. According to the SynergyAge database http://synergyage.info/ [18], the above-mentioned genetic interventions were among the most promising in terms of lifespan extension, when combined with IISdefective daf-2 mutants [39,42,46], a highly desirable property. SynergyAge hosts 133 unique synergistic interactions, involving 108 genes and of these, 62 gene combinations include daf-2. Much less combinations are antagonistic (32 gene combinations) while for 156 gene combinations the effect is somewhere between the two individual effects. Although this summary does not provide information on the number of negative results (which would probably be classified in one of the last 2 categories) it gives a sense of the scarcity of synergistic interactions discovered so far -considering the total number of potential gene combinations that could encompass for example the 889 worm longevityassociated genes from GenAge [9] (even for two gene combinations). In C. elegans, ODR-3, IFE-2 and CKU-70 have different functions, and although inactivation of each of them extends lifespan, the mechanisms by which this occurs are at least partially different from each other. In our study, the lifespan and healthspan assays were carried out for all these 3 genes and their combinations. As a result, we obtained a clear perspective of the lifespan and healthspan changes from all worm strains, which is represented in Figure 2 and Supplementary Figure 4, respectively. Using this representation, it can be easily observed that the odr-3; ife-2 double inactivation leads to an additive increase (i.e. the effect of the joint interventions is equal to the sum of the individual effects) in lifespan compared with single gene inactivations (Figure 2), as well as a synergistic effect on healthspan (Supplementary Figure 4). On the other hand, cku-70 down regulation does not significantly affect the lifespan of ife-2 mutants ( Figure 1C) and is detrimental to the long-lived odr-3 and odr-3; ife-2 strains ( Figure 1B, 1D). This detrimental effect is not seen in the daf-16 mutants, where the loss of both odr-3 and daf-16 seems to be dominant compared to the influence of cku-70 downregulation ( Figure 1F). It is difficult to explain the effect of cku-70 down regulation on odr-3 and odr-3; ife-2 only through its function in the DNA repair process. In addition to its role in DNA repair, the conserved Ku heterodimer was found to participate in other cellular processes such as transcriptional regulation, apoptosis, DNA replication, RNA metabolism and other [66,67]. Therefore, a more complex interaction between cku-70, odr-3 and ife-2 might exist. In our experiments we double inactivated ife-2 and cku-70 by RNAi. This raises the possibility that the knockdown efficiency of one or both genes could be reduced to half. For this reason, for an appropriate comparison with single RNAi down regulation we exposed the worms to a half concentration of dsRNA by mixing RNAi clone with EV clone. Unfortunately, we could not verify by quantitative PCR the efficiency of AGING ife-2 RNAi down regulation, which gave us the most notable effect, because the DNA fragment used for expression of dsRNA from L4440 vector encompasses the full gene (ORF and UTR) and due to the uptake of dsRNA in the cells, the endogenous ife-2 mRNA cannot be distinguished and targeted for PCR amplification. However, knock down of ife-2 RNAi; EV increased the lifespan of odr-3 mutants, suggesting efficient down regulation even with half concentration of ife-2 dsRNA. We have to point out that in the case of double RNAi there is a possibility that worms are not equally exposed to both dsRNA, and down regulation of one or both genes might be less efficient. Hence, although single downregulation of cku-70 RNAi gave only a very modest increase of lifespan, the effect of double inactivation ife-2(RNAi; cku-70(RNAi) should be considered with care. The increased lifespan of odr-3; ife-2 might be simply explained by the combined effect of two genes acting in distinct pathways. However, since the daf-16 mutation is epistatic to the odr-3 mutation and greatly reduces the lifespan of ife-2 knock-down animals, a more complex interaction is also possible. ODR-3 is expressed in 5 pairs of sensory neurons. In the AWA and AWC neurons, ODR-3 functions in perception and transduction of odor signals [37], in ADF and ASH mediates gustatory plasticity and detection of nociceptive stimuli, respectively [37,68], whereas in AWC neurons is involved in temperature sensing [69]. Among these neurons, only AWA and AWC neurons were found to modulate lifespan [36], therefore it is thought that ODR-3 regulates longevity by functioning in these neurons. However, ODR-3 has a role in modulating the adaptive behavior to different stimuli by adjusting the levels of second messenger cGMP in response to environmental cues. We found that WT C. elegans grown on OP50 and HT115 diets have similar lifespans (20.6 mean lifespan on HT115 vs. 19.9 mean lifespan on OP50), indicating an efficient adaptive response of WT animals to these bacterial diets, as previously reported [70,71]. In contrast, odr-3(n1605) mutants have an increased lifespan on HT115 diet (26% increase in mean lifespan on HT115 diet, comparative to 7.5% increase of mean lifespan on OP50 diet), which implicates ODR-3 in metabolic adaptation as it was found in the case of other metabolic genes [70,71]. IFE-2 is expressed in all soma cells, including neurons [42]. The most pronounced lifespan and healthspan extension of odr-3; ife-2 animals was observed when ife-2 was down regulated by RNAi. Since RNAi interference is known to affect all tissues of the WT animals with the exception of neurons, these findings raise the possibility that non-neuronal silencing of ife-2 by RNAi might be primarily responsible for the improved healthspan and extended longevity of the odr-3; ife-2(RNAi) animals. Alternatively, impaired protein synthesis in neurons due to ife-2 deficiency might be detrimental to nematode health by affecting neuronal proteostasis [72]. Growing evidence unravels an important role for cell non-autonomous regulation of proteostasis in aging in which neuronal activation of stress response pathways such as heat shock response, mitochondrial and ER unfolded protein responses regulate nematode longevity by modulating cellular proteostasis in distal cells [73,74]. DAF-16 stabilizes the transcriptome against the proteostasis collapse during aging by controlling the activity of hundreds of genes, integrating inputs from the DAF-2 pathway and from pathways that appear to regulate lifespan independently of DAF-2 [34,75]. Therefore, the genetic interaction between odr-3, ife-2 and daf-16 could take many forms. In our experiments, although ife-2 inactivation increased the lifespan of daf-16(mu86) mutants, a result that is in accordance with a previous report [42], it did not extend the lifespan beyond that of WT controls. We found that in contrast to odr-3 mutation, which weakly activates DAF-16 in posterior intestine, down regulation of ife-2 does not induce DAF-16 nuclear translocalization, implying that DAF-16 activity is not directly modulated by IFE-2. Both DAF-16 and IFE-2 could affect common processes such as metabolic remodeling and maintenance of cellular proteostasis that modulate longevity. Several metabolic changes were identified as fingerprints for long-lived mutants including the shift from carbon to amino acid catabolism as an alternative energy source, upregulation of lipid storage, increased purine metabolism and increased trehalose stores [43,76,77]. Many of these processes were found to be regulated in a DAF-16-dependent manner [75,[78][79][80][81]. In mev-1 mutants, which lack succinate dehydrogenase cytochrome b, depletion of ife-2 induces stress resistance but also restores WT lifespan [42]. A metabolomic study revealed that ife-2 deficiency does not revert the mitochondrial mev-1 defects, but rather restores the catabolism of purine nucleotides (e.g. GMP and AMP) and the metabolism of very long-chain fatty acids (VLFA) [82], processes related to peroxisomes. Since beta-oxidation of VLFA is a source of reactive oxygen species, and peroxisomes are sensitive to increased oxidative stress, ife-2 depletion could also protect peroxisomes from oxidative stress, hence ameliorating peroxisomal function. We found that the odr-3; ife-2 double mutants are less sensitive to induced ROS or H2O2, however a relationship between this and the additive/synergistic nature of the combined intervention cannot be directly AGING inferred. First, it was previously shown that stress resistance and lifespan can be experimentally dissociated and the magnitudes of changes in these two parameters produced by mutations are not identical [83]. Second, while it might be intuitive to suggest that the lack of additivity in the oxidative stress defence could mean the two genetic interventions activate the same mechanism, this is highly speculative, and small added differences in stress resistance could in fact affect longevity non linearly. We also found that the decrease of motility and pharyngeal pumping, which decline in an age-related manner, were delayed in the odr-3; ife-2 mutants. As seen in Figure 4E Table 1), it suggests that animals remain healthy for a longer period, while the physiological decline that occurs during the advanced stage of aging is seemingly unaffected. Using an analogy to the socio-economic implications in a human population (if such an intervention could be translatable), such a therapy would probably not reduce the healthcare costs during late senescence, however it would increase the Healthy Life Years (HLY) indicator, which is a measure of productivity during life and an important economic factor. Among the three genes that we investigated, the role of ife-2 in aging was the most comprehensively analyzed, so far. Thus, it was shown that the long-lived mutants, daf-2, age-1, let-363, clk-1, eat-2, dramatically extended the lifespan of ife-2 impaired animals [21,42]. There is limited information about interaction of odr-3 or cku-70 with other long-lived mutants. Both, odr-3(n1605) and cku-70RNAi extended the lifespan of daf-2(e1370) mutants [39,46]. We found that mutation in odr-3 extended the healthspan of ife-2 downregulated animals with a higher magnitude than it extended ife-2 lifespan, suggesting that the effect of odr-3 and ife-2 impairment may not be due to a role of these genes in the control of longevity per se, but rather a consequence of a longer healthspan due to amelioration of agerelated decline of physiological processes. This is supported by the observation that in contrast to eat-2 mutants which have reduced pharyngeal pumping, the odr-3; ife-2 animals exhibit a delay in the pharyngeal pumping decline, in older animals. While much more work is probably needed to fully explore the mechanistic way in which the interaction between odr-3 and ife-2 modulates longevity, our results show that the knock-down of both odr-3 and ife-2 increases resistance to some types of stress and additively extends lifespan and healthspan. MATERIALS AND METHODS Strains and culture conditions The following strains used in this study were provided by the Caenorhabditis Genetic Center (CGC): C. elegans wild-type Bristol strain (N2), CX3222 odr-3(n1605)V, RB579 ife-2(ok306), CF1038 daf-16(mu86)I, OH16024 daf- 16(ot971 [daf-16::GFP])I, E. coli OP50 and HT115(DE3) strains. The C. elegans strains were maintained at 20° C using standard methods [84]. Multiple mutants were obtained by standard genetic methods and the presence of mutations was tested either by screening for characteristic phenotypes or via PCR genotyping. The homozygous odr-3(n1605) allele was confirmed by negative chemotaxis tests to isoamyl alcohol. To confirm the presence of homozygous daf-16(mu86) allele, high density populations were allowed dauer formation and subsequently tested for resistance to SDS 1%. Presence of ife-2(ok306) deletion was confirmed by PCR genotyping. ife-2 and cku-70 RNAi For the RNAi-mediated gene knock-down by feeding method, a slightly modified protocol of the Ahringer technique was used [85]. Briefly, bacteria were grown overnight in LB medium supplemented with 50 µg/ml ampicillin and seeded onto NGM plates supplemented with 25 µg/ml carbenicillin and 1 mM IPTG. The plates were kept at room temperature for two days before use. Several L4 hermaphrodites picked from plates seeded with OP50 were placed onto RNAi plates, transferred the next day to other fresh RNAi plates, allowed to lay eggs for 24 h, then removed. The L4 hermaphrodites developed from eggs laid onto RNAi plates were used for longevity and healthspan assays. For double RNAi experiments, the plates were prepared in a similar way, with the exception that the plates were seeded with a 1:1 mixture of both RNAi bacterial clones. The ife-2 and cku-70 RNAi clones were obtained from the Ahringer RNAi library (Source BioScience, Nottingham, UK); both clones were validated by sequencing. The HT115 bacteria transformed with the L4440 empty vector, HT115 (EV), was used as control for RNAi experiments unless otherwise specified. When ife-2; cku-70 double RNAi was employed, the control worms were grown on plates seeded with a 1:1 density mixture of HT115 (EV) bacteria and ife-2 or cku-70 RNAi clone, respectively, to maintain the same concentration of each double strand RNA as in the strains subjected to double RNAi. AGING Lifespan assays Since the age at which a treatment is started can significantly influence the outcome [86], all worm cohorts used in this work have been age-synchronized (L4 larvae stage). For all RNAi experiments, agesynchronized L4 larvae were manually transferred to RNAi agar plates containing 15 µM 5-fluorodeoxyuridine (FUdR). For lifespan assays of mutant animals, age-synchronized L4 larvae were manually transferred to NGM plates, and seeded with OP50. In case of lifespan assays without FUdR, the worms were transferred to a new plate every day until they ceased laying eggs, then when needed. For mutant animals cultured with FUdR, a 15 µM FUdR concentration (same as in the RNAi experiments) was used. In all cases, worms were kept at 20° C and scored daily as dead or alive based on their response to a gentle touch with a wire. Worms that presented externalization of internal organs, died because of bagging, or crawled up the wall of the dish were censored. For RNAi experiments, the WT control and odr-3 animals were fed with HT115 (EV) bacteria. 85 worms were assayed per experiment. While the lifespan assays were not conducted in a blinded manner, as suggested by Gruber et al., [87], the experiments were carried out by 3 operators, working with the data independently and results were evaluated for consistency. From the beginning of the study, all operators aimed to treat worm cohorts in an unbiased fashion and keep them in the same conditions. Locomotion assay Animals were scored for free movement and for a response to prodding with a wire, daily, during the lifespan assay until death. Worms were classified into three motion stages, based on ability to engage and coordinate the body wall muscle in a forward or backward movement according to a previously described method [88], with slight modifications. The three stages considered were: 1) state A, corresponding to a physiological, fully mobile state, which included animals that could move more than 0.5 cm (freely or upon prodding); 2) state B, representing impaired animals that responded to the prodding, but did not have enough strength to move more than 0.5 cm, and 3) state C, which encompassed animals in a frailty state that barely exhibited head or tail movements or twitches upon prodding. Pharyngeal pumping assay For the pharyngeal pumping assay, separate worm cohorts were cultured, in three independent experiments, each with 60 animals grown on FUdRsupplemented plates seeded with OP50 bacteria. Pumping was monitored and recorded at 1, 5, 10 and 15 days post-L4 moult by filming 13-15 randomly selected worms at each time point. Time lapse recordings were obtained on the Zeiss SteREO Discovery. V20 stereomicroscope (Carl Zeiss AG, Jena, Germany) using the AxioVs40 V4.8.2.0 software (Carl Zeiss AG), using the 1X Plan Apo S objective, at a magnification of 150X; the time lapse captures were converted into.mp4 files using an in-house developed script. Pharyngeal contractions were then accurately counted during a 30seconds interval. For each strain the average number of pharyngeal movements per minute, standard deviation and standard error of the mean were computed. Stress assays and fecundity Tolerance to heat and oxidative stress was tested for late L4 animals, since responses to both stresses become repressed early in adulthood (starting as early as 4 hours post L4) [89], suggesting that collapse of cellular stress response could represent an early molecular event in the aging process. Heat stress assay To obtain synchronized populations, five hermaphrodites were let to lay eggs for about two hours on three replicate plates. The larvae were reared at 20° C up to late L4 stage, shifted to 35° C for four hours, then returned to 20° C. The percentage of alive animals was scored 48 h later. More than 420 animals were tested for each strain in three individual experiments. Oxidative stress assay Tolerance to oxidative stress was tested by exposure to paraquat and NaN3. Both assays were essentially performed as previously described [42]. Briefly, synchronized L4 larvae were transferred to NGM plates containing 2 mM paraquat and survival was scored on day 5 of exposure. For NaN3 tolerance, synchronized L4 larvae were collected, washed with M9, incubated for one hour with 0.5 M freshly made NaN3 in M9, washed with M9 and placed on a NGM plate to recover. Survival rates were determined after 24 hours. Brood size To analyze the total brood size, L4 worms were placed on NGM plates seeded with OP50 and transferred every 12h to a new plate until they ceased laying eggs. The number of eggs laid by each worm was counted after removal of the parent. DAF-16::GFP nuclear translocation To verify DAF-16 activation we used the CRISPR allele of daf-16 tagged at the C-terminus with GFP [90]. Since AGING DAF-16 translocation from cytoplasm to nucleus is induced by heat stress and common drugs used to anesthetize the worms [91], the young adult worms were fixed in 4% paraformaldehyde. The worms were grown at 20o C on RNAi plates, at L4 stage 15 µM FUdR was added and next day, the young adults were fixed and immediately visualized. Images were acquired with a Zeiss LSM 710 laser scanning confocal microscope using 10x objective, Argon 488 laser, and identical acquisition setting. Statistical analysis Comparisons between the lifespan values of different strains were carried out by analyzing Kaplan-Meier survival curves. For the statistical analysis and graphical representation of the curves, the R package "survival" was used (https://cran.r-project.org/package=survival). For the statistical analysis of locomotion, animals were scored by motion stage and plotted as motility curves, similar to the lifespan curves, to evaluate the decline rate of motility in each strain. While in the survival analysis an event represents the death of one worm, in the locomotion analysis an event was defined as the transition between motility states, modeling the population dynamics from increased to low motility (Supplementary Figure 4). Since three motility categories exist, two different motility analyses were conducted, corresponding to two types of events: i) transitions from the mobile state A to either the impaired state B or to the frail state C, and ii) transitions from either state A or B to state C. In the manuscript, only the results from the motility analysis of transitions from A vs. cumulated B and C is included, however the two analyses produced very similar results (data not shown). For all survival and locomotion analyses, the statistical significance was tested using the log-rank test (Mantel-Cox). Comparisons were performed against WT, WT(EV) or daf-16, as appropriately. If not otherwise specified, p<0.0001 was considered significant. The pvalues were corrected for multiple testing using the Benjamini-Hochberg method, at alpha=0.05. One-way ANOVA was used for the analysis of pharyngeal pumping, heat, paraquat and sodium azide stress resistance assays. All mutants were compared to wild type and statistical significance was assessed using Dunnett's test. represent either the death of a worm (A) or the transition from a healthy state to a sick state (becoming impaired or frail) (B). For the healthspan analysis, censored and dead worms are removed. (C-F) Comparison between healthspan curves. Statistical significance was determined using the log rank test. AUTHOR CONTRIBUTIONS Conceptualization Figure 1 . 1Kaplan-Meier survival curves depicting the effects of combined genetic interventions on odr-3, ife-2 and cku-70 at 20° C. (A-D) Lifespan comparisons in the WT background (continuous lines). (E-H) Lifespan comparisons in the daf-16(mu86) background (dashed lines). Survival curves represent: (A, E) odr-3(n1605) and ife-2(RNAi) single and double genetic interventions; (B, F) odr-3(n1605) and cku-70(RNAi) single and double genetic interventions; (C, G) ife-2(RNAi) and cku-70(RNAi) single and double genetic interventions; (D, H) odr-3(n1605), ife-2(RNAi) and cku-70(RNAi) double and triple genetic interventions. (C, D) Control in the case of single RNAi knock-downs refers to treatment with a 1:1 mixture of RNAi bacteria and EV bacteria, in order to be comparable to the double RNAi intervention. (A-H) The survival plots in the WT background represent pooled populations from 3 independent experiments, whereas survival plots in the daf-16(m28) background represent pooled populations from 2 independent experiments. odr-3 denotes odr-3(n1605) fed with EV; daf-16 denotes daf-16(mu86) fed with EV; all strains in these experiments were grown on agar plates with E. coli HT115(DE3) and FUdR. Figure 2 . 2Network schematic representation of the strains analyzed in this study and of the effects of each genetic intervention. Nodes represent the strains as follows: diamond for WT, circle for single gene interventions, square for double gene interventions, hexagon for triple gene interventions, and octagon for quadruple gene interventions. Nodes are positioned on the vertical axis according to their respective mean lifespan. Edges between worm strains are colored depending on the gain (or loss) in lifespan extension: increase (green), decrease (red) and small or non-significant change (gray). The extent of the change is included on the edge as a percentage increase/decrease between the origin and destination nodes of the edge. odr-3 and daf-16 denote mutants containing the odr-3(n1605) and daf-16(mu86) mutations; ife-2 and cku-70 denote animals in which these genes were modulated by RNAi bacteria. The white bars inside of the nodes indicate the mean ± SEM. Figure 3 .Figure 4 . 34Kaplan-Meier survival curves for animals containing the odr-3(n1605) and ife-2(ok306) mutations. (A) odr-3(n1605); ife-2(ok306) single and double mutants, cultivated in the presence of FUdR. (B) odr-3(n1605); ife-2(ok306) single and double mutants, cultured without FUdR. (C) Lifespan comparisons for odr-3 and ife-2 in the daf-16(mu86) background. (A-C) Dashed lines are used for odr-3(n1605) and ife-2(ok306) mutants tested in the daf16(mu86) genetic background, while continuous lines are used for WT or single/double odr-3 and ife-2 mutants tested in the WT background. All cohorts were fed OP50 and kept at 20° C. All lifespan values can be viewed in the Supplementary Table 1. AGING Healthspan of combined genetic interventions on odr-3 and ife-2 at 20° C. (A-D) Bar chart representation of motilityassessed healthspan illustrating the fraction of each category upon daily monitorization. Worms are grouped into three categories: mobile (white), impaired (light gray) and frail (dark gray). Dead and censored animals were subtracted from these analyses. (E) Mean number of days in each motility state throughout lifespan. The mean time spent in the impaired state is computed as the difference between the mean time spent as mobile or impaired, and the mean time spent in the mobile state. The mean time spent in the frail state is computed as the difference between the mean lifespan and mean time spent as mobile or impaired. The values within brackets represent the distribution of motion stages during the lifespan. (A-E) WT(EV) and odr-3 denote worms fed with RNAi(EV). (C, D) ife-2 and odr-3; ife-2 denote worms fed with ife-2 RNAi bacteria. (F) The pharyngeal pumping rate (average number of contractions per minute) of WT, odr-3(n1605), ife-2(ok306) and odr-3(n1605); ife-2(ok306) mutants were recorded on days 1, 5, 10 and 15 post-L4 moult. odr-3(n1605); ife-2(ok306) worms show a significantly slower decline of pharyngeal pumping with age, compared to WT. For simplicity, only significant differences among groups are indicated (one way ANOVA with Dunnett's test); ** denotes p < 0.01; *** denotes p < 0.001. AGING motility data of odr-3(n1605); ife-2(ok306) mutants, see Supplementary Figure 3A-3H; for the daf-16(mu86); odr-3(n1605); ife-2(ok306) mutants see Supplementary Figure 3I-3L). . Overall, our findings show that the ife-2 RNAi treatment and the double intervention odr-3(n1605); ife-2(RNAi) have a beneficial effect on motility-assessed healthspan, significantly increasing the period of full motility (SupplementaryFigure 5C; p < 2.0E-16) and exhibiting a decrease in the decrepit period of life. Similar data for the odr-3(n1605); ife-2(ok306) double mutant can be seen in SupplementaryFigure 5D, 5E and for the daf-16(mu86); odr-3(n1605); ife-2(ok306) triple mutant in SupplementaryFigure 5F. Figure 5 . 5Loss of odr-3 and ife-2 activity enhances oxidative stress tolerance. (A) Survival fraction of the indicated L4 larvae upon 5 days exposure to 0.2M paraquat (the experiment was repeated independently three times). (B) Survival fraction of the indicated strains upon 1 hour treatment with 0.5M NaN3 (the experiment was repeated independently four times). (C) Survival fraction of the indicated strains upon 4h heat stress at 35° C. Each strain was scored on three replicate plates and the experiment was repeated independently four times. (D) Egg-laying rate of the indicated strains. The average number of eggs laid by each strain was determined by transferring worms to new E. coli plates every 12 hours from L4 stage. (E) Brood size of the indicated strains at 20° C. Each point represents the total brood of one hermaphrodite. (A-E) Bars indicate the mean ± SEM. For simplicity, only significant differences among groups are indicated (one way ANOVA with Dunnett's test); * denotes p < 0.05; ** denotes p < 0.01. , the time spent by odr-3; ife-2 animals (in absolute values) in a frail state does not increase, although their lifespan increases compared to both WT and single mutants. Together with the fact that the double intervention extends both mean and maximum lifespan (Supplementary Figure 1 . 1, RT and SG; methodology, SG; lab work and validations, IVM, VNCS, SG; bioinformatics and programmatic tools, GB; all authors have participated to the analysis of results and to the writing Kaplan-Meier survival curves for worms fed bacteria expressing the target dsRNA or an equal mix of target dsRNA and EV. All survival plots represent pooled populations from 3 independent experiments. (A, B) Lifespan comparison of WT (A) or odr-3(n1605) (B) worms fed bacteria expressing ife-2 dsRNA or a 1:1 mixture of ife-2 dsRNA and empty-vector (EV). (C, D) Lifespan comparison of WT (C) or odr-3(n1605) (D) worms fed bacteria expressing cku-70 dsRNA or a mixture of cku-70 dsRNA and EV. Lifespan values are given in Supplementary Table 1. AGING Supplementary Figure 2. DAF-16::GFP nuclear translocation in odr-3; ife-2 (RNAi) mutant worms. Worms expressing daf-16 (ot971 [daf-16::GFP]) fluorescent marker in WT animals and odr-3(n1605), ife-2(RNAi) or odr-3(n1605); ife-2(RNAi) mutant background show nuclear accumulation of DAF-16::GFP in odr-3(n1605) mutants but not in WT, ife-2 or odr-3; ife-2 animals. Left panels show GFP images and right panels show Differential Interference Contrast (DIC) images captured with a confocal microscope. The arrow points to the nuclear accumulation of DAF-16::GFP in intestinal cells of odr-3 mutant. Images were obtained using the same confocal settings and exposure adjustments were uniformly applied in all images for better visualization. AGINGSupplementary Figure 4 . 4Network schematic representation of the strains analyzed in this study and of the effects of each genetic intervention. Nodes represent the strains as follows: diamond for WT, circle for single gene interventions, square for double gene interventions, hexagon for triple gene interventions, and octagon for quadruple gene interventions. Nodes are positioned on the vertical axis according to their respective mean healthspan. Edges between worm strains are colored depending on the gain (or loss) in lifespan extension: increase (green), decrease (red) and small or non-significant change (gray). The extent of the change is included on the edge as a percentage increase/decrease between the origin and destination nodes of the edge. odr-3 and daf-16 denote mutants containing the odr-3(n1605) and daf-16(mu86) mutations; ife-2 and cku-70 denote animals in which these genes were modulated by RNAi bacteria. The white bars inside of the nodes indicate the mean ± SEM. AGING Supplementary Figure 5. Method of quantifying the statistical significance for the healthspan difference between two worm populations. Kaplan Meier curves of survival (A) and healthspan (B), showing the probability of an event occurring over time. The events Table 1 . 1Mean lifespan of C. elegans strains with genetic interventions in the odr-3, ife-2 and cku-70.Strain RNAi* Mean lifespan days ± SD Effect vs control p-value Strain RNAi* Mean lifespan days ± SD Effect vs control p-value WT EV 20.6±0.2 [control] WT EV 20.6±0.2 [control] odr-3 26.0±0.3 26.2 % <2.0E-16 odr-3 26.0±0.3 26.2 % <2.00E-16 WT ife-2 24.3±0.2 18.0 % <2.0E-16 WT ife-2;EV 22.9±0.3 11.2 % 1.00E-10 WT cku-70 21.5±0.2 4.4 % 7.00E-03 WT cku-70;EV 21.9±0.2 6.3 % 1.00E-04 odr-3 ife-2 28.9±0.4 40.3 % <2.0E-16 odr-3 ife-2;EV 30.1±0.5 46.1 % <2.00E-16 odr-3 cku-70 23.4±0.5 13.6 % 4.00E-11 odr-3 cku-70;EV 26.5±0.4 28.6 % <2.00E-16 WT ife-2;cku-70 23.7±0.3 15.0 % 3.00E-16 odr-3 ife-2;cku-70 27.9±0.3 35.4 % <2.00E-16 *ife-2 denotes animals fed only with ife-2 RNAi bacteria, whereas ife-2;EV denotes animals fed with a mixture of ife- 2RNAi/RNAi(EV). Similar for cku-70. odr-3 denotes animals fed with RNAi(EV). (mu86) background. Similar to the RNAi experiments, the ok306 extended the lifespan of daf-16 single mutants (Figure 3C, 6.25% increase), while the double daf-16(mu86); odr-3(n1605) ACKNOWLEDGMENTSSome of the strains used in this study were provided by the CGC, which is funded by NIH Office of Research Infrastructure Programs (P40 OD010440). The authors would like to thank Anton Kulaga for helping with the development of the time-lapse conversion script.CONFLICTS OF INTERESTThe authors declare that they have no conflicts of interest.Please browse FullText version to see the data of Supplementary Tables 1, 2 The Mechanobiology of Aging. J M Phillip, I Aifuwa, J Walston, D Wirtz, 10.1146/annurev-bioeng-071114-04082926643020Annu Rev Biomed Eng. 17Phillip JM, Aifuwa I, Walston J, Wirtz D. The Mechanobiology of Aging. Annu Rev Biomed Eng. 2015; 17:113-41. https://doi.org/10.1146/annurev-bioeng-071114- 040829 PMID:26643020 Defining aging. M Lemoine, 10.1007/s10539-020-09765-zBiol Philos. 3546Lemoine M. Defining aging. Biol Philos. 2020; 35:46. https://doi.org/10.1007/s10539-020-09765-z New model of health promotion and disease prevention for the 21st century. R N Butler, R A Miller, D Perry, B A Carnes, T F Williams, C Cassel, J Brody, M A Bernard, L Partridge, T Kirkwood, G M Martin, S J Olshansky, 10.1136/bmj.a39918614506BMJ. 337399Butler RN, Miller RA, Perry D, Carnes BA, Williams TF, Cassel C, Brody J, Bernard MA, Partridge L, Kirkwood T, Martin GM, Olshansky SJ. New model of health promotion and disease prevention for the 21st century. BMJ. 2008; 337:a399. https://doi.org/10.1136/bmj.a399 PMID:18614506 Dividends" from research on aging--can biogerontologists, at long last, find something useful to do?. R A Miller, 10.1093/gerona/gln06219225032J Gerontol A Biol Sci Med Sci. 64Miller RA. "Dividends" from research on aging--can biogerontologists, at long last, find something useful to do? J Gerontol A Biol Sci Med Sci. 2009; 64:157-60. https://doi.org/10.1093/gerona/gln062 PMID:19225032 Infrastructure and resources for an aging population: embracing complexity in translational research. K P High, 10.1016/j.trsl.2013.09.00124095640Transl Res. 163High KP. Infrastructure and resources for an aging population: embracing complexity in translational research. Transl Res. 2014; 163:446-55. https://doi.org/10.1016/j.trsl.2013.09.001 PMID:24095640 The genetics of ageing. C J Kenyon, 10.1038/nature0898020336132Nature. 464Kenyon CJ. The genetics of ageing. Nature. 2010; 464:504-12. https://doi.org/10.1038/nature08980 PMID:20336132 AGING Facing up to the global challenges of ageing. L Partridge, J Deelen, P E Slagboom, 10.1038/s41586-018-0457-830185958Nature. 561Partridge L, Deelen J, Slagboom PE. Facing up to the global challenges of ageing. Nature. 2018; 561:45-56. https://doi.org/10.1038/s41586-018-0457-8 PMID:30185958 Developmental drift as a mechanism for aging: lessons from nematodes. M Lezzerini, R L Smith, Y Budovskaya, 10.1007/s10522-013-9462-3Biogerontology. 1424122213Lezzerini M, Smith RL, Budovskaya Y. Developmental drift as a mechanism for aging: lessons from nematodes. Biogerontology. 2013; 14:693-701. https://doi.org/10.1007/s10522-013-9462-3 PMID:24122213 Human Ageing Genomic Resources: new and updated databases. R Tacutu, D Thornton, Johnson E Budovsky, A Barardo, D Craig, T , Diana E Lehmann, G Toren, D Wang, J Fraifeld, V E De Magalhães, J P , 10.1093/nar/gkx104229121237Nucleic Acids Res. 46Tacutu R, Thornton D, Johnson E, Budovsky A, Barardo D, Craig T, Diana E, Lehmann G, Toren D, Wang J, Fraifeld VE, de Magalhães JP. Human Ageing Genomic Resources: new and updated databases. Nucleic Acids Res. 2018; 46:D1083-90. https://doi.org/10.1093/nar/gkx1042 PMID:29121237 Wide-scale comparative analysis of longevity genes and interventions. H Yanai, A Budovsky, T Barzilay, R Tacutu, V E Fraifeld, 10.1111/acel.1265928836369Aging Cell. 16Yanai H, Budovsky A, Barzilay T, Tacutu R, Fraifeld VE. Wide-scale comparative analysis of longevity genes and interventions. Aging Cell. 2017; 16:1267-75. https://doi.org/10.1111/acel.12659 PMID:28836369 Gray whale transcriptome reveals longevity adaptations associated with DNA repair and ubiquitination. D Toren, A Kulaga, M Jethva, E Rubin, A V Snezhkina, A V Kudryavtseva, D Nowicki, R Tacutu, A A Moskalev, V E Fraifeld, 10.1111/acel.1315832515539Aging Cell. 19Toren D, Kulaga A, Jethva M, Rubin E, Snezhkina AV, Kudryavtseva AV, Nowicki D, Tacutu R, Moskalev AA, Fraifeld VE. Gray whale transcriptome reveals longevity adaptations associated with DNA repair and ubiquitination. Aging Cell. 2020; 19:e13158. https://doi.org/10.1111/acel.13158 PMID:32515539 Machine Learning Analysis of Longevity-Associated Gene Expression Landscapes in Mammals. A Y Kulaga, E Ursu, D Toren, V Tyshchenko, R Guinea, M Pushkova, V E Fraifeld, R Tacutu, 10.3390/ijms2203107333499037Int J Mol Sci. 221073Kulaga AY, Ursu E, Toren D, Tyshchenko V, Guinea R, Pushkova M, Fraifeld VE, Tacutu R. Machine Learning Analysis of Longevity-Associated Gene Expression Landscapes in Mammals. Int J Mol Sci. 2021; 22:1073. https://doi.org/10.3390/ijms22031073 PMID:33499037 elegans mutant that lives twice as long as wild type. C Kenyon, J Chang, E Gensch, A Rudner, R Tabtiang, 10.1038/366461a08247153Nature. 366Kenyon C, Chang J, Gensch E, Rudner A, Tabtiang R. A C. elegans mutant that lives twice as long as wild type. Nature. 1993; 366:461-64. https://doi.org/10.1038/366461a0 PMID:8247153 Sir2 blocks extreme life-span extension. P Fabrizio, C Gattazzo, L Battistella, M Wei, C Cheng, K Mcgrew, V D Longo, 10.1016/j.cell.2005.08.04216286010Cell. 123Fabrizio P, Gattazzo C, Battistella L, Wei M, Cheng C, McGrew K, Longo VD. Sir2 blocks extreme life-span extension. Cell. 2005; 123:655-67. https://doi.org/10.1016/j.cell.2005.08.042 PMID:16286010 Remarkable longevity and stress resistance of nematode PI3K-null mutants. S Ayyadevara, Alla R Thaden, J J , Shmookler Reis, R J , 10.1111/j.1474-9726.2007.00348.x17996009Aging Cell. 7Ayyadevara S, Alla R, Thaden JJ, Shmookler Reis RJ. Remarkable longevity and stress resistance of nematode PI3K-null mutants. Aging Cell. 2008; 7:13-22. https://doi.org/10.1111/j.1474-9726.2007.00348.x PMID:17996009 Epistasis--the essential role of gene interactions in the structure and evolution of genetic systems. P C Phillips, 10.1038/nrg245218852697Nat Rev Genet. 9Phillips PC. Epistasis--the essential role of gene interactions in the structure and evolution of genetic systems. Nat Rev Genet. 2008; 9:855-67. https://doi.org/10.1038/nrg2452 PMID:18852697 Epistasis and quantitative traits: using model organisms to study gene-gene interactions. T F Mackay, 10.1038/nrg362724296533Nat Rev Genet. 15Mackay TF. Epistasis and quantitative traits: using model organisms to study gene-gene interactions. Nat Rev Genet. 2014; 15:22-33. https://doi.org/10.1038/nrg3627 PMID:24296533 SynergyAge, a curated database for synergistic and antagonistic interactions of longevity-associated genes. G Bunu, D Toren, C F Ion, D Barardo, L Sârghie, L G Grigore, J P De Magalhães, V E Fraifeld, R Tacutu, 10.1038/s41597-020-00710-zSci Data. 733106474Bunu G, Toren D, Ion CF, Barardo D, Sârghie L, Grigore LG, de Magalhães JP, Fraifeld VE, Tacutu R. SynergyAge, a curated database for synergistic and antagonistic interactions of longevity-associated genes. Sci Data. 2020; 7:366. https://doi.org/10.1038/s41597-020-00710-z PMID:33106474 Long live FOXO: unraveling the role of FOXO proteins in aging and longevity. R Martins, G J Lithgow, W Link, 10.1111/acel.1242726643314Aging Cell. 15Martins R, Lithgow GJ, Link W. Long live FOXO: unraveling the role of FOXO proteins in aging and longevity. Aging Cell. 2016; 15:196-207. https://doi.org/10.1111/acel.12427 PMID:26643314 Biochemical Genetic Pathways that Modulate Aging in Multiple Species. A Bitto, A M Wang, C F Bennett, M Kaeberlein, 10.1101/cshperspect.a02511426525455Cold Spring Harb Perspect Med. 525114Bitto A, Wang AM, Bennett CF, Kaeberlein M. Biochemical Genetic Pathways that Modulate Aging in Multiple Species. Cold Spring Harb Perspect Med. 2015; 5:a025114. https://doi.org/10.1101/cshperspect.a025114 PMID:26525455 Lifespan extension by conditions that inhibit translation in Caenorhabditis elegans. M Hansen, S Taubert, D Crawford, N Libina, S J Lee, C Kenyon, 10.1111/j.1474-9726.2006.00267.xAging Cell. 617266679Hansen M, Taubert S, Crawford D, Libina N, Lee SJ, Kenyon C. Lifespan extension by conditions that inhibit translation in Caenorhabditis elegans. Aging Cell. 2007; 6:95-110. https://doi.org/10.1111/j.1474-9726.2006.00267.x PMID:17266679 Deletion of the mitochondrial superoxide dismutase sod-2 extends lifespan in Caenorhabditis elegans. J M Van Raamsdonk, S Hekimi, 10.1371/journal.pgen.100036119197346PLoS Genet. 51000361Van Raamsdonk JM, Hekimi S. Deletion of the mitochondrial superoxide dismutase sod-2 extends lifespan in Caenorhabditis elegans. PLoS Genet. 2009; 5:e1000361. https://doi.org/10.1371/journal.pgen.1000361 PMID:19197346 Determination of life-span in Caenorhabditis elegans by four clock genes. B Lakowski, S Hekimi, 10.1126/science.272.5264.10108638122Science. 272Lakowski B, Hekimi S. Determination of life-span in Caenorhabditis elegans by four clock genes. Science. 1996; 272:1010-13. https://doi.org/10.1126/science.272.5264.1010 PMID:8638122 Germline signaling mediates the synergistically prolonged longevity produced by double mutations in daf-2 and rsks-1 in C. elegans. D Chen, P W Li, B A Goldstein, W Cai, E L Thomas, F Chen, A E Hubbard, S Melov, P Kapahi, 10.1016/j.celrep.2013.11.01824332851Cell Rep. 5Chen D, Li PW, Goldstein BA, Cai W, Thomas EL, Chen F, Hubbard AE, Melov S, Kapahi P. Germline signaling mediates the synergistically prolonged longevity produced by double mutations in daf-2 and rsks-1 in C. elegans. Cell Rep. 2013; 5:1600-10. https://doi.org/10.1016/j.celrep.2013.11.018 PMID:24332851 The dynaminrelated protein DRP-1 and the insulin signaling pathway cooperate to modulate Caenorhabditis elegans longevity. C C Yang, D Chen, S S Lee, L Walter, 10.1111/j.1474-9726.2011.00711.x21463460Aging Cell. 10Yang CC, Chen D, Lee SS, Walter L. The dynamin- related protein DRP-1 and the insulin signaling pathway cooperate to modulate Caenorhabditis elegans longevity. Aging Cell. 2011; 10:724-28. https://doi.org/10.1111/j.1474-9726.2011.00711.x PMID:21463460 Decreased energy metabolism extends life span in Caenorhabditis elegans without reducing oxidative damage. J M Van Raamsdonk, Y Meng, D Camp, W Yang, X Jia, C Bénard, S Hekimi, 10.1534/genetics.110.11537820382831Genetics. 185Van Raamsdonk JM, Meng Y, Camp D, Yang W, Jia X, Bénard C, Hekimi S. Decreased energy metabolism extends life span in Caenorhabditis elegans without reducing oxidative damage. Genetics. 2010; 185:559-71. https://doi.org/10.1534/genetics.110.115378 PMID:20382831 Regulation of lifespan by sensory perception in Caenorhabditis elegans. J Apfeld, C Kenyon, 10.1038/4554410617200Nature. 402Apfeld J, Kenyon C. Regulation of lifespan by sensory perception in Caenorhabditis elegans. Nature. 1999; 402:804-09. https://doi.org/10.1038/45544 PMID:10617200 daf-2, daf-16 and daf-23: genetically interacting genes controlling Dauer formation in Caenorhabditis elegans. S Gottlieb, G Ruvkun, 10.1093/genetics/137.1.1078056303Genetics. 137Gottlieb S, Ruvkun G. daf-2, daf-16 and daf-23: genetically interacting genes controlling Dauer formation in Caenorhabditis elegans. Genetics. 1994; 137:107-20. https://doi.org/10.1093/genetics/137.1.107 PMID:8056303 Two pleiotropic classes of daf-2 mutation affect larval arrest, adult behavior, reproduction and longevity in Caenorhabditis elegans. D Gems, A J Sutton, M L Sundermeyer, P S Albert, K V King, M L Edgley, P L Larsen, D L Riddle, 10.1093/genetics/150.1.1299725835Genetics. 150Gems D, Sutton AJ, Sundermeyer ML, Albert PS, King KV, Edgley ML, Larsen PL, Riddle DL. Two pleiotropic classes of daf-2 mutation affect larval arrest, adult behavior, reproduction and longevity in Caenorhabditis elegans. Genetics. 1998; 150:129-55. https://doi.org/10.1093/genetics/150.1.129 PMID:9725835 daf-2, an insulin receptor-like gene that regulates longevity and diapause in Caenorhabditis elegans. K D Kimura, H A Tissenbaum, Y Liu, G Ruvkun, 10.1126/science.277.5328.9429252323Science. 277Kimura KD, Tissenbaum HA, Liu Y, Ruvkun G. daf-2, an insulin receptor-like gene that regulates longevity and diapause in Caenorhabditis elegans. Science. 1997; 277:942-46. https://doi.org/10.1126/science.277.5328.942 PMID:9252323 The C. elegans DAF-2 insulin-like receptor is abundantly expressed in the nervous system and regulated by nutritional status. K D Kimura, D L Riddle, G Ruvkun, 10.1101/sqb.2011.76.01066022123849Cold Spring Harb Symp Quant Biol. 76Kimura KD, Riddle DL, Ruvkun G. The C. elegans DAF-2 insulin-like receptor is abundantly expressed in the nervous system and regulated by nutritional status. Cold Spring Harb Symp Quant Biol. 2011; 76:113-20. https://doi.org/10.1101/sqb.2011.76.010660 PMID:22123849 daf-16: An HNF-3/forkhead family member that can function to double the life-span of Caenorhabditis elegans. K Lin, J B Dorman, A Rodan, C Kenyon, 10.1126/science.278.5341.13199360933Science. 278Lin K, Dorman JB, Rodan A, Kenyon C. daf-16: An HNF- 3/forkhead family member that can function to double the life-span of Caenorhabditis elegans. Science. 1997; 278:1319-22. https://doi.org/10.1126/science.278.5341.1319 PMID:9360933 The Fork head transcription factor DAF-16 transduces insulin-like metabolic and longevity signals in C. elegans. S Ogg, S Paradis, S Gottlieb, G I Patterson, L Lee, H A Tissenbaum, G Ruvkun, 10.1038/401949353126Nature. 389Ogg S, Paradis S, Gottlieb S, Patterson GI, Lee L, Tissenbaum HA, Ruvkun G. The Fork head transcription factor DAF-16 transduces insulin-like metabolic and longevity signals in C. elegans. Nature. 1997; 389:994-99. https://doi.org/10.1038/40194 PMID:9353126 DAF-16: FOXO in the Context of C. elegans. H A Tissenbaum, 10.1016/bs.ctdb.2017.11.00729433733Curr Top Dev Biol. 127Tissenbaum HA. DAF-16: FOXO in the Context of C. elegans. Curr Top Dev Biol. 2018; 127:1-21. https://doi.org/10.1016/bs.ctdb.2017.11.007 PMID:29433733 Odorantselective genes and neurons mediate olfaction in C. elegans. C I Bargmann, E Hartwieg, H R Horvitz, 10.1016/0092-8674(93)80053-h8348618Cell. 74Bargmann CI, Hartwieg E, Horvitz HR. Odorant- selective genes and neurons mediate olfaction in C. elegans. Cell. 1993; 74:515-27. https://doi.org/10.1016/0092-8674(93)80053-h PMID:8348618 Regulation of C. elegans longevity by specific gustatory and olfactory neurons. J Alcedo, C Kenyon, 10.1016/s0896-6273(03)00816-x14715134Neuron. 41Alcedo J, Kenyon C. Regulation of C. elegans longevity by specific gustatory and olfactory neurons. Neuron. 2004; 41:45-55. https://doi.org/10.1016/s0896-6273(03)00816-x PMID:14715134 The G alpha protein ODR-3 mediates olfactory and nociceptive function and controls cilium morphogenesis in C. elegans olfactory neurons. K Roayaie, J G Crump, A Sagasti, C I Bargmann, 10.1016/s0896-6273(00)80434-19459442Neuron. 20Roayaie K, Crump JG, Sagasti A, Bargmann CI. The G alpha protein ODR-3 mediates olfactory and nociceptive function and controls cilium morphogenesis in C. elegans olfactory neurons. Neuron. 1998; 20:55-67. https://doi.org/10.1016/s0896-6273(00)80434-1 PMID:9459442 A network of stimulatory and inhibitory Galpha-subunits regulates olfaction in Caenorhabditis elegans. H Lans, S Rademakers, G Jansen, 10.1534/genetics.103.02478615342507Genetics. 167Lans H, Rademakers S, Jansen G. A network of stimulatory and inhibitory Galpha-subunits regulates olfaction in Caenorhabditis elegans. Genetics. 2004; 167:1677-87. https://doi.org/10.1534/genetics.103.024786 PMID:15342507 Multiple sensory G proteins in the olfactory, gustatory and nociceptive neurons modulate longevity in Caenorhabditis elegans. H Lans, G Jansen, 10.1016/j.ydbio.2006.11.02817187771Dev Biol. 303Lans H, Jansen G. Multiple sensory G proteins in the olfactory, gustatory and nociceptive neurons modulate longevity in Caenorhabditis elegans. Dev Biol. 2007; 303:474-82. https://doi.org/10.1016/j.ydbio.2006.11.028 PMID:17187771 Polygenic control of Caenorhabditis elegans fat storage. H Y Mak, L S Nelson, M Basson, C D Johnson, G Ruvkun, 10.1038/ng173916462744Nat Genet. 38Mak HY, Nelson LS, Basson M, Johnson CD, Ruvkun G. Polygenic control of Caenorhabditis elegans fat storage. Nat Genet. 2006; 38:363-68. https://doi.org/10.1038/ng1739 PMID:16462744 Beneficial and Detrimental Effects of Reactive Oxygen Species on Lifespan: A Comprehensive Review of Comparative and Experimental Studies. H J Shields, A Traa, J M Van Raamsdonk, 10.3389/fcell.2021.62815733644065Front Cell Dev Biol. 9628157Shields HJ, Traa A, Van Raamsdonk JM. Beneficial and Detrimental Effects of Reactive Oxygen Species on Lifespan: A Comprehensive Review of Comparative and Experimental Studies. Front Cell Dev Biol. 2021; 9:628157. https://doi.org/10.3389/fcell.2021.628157 PMID:33644065 eIF4E function in somatic cells modulates ageing in Caenorhabditis elegans. P Syntichaki, K Troulinaki, N Tavernarakis, 10.1038/nature0560317277769Nature. 445Syntichaki P, Troulinaki K, Tavernarakis N. eIF4E function in somatic cells modulates ageing in Caenorhabditis elegans. Nature. 2007; 445:922-26. https://doi.org/10.1038/nature05603 PMID:17277769 A metabolic signature of long life in Caenorhabditis elegans. S Fuchs, J G Bundy, S K Davies, J M Viney, J S Swire, A M Leroi, 10.1186/1741-7007-8-1420146810BMC Biol. 814Fuchs S, Bundy JG, Davies SK, Viney JM, Swire JS, Leroi AM. A metabolic signature of long life in Caenorhabditis elegans. BMC Biol. 2010; 8:14. https://doi.org/10.1186/1741-7007-8-14 PMID:20146810 Disruption of DNA-PK in Ku80 mutant xrs-6 and the implications in DNA double-strand break repair. F Chen, S R Peterson, M D Story, D J Chen, 10.1016/0921-8777(95)00026-78538653Mutat Res. 362Chen F, Peterson SR, Story MD, Chen DJ. Disruption of DNA-PK in Ku80 mutant xrs-6 and the implications in DNA double-strand break repair. Mutat Res. 1996; 362:9-19. https://doi.org/10.1016/0921-8777(95)00026-7 PMID:8538653 Developmental modulation of nonhomologous end joining in Caenorhabditis elegans. I Clejan, J Boerckel, S Ahmed, 10.1534/genetics.106.05862816702421Genetics. 173Clejan I, Boerckel J, Ahmed S. Developmental modulation of nonhomologous end joining in Caenorhabditis elegans. Genetics. 2006; 173:1301-17. https://doi.org/10.1534/genetics.106.058628 PMID:16702421 The C. elegans ortholog of mammalian Ku70, interacts with insulinlike signaling to modulate stress resistance and life span. G Mccoll, M C Vantipalli, G J Lithgow, 10.1096/fj.04-2447fje16099946FASEB J. 19McColl G, Vantipalli MC, Lithgow GJ. The C. elegans ortholog of mammalian Ku70, interacts with insulin- like signaling to modulate stress resistance and life span. FASEB J. 2005; 19:1716-18. https://doi.org/10.1096/fj.04-2447fje PMID:16099946 DAF-16/FOXO Transcription Factor in Aging and Longevity. X Sun, W D Chen, Y D Wang, 10.3389/fphar.2017.0054828878670Front Pharmacol. 8548Sun X, Chen WD, Wang YD. DAF-16/FOXO Transcription Factor in Aging and Longevity. Front Pharmacol. 2017; 8:548. https://doi.org/10.3389/fphar.2017.00548 PMID:28878670 Longevity Genes Revealed by Integrative Analysis of Isoform-Specific daf-16/FoxO Mutants of Caenorhabditis elegans. A T Chen, C Guo, O A Itani, B G Budaitis, T W Williams, C E Hopkins, R C Mceachin, M Pande, A R Grant, S Yoshina, S Mitani, P J Hu, 10.1534/genetics.115.17799826219299Genetics. 201Chen AT, Guo C, Itani OA, Budaitis BG, Williams TW, Hopkins CE, McEachin RC, Pande M, Grant AR, Yoshina S, Mitani S, Hu PJ. Longevity Genes Revealed by Integrative Analysis of Isoform-Specific daf-16/FoxO Mutants of Caenorhabditis elegans. Genetics. 2015; 201:613-29. https://doi.org/10.1534/genetics.115.177998 PMID:26219299 The use of FUdR can cause prolonged longevity in mutant nematodes. L Aitlhadj, S R Stürzenbaum, 10.1016/j.mad.2010.03.00220236608Mech Ageing Dev. 131Aitlhadj L, Stürzenbaum SR. The use of FUdR can cause prolonged longevity in mutant nematodes. Mech Ageing Dev. 2010; 131:364-65. https://doi.org/10.1016/j.mad.2010.03.002 PMID:20236608 FUdR causes a twofold increase in the lifespan of the mitochondrial mutant gas-1. J M Van Raamsdonk, S Hekimi, 10.1016/j.mad.2011.08.00621893079Mech Ageing Dev. 132Van Raamsdonk JM, Hekimi S. FUdR causes a twofold increase in the lifespan of the mitochondrial mutant gas-1. Mech Ageing Dev. 2011; 132:519-21. https://doi.org/10.1016/j.mad.2011.08.006 PMID:21893079 Fluorodeoxyuridine affects the identification of metabolic responses to daf-2 status in Caenorhabditis elegans. S K Davies, A M Leroi, J G Bundy, 10.1016/j.mad.2011.11.00222116032Mech Ageing Dev. 133Davies SK, Leroi AM, Bundy JG. Fluorodeoxyuridine affects the identification of metabolic responses to daf- 2 status in Caenorhabditis elegans. Mech Ageing Dev. 2012; 133:46-49. https://doi.org/10.1016/j.mad.2011.11.002 PMID:22116032 Measurements of agerelated changes of physiological processes that predict lifespan of Caenorhabditis elegans. C Huang, C Xiong, K Kornfeld, 10.1073/pnas.040084810115141086Proc Natl Acad Sci USA. 101Huang C, Xiong C, Kornfeld K. Measurements of age- related changes of physiological processes that predict lifespan of Caenorhabditis elegans. Proc Natl Acad Sci USA. 2004; 101:8084-89. https://doi.org/10.1073/pnas.0400848101 PMID:15141086 Aging causes decreased resistance to multiple stresses and a failure to activate specific stress response pathways. D J Dues, E K Andrews, C E Schaar, A L Bergsma, M M Senchuk, J M Van Raamsdonk, 10.18632/aging.10093927053445Aging (Albany NY). 8Dues DJ, Andrews EK, Schaar CE, Bergsma AL, Senchuk MM, Van Raamsdonk JM. Aging causes decreased resistance to multiple stresses and a failure to activate specific stress response pathways. Aging (Albany NY). 2016; 8:777-95. https://doi.org/10.18632/aging.100939 PMID:27053445 Hallmarks of Health. C López-Otín, G Kroemer, 10.1016/j.cell.2020.11.03433340459Cell. 184López-Otín C, Kroemer G. Hallmarks of Health. Cell. 2021; 184:33-63. https://doi.org/10.1016/j.cell.2020.11.034 PMID:33340459 Evidence for a trade-off between survival and fitness caused by resveratrol treatment of Caenorhabditis elegans. J Gruber, S Y Tang, B Halliwell, 10.1196/annals.1395.05917460219Ann N Y Acad Sci. 1100Gruber J, Tang SY, Halliwell B. Evidence for a trade-off between survival and fitness caused by resveratrol treatment of Caenorhabditis elegans. Ann N Y Acad Sci. 2007; 1100:530-42. https://doi.org/10.1196/annals.1395.059 PMID:17460219 The mysterious relationship between reproduction and longevity. H Aguilaniu, 10.1080/21624054.2015.102027626430561Worm. 41020276Aguilaniu H. The mysterious relationship between reproduction and longevity. Worm. 2015; 4:e1020276. https://doi.org/10.1080/21624054.2015.1020276 PMID:26430561 Reproduction and longevity: secrets revealed by C. elegans. A Mukhopadhyay, H A Tissenbaum, 10.1016/j.tcb.2006.12.00417187981Trends Cell Biol. 17Mukhopadhyay A, Tissenbaum HA. Reproduction and longevity: secrets revealed by C. elegans. Trends Cell Biol. 2007; 17:65-71. https://doi.org/10.1016/j.tcb.2006.12.004 PMID:17187981 The hallmarks of aging. C López-Otín, M A Blasco, L Partridge, M Serrano, G Kroemer, 10.1016/j.cell.2013.05.03923746838Cell. 153López-Otín C, Blasco MA, Partridge L, Serrano M, Kroemer G. The hallmarks of aging. Cell. 2013; 153:1194-217. https://doi.org/10.1016/j.cell.2013.05.039 PMID:23746838 Healthspan pathway maps in C. elegans and humans highlight transcription, proliferation/biosynthesis and lipids. S Möller, N Saul, A A Cohen, R Köhling, S Sender, Murua Escobar, H Junghanss, C Cirulli, F Berry, A Antal, P Adler, P Vilo, J Boiani, M , 10.18632/aging.10351432634117Aging (Albany NY). 12Möller S, Saul N, Cohen AA, Köhling R, Sender S, Murua Escobar H, Junghanss C, Cirulli F, Berry A, Antal P, Adler P, Vilo J, Boiani M, et al. Healthspan pathway maps in C. elegans and humans highlight transcription, proliferation/biosynthesis and lipids. Aging (Albany NY). 2020; 12:12534-81. https://doi.org/10.18632/aging.103514 PMID:32634117 Prediction of C. elegans longevity genes by human and worm longevity networks. R Tacutu, D E Shore, A Budovsky, J P De Magalhães, G Ruvkun, V E Fraifeld, S P Curran, 10.1371/journal.pone.004828223144747PLoS One. 7Tacutu R, Shore DE, Budovsky A, de Magalhães JP, Ruvkun G, Fraifeld VE, Curran SP. Prediction of C. elegans longevity genes by human and worm longevity networks. PLoS One. 2012; 7:e48282. https://doi.org/10.1371/journal.pone.0048282 PMID:23144747 A network pharmacology approach reveals new candidate caloric restriction mimetics in C. elegans. S Calvert, R Tacutu, S Sharifi, R Teixeira, P Ghosh, J P De Magalhães, 10.1111/acel.1243226676933Aging Cell. 15Calvert S, Tacutu R, Sharifi S, Teixeira R, Ghosh P, de Magalhães JP. A network pharmacology approach reveals new candidate caloric restriction mimetics in C. elegans. Aging Cell. 2016; 15:256-66. https://doi.org/10.1111/acel.12432 PMID:26676933 The signaling hubs at the crossroad of longevity and agerelated disease networks. M Wolfson, A Budovsky, R Tacutu, V Fraifeld, Int J Biochem Cell Biol. 41Wolfson M, Budovsky A, Tacutu R, Fraifeld V. The signaling hubs at the crossroad of longevity and age- related disease networks. Int J Biochem Cell Biol. 2009; 41:516-20. . 10.1016/j.biocel.2008.08.02618793745https://doi.org/10.1016/j.biocel.2008.08.026 PMID:18793745 The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. R Tacutu, A Budovsky, V E Fraifeld, 10.1007/s10522-010-9265-820186480Biogerontology. 11Tacutu R, Budovsky A, Fraifeld VE. The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. Biogerontology. 2010; 11:513-22. https://doi.org/10.1007/s10522-010-9265-8 PMID:20186480 MicroRNA-regulated protein-protein interaction networks: how could they help in searching for prolongevity targets?. R Tacutu, A Budovsky, M Wolfson, V E Fraifeld, 10.1089/rej.2009.098020367577Rejuvenation Res. 13Tacutu R, Budovsky A, Wolfson M, Fraifeld VE. MicroRNA-regulated protein-protein interaction networks: how could they help in searching for pro- longevity targets? Rejuvenation Res. 2010; 13:373-77. https://doi.org/10.1089/rej.2009.0980 PMID:20367577 Drug Synergy Slows Aging and Improves Healthspan through IGF and SREBP Lipid Signaling. T D Admasu, Chaithanya Batchu, K Barardo, D Ng, L F Lam, V Y Xiao, L Cazenave-Gassiot, A Wenk, M R Tolwinski, N S Gruber, J , 10.1016/j.devcel.2018.09.00130269951Dev Cell. 47Admasu TD, Chaithanya Batchu K, Barardo D, Ng LF, Lam VY, Xiao L, Cazenave-Gassiot A, Wenk MR, Tolwinski NS, Gruber J. Drug Synergy Slows Aging and Improves Healthspan through IGF and SREBP Lipid Signaling. Dev Cell. 2018; 47:67-79.e5. https://doi.org/10.1016/j.devcel.2018.09.001 PMID:30269951 The Ku complex: recent advances and emerging roles outside of non-homologous endjoining. S Abbasi, G Parmar, R D Kelly, N Balasuriya, C Schild-Poulter, 10.1007/s00018-021-03801-1Cell Mol Life Sci. 7833855626Abbasi S, Parmar G, Kelly RD, Balasuriya N, Schild- Poulter C. The Ku complex: recent advances and emerging roles outside of non-homologous end- joining. Cell Mol Life Sci. 2021; 78:4589-613. https://doi.org/10.1007/s00018-021-03801-1 PMID:33855626 Apoptosis Induction byHistone Deacetylase Inhibitors in Cancer Cells: Role of Ku70. P Gong, Y Wang, Y Jing, 10.3390/ijms2007160130935057Int J Mol Sci. 201601Gong P, Wang Y, Jing Y. Apoptosis Induction byHistone Deacetylase Inhibitors in Cancer Cells: Role of Ku70. Int J Mol Sci. 2019; 20:1601. https://doi.org/10.3390/ijms20071601 PMID:30935057 Antagonistic sensory cues generate gustatory plasticity in Caenorhabditis elegans. R K Hukema, S Rademakers, M P Dekkers, J Burghoorn, G Jansen, 10.1038/sj.emboj.760094016407969EMBO J. 25Hukema RK, Rademakers S, Dekkers MP, Burghoorn J, Jansen G. Antagonistic sensory cues generate gustatory plasticity in Caenorhabditis elegans. EMBO J. 2006; 25:312-22. https://doi.org/10.1038/sj.emboj.7600940 PMID:16407969 Temperature sensing by an olfactory neuron in a circuit controlling behavior of C. elegans. A Kuhara, M Okumura, T Kimata, Y Tanizawa, R Takano, K D Kimura, H Inada, K Matsumoto, I Mori, 10.1126/science.114892218403676Science. 320Kuhara A, Okumura M, Kimata T, Tanizawa Y, Takano R, Kimura KD, Inada H, Matsumoto K, Mori I. Temperature sensing by an olfactory neuron in a circuit controlling behavior of C. elegans. Science. 2008; 320:803-07. https://doi.org/10.1126/science.1148922 PMID:18403676 Rictor/TORC2 regulates fat metabolism, feeding, growth, and life span in Caenorhabditis elegans. A A Soukas, E A Kane, C E Carr, J A Melo, G Ruvkun, 10.1101/gad.177540919240135Genes Dev. 23Soukas AA, Kane EA, Carr CE, Melo JA, Ruvkun G. Rictor/TORC2 regulates fat metabolism, feeding, growth, and life span in Caenorhabditis elegans. Genes Dev. 2009; 23:496-511. https://doi.org/10.1101/gad.1775409 PMID:19240135 The influence of bacterial diet on fat storage in C. elegans. K K Brooks, B Liang, J L Watts, 10.1371/journal.pone.000754519844570PLoS One. 4Brooks KK, Liang B, Watts JL. The influence of bacterial diet on fat storage in C. elegans. PLoS One. 2009; 4:e7545. https://doi.org/10.1371/journal.pone.0007545 PMID:19844570 Pathways of cellular proteostasis in aging and disease. C L Klaips, G G Jayaraj, F U Hartl, 10.1083/jcb.20170907229127110J Cell Biol. 217Klaips CL, Jayaraj GG, Hartl FU. Pathways of cellular proteostasis in aging and disease. J Cell Biol. 2018; 217:51-63. https://doi.org/10.1083/jcb.201709072 PMID:29127110 Systemic stress signalling: understanding the cell non-autonomous control of proteostasis. R C Taylor, K M Berendzen, A Dillin, 10.1038/nrm375224556842Nat Rev Mol Cell Biol. 15Taylor RC, Berendzen KM, Dillin A. Systemic stress signalling: understanding the cell non-autonomous control of proteostasis. Nat Rev Mol Cell Biol. 2014; 15:211-17. https://doi.org/10.1038/nrm3752 PMID:24556842 Cell-Nonautonomous Regulation of Proteostasis in Aging and Disease. R I Morimoto, 10.1101/cshperspect.a03407430962274Cold Spring Harb Perspect Biol. 12Morimoto RI. Cell-Nonautonomous Regulation of Proteostasis in Aging and Disease. Cold Spring Harb Perspect Biol. 2020; 12:a034074. https://doi.org/10.1101/cshperspect.a034074 PMID:30962274 DAF-16 stabilizes the aging transcriptome and is activated in mid-aged Caenorhabditis elegans to cope with internal stress. S T Li, H Q Zhao, P Zhang, C Y Liang, Y P Zhang, A L Hsu, M Q Dong, 10.1111/acel.1289630773782Aging Cell. 1812896Li ST, Zhao HQ, Zhang P, Liang CY, Zhang YP, Hsu AL, Dong MQ. DAF-16 stabilizes the aging transcriptome and is activated in mid-aged Caenorhabditis elegans to cope with internal stress. Aging Cell. 2019; 18:e12896. https://doi.org/10.1111/acel.12896 PMID:30773782 Identification of key pathways and metabolic fingerprints of longevity in C. elegans. A W Gao, R L Smith, M Van Weeghel, R Kamble, G E Janssens, R H Houtkooper, 10.1016/j.exger.2018.10.00330300667Exp Gerontol. 113Gao AW, Smith RL, van Weeghel M, Kamble R, Janssens GE, Houtkooper RH. Identification of key pathways and metabolic fingerprints of longevity in C. elegans. Exp Gerontol. 2018; 113:128-40. https://doi.org/10.1016/j.exger.2018.10.003 PMID:30300667 Mono-unsaturated fatty acids link H3K4me3 modifiers to C. elegans lifespan. S Han, E A Schroeder, C G Silva-García, K Hebestreit, W B Mair, A Brunet, 10.1038/nature2168628379943Nature. 544Han S, Schroeder EA, Silva-García CG, Hebestreit K, Mair WB, Brunet A. Mono-unsaturated fatty acids link H3K4me3 modifiers to C. elegans lifespan. Nature. 2017; 544:185-90. https://doi.org/10.1038/nature21686 PMID:28379943 daf-16/FoxO promotes gluconeogenesis and trehalose synthesis during starvation to support survival. J D Hibshman, A E Doan, B T Moore, R E Kaplan, A Hung, A K Webster, D P Bhatt, R Chitrakar, M D Hirschey, L R Baugh, 10.7554/eLife.3005729063832Elife. 6Hibshman JD, Doan AE, Moore BT, Kaplan RE, Hung A, Webster AK, Bhatt DP, Chitrakar R, Hirschey MD, Baugh LR. daf-16/FoxO promotes gluconeogenesis and trehalose synthesis during starvation to support survival. Elife. 2017; 6:e30057. https://doi.org/10.7554/eLife.30057 PMID:29063832 Metabolic shift from glycogen to trehalose promotes lifespan and healthspan in Caenorhabditis elegans. Y Seo, S Kingsley, G Walker, M A Mondoux, H A Tissenbaum, 10.1073/pnas.171417811529511104Proc Natl Acad Sci USA. 115Seo Y, Kingsley S, Walker G, Mondoux MA, Tissenbaum HA. Metabolic shift from glycogen to trehalose promotes lifespan and healthspan in Caenorhabditis elegans. Proc Natl Acad Sci USA. 2018; 115:E2791-800. https://doi.org/10.1073/pnas.1714178115 PMID:29511104 DAF-16 and TCER-1 Facilitate Adaptation to Germline Loss by Restoring Lipid Homeostasis and Repressing Reproductive Physiology in C. elegans. F R Amrit, E M Steenkiste, R Ratnappan, S W Chen, T B Mcclendon, D Kostka, J Yanowitz, C P Olsen, A Ghazi, 10.1371/journal.pgen.100578826862916PLoS Genet. 12Amrit FR, Steenkiste EM, Ratnappan R, Chen SW, McClendon TB, Kostka D, Yanowitz J, Olsen CP, Ghazi A. DAF-16 and TCER-1 Facilitate Adaptation to Germline Loss by Restoring Lipid Homeostasis and Repressing Reproductive Physiology in C. elegans. PLoS Genet. 2016; 12:e1005788. https://doi.org/10.1371/journal.pgen.1005788 PMID:26862916 Metabolomic signature associated with reproduction-regulated aging in Caenorhabditis elegans. Q L Wan, X Shi, J Liu, A J Ding, Y Z Pu, Z Li, G S Wu, H R Luo, 10.18632/aging.10117028177875Aging (Albany NY). 9Wan QL, Shi X, Liu J, Ding AJ, Pu YZ, Li Z, Wu GS, Luo HR. Metabolomic signature associated with reproduction-regulated aging in Caenorhabditis elegans. Aging (Albany NY). 2017; 9:447-74. https://doi.org/10.18632/aging.101170 PMID:28177875 Metabolomic changes in Caenorhabditis elegans lifespan mutants as evident from GC-EI-MS and GC-APCI-TOF-MS profiling. C Jaeger, V Tellström, G Zurek, S König, S Eimer, B Kammerer, 10.1007/s11306-014-0637-yMetabolomics. 10Jaeger C, Tellström V, Zurek G, König S, Eimer S, Kammerer B. Metabolomic changes in Caenorhabditis elegans lifespan mutants as evident from GC-EI-MS and GC-APCI-TOF-MS profiling. Metabolomics. 2014; 10:859-76. https://doi.org/10.1007/s11306-014-0637-y Resistance to Stress Can Be Experimentally Dissociated From Longevity. D J Dues, E K Andrews, M M Senchuk, J M Van Raamsdonk, 10.1093/gerona/gly21330247515J Gerontol A Biol Sci Med Sci. 74Dues DJ, Andrews EK, Senchuk MM, Van Raamsdonk JM. Resistance to Stress Can Be Experimentally Dissociated From Longevity. J Gerontol A Biol Sci Med Sci. 2019; 74:1206-14. https://doi.org/10.1093/gerona/gly213 PMID:30247515 Maintenance of C. elegans. WormBook. T Stiernagle, 10.1895/wormbook.1.101.118050451Stiernagle T. Maintenance of C. elegans. WormBook. 2006. https://doi.org/10.1895/wormbook.1.101.1 PMID:18050451 Effectiveness of specific RNA-mediated interference through ingested double-stranded RNA in Caenorhabditis elegans. R S Kamath, M Martinez-Campos, P Zipperlen, A G Fraser, J Ahringer, 10.1186/gb-2000-2-1-research000211178279Genome Biol. 22Kamath RS, Martinez-Campos M, Zipperlen P, Fraser AG, Ahringer J. Effectiveness of specific RNA-mediated interference through ingested double-stranded RNA in Caenorhabditis elegans. Genome Biol. 2001; 2:RESEARCH0002. https://doi.org/10.1186/gb-2000-2-1-research0002 PMID:11178279 Health and longevity studies in C. elegans: the "healthy worm database" reveals strengths, weaknesses and gaps of test compound-based studies. N Saul, S Möller, F Cirulli, A Berry, W Luyten, G Fuellen, 10.1007/s10522-021-09913-2Biogerontology. 2233683565Saul N, Möller S, Cirulli F, Berry A, Luyten W, Fuellen G. Health and longevity studies in C. elegans: the "healthy worm database" reveals strengths, weaknesses and gaps of test compound-based studies. Biogerontology. 2021; 22:215-36. https://doi.org/10.1007/s10522-021-09913-2 PMID:33683565 Deceptively simple but simply deceptive--Caenorhabditis elegans lifespan studies: considerations for aging and antioxidant effects. J Gruber, L F Ng, S K Poovathingal, B Halliwell, 10.1016/j.febslet.2009.09.05119815017FEBS Lett. 583Gruber J, Ng LF, Poovathingal SK, Halliwell B. Deceptively simple but simply deceptive-- Caenorhabditis elegans lifespan studies: considerations for aging and antioxidant effects. FEBS Lett. 2009; 583:3377-87. https://doi.org/10.1016/j.febslet.2009.09.051 PMID:19815017 Stochastic and genetic factors influence tissuespecific decline in ageing C. elegans. L A Herndon, P J Schmeissner, J M Dudaronek, P A Brown, K M Listner, Y Sakano, M C Paupard, D H Hall, M Driscoll, 10.1038/nature0113512397350Nature. 419Herndon LA, Schmeissner PJ, Dudaronek JM, Brown PA, Listner KM, Sakano Y, Paupard MC, Hall DH, Driscoll M. Stochastic and genetic factors influence tissue- specific decline in ageing C. elegans. Nature. 2002; 419:808-14. https://doi.org/10.1038/nature01135 PMID:12397350 Repression of the Heat Shock Response Is a Programmed Event at the Onset of Reproduction. J Labbadia, R I Morimoto, 10.1016/j.molcel.2015.06.02726212459Mol Cell. 59Labbadia J, Morimoto RI. Repression of the Heat Shock Response Is a Programmed Event at the Onset of Reproduction. Mol Cell. 2015; 59:639-50. https://doi.org/10.1016/j.molcel.2015.06.027 PMID:26212459 A panel of fluorophore-tagged daf-16 alleles. U Aghayeva, A Bhattacharya, O Hobert, 10.17912/micropub.biology.00021032550509MicroPubl Biol. Aghayeva U, Bhattacharya A, Hobert O. A panel of fluorophore-tagged daf-16 alleles. MicroPubl Biol. 2020; 2020:10.17912. https://doi.org/10.17912/micropub.biology.000210 PMID:32550509 Stress and timing associated with Caenorhabditis elegans immobilization methods. J R Manjarrez, R Mailler, 10.1016/j.heliyon.2020.e0426332671240Heliyon. 64263Manjarrez JR, Mailler R. Stress and timing associated with Caenorhabditis elegans immobilization methods. Heliyon. 2020; 6:e04263. https://doi.org/10.1016/j.heliyon.2020.e04263 PMID:32671240 ife-2(ok306) and daf-16(mu86) mutations. (A-L) Bar chart representation of motility-assessed healthspan illustrating the fraction of each category upon daily monitorization. Worms are grouped into three categories: mobile (white), impaired (light gray) and frail (dark gray). Dead and censored animals were subtracted from these analyses. Supplementary Figure 3. Motility-assessed healthspan for mutants containing the odr-3(n1605). All cohorts were kept at 20° C and fed OP50 E. coli. (A-D) WT, odr-3(n1605), ife-2(ok306) and odr-3(n1605)Supplementary Figure 3. Motility-assessed healthspan for mutants containing the odr-3(n1605), ife-2(ok306) and daf- 16(mu86) mutations. (A-L) Bar chart representation of motility-assessed healthspan illustrating the fraction of each category upon daily monitorization. Worms are grouped into three categories: mobile (white), impaired (light gray) and frail (dark gray). Dead and censored animals were subtracted from these analyses. All cohorts were kept at 20° C and fed OP50 E. coli. (A-D) WT, odr-3(n1605), ife-2(ok306) and odr-3(n1605); ife-2(ok306) and odr-3(n1605);ife-2(ok306) strains grown on plates with no FUdR. (I-L) odr-3(n1605), ife-2(ok306) and odr-3(n1605). ife-2(ok306) strains on FUdR supplemented plates. (E-H) WT, odr-3(n1605). ife-2(ok306) in the daf-16(mu86) background, grown on FUdR supplemented platesife-2(ok306) strains on FUdR supplemented plates. (E-H) WT, odr-3(n1605), ife-2(ok306) and odr-3(n1605);ife-2(ok306) strains grown on plates with no FUdR. (I-L) odr-3(n1605), ife-2(ok306) and odr-3(n1605);ife-2(ok306) in the daf-16(mu86) background, grown on FUdR supplemented plates.

2022-07-14T06:45:02Z

CCBY

https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0048282&type=printable

GOLD

76287e8c5b1b81b09d7fbae1b00b4152d1943271

10.1371/journal.pone.0048282

2060694610

23144747

3483217

Prediction of C. elegans Longevity Genes by Human and Worm Longevity Networks Published October 29, 2012 Robi Tacutu The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael David E Shore Department of Molecular Biology Massachusetts General Hospital BostonMassachusetts United States of America Department of Genetics Harvard Medical School BostonMassachusetts United States of America Arie Budovsky The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Joã Pedro De Magalhã Es Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool LiverpoolUnited Kingdom Gary Ruvkun Department of Molecular Biology Massachusetts General Hospital BostonMassachusetts United States of America Department of Genetics Harvard Medical School BostonMassachusetts United States of America Vadim E Fraifeld [email protected] The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Sean P Curran [email protected] Division of Biogerontology Davis School of Gerontology United States of America University of Southern California Los AngelesCalifornia Department of Molecular and Computational Biology Arts, and Sciences United States of America Dornsife College of Letters University of Southern California Los AngelesCalifornia Department of Biochemistry and Molecular Biology Keck School of Medicine University of Southern California Los AngelesCaliforniaUnited States of America Prediction of C. elegans Longevity Genes by Human and Worm Longevity Networks Published October 29, 2012Received June 7, 2012; Accepted September 21, 2012;Citation: Tacutu R, Shore DE, Budovsky A, de Magalhães JP, Ruvkun G, et al. (2012) Prediction of C. elegans Longevity Genes by Human and Worm Longevity Networks. PLoS ONE 7(10): e48282. Editor: Yousin Suh, Albert Einstein College of Medicne, United States of America The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing Interests: The authors have declared that no competing interests exist. * Intricate and interconnected pathways modulate longevity, but screens to identify the components of these pathways have not been saturating. Because biological processes are often executed by protein complexes and fine-tuned by regulatory factors, the first-order protein-protein interactors of known longevity genes are likely to participate in the regulation of longevity. Data-rich maps of protein interactions have been established for many cardinal organisms such as yeast, worms, and humans. We propose that these interaction maps could be mined for the identification of new putative regulators of longevity. For this purpose, we have constructed longevity networks in both humans and worms. We reasoned that the essential first-order interactors of known longevity-associated genes in these networks are more likely to have longevity phenotypes than randomly chosen genes. We have used C. elegans to determine whether post-developmental inactivation of these essential genes modulates lifespan. Our results suggest that the worm and human longevity networks are functionally relevant and possess a high predictive power for identifying new longevity regulators. Introduction Numerous pathways contribute to longevity, but the identification of their components has not been saturating [1]. Because of their short lifespan and genetic tractability, C. elegans have proven indispensable in the study of longevity. The first screen to identify C. elegans genes that regulate longevity was an EMS mutagenesis that isolated eight mutants, each of which modulated the dauer developmental pathway or caloric intake [2]. The relationship between these functions and lifespan is now well established [3,4]. Two subsequent studies utilized genome-wide RNA interference (RNAi) to identify genes that increase longevity when inactivated [5,6]. These screens identified 89 and 29 genes respectively, with an overlap of only 3 genes, strongly suggesting that neither was saturating. This likely reflects the high false negative rate associated with high-throughput RNAi screening, as well as technical limitations of the screen designs [1]. For instance, because the screens inactivated genes of interest during development, genes required for development but capable of modulating adult lifespan would be missed. Curran and Ruvkun explored this overlooked gene set by inactivating essential genes postdevelop-mentally, revealing 64 genes required for development that extend lifespan when inactivated during adulthood [7]. Nevertheless, many important longevity genes likely remain unidentified. Known longevity genes are enriched for roles in stress tolerance and development. Many mutations that modulate longevity were identified by virtue of their contribution to stress response pathways or by homology to genes of this kind. A study of over 50 long-lived C. elegans mutants reveals that each is resistant to one or more stressors, such as oxidative damage, heat stress or irradiation [8,9]. Many genes required for the successful extension of lifespan in one or more long-lived mutants also contribute to the longevity of wild-type animals, but are distinguished by a proportionally greater change in the mutant background. Examples of such genes include daf-16, hsf-1, smk-1, jnk-1, cst-1, skn-1, and pha-4 [10,11,12,13,14,15]. Overexpression of most of these genes extends lifespan and, where tested, increases stress tolerance. Network analyses have also revealed a link between aging-related genes and development; known longevity-associated genes (LAGs) are enriched for essential genes or those required for development, and essential genes are likewise enriched for LAGs [7,16,17,18,19]. This finding appears to illustrate the antagonistic pleiotropy theory of aging, which suggests that the postreproductive decrescendo of the force of natural selection permits the evolution of genes that are essential early in life but detrimental late in life [20]. Samuelson et al. (2007) screened for gene inactivations that suppress lifespan extension in daf-2 mutant C. elegans and identified 159 genes contributing to daf-2 lifespan and to stress tolerance [21]. The majority of the suppressors decrease the longevity of a control strain, but decrease daf-2 longevity by a greater margin. Based upon the efficacy of other genome-wide screens and technical limitations, it is unlikely that this screen saturated the breadth of genes that contribute to lifespan extension. Network biology is one approach to gaining insight regarding the interactions of known LAGs and identifying new longevity regulators [22,23]. Network approaches provide a conceptual framework for the study of the complex interactions amongst the components of biological systems [24]. Networks may be constructed from many kinds of data, including, but not limited to, protein-protein interactions, transcriptional co-regulation, putative microRNA targets, or participation in annotated biological pathways [16,18,25]. Databases of such interactions exist for many species including yeast, worm, fly, mouse, and human [26]. Often, genes that serve essential cell functions are more connected than others and genes that contribute to a particular phenotype are more interconnected than would be expected by chance [19,22,27]. Previous network analyses have demonstrated that LAGs, on average, have more protein-protein interactions (PPIs) with other proteins and amongst each other than non-LAGs in the interactome [28]. This is consistent with the fact that many LAGs play significant roles in development, participate in complex stress response cascades or are otherwise essential. Given the wealth of LAGs, an effort to understand the regulation of longevity from a biological network perspective may provide new insights into longevity pathways. Networks may be enriched by the integration of information from diverse species using homology as a means to overlay speciesspecific findings [26]. This technique could be applicable to aging because LAGs are highly conserved across species [7,22,28]. Such an approach may be especially fruitful in the study of human aging because aging has been extensively studied in model organisms. The profusion of data from non-mammalian systems renders broader analyses increasingly powerful and informative. An interaction map enriched with data across all species and accounting for cross-species homology could generate a robust functional network and be used to identify new genes in lifespan extension pathways. In this study, we performed a network analysis of LAGs and their interacting partners in worms and humans. We found that LAGs and their first-order partners form tightly interconnected networks. The partners of known LAGs in the worm and human longevity networks may participate in the intricate pathways and complexes that regulate lifespan, and are therefore candidate longevity genes. Essential genes are particularly interesting in this regard because known LAGs are enriched for developmental functions, consistent with the concept of antagonistic pleiotropy [7,16,22]. To functionally verify this prediction, we post-developmentally inactivated 374 of these genes or their orthologs in C. elegans. In our primary analysis, 156 of these inactivations resulted Figure 1. Worm Longevity Network. The worm longevity network (WLN) includes 205 previously identified LAGs (dark green) and their 666 firstorder protein interaction partners (light green). The graphical output of the network was generated using Cytoscape 2.8.0 [51]. Size of nodes is proportional with the number of PPIs in the BioGRID interactome. Genes in the network are more connected and more interconnected than would be expected by chance, suggesting shared functionality. Because many genes that influence longevity function in complexes, signaling networks or in conjunction with cofactors, the 666 first-order interacting LAG partners may be enriched with previously unidentified longevity genes. doi:10.1371/journal.pone.0048282.g001 in extended (101) or decreased (55) lifespan. We confirmed a subset (30 genes) of these phenotypes in rigorous longitudinal analyses. Our results are consistent with the idea that genes involved in development and translation have a role in longevity regulation. Collectively, this study presents a proof of concept that by combining a network-based approach with the selection of genes fitting antagonistic pleiotropy, new worm lifespan regulators could be identified with an unprecedentedly high predictive power. Materials and Methods Data Sources C. elegans LAGs were compiled from scientific literature and manually curated. The list of LAGs includes genes reported to promote longevity or cause premature aging following genetic intervention (partial or full loss-of-function mutations, gene overexpression) or RNA interference-induced gene silencing [17,22,29]. The collection contains 555 entries and is accessible in the Human Ageing Genomic Resources -GenAge database (build 14), http://genomics.senescence.info/genes/index.html [29]. This list of genes was used as a ''core set'' for the construction of the worm longevity network (WLN). In addition to established human LAGs -the number of which is still very limited -the ''core set'' for the construction of human longevity network (HLN) also included non-redundant orthologs of LAGs from model organisms (S. cerevisae, C. elegans, D. melanogaster, and M. musculus). In total, this list consists of 662 human genes. All LAG lists for model organisms were compiled using the same method described for C. elegans and are available in the GenAge database [29]. PPI data used for the constructions of WLN and HLN were extracted from the BioGRID database, release 2.0.53 [30]. Orthology information was obtained from the InParanoid database -Eukaryotic Ortholog Groups, release 6.1 [47] and worm lethal phenotypes were retrieved from the WormBase database [48,49]. Network Construction The approach for constructing the longevity networks was described in detail elsewhere [22,28]. The longevity networks for worms and humans were created using YABNA (Yet Another Biological Networks Analyzer), a flexible software program developed in Vadim Fraifeld's lab. Current versions of human and model organism longevity networks are available in the NetAge database [28]. For both worm and human gene sets, the network construction algorithm included: 1) keeping all genes (LAGs) with reported PPIs from the ''core set''; 2) adding all firstorder PPI partners of core genes; and 3) taking the largest interconnected sub-graph as a longevity network. Prediction of New Worm Longevity Regulators Based on WLN/HLN Prediction of new LAGs in C. elegans was based on all the following criteria: 1) belonging to the WLN or being the C. elegans ortholog of a gene from the HLN; 2) not reported previously as a LAG in C. elegans or other model organisms; 3) being essential for the development and growth of C. elegans (essential genes). Thus, there were two sources for selection of candidate genes_the firstorder partners of LAGs from WLN and HLN. In total, 500 essential worm genes were included as candidates for longevity analysis. Two hundred twenty-eight of these genes were derived from the WLN and the remainder from the HLN, with an overlap of 54. RNAi clones were available to target 374 of these 500 genes. Detection of Functional Enrichment To detect enriched functions and processes, the Database for Annotation, Visualization, and Integrated Discovery (DAVID) was employed using default settings [50]. For detecting functional Table 3. A network-based approach verifies new worm longevity genes at a greater frequency than genome-wide RNAi screens. enrichment in the genes tested, the worm genome was used as background. For detecting enrichment in genes resulting in a short-or long-lived phenotype, the genes tested were used as background. Strains and Culture Conditions C. elegans were cultured on Escherichia coli OP50 or HT115 by standard techniques. C. elegans N2 Bristol (WT) and GR1373 eri-1(mg366) IV were utilized in the described experiments. Post-developmental RNAi Post-developmental RNAi lifespan analysis was performed as described by Curran and Ruvkun, 2007 [7]. RNAi clones were grown overnight in LB with carbenicillin and seeded to 6-well plates containing 5 mM isopropylthiogalactoside (IPTG). 400 ml bacteria were seeded to each well. Expression of double-stranded RNA was induced overnight. eri-1(mg366) worms raised to the L4 larval stage on HT115 E. coli were washed twice in M9 buffer with tetracycline, carbenicillin and streptomycin, resuspended in M9 carbenicillin buffer, and seeded onto the RNAi induced bacteria. Wells were treated with 80 mg/ml 5-fluorodeoxyuridine (FUDR) to inhibit progeny production. On the third day of adulthood, wells were supplemented with 50 ml of 10x concentrated RNAi bacteria grown overnight and induced for two hours at room temperature in LB with carbenicillin and 5 mM IPTG. Survival Analysis Post-developmental gene inactivation was carried out as described above. Primary survival analysis was accomplished by a thrashing assay in which replicate wells were flooded with M9 buffer on the 15 th , 17 th , 19 th and 21 st days of adulthood at 20uC [21]. One of four replicates was scored and subsequently discarded for each time point. Preliminary survival phenotypes were assigned to gene inactivations that produced a consistent 10% increase or decrease in survival on at least three of the four time points. Strong candidates were selected for longitudinal lifespan analysis. In each biological replicate and at each time point in the thrashing assay, at least 30 worms were scored for each condition. Adult Lifespan Analysis From the four biological replicates in the thrashing assay we selected RNAi clones that consistently yielded the strongest phenotypes (both long and short lived) at each of the time points screened. Post-developmental gene inactivation was carried out as described above. Longitudinal lifespan analyses were conducted as described with incubation at 20uC or 25uC. Survival was assayed by gently tapping worms with a thin length of platinum every other day from young adulthood onward. Statistical analyses were performed using the software SPSS (SPSS Inc.). Survival of each RNAi-treated population was compared with that of a population treated with control RNAi, and significance was determined by Kaplan-Meier analysis using the log-rank test. A significance threshold of p,0.05 was applied. On average, 90 animals were scored longitudinally across three biological replicates for each clone. Results and Discussion The C. elegans LAG set contains 555 previously identified positive and negative regulators of longevity [29]. Of these, 218 are present in the C. elegans interactome in the BioGRID database [30]. These LAGs display a higher degree of connectivity than would be expected by chance and more importantly, a very high degree of interconnectivity. Almost a quarter of worm LAGs found in the interactome are interconnected, whereas simulations with randomized gene sets of the same size predict only 5% interconnectivity (p,0.001). Because PPI databases are inherently incomplete, the distribution of interactions may change as entries accumulate. This does not detract from the significance of individual PPIs currently found within the database. To exploit LAG connectivity as a means to identify new candidate longevity genes, we explored the first-order partners of C. elegans LAGs. Taken together with their first-order partners, 94% of the LAGs in the interactome form a continuous Worm Longevity Network (WLN) (Figure 1). The network encompasses 871 genes, of which 205 are known LAGs at the core of the network and 666 are their first-order partners not previously associated with longevity. Because many longevity genes function within complex signaling pathways or functional groups, we propose that the 666 first-order partners of LAG network genes are likely to include factors that regulate longevity. To verify the functional significance of LAG PPIs, we inactivated first-order partners of LAGs and assayed their effect on lifespan in C. elegans. We focused our search on a subset of 228 of the 666 first-order partners, selected because they are essential in development, a process closely associated with longevity regulation [7,17,18,19,28]. RNAi clones were available for 190 of these 228 genes (Tables S1 and S3). Evidence that previous screens have not been saturating supports the possibility that a rigorous analysis of our systematically selected candidate gene set will reveal previously unidentified LAGs. Because LAGs are highly evolutionary conserved, we also utilized a Human Longevity Network (HLN) to identify candidate regulators of lifespan [17,22,28]. The HLN was previously constructed by compiling a set of genes directly associated with human aging, such as those responsible for progeria, and the orthologs of LAGs experimentally defined in other species [18,22]. Together with their first-order partners, these genes form a continuous PPI network [22,28]. Like the first-order partners of the WLN, the first-order partners of HLN LAGs could also be considered putative new longevity regulators. As human studies to test network predictions experimentally are not possible, we elected to test the C. elegans orthologs of the first-order partners of LAGs in the HLN. We further narrowed our candidate list from the HLN to 272 orthologs previously found to be required for development in C. elegans. RNAi clones targeting 227 genes were available (Tables S2 and S3). Forty-three of the C. elegans orthologs of HLN genes with available RNAi clones were already present amongst the 190 WLN candidates, suggesting that the first-order interactions of LAGs, like the genes themselves, may be highly conserved. We chose to inactivate the 374 putative C. elegans LAGs identified by the above network analysis post-developmentally (Tables S1-S3). This approach was necessitated by our focus upon genes required for development, an ontology enriched with known LAGs. Post-developmental gene inactivation circumvents developmental pleiotropies, thus simplifying the interpretation of longevity phenotypes. Genes essential for development are Figure 3. Interaction of new LAGs with known worm longevity genes. The LAGs identified in our screen interact extensively with known longevity genes. Nearly two-thirds of these genes interact with daf-2 and bar-1, both involved in the daf-16-mediated lifespan extension, and clk-2, a gene whose mutations may extend lifespan by slowing down development. Size of nodes is proportional with the number of PPIs in the BioGRID interactome. Bordered circles depict genes essential for growth and development. doi:10.1371/journal.pone.0048282.g003 enriched approximately 5-fold for LAGs and 15-fold for gene inactivations that extend lifespan by greater than 20% [5,6,7]. In addition, evidence suggests that many lifespan regulatory treatments that are effective when applied during development are also effective if initiated during adulthood [7,31,32,33,34,35]. Therefore, post-developmental gene inactivation maximizes our capacity to detect genes that regulate longevity. In our primary screen consisting of the post-developmental inactivation of 374 candidate LAGs, lifespan was approximated by survival at four time points distributed across the period of mortality. In this primary analysis, 101 gene inactivations conferred a consistent increase in survival and 55 a decrease, compared to controls (Tables S1-S3). To confirm these results in a more rigorous analysis, we selected 57 novel putative LAGs (45 long-lived and 12 short-lived phenotypes) for longitudinal analysis on the basis of phenotypic strength. Longitudinal analysis confirmed 19 of the gene inactivations that increase lifespan and 11 that decrease lifespan, thus identifying 30 new C. elegans LAGs (Tables 1 and 2). The rate of identification for lifespan-promoting gene inactivations (19/374; 5.1%) is much greater than that achieved in genome-wide screens (average 0.38%), as well as that of an unbiased screen that canvased genes known to play critical roles in development (2.4%, Tables 3 and 4) [5,6,7]. This comparison is crude because conditions and selection criteria utilized in each screen were distinct. However, if the rate at which our preliminary results were confirmed longitudinally could be extrapolated to the untested preliminary hits, the set of 374 candidate LAGs would include 42 (11.2%) that increase lifespan when inactivated and 50 (13.4%) that decrease it. This is likely to be an overestimate, however, because the genes selected for secondary analysis represent the strongest preliminary scores. Most of the longitudinally confirmed gene inactivations that increased mean lifespan did so by a small but statistically significant percentage when performed post-developmentally (average 9.7%). Previous screens for increased lifespan have identified numerous genes that extend lifespan by upwards of 20%. In this screen, we report only one gene at that level, T23D8.3, which extends mean lifespan by 26% (Table 1, Figure 2). T23D8.3 is required for embryonic and larval development. This is also true of its ortholog, LTV1, which is essential for cell growth in yeast and cultured human cells [36]. LTV1 participates in the nuclear export and processing of the 40S ribosomal subunit, suggesting that T23D8.3 inactivation inhibits translation in C. elegans. The second greatest extension of mean lifespan resulted from inhibition of a protein subunit of the 40S ribosome (rps-14), another component of translation, with 16.6% extension of mean lifespan (Table 1, Figure 2). The identification of gene inactivations with modest lifespan extension phenotypes (12 of 19 gene inactivations ,10%, Tables 1 and 2) is consistent with our hypothesis in several ways. First, we predicted that we would identify gene inactivations missed in previous screens (false negatives). The subtlety of some of the observed phenotypes may explain, in part, how they were overlooked. Second, previous LAG network studies have highlighted the existence of nodes representing key lifespan regulatory factors [16,37]. Our results suggest that genes interacting directly or indirectly with these nodes, such as those tested here, contribute to their activity, perhaps additively, but are not strictly required. We performed a functional enrichment analysis using DAVID to determine whether translation or other processes were enriched within our results. In total, 17 of the gene inactivations that conferred increased longevity in our preliminary analysis are involved in translation, including initiation factors, tRNA synthetases, and ribosomal subunits. This ontology was not present for genes found to decrease lifespan when inactivated. Therefore genes required for translation are specifically enriched amongst those that extend lifespan when inactivated in comparison to both the set of genes tested (p = 4.1E-4) and the C. elegans genome as a whole (p = 3.3E-8). Previous studies have demonstrated that translation inhibition is a potent mechanism of longevity extension [7,38,39,40,41]. Current models of longevity extension thereby substantiate the results of our combined in silico and in vivo analyses. We longitudinally confirmed 11 positive regulators of lifespan; these genes decrease C. elegans longevity when inactivated postdevelopmentally. The frequency at which these genes were identified (11/374, 2.9%) is comparable to that achieved in a previous genome wide screen for positive regulators of lifespan in daf-2 mutants (3.2%) [21]. Our results, however, represent the first systematic analysis of short-lifespan phenotypes in a wild-type background. The actual number of positive regulators in our candidate LAG set may have been significantly greater than 11 because 92% of candidates retested were confirmed and 44 preliminary candidates were not retested. Mutation or inactivation of many genes with key longevity-regulatory functions reduces wild-type longevity while abrogating lifespan extension in one or more long-lived mutants, resulting in the equalization of wild-type and mutant lifespans [10,11,12,13,14,15]. The genes identified in this screen are particularly intriguing because they interact with known components of the LAG network. It would be interesting to determine whether the positive regulators we have identified contribute specifically to particular mechanisms of lifespan extension, such as insulin/IGF-1 signaling or the disruption of translation. Determining the molecular roles these genes might play in lifespan extension will be a topic of continued research. Analyzing the interactions of the LAGs identified in this screen within the existing WLN reveals that almost two thirds of the new LAGs are connected to at least one of the prominent lifespan regulatory genes daf-2, bar-1, or clk-2. Moreover, 5 genes (rnr-2, asb-1, rpb-8, hmg-4, and B0285.1) interact with all 3. Multiple connections to established LAGs were also observed for gei-4. The WLN LAGs identified in this screen and their first-order LAG partners form a continuous network -a WLN module, which is presented in Figure 3. We separated results from the WLN (Table S1), the C. elegans orthologs from the HLN (Table S2) and genes common to both sets to determine the functional efficacy of each selection (Table S3). For this analysis, we considered only the primary results because the number of genes selected for inclusion in the longitudinal analysis was independent of these groupings. Results from each class are similar. Primary analysis detected lifespan phenotypes for 46% (32% long-lived, 14% short lived) of WLNspecific candidates (Table S1), 39% (24% long-lived, 15% shortlived) of HLN-specific candidates (Table S2) and 37% (21% longlived and 16% short-lived) of the candidates common to both groups (Table S3). Although the modest increase in efficacy with which the WLN predicted increased lifespan may indicate superior accuracy, failure to replicate that observation amongst shared LAGs suggests this is not the case. Importantly however, the orthologs of human interactors were similarly predictive with the interactors endogenous to C. elegans, underscoring the remarkable conservation of genes that regulate longevity as well as their protein-protein interactions. We confirm 30 new LAGs identified through network analysis of C. elegans and human candidates. Our results are consistent with several empirical observations regarding genes found to regulate longevity in previous screens. First, in agreement with previous studies, we identify a strong association of developmental and lifespan-regulatory functions [7,17,18,19,28,42,43,44,45]. By uti-lizing PPI networks and applying developmental ontology filters, we identified new lifespan regulators with a 13-fold greater frequency than has been reported in previous genome-wide screens (5.08% of long-lived phenotypes vs. an average of 0.38% in two genome-wide screens [5,6]; Tables 3 and 4). This is particularly supportive of our approach because we excluded all previously identified LAGs, suggesting we achieved greater efficacy despite systematically omitting the most robust, and therefore most easily identified, LAG inactivations. The verification of new LAGs resulting from our analysis probably underestimates the total present in our candidate pool, as 56 gene inactivations that extend lifespan and 43 that decrease lifespan from our preliminary analysis were not retested longitudinally. Second, the selection of candidate LAGs from the HLN and WLN were similarly effective, suggesting that both LAGs and their PPIs are highly conserved. Third, we identify the inhibition of translation as a means of lifespan regulation, consistent with results from several laboratories [5,6,7,39,46]. Finally, our results demonstrate that as expected, previous screens for LAGs have not saturated the search for gene inactivations that influence longevity. This is the first study to experimentally test a large set of novel LAG predictions generated based on the concept of a longevity network. The future application of increasingly detailed biological networks to the study of aging/longevity and the potential to synergize those networks with experimental studies in C. elegans and other cardinal organisms are promising. In summary, we reasoned that the first-order interacting partners of known LAGs would be more likely modulate the longevity than a random set of genes. We used previously constructed worm and human longevity networks to identify candidate lifespan regulatory genes [28]. We then narrowed this candidate gene set in a manner consistent with the principles of antagonistic pleiotropy by focusing on genes essential to development. By combining a network-based approach with the selection of genes required for development, we identified new lifespan regulatory genes at a frequency far exceeding that achieved in genome-wide screens. Though the effect of the new LAGs on lifespan is relatively modest, one can speculate that they might function in pathways or complexes that modulate core longevity functions. The interaction of genes identified in this screen with key nodes of the WLN is strongly suggestive in this regard ( Figure 3). This work establishes both specifically and in proof of concept that biological networks enriched with experimental data are empowered to generate valuable candidate gene sets. Such an approach may prove broadly applicable as a tool to improve the efficiency of screening efforts. Moreover, the applicability of our method across diverse organisms or phenotypes is limited only by the availability of sufficient data to construct relevant networks. Supporting Information Table S1 First-order interactors of LAGs in the WLN (without shared genes with HLN) assayed in C. elegans. (DOCX) Figure 2 . 2Disruption of translation extends longevity in C. elegans. The gene inactivations found to extend lifespan by the greatest percentage function in translation. T23D8.3 is the C. elegans ortholog of human LTV1, which is required for the nuclear export and processing f the 40S ribosomal subunit. The ribosomal protein subunit rps-14 directly participates in the 40S ribosome and is required for translation. Mean lifespan extension following postdevelopmental inactivation of these genes in an enhanced RNAi strain, eri-1, is 26% for the LTV1 ortholog (dashed line) and16.6% for rps-14 (dotted line) in comparison to an empty vector RNAi control (solid line). The lifespan extension phenotypes of these genes are consistent with the phenotypes of other translation genes in our screen, including rps-12 and 3 RNA polymerases(Tables 1 and 2). doi:10.1371/journal.pone.0048282.g002 Table 1 .Table 2 . 12First-order interactors of LAGs in the WLN regulate lifespan in C. elegans. First-order interactors of LAGs in the HLN regulate lifespan in C. elegans.Gene Common name Lifespan (%D mean) a Lifespan (%D max) a Function T23D8.3 T23D8.3 25.7 34.7 Translation F37C12.9 rps-14 16.6 21.7 Translation T09A5.10 lin-5 16.4 8.7 Cell division C03C10.3 b rnr-2 14.2 21.7 DNA biosynthesis F54E7.2 rps-12 11.6 21.7 Translation F09F7.3 F09F7.3 8.1 21.7 RNA polymerase III C38D4.3 mel-28 6.5 8.7 Cell division T20B12.8 hmg-4 6.5 0.0 Transcription elongation F35G12.10 asb-1 6.4 0.0 ATP synthase F26F4.11 b rpb-8 6.4 0.0 RNA polymerase II C47D12.2 C47D12.2 5.5 8.7 Unknown K10B3.7 gpd-3 5.3 0.0 Glycolysis R144.2 R144.2 218.9 225.0 mRNA cleavage and polyadenylation K07D4.3 rpn-11 219.3 225.0 Proteasome B0285.1 B0285.1 222.2 225.0 Kinase R13G10.1 dpy-27 223.5 225.0 Condensin W07B3.2 gei-4 227.0 238.0 Filament regulation a %D mean and %D maximum lifespan (last quartile) were calculated in relation to control. b Shared genes in WLN and HLN. doi:10.1371/journal.pone.0048282.t001 Gene Common name Lifespan (%D mean) a Lifespan (%D max) a Function C03C10.3 b rnr-2 14.2 21.7 DNA biosynthesis R02D3.3 R02D3.3 13.0 21.7 RNA polymerase II F10B5.6 emb-27 10.5 8.7 Cell division F29G9.3 aps-1 7.9 0.0 Adaptin T09B4.10 chn-1 7.1 8.7 Ubiquitin ligase F26F4.11 b rpb-8 6.4 0.0 RNA polymerase II F28D9.1 rsr-1 6.1 0.0 Splicing F18A1.5 rpa-1 5.1 21.7 DNA replication C01H6.5 nhr-23 4.9 8.7 Molting Y40B1A.4 sptf-3 222.9 237.0 Transcription factor ZK1058.2 pat-3 228.1 250.0 Integrin C52E4.4 rpt-1 228.4 250.0 Proteasome F26H9.6 rab-5 238.0 250.0 Endocytosis D1014.3 snap-1 239.9 250.0 Vesicle fusion K02D10.5 K02D10.5 248.6 272.0 SNARE complex a %D mean and %D maximum lifespan (last quartile) were calculated in relation to control. b Shared genes in WLN and HLN. doi:10.1371/journal.pone.0048282.t002 The frequency with which longevity genes were verified in each screen is presented as a percentage of the total number of genes screened. doi:10.1371/journal.pone.0048282.t003Gene set Phenotype Genes screened Preliminary candidates Retested Verified Frequency a Genome-wide [6] Long-lived 16475 600 600 89 0.54 Genome-wide [5] Long-lived 13300 94 94 29 0.22 Essential genes [7] Long-lived 2700 470 470 64 2.37 Genome-wide [21] Short-lived 15718 500 500 159 1.01 LAG interacting partners (this study) Long-lived 374 101 45 19 5.08 Short-lived 374 55 12 11 2.94 a Table 4 . 4Distribution of long-lived and short-lived phenotypes and the frequency of verified longevity regulators (LAGs) in this study.The frequency with which longevity genes were verified in each gene set is presented as a percentage of the total number of genes screened.Gene set Genes screened Phenotype Preliminary candidates Retested Verified Frequency a WLN b 147 Long-lived 47 24 10 6.8 Short-lived 21 5 5 3.4 HLN b 184 Long-lived 45 17 7 3.8 Short-lived 27 6 6 3.3 Shared 43 Long-lived 9 4 2 4.7 Short-lived 7 1 0 0.0 Total 374 Long-lived 101 45 19 5.1 Short-lived 55 12 11 2.9 a b WLN and HLN without shared genes. doi:10.1371/journal.pone.0048282.t004 Table S2 S2First-order interactors of LAGs in the HLN (without shared genes with WLN) assayed in C. elegans. (DOCX)Table S3 Shared first-order interactors of LAGs found in WLN and HLN assayed in C. elegans. (DOCX) PLOS ONE | www.plosone.org October 2012 | Volume 7 | Issue 10 | e48282 AcknowledgmentsWe thank the Ellison Medical Foundation and the Biology of Aging Woods Hole summer course participants for assistance in screening the candidate longevity regulators. We are thankful to Lori E. Thomas and the members of the Curran laboratory for technical assistance in lifespan experiments and Jacqueline Y. Lo for critical reading of the manuscript. SPC is an Ellison Medical Foundation New Scholar in Aging.Author Contributions Whole genome RNAi screens for increased longevity: important new insights but not the whole story. S S Lee, Exp Gerontol. 41Lee SS (2006) Whole genome RNAi screens for increased longevity: important new insights but not the whole story. Exp Gerontol 41: 968-973. A method for the isolation of longevity mutants in the nematode Caenorhabditis elegans and initial results. M R Klass, Mech Ageing Dev. 22Klass MR (1983) A method for the isolation of longevity mutants in the nematode Caenorhabditis elegans and initial results. Mech Ageing Dev 22: 279- 286. A C. elegans mutant that lives twice as long as wild type. C Kenyon, J Chang, E Gensch, A Rudner, R Tabtiang, Nature. 366Kenyon C, Chang J, Gensch E, Rudner A, Tabtiang R (1993) A C. elegans mutant that lives twice as long as wild type. Nature 366: 461-464. Different dietary restriction regimens extend lifespan by both independent and overlapping genetic pathways in C. elegans. E L Greer, A Brunet, Aging Cell. 8Greer EL, Brunet A (2009) Different dietary restriction regimens extend lifespan by both independent and overlapping genetic pathways in C. elegans. Aging Cell 8: 113-127. New Genes Tied to Endocrine, Metabolic, and Dietary Regulation of Lifespan from a Caenorhabditis elegans Genomic RNAi Screen. M Hansen, A L Hsu, A Dillin, C Kenyon, PLoS Genet. 117Hansen M, Hsu AL, Dillin A, Kenyon C (2005) New Genes Tied to Endocrine, Metabolic, and Dietary Regulation of Lifespan from a Caenorhabditis elegans Genomic RNAi Screen. PLoS Genet 1: e17. A systematic RNAi screen for longevity genes in C. elegans. B Hamilton, Y Dong, M Shindo, W Liu, I Odell, Genes Dev. 19Hamilton B, Dong Y, Shindo M, Liu W, Odell I, et al. (2005) A systematic RNAi screen for longevity genes in C. elegans. Genes Dev 19: 1544-1555. Lifespan regulation by evolutionarily conserved genes essential for viability. S Curran, G Ruvkun, PLoS Genet. 356Curran S, Ruvkun G (2007) Lifespan regulation by evolutionarily conserved genes essential for viability. PLoS Genet 3: e56. Longevity genes in the nematode Caenorhabditis elegans also mediate increased resistance to stress and prevent disease. T E Johnson, S Henderson, S Murakami, E De Castro, S H De Castro, Journal Inherit Metab Dis. 25Johnson TE, Henderson S, Murakami S, de Castro E, de Castro SH, et al. (2002) Longevity genes in the nematode Caenorhabditis elegans also mediate increased resistance to stress and prevent disease. Journal Inherit Metab Dis 25: 197-206. Relationship between increased longevity and stress resistance as assessed through gerontogene mutations in Caenorhabditis elegans. T E Johnson, E De Castro, Hegi De Castro, S Cypser, J Henderson, S , Exp Gerontol. 36Johnson TE, de Castro E, Hegi de Castro S, Cypser J, Henderson S, et al. (2001) Relationship between increased longevity and stress resistance as assessed through gerontogene mutations in Caenorhabditis elegans. Exp Gerontol 36: 1609-1617. SMK-1, an essential regulator of DAF-16-mediated longevity. S Wolff, H Ma, D Burch, G Maciel, T Hunter, Cell. 124Wolff S, Ma H, Burch D, Maciel G, Hunter T, et al. (2006) SMK-1, an essential regulator of DAF-16-mediated longevity. Cell 124: 1039-1053. Direct inhibition of the longevity-promoting factor SKN-1 by insulin-like signaling in C. elegans. J M Tullet, M Hertweck, J H An, J Baker, J Y Hwang, Cell. 132Tullet JM, Hertweck M, An JH, Baker J, Hwang JY, et al. (2008) Direct inhibition of the longevity-promoting factor SKN-1 by insulin-like signaling in C. elegans. Cell 132: 1025-1038. PHA-4/Foxa mediates diet-restriction-induced longevity of C. elegans. S Panowski, S Wolff, H Aguilaniu, J Durieux, A Dillin, Nature. 447Panowski S, Wolff S, Aguilaniu H, Durieux J, Dillin A (2007) PHA-4/Foxa mediates diet-restriction-induced longevity of C. elegans. Nature 447: 550-555. JNK regulates lifespan in Caenorhabditis elegans by modulating nuclear translocation of forkhead transcription factor/DAF-16. S W Oh, A Mukhopadhyay, N Svrzikapa, F Jiang, R J Davis, Proc Natl Acad Sci U S A. 102Oh SW, Mukhopadhyay A, Svrzikapa N, Jiang F, Davis RJ, et al. (2005) JNK regulates lifespan in Caenorhabditis elegans by modulating nuclear translocation of forkhead transcription factor/DAF-16. Proc Natl Acad Sci U S A 102: 4494- 4499. A conserved MST-FOXO signaling pathway mediates oxidative-stress responses and extends life span. M K Lehtinen, Z Yuan, P R Boag, Y Yang, J Villen, Cell. 125Lehtinen MK, Yuan Z, Boag PR, Yang Y, Villen J, et al. (2006) A conserved MST-FOXO signaling pathway mediates oxidative-stress responses and extends life span. Cell 125: 987-1001. Regulation of aging and age-related disease by DAF-16 and heat-shock factor. A L Hsu, C T Murphy, C Kenyon, Science. 300Hsu AL, Murphy CT, Kenyon C (2003) Regulation of aging and age-related disease by DAF-16 and heat-shock factor. Science 300: 1142-1145. MicroRNA-regulated protein-protein interaction networks: how could they help in searching for prolongevity targets?. R Tacutu, A Budovsky, M Wolfson, V E Fraifeld, Rejuvenation Res. 13Tacutu R, Budovsky A, Wolfson M, Fraifeld VE (2010) MicroRNA-regulated protein-protein interaction networks: how could they help in searching for pro- longevity targets? Rejuvenation Res 13: 373-377. Common gene signature of cancer and longevity. A Budovsky, R Tacutu, H Yanai, A Abramovich, M Wolfson, Mech Ageing Dev. 130Budovsky A, Tacutu R, Yanai H, Abramovich A, Wolfson M, et al. (2009) Common gene signature of cancer and longevity. Mech Ageing Dev 130: 33-39. GenAge: a genomic and proteomic network map of human ageing. J P De Magalhaes, O Toussaint, FEBS Lett. 571de Magalhaes JP, Toussaint O (2004) GenAge: a genomic and proteomic network map of human ageing. FEBS Lett 571: 243-247. Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. R Tacutu, A Budovsky, H Yanai, V E Fraifeld, Aging (Albany NY). 3Tacutu R, Budovsky A, Yanai H, Fraifeld VE (2011) Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. Aging (Albany NY) 3: 1178-1191. Pleiotropy, natural selection, and the evolution of senescence. Gc W, Evolution. 11GC W (1957) Pleiotropy, natural selection, and the evolution of senescence. Evolution 11: 398-809. Gene activities that mediate increased life span of C. elegans insulin-like signaling mutants. A V Samuelson, C E Carr, G Ruvkun, Genes Dev. 21Samuelson AV, Carr CE, Ruvkun G (2007) Gene activities that mediate increased life span of C. elegans insulin-like signaling mutants. Genes Dev 21: 2976-2994. Longevity network: construction and implications. A Budovsky, A Abramovich, R Cohen, V Chalifa-Caspi, V Fraifeld, Mech Ageing Dev. 128Budovsky A, Abramovich A, Cohen R, Chalifa-Caspi V, Fraifeld V (2007) Longevity network: construction and implications. Mech Ageing Dev 128: 117- 124. Interaction networks as a tool to investigate the mechanisms of aging. E Chautard, N Thierry-Mieg, Ricard - Blum, S , Biogerontology. 11Chautard E, Thierry-Mieg N, Ricard-Blum S (2010) Interaction networks as a tool to investigate the mechanisms of aging. Biogerontology 11: 463-473. Network medicine: a networkbased approach to human disease. A L Barabasi, N Gulbahce, J Loscalzo, Nat Rev Genet. 12Barabasi AL, Gulbahce N, Loscalzo J (2011) Network medicine: a network- based approach to human disease. Nat Rev Genet 12: 56-68. A gene expression map for Caenorhabditis elegans. S K Kim, J Lund, M Kiraly, K Duke, M Jiang, Science. 293Kim SK, Lund J, Kiraly M, Duke K, Jiang M, et al. (2001) A gene expression map for Caenorhabditis elegans. Science 293: 2087-2092. Cataloging the relationships between proteins: a review of interaction databases. C Rohl, Y Price, T B Fischer, M Paczkowski, M F Zettel, Mol Biotechnol. 34Rohl C, Price Y, Fischer TB, Paczkowski M, Zettel MF, et al. (2006) Cataloging the relationships between proteins: a review of interaction databases. Mol Biotechnol 34: 69-93. A single gene network accurately predicts phenotypic effects of gene perturbation in Caenorhabditis elegans. I Lee, B Lehner, C Crombie, W Wong, A G Fraser, Nat Genet. 40Lee I, Lehner B, Crombie C, Wong W, Fraser AG, et al. (2008) A single gene network accurately predicts phenotypic effects of gene perturbation in Caenorhabditis elegans. Nat Genet 40: 181-188. The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. R Tacutu, A Budovsky, V E Fraifeld, Biogerontology. 11Tacutu R, Budovsky A, Fraifeld VE (2010) The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. Biogerontology 11: 513-522. The Human Ageing Genomic Resources: online databases and tools for biogerontologists. J P De Magalhaes, A Budovsky, G Lehmann, J Costa, Y Li, Aging Cell. 8de Magalhaes JP, Budovsky A, Lehmann G, Costa J, Li Y, et al. (2009) The Human Ageing Genomic Resources: online databases and tools for bioger- ontologists. Aging Cell 8: 65-72. The BioGRID Interaction Database: 2008 update. B J Breitkreutz, C Stark, T Reguly, L Boucher, A Breitkreutz, Nucleic Acids Res. 36Breitkreutz BJ, Stark C, Reguly T, Boucher L, Breitkreutz A, et al. (2008) The BioGRID Interaction Database: 2008 update. Nucleic Acids Res 36: D637-640. Ageand calorie-independent life span extension from dietary restriction by bacterial deprivation in Caenorhabditis elegans. E D Smith, T L Kaeberlein, B T Lydum, J Sager, K L Welton, BMC Dev Biol. 849Smith ED, Kaeberlein TL, Lydum BT, Sager J, Welton KL, et al. (2008) Age- and calorie-independent life span extension from dietary restriction by bacterial deprivation in Caenorhabditis elegans. BMC Dev Biol 8: 49. Demography of dietary restriction and death in Drosophila. W Mair, P Goymer, S D Pletcher, L Partridge, Science. 301Mair W, Goymer P, Pletcher SD, Partridge L (2003) Demography of dietary restriction and death in Drosophila. Science 301: 1731-1733. Rapamycin fed late in life extends lifespan in genetically heterogeneous mice. D E Harrison, R Strong, Z D Sharp, J F Nelson, C M Astle, Nature. 460Harrison DE, Strong R, Sharp ZD, Nelson JF, Astle CM, et al. (2009) Rapamycin fed late in life extends lifespan in genetically heterogeneous mice. Nature 460: 392-395. Timing requirements for insulin/IGF-1 signaling in C. elegans. A Dillin, D Crawford, C Kenyon, Science. 298Dillin A, Crawford D, Kenyon C (2002) Timing requirements for insulin/IGF-1 signaling in C. elegans. Science 298: 830-834. Temporal linkage between the phenotypic and genomic responses to caloric restriction. J M Dhahbi, H J Kim, P L Mote, R J Beaver, S R Spindler, Proc Natl Acad Sci U S A. 101Dhahbi JM, Kim HJ, Mote PL, Beaver RJ, Spindler SR (2004) Temporal linkage between the phenotypic and genomic responses to caloric restriction. Proc Natl Acad Sci U S A 101: 5524-5529. Dominant mutations in the late 40S biogenesis factor Ltv1 affect cytoplasmic maturation of the small ribosomal subunit in Saccharomyces cerevisiae. C A Fassio, B J Schofield, R M Seiser, A W Johnson, D E Lycan, Genetics. 185Fassio CA, Schofield BJ, Seiser RM, Johnson AW, Lycan DE (2010) Dominant mutations in the late 40S biogenesis factor Ltv1 affect cytoplasmic maturation of the small ribosomal subunit in Saccharomyces cerevisiae. Genetics 185: 199- 209. Transcriptional profile of aging in C. elegans. J Lund, P Tedesco, K Duke, J Wang, S K Kim, Curr Biol. 12Lund J, Tedesco P, Duke K, Wang J, Kim SK, et al. (2002) Transcriptional profile of aging in C. elegans. Curr Biol 12: 1566-1573. RNAi screening implicates a SKN-1-dependent transcriptional response in stress resistance and longevity deriving from translation inhibition. J Wang, S Robida-Stubbs, J M Tullet, J F Rual, M Vidal, PLoS Genet. 6Wang J, Robida-Stubbs S, Tullet JM, Rual JF, Vidal M, et al. (2010) RNAi screening implicates a SKN-1-dependent transcriptional response in stress resistance and longevity deriving from translation inhibition. PLoS Genet 6. Inhibition of mRNA translation extends lifespan in Caenorhabditis elegans. K Pan, J Palter, A Rogers, A Olsen, D Chen, Aging Cell. 6Pan K, Palter J, Rogers A, Olsen A, Chen D, et al. (2007) Inhibition of mRNA translation extends lifespan in Caenorhabditis elegans. Aging Cell 6: 111-119. Lifespan extension by conditions that inhibit translation in Caenorhabditis elegans. M Hansen, S Taubert, D Crawford, N Libina, S Lee, Aging Cell. 6Hansen M, Taubert S, Crawford D, Libina N, Lee S, et al. (2007) Lifespan extension by conditions that inhibit translation in Caenorhabditis elegans. Aging Cell 6: 95-110. Synthesis of insulin-dependent activator of hepatocyte plasmatic membrane Na,K-ATPase decreases in aging. V V Frolkis, A L Shkapenko, Mech Ageing Dev. 60Frolkis VV, Shkapenko AL (1991) Synthesis of insulin-dependent activator of hepatocyte plasmatic membrane Na,K-ATPase decreases in aging. Mech Ageing Dev 60: 55-60. Genomes optimize reproduction: aging as a consequence of the developmental program. J P De Magalhaes, G M Church, Physiology (Bethesda). 20de Magalhaes JP, Church GM (2005) Genomes optimize reproduction: aging as a consequence of the developmental program. Physiology (Bethesda) 20: 252- 259. MicroRNA, mRNA, and protein expression link development and aging in human and macaque brain. M Somel, S Guo, N Fu, Z Yan, H Y Hu, Genome Res. 20Somel M, Guo S, Fu N, Yan Z, Hu HY, et al. (2010) MicroRNA, mRNA, and protein expression link development and aging in human and macaque brain. Genome Res 20: 1207-1218. Progressive age-dependent DNA methylation changes start before adulthood in mouse tissues. M Takasugi, Mech Ageing Dev. 132Takasugi M (2011) Progressive age-dependent DNA methylation changes start before adulthood in mouse tissues. Mech Ageing Dev 132: 65-71. Changes in gene expression associated with aging commonly originate during juvenile growth. J C Lui, W Chen, K M Barnes, J Baron, Mech Ageing Dev. 131Lui JC, Chen W, Barnes KM, Baron J (2010) Changes in gene expression associated with aging commonly originate during juvenile growth. Mech Ageing Dev 131: 641-649. . V V Frolkis, Stress-age syndromeFrolkis VV (1991) [Stress-age syndrome]. . Fiziol Zhl. 37Fiziol Zhl 37: 3-11. InParanoid 6: eukaryotic ortholog clusters with inparalogs. A C Berglund, E Sjolund, G Ostlund, E L Sonnhammer, Nucleic Acids Res. 36Berglund AC, Sjolund E, Ostlund G, Sonnhammer EL (2008) InParanoid 6: eukaryotic ortholog clusters with inparalogs. Nucleic Acids Res 36: D263-266. WormBase 2012: more genomes, more data, new website. K Yook, T W Harris, T Bieri, A Cabunoc, J Chan, Nucleic Acids Res. 40Yook K, Harris TW, Bieri T, Cabunoc A, Chan J, et al. (2012) WormBase 2012: more genomes, more data, new website. Nucleic Acids Res 40: D735-741. . A Rogers, I Antoshechkin, T Bieri, D Blasiar, C Bastiani, Nucleic Acids Res. 36Rogers A, Antoshechkin I, Bieri T, Blasiar D, Bastiani C, et al. (2008) WormBase 2007. Nucleic Acids Res 36: D612-617. DAVID: Database for Annotation, Visualization, and Integrated Discovery. G DennisJr, B T Sherman, D A Hosack, J Yang, W Gao, Genome Biol. 43Dennis G, Jr., Sherman BT, Hosack DA, Yang J, Gao W, et al. (2003) DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol 4: P3. Cytoscape: a software environment for integrated models of biomolecular interaction networks. P Shannon, A Markiel, O Ozier, N S Baliga, J T Wang, Genome Res. 13Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, et al. (2003) Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res 13: 2498-2504.

2022-03-13T19:59:13Z

CCBY

https://doi.org/10.1111/acel.12659

GOLD

93d687e8ff81b1ab32374d3bca6c7f4ec9e9c429

10.1111/acel.12659

2745510249

28836369

5676071

Wide-scale comparative analysis of longevity genes and interventions Hagai Yanai The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev POB 6538410501Beer ShevaIsrael Arie Budovsky The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev POB 6538410501Beer ShevaIsrael Biotechnology Unit Technological Center 8489101Beer ShevaIsrael Thomer Barzilay The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev POB 6538410501Beer ShevaIsrael Robi Tacutu Computational Biology of Aging Group Institute of Biochemistry Romanian Academy 060031BucharestRomania Chronos Biosystems SRL BucharestRomania Vadim E Fraifeld The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev POB 6538410501Beer ShevaIsrael Wide-scale comparative analysis of longevity genes and interventions 10.1111/acel.12659comparative analysisevolutionary conservationgene enrichmentgene orthologylongevity genespro- teomepublic and private mechanisms of aging/longevity Hundreds of genes, when manipulated, affect the lifespan of model organisms (yeast, worm, fruit fly, and mouse) and thus can be defined as longevity-associated genes (LAGs). A major challenge is to determine whether these LAGs are model-specific or may play a universal role as longevity regulators across diverse taxa. A wide-scale comparative analysis of the 1805 known LAGs across 205 species revealed that (i) LAG orthologs are substantially overrepresented, from bacteria to mammals, compared to the entire genomes or interactomes, and this was especially noted for essential LAGs; (ii) the effects on lifespan, when manipulating orthologous LAGs in different model organisms, were mostly concordant, despite a high evolutionary distance between them; (iii) LAGs that have orthologs across a high number of phyla were enriched in translational processes, energy metabolism, and DNA repair genes; (iv) LAGs that have no orthologs out of the taxa in which they were discovered were enriched in autophagy (Ascomycota/Fungi), G proteins (Nematodes), and neuroactive ligand-receptor interactions (Chordata). The results also suggest that antagonistic pleiotropy might be a conserved principle of aging and highlight the importance of overexpression studies in the search for longevity regulators. Introduction The role of genetic factors in determination of longevity and aging patterns is an intensively studied issue (Vijg & Suh, 2005;Kenyon, 2010). Hundreds of genes, when manipulated, have been shown to affect the lifespan of model organisms (yeast, worm, fruit fly, and mouse) (Tacutu et al., 2013). These genes (further denoted as longevity-associated genes, LAGs) could be defined as those whose modulation of function or expression (such as gene knockout, overexpression, partial or full loss-offunction mutations, RNA interference, and genetic polymorphisms) results in noticeable changes in longevity-lifespan extension or accelerated aging (Budovsky et al., 2007;Tacutu et al., 2013). We have previously investigated the characteristic features of LAGs and found that (i) they display a marked diversity in their basic function and primary cellular location of the encoded proteins (Budovsky et al., 2007); and (ii) LAG-encoded proteins display a high connectivity and interconnectivity. As a result, they form a scale-free protein-protein interaction network ('longevity network'), indicating that LAGs could act in a cooperative manner (Budovsky et al., 2007;Wolfson et al., 2009;Tacutu et al., 2010aTacutu et al., , 2011Tacutu et al., , 2012. (iii) Many LAGs, particularly those that are hubs in the 'longevity network', are involved in age-related diseases (including atherosclerosis, type 2 diabetes, cancer, and Alzheimer's disease), and in aging-associated conditions (such as oxidative stress, chronic inflammation, and cellular senescence) (Budovsky et al., 2007Wolfson et al., 2009;Tacutu et al., 2010aTacutu et al., , 2011. (iv) The majority of LAGs established by that time in yeast, worms, flies, and mice have human orthologs, indicating their conservation 'from yeast to humans' (Budovsky et al., 2007). This assumption was also supported by studies on specific LAGs or pathways such as Foxo (Martins et al., 2016), insulin/IGF1/mTOR signaling (Tatar et al., 2003;Warner, 2005;Piper et al., 2008;Ziv & Hu, 2011;Gems & Partridge, 2013;Zhang & Liu, 2014;Pitt & Kaeberlein, 2015), Gadd45 (Moskalev et al., 2012), and cell-cell and cell-extracellular matrix interaction proteins . Again, the above studies were limited only to the four model organisms and humans. Now, the existing databases on orthologs allow for an essential extension of the analysis of LAG orthology, far beyond the traditional model organisms and humans. In particular, the data deposited in the InParanoid database-Eukaryotic Ortholog Groups (http://inparanoid. sbc.su.se/, Sonnhammer & Ostlund, 2015) include orthologs for the complete proteomes of 273 species. Here, we report the results of an unprecedentedly wide-scale analysis of 1805 LAGs established in model organisms (available at Human Ageing Genomic Resources (HAGR)-GenAge database; http://genomics.senescence.info/genes/longevity. html, Tacutu et al., 2013), with regard to their putative relevance to public and private mechanisms of aging. Results Orthology of longevity-associated genes Our first question was how LAGs orthologs are distributed across diverse taxonomic groups. For that purpose, we extracted the LAG orthologs for all the species in the InParanoid database, using a software developed in our laboratory (see Methods). For each gene of interest, the evolutionary conservation was evaluated as the presence or absence of orthologs across 205 proteomes (all species available excluding parasites) for a high InParanoid score of 1.0. Parasites were excluded from the analysis because they usually keep the minimal set of genes required for survival in the hosts, and thus, their inclusion could bias the results into overstating the conservation of these genes and diminish the conservation of others. As seen in Fig. 1, for the vast majority of InParanoid species, the fraction of conserved genes was significantly higher for LAGs than for the entire proteome of the same model organism. The few exceptions were fringe cases where the baseline orthology was either very high (phylogenetically very close species, for example, Caenorhabditis elegans and Caenorhabditis briggsae), or very low (phylogenetically very distant species, for example, Mus musculus and Korarchaeum cryptofilum) (Table S1). Remarkably, despite the high diversity of the species under analysis, the ratio between the LAG orthologs and the orthologs of the entire proteome was relatively constant along the evolutionary axis (Fig. 2). This could indicate that the high conservation of LAGs is relatively independent of evolutionary distance. As lifespan extension experiments can be regarded as more robust, we placed a special focus on LAGs that extended lifespan, when manipulated (LSE-LAGs). The results indicate that the major principles still hold (Fig. 1). For the three lower model organisms (S. cerevisiae, C. elegans, D. melanogaster), the distribution of the LSE-LAGs orthologs was almost identical to all LAGs; a clear trend was also observed for M. musculus, although it did not reach a statistically significant value, most likely because of the relatively low number of lifespan-extending interventions in this model. It should be taken into account that genes which a priori have orthologs in humans and are involved in basic biological processes or major diseases are more often tested for their potential effect on lifespan. Despite this obvious bias, an important point is that among the model organisms examined, the highest conservation ratio was observed for C. elegans (P < E-8 for all comparisons; Fig. 2a), where the majority of LAGs were identified by means of an unbiased genome-wide RNA interference (RNAi) screens (Lee et al., 2003;Hamilton et al., 2005;Hansen et al., 2005;Yanos et al., 2012). Postdevelopmental gene inactivation using RNAi is of special interest. This is because it allows for discovering longevity regulators that could not be discovered otherwise, because their predevelopmental inactivation causes a lethal phenotype . According to WormBase (http://www.wormbase.org/; Howe et al., 2016), 127 of the 733 known C. elegans LAGs are essential for development and growth, which means that worm LAGs are enriched in essential genes by approximately fivefold compared to the entire genome . This is even more pronounced among LAGs that extend worm lifespan by more than 20% when inactivated: They are enriched 15-fold in essential genes. As essential genes are generally more evolutionary conserved than nonessential ones (Tacutu et al., 2011), we looked at the ortholog distribution of the 127 essential worm LAGs and found that they are indeed dramatically more conserved than all LAGs (Fig. 1b). The same is also true for essential LAGs where the postdevelopmental RNAi resulted in lifespan extension (Fig. 1b). Remarkably, postdevelopmental inactivation of worm essential LAGs has been shown to predominantly Fig. 1 Percentage of orthologs of longevity-associated genes (LAGs) from the four model organisms across 205 species. Each graph represents one of the four model organisms and the LAGs discovered for that species. Each dot represents the percentage of orthologs between the model species and a single other species (total of 205 species from all Kingdoms; for a full list of species see Table S1). The entire proteome of the model species (extracted from the InParanoid database) was used as control. The species (X-axis) are ordered in descending order of the percentage of orthologs for the entire proteome. Presented are the ortholog percentage of the entire proteome (gray triangle), LAGs (black circle), LAGs discovered by lifespan extension (LSE-LAGs, gray circle), Caenorhabditis elegans essential LAGs discovered by postdevelopmental RNAi (PD RNAi LAGs, gray x), and C. elegans essential LAGs discovered by postdevelopmental RNAi that resulted in lifespan extension (PD RNAi LSE-LAGs, black +). extend lifespan rather than reduce it (Curran & Ruvkun, 2007;Tacutu et al., 2012), which means that they have detrimental effects later in life. As these genes are essential for the early stages of life, but their postdevelopmental inactivation resulted in lifespan extension, this by definition is consistent with Williams's idea of antagonistic pleiotropy (Williams 1957). In support of this notion are also our previous studies (Budovsky et al., 2007Tacutu et al., 2010bTacutu et al., , 2011Tacutu et al., , 2012 and the study of Promislow (2004). All in all, the results suggest that antagonistic pleiotropy might be a conserved principle of aging. One of the strong features of InParanoid is that it provides the best balance between sensitivity and specificity (Chen et al., 2007). Yet, the proteomes found in the InParanoid database contain many poorly annotated proteins and predicted transcripts that were not experimentally verified (Sonnhammer & Ostlund, 2015). These proteins have relatively few orthologs in other species and therefore could influence the results. In contrast, the interactomes from the BioGrid database (http://www.thebiogrid.org), the largest repository of validated PPIs, almost exclusively include experimentally verified proteins (Chatr-Aryamontri et al., 2015). Therefore, the BioGrid data could serve as an additional, high quality control for a more rigorous testing of the evolutionary conservation of LAGs. For this purpose, we used the interactomes of S. cerevisae, C. elegans, and D. melanogaster. As seen in Fig. S1, the same trend of over-conservation of LAGs was also observed in comparison with the BioGrid control. Mouse was not included in the analysis because its BioGrid gene list still contains a relatively small portion of the entire genome and thus could not provide a reliable control. Altogether, the results clearly show a high evolutionary conservation of LAGs across distant species. With regard to this, a question arises as to whether this observation is attributed to an enrichment of specific categories that are known to be strongly preserved in the course of evolution. From the available data on gene and protein annotations for the four model species, we noted that LAGs are enriched in genes that belong to categories known to be extraordinarily conserved in evolution, such as the ribosomal or mitochondrial genes (Table S2). However, exclusion of LAGs belonging to these categories from the analysis had almost no impact (Fig. S2). Therefore, we conclude that the high evolutionary conservation of LAGs is not solely attributed to an enrichment of proteins from exceptionally conserved categories, but rather reflects a general trend. 'Public' and 'private' LAG categories The distinction between public and private mechanisms of aging and longevity is a fundamental question in comparative studies of biogerontology (Gems & Partridge, 2013). We attempted to shed some light on this subject based on the ortholog distribution of LAGs in different taxa. Yet, it is important to note that if a given LAG is highly evolutionary conserved, it does not automatically translate to its role in a public mechanism of aging. In fact, in order to draw conclusions on public or private mechanisms from the presence or lack of orthologs, one must (i) have a context on the mode of operation of a given protein as its function could differ between species; or (ii) compare groups of proteins belonging to a certain pathway or category, so that generalized Fig. 2 Ratio of LAGs orthologs to the entire proteome. Each graph represents the LAGs discovered in the indicated model species. Each dot represents the ratio between the number of LAG orthologs to the orthologs from the entire proteome, for a single other species (total of 205 species from all Kingdoms; for a full list of species see Table S1). The species (X-axis) are ordered in descending order of ortholog percentage for the entire proteome. assumptions may be made. The data used in this study only allow for the second approach. Thus, we comprised lists of proteins under different conservation criteria, for example, proteins that have orthologs across at least 12 phyla or have orthologs in a limited number of taxa (for more details, see Table S3). As shown in Fig. 3, LAG orthologs are distributed over more phyla than the entire genome, again indicating their wider evolutionary conservation. Nevertheless, while most LAGs are broadly presented across phyla, a considerable portion of them (around 10-20%) are specific to a relatively small number of phyla (Fig. 3). To get further insight into the universality of longevity-associated pathways, we carried out an enrichment analysis for LAGs with orthologs across a large number of phyla ('public') and those that are limited to specific phyla ('private'). For the 'private' analysis, we used the phylum of the corresponding model organism (as depicted in the Table 1), as smaller taxonomic groups did not yield statistically conclusive results. The detailed results of the enrichment analysis are available in Table S4 and Fig. S3. Overall, the analysis of the most conserved (public) LAGs revealed that they, not surprisingly, fall under three major categories (Table 1): (i) ribosome and translational processes, (ii) mitochondria and energy metabolism pathways (including the FoxO pathway), and (iii) DNA repair. At a first view, it may seem that the most conserved LAGs are enriched for these categories just because of their function, regardless of their role in aging, that is, simply because they belong to very basic and therefore highly conserved biological processes. However, comparing the most conserved LAGs against different backgrounds showed that it is not the case. Indeed, these pathways were not only overrepresented when compared to the entire genome or to all LAGs, but-what is most important-when compared to all highly conserved genes (i.e., all genes with orthologs across at least 10 or 12 phyla; Table S4). The results provide a strong support for previous studies (in particular those by McElwee et al., 2007;Smith et al., 2008;Freitas et al., 2011;MacRae et al., 2015;Ma et al., 2016;) and highlight the public role of the above categories in the control of lifespan. It should however be noted that the number of LAGs was not always sufficient for a robust enrichment analysis, especially for the mouse and fly models (see Table S3); the results from the yeast and worm models were more significant and thus more reliable. LAGs were discovered in either lifespan-extending or lifespanreducing experiments (or sometimes both). Do these two groups display any specificity in their enriched pathways/categories? Not surprisingly, the common public categories for both groups included the ribosomal and mitochondrial genes ( Table 1, Table S5 and S6). However, we surprisingly found that the division of all LAGs in such a way also revealed a distinct pattern of enrichment. In particular, an overrepresentation of autophagy-related and DNA repair genes was observed only among the LAGs discovered in lifespan-decreasing experiments (LSD-LAGs; Table 1 and Table S5). In contrast, LAGs discovered by lifespanextending experiments (LSE-LAGs; Table 1 and Table S6) were specifically enriched for oxidative phosphorylation and oxidoreductase. Due to the high evolutionary conservation of LAGs, those that have orthologs only in the same phylum as the model species in which they were discovered are relatively small in number. Because of that, the enrichment analysis of these genes yielded less significant results (Table S4). Nevertheless, the 'private' list for S. cerevisae (i.e., yeast LAGs with orthologs only in Ascomycota/Fungi) was found to be enriched with autophagy-related genes, which can be attributed mostly to LSD-LAGs (Table 1). For LAGs that have orthologs only in Nematoda, we found enrichment in G protein-related genes, apparently attributed to LSE-LAGs. This is surprising because both autophagy and G protein signaling represent basic and highly conserved processes which were shown to be involved in aging and longevity in various model organisms (Lans & Jansen, 2007;Hahm et al., 2009;Rubinsztein et al., 2011;Schneider & Cuervo, 2014). Yet, the unusual enrichment of these pathways in yeast and worms definitely highlights their importance in determination of longevity for these taxa specifically, although it does not exclude their role in mechanisms of aging for higher taxa. For vertebrates, we found a significant enrichment of LSE-LAGs in Neuroactive ligand-receptor interaction, which could reflect the importance of neuroendocrine regulation of aging and longevity in higher organisms (Dilman et al., 1986;Frolkis, 1988;Blagosklonny, 2013). All of the above results obtained by David were also validated by the WebGestalt, EnrichR and Panther tools (see Methods; C. elegans data are presented as an example in Fig. S3). Stress response genes and the importance of overexpression interventions The vast majority of LAGs were discovered by downregulating gene activity (e.g., knockout and RNAi). For example, in C. elegans, only 52 of hundreds of LAGs were discovered by overexpression assays. This could potentially create a bias toward overestimation of certain categories for longevity regulation on the expanse of others. A good illustration of that point is that the 'stress response' category was noticeably absent from the enrichment analysis. Yet, it could be expected that many stress response genes would extend lifespan when upregulated rather than downregulated (Moskalev et al., 2014). Indeed, our analysis shows that overexpression of LAGs listed in the GO database as stress response genes (n = 81; Table S7) almost exclusively (95%) resulted in lifespan extension (Fig. 4). Of these LAGs, 19 were also tested by knockout or knockdown experiments. Remarkably, in 18 cases, this resulted in lifespan reduction and only in the case of Sirtuin-1 in yeast-in lifespan extension. Apart from the longevity value of stress response genes, these observations clearly demonstrate the importance of overexpression experiments in longevity studies. Concordancy and discordancy in lifespan-modulating genetic interventions Considering the conservation of many LAGs over a broad evolutionary distance, a valid question is whether modulating a given LAG in different species has a similar impact on longevity, that is, lifespan extension or Fig. 4 Percentage of manipulations on stress response LAGs that extended the lifespan. Only LAGs which were termed as 'stress response' under the GO classification system are included. Depicted is the percentage of overexpression (black) and knockout/knockdown (gray) interventions that resulted in lifespan extension. All intraspecies differences between the effects of overexpression and knockout/knockdown on lifespan were significant (P < 0.001). The full list of stress response LAGs is available in Table S7. Longevity genes and interventions, H. Yanai et al. 1271 reduction. This was previously addressed for worm and yeast, where the genetic component of lifespan determination was found to be significantly conserved (Smith et al., 2008). Here, we broadened the question to all available model organisms. Namely, we compared all orthologs which were shown to have an impact on longevity in more than one species. Overall, we found that approximately 10% of LAGs' orthologs (n = 184) were identified as such in at least two model organisms; 36 LAGs' orthologs were identified in three and 20 in four model organisms. The number of concordant effects was significantly higher than the discordant ones (P < 0.003). That is, manipulation of LAGs has, more than often, the same effect in different species (Fig. 5, Table S8). Unfortunately, a substantial portion of the genetic interventions in yeast and worms could not be clearly defined as concordant or discordant with other model organisms (Fig. 5a, white), mostly due to a major difference in methods and evaluation criteria (Tacutu et al., 2013). When looking at pairwise comparisons (Fig. 5b), it is evident that the level of concordancy is very high for some pairs of species (for example, M. musculus and D. melanogaster) and lower for others (for example, M. musculus and C. elegans). In order to discern what could bring about this difference, we calculated a conservation index for each pair of orthologs (as previously described by Huang et al., 2004) and compared the results to the concordancy/discordancy of the effects. As seen in Fig. S4, the observed discordancy could not be explained by sequence dissimilarity. One of the possible explanations for the observed discordancy is that in these cases orthologous LAGs were discovered by interventions which greatly differ from one another (e.g., knockout and Fig. 5 Concordancy in LAG manipulations across model organisms. Concordancy was determined according to the classification of LAGs as pro-or anti-longevity genes (Tacutu et al., 2013). That is, if a given LAG was determined as a pro-or anti-longevity gene in two or more species, it was termed 'concordant'; otherwise, it was termed 'discordant'. A detailed table is available in Table S8. (a) Summary of the concordancy for LAGs from each model species which have also been tested in two or more species (interspecies). (b) Venn diagram of the concordancy between species. (c+d) Summary of the concordancy of LAG manipulations within the same species (intraspecies). Longevity genes and interventions, H. Yanai et al. overexpression). As such, if a given LAG is knocked out and as a result the animal ages more rapidly, that LAG is defined as a 'pro-longevity' gene; however, an overexpression of the same LAG will not necessarily increase lifespan. For example, a knockout of G protein, alpha subunit (gpa-9) in C. elegans increases maximum lifespan by up to 50%, but paradoxically, its overexpression also increases the worm maximum lifespan (by 20%) (Schneider & Cuervo, 2014). If such a difference can occur in the same species, more so could be expected when testing for effects on lifespan between different model organisms. Indeed, as is evident from Fig. S5, the concordancy increased significantly (from 73% to 88%) when a similar intervention was performed. Then, at least some of the discordancy could be explained by a variety in the methods of intervention. It should however be noted that the vast majority of intraspecies comparisons of opposite interventions have brought about concordant effects (Fig. 5c), so that we cannot rule out interspecies differences that caused some of the inconsistencies. Interestingly, only five LAGs (Sod2, Sirt1, Mtor, Fxn, and Rps6 kb1; in total, 20 orthologs) were tested for their impact on longevity in all four model species. The manipulations of these genes showed a predominantly concordant effect on longevity, with the exception of Fxn (Frh-1) which has an opposite effect only in C. elegans (Table S7). Altogether, the results indicate a clear trend of concordancy in the effects of LAG manipulations across model species despite a high evolutionary distance between them. The much smaller portion of discordant cases could be attributed to either technical or biological issues, or both. Discussion Our wide-scale analysis of longevity-associated genes (LAGs) shows that their orthologs are consistently overrepresented across diverse taxa, compared with the orthologs of other genes, and this conservation was relatively independent of evolutionary distance (Figs 1-3). The high evolutionary conservation was evident for LAGs discovered in all of the four major model organisms (yeast, S. cerevisae; worm, C. elegans; fly, D. melanogaster; mouse, M. musculus), but was especially relevant for C. elegans, where a large portion of LAGs were identified by genome-wide screens, thus minimizing potential biases. Moreover, many worm LAGs were discovered by postdevelopmental RNAi on genes essential for growth and development, and this predominantly resulted in lifespan extension (Curran & Ruvkun, 2007;Tacutu et al., 2012). That is, postdevelopmental suppression of genes that are vital early in life but are detrimental later in life, can be beneficial for longevity. The orthologs of these LAGs are also highly overrepresented across diverse taxa. Altogether, the C. elegans analysis suggests that antagonistic pleiotropy might be a highly conserved principle of aging. As one of Niven's laws states: 'It is easier to destroy than to create'. Indeed, there are many more ways to make an organism live shorter than to make it live longer. The enrichment analysis demonstrated the difference in public and private pathways/categories, with potential importance for lifespan extension or development of an early aging phenotype (Table 1). It is worthwhile to note that while the enrichment analysis definitely highlights the importance of the overrepresented categories, it does not exclude the importance of nonenriched ones. For example, a well-recognized longevity pathway, insulin/IGF signaling, does not appear as an enriched pathway in our analysis but has been previously shown to be a public mechanism of aging (Piper et al., 2008). Another important category that did not fall into the enriched ones is stress response genes. Stress resistance has long been linked to longevity in many animal models (Johnson et al., 2002;Moskalev et al., 2014). In this study, we have specifically addressed stress response genes and showed that their overexpression mostly results in lifespan extension (Fig. 4). This further emphasizes the importance of overexpression interventions in longevity studies, which should be a point for future investigations. The latter would not only drive the discovery of new longevity regulators but could also strengthen the validity of LAGs that were discovered by knockout or knockdown experiments, as we have shown in this study (Fig. 5c). The recent development of novel CRISPRbased gene activation technologies could provide a strong platform and push toward this approach. An important observation in our study was that the majority of manipulations on LAG orthologs in more than one model animal resulted in concordant effects on longevity (Fig. 5). This strengthens the paradigm of 'public' longevity pathways and of using model animals to study longevity, even across a large evolutionary distance. This notion is further strengthened when combined with the observation of Smith et al. (2008) who demonstrated that the existence of an ortholog is probably accompanied by a preserved role in longevity. Yet, we also observed LAGs with ortholog presence only in a limited number of taxa, or that displayed discordant effects when tested in more than one species (Fig. 5), which could, at least in part, be attributed to 'private' mechanisms of aging. Definitely, more comparative studies are warranted to better discriminate between private and public mechanisms, with unified methods of intervention and evaluation in mind. A recent study by Harel et al. (2015) could serve as a step in that direction by offering a new model of short-lived vertebrate species. In perspective, the combination of the existing data on LAGs with the emerging data on their expression throughout lifespan could bring about a deeper understanding of the role of genetic factors in aging and longevity. Experimental procedures Gene lists Longevity-associated genes The longevity-associated genes (LAGs) are defined as genes whose genetic manipulation in model organisms (M. musculus, D. melanogaster, C. elegans and S. cerevisiae) was shown to significantly affect their lifespan. The list was obtained from Human Ageing Genomic Resources (HAGR)-GenAge database (http://genomics.senescence.info/gene s/longevity.html; Tacutu et al., 2013). Interactome genes Interactome genes were extracted from the BioGrid database (http:// www.thebiogrid.org; Chatr-Aryamontri et al., 2015) and were used as additional, high quality control for a more rigorous testing of the evolutionary conservation of LAGs. Essential genes Genes essential for the development and growth of C. elegans were extracted from WormBase (http://www.wormbase.org/; Howe et al., 2016). Stress response genes Genes classified under the category of 'stress response' were extracted by the UniProt Retrieve ID/Mapping service (http://www.uniprot.org/ uploadlists/; Pundir et al., 2017) Determination of orthology Ortholog determination for each gene was based on the InParanoid database-Eukaryotic Ortholog Groups (http://inparanoid.sbc.su.se/; Sonnhammer & Ostlund, 2015). The analysis was performed for 205 species (all species available excluding parasites; for a full list, see Table S1). The ortholog extraction was performed automatically using software developed in our laboratory. The taxonomy of the species examined was based on the ITIS database (http://www.itis.gov/). The statistical significance of conservation for a group of genes was evaluated with the chi-squared goodness of fit test. Gene set enrichment Enrichment analysis was performed using David Bioinformatics Resources 6.8 (https://david-d.ncifcrf.gov/; Huang et al., 2009), Web-Gestalt (http://www.webgestalt.org/; Wang et al., 2013), and EnrichR (http://amp.pharm.mssm.edu/Enrichr/; Kuleshov et al., 2016). The enrichment analysis was performed against three different backgrounds, including the whole genome, all LAGs, and the genes of the model organism under the same conservation criteria depicted in Tables S4-S6. Supporting Information Additional Supporting Information may be found online in the supporting information tab for this article. Table S1 List of species used for orthology analysis. (a) Saccharomyces cerevisiae, n = 6590 for control, 824 for all LAGs, and 277 for LSE-LAGs. (b) C. elegans, n = 20 325 for control, 733 for all LAGs, 491 for LSE-LAGs, 127 for PD RNAi LAGs, and 107 for PD RNAi LSE-LAGs. (c) Drosophila melanogaster, n = 13 250 for control, 136 for all LAGs, and 85 for LSE-LAGs. (d) Mus musculus, n = 21 895 for control, 112 for all LAGs, and 42 for LSE-LAGs. The vast majority of pairwise differences between LAGs and the entire proteome are significant (P < 0.05), with a few exceptions of fringe cases as described in the text. For most M. musculus LSE-LAGs, the pairwise differences are insignificant (P > 0.05), with a few exceptions where the number of orthologs was relatively high. Longevity genes and interventions, H. Yanai et al. (a) Saccharomyces cerevisiae, n = 6590 for control and 824 for LAGs; (b) Caenorhabditis elegans, n = 20 325 for control and 733 for LAGs; (c) Drosophila melanogaster, n = 13 250 for control and 136 for LAGs; (d) Mus musculus, n = 21 895 for control and 112 for LAGs. Fig. 3 3Distribution of LAGs according to the number of phyla in which LAGs have orthologs. Each graph represents the distribution of LAGs (gray area) discovered in the indicated model species. The entire proteome was used as a control (dotted line). X-axis depicts the number of phyla in which the genes have orthologs. The medians of the distributions are presented as vertical lines: dotted line for all genes and smooth black line for LAGs. Longevity genes and interventions, H. Yanai et al. Fig. S1 S1Percentage of Interactome LAG orthologs from the four model species. Fig. S2 S2Percentage of LAG orthologs from the four model species after exclusion of proteins from enriched categories. Fig. S3 S3GO Slim summary and enrichment analysis. Fig. S4 S4Conservation index (CI) compared to concordancy of longevity effects. Fig. S5 S5Method similarity score compared to concordancy of longevity effects. Table 1 ' 1Public' and 'private' enriched categories. The table depicts the most enriched categories for lists of proteins of all longevity-associated genes (all LAGs) and LAGs discovered by either lifespan extension (LSE-LAGs) or lifespan reduction (LSD-LAGs), under different evolutionary conservation criteria (defined as the presence of orthologs across a listed number of phyla). For a detailed enrichment analysis, seeTables S4-S6Public/Private Saccharomyces cerevisiae Caenorhabditis elegans Drosophila melanogaster Mus musculus Public Taxa groups where orthologs are present at least 12 phyla at least 12 phyla at least 10 phyla at least 10 phyla All LAGs Ribosome and translation Mitochondria Citrate cycle (TCA cycle) Ribosome and translation Mitochondria Oxidative phosphorylation NADH activity FoxO signaling Autophagy DNA repair, especially Nucleotide excision repair LSE-LAGs Ribosome and translation Ribosome and translation Mitochondria Oxidative phosphorylation Development Oxidoreductase No enrichment LSD-LAGs Ribosome and translation Mitochondria Citrate cycle (TCA cycle) Autophagy DNA repair Autophagy Autophagy FoxO signaling DNA repair, especially Nucleotide excision repair Private (indicated taxa) Taxa groups where orthologs are present only in Fungi/Ascomycota only in Nematoda only in Arthropoda only in Chordata All LAGs Autophagy G protein related No enrichment Neuroactive ligand-receptor interaction LSE-LAGs Meiosis G protein related No enrichment Neuroactive ligand-receptor interaction LSD-LAGs Autophagy Mitochondrion Mitophagy Transcription regulation No enrichment No enrichment Table S2 S2Enrichment of KEGG and GOCC pathways in Longevity Associated Genes (LAGs) from different species. Table S3 S3List of proteins under different evolutionary conservation criteria. Table S4 S4Enrichment analysis of Longevity-associated genes (LAGs) under different criteria and backgrounds. Table S5 S5Enrichment analysis of lifespan decreasing Longevity-associated genes (LAGs) under different criteria and backgrounds. Table S6 S6Enrichment analysis of lifespan extending Longevity-associated genes (LAGs) under different criteria and backgrounds. Table S7 S7List of LAGs that are listed as stress response genes under the GO classification system. Table S8 S8Pairwise concordance of LAG manipulation. Longevity genes and interventions, H. Yanai et al. 1275 ª 2017 The Authors. Aging Cell published by the Anatomical Society and John Wiley & Sons Ltd. AcknowledgmentsWe would like to thank Prof. Marina Wolfson for her helpful suggestions and insightful comments. We also thank the anonymous reviewers for their valuable comments and suggestions which allowed us to greatly improve the manuscript.Conflict of interestNone declared.Author contributionsAll authors participated in data collection and analysis. In addition, TB wrote all the programs for data extraction from the databases used and, in particular, the ortholog extraction for large sets of genes and species and the calculation of the conservation index. HY wrote the manuscript and prepared the figures and tables. AB participated in writing the manuscript. RT was involved in the computational aspects of analysis. VEF coordinated the study.Aging Cell M(o)TOR of aging: MTOR as a universal molecular hypothalamus. M V Blagosklonny, Aging (Albany NY). 5Blagosklonny MV (2013) M(o)TOR of aging: MTOR as a universal molecular hypothalamus. Aging (Albany NY) 5, 490-494. Longevity network: construction and implications. A Budovsky, A Abramovich, R Cohen, V Chalifa-Caspi, V Fraifeld, Mech. Ageing Dev. 128Budovsky A, Abramovich A, Cohen R, Chalifa-Caspi V, Fraifeld V (2007) Longevity network: construction and implications. Mech. Ageing Dev. 128, 117-124. Common gene signature of cancer and longevity. A Budovsky, R Tacutu, H Yanai, A Abramovich, M Wolfson, V Fraifeld, Mech. Ageing Dev. 130Budovsky A, Tacutu R, Yanai H, Abramovich A, Wolfson M, Fraifeld V (2009) Common gene signature of cancer and longevity. Mech. Ageing Dev. 130, 33-39. The BioGRID interaction database: 2015 update. A Chatr-Aryamontri, B J Breitkreutz, R Oughtred, L Boucher, S Heinicke, D Chen, C Stark, A Breitkreutz, N Kolas, O&apos; Donnell, L Reguly, T Nixon, J Ramage, L Winter, A Sellam, A Chang, C Hirschman, J Theesfeld, C Rust, J Livstone, M S Dolinski, K Tyers, M , Nucleic Acids Res. 43Chatr-Aryamontri A, Breitkreutz BJ, Oughtred R, Boucher L, Heinicke S, Chen D, Stark C, Breitkreutz A, Kolas N, O'Donnell L, Reguly T, Nixon J, Ramage L, Winter A, Sellam A, Chang C, Hirschman J, Theesfeld C, Rust J, Livstone MS, Dolinski K, Tyers M (2015) The BioGRID interaction database: 2015 update. Nucleic Acids Res. 43, D470-D478. Assessing performance of orthology detection strategies applied to eukaryotic genomes. F Chen, A J Mackey, J K Vermunt, D S Roos, PLoS ONE. 2383Chen F, Mackey AJ, Vermunt JK, Roos DS (2007) Assessing performance of orthology detection strategies applied to eukaryotic genomes. PLoS ONE 2, e383. Lifespan regulation by evolutionarily conserved genes essential for viability. S P Curran, G Ruvkun, PLoS Genet. 356Curran SP, Ruvkun G (2007) Lifespan regulation by evolutionarily conserved genes essential for viability. PLoS Genet. 3, e56. Neuroendocrine-ontogenetic mechanism of aging: toward an integrated theory of aging. V M Dilman, S Y Revskoy, A G Golubev, Int. Rev. Neurobiol. 28Dilman VM, Revskoy SY, Golubev AG (1986) Neuroendocrine-ontogenetic mechanism of aging: toward an integrated theory of aging. Int. Rev. Neurobiol. 28, 89-156. A data mining approach for classifying DNA repair genes into ageing-related or non-ageing-related. A A Freitas, O Vasieva, J P De Magalhaes, BMC Genom. 1227Freitas AA, Vasieva O, de Magalhaes JP (2011) A data mining approach for classifying DNA repair genes into ageing-related or non-ageing-related. BMC Genom. 12, 27. A hundred questions on neurohumoral mechanisms of aging. V V Frolkis, Gerontology. 34Frolkis VV (1988) A hundred questions on neurohumoral mechanisms of aging. Gerontology 34, 6-13. Genetics of longevity in model organisms: debates and paradigm shifts. D Gems, L Partridge, Annu. Rev. Physiol. 75Gems D, Partridge L (2013) Genetics of longevity in model organisms: debates and paradigm shifts. Annu. Rev. Physiol. 75, 621-644. Endogenous cGMP regulates adult longevity via the insulin signaling pathway in Caenorhabditis elegans. J H Hahm, S Kim, Y K Paik, Aging Cell. 8Hahm JH, Kim S, Paik YK (2009) Endogenous cGMP regulates adult longevity via the insulin signaling pathway in Caenorhabditis elegans. Aging Cell 8, 473-483. A systematic RNAi screen for longevity genes in C. elegans. B Hamilton, Y Dong, M Shindo, W Liu, I Odell, G Ruvkun, S S Lee, Genes Dev. 19Hamilton B, Dong Y, Shindo M, Liu W, Odell I, Ruvkun G, Lee SS (2005) A systematic RNAi screen for longevity genes in C. elegans. Genes Dev. 19, 1544- 1555. New genes tied to endocrine, metabolic, and dietary regulation of lifespan from a Caenorhabditis elegans genomic RNAi screen. M Hansen, A L Hsu, A Dillin, C Kenyon, PLoS Genet. 1Hansen M, Hsu AL, Dillin A, Kenyon C (2005) New genes tied to endocrine, metabolic, and dietary regulation of lifespan from a Caenorhabditis elegans genomic RNAi screen. PLoS Genet. 1, 119-128. A platform for rapid exploration of aging and diseases in a naturally short-lived vertebrate. I Harel, B A Benayoun, B Machado, P P Singh, C K Hu, M F Pech, D R Valenzano, E Zhang, S C Sharp, S E Artandi, A Brunet, Cell. 160Harel I, Benayoun BA, Machado B, Singh PP, Hu CK, Pech MF, Valenzano DR, Zhang E, Sharp SC, Artandi SE, Brunet A (2015) A platform for rapid exploration of aging and diseases in a naturally short-lived vertebrate. Cell 160, 1013-1026. WormBase 2016: expanding to enable helminth genomic research. K L Howe, B J Bolt, S Cain, J Chan, W J Chen, P Davis, J Done, T Down, S Gao, C Grove, T W Harris, R Kishore, R Lee, J Lomax, Y Li, H M Muller, C Nakamura, P Nuin, M Paulini, D Raciti, G Schindelman, E Stanley, M A Tuli, K Van Auken, D Wang, X Wang, G Williams, A Wright, K Yook, M Berriman, P Kersey, T Schedl, L Stein, P W Sternberg, Nucleic Acids Res. 44Howe KL, Bolt BJ, Cain S, Chan J, Chen WJ, Davis P, Done J, Down T, Gao S, Grove C, Harris TW, Kishore R, Lee R, Lomax J, Li Y, Muller HM, Nakamura C, Nuin P, Paulini M, Raciti D, Schindelman G, Stanley E, Tuli MA, Van Auken K, Wang D, Wang X, Williams G, Wright A, Yook K, Berriman M, Kersey P, Schedl T, Stein L, Sternberg PW (2016) WormBase 2016: expanding to enable helminth genomic research. Nucleic Acids Res. 44, D774-D780. Evolutionary conservation and selection of human disease gene orthologs in the rat and mouse genomes. H Huang, E E Winter, H Wang, K G Weinstock, H Xing, L Goodstadt, P D Stenson, D N Cooper, D Smith, M M Alba, C P Ponting, K Fechtel, Genome Biol. 547Huang H, Winter EE, Wang H, Weinstock KG, Xing H, Goodstadt L, Stenson PD, Cooper DN, Smith D, Alba MM, Ponting CP, Fechtel K (2004) Evolutionary conservation and selection of human disease gene orthologs in the rat and mouse genomes. Genome Biol. 5, R47. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. W Huang, B T Sherman, R A Lempicki, Nat. Protoc. 4Huang W, Sherman BT, Lempicki RA (2009) Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc. 4, 44-57. Longevity genes in the nematode Caenorhabditis elegans also mediate increased resistance to stress and prevent disease. T E Johnson, S Henderson, S Murakami, E De Castro, S H De Castro, J Cypser, B Rikke, P Tedesco, C Link, J. Inherit. Metab. Dis. 25Johnson TE, Henderson S, Murakami S, de Castro E, de Castro SH, Cypser J, Rikke B, Tedesco P, Link C (2002) Longevity genes in the nematode Caenorhabditis elegans also mediate increased resistance to stress and prevent disease. J. Inherit. Metab. Dis. 25, 197-206. The genetics of ageing. C J Kenyon, Nature. 464Kenyon CJ (2010) The genetics of ageing. Nature 464, 504-512. Enrichr: a comprehensive gene set enrichment analysis web server update. M V Kuleshov, M R Jones, A D Rouillard, N F Fernandez, Q Duan, Z Wang, S Koplev, S L Jenkins, K M Jagodnik, A Lachmann, M G Mcdermott, C D Monteiro, G W Gundersen, Ma , Nucleic Acids Res. 44W1Kuleshov MV, Jones MR, Rouillard AD, Fernandez NF, Duan Q, Wang Z, Koplev S, Jenkins SL, Jagodnik KM, Lachmann A, McDermott MG, Monteiro CD, Gundersen GW, Ma'ayan A (2016) Enrichr: a comprehensive gene set enrichment analysis web server update. Nucleic Acids Res. 44(W1), W90-W97. Multiple sensory G proteins in the olfactory, gustatory and nociceptive neurons modulate longevity in Caenorhabditis elegans. H Lans, G Jansen, Dev. Biol. 303Lans H, Jansen G (2007) Multiple sensory G proteins in the olfactory, gustatory and nociceptive neurons modulate longevity in Caenorhabditis elegans. Dev. Biol. 303, 474-482. A systematic RNAi screen identifies a critical role for mitochondria in C. elegans longevity. S S Lee, R Y Lee, A G Fraser, R S Kamath, J Ahringer, G Ruvkun, Nat. Genet. 33Lee SS, Lee RY, Fraser AG, Kamath RS, Ahringer J, Ruvkun G (2003) A systematic RNAi screen identifies a critical role for mitochondria in C. elegans longevity. Nat. Genet. 33, 40-48. Cell culturebased profiling across mammals reveals DNA repair and metabolism as determinants of species longevity. S Ma, A Upneja, A Galecki, Y M Tsai, C F Burant, S Raskind, Q Zhang, Z D Zhang, A Seluanov, V Gorbunova, C B Clish, R A Miller, V N Gladyshev, 519130Ma S, Upneja A, Galecki A, Tsai YM, Burant CF, Raskind S, Zhang Q, Zhang ZD, Seluanov A, Gorbunova V, Clish CB, Miller RA, Gladyshev VN (2016) Cell culture- based profiling across mammals reveals DNA repair and metabolism as determinants of species longevity. Elife 5, e19130. DNA repair in species with extreme lifespan differences. S L Macrae, M M Croken, R B Calder, A Aliper, B Milholland, R R White, A Zhavoronkov, V N Gladyshev, A Seluanov, V Gorbunova, Z D Zhang, J Vijg, Aging (Albany NY). 7MacRae SL, Croken MM, Calder RB, Aliper A, Milholland B, White RR, Zhavoronkov A, Gladyshev VN, Seluanov A, Gorbunova V, Zhang ZD, Vijg J (2015) DNA repair in species with extreme lifespan differences. Aging (Albany NY) 7, 1171-1184. Long live FOXO: unraveling the role of FOXO proteins in aging and longevity. R Martins, G J Lithgow, W Link, Aging Cell. 15Martins R, Lithgow GJ, Link W (2016) Long live FOXO: unraveling the role of FOXO proteins in aging and longevity. Aging Cell 15, 196-207. Evolutionary conservation of regulated longevity assurance mechanisms. J J Mcelwee, E Schuster, E Blanc, M D Piper, J H Thomas, D S Patel, C Selman, D J Withers, J M Thornton, L Partridge, D Gems, Genome Biol. 8132McElwee JJ, Schuster E, Blanc E, Piper MD, Thomas JH, Patel DS, Selman C, Withers DJ, Thornton JM, Partridge L, Gems D (2007) Evolutionary conservation of regulated longevity assurance mechanisms. Genome Biol. 8, R132. Gadd45 proteins: relevance to aging, longevity and age-related pathologies. A A Moskalev, Z Smit-Mcbride, M V Shaposhnikov, E N Plyusnina, A Zhavoronkov, A Budovsky, R Tacutu, V E Fraifeld, Ageing Res. Rev. 11Moskalev AA, Smit-McBride Z, Shaposhnikov MV, Plyusnina EN, Zhavoronkov A, Budovsky A, Tacutu R, Fraifeld VE (2012) Gadd45 proteins: relevance to aging, longevity and age-related pathologies. Ageing Res. Rev. 11, 51-66. Genetics and epigenetics of aging and longevity. A A Moskalev, A M Aliper, Z Smit-Mcbride, A Buzdin, A Zhavoronkov, Cell Cycle. 13Moskalev AA, Aliper AM, Smit-McBride Z, Buzdin A, Zhavoronkov A (2014) Genetics and epigenetics of aging and longevity. Cell Cycle 13, 1063-1077. Separating cause from effect: how does insulin/IGF signalling control lifespan in worms, flies and mice?. M D Piper, C Selman, J J Mcelwee, L Partridge, J. Intern. Med. 263Piper MD, Selman C, McElwee JJ, Partridge L (2008) Separating cause from effect: how does insulin/IGF signalling control lifespan in worms, flies and mice? J. Intern. Med. 263, 179-191. Why is aging conserved and what can we do about it?. J N Pitt, M Kaeberlein, PLoS Biol. 131002131Pitt JN, Kaeberlein M (2015) Why is aging conserved and what can we do about it? PLoS Biol. 13, e1002131. Protein networks, pleiotropy and the evolution of senescence. D E Promislow, Proc. Biol. Sci. 271Promislow DE (2004) Protein networks, pleiotropy and the evolution of senes- cence. Proc. Biol. Sci. 271, 1225-1234. . S Pundir, M J Martin, C O&apos;donovan, UniProt Protein Knowledgebase. Methods Mol. Biol. 1558Pundir S, Martin MJ, O'Donovan C (2017) UniProt Protein Knowledgebase. Methods Mol. Biol. 1558, 41-55. . D C Rubinsztein, G Marino, G Kroemer, Autophagy and aging. Cell. 146Rubinsztein DC, Marino G, Kroemer G (2011) Autophagy and aging. Cell 146, 682-695. Autophagy and human disease: emerging themes. J L Schneider, A M Cuervo, Curr. Opin. Genet. Dev. 26Schneider JL, Cuervo AM (2014) Autophagy and human disease: emerging themes. Curr. Opin. Genet. Dev. 26, 16-23. Quantitative evidence for conserved longevity pathways between divergent eukaryotic species. E D Smith, M Tsuchiya, L A Fox, N Dang, D Hu, E O Kerr, E D Johnston, B N Tchao, D N Pak, K L Welton, D E Promislow, J H Thomas, M Kaeberlein, B K Kennedy, Genome Res. 18Smith ED, Tsuchiya M, Fox LA, Dang N, Hu D, Kerr EO, Johnston ED, Tchao BN, Pak DN, Welton KL, Promislow DE, Thomas JH, Kaeberlein M, Kennedy BK (2008) Quantitative evidence for conserved longevity pathways between divergent eukaryotic species. Genome Res. 18, 564-570. InParanoid 8: orthology analysis between 273 proteomes, mostly eukaryotic. E L Sonnhammer, G Ostlund, Nucleic Acids Res. 43Sonnhammer EL, Ostlund G (2015) InParanoid 8: orthology analysis between 273 proteomes, mostly eukaryotic. Nucleic Acids Res. 43, D234-D239. The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. R Tacutu, A Budovsky, V E Fraifeld, Biogerontology. 11Tacutu R, Budovsky A, Fraifeld VE (2010a) The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. Biogeron- tology 11, 513-522. MicroRNA-regulated protein-protein interaction networks: how could they help in searching for prolongevity targets?. R Tacutu, A Budovsky, M Wolfson, V E Fraifeld, Rejuvenation Res. 13Tacutu R, Budovsky A, Wolfson M, Fraifeld VE (2010b) MicroRNA-regulated protein-protein interaction networks: how could they help in searching for pro- longevity targets? Rejuvenation Res. 13, 373-377. Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. R Tacutu, A Budovsky, H Yanai, V E Fraifeld, Aging (Albany NY). 3Tacutu R, Budovsky A, Yanai H, Fraifeld VE (2011) Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. Aging (Albany NY) 3, 1178-1191. Prediction of C. elegans longevity genes by human and worm longevity networks. R Tacutu, D E Shore, A Budovsky, J P De Magalhaes, G Ruvkun, V E Fraifeld, S P Curran, PLoS ONE. 748282Tacutu R, Shore DE, Budovsky A, de Magalhaes JP, Ruvkun G, Fraifeld VE, Curran SP (2012) Prediction of C. elegans longevity genes by human and worm longevity networks. PLoS ONE 7, e48282. Human Ageing Genomic Resources: integrated databases and tools for the biology and genetics of ageing. R Tacutu, T Craig, A Budovsky, D Wuttke, G Lehmann, D Taranukha, J Costa, V E Fraifeld, J P De Magalhaes, Nucleic Acids Res. 41Tacutu R, Craig T, Budovsky A, Wuttke D, Lehmann G, Taranukha D, Costa J, Fraifeld VE, de Magalhaes JP (2013) Human Ageing Genomic Resources: integrated databases and tools for the biology and genetics of ageing. Nucleic Acids Res. 41, D1027-D1033. The endocrine regulation of aging by insulinlike signals. M Tatar, A Bartke, A Antebi, Science. 299Tatar M, Bartke A, Antebi A (2003) The endocrine regulation of aging by insulin- like signals. Science 299, 1346-1351. Genetics of longevity and aging. J Vijg, Y Suh, Annu. Rev. Med. 56Vijg J, Suh Y (2005) Genetics of longevity and aging. Annu. Rev. Med. 56, 193-212. WEB-based GEne SeT AnaLysis Toolkit (WebGestalt): update 2013. J Wang, D Duncan, Z Shi, B Zhang, Nucleic Acids Res. 41Wang J, Duncan D, Shi Z, Zhang B (2013) WEB-based GEne SeT AnaLysis Toolkit (WebGestalt): update 2013. Nucleic Acids Res. 41, W77-W83. Longevity genes: from primitive organisms to humans. H R Warner, Mech. Ageing Dev. 126Warner HR (2005) Longevity genes: from primitive organisms to humans. Mech. Ageing Dev. 126, 235-242. Pleiotropy, natural selection, and the evolution of senescence. G C Williams, 10.2307/2406060Evolution. 11Williams GC (1957) Pleiotropy, natural selection, and the evolution of senescence. Evolution 11, 398-411. https://doi.org/10.2307/2406060. The signaling hubs at the crossroad of longevity and age-related disease networks. M Wolfson, A Budovsky, R Tacutu, V Fraifeld, Int. J. Biochem. Cell Biol. 41Wolfson M, Budovsky A, Tacutu R, Fraifeld V (2009) The signaling hubs at the crossroad of longevity and age-related disease networks. Int. J. Biochem. Cell Biol. 41, 516-520. Genome-Wide RNAi Longevity Screens in Caenorhabditis elegans. M E Yanos, C F Bennett, M Kaeberlein, Curr. Genomics. 13Yanos ME, Bennett CF, Kaeberlein M (2012) Genome-Wide RNAi Longevity Screens in Caenorhabditis elegans. Curr. Genomics 13, 508-518. Tissue-specific insulin signaling in the regulation of metabolism and aging. J Zhang, F Liu, IUBMB Life. 66Zhang J, Liu F (2014) Tissue-specific insulin signaling in the regulation of metabolism and aging. IUBMB Life 66, 485-495. Genetic variation in insulin/IGF-1 signaling pathways and longevity. E Ziv, D Hu, Ageing Res. Rev. 10Ziv E, Hu D (2011) Genetic variation in insulin/IGF-1 signaling pathways and longevity. Ageing Res. Rev. 10, 201-204.

2022-03-01T11:08:03Z

CCBY

https://doi.org/10.1093/nar/gkx1042

GOLD

656ed0833088fbdfccecd4808caebe1925fdadc9

journals/nar/TacutuTJBBCDLTW18

10.1093/nar/gkx1042

2767517408

29121237

5753192

Human Ageing Genomic Resources: new and updated databases 2018 Robi Tacutu Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool L7 8TXLiverpoolUK Institute of Biochemistry Computational Biology of Aging Group Romanian Academy 060031BucharestRomania Daniel Thornton Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool L7 8TXLiverpoolUK Emily Johnson Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool L7 8TXLiverpoolUK Arie Budovsky The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 84105Beer-ShevaIsrael Judea Regional Research & Development Center 90404CarmelIsrael Diogo Barardo Department of Biochemistry Yong Loo Lin School of Medicine National University of Singapore 117597Singapore CitySingapore Science Division Yale-NUS College 138527Singapore City, Singapore Thomas Craig Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool L7 8TXLiverpoolUK Eugene Diana Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool L7 8TXLiverpoolUK Gilad Lehmann The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 84105Beer-ShevaIsrael Dmitri Toren The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 84105Beer-ShevaIsrael Jingwei Wang Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool L7 8TXLiverpoolUK Vadim E Fraifeld The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 84105Beer-ShevaIsrael João P De Magalhães Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool L7 8TXLiverpoolUK Human Ageing Genomic Resources: new and updated databases Nucleic Acids Research 46201810.1093/nar/gkx1042Received September 18, 2017; Revised October 15, 2017; Editorial Decision October 17, 2017; Accepted October 18, 2017 In spite of a growing body of research and data, human ageing remains a poorly understood process. Over 10 years ago we developed the Human Ageing Genomic Resources (HAGR), a collection of databases and tools for studying the biology and genetics of ageing. Here, we present HAGR's main functionalities, highlighting new additions and improvements. HAGR consists of six core databases: (i) the GenAge database of ageing-related genes, in turn composed of a dataset of >300 human ageing-related genes and a dataset with >2000 genes associated with ageing or longevity in model organisms; (ii) the AnAge database of animal ageing and longevity, featuring >4000 species; (iii) the GenDR database with >200 genes associated with the life-extending effects of dietary restriction; (iv) the LongevityMap database of human genetic association studies of longevity with >500 entries; (v) the DrugAge database with >400 ageing or longevity-associated drugs or compounds; (vi) the CellAge database with >200 genes associated with cell senescence. All our databases are manually curated by experts and regularly updated to ensure a high quality data. Cross-links across our databases and to external resources help researchers locate and integrate relevant information. HAGR is freely available online (http://genomics. senescence.info/). INTRODUCTION Ageing is a complex biological process that, despite decades of research, is not yet well understood. Many age-related changes have been described, however the theories regarding which mechanisms drive ageing changes are still controversial (1). Since their conception, the Human Ageing Genomic Resources (HAGR) have aimed to tackle this complex problem, rapidly becoming a leading online resource for biogerontologists. With the advent of large scale sequencing and breakthroughs in the genetics of ageing, HAGR has a particular (but not exclusive) focus on genomics. As the field of ageing research has grown the amount of data being generated has rapidly increased. Since its first publication in 2005 (2), HAGR has expanded considerably to match this increase. Having started with only two databases, GenAge, a database of genes potentially associated with human ageing, and AnAge, a database of ageing and longevity in animals (2), HAGR now consists of six databases and a wide range of tools and resources tackling different aspects of ageing. This article provides a non-technical description of the various databases, tools and projects in HAGR and their research applications. New resources created since the 2013 publication (3) are highlighted alongside updates to the remaining resources. In doing so we hope to provide a guide to HAGR so they can remain the most accessible and indepth resources available online in the field of biogerontology. HAGR is freely available online (with no registration required) at http://genomics.senescence.info/. DATABASE CONTENT GenAge--the ageing gene database The GenAge database (http://genomics.senescence.info/ genes/) is the benchmark database of genes related to ageing. Since its first publication in 2005 (2), GenAge has progressed considerably (Table 1). At first, GenAge only included human genes potentially associated with ageing. Now the database is divided into two main sections: human potential ageing-associated genes and longevity-associated genes in model organisms. When the first HAGR paper was published in 2005 (2), GenAge contained 220 entries for human genes. Presently, build 19 (24/06/2017) of GenAge contains 307 human gene entries and 2152 entries for model organisms. GenAge--human genes (http://genomics.senescence. info/genes/human.html) contains a selection of genes which might affect the human ageing process. The focus is on genes implicated in multiple processes and pathologies related to ageing, so those genes affecting only a single age-related disease are excluded. Each gene in the dataset is annotated to indicate how it has been linked to human ageing and why it has been selected for inclusion in the database. The strongest level of evidence is for those genes directly linked to human ageing, typically those resulting in progeroid syndromes when mutated. Since our previous publication in 2013 (3), in addition to new gene entries, older gene entries have been updated to reflect additional findings from new publications. Currently >2500 publications are cited. Also included is a list of 73 genes whose expression is commonly altered during mammalian ageing (4). Using data from GenAge--human genes we tracked patterns of ageing research over time. Research into specific genes in the context of ageing mostly began in the 1990s. Certain genes have become well-known through their role in ageing. Examples of these include WRN, the mutation of which results in Werner syndrome, possibly the most dramatic progeroid syndrome (5), LMNA, the mutation of which leads to Hutchinson-Gilford's progeroid syndrome (6), and SIRT1, linked to several processes involved in ageing (7) (Figure 1). For well-studied genes, like TP53, an additional role in the ageing process has emerged over time. Examples of these include MYC, an oncogene mainly studied in the context of cancer (8), MTOR, a regulator of several cellular processes which was found to play a role in ageing in various model organisms (9), and TP53, a well-known tumour suppressor (10) (Figure 1). GenAge--model organisms (http://genomics.senescence. info/genes/models.html) is a database of genes in model organisms which, if genetically modulated, result in significant changes in the ageing phenotype (e.g. progeroid syndromes in mice) and/or longevity (4). Most observations are from the four most popular biomedical model organisms: mice, worms, fruit flies and yeast (Table 1); however, several results from other model organisms such as zebrafish and golden hamsters are also included. For experiments using transgenic organisms, entries are classified according to the species in which the experiments were conducted in, not the species source of DNA. Where reported, the effects of the genetic manipulation on mean and/or maximal lifespan are included to provide quantitative data. As previously detailed (3), most genes are categorized as either pro-or anti-longevity depending on their effects on longevity; where studies report conflicting results, the genes are annotated as 'unclear'. The size of the model organisms' dataset has increased almost 3-fold since its conception in 2009, with over 400 genes added in the current update (Table 1), including miRNAs for the first time. GenAge has proven a valuable resource for ageing research, as evidence by many publications. A systems level analysis of the GenAge human genes database identified a robust group of ageing-specific network characteristics, revealing ageing genes as network hubs (11). Moreover, in an analysis of genes in the ageing human brain, 54 genes with sustained, consistent expression and 23 genes with DNA methylation changes were found in GenAge (12). GenAge was also used to validate the targets of a serum miRNA profile of human longevity (13). The data from GenAge has been incorporated into other databases, including AgeFactDB (http://agefactdb.jenage.de/) (14) and the Ne-tAge database (http://netage-project.org) (15). Therefore, although other databases with ageing-related genes exist, GenAge is the benchmark in the field. AnAge--the database of animal ageing and longevity Comparative biology is an essential and growing approach in the biology of ageing (16). AnAge (http://genomics. senescence.info/species/) is an integrative database of longevity records for over 4000 organisms. It includes, if available, maximum longevity, taxonomy, metabolic characteristics, development schedules and a multitude of additional life history data. AnAge is now over a decade old (2), becoming the most widely used resource in HAGR (see below). Although other datasets with longevity data exist (17), AnAge is arguably the 'gold standard' for longevity data in animals given its regular updating and quality data from manual curation. Build 14 (14 October 2017) contains 4244 entries, mostly individual species but also entries for higher taxa like primates and mammals. AnAge has previously been described in depth (3,18), and so its utility will only be briefly described. Entries contain maximum longevity and, where available, mortality parameters. Entries indicate whether the maximum longevity value comes from specimens kept in captivity or from the wild. Each entry includes a qualifier of confidence in the data and an estimate of sample size (3,18). Anecdotal evidence is not used to estimate maximum longevity but may be included in the observations. Factors that might introduce bias into comparative ageing studies, such as body size, metabolic rate, and development schedules, are also included where available (3,4,18). A list of species with negligible senescence is also provided. D1086 Nucleic Acids Research, 2018, Vol. 46, Database issue The primary goal of AnAge is as a data source for comparative and evolutionary biogerontological studies, thus enabling researchers to study what factors influence differences in phenotype and longevity across phylogeny. For example, one large-scale study using data from AnAge investigated how multiple ecological and mode-of-life traits affect lifespan (19). The data from AnAge have also been incorporated into the Comparative Cellular and Molecular Biology of Longevity Database (http://genomics.brocku.ca/ ccmbl/) (20), the MitoAge database (http://www.mitoage. info) (21), the Encyclopedia of Life (http://eol.org/), and the Animal Diversity Web (http://animaldiversity.ummz. umich.edu), demonstrating the versatility of this resource. Recent updates for AnAge have mostly been qualitative. The rate of new species added has reduced over time--we have added 40 new species since the last HAGR publication (3)--but older entries are kept up to date with new findings in the field. Our latest update, build 14, included ∼150 new references. As evidence of the substantial curation efforts in AnAge, the observations in AnAge now total >50 000 words. While the main focus of AnAge remains on data in animals, particularly chordates, the database contains entries for traditional biomedical models, including invertebrates and fungi. GenDR--a database of dietary restriction-related genes Dietary restriction (DR) delays the ageing process and extends lifespan in a multitude of species from yeast to mammals (22). However, the exact mechanisms of how DR extends lifespan are still unknown. As previously described (23), GenDR (http://genomics.senescence.info/diet/) is a database of DR-related genes. Herein, the use and function of GenDR will be briefly outlined along with updates since the 2013 HAGR paper (3). DR-essential genes are defined in GenDR as those which, if genetically modified, interfere with DR-mediated lifespan extension (3,23). GenDR has entries for nematodes, fruit flies, mice, budding yeast and fission yeast. We recently (24 June 2017) released a new build of GenDR, which contains 214 DR-essential genes, a 35% increase (56 new genes) since our previous update (3). GenDR also contains a complimentary dataset of 173 genes consistently differentially expressed in mammals under DR (24). GenDR is the first and, to our knowledge, only database of DR genes. We hope that GenDR may aid in the development of pharmacological DR mimetics. Indeed, GenDR was used to validate the gene targets of candidate DR mimetics in worms (25). In an analysis of the downstream targets of daf-16, a gene involved in DR in worms, four of the targets overlapped with the GenDR database, demonstrating the involvement of different components of the pathway in DR (26). LongevityMap--human genetic variants associated with longevity Variation in human lifespan has been found to be 20-30% heritable, with increasing heritability at advanced ages (27). As next-generation sequencing and genome-wide approaches advance, so does the capacity for performing longevity association studies. To catalog the increasing volume of data in genetic studies of human longevity, we created LongevityMap (http://genomics.senescence.info/ longevity/), a database of genes, gene variants and chromosomal locations associated with longevity (28). This differs from the GenAge database, which focuses mostly on data from model organisms and the few genes associated with human ageing (e.g. genes causing progeroid syndromes). Entries in LongevityMap were compiled from the literature (28). Negative results are included to provide information regarding each gene and variant previously studied in the context of human longevity. Both large and smallscale studies are included, along with several cross-sectional studies and studies of extreme human longevity (e.g., in centenarians). Due to the diversity of data, details about the study design are outlined for each entry, such as population and sample size (28). Build 3 (24 June 2017) of Longevi-tyMap contains 550 entries (a 9% increase in this latest update), 884 genes (18% increase) and 3144 variants (58% increase). Of the 550 entries, 275 are reported as significant findings. We hope that LongevityMap will act as a reference to help researchers parse the increasing quantities of data related to the genetics of human longevity. DrugAge--a database of ageing-related drugs Identifying drugs that could extend lifespan in model organisms has received considerable interest (29). Our new DrugAge database (http://genomics.senescence.info/drugs/) is a curated database of drugs, compounds and supplements with anti-ageing effects that extend longevity in model organisms. Although another database of candidate geroprotectors exists, called Geroprotectors.org (30), DrugAge provides a more comprehensive and systematic dataset of lifeextending drugs and compounds (31). DrugAge was developed to allow researchers to prioritize drugs and compounds relevant to ageing, providing highquality summary data in model organisms. As described (31), the data were primarily compiled from the literature, in addition to other databases and submissions from the scientific community. Build 2 (01 September 2016) of DrugAge contains 418 distinct compounds across 1316 lifespan assays on 27 unique model organisms. Hundreds of genes in several pathways act as regulators of ageing (1,32). However, analysis of DrugAge and other HAGR databases has revealed that the overlap between the targets of lifespan-extending drugs and known ageing related genes is modest (31). This indicates that most ageing-related pathways have yet to be targeted pharmacologically; DrugAge may aid in guiding further assays. This was recently demonstrated in one study where machine learning was used to predict whether a compound would increase lifespan in worms using data from Dru-gAge. The best model had 80% prediction accuracy and the top hit compounds could broadly be divided into compounds affecting mitochondria, inflammation, cancer, and gonadotropin-releasing hormone (33). CellAge--a database of cell senescence genes Cell senescence, also known as cellular senescence (CS), is the irreversible cessation of cell division of normally prolif-Nucleic Acids Research, 2018, Vol. 46, Database issue D1087 erating cells. Senescent cells accumulate as an organism ages and may be an important contributor to ageing and agerelated disease (34). However, the connection between organismal ageing and CS remains controversial (35). CellAge (http://genomics.senescence.info/cells/) is a new database of CS-associated genes, built to elucidate mechanisms of CS and its role in ageing. It is described here for the first time. To develop CellAge, a list of CS-associated genes was manually curated from the literature. Selection was based on gene manipulation experiments in human cells, which caused cells to induce or inhibit CS. The type of CS (replicative, stress-induced, or oncogene-induced), cell line, cell type and manipulation methods were standardized and recorded, facilitating the search and grouping of records of interest. The database includes data from primary cells in addition to immortalized cell lines and cancer cell lines. Each record contains observations about the evidence. Where reported, common markers of CS (36) such as growth arrest, increased SA-␤-galactosidase activity, SAheterochromatin foci, a decrease in BrdU incorporation, changes in morphology and specific gene expression signatures are described. A Human Cellular Senescence Gene Database (HCSGD) has been recently described by others (37), yet it combines information from many distinct sources and types of evidence, while CellAge has a more clear and rigorous selection procedure as well as manual curation. The first build of CellAge contains 279 entries, in which experiments in lung fibroblasts, embryonic kidney cells and foreskin fibroblasts are the most widely represented in the data. The majority of genes are associated with replicative senescence (232 genes), followed by stress-induced senescence (34 genes) and oncogene-induced senescence (28 genes). It is hoped that CellAge will aid in understanding the various types of CS and that analysis of the data will lead to the discovery of further CS-associated genes and their regulatory mechanisms. Analysis of the CellAge dataset is currently being carried out by our group and will be published in a future publication. TOOLS, PROJECTS AND OTHER INFORMATION RE-SOURCES Ageing-related disease genes In industrialized societies, ageing is the main risk factor for many debilitating and life-threatening diseases including cancer, cardiovascular disease, arthritis, diabetes and neurodegeneration. As lifespan increases so too does the prevalence of these diseases (38). An understanding of how these diseases are linked to the ageing process is needed to help tackle this growing problem (39). Our new ageingrelated disease genes tool (http://genomics.senescence.info/ diseases/), first described here, makes available a set of agerelated disease genes and permits their integration with ageing-related genes from our other databases. The genes were assembled using data compiled by a National Institute of Ageing study (40), as described (41). Diseases with fewer than 20 genes associated were excluded from the gene list to avoid the dilution of findings. Processes and conditions such as insulin resistance and hyperlipidaemia were classified as dysfunctions and excluded from the list. Users can browse genes and diseases by MeSH disease terms, MeSH disease class and by gene symbol. The disease classes are cardiovascular diseases, immune system diseases, musculoskeletal diseases, neoplasms, nervous system diseases, and nutritional and metabolic diseases. Results can be grouped by gene or disease. There are 769 genes associated with 20 age-related diseases in total. Our tool was designed so that age-related disease genes can be viewed, analyzed and downloaded in the context of ageing genes to understand potential functional overlap. The tool allows users to create a merged data set between age-related disease genes and ageing genes, according to user-defined filters. Where applicable, genes in HAGR databases can be converted into human homologs before merging. Cross-links and complementary resources All our databases are fully integrated, allowing users to gain a deeper understanding of the genes and pathways involved in ageing. In particular genetic databases have extensive crosslinks between them, linking each entry in a database to entries in other databases where available. DrugAge is also integrated with other HAGR databases using drug-gene interaction data from DGIdb (42). Moreover, for a greater understanding of ageing there are several additional resources to HAGR. Succinctly, the Digital Ageing Atlas (DAA, http://ageing-map.org/) is a database of human age-related changes at different biological levels (43). HAGR links to the DAA on its homepage and searches within HAGR also show results from the DAA where available. The Ageing Research Computational Tools (http://genomics.senescence.info/software/) are a toolkit of Perl modules aimed at parsing files, datamining, and searching and downloading data from the Internet (2,4). An SPSS script is also available, which can be used to determine the demographic rate of ageing for a given population (44). Senescence.info (http://www.senescence.info) is an informational repository on the science of ageing which aims to highlight the importance of ageing research and give an overview of current knowledge on the biology and genetics of ageing. Unlike HAGR, senescence.info is developed by a single person (J.P.M). Also in a informational and educational context, the WhosAge database (http://whoswho. senescence.info/) is a non-exhaustive list of individuals and companies working on ageing and longevity science, featuring 26 companies and 291 researchers. Lastly, since the last update on HAGR (3), two social media resources have been made available on Facebook (https: //www.facebook.com/pg/BiologyAgingNews) and Twitter (https://twitter.com/AgingBiology) which report on the latest news and findings in the field. The updates detail research in longevity, life-extension and rejuvenation technologies and link to articles/papers for further reading. These resources usually post several times a week to >6000 followers. DOWNLOADS AND AVAILABILITY Our access policy remains the same as in our previous publications (2,3). All HAGR databases and resources are freely available at http://genomics.senescence.info/. All databases allow users the opportunity to export, download and reuse data for their own analyses, under a Creative Commons Attribution licence. Of note, for data from model organisms in GenAge and GenDR, users can not only download genes from each model organism but also homologs from other model organisms for each dataset. Lists of human homologs for all the genes from model organisms are also available. Feedback from users and colleagues is welcome and encouraged via email. DISCUSSION Over the last decade, HAGR has expanded to include several new databases, datasets, tools and additional resources. Specifically, compared to our previous HAGR update (3), HAGR now includes the LongevityMap, DrugAge and Cel-lAge databases. The older databases--GenAge, AnAge and GenDR--have been updated and enhanced with significant information and data. Overall, the databases in HAGR organize large quantities of complex data, putting the findings into context and aiding further analysis. Having organized databases is necessary for employing computational approaches to ageing (45), including machine learning (46) and systems biology approaches (47). HAGR emphasizes high quality data on ageing and our databases are under continuous curation by experts in the field. AnAge provides information on data quality and sample size and prioritizes the reliability of the data over the most extreme values. GenAge--Model organisms, GenDR and CellAge all focus on genes from genetic manipulation experiments to ensure the selection process is as unbiased as possible. Nonetheless, some subjectivity is unavoidable and conflicting results can emerge. To cope with this, our policy is to be inclusive, providing evidence and links to the relevant literature and thus providing a balanced and comprehensive overview to the reader. HAGR has been cited over 500 times since it was first published in 2005 and has seen a continuous rise in the number of citations over recent years. From 2006 our resources received over 10 000 unique visitors per month, and they now receive over 30 000 unique visitors per month, thus indicating HAGR's growing importance in the field (Figure 2A). Out of all the HAGR databases, AnAge is the most popular ( Figure 2B). GenAge--Human genes and GenAge--Model Organisms have collectively also maintained high levels of use. Since DrugAge was released in 2016 its usage has greatly increased, becoming one of the most widely used databases. CellAge is the newest HAGR Nucleic Acids Research, 2018, Vol. 46, Database issue D1089 resource, released in late 2016, hence not surprisingly still one of the least popular. In conclusion, HAGR covers many aspects of ageing, acting as a science of ageing portal aimed at an audience from beginners to experts in biogerontology. Visitors are encouraged to send feedback and propose enhancements/features they would like to see in future. Over time as data continues to be generated, we anticipate that HAGR will continue to grow to meet this influx, maintaining its status as a leading online resource for studying the biology and genetics of ageing. Figure 1 . 1Patterns of ageing research for different human genes over time. The proportion of publications that focused mainly on ageing research are shown in red, the remaining publications are shown in blue. Figure 2 . 2(A) Combined HAGR and senescence.info unique visitors per month (black line). The secondary axis and grey bars show the yearly growth in HAGR citations *HAGR became publicly available online in mid-2004. **Data to end of August 2017. (B) Usage by percentage of the different HAGR databases in 2017. Table 1 . 1Growth over time of the various datasets in the GenAge databaseDatabase Species ACKNOWLEDGEMENTSWe would like to thank the many people who have helped in this project, providing valuable advice and suggestions, including former and current members of the Integrative Genomics of Ageing Group. In addition, we are grateful to the many users who have provided feedback on our resources and to the many scientists who have contributed data and expertise to our databases, specially Steve Austad for helping curate AnAge. The hallmarks of aging. C Lopez-Otin, M A Blasco, L Partridge, M Serrano, G Kroemer, Cell. 153Lopez-Otin,C., Blasco,M.A., Partridge,L., Serrano,M. and Kroemer,G. (2013) The hallmarks of aging. Cell, 153, 1194-1217. HAGR: the Human Ageing Genomic Resources. J P De Magalhaes, J Costa, O Toussaint, Nucleic Acids Res. 33de Magalhaes,J.P., Costa,J. and Toussaint,O. (2005) HAGR: the Human Ageing Genomic Resources. Nucleic Acids Res., 33, D537-D543. Human Ageing Genomic Resources: integrated databases and tools for the biology and genetics of ageing. R Tacutu, T Craig, A Budovsky, D Wuttke, G Lehmann, D Taranukha, J Costa, V E Fraifeld, De Magalhaes, Nucleic Acids Res. 41J.P.Tacutu,R., Craig,T., Budovsky,A., Wuttke,D., Lehmann,G., Taranukha,D., Costa,J., Fraifeld,V.E. and de Magalhaes,J.P. (2013) Human Ageing Genomic Resources: integrated databases and tools for the biology and genetics of ageing. Nucleic Acids Res., 41, D1027-D033. The Human Ageing Genomic Resources: online databases and tools for biogerontologists. J P De Magalhaes, A Budovsky, G Lehmann, J Costa, Y Li, V Fraifeld, G M Church, Aging Cell. 8de Magalhaes,J.P., Budovsky,A., Lehmann,G., Costa,J., Li,Y., Fraifeld,V. and Church,G.M. (2009) The Human Ageing Genomic Resources: online databases and tools for biogerontologists. Aging Cell, 8, 65-72. Hierarchical deterioration of body systems in Werner's syndrome: implications for normal ageing. M Goto, Mech. Ageing Dev. 98Goto,M. (1997) Hierarchical deterioration of body systems in Werner's syndrome: implications for normal ageing. Mech. Ageing Dev., 98, 239-254. Recurrent de novo point mutations in lamin a cause Hutchinson-Gilford progeria syndrome. M Eriksson, W T Brown, L B Gordon, M W Glynn, J Singer, L Scott, M R Erdos, C M Robbins, T Y Moses, P Berglund, Nature. 423Eriksson,M., Brown,W.T., Gordon,L.B., Glynn,M.W., Singer,J., Scott,L., Erdos,M.R., Robbins,C.M., Moses,T.Y., Berglund,P. et al. (2003) Recurrent de novo point mutations in lamin a cause Hutchinson-Gilford progeria syndrome. Nature, 423, 293-298. Calorie restriction, SIRT1 and metabolism: understanding longevity. L Bordone, L Guarente, Nat. Rev. Mol. Cell Biol. 6Bordone,L. and Guarente,L. (2005) Calorie restriction, SIRT1 and metabolism: understanding longevity. Nat. Rev. Mol. Cell Biol., 6, 298-305. Proteins of the Myc network: essential regulators of cell growth and differentiation. M Henriksson, B Luscher, Adv. Cancer Res. 68Henriksson,M. and Luscher,B. (1996) Proteins of the Myc network: essential regulators of cell growth and differentiation. Adv. Cancer Res., 68, 109-182. mTOR is a key modulator of ageing and age-related disease. S C Johnson, P S Rabinovitch, M Kaeberlein, Nature. 493Johnson,S.C., Rabinovitch,P.S. and Kaeberlein,M. (2013) mTOR is a key modulator of ageing and age-related disease. Nature, 493, 338-345. ) p53: puzzle and paradigm. L J Ko, C Prives, Genes Dev. 10Ko,L.J. and Prives,C. (1996) p53: puzzle and paradigm. Genes Dev., 10, 1054-1072. Systems-level analysis of human aging genes shed new light on mechanisms of aging. Q Zhang, R Nogales-Cadenas, J R Lin, W Zhang, Y Cai, J Vijg, Z D Zhang, Hum. Mol. Genet. 25Zhang,Q., Nogales-Cadenas,R., Lin,J.R., Zhang,W., Cai,Y., Vijg,J. and Zhang,Z.D. (2016) Systems-level analysis of human aging genes shed new light on mechanisms of aging. Hum. Mol. Genet., 25, 2934-2947. A systematic investigation into aging related genes in brain and their relationship with Alzheimer's disease. G Meng, X Zhong, H Mei, PLoS One. 11150624Meng,G., Zhong,X. and Mei,H. (2016) A systematic investigation into aging related genes in brain and their relationship with Alzheimer's disease. PLoS One, 11, e0150624. A serum miRNA profile of human longevity: findings from the Baltimore Longitudinal Study of Aging (BLSA). T Smith-Vikos, Z Liu, C Parsons, M Gorospe, L Ferrucci, T M Gill, F J Slack, Aging (Albany NY). 8Smith-Vikos,T., Liu,Z., Parsons,C., Gorospe,M., Ferrucci,L., Gill,T.M. and Slack,F.J. (2016) A serum miRNA profile of human longevity: findings from the Baltimore Longitudinal Study of Aging (BLSA). Aging (Albany NY), 8, 2971-2987. AgeFactDB-the JenAge Ageing Factor Database-towards data integration in ageing research. R Huhne, T Thalheim, J Suhnel, Nucleic Acids Res. 42Huhne,R., Thalheim,T. and Suhnel,J. (2014) AgeFactDB-the JenAge Ageing Factor Database-towards data integration in ageing research. Nucleic Acids Res., 42, D892-D896. The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. R Tacutu, A Budovsky, V E Fraifeld, Biogerontology. 11Tacutu,R., Budovsky,A. and Fraifeld,V.E. (2010) The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. Biogerontology, 11, 513-522. Cats, "rats," and bats: the comparative biology of aging in the 21st century. S N Austad, Integr. Comp. Biol. 50Austad,S.N. (2010) Cats, "rats," and bats: the comparative biology of aging in the 21st century. Integr. Comp. Biol., 50, 783-792. Cats, "Rats. J R Carey, D S Judge, and Bats: the Comparative Biology of Aging in the 21st Century. OdenseOdense University PressCarey,J.R. and Judge,D.S. (2000) Cats, "Rats," and Bats: the Comparative Biology of Aging in the 21st Century. Odense University Press, Odense. A database of vertebrate longevity records and their relation to other life-history traits. J P De Magalhaes, J Costa, J. Evol. Biol. 22de Magalhaes,J.P. and Costa,J. (2009) A database of vertebrate longevity records and their relation to other life-history traits. J. Evol. Biol., 22, 1770-1774. Ecology and mode-of-life explain lifespan variation in birds and mammals. K Healy, T Guillerme, S Finlay, A Kane, S B Kelly, D Mcclean, D J Kelly, I Donohue, A L Jackson, N Cooper, Proc. Biol. Scie. 281Healy,K., Guillerme,T., Finlay,S., Kane,A., Kelly,S.B., McClean,D., Kelly,D.J., Donohue,I., Jackson,A.L. and Cooper,N. (2014) Ecology and mode-of-life explain lifespan variation in birds and mammals. Proc. Biol. Scie., 281, 20140298. A comparative cellular and molecular biology of longevity database. J A Stuart, P Liang, X Luo, M M Page, E J Gallagher, C A Christoff, E L Robb, Age (Dordr). 35Stuart,J.A., Liang,P., Luo,X., Page,M.M., Gallagher,E.J., Christoff,C.A. and Robb,E.L. (2013) A comparative cellular and molecular biology of longevity database. Age (Dordr), 35, 1937-1947. MitoAge: a database for comparative analysis of mitochondrial DNA, with a special focus on animal longevity. D Toren, T Barzilay, R Tacutu, G Lehmann, K K Muradian, V E Fraifeld, Nucleic Acids Res. 44Toren,D., Barzilay,T., Tacutu,R., Lehmann,G., Muradian,K.K. and Fraifeld,V.E. (2016) MitoAge: a database for comparative analysis of mitochondrial DNA, with a special focus on animal longevity. Nucleic Acids Res., 44, D1262-D1265. Extending healthy life span-from yeast to humans. L Fontana, L Partridge, V D Longo, Science. 328Fontana,L., Partridge,L. and Longo,V.D. (2010) Extending healthy life span-from yeast to humans. Science, 328, 321-326. Dissecting the gene network of dietary restriction to identify evolutionarily conserved pathways and new functional genes. D Wuttke, R Connor, C Vora, T Craig, Y Li, S Wood, O Vasieva, R Shmookler Reis, F Tang, J P De Magalhaes, PLoS Genet. 81002834Wuttke,D., Connor,R., Vora,C., Craig,T., Li,Y., Wood,S., Vasieva,O., Shmookler Reis,R., Tang,F. and de Magalhaes,J.P. (2012) Dissecting the gene network of dietary restriction to identify evolutionarily conserved pathways and new functional genes. PLoS Genet., 8, e1002834. A meta-analysis of caloric restriction gene expression profiles to infer common signatures and regulatory mechanisms. M Plank, D Wuttke, S Van Dam, S A Clarke, J P De Magalhaes, Mol. Biosyst. 8Plank,M., Wuttke,D., van Dam,S., Clarke,S.A. and de Magalhaes,J.P. (2012) A meta-analysis of caloric restriction gene expression profiles to infer common signatures and regulatory mechanisms. Mol. Biosyst., 8, 1339-1349. A network pharmacology approach reveals new candidate caloric restriction mimetics in C. elegans. S Calvert, R Tacutu, S Sharifi, R Teixeira, P Ghosh, J P De Magalhaes, Aging Cell. 15Calvert,S., Tacutu,R., Sharifi,S., Teixeira,R., Ghosh,P. and de Magalhaes,J.P. (2016) A network pharmacology approach reveals new candidate caloric restriction mimetics in C. elegans. Aging Cell, 15, 256-266. Towards understanding the lifespan extension by reduced insulin signaling: bioinformatics analysis of DAF-16/FOXO direct targets in Caenorhabditis elegans. Y H Li, G G Zhang, Oncotarget. 7Li,Y.H. and Zhang,G.G. (2016) Towards understanding the lifespan extension by reduced insulin signaling: bioinformatics analysis of DAF-16/FOXO direct targets in Caenorhabditis elegans. Oncotarget, 7, 19185-19192. The quest for genetic determinants of human longevity: challenges and insights. K Christensen, T E Johnson, J W Vaupel, Nat. Rev. Genet. 7Christensen,K., Johnson,T.E. and Vaupel,J.W. (2006) The quest for genetic determinants of human longevity: challenges and insights. Nat. Rev. Genet., 7, 436-448. LongevityMap: a database of human genetic variants associated with longevity. A Budovsky, T Craig, J Wang, R Tacutu, A Csordas, J Lourenco, V E Fraifeld, De Magalhaes, Trends Genet. 29J.P.Budovsky,A., Craig,T., Wang,J., Tacutu,R., Csordas,A., Lourenco,J., Fraifeld,V.E. and de Magalhaes,J.P. (2013) LongevityMap: a database of human genetic variants associated with longevity. Trends Genet., 29, 559-560. The business of anti-aging science. J P De Magalhaes, M Stevens, D Thornton, Trends Biotechnol. 35de Magalhaes,J.P., Stevens,M. and Thornton,D. (2017) The business of anti-aging science. Trends Biotechnol., 35, 1062-1073. Geroprotectors.org: a new, structured and curated database of current therapeutic interventions in aging and age-related disease. A Moskalev, E Chernyagina, J P De Magalhaes, D Barardo, H Thoppil, M Shaposhnikov, A Budovsky, V E Fraifeld, A Garazha, V Tsvetkov, Aging (Albany NY). 7Moskalev,A., Chernyagina,E., de Magalhaes,J.P., Barardo,D., Thoppil,H., Shaposhnikov,M., Budovsky,A., Fraifeld,V.E., Garazha,A., Tsvetkov,V. et al. (2015) Geroprotectors.org: a new, structured and curated database of current therapeutic interventions in aging and age-related disease. Aging (Albany NY), 7, 616-628. The DrugAge database of aging-related drugs. D Barardo, D Thornton, H Thoppil, M Walsh, S Sharifi, S Ferreira, A Anzic, M Fernandes, P Monteiro, T Grum, Aging Cell. 16Barardo,D., Thornton,D., Thoppil,H., Walsh,M., Sharifi,S., Ferreira,S., Anzic,A., Fernandes,M., Monteiro,P., Grum,T. et al. (2017) The DrugAge database of aging-related drugs. Aging Cell, 16, 594-597. Genome-environment interactions that modulate aging: powerful targets for drug discovery. J P De Magalhaes, D Wuttke, S H Wood, M Plank, C Vora, Pharmacol. Rev. 64de Magalhaes,J.P., Wuttke,D., Wood,S.H., Plank,M. and Vora,C. (2012) Genome-environment interactions that modulate aging: powerful targets for drug discovery. Pharmacol. Rev., 64, 88-101. Machine learning for predicting lifespan-extending chemical compounds. D G Barardo, D Newby, D Thornton, T Ghafourian, J P De Magalhaes, A A Freitas, Aging (Albany NY). 9Barardo,D.G., Newby,D., Thornton,D., Ghafourian,T., de Magalhaes,J.P. and Freitas,A.A. (2017) Machine learning for predicting lifespan-extending chemical compounds. Aging (Albany NY), 9, 1721-1737. Cellular senescence: when bad things happen to good cells. J Campisi, F Di Fagagna, Nat. Rev. Mol. Cell Biol. 8Campisi,J. and d'Adda di Fagagna,F. (2007) Cellular senescence: when bad things happen to good cells. Nat. Rev. Mol. Cell Biol., 8, 729-740. Stress, cell senescence and organismal ageing. J P De Magalhaes, J F Passos, 10.1016/j.mad.2017.07.001Mech. Ageing Dev. de Magalhaes,J.P. and Passos,J.F. (2017) Stress, cell senescence and organismal ageing. Mech. Ageing Dev., doi:10.1016/j.mad.2017.07.001. Markers of cellular senescence. A Carnero, Methods Mol. Biol. 965Carnero,A. (2013) Markers of cellular senescence. Methods Mol. Biol., 965, 63-81. HCSGD: an integrated database of human cellular senescence genes. Q Dong, H Han, X Liu, L Wei, W Zhang, Z Zhao, M Q Zhang, X Wang, J. Genet. Genomics. 44Dong,Q., Han,H., Liu,X., Wei,L., Zhang,W., Zhao,Z., Zhang,M.Q. and Wang,X. (2017) HCSGD: an integrated database of human cellular senescence genes. J. Genet. Genomics, 44, 227-234. Ageing as a risk factor for disease. T Niccoli, L Partridge, Curr. Biol. 22Niccoli,T. and Partridge,L. (2012) Ageing as a risk factor for disease. Curr. Biol., 22, R741-R752. Geroscience: linking aging to chronic disease. B K Kennedy, S L Berger, A Brunet, J Campisi, A M Cuervo, E S Epel, C Franceschi, G J Lithgow, R I Morimoto, J E Pessin, Cell. 159Kennedy,B.K., Berger,S.L., Brunet,A., Campisi,J., Cuervo,A.M., Epel,E.S., Franceschi,C., Lithgow,G.J., Morimoto,R.I., Pessin,J.E. et al. (2014) Geroscience: linking aging to chronic disease. Cell, 159, 709-713. Systematic analysis, comparison, and integration of disease based human genetic association data and mouse genetic phenotypic information. Y Zhang, S De, J R Garner, K Smith, S A Wang, K G Becker, BMC Med. Genomics. 31Zhang,Y., De,S., Garner,J.R., Smith,K., Wang,S.A. and Becker,K.G. (2010) Systematic analysis, comparison, and integration of disease based human genetic association data and mouse genetic phenotypic information. BMC Med. Genomics, 3, 1. Systematic analysis of the gerontome reveals links between aging and age-related diseases. M Fernandes, C Wan, R Tacutu, D Barardo, A Rajput, J Wang, H Thoppil, D Thornton, C Yang, A Freitas, Hum. Mol. Genet. 25Fernandes,M., Wan,C., Tacutu,R., Barardo,D., Rajput,A., Wang,J., Thoppil,H., Thornton,D., Yang,C., Freitas,A. et al. (2016) Systematic analysis of the gerontome reveals links between aging and age-related diseases. Hum. Mol. Genet., 25, 4804-4818. DGIdb 2.0: mining clinically relevant drug-gene interactions. A H Wagner, A C Coffman, B J Ainscough, N C Spies, Z L Skidmore, K M Campbell, K Krysiak, D Pan, J F Mcmichael, J M Eldred, Nucleic Acids Res. 44Wagner,A.H., Coffman,A.C., Ainscough,B.J., Spies,N.C., Skidmore,Z.L., Campbell,K.M., Krysiak,K., Pan,D., McMichael,J.F., Eldred,J.M. et al. (2016) DGIdb 2.0: mining clinically relevant drug-gene interactions. Nucleic Acids Res., 44, D1036-D1044. The Digital Ageing Atlas: integrating the diversity of age-related changes into a unified resource. T Craig, C Smelick, R Tacutu, D Wuttke, S H Wood, H Stanley, G Janssens, E Savitskaya, A Moskalev, R Arking, Nucleic Acids Res. 43Craig,T., Smelick,C., Tacutu,R., Wuttke,D., Wood,S.H., Stanley,H., Janssens,G., Savitskaya,E., Moskalev,A., Arking,R. et al. (2015) The Digital Ageing Atlas: integrating the diversity of age-related changes into a unified resource. Nucleic Acids Res., 43, D873-D878. The influence of genes on the aging process of mice: a statistical assessment of the genetics of aging. J P De Magalhaes, J A Cabral, D Magalhaes, Genetics. 169de Magalhaes,J.P., Cabral,J.A. and Magalhaes,D. (2005) The influence of genes on the aging process of mice: a statistical assessment of the genetics of aging. Genetics, 169, 265-274. Computational biology for ageing. D Wieser, I Papatheodorou, M Ziehm, J M Thornton, Philos. Trans. R. Soc. Lond. B. Biol. Sci. 366Wieser,D., Papatheodorou,I., Ziehm,M. and Thornton,J.M. (2011) Computational biology for ageing. Philos. Trans. R. Soc. Lond. B. Biol. Sci., 366, 51-63. A review of supervised machine learning applied to ageing research. F Fabris, J P Magalhaes, A A Freitas, Biogerontology. 18Fabris,F., Magalhaes,J.P. and Freitas,A.A. (2017) A review of supervised machine learning applied to ageing research. Biogerontology, 18, 171-188. Computational systems biology of aging. A Kriete, M Lechner, D Clearfield, D Bohmann, Wiley Interdiscip. Rev. Syst. Biol. Med. 3Kriete,A., Lechner,M., Clearfield,D. and Bohmann,D. (2011) Computational systems biology of aging. Wiley Interdiscip. Rev. Syst. Biol. Med., 3, 414-428.

2023-02-22T16:52:05Z

CCBY

https://www.nature.com/articles/s41598-021-98674-6.pdf

GOLD

162da7fa23c3a96dfd25c59a41eeb72d79adf8b7

10.1038/s41598-021-98674-6

34588506

8481473

Systems biology analysis of lung fibrosis-related genes in the bleomycin mouse model 0123456789. 2021 Dmitri Toren Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania Hagai Yanai Epigenetics and Stem Cell Unit Translational Gerontology Branch National Institute on Aging, NIH 21224BaltimoreMDUSA Reem Abu Taha Gabriela Bunu Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania Eugen Ursu Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania Rolf Ziesche Internal Medicine II/Pulmonology Medical University of Vienna 27271WienAustria Robi Tacutu Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy 060031BucharestRomania E Vadim Fraifeld The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 8410501Beer-ShevaIsrael Systems biology analysis of lung fibrosis-related genes in the bleomycin mouse model Scientific Reports | 11192690123456789. 202110.1038/s41598-021-98674-61 www.nature.com/scientificreports Tissue fibrosis is a major driver of pathology in aging and is involved in numerous age-related diseases. The lungs are particularly susceptible to fibrotic pathology which is currently difficult to treat. The mouse bleomycin-induced fibrosis model was developed to investigate lung fibrosis and widely used over the years. However, a systematic analysis of the accumulated results has not been performed. We undertook a comprehensive data mining and subsequent manual curation, resulting in a collection of 213 genes (available at the TiRe database, www. tiredb. org), which when manipulated had a clear impact on bleomycin-induced lung fibrosis. Our meta-analysis highlights the age component in pulmonary fibrosis and strong links of related genes with longevity. The results support the validity of the bleomycin model to human pathology and suggest the importance of a multi-target therapeutic strategy for pulmonary fibrosis treatment.Tissue fibrosis is a major cause of frailty in aging and is involved in numerous age-related pathologies 1,2 . Among adult tissues, the lungs seem to be especially susceptible to age-related fibrotic pathology which is often poorly mendable 3-5 . As such, there is a great need for treatment and drug development to cope with this problem 6 . Unfortunately, experimental models of lung fibrosis are still few 7 , and it is still debatable to what extent these models adequately reflect pulmonary fibrosis in humans, in particular, idiopathic pulmonary fibrosis (IPF) 3,8,9 . The most popular experimental model, due to its ease of use, has been the mouse bleomycin-induced model 7 , in which a chemotherapeutic agent elicits a quick and robust fibrotic effect when inhaled 10 . While bleomycin does not model the disease perfectly 11,12 , it is still extremely useful for research and a widely-used model which has undoubtedly increased our understanding of fibrotic pathology 13,14 . Subsequently, a large body of data on the genetic factors that influence lung fibrosis has been accumulated based on the bleomycin-induced model. Yet, a systematic analysis of these data has not been performed to date. With this in mind, we conducted a meta-analysis of highly curated genes that have been rigorously shown to impact lung fibrosis in the bleomycin mouse model. The list containing these pulmonary fibrosis-related genes (PFRGs) is now part of the TiRe database (http:// www. tiredb. org), which contains curated genetic information on wound healing and fibroproliferative processes 15 .Apart from the collection and detailed characterization of PFRGs, we placed a special emphasis on (i) the consistency between different types of manipulations; (ii) the consistency of bleomycin data with the expression of corresponding genes in lungs of IPF patients, and (iii) the relationships between lung fibrosis-related and longevity-associated genes (LAGs). www.nature.com/scientificreports/ two or more manipulations. As seen in Table 2, over 85% of the different types of manipulations, applied either in the same study or in independent studies, are fully consistent with one another with regard to their effects on bleomycin-induced lung fibrosis. The rest were either partially consistent or consistency was unclear based on available data. Along with the bleomycin model to test interventions, several other murine models for inducing lung fibrosis have also been used (Table 3). In all cases, when a "non-bleomycin" murine model was used for investigating the role of a given gene or protein in lung fibrosis, the effects observed were consistent with those seen in the bleomycin model (Supplementary Table ST1). Overall, the curated list of PFRGs consists of 216 unique (non-redundant) genes that were examined regarding their role in bleomycin-induced lung fibrosis in mice (Supplementary Table ST1). Of them, only 3 genes Table 1. Gene-specific manipulation types in bleomycin-induced lung fibrosis studies. For a full detailed list see Supplementary Table ST1. Type of manipulation Number of studies Knockout 185 Overexpression 77 Overexpression and knockout 20 Protein downregulation (inhibitors, Abs) 20 Protein upregulation (agonists, external) 8 MiR knockout/knockdown 7 MiR overexpression 1 Genetic and non-genetic manipulation 22 Genetic and non-genetic (protein) 18 Genetic and epigenetic (microRNA) 3 Protein and epigenetic (microRNA) 1 Combined studies 44 Table 3. Non-bleomycin models of lung fibrosis in mice. 21 of 23 non-bleomycin models were used in addition to the bleomycin model in the same studies; in the case of Lox1 and Timp1 genes, only non-bleomycin models were used. www.nature.com/scientificreports/ (Ptgs1, Rps6kb1, Rps6kb2) did not show a definite impact. The PFRGs that displayed an effect could be further divided into two major groups: anti-fibrotic and pro-fibrotic. We considered a gene as anti-fibrotic if its upregulation reduced fibrosis and/or its downregulation had an opposite effect. Conversely, if upregulation promoted lung fibrosis and/or downregulation reduced fibrosis, the gene was considered pro-fibrotic. In our dataset, the number of anti-fibrotic genes was approximately equal to that of pro-fibrotic genes: 43.5% of genes displayed a clear anti-fibrotic activity, 50% of genes displayed pro-fibrotic activity, and 6.5% of genes showed inconsistent results (both pro-and anti-fibrotic). Remarkably, we noted a high consistency (~ 80%) between the effects of genetic manipulations of PFRGs in bleomycin-induced lung fibrosis in mice and the expression of their human orthologs in the lungs of patients with IPF (Table 2). Additional models of lung fibrosis Number of studies Evolutionary conservation of pulmonary fibrosis-related genes. PFRGs were studied in mice but are most interesting for their potential impact on humans. Consequently, we also evaluated their evolutionary conservation in more depth. For this purpose, we extracted the PFRG orthologs for all species available in the InParanoid database 16 ; http:// inpar anoid. sbc. su. se/). Some of the PFRGs (n = 18) were not represented in InParanoid and the analysis was performed on 195 genes only. As seen in Fig. 1, PFRGs are differentially conserved among vertebrates and invertebrates: they are over-represented in vertebrates and under-represented in invertebrates. Enrichment analysis. Next, we looked to see whether the PFRG list is enriched for certain processes and pathways that dominate in pulmonary fibrosis. We found that PFRGs are enriched for processes such as regulation of proliferation, inflammation and immune functions, and aging processes (Table 4). Interestingly, there are marked enrichment differences between pro-fibrotic and anti-fibrotic genes: (i) While pro-fibrotic genes are enriched for genes that positively regulate proliferation, anti-fibrotic genes are composed of those that negatively regulate it; (ii) pro-fibrotic genes are also enriched for pathways that relate to inflammation and immune function while anti-fibrotic genes are not. An important question is whether there is any association between pro-fibrotic or anti-fibrotic genes and specific cell types. Since the vast majority of data on gene expression in pulmonary fibrosis were obtained for the whole lung tissue, a direct answer to this question is at the moment impossible. Nevertheless, we took advantage of the Enrichr tools to estimate what tissues and cell types are more likely to be associated, if any, with PFRGs. The Human Gene Atlas database 17 (http:// biogps. org) analysis revealed that our set of anti-fibrotic genes is more likely to be expressed in lungs, while the pro-fibrotic genes are more likely to be expressed in vessels, lungs, and blood. We further used the PanglaoDB database 18 (http:// pangl aodb. se) for determining cell types associated with PFRGs. As seen in Fig. 2, the pro-fibrotic genes are mostly associated with immune cells. Apart from immune cells, the anti-fibrotic genes are mostly associated with connective tissue and endothelial cells. Functional module analysis. To further understand the characteristics of the PFRG network, we conducted a functional module analysis using the HumanBase online tool 19 (https:// hb. flati ronin stitu te. org). This tool allows for identifying functional modules in a gene-set in a manner specific to a given tissue (in this case, lungs). The analysis revealed six functional modules formed by PFRGs and their immediate partners ( Fig. 3a and Supplementary Table ST5). These include: (i) a cluster encompassing cell signaling, cell migration, proliferation and programmed cell death (MA1, 643 GO terms; p < 0.05); (ii) a cluster of immuno-inflammatory responses (MA2, 177 GO terms; p < 0.05); (iii) a cluster pertaining to nucleotide-related biosynthesis and metabolism Figure 1. Evolutionary conservation of pulmonary fibrosis-related genes (PFRGs). The graph summarizes the percentages of orthologs between humans and given species. Each dot corresponds to a single species from InParanoid (in descending order by the percentage of orthology). A total of 268 species from all kingdoms of life are presented. Orange triangles stand for the percentage of orthologs for PFRGs (n = 195); green circles stand for the percentage for the entire human genome (n = 20,297). The difference in ortholog percentage between the entire genome and PFRGs is significant for the vast majority of species: Chi-square (χ 2 ) goodness of fit, p < 0.05. For anti-fibrotic genes, there are two functional modules that include at least ten genes each. The first cluster (MB1, 65 GO terms; p < 0.05) includes mostly nucleotide metabolism, and the second module (MB2, 137 GO terms; p < 0.05) encompasses response to hypoxia and MAPK signaling (Fig. 3b). The list of pro-fibrotic genes includes three functional modules: the first one, (MC1, 356 GO terms; p < 0.05) covers regulation of cytokine production, regulation of cell migration and inflammatory response; the second module (MC2, 92 GO terms; p < 0.05) is in relation to actomyosin activity and the third module (MC3, 163 GO terms; p < 0.05) includes cytokine regulation. Figure 2. Cell type enrichment analysis of the PFRGs. Pro-and anti-fibrotic genes were tested for expression enrichment in specific cell types. Analysis was performed against the PanglaoDB database on the Enrichr platform. Presented are the main enriched cell types for both lists and their combined enrichment score (see methods) for pro-fibrotic (red) and anti-fibrotic (green). *Non-significant enrichment in anti-fibrotic genes (p > 0.05) and **Non-significant enrichment in pro-fibrotic genes (p > 0.05). www.nature.com/scientificreports/ www.nature.com/scientificreports/ Network analysis. To further understand the features of PFRGs and how they relate to each other, we performed a protein-protein interaction (PPI) network analysis on the protein products of PFRGs. As the human proteome is more comprehensively mapped 20 , the analysis was done on the human orthologs of the mouse PFRGs. The analysis shows that PFRGs are highly connected in the interactome (Fig. 4, panel a1) and strongly interact among themselves, forming a large, directly connected component of 107 nodes (56.3% of all PFRGs). This interconnectivity, i.e. the fraction of genes/nodes forming the largest connected subnetwork from a certain gene set, is higher for PFRGs than expected by chance ( Fig. 4, panel a2). The statistical significance of this observation was validated by comparing it with the results from random gene set samples of equal size (Fig. 4, panel a3). Considering this, it is not surprising that several genes in the network are also important hubs, i.e., display very high connectivity with other genes in the network. For example, the topmost 5% network hubs (with 13-21 PFRG interactions and > 150 interactome PPIs), depicted in bold in Fig. 4, include AKT1 (a regulator of mTOR signaling), CAV1, SIRT1 (epigenetic regulator), HSPA5 (heat shock protein), SMAD3 (plays a role in TGFβ signaling), and HIF1A (response to oxygen). Of note, the CAV1 gene was found to be a large hub in both the functional anti-fibrotic modules (Fig. 3b, MB) and in the PFRG PPI network (Fig. 4a, b). Additionally, the two network hubs HIF1A, which is linked to aging and response to hypoxia 21 , and SMAD3, known as an important player in wound healing and an anti-longevity gene 15,21 , were also found in the pro-fibrotic module (Fig. 3c, MC1). Another hub of the PFRG network, the AKT1 protein ( Fig. 4a) is well-known for its anti-apoptotic activity and has been linked to cellular senescence 22 . Scientific Links between pulmonary fibrosis genes and longevity. Interestingly, many of the enriched categories in the functional modules analysis (Fig. 3., e.g. MA2) are also relevant to aging, while at the same time, many genes from MA3 ( Fig. 3) are in fact longevity-associated genes (LAGs). This prompted the following question: Do genes that influence lung fibrosis have any impact on longevity in mice? To get insight into this issue, we compared the list of PFRGs with the list of LAGs from GenAge, which were reported to affect the lifespan of mice 21,23 . The comparison yielded 18 genetic mouse models of extended lifespan (longevity phenotype) or reduced lifespan (premature aging phenotype), which were also tested for their role in bleomycin-induced pulmonary fibrosis. The results summarized in Fig. 5a and Table 5, clearly show that pro-longevity genetic manipulations also reduce pulmonary fibrosis, while anti-longevity genetic manipulations have the opposite effect. That is, pro-longevity genes tend to be associated with anti-fibrosis (11 out of 12 pro-LAGs are anti-fibrotic), while anti-longevity genes with pro-fibrosis (5 out of 6 anti-LAGs are clearly pro-fibrotic with an additional anti-LAG showing both anti-and pro-fibrotic effects; Fisher's exact test, p = 0.001). The analysis above suggests that the anti-and pro-fibrotic genes have evolved in tight relation to the anti-and pro-longevity genes, with many of these genes even having direct roles in both lifespan determination and fibrosis development. To further test this notion, we decided to investigate to what extent the gene expression levels of PFRGs correlate with maximum lifespan (MLS). For this we used the tools developed in our lab and collected data, comprising 28 samples from 14 mammalian species, analyzed in our recent cross-species lifespan study 24 . The analysis showed that the expression of many PFRGs (n = 34) correlates with MLS in mammals ( Fig. 5b; Supplementary Table ST6 and ST7). The number of these MLS-associated genes is 2.34 times more than expected by chance (Fisher's exact test, p = 6.4E−05), with R 2 being even greater than 0.6 for some of the correlations, overall suggesting that many fibrosis-related genes might have had a role in the determination of MLS as well. Discussion Although many animal models have thus far been established for investigating IPF 7 , the bleomycin model, despite its limitations and disadvantages, is the most widely used and generally viewed as the standard in modeling pulmonary fibrosis 25 . In this study, we collected a list of over 200 genes (PFRGs) influencing the course and/or outcome of bleomycin-induced pulmonary fibrosis, and performed a comprehensive analysis on their role in the bleomycin mouse model. The PFRGs are currently available in the TiRe database, which contains curated (a) Protein-protein interaction network of PFRGs. The network was constructed using the BioGRID database. Depicted is the largest continuous component of the pulmonary fibrosis network (107 nodes). Green nodes represent anti-fibrotic genes (N = 41), red are pro-fibrotic genes (N = 55), and blue nodes are genes with an unclear (both pro-and anti-) effect (N = 11). Pulmonary fibrosis hubs, the topmost 5% connected genes in the network, are highlighted with bold text and a black border. (a1) PFRGs (blue bar) are significantly more connected than random interactome genes (green bar), with a higher average number of protein-protein interactions (64.4 for PFRGs vs. 45.5 for all genes) and a GSEA-based enrichment score of 0.6 for degree connectivity (p < 0.05). (a2) The observed interconnectivity of PFRGs in the interactome, depicted by the red dot in the scatter plot, can be compared to the percentage of interconnected nodes (on the Y-axis), found in the largest continuous component, for randomly sampled node sets. The plot shows the sampling of subsets of random interactome nodes, of various sizes (represented in a log10 scale on the X-axis, from 50 to 17,600 nodes), for which the interconnectivity was computed 100 times. (a3) The interconnectivity of the PFRG network (blue line), compared to the histogram of frequencies of interconnectivity, per one thousand random samples of the same geneset size (Y-axis). PFRG interconnectivity (56.31%) is significantly larger than expected, with a Z-score of 5.37 (distribution average 13.9%, SD: 7.9%). (b) Anti-fibrotic genes in the PFRG network form a continuous subnetwork (26 out of 107, 24.3%). (c) Pro-fibrotic genes in the PFRG network form a continuous subnetwork (42 out of 107, 39.3%). The networks were generated by Cytoscape 3.8.0. (https:// www. cytos cape. org). The panels a1-a3 were generated using a custom R script developed in-house. www.nature.com/scientificreports/ genetic information on wound healing and fibroproliferative processes 15 (hosted at http:// www. tiredb. org). The results of the analysis pointed out several important findings: (i) A high consistency between the different types of genetic and non-genetic manipulations in their effects on bleomycin-induced lung fibrosis. When the same manipulation was used in different studies or different manipulations were used in the same study, the consistency between results supports their reliability. In interpreting and evaluating these results, it should be considered that the list of PFRGs relies mostly on studies that employ loss-of-function interventions (Table 1) and thus might cover only a part of the molecular mechanisms involved in fibrosis. Still, even if the collected PFRGs do not encompass the whole picture of fibrosis, they seem to provide a coherent image as they form a highly interconnected PPI network. Another important point is that PFRGs might be involved in a more general response, i.e. tissue repair after injury. In particular, we compared PFRGs with skin wound healing-related genes 15 and found a significant overlap of over 20% (p < 0.001). This observation suggests that fibrosis and wound healing have much in common and that PFRGs are not exclusively associated with lung fibrosis but rather many of them are involved in a more general response. Yet, it seems the relationship between these processes is more complex than a simple "accelerated/slower wound healing-reduced/promoted fibrosis". These relationships are an excellent point for future investigations. (ii) PFRGs are overall enriched for regulation of cellular proliferation, inflammation and immune functions, and aging-related processes, with a prominent difference between anti-and pro-fibrotic genes. That is, when pro-and anti-fibrotic genes were analyzed separately, they displayed definite enrichment patterns that were distinct from one another. We found that the pro-fibrotic genes are dominated by positive regulation of cellular proliferation, inflammatory processes and immune responses, including related processes and pathways such as Cytokine signaling, Jak-STAT signaling pathway, TNF signaling pathway, or Reaction to pathogen. These findings are not unexpected when considering that fibrosis is a wound healing response gone awry, which would most likely be the case for fibrosis of any tissue. The above indirectly highlights the role of immunity and inflammatory responses in the induction and development of pulmonary fibrosis 9 , a conclusion that is also supported by the results of our functional module analysis. Remarkably, pro-fibrotic genes are specifically enriched in processes and pathways closely linked to aging, such as the insulin-FoxO signaling pathway 26 , PI3K-Akt signaling pathway 22 , etc. Not surprisingly, PFRGs and particularly profibrotic genes are enriched for the Oxygen homeostasis and Stress response categories. This is in line with our previous finding of high resistance to oxidative-stress-induced cytotoxicity in lung fibroblasts from IPF patients 27 . Huang et al. 28 have also shown that fibroblasts from the lungs of bleomycin-treated old mice, displayed a stronger fibrotic response and were more resistant to H 2 O 2 -induced apoptosis than those from the young. In the long run, this may result in the accumulation of damaged/senescent cells that would otherwise be eliminated, as was actually observed in vivo by Hecker et al. 29 . In that regard, we were surprised to find that cellular senescence per se was not one of the terms found in the enrichment analysis. However, many of the pathways that we found can potentially converge to it (e.g. regulation of proliferation, cancer, inflammation, response to stress, etc.), thus supporting the idea that cellular senescence may indeed play an important role in lung fibrosis 6,27,30,31 . The results of enrichment analysis are further strengthened by the functional module network analysis in which at least two modules are enriched for processes related to the response to oxidative stress and hypoxia (see Fig. 3 and Supplementary Table ST5). This could be specifically relevant to lung vs other tissues, because of the high oxygen environment 32,33 . In contrast to pro-fibrotic genes, anti-fibrotic genes were found to be enriched for negative regulation of cell proliferation, MAPK signaling pathway and response to mechanical stimulus. Of note, the lungs experience ongoing mechanical stress and areas with a higher pressure are more prone to fibrotic changes 34 . With this in mind, our previous finding of actin-organization aberrations in IPF fibroblasts 27 may highlight the critical role of the response to mechanical stress in lung integrity and functionality. Interestingly, PFRGs are most likely to be expressed in immune cells and connective tissues cells, pro-fibrotic genes being mainly associated with immune cells, whereas anti-fibrotic genes with connective tissues, pulmonary vascular smooth muscle and endothelial cells (Fig. 2). (iii) The high consistency between the expression of PFRGs in the mouse model of bleomycin-induced lung fibrosis and the expression of their human orthologs in the lungs of IPF patients indicates that, despite its disadvantages 11,12 , the bleomycin model is still highly relevant for the study of human lung fibrosis 10,25,35 . In particular, the human orthologs of murine PFRGs could be the targets for therapeutic interventions. To some extent, it is also supported by our findings that PFRGs are highly conserved among vertebrates but much less in invertebrate species. This implies that many PFRGs are a relatively recent acquisition in the course of evolution and that the genetic basis of pulmonary fibrosis may have a common platform among vertebrates. Yet, the targets for manipulation might have been, to some extent, selected a priori based on indications for their potential involvement in IPF. If so, it may cause a bias for the PFRG list. (iv) Our network-based analysis clearly showed that PFRGs are highly interconnected and hence interacting, thus significantly reducing the odds to treat pulmonary fibrosis by targeting a single gene 36 . In other words, it means that a multi-target therapy approach would definitely be preferable. (v) Comparing the mouse PFRGs and longevity-associated genes (LAGs) brought another remarkable finding: pro-longevity genes are dominated by anti-fibrotic genes, whereas the anti-longevity genes are dominated by pro-fibrotic genes. Congruent with this finding is the observation that the functional modules for antifibrotic genes contain pro-LAGs but no anti-LAGs (Fig. 3b). Conversely, the modules for pro-fibrotic genes www.nature.com/scientificreports/ include mostly anti-LAGs (Fig. 3c). That is, the anti-fibrotic genes are associated with lifespan extension while the pro-fibrotic ones are associated with premature aging. Both findings support the notion that pulmonary fibrosis is a disease of aging. It would be worth mentioning that many age-related processes are linked to signaling pathways crucially involved in organ regeneration/repair (though on various levels of the metabolic hierarchy) 37 . They concern structural and functional differentiation as well as de-differentiation of organ tissue, chronic inflammation, and energy supply (or more precisely, its depletion). However, at this point, we should still keep in mind that our analysis predominantly represents pulmonary fibrosis induced by bleomycin and does not necessarily reflect (in both quality and quantity) the entirety of the most frequent and threatening form of age-related human lung fibrosis, IPF. In addition, it is difficult to assess the "chicken and egg" dilemma at this point. That is, we cannot definitely state whether manipulation of these genes affects fibrosis primarily, which in turn drives tissue aging, or the more likely case, that these genes are integral to important processes that affect both aging and fibrosis in parallel. It should be noted that the vast majority of the bleomycin studies were done using young mice, whereas the logic of lung fibrosis, and in particular IPF, requires including aged animals in the study. Thus far, these studies are sporadic. As an example, young WT and Zmpste24-deficient progeroid mice developed a similar fibrotic response to BLM. In contrast, old WT mice but not old Zmpste24-deficient mice developed severe lung fibrosis 38 . Unexpected protection of Zmpste24-null old lungs against BLM was apparently attributed to the upregulation of several extracellular matrix-related miRNAs (miR23a, 27a, 29a, 145a), thus resulting in downregulation of targeted profibrotic pathways of TGF-β/SMAD3/NF-κB and Wnt3a/β-catenin signaling axes. Of note, similar age-dependent responses were observed when the rate of skin wound healing was investigated in longevity/ premature-aging phenotypes 15,39,40 . Nevertheless, the mouse BML model, even when used in young animals alone, appears to be a valuable tool for investigating lung fibrosis. Finally, we would like to stress again that, based on our analysis, a multi-target therapy of pulmonary fibrosis should become the major strategy. Furthermore, although the collected list of PFRGs is quite extensive, the application of novel techniques, such as CRISPR and modified RNA, would be an important point for future investigations. It is also worthwhile to extend the experimental studies by including aged animals. Methods Data sources. The list of the PFRGs was compiled from peer-reviewed literature and the extracted data were manually curated by the authors. The data were organized in a tabular format (and is available as an Excel table in Supplementary Table ST1, ST2 and ST3), and the curation process focused on the extraction of the following characteristics: targeted gene/protein, Ensembl ID, mouse strain, manipulation type, gender, age, dose/route of administration, regimen, main effects, pro-or anti-fibrotic effect, other effects, and relevant references. In order for a paper to be considered for the analysis, each article had to meet the following criteria: (1) To use the mouse model of BML-induced lung fibrosis, with sufficient fibrosis markers and follow-up description; (2) To contain the data on genetic or protein manipulations resulted in a significant promotion or suppression of BML-induced lung fibrosis. In addition, for comparative analysis between BLM-induced lung fibrosis and IPF, only the IPF papers with the expression of the gene of interest (i.e. genetically manipulated in BLM-induced lung fibrosis) were included. Evolutionary conservation. The evolutionary conservation analysis was performed using Python scripts developed in our lab, which automatically extract and analyze data from the InParanoid database 16 , version 8 (http:// inpar anoid. sbc. su. se/ cgi-bin/ index. cgi). For each mouse gene, the presence or absence of orthologs across 268 proteomes (all species available in InParanoid) was assessed and the overall evolutionary conservation was defined as the percentage of species in which at least one ortholog exists. The evaluation of orthology in InParanoid was performed with a threshold for the inparalog score of 1.0 (highest stringency). All comparisons were statistically significant unless otherwise mentioned (Chi-Square χ 2 test; p < 0.05). For the network analysis, the human orthologs for mouse PFRGs were computed using a similar method (InParanoid 8 inparalogs with scores of 1.0). Enrichment analysis. Enrichment analysis of PFRGs and related pathways was performed using the DAVID Bioinformatics Resources tool 41 , version 6.8, https:// david. ncifc rf. gov. As the data on human genes and proteins is the most abundant, the human orthologs of PFRGs defined in mice were used for the analysis. Statistical significance of enrichment was evaluated using the default parameters set in DAVID. Cell type enrichment analysis of the PFRGs was performed against the PanglaoDB database 18 (https:// pangl aodb. se) using the Enrichr platform 42 (https:// maaya nlab. cloud/ Enric hr/). The presented combined score is the multiplication of the natural logarithm of p-value (Fisher exact test) and the z-score of the deviation from the expected rank (for more details, please see: https:// maaya nlab. cloud/ Enric hr/). To determine the most likely tissue, we used the Human Gene Atlas database 17 (http:// biogps. org/). Protein-protein interaction network. Protein-protein interaction (PPI) data were taken from the BioGRID database 20 , http:// thebi ogrid. org, human interactome, Build 3.5.188. The PPI network construction and analysis were performed using Cytoscape 43 , http:// www. cytos cape. org, version 3.8.0. Prior to any network analyses, genetic interactions, self-loops, duplicate edges and interactions with proteins from other species were removed from the interactome, and the remaining network was used as a control. The interconnectivity was computed as the fraction of nodes in the largest connected component out of the input gene set, by using the www.nature.com/scientificreports/ breadth-first search algorithm. Modeling the relationship between node subset size and interconnectivity in the human interactome (Fig. 4, panel a2) was carried out by randomly sampling subsets of nodes in the interactome, with a sample size varying from 50 to 17,600 nodes (step of 50). In this case, sampling was performed 100 times for each subset size. To evaluate the statistical significance of the observed network interconnectivity, random sampling of 190 nodes from the BioGRID network was performed 1000 times. The enrichment score for the degree of PFRGs in the PPI network was computed using the GSEA method 44 . Functional module analysis. The construction of a network with functional modules for FPRGs was carried out using the HumanBase tool 45 , https:// hb. flati ronin stitu te. org, with a minimum module size set to 10 genes. Briefly, HumanBase provides the possibility to identify, at the tissue level, functional modules containing genes and their interaction partners which specifically work together, by grouping them into clusters of relevant biological processes. HumanBase detects modules of genes from tissue-specific functional association gene networks built by integrating vast omics datasets and associates terms (e.g. processes, pathways) to the detected modules based on overrepresentation. Linear models linking longevity and gene expression. The linear longevity models for the PFRGs dataset included 14 species (Homo sapiens not included) with reported maximum lifespan and a total of 28 lung transcriptome samples (Supplementary Table ST6). 8205 genes were selected based on the orthology relationships and lung expression. Raw gene expression cross-species data was extracted from public archives and reanalyzed with an internal pipeline described elsewhere 24 . The orthology relationships were obtained from the 99th release of Ensembl Compara Database 46 ; https:// doi. org/ 10. 1093/ datab ase/ bav096. 90 of the PFRGs were considered for analysis based on orthology relationships among the 15 species, as described in Kulaga et al. 24 . Maximum lifespan data were extracted from the AnAge database 21 ; https:// genom ics. senes cence. info/ speci es. The analysis was performed using python scripts developed in our lab using several packages including Pandas, Seaborn, and Statsmodels. The analysis includes species with good quality assemblies and annotations and genes with orthologs in the selected species. For the evaluation of statistical significance (p < 0.05), the adjusted p-values with Benjamini-Hochberg correction were used. The models were defined and fitted using the "statsmodels" Python module 47 . Received: 28 April 2021; Accepted: 13 September 2021 Figure 3 . 3Functional module network of the pulmonary fibrosis-related genes (PFRGs) in lung tissue. The network was built for the lung tissue, using HumanBase, with the human orthologs of PFRGs as input. The interaction network is built using the closest gene neighbors and then clustered based on enrichment in GO categories. (a) All PFRGs. (b) Anti-fibrotic genes. (c) Pro-fibrotic genes. The networks were generated with HumanBase online tool (https:// hb. flati ronin stitu te. org). Figure 4 . 4Network interactions between the human orthologs of pulmonary fibrosis-related genes (PFRGs). Figure 5 . 5Links between pulmonary fibrosis-related genes (PFRGs) and longevity. (a) Distribution of longevityassociated genes (LAGs) by their role in bleomycin-induced lung fibrosis in mice. 18 genes form the overlap between known LAGs and PFRGs. Among 12 pro-LAGs, 11 have an anti-fibrotic effect, while at least five out of the six anti-LAGs are pro-fibrotic. (b) Linear models of PFRG expression in lung tissue relative to maximum lifespan (MLS) identified 34 gene correlations (out of 90 PFRGs with orthologs in all considered species), for both pro-and anti-fibrotic genes (R 2 between 0.18 and 0.63). Table 2 . 2Consistency between the effects of genetic/protein manipulations on bleomycin-induced lung fibrosis in mice and expression of corresponding genes in human IPF. For a detailed list seeSupplementary Table ST1, ST2 and ST3.Consistency between bleomycin model and IPF Number of manipulations Percentage (%) Full 93 79.5 Partial 2 1.7 Inconsistent 10 8.5 Not clear 12 10.3 Consistency between the effects of different manipulations Full 54 85.7 Partial 7 11.1 Not clear 2 3.2 Table 4 . 4Processes and pathways enriched in pulmonary fibrosis-related genes (PFRGs). For a full enrichment analysis seeSupplementary Table ST4. Italicised areas that are common to all, pro-fibrotic, and anti-fibrotic PFRGs.All PFRGs Pro-fibrotic PFRGs Anti-fibrotic PFRGs Regulation of proliferation Positive regulation of proliferation Negative regulation of proliferation Cytokine signaling Cytokine signaling Inflammation Inflammation Immune function Immune function Cancer Cancer Cancer Reaction to pathogen Reaction to pathogen Oxygen homeostasis Oxygen homeostasis MAPK signaling pathway MAPK signaling pathway TNF signaling pathway TNF signaling pathway Jak-STAT signaling pathway Jak-STAT signaling pathway Asthma Aging Insulin resistance Stress response Estrogen signaling pathway Response to mechanical stimulus PI3K-Akt signaling pathway VEGF signaling pathway FoxO signaling pathway Apoptosis Table 5 . 5Genes/proteins that when manipulated had an effect on both longevity and lung fibrosis in mice. *Tert-The final impact on longevity is not entirely clear (for details please see the HAGR-GenAge database for details; https:// genom ics. senes cence. info/ genes/ index. html). For detailed description, see Suppl.Table ST8.Targeted gene/protein Impact on longevity Impact on lung fibrosis Akt1 Anti-Longevity Pro-fibrotic Akt2 Anti-Longevity Pro-fibrotic Cav1 Pro-Longevity Anti-fibrotic Fgf2 Pro-Longevity Anti-fibrotic Foxm1 Pro-Longevity Anti-fibrotic Kl (Klotho) Pro-Longevity Anti-fibrotic Mtor Anti-Longevity Pro-fibrotic Nos3 Pro-Longevity Anti-fibrotic Parp1 Anti-Longevity Pro-fibrotic Plau Pro-Longevity Anti-fibrotic Pparg Pro-Longevity Anti-fibrotic Rps6kb1 Anti-Longevity unclear Serpine1 (PAI-1) Anti-Longevity Pro-fibrotic Sirt1 Pro-Longevity Anti-fibrotic Sod3 Pro-Longevity Anti-fibrotic Tert Pro-Longevity* Pro-fibrotic Txn1 Pro-Longevity Anti-fibrotic Zmpste24 Pro-Longevity Anti-fibrotic (in old age) https://doi.org/10.1038/s41598-021-98674-6Scientific Reports | (2021) 11:19269 | © The Author(s) 2021 Author contributionsThis study was carried out by the V.E.F. and R.T. research groups. Data collection and processing were done by R.A.T., D.T. and V.E.F. V.E.F., R.T., H.Y. and D.T. conceptually designed the analyses. EvolutionaryCompeting interestsThe authors declare no competing interests.Additional informationSupplementary InformationThe online version contains supplementary material available at https:// doi. org/ 10. 1038/ s41598-021-98674-6.Correspondence and requests for materials should be addressed to R.T. or V.E.F.Reprints and permissions information is available at www.nature.com/reprints.Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Scientific Reports Aging, antagonistic pleiotropy and fibrotic disease. V J Thannickal, Int. J. Biochem. Cell Biol. 42Thannickal, V. J. Aging, antagonistic pleiotropy and fibrotic disease. Int. J. Biochem. Cell Biol. 42, 1398-1400 (2010). The aging lung: Physiology, disease, and immunity. J L Schneider, Cell. 184Schneider, J. L. et al. The aging lung: Physiology, disease, and immunity. Cell 184, 1990-2019 (2021). The RESOLVE concept: approaching pathophysiology of fibroproliferative disease in aged individuals. R Ziesche, M Golec, E Samaha, Biogerontology. 14Ziesche, R., Golec, M. & Samaha, E. The RESOLVE concept: approaching pathophysiology of fibroproliferative disease in aged individuals. Biogerontology 14, 679-685 (2013). The aging lung and idiopathic pulmonary fibrosis. S Gulati, V J Thannickal, Am. J. Med. Sci. 357Gulati, S. & Thannickal, V. J. The aging lung and idiopathic pulmonary fibrosis. Am. J. Med. Sci. 357, 384-389 (2019). Idiopathic pulmonary fibrosis: Aging, mitochondrial dysfunction, and cellular bioenergetics. D C Zank, M Bueno, A L Mora, M Rojas, Front. Med (Lausanne). 510Zank, D. C., Bueno, M., Mora, A. L. & Rojas, M. Idiopathic pulmonary fibrosis: Aging, mitochondrial dysfunction, and cellular bioenergetics. Front. Med (Lausanne) 5, 10 (2018). Emerging therapies for idiopathic pulmonary fibrosis, a progressive age-related disease. A L Mora, M Rojas, A Pardo, M Selman, Nat. Rev. Drug Discov. 16810Mora, A. L., Rojas, M., Pardo, A. & Selman, M. Emerging therapies for idiopathic pulmonary fibrosis, a progressive age-related disease. Nat. Rev. Drug Discov. 16, 810 (2017). Animal models of fibrotic lung disease. B Moore, Am. J. Respir. Cell Mol. Biol. 49Moore, B. et al. Animal models of fibrotic lung disease. Am. J. Respir. Cell Mol. Biol. 49, 167-179 (2013). Expression profiling suggests loss of surface integrity and failure of regenerative repair as major driving forces for COPD progression. E Samaha, Am. J. Respir. Cell Mol. Biol. 64Samaha, E. et al. Expression profiling suggests loss of surface integrity and failure of regenerative repair as major driving forces for COPD progression. Am. J. Respir. Cell Mol. Biol. 64, 441-452. (2021). Mechanistic links between aging and lung fibrosis. V J Thannickal, Biogerontology. 14Thannickal, V. J. Mechanistic links between aging and lung fibrosis. Biogerontology 14, 609-615 (2013). The bleomycin model of pulmonary fibrosis. T Liu, F G De Los Santos, S H Phan, Methods Mol. Biol. 1627Liu, T., De Los Santos, F. G. & Phan, S. H. The bleomycin model of pulmonary fibrosis. Methods Mol. Biol. 1627, 27-42 (2017). The bleomycin animal model: a useful tool to investigate treatment options for idiopathic pulmonary fibrosis?. A Moeller, K Ask, D Warburton, J Gauldie, M Kolb, Int. J. Biochem. Cell Biol. 40Moeller, A., Ask, K., Warburton, D., Gauldie, J. & Kolb, M. The bleomycin animal model: a useful tool to investigate treatment options for idiopathic pulmonary fibrosis?. Int. J. Biochem. Cell Biol. 40, 362-382 (2008). Human ex vivo lung perfusion: A novel model to study human lung diseases. N Cárdenes, Sci. Rep. 11490Cárdenes, N. et al. Human ex vivo lung perfusion: A novel model to study human lung diseases. Sci. Rep. 11, 490 (2021). Bleomycin in the setting of lung fibrosis induction: From biological mechanisms to counteractions. V Della Latta, A Cecchettini, S Del Ry, M A Morales, Pharmacol. Res. 97Della Latta, V., Cecchettini, A., Del Ry, S. & Morales, M. A. Bleomycin in the setting of lung fibrosis induction: From biological mechanisms to counteractions. Pharmacol. Res. 97, 122-130 (2015). The importance of interventional timing in the bleomycin model of pulmonary fibrosis. P Kolb, Eur. Respir. J. 551901105Kolb, P. et al. The importance of interventional timing in the bleomycin model of pulmonary fibrosis. Eur. Respir. J. 55, 1901105 (2020). Tissue repair genes: The TiRe database and its implication for skin wound healing. H Yanai, Oncotarget. 7Yanai, H. et al. Tissue repair genes: The TiRe database and its implication for skin wound healing. Oncotarget 7, 21145-21155 (2016). Orthology analysis between 273 proteomes, mostly eukaryotic. E L L Sonnhammer, G Östlund, Nucleic Acids Res. 8Sonnhammer, E. L. L. & Östlund, G. InParanoid 8: Orthology analysis between 273 proteomes, mostly eukaryotic. Nucleic Acids Res. 43, D234-D239 (2015). Building your own mash-up of gene annotations and expression profiles. C Wu, X Jin, G Tsueng, C Afrasiabi, A I Su, Biogps, Nucleic Acids Res. 44Wu, C., Jin, X., Tsueng, G., Afrasiabi, C. & Su, A. I. BioGPS: Building your own mash-up of gene annotations and expression profiles. Nucleic Acids Res. 44, D313-D316 (2016). A web server for exploration of mouse and human single-cell RNA sequencing data. O Franzén, L.-M Gan, J L M Björkegren, Panglaodb, Database (Oxford). 19Franzén, O., Gan, L.-M. & Björkegren, J. L. M. PanglaoDB: A web server for exploration of mouse and human single-cell RNA sequencing data. Database (Oxford) 1, 9 (2019). Genome-wide prediction and functional characterization of the genetic basis of autism spectrum disorder. A Krishnan, Nat. Neurosci. 19Krishnan, A. et al. Genome-wide prediction and functional characterization of the genetic basis of autism spectrum disorder. Nat. Neurosci. 19, 1454-1462 (2016). The BioGRID interaction database: 2019 update. R Oughtred, Nucleic Acids Res. 47Oughtred, R. et al. The BioGRID interaction database: 2019 update. Nucleic Acids Res. 47, D529-D541 (2019). Human Ageing Genomic Resources: New and updated databases. R Tacutu, Nucleic Acids Res. 46Tacutu, R. et al. Human Ageing Genomic Resources: New and updated databases. Nucleic Acids Res. 46, D1083-D1090 (2018). Met as a new marker of cellular senescence. M Boichuck, J Zorea, M Elkabets, M Wolfson, V E Fraifeld, Aging. 11Boichuck, M., Zorea, J., Elkabets, M., Wolfson, M. & Fraifeld, V. E. c-Met as a new marker of cellular senescence. Aging (Albany, NY) 11, 2889-2897 (2019). Wide-scale comparative analysis of longevity genes and interventions. H Yanai, A Budovsky, T Barzilay, R Tacutu, V E Fraifeld, Aging Cell. 16Yanai, H., Budovsky, A., Barzilay, T., Tacutu, R. & Fraifeld, V. E. Wide-scale comparative analysis of longevity genes and interven- tions. Aging Cell 16, 1267-1275 (2017). . 10.1038/s41598-021-98674-6Scientific Reports |. 11Scientific Reports | (2021) 11:19269 | https://doi.org/10.1038/s41598-021-98674-6 Machine learning analysis of longevity-associated gene expression landscapes in mammals. A Y Kulaga, Int. J. Mol. Sci. 221073Kulaga, A. Y. et al. Machine learning analysis of longevity-associated gene expression landscapes in mammals. Int. J. Mol. Sci. 22, 1073 (2021). An official American thoracic society workshop report: Use of animal models for the preclinical assessment of potential therapies for pulmonary fibrosis. R G Jenkins, Am. J. Respir. Cell Mol. Biol. 56Jenkins, R. G. et al. An official American thoracic society workshop report: Use of animal models for the preclinical assessment of potential therapies for pulmonary fibrosis. Am. J. Respir. Cell Mol. Biol. 56, 667-679 (2017). Long live FOXO: Unraveling the role of FOXO proteins in aging and longevity. R Martins, G J Lithgow, W Link, Aging Cell. 15Martins, R., Lithgow, G. J. & Link, W. Long live FOXO: Unraveling the role of FOXO proteins in aging and longevity. Aging Cell 15, 196-207 (2016). Cellular senescence-like features of lung fibroblasts derived from idiopathic pulmonary fibrosis patients. H Yanai, Aging. 7Yanai, H. et al. Cellular senescence-like features of lung fibroblasts derived from idiopathic pulmonary fibrosis patients. Aging (Albany, NY) 7, 664-672 (2015). Plasminogen activator inhibitor 1, fibroblast apoptosis resistance, and aging-related susceptibility to lung fibrosis. W.-T Huang, Exp. Gerontol. 61Huang, W.-T. et al. Plasminogen activator inhibitor 1, fibroblast apoptosis resistance, and aging-related susceptibility to lung fibrosis. Exp. Gerontol. 61, 62-75 (2015). Reversal of persistent fibrosis in aging by targeting Nox4-Nrf2 redox imbalance. L Hecker, Sci. Transl. Med. 6Hecker, L. et al. Reversal of persistent fibrosis in aging by targeting Nox4-Nrf2 redox imbalance. Sci. Transl. Med. 6, 231ra47 (2014). The role of cellular senescence in aging through the prism of Koch-like criteria. H Yanai, V E Fraifeld, Ageing Res. Rev. 41Yanai, H. & Fraifeld, V. E. The role of cellular senescence in aging through the prism of Koch-like criteria. Ageing Res. Rev. 41, 18-33 (2018). Transcriptomic profile of the mice aging lung is associated with inflammation and apoptosis as important pathways. J Calyeca, Y I Balderas-Martínez, M Selman, A Pardo, Aging. 13Calyeca, J., Balderas-Martínez, Y. I., Selman, M. & Pardo, A. Transcriptomic profile of the mice aging lung is associated with inflammation and apoptosis as important pathways. Aging (Albany, NY) 13, 12378-12394 (2021). Energy sensing pathways in aging and chronic lung disease. V J Thannickal, Trans. Am. Clin. Climatol. Assoc. 131Thannickal, V. J. Energy sensing pathways in aging and chronic lung disease. Trans. Am. Clin. Climatol. Assoc. 131, 286-293 (2020). Oxidative stress in pulmonary fibrosis. E Otoupalova, S Smith, G Cheng, V J Thannickal, Compr. Physiol. 10Otoupalova, E., Smith, S., Cheng, G. & Thannickal, V. J. Oxidative stress in pulmonary fibrosis. Compr. Physiol. 10, 509-547 (2020). Heterogeneous distribution of mechanical stress in human lung: A mathematical approach to evaluate abnormal remodeling in IPF. A Carloni, V Poletti, L Fermo, N Bellomo, M Chilosi, J. Theor. Biol. 332Carloni, A., Poletti, V., Fermo, L., Bellomo, N. & Chilosi, M. Heterogeneous distribution of mechanical stress in human lung: A mathematical approach to evaluate abnormal remodeling in IPF. J. Theor. Biol. 332, 136-140 (2013). Exploring animal models that resemble idiopathic pulmonary fibrosis. J Tashiro, Front Med. (Lausanne). 4118Tashiro, J. et al. Exploring animal models that resemble idiopathic pulmonary fibrosis. Front Med. (Lausanne) 4, 118 (2017). Systems biology approaches and tools for analysis of interactomes and multi-target drugs. A Schrattenholz, K Groebe, V Soskic, Methods Mol. Biol. 662Schrattenholz, A., Groebe, K. & Soskic, V. Systems biology approaches and tools for analysis of interactomes and multi-target drugs. Methods Mol. Biol. 662, 29-58 (2010). M(o)TOR of aging: MTOR as a universal molecular hypothalamus. M V Blagosklonny, Aging. 5Blagosklonny, M. V. M(o)TOR of aging: MTOR as a universal molecular hypothalamus. Aging (Albany, NY) 5, 490-494 (2013). Accelerated aging induced by deficiency of Zmpste24 protects old mice to develop bleomycin-induced pulmonary fibrosis. J Calyeca, Aging. 10Calyeca, J. et al. Accelerated aging induced by deficiency of Zmpste24 protects old mice to develop bleomycin-induced pulmonary fibrosis. Aging (Albany, NY) 10, 3881-3896 (2018). Wound healing and longevity: Lessons from long-lived αMUPA mice. H Yanai, Aging. 7Yanai, H. et al. Wound healing and longevity: Lessons from long-lived αMUPA mice. Aging (Albany, NY) 7, 167-176 (2015). Is rate of skin wound healing associated with aging or longevity phenotype. H Yanai, A Budovsky, R Tacutu, V E Fraifeld, Biogerontology. 12Yanai, H., Budovsky, A., Tacutu, R. & Fraifeld, V. E. Is rate of skin wound healing associated with aging or longevity phenotype?. Biogerontology 12, 591-597 (2011). DAVID-WS: A stateful web service to facilitate gene/protein list analysis. X Jiao, Bioinformatics. 28Jiao, X. et al. DAVID-WS: A stateful web service to facilitate gene/protein list analysis. Bioinformatics 28, 1805-1806 (2012). Enrichr: A comprehensive gene set enrichment analysis web server 2016 update. M V Kuleshov, Nucleic Acids Res. 44Kuleshov, M. V. et al. Enrichr: A comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res. 44, W90-W97 (2016). Cytoscape: A software environment for integrated models of biomolecular interaction networks. P Shannon, Genome Res. 13Shannon, P. et al. Cytoscape: A software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498-2504 (2003). Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. A Subramanian, Proc. Natl. Acad. Sci. USA. 102Subramanian, A. et al. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA 102, 15545-15550 (2005). Understanding multicellular function and disease with human tissue-specific networks. C S Greene, Nat. Genet. 47Greene, C. S. et al. Understanding multicellular function and disease with human tissue-specific networks. Nat. Genet. 47, 569-576 (2015). Ensemble comparative genomics resources. J Herrero, Database (Oxford). 117Herrero, J. et al. Ensemble comparative genomics resources. Database (Oxford) 1, 17 (2016). Econometric and statistical modeling with Python. S Seabold, J Perktold, Statsmodels, 10.25080/Majora-92bf1922-011Proceedings of the 9th Python in Science Conference 92-96. the 9th Python in Science Conference 92-96SciPySeabold, S. & Perktold, J. Statsmodels: Econometric and statistical modeling with Python. in Proceedings of the 9th Python in Sci- ence Conference 92-96 (SciPy, 2010). https:// doi. org/ 10. 25080/ Majora-92bf1 922-011.

2022-08-05T22:03:00Z

CCBY

https://academic.oup.com/hmg/article-pdf/25/21/4804/10018079/ddw307.pdf

HYBRID

aeb03017edd17faff128c007ed3ed54e86ada8cc

10.1093/hmg/ddw307

2513896837

28175300

5418736

Systematic analysis of the gerontome reveals links between aging and age-related diseases Maria Fernandes Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool LiverpoolUK Faculty of Sciences LaSIGE -Large-Scale Informatics Systems Laboratory University of Lisbon Portugal Cen Wan School of Computing University of Kent CanterburyUK Department of Computer Science University College London LondonUK Robi Tacutu Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool LiverpoolUK Diogo Barardo Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool LiverpoolUK Ashish Rajput German Center for Neurodegenerative Diseases (DZNE) Research Group for Computational Systems Biology Gö ttingenGermany Institute of Ageing and Chronic Disease Integrative Genomics of Ageing Group University of Liverpool William Henry Duncan Building, Room 281, 6 West Derby StreetLiverpool, UK Jingwei Wang Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool LiverpoolUK Harikrishnan Thoppil Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool LiverpoolUK Daniel Thornton Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool LiverpoolUK Chenhao Yang Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool LiverpoolUK Alex Freitas School of Computing University of Kent CanterburyUK João Pedro De Magalhães Integrative Genomics of Ageing Group Institute of Ageing and Chronic Disease University of Liverpool LiverpoolUK Systematic analysis of the gerontome reveals links between aging and age-related diseases 10.1093/hmg/ddw307O R I G I N A L A R T I C L E *To whom correspondence should be addressed at: In model organisms, over 2,000 genes have been shown to modulate aging, the collection of which we call the 'gerontome'. Although some individual aging-related genes have been the subject of intense scrutiny, their analysis as a whole has been limited. In particular, the genetic interaction of aging and age-related pathologies remain a subject of debate. In this work, we perform a systematic analysis of the gerontome across species, including human aging-related genes. First, by classifying aging-related genes as pro-or anti-longevity, we define distinct pathways and genes that modulate aging in different ways. Our subsequent comparison of aging-related genes with age-related disease genes reveals species-specific effects with strong overlaps between aging and age-related diseases in mice, yet surprisingly few overlaps in lower model organisms. We discover that genetic links between aging and age-related diseases are due to a small fraction of aging-related genes which also tend to have a high network connectivity. Other insights from our systematic analysis include assessing how using datasets with genes more or less studied than average may result in biases, showing that age-related disease genes have faster molecular evolution rates and predicting new aging-related drugs based on drug-gene interaction data. Overall, this is the largest systems-level analysis of the genetics of aging to date and the first to discriminate anti-and pro-longevity genes, revealing new insights on aging-related genes as a whole and their interactions with age-related diseases. Introduction Aging is a major social and medical challenge of the 21 st century. The most accepted mechanisms of aging include inflammation (1), apoptosis, oxidative stress, accumulation of DNA damage, cell cycle deregulation and mitochondrial dysfunction (2)(3)(4). In addition, one of the major breakthroughs in the field of aging research is the discovery that, in model organisms, aging is under genetic regulation (5). In the past 20 years, aging has been shown to be under genetic control in various short-lived model organisms, and in particular in yeast, worms, flies and mice. According to the GenAge database (6), over 2,000 genes can modulate aging and/or longevity in model organisms. We call the collection of these aging-related genes the 'gerontome' (7). Many of these genes work in common pathways (4), which include the insulin-like growth factor (IGF-1) signalling pathway, the target of rapamycin (TOR) pathway and the AMP kinase pathway (5). Although some individual aging-related genes have been the subject of intense scrutiny, their analysis as a whole has been limited (8)(9)(10)(11). Yet genes and proteins do not act individually. Therefore, biological networks provide a more realistic description of biological systems than single-molecule studies and give way to the integration of several types of data (12). Indeed, network analyses have already revealed insights on aging and its manipulation (13)(14)(15). Aging is associated with various diseases. The main categories of aging-related pathologies are: cancer, cardiovascular diseases, neurodegenerative diseases, nutritional and metabolic diseases (16)(17)(18). The relationship between aging and agerelated diseases has long been a contentious topic. A previous study has shown that the analysis of networks can uncover links between aging-related genes and age-related diseases (19), but many questions remain unanswered, like which agingrelated genes and pathways are important in these interactions? Moreover, we have further classified aging-related genes as anti-or pro-longevity, depending on how they are genetically manipulated and whether they increase or decrease lifespan in model organisms (6). Whether and how anti-and pro-longevity genes interact with aging disease-related genes is unknown. In this work, we performed a systematic analysis of the gerontome, the largest such analysis to date and the first to discriminate anti-and pro-longevity genes. Our analysis of pathways common to aging-related genes allows us to systematically classify pathways as anti-or pro-longevity, even though these mostly recapitulate previous findings. By contrast, our comparison of aging-related genes with age-related disease genes reveals several unexpected results: we found an association between aging-related genes and age-related diseases, yet this association is surprisingly organism-specific and driven by a small cluster of genes. Besides, one major issue in network analysis is whether some genes being better studied than others (what we call publication bias) impacts the underlying datasets and subsequent results. We correct for publication bias and show that a small but detectable fraction of results from protein-protein interaction network analysis is indeed influenced by whether genes are more studied than others. Lastly, we identify and rank drugs being targeted by aging-related genes which are promising for additional studies. Results Our systematic analysis of the gerontome employed the GenAge database developed by our lab (6). This includes 298 human candidate aging-related genes and genes associated with aging and/or longevity in model organisms of which over 1,000 can be converted to human homologs (see Materials and Methods). Model organism aging-related genes were further classified as pro-or anti-longevity depending on their effects: pro-longevity genes are defined as genes whose decreased expression reduces lifespan and/or whose overexpression extends lifespan; accordingly, anti-longevity genes are those whose decreased expression extends lifespan and/or whose overexpression decreases it (6) (Materials and Methods). This work is the first to consider such classification in a systematic way. Processes and pathways overrepresented in pro-and anti-longevity genes First, we performed a functional enrichment analysis of proand anti-longevity genes in each of the major model organisms. For pro-longevity genes, the most significant enriched pathways were p53-signalling pathway and cell cycle in mice; hypoxia response via HIF activation in Drosophila melanogaster; regulation of autophagy and oxidative phosphorylation in C. elegans (Supplementary Dataset 1). On the other hand, for antilongevity genes, insulin signalling, growth hormone signalling and IGF-1 receptor pathways were overrepresented in mice; the PI3 kinase pathway, oxidative phosphorylation and IGF pathway in Drosophila; oxidative phosphorylation, mTOR signalling pathway in C. elegans; ribosome in Saccharomyces cerevisiae. Some pathways like mTOR signalling, autophagy, insulin signalling and ribosome were enriched in more than one model organism (Supplementary Dataset 1). In addition to the more traditional functional enrichment, we also used a recently proposed feature selection method, from the area of data mining (or machine learning) to select relevant biological process Gene Ontology (GO) terms for predicting the pro-longevity or anti-longevity effect of a gene on a model organism (8). Among the top ranking GO terms identified by that feature selection method, terms associated with prolongevity included apoptotic signalling pathway and cell cycle checkpoint in mice, lipid metabolic process in Drosophila, autophagy in C. elegans and telomere organization in S. cerevisiae. By contrast, top ranking GO terms associated with antilongevity included positive regulation of multicellular organism growth in mice, sensory perception in Drosophila and translation in C. elegans (Supplementary Material, Dataset 2). Although the two aforementioned methods work in very different ways, there is some overlap between their results. In particular, in the results for mice, both methods found terms related to insulin signalling or growth to be significantly associated with anti-longevity; and terms related to the cell cycle were found to be significantly associated with pro-longevity. In addition, some terms related to autophagy were found to be significantly associated with pro-longevity in C. elegans by both methods. These results mostly recapitulate current knowledge of pathways associated with longevity manipulation in model organisms. Nonetheless, our results highlight pathways with proand anti-longevity effects and allow us to classify such pathways in a more consistent, systematic way. Pro-and anti-longevity networks are interwined Next, we tested if aging-related genes interact with each other and if we can observe the differences between the ways that pro-and anti-longevity genes form protein interaction networks. To perform this analysis, we employed protein-protein interaction data from BioGRID (see Materials and Methods) and focused our attention on worm genes, as the dataset of agingrelated genes in worms is by far the largest among the animal models in GenAge ( Table 1). Out of all the worm genes classified as anti-or pro-longevity genes (n ¼ 719), 283 genes had interactions in the BioGRID interactome (109 pro-and 174 anti-longevity genes). The average connectivity degree of pro-longevity genes was slightly higher than that of anti-longevity genes (8.42 compared to 5.43), and on average, both sets of aging-associated genes included more connected genes than similarly-sized random sets from the interactome (the connectivity degree for the entire interactome is 3.8). The clustering coefficient of pro-longevity genes was also higher than that of anti-longevity genes (0.108 compared to 0.063), showing that on average pro-longevity genes tend to cluster better than anti-longevity genes. In addition, we found that pro-and anti-longevity genes are much intertwined, with almost as many protein-protein interactions between genes from opposite categories (80 interactions) as between genes from the same category (43 interactions between pro-longevity genes and 56 interactions between antilongevity genes) (Fig. 1). While pro-and anti-longevity genes can form two network cores by themselves (28 genes are interconnected for each set), they also form a much larger network when taken together (90 genes), suggesting that the way in which pro-and anti-longevity genes determine lifespan is in many cases dependent on one another. Various previous studies have shown aging-related genes to form strong networks (9,10,14), as is normal in biology, but our results expand these observations to pro-and anti-longevity subnetworks and support substantial interactions between proand anti-longevity genes. Comparison with longevity-associated human genes In addition to genetic manipulations in model organisms, a number of genes have been associated with longevity in human populations (20). We therefore also performed a functional enrichment analysis of these genes using data from the LongevityMap, which consists of 755 human genes, 328 of which associated with longevity in at least one genetic association study (20). For a first functional enrichment analysis using DAVID (see Materials and Methods), with the background set as default, 49 clusters showed an enrichment score greater than 2.5 (Supplementary Dataset 3). With the background set as LongevityMap genes, 62 clusters with an enrichment score > 2.5 were obtained (Supplementary Material, Dataset 3). A similar functional annotation clustering pattern as seen in the first run was observed (Supplementary Material, Fig. S1), and major enriched clusters consisted of: regulation of apoptosis, regulation of phosphorylation, response to environment, regulation of locomotion and response to hormone stimulus. Results from human longevity-associated genes only modestly overlap with the above results for model organisms, although some pathways thought to be related to aging (e.g., apoptosis, response to oxidative stress and mTOR signalling) were found. Enriched clusters also included terms related to age-related diseases like cancer and diabetes mellitus (Supplementary Dataset 3). This may reflect how researchers choose candidate genes for longevity association studies, however. Perhaps researchers tend to select candidate genes for their studies that are suspected of playing important roles in human longevity, or in severe pathological processes that can significantly impair longevity. Overlap between aging-related genes and age-related diseases Next we aimed to study the genetic overlap between aging and age-related diseases (ARDs). For this analysis, we used human genes associated with ARDs from public databases (see Materials and Methods), human candidate aging-related genes and human homologs of genes associated with aging in model organisms from GenAge (6). Common or shared genes between ARDs and aging gene sets are referred to as common aging and disease (CAD) genes. In addition to an analysis focused on individual age-related diseases, a set named 'all diseases' and another named 'all classes' were created, composed of all genes considered in the analyses per individual age-related disease and per age-related disease class, respectively (Materials and Methods). As expected, the human aging-related gene set has the most associations with ARD genes. In addition, the immune system and respiratory tract disease classes only show a relation with aging in human aging-related genes ( Fig. 2A). Among the human homologs of genes associated with aging in model organisms, the musculoskeletal disease class only exhibits a significant overlap with aging-related genes in the mouse. A decrease in the overlap between aging and age-related diseases as evolutionary distance increases is also clear from our results ( Fig. 2A) with the mouse showing an overlap with more ARDs when compared with the other model organisms, even though fewer aging-related genes are known in mice than in flies or worms. This is also clear from looking at individual ARDs (Fig. 3) since in mice there is a significant overlap with 9 and 17 ARDs for anti-and pro-longevity sets, respectively, while the remaining model organisms present the following values: Drosophila -8 and 3; C. elegans -5 and 0; and finally S. Berevisiae -1 and 3. We also compared overlaps between anti-and pro-longevity genes. Pro-longevity genes present a higher number of overlapping age-related diseases than anti-longevity genes for all orthologs (i.e., combining human orthologs of all genes from model organisms), mouse and S. cerevisiae. The opposite is verified for Drosophila and C. elegans sets ( Fig. 2A). Supplementary Materials, Tables S6 and S7 include the P-values and the number of CAD-genes for, respectively, age-related disease classes and individual diseases. Neoplasms (1.35E-56), nutritional and metabolic (9.70E-34), cardiovascular (2.00E-23) and nervous system (1.78E-18) classes have the strongest associations with human aging. There is an additional class not considered in the individual ARD analysis, the eye diseases, which presents a positive association with aging only for human aging-related genes and mouse pro-longevity genes (Supplementary Materials, Table S6). Publication bias effects and correction The inclusion of more and less studied genes may reduce the accuracy of the results. This is an issue when using large datasets that may contain systematic biases. Indeed, we observed a moderate correlation between the number of publications associated with a gene and its number of annotated protein-protein interactions (Spearman correlation coefficient ¼ 0.67). While this is not unexpected, it could result in biases in systems biology analysis. To minimize this issue, a publication bias correction (PBC) based on the number of publications per gene was tested. The first step of the PBC was setting a threshold for differentiation between more and less studied genes. Table 1 shows the average and the median number of publications computed for the following gene sets: human genome, human interactome, human aging-related genes and human homologs of agingrelated genes from model organisms. The sets of aging-related genes have a higher average (range 13.0 to 30.3) and median (range 9 to 23) values when compared with the whole genome (average of 8.7 and median of 6) and the interactome (average of 10.4 and median of 8). This is expected but it shows that agingrelated genes are more studied than average. Thresholds between 8 and 20 publications were assessed in order to define the value with which the subsequent analyses were performed (see Supplementary Material, Fig. S2). Overall, we used 10 publications as a threshold. Overlap between aging-related genes and age-related diseases with publication bias correction The overlap analysis between aging-related genes and ARD genes was repeated after applying a filter for PBC (i.e., only containing genes with at least 10 publications). After PBC, the human aging-related gene set presents a significant overlap with ARD from all classes except for the immune system class (Fig. 2B). For the human homologs of aging-related genes from model organisms, only the mouse and the Drosophila present significant overlaps with ARDs and the latter only presents significant results for the anti-longevity gene set. In the mouse, pro-longevity genes have a higher number of ARDs overlapping compared to anti-longevity genes. In the all orthologs gene set, both anti-and pro-longevity genes show an association with the neoplasms class. These results are supported by P-values in Supplementary Materials, Tables S8 and S9, also suggesting a general stronger overlap with aging of genes associated with neoplasms and nutritional and metabolic diseases. Comparing the two analyses without and with PBC, respectively, Figure 2A and B, we observe a decrease in the number of significant overlapping ARDs with all aging-related gene sets due to the exclusion of less studied genes. The human aging gene set is the least affected by the exclusion of less studied genes since after PBC it presents only a loss of 16% of its genes (298 to 253 genes). Small reductions are verified in small gene sets, such as the baker's yeast and the mouse. The opposite is verified in bigger gene sets, such as worms, which lose 47% and 50% of genes in anti-and pro-longevity sets, respectively. Finally, these same patterns are observed comparing Supplementary Materials, Tables S6-S9, which show statistical tests for the various overlaps. Overlap between aging-related genes and age-related diseases in the interactome The interactome (15000 genes) is a subset of the genome (20183 genes) within which only genes for which protein-protein interaction data is available are present (Table 1). We assessed the overlap between aging and ARDs genes when restricting the analysis to genes in the human interactome (Supplementary Material, Fig. S3). The distribution of aging-related gene overlaps with ARDs is similar in the interactome (Supplementary Material, Fig. S3) to the distribution in the whole genome ( Fig. 2A), both without and with PBC. The analyses present similarities in the total number of genes, the overlap significance (P-values), the number of CAD-genes, and the relations between age-related diseases and aging shown by the anti-and pro-longevity sets. When comparing the whole genome and the interactome analyses without PBC (Supplementary Materials, Tables S6, S7, S10 and S11) and the whole genome and the interactome analyses with PBC (Supplementary Materials, Tables S8, S9, S12 and S13), there is a slight drop in the significance of overlaps with PBC, suggesting that some (but not all) results are indeed due to publication bias. Looking at the effect of PBC on the number of CAD-genes, there is again a slight decrease with PBC in the majority of ARD classes and individual diseases. We conclude that publication bias has a modest but noticeable effect on our results. Since genes function in combination with other genes, studying genes and proteins as part of interaction networks is essential (21). To study the effect of genes which interact with agingrelated genes in the links between aging and ARDs, we performed an analysis in which aging gene sets were composed of gene sets from the genome with PBC plus the genes directly connected to them (first order partners). This analysis revealed that aging-related gene sets including the first order partners are 18 to 51 times larger than the original aging-related gene sets with PBC (Fig. 4A). This increase in the number of genes is not proportional to the initial gene set size, and human aging-related genes are the ones that interact more with other genes. Regarding ARDs classes overlapping with aging-related genes, neoplasms and nervous system classes do so in all gene sets analysed. Cardiovascular, as well nutritional and metabolic classes are also present. Musculoskeletal diseases overlap with human aging-related genes and then they only overlap with anti-longevity genes of mice and Drosophila. Finally, respiratory tract diseases present a significant overlap with aging for the anti-longevity gene set in C. elegans. Overall, there is a clear increase in the number of overlapping ARD genes with aging-related genes by including first order interaction partners, as well as in the number of CAD-genes, which is supported by a statistical significance analysis (Supplementary Materials, Tables S14 and S15). Given the large increase in the number of genes by including first order partners, these results are not surprising but they underscore the large interconnection of biological networks, including between aging and age-related diseases. Co-expression network analysis Co-expression networks offer a complementary perspective on biological interactions from protein-protein interaction networks. To study co-expression, data were downloaded from the GeneFriends database (22) and genes co-expressed with agingrelated genes (human candidate genes plus human homologs of genes associated with aging in model organisms, all following PBC) were considered (see Materials and Methods). The inclusion of genes co-expressed with aging-related genes again changes the number of genes per set; and there is an increase which is proportional to the size of the initial set, i.e., larger sets have a greater increase in their sizes (Fig. 4B). At the level of age-related disease classes, neoplasms is the main class with a significant overlap between aging-related genes and ARDs genes, followed by the cardiovascular and nervous system classes. As before, the human aging-related gene set shows the highest association with ARDs. Immune system disease genes seem associated with aging-related genes when considering the anti-longevity genes of all orthologs together. Interestingly, there is a difference in overlapping classes between anti-and prolongevity gene sets and, except in the mouse, anti-longevity genes . The y-axis quantifies the number of age-related diseases which significantly overlap with aging-related genes; the x-axis describes the aging-related gene sets studied according to the source organism (i.e., human plus human homologs of aging-related genes from each model organism). The columns have seven different colours to represent each age-related disease classe analysed: Neoplasms (light blue), Nutritional and Metabolic diseases (orange), Nervous System diseases (light grey), Cardiovascular diseases (yellow), Musculoskeletal diseases (blue), Respiratory Tract diseases (green) and Immune System diseases (dark blue). The first column represents the number of age-related diseases with a significant overlap with candidate human aging-associated genes. Model organisms are ordered by evolutionary proximity to humans. This analysis was performed with PBC. The secondary y-axis displays the number of genes from the respective gene sets. (A) shows the number of significant overlapping aging-related genes with age-related diseases, including first order interaction partners. The interactome plus aging-related and age-related disease genes was considered as background. (B) shows the number of significant overlapping aging-related genes with age-related diseases, including co-expressed genes. The genome was considered as background. present a higher relation to ARDs than pro-longevity genes, which is very clear in flies and worms (Fig. 5). In the mouse, the results show that pro-longevity genes slightly overlap with nervous system disease genes, but the association is verified due to just one disease. In Drosophila, antilongevity genes are associated with nutritional and metabolic, cardiovascular and respiratory tract diseases, while the prolongevity genes are associated only with nervous system diseases. Finally, there is no significant overlap with any age-related diseases in the pro-longevity gene set of worms, even though it contains more than two thousand genes. Moreover, there are a few CAD-genes (up to 15) which show non-significant overlaps with the assessed ARDs (Supplementary Materials, Tables S16 and S17). Because including co-expressed genes increases the number of starting genes, there is an increase in the number of CAD-genes when including co-expressed genes. Using the human agingrelated gene set as an example, there are 65 CAD-genes in the overlap with neoplasms genes but when including the co-expressed genes the overlap with neoplasms increases to 131 CAD-genes. However, the percentage of overlapping genes drops dramatically. For human aging-related genes, 22% and 0.06% are associated with ARDs, with and without co-expressed genes, respectively. Properties of common genes between aging and age-related diseases Common genes or CAD-genes from the overlap analyses can highlight clues about pathways which link aging to disease processes. CAD-genes were obtained from the overlap between the human aging set and ARDs genes from analyses per individual age-related diseases or per diseases class, both with PBC. A small subset of aging-related genes are also associated with agerelated diseases The number of times that each CAD-gene overlaps with ARDs was defined as its frequency, and allows us to determine if some genes are involved in several disease processes. Figure 6A shows the frequency of CAD-genes across all the age-related disease classes. A total of 94 genes were obtained from the overlap between the human aging-related genes and all the ARDs genes per class. A majority of these genes (approx. 83% genes) overlap with up to three classes (Fig. 6A). Regarding genes which overlap with a great number of classes, PON1 and APOE are at the top (Fig. 6B), as well as some other genes involved in agerelated changes, for example, VEGFA, IL6 and AR. One gene present in all ARDs analysed is TNF (tumour necrosis factor). It is also interesting to explore aging-related genes which are not associated with any ARDs. The 94 CAD genes represent 37% of the human aging-related genes with PBC (253 genes), which means that most (63%) aging-related genes are not associated with any ARD class. From the perspective of ARDs genes (639 genes), about 15% have been related to human aging. An analysis of the CAD-genes distribution was also performed by individual ARDs. A total of 90 genes were found to overlap between human aging-related genes and ARD genes. Figure 7A shows that the number of genes involved in several ARDs is small, and about 59% (53 genes) of the 90 genes are associated with up to three ARDs. The pattern of distribution is similar to the analysis by ARD class and TNF, PON1, APOE and VEGFA are the top of CAD-genes among ARDs for both analyses (per age-related disease class and per individual disease) (Fig. 7B). In this analysis, the percentage of aging-related genes not associated with any age-related disease is about 64%. Similar to above, from the perspective of ARD genes (596 genes), only 15% have been related to human aging. Pathways and processes linking aging and age-related diseases A functional enrichment analysis was performed on CADgenes. The background used was the set of human agingrelated genes plus ARDs sets. Functional enrichment for CADgenes from all analysed ARD classes shows that these genes are Figure 5. Overlapping aging-related genes and their co-expressed partners with age-related diseases for various classes and organisms. Green means there is at least one age-related disease from that class that significantly overlaps with aging-related genes and red means no association. Model organisms are in descending order of their proximity to humans. This analysis was performed without PBC. associated with: negative regulation of apoptosis, cell cycle, positive regulation of DNA, positive regulation of protein metabolic process and response to stimuli (Supplementary Material, Table S1). Looking at CAD-genes in individual ARDs, only neoplasms, nutritional and metabolic, musculoskeletal and eye classes have significant functional clusters. Relative to neoplasms, CAD-genes are more associated with negative regulation of apoptosis, DNA repair, regulation of cell cycle and cancer, which is in line with cancer aetiology and its relationship to aging (17). CADgenes from the nutritional and metabolic class are related to response to insulin stimulus and positive regulation of lipid process, while CAD-genes for musculoskeletal diseases only show an association with the extracellular region. Finally, eye diseases CAD-genes seem to be associated with positive regulation of RNA metabolic process (Supplementary Material, Table S1). Increased network connectivity in genes common to aging and agerelated diseases Network approaches consider as a measure of node (gene) relevance the node's degree, which represents the number of connections of each node. This measure helps to define hubs, which in general are deemed essential nodes with many connections. To understand if CAD-genes are likely to be hubs, a comparison between the degree of CAD-genes and ARD genes or aging-related genes (the non-common genes) was made using protein-protein interaction data (Supplementary Material, Table S2). Age-related disease class analysis shows significant differences in node degree between CAD-genes and controls (P-value < 0.001). The median node degree of CAD-genes (47) is substantially higher than the median for the control set (11). Looking at ARD classes, only two classes have a significant (P < 0.05) difference between the two sets investigated: neoplasms and immune system diseases. Neoplasms present a higher median for CAD-genes (47) compared to the control set (23.5), while for the immune system class the opposite is verified (8.5 vs 43) (Supplementary Material, Table S2). The results from the analysis per individual ARDs show a significant difference in the number of node connections for four diseases: atherosclerosis (P ¼ 0.002), breast neoplasm (P ¼ 0.020), hypersensitivity (P ¼ 0.019) and osteoporosis (P ¼ 0.039). Except for breast neoplasm, the median for CADgenes is lower when compared to the ARD genes (Supplementary Material, Table S2). Processes associated with aging-related genes not associated with age-related diseases A functional enrichment was performed for genes from the human aging gene set which are not associated with any ARD. The main processes in the functional enrichment are: response to DNA damage, negative regulation of apoptosis, ATP-binding, negative regulation of transcription, DNA repair, aging, telomere maintenance, response to several stimuli, negative regulation of gene expression, cancer and signalling pathways (for examples, insulin, IL3 and MAPKinase). Of these terms, the ones with the higher cluster scores are response to DNA damage and negative regulation of apoptosis. The full list of significantly enriched terms is in Supplementary Material, Table S3. Molecular evolutionary rates of aging-and diseaserelated genes Aging-related and disease-related genes are also known to differ from the genome-wide average at the level of selection pressures. The study this, the dN/dS ratio between humans and mice of the human aging-related genes and the ARD genes sets was analysed and compared to the remaining genome (see Materials and Methods). Results show a significant (p < 0.001) difference between ARD genes and the other genes in the genome, wherein ARD genes have a higher median dN/dS ratio (0.137) than the whole genome (0.091). Although there was a difference between aging (median of 0.079) and non-aging genes (median of 0.093), this was not statistically significant (pvalue ¼ 0.155). The dN/dS ratio was also assessed in anti-and pro-longevity genes. A difference in dN/dS ratio between anti-and prolongevity genes was only observed in C. elegans (p-value ¼ 0.046, which is not significant after Bonferroni correction), so we find no evidence of differences in molecular evolution rates between anti-and pro-longevity genes. Searching for patterns and features which could define CADgenes, their molecular evolutionary (dN/dS) rate was analysed in comparison with aging-related genes and ARD genes. The CAD-genes used were from the overlaps of the three ARD classes with more genes: all classes together, neoplasms and nutritional and metabolic diseases. No statistically significant differences were found, suggesting that molecular evolution rates of CAD-genes are not different from other aging and ARD genes. Drugs predicted from aging-related gene interactions with drugs Given the large number of aging-related genes and pathways identified, there is great interest in identifying drugs that target them and may potentially have clinical benefits (23). To obtain candidate drugs affecting the aging process, we employed publicly available drug-gene interaction data (see Materials and Methods). In total, 376 drugs whose targets overlapped with aging-related genes were obtained. Twenty statisticallysignificant drugs that have more interactions with aging-related genes than expected by chance were obtained after Bonferroni correction (Supplementary Material, Table S4). The majority of the drugs obtained from this analysis were histone deacetylase inhibitors used for the treatment of cancer. This might be due to an overrepresentation of cancer drugs in public databases. Nonetheless, three known lifespan-extending drugs were identified: sodium phenylbutyrate, valproic acid and everolimus (Supplementary Material, Table S4). The fact that experimentally validated aging-related drugs are detected by our methodology suggests that this approach may be useful to identify new candidate drugs with effects on aging. Discussion To our knowledge, ours is the largest analysis of the gerontome to date, and the first to consider pro-and anti-longevity genes in a systematic fashion. We first characterized functions and pathways overrepresented in pro-and anti-longevity genes. Major anti-longevity pathways and processes include insulin signalling, growth hormone signalling and mTOR signalling. Key pro-longevity pathways include p53, cell cycle and autophagy. Although such pathways and processes are known to be related to aging (2,4,5,24), it is interesting that they are classified as anti-and pro-longevity in our systematic analysis of the genetics of aging. Differentiation between anti-longevity and prolongevity genes and processes can provide additional clues about aging-related processes and can help identify other genes with a similar effect on aging. In order to find relations between aging and ARDs, we compared aging-related gene sets with ARD genes. Limitations of our study include the fact that possibly many genes associated with longevity and diseases remain to be identified, and the causal genes in many genetic associations with disease are still unknown. In spite of these caveats, our results show an association between aging and ARDs at the genetic level, although this is surprisingly species-specific with a stronger overlap in mice than in invertebrates (flies and worms) and practically no overlap in yeast. The overlap analyses of anti-and pro-longevity genes shows differences in musculoskeletal, nervous system and cardiovascular diseases. The identified overlaps suggest that the musculoskeletal and nervous systems are related to pro-longevity genes while anti-longevity genes seem more associated with cardiovascular diseases. Looking at ARD classes which overlap with human aging-related genes, a significant overlap is verified for all classes as expected, except for immune system diseases in the analysis with PBC. The nutritional and metabolic diseases, the neoplasms, the cardiovascular diseases and the nervous system diseases have the most significant overlap with human aging-related genes. Eye diseases, respiratory tract diseases (which we considered a negative control) and immune system diseases had the least overlap, but it is important to mention that these are (together with musculoskeletal diseases) the age-related disease classes with fewer genes (Fig. 8). Genes historically associated with diseases are more likely to be studied. A publication bias correction approach, based on the number of publications associated with each gene, was applied in order to explore and reduce such biases. The analyses with and without PBC, when compared, show the effect of the removal of less studied genes ( Fig. 2A vs. B). The overlaps for C. elegans and S. cerevisiae disappear when the PBC is applied, which supports the hypothesis that some overlaps are statistically significant only due to an overrepresentation of betterstudied genes. The comparison of analyses with and without PBC proves that systematic researcher biases can influence the results in large-scale systems biology, genomic and genetic analysis. From our network analysis including the first order proteinprotein interaction partners, it is possible to conclude that aging-related genes are widely connected to other genes, which is supported by the huge increase in gene sets' sizes (Fig. 4A). There is an increase in the number of CAD-genes (the common or shared genes by aging and ARD) when including the first order partners, which suggests widespread interactions between aging-related genes and genes associated with age-related diseases. The results are also in agreement with recent research using genome-wide association studies (GWAS) data, which showed the same conclusion for five age-related categories: neurodegenerative, cancers, cardiovascular, metabolic and other diseases (25). A co-expression analysis of the links between aging and ARDs supports the idea of species-specific effects, but with more anti-longevity genes in invertebrates being related to ARDs. It is tempting to speculate that perhaps antilongevity genes work together more tightly in transcriptional networks than pro-longevity genes. Previous studies of the association between aging and diseases have demonstrated that the association is established by a small number of genes (25). Indeed, in the present analysis, CAD-genes represent a minority of the aging-related genes. CAD-genes are mainly related to apoptosis, metabolic regulation and DNA damage. These processes are similar to those previously reported to be associated with aging and may hint at underlying mechanisms important in various age-related diseases. CAD-genes also showed a higher number of connections with other genes than the remaining genome, which suggests that those genes tend to be hubs in networks. TNF, PON1, APOE and VEGFA are present in a great number of ARDs, which is in line with their involvement in some of the essential pathways whose disruption compromises metabolism and can lead to pathologies (26,27). The dN/dS ratio analysis showed a statistically significant higher dN/dS ratio of ARD genes when compared to the remaining genome, while aging-related genes had a lower dN/dS ratio that was not statistically significant. Therefore, we can affirm that ARD genes have a higher predisposition to changes in their sequence than aging-related genes. These results are in line with previous findings: an analysis using a previous version of GenAge found that aging-related genes have a lower dN/dS ratio (28). One previous study found a higher molecular evolutionary rate in disease genes (29). Our results further suggest that aging-related genes tend to be evolutionarily conserved, perhaps because they are part of essential pathways and conserved pleiotropic effects on aging (28), while genes associated with age-related diseases may be under relaxed selection given that they impact later in life. Finally, taking advantage of a database of gene-drug interactions, we mapped GenAge's genes to drugs and obtained a list of 20 candidate drugs for aging effects. Of these, three are already experimentally validated and the rest is yet to be explored. As such, these compounds are promising for future studies. Concluding Remarks The main conclusion from this work is that aging and agerelated diseases are related and share more genes than expected by chance. Human aging-related genes showed a considerable overlap with ARDs. These overlaps are driven by a small subset of aging-related genes which are associated with various age-related diseases and are hubs in networks. Besides, the extent of overlaps decreases with evolutionary distance, and yeast aging-related genes show practically no overlap with ARDs. Novel differences in overlapping age-related disease classes between anti-and pro-longevity genes were observed: Nervous system and musculoskeletal diseases seem more associated with pro-longevity, while cardiovascular diseases have a stronger association with anti-longevity genes. Moreover, network analyses (protein-protein interactions (PPI) and co-expression) suggest the existence of intermediate genes which promote the associations between aging and age-related disease genes. Overall, our work establishes a new standard in the analysis of aging-related genes in a systematic way. Materials and Methods Aging-and longevity-associated genes Aging-associated genes were obtained from GenAge Build 17 (6). These include 298 human candidate aging-related genes. GenAge also includes aging-or longevity-related genes in model organisms. For use with human datasets, human orthologs of model organism genes were used, composed of 1037 genes from the four main biomedical model organisms: mouse, fruit fly, roundworm and baker's yeast. The genes of each model organism were separated by their longevity classification: anti-or pro-longevity. Pro-longevity genes are genes whose decreased expression (due to knockout, mutations or RNA interference) reduces lifespan and/or whose overexpression extends lifespan; conversely, anti-longevity genes are those whose decreased expression extends lifespan and/or whose overexpression decreases it (6). Genes which were not included in one of these two longevity classes were excluded. A small number (19) of genes with both anti-and pro-longevity classifications were also excluded. To sum up the data, aging-related genes were divided into 11 gene sets: one set of 298 human aging-related genes, two sets (anti-and pro-longevity) of human orthologs from each model organism and two sets with all human orthologs of genes in all model organisms. The mouse sets (i.e. human orthologs of genes associated with aging in mice), have 23 and 59 genes, the fruit fly sets have 48 and 87 genes, the roundworm sets have 381 and 290 genes, and lastly the baker's yeast sets have 41 and 13 genes, respectively anti-and pro-longevity (Supplementary Materials, Table S18). Finally, sets with all orthologs have 448 and 421 genes, anti-and pro-longevity sets, respectively. The full lists of human aging-related genes and human orthologs are available in the sup plementary material (Supplementary Dataset 4). Data on human genes associated with longevity in genetic association studies were obtained from the LongevityMap build 1 (20). In the full set of 755 genes, there were 328 genes with at least one significant result reported. Age-related disease genes Age-related disease (ARDs) genes were assembled on 15-04-2015 from a diseases list compiled by a National Institute of Aging study. The list only includes genes with an association with the disease phenotype and with a MeSH annotation (30). This list is available online (https://www.irp.nia.nih.gov/ branches/rrb/dna/gene_sets.htm) and it was compiled using information from the Genetic Association Database (30). The original list includes many diseases not relevant for the present analysis since our interest focuses on complex ARDs. To select relevant ARDs, diseases with fewer than 20 genes associated and diseases of non-age-related disease classes were excluded. We chose a threshold of 20 genes because it captures the major age-related diseases yet not so many diseases that our findings end up being diluted (Supplementary Material, Table S19). The original list also includes processes and conditions, for example, insulin resistance and hyperlipidemia, which are dysfunctions, and for that reason were also excluded. The following analysed classes were described as age-related in the literature: cardiovascular diseases, eye diseases, immune system diseases, musculoskeletal diseases, nervous system diseases, nutritional and metabolic diseases and neoplasms (2,31). Respiratory tract diseases were considered as negative controls since the two diseases (after application of the described selection criteria) in this class are asthma, which is not considered an age-related disease (32), and chronic obstruction pulmonary disease, which is primarily environmental. Selection of individually studied ARDs was made based on two criteria: first, the number of genes, to have larger sample sizes and increase statistical power. The second criterion was how often and common was each disease. An example of selection is the case of ovarian neoplasm, which presents a smaller number of genes and is better known than head and neck neoplasms. In order to have a representative selection, seven diseases classes were included in the individual age-related disease analysis; for classes with a large number of diseases, we selected the top five or six most representative individual diseases of the class. Eye diseases were excluded from the individual disease analysis, since they include only non-common diseases with a small number of genes. Diseases that are primarily driven by environmental factors, like chronic obstruction pulmonary disease, were also not studied. In total, 893 different genes associated with ARDs were considered. Figure 8 shows the number of genes per ARD class and the number of genes shared with each one of the remaining classes. The list of ARDs used in the present analysis is summarized in Supplementary Material, Protein-protein interaction and gene co-expression data Protein-protein interactions were obtained using the BioGRID plug-in available in Cytoscape, on 16-04-2015, by downloading the available node and edge tables. The two main types of interactions ('physical association' and 'direct interaction') represent 124,238 of 140,891 interactions, involving 14,721 of 15,000 proteins. As such, the interactome analysis was performed using the full interactome. To obtain the first-order partners of agingrelated genes, a Python script was used to compile connections between all genes in the interactome and return the merged list of seed genes and genes connected to them. Co-expression data from RNA-Seq was obtained using GeneFriends (22) on 03-06-2015. To obtain the co-expressed genes, a significance threshold of 2.5E-06 was applied to the pvalue retrieved from GeneFriends. The threshold was defined by correction of standard a (0.05) using a Bonferroni correction where N represents the genome size (20183 genes). Publication bias correction (PBC) The number of publications was compiled from the Swiss-Prot PubMed annotation list, downloaded on 23-04-2015. Only human and reviewed genes were considered for this analysis. Although PubMed publications annotated in Swiss-Prot are not the total number of publications for each entry, they represent a curated selection. Thus, Swiss-Prot was selected as the source for the number of publications due to its curated nature, which makes it a reliable source of annotated data for protein coding genes. Gene features The list of all human genes was collected from GenBank on 31-01-2015. For this analysis, only human annotated and protein-coding genes were considered, which represent a set of 20,183 genes. Molecular evolutionary rate (dN/dS) was calculated from the number of synonymous (dN) and non-synonymous (dS) substitutions downloaded from Ensembl BioMart, selecting the Ensembl Genes 80 database and Homo sapiens genes (GRCh38.p2) dataset on 17-05-2015. Of all the model organisms considered in the present work, only mouse orthologs present dN and dS values, due to the great evolutionary distance shown by the other organisms. Thus, all dN/dS ratio comparisons consider the evolution between mice and humans. Feature selection method We used a recently proposed feature selection method, from the area of data mining (or machine learning) to select the relevant Gene Ontology (GO) terms for predicting the pro-longevity or antilongevity effect of a gene on a model organism (8). In essence, we addressed the classification task of data mining, where the goal is to predict the class (pro-or anti-longevity effect) of an instance (a gene) based on predictive features (GO terms) associated with that gene. The used feature selection method differs from other feature selection methods typically used in data mining in two important ways, as follows. First, it selects a specific set of features relevant for the classification of each instance, instead of selecting the same set of features for all instances, as usual in data mining. This increases the flexibility of the feature selection process, recognizing that the optimal set of GO terms for predicting the proor anti-longevity effect of a gene varies across different genes. Second, the used method performs 'hierarchical' feature selection, in the sense that it takes into account the hierarchical structure of the GO in order to improve the feature selection process; unlike conventional ('flat') feature selection methods. That feature selection method was applied to datasets with data about aging-related genes from the four traditional model organisms, namely: mouse, fly, roundworm and yeast. The results of the feature selection method were transformed into a rank of GO terms as follows. For each dataset (model organism), for each GO term, we counted the number of instances (genes) where that GO term was one of the relevant features selected by the feature selection method for predicting the pro-or anti-longevity effect of the gene. Then, we ranked the GO terms in decreasing order of this frequency of selection. We also used a statistical test of significance based on the binomial distribution to detect which GO terms were significantly associated with the class being predicted. A detailed description of the feature selection method and how its results were used to rank the GO terms can be found in (8). Overlap analysis A significant overlap between aging-related genes and ARDs is defined as: i) an observed number of CAD-genes above the number of CAD-genes expected by chance; and ii) a p-value below 0.05 (Fisher's exact test). Genome and interactome analysis used whole genome and interactome as background, respectively. The background for the first order partners analysis was the seed list plus the interactome and was adjusted for each aging gene set since the seed list varies between different aging sets. In this analysis, the interactome was added to the background since the first order partners were selected from that group. Functional enrichment analysis Functional enrichment analysis using the Database of Annotation Visualization and Integrated Discovery (DAVID) (33) was performed to identify overrepresented categories. The analysis was done by running the Functional Annotation Clustering module under default parameters. Unless otherwise stated, the whole genome was used as background. Enrichment scores (E. Score) above 1.3 (which corresponds to P ¼ 0.05) are widely accepted as relevant (33); however, in this analysis a threshold of 2.5 (corresponding to P ¼ 0.003) was used for more significant results. A Benjamini correction was applied for correcting for multiple hypothesis testing. Candidate drugs from GenAge targets To identify candidate drugs with possible anti-aging properties, the Drug Gene Interaction Database (DGIdb) version 2 (34) was used. We classified all 44 types of drug-gene interactions in DGIDB into either 'Anti' (decrease gene expression or activity) or 'Pro' (increase gene expression or activity) or 'Neither' (non-applicable or undefined effects), so that they can be matched with GenAge genes to obtain a putative lifespan-extending effect (Supplementary Material, Table S5). Drugs were obtained by considering if they interact with GenAge genes in a way that would be predicted to extend lifespan. That is, for an 'Anti' drug, only the interactions with anti-longevity genes are scored; and vice-versa for 'Pro' drugs. In total, 376 drugs were obtained which were ordered based on ascending P-value obtained using a one-tailed hypergeometric test. After Bonferroni correction (a ¼ 0.05), 20 statistically significant drugs were obtained. Supplementary Material Supplementary Material is available at HMG online. Figure 1 . 1Protein-protein interactions between worm aging-related genes. Pro-longevity genes are depicted in red and anti-longevity genes in green. For each of the two gene sets, the smaller inside ellipse indicates genes that form a continuously connected network. Left right straight and curved arrows are used to summarize undirected interactions between genes from different and the same gene set, respectively. Figure 3 . 3Overlapping aging-related genes for various organisms with age-related disease genes sets. Green means significant overlap between aging-related and agerelated disease genes and red means there is no significant overlap. Model organisms are in descending order of their proximity to humans. Figure 4 4Figure 4. The y-axis quantifies the number of age-related diseases which significantly overlap with aging-related genes; the x-axis describes the aging-related gene sets studied according to the source organism (i.e., human plus human homologs of aging-related genes from each model organism). The columns have seven different Figure 6 . 6(A) CAD-genes distribution as associated with age-related disease classes. (B) shows the genes involved in half or more disease classes. TNF is associated with all the age-related disease classes analysed. This analysis was performed with PBC. Figure 7 . 7(A) CAD-gene distribution as associated with individual age-related diseases. (B) shows the CAD-genes involved in ten or more individual diseases, with PON1, TNF, APOE the top 3 genes associated with the greatest number of age-related diseases. This analysis was performed with PBC. Figure 8 . 8Number of genes by age-related disease class (Total column) and shared with each other disease classes. The white cells present the number of genes shared between disease classes and the darker grey cells show the number of genes not shared with any other disease class. Table 1 . 1Number of genes plus average and median number of publications per gene in each datasetDataset a Num. of genes Average num. pubs. Median num. pubs. Human genome (NCBI) 20183 b 8.7 6 Human interactome (BioGRID) 15000 c 10.4 8 Human aging-related genes 298 30.3 23 All aging-related orthologs 894 14.5 10 anti-longevity 448 13.2 9 pro-longevity 421 15.9 11 M. musculus 84 26.8 19 anti-longevity 23 22.7 13 pro-longevity 59 28.6 21 D. melanogaster 135 19.9 13 anti-longevity 48 20.1 12 pro-longevity 87 19.6 13 C. elegans 693 13.1 9 anti-longevity 381 13.0 9 pro-longevity 290 13.1 9 S. cerevisiae 62 17.9 14 anti-longevity 41 14.7 10 pro-longevity 13 23.2 15 Notes:. a All datasets refer to human genes, including human orthologs of genes from various model organisms. b Genome has 20183 annotated genes in NCBI but only 19071 are in the Swiss- Prot database. c Interactome has 15000 annotated genes in NCBI but only 14498 are in the Swiss-Prot database. Table S20 . S20In total, 40 diseases were part of ARD classes, of which 22 ARDs were analysed individually. The list of all age-related diseases and their related Supplementary Dataset 5). Our full datasets are also available on GitHub (https://github.com/maglab/genage-analysis).genes is available in the supplementary material ( Figure 2. The y-axis quantifies the number of age-related diseases which significantly overlap with aging-related genes; the x-axis describes the aging-related gene sets studied according to the source organism (i.e., human plus human homologs of aging-related genes from each model organism). The columns have seven different colours to represent each age-related disease classe analysed: Neoplasms (light blue), Nutritional and Metabolic diseases (orange), Nervous System diseases (light grey), Cardiovascular diseases (yellow), Musculoskeletal diseases (blue), Respiratory Tract diseases (green) and Immune System diseases (dark blue). The first column represents the number of age-related diseases with a significant overlap with candidate human aging-associated genes. Model organisms are ordered by evolutionary proximity to humans. The genome was considered as background. The secondary y-axis displays the number of genes from the respective gene sets. (A) shows the number of significant overlapping aging-related genes with age-related diseases. (B) shows the number of significant overlapping aging-related genes with age-related diseases with PBC (i.e., only genes with more than 10 publications were used). | Human Molecular Genetics, 2016, Vol. 25, No. 21 AcknowledgementsWe thank Louise Crompton for assistance in compiling the relevant literature.Conflict of Interest statement. None declared.FundingGenAge is supported by a Wellcome Trust grant (104978/Z/14/Z) to JPM. Funding to pay the Open Access publication charges for this article was provided by Wellcome Trust. Inflamm-aging -An evolutionary perspective on immunosenescence. C Franceschi, M Bonafe, S Valensin, F Olivieri, M De Luca, E Ottaviani, G De Benedictis, Mol. Cell. Gerontol. 908Franceschi, C., Bonafe, M., Valensin, S., Olivieri, F., De Luca, M., Ottaviani, E. and De Benedictis, G. (2000) Inflamm-aging - An evolutionary perspective on immunosenescence. Mol. Cell. Gerontol., 908, 244-254. . C Lopez-Otin, M A Blasco, L Partridge, M Serrano, G Kroemer, The Hallmarks of Aging. Cell. 153Lopez-Otin, C., Blasco, M.A., Partridge, L., Serrano, M. and Kroemer, G. (2013) The Hallmarks of Aging. Cell, 153, 1194-1217. J P De Magalhaes, I Stuart-Hamilton, An Introduction to Gerontology. CambridgeCambridge University Pressde Magalhaes, J.P. (2011) Stuart-Hamilton, I. (ed.), In An Introduction to Gerontology. Cambridge University Press, Cambridge, pp. 21-47. Open-minded scepticism: inferring the causal mechanisms of human ageing from genetic perturbations. J P De Magalhaes, Ageing Res. Rev. 4de Magalhaes, J.P. (2005) Open-minded scepticism: inferring the causal mechanisms of human ageing from genetic per- turbations. Ageing Res. Rev., 4, 1-22. The genetics of ageing. C J Kenyon, Nature. 464Kenyon, C.J. (2010) The genetics of ageing. Nature, 464, 504-512. Human Ageing Genomic Resources: Integrated databases and tools for the biology and genetics of ageing. R Tacutu, T Craig, A Budovsky, D Wuttke, G Lehmann, D Taranukha, J Costa, V E Fraifeld, De Magalhaes, Nucleic Acids Res. 41J.P.Tacutu, R., Craig, T., Budovsky, A., Wuttke, D., Lehmann, G., Taranukha, D., Costa, J., Fraifeld, V.E. and de Magalhaes, J.P. (2013) Human Ageing Genomic Resources: Integrated data- bases and tools for the biology and genetics of ageing. Nucleic Acids Res., 41, D1027-D1033. Gerontome: a webbased database server for aging-related genes and analysis pipelines. J Kwon, B Lee, H Chung, BMC Genomics. 1120Kwon, J., Lee, B. and Chung, H. (2010) Gerontome: a web- based database server for aging-related genes and analysis pipelines. BMC Genomics, 11, S20. Predicting the Pro-Longevity or Anti-Longevity Effect of Model Organism Genes with New Hierarchical Feature Selection Methods. C Wan, A A Freitas, J P De Magalhaes, IEEE-ACM Transact. Comput. Biol. Bioinform. 12Wan, C., Freitas, A.A. and de Magalhaes, J.P. (2015) Predicting the Pro-Longevity or Anti-Longevity Effect of Model Organism Genes with New Hierarchical Feature Selection Methods. IEEE- ACM Transact. Comput. Biol. Bioinform., 12, 262-275. GenAge: a genomic and proteomic network map of human ageing. J P De Magalhaes, O Toussaint, FEBS Lett. 571de Magalhaes, J.P. and Toussaint, O. (2004) GenAge: a geno- mic and proteomic network map of human ageing. FEBS Lett, 571, 243-247. Systems-level analysis of human aging genes shed new light on mechanisms of aging. Q Zhang, R Nogales-Cadenas, J R Lin, W Zhang, Y Cai, J Vijg, Z D Zhang, Hum Mol Genet. 25Zhang, Q., Nogales-Cadenas, R., Lin, J.R., Zhang, W., Cai, Y., Vijg, J., and Zhang, Z.D. (2016) Systems-level analysis of hu- man aging genes shed new light on mechanisms of aging. Hum Mol Genet., 25, 2934-2947. A data mining approach for classifying DNA repair genes into ageing-related or non-ageing-related. A A Freitas, O Vasieva, J P De Magalhaes, BMC Genomics. 1227Freitas, A.A., Vasieva, O. and de Magalhaes, J.P. (2011) A data mining approach for classifying DNA repair genes into age- ing-related or non-ageing-related. BMC Genomics, 12, 27. Dynamic networks reveal key players in aging. F E Faisal, T Milenkovic, Bioinformatics. 30Faisal, F.E. and Milenkovic, T. (2014) Dynamic networks re- veal key players in aging. Bioinformatics, 30, 1721-1729. Dissecting the gene network of dietary restriction to identify evolutionarily conserved pathways and new functional genes. D Wuttke, R Connor, C Vora, T Craig, Y Li, S Wood, O Vasieva, R Shmookler Reis, F Tang, J P De Magalhaes, PLoS Genet. 81002834Wuttke, D., Connor, R., Vora, C., Craig, T., Li, Y., Wood, S., Vasieva, O., Shmookler Reis, R., Tang, F. and de Magalhaes, J.P. (2012) Dissecting the gene network of dietary restriction to identify evolutionarily conserved pathways and new functional genes. PLoS Genet., 8, e1002834. Prediction of C. elegans longevity genes by human and worm longevity networks. R Tacutu, D E Shore, A Budovsky, J P De Magalhaes, G Ruvkun, V E Fraifeld, S P Curran, PLoS One. 748282Tacutu, R., Shore, D.E., Budovsky, A., de Magalhaes, J.P., Ruvkun, G., Fraifeld, V.E. and Curran, S.P. (2012) Prediction of C. elegans longevity genes by human and worm longevity networks. PLoS One, 7, e48282. GeneFriends: an online coexpression analysis tool to identify novel gene targets for aging and complex diseases. S Van Dam, R Cordeiro, T Craig, J Van Dam, S H Wood, De Magalhaes, BMC Genomics. 535J.P.van Dam, S., Cordeiro, R., Craig, T., van Dam, J., Wood, S.H. and de Magalhaes, J.P. (2012) GeneFriends: an online co- expression analysis tool to identify novel gene targets for ag- ing and complex diseases. BMC Genomics, 13, 535. Gene expression reveals overlap between normal aging and Alzheimer's disease genes. D Avramopoulos, M Szymanski, R H Wang, S Bassett, Neurobiol. Aging. 32Avramopoulos, D., Szymanski, M., Wang, R.H. and Bassett, S. (2011) Gene expression reveals overlap between normal aging and Alzheimer's disease genes. Neurobiol. Aging, 32, 2319.e27-34. How ageing processes influence cancer. J P De Magalhaes, Nat. Rev. Cancer. 13de Magalhaes, J.P. (2013) How ageing processes influence cancer. Nat. Rev. Cancer, 13, 357-365. Understanding the biology of aging with interaction networks. F Peysselon, S Ricard-Blum, Maturitas. 69Peysselon, F. and Ricard-Blum, S. (2011) Understanding the biol- ogy of aging with interaction networks. Maturitas, 69, 126-130. Disease-Aging Network Reveals Significant Roles of Aging Genes in Connecting Genetic Diseases. J G Wang, S H Zhang, Y Wang, L N Chen, X S Zhang, Plos Comput. Biol. 51000521Wang, J.G., Zhang, S.H., Wang, Y., Chen, L.N.,. and Zhang, X.S. (2009) Disease-Aging Network Reveals Significant Roles of Aging Genes in Connecting Genetic Diseases. Plos Comput. Biol., 5, e1000521. LongevityMap: a database of human genetic variants associated with longevity. A Budovsky, T Craig, J Wang, R Tacutu, A Csordas, J Lourenco, V E Fraifeld, De Magalhaes, Trends Genet. 29J.P.Budovsky, A., Craig, T., Wang, J., Tacutu, R., Csordas, A., Lourenco, J., Fraifeld, V.E. and de Magalhaes, J.P. (2013) LongevityMap: a database of human genetic variants associ- ated with longevity. Trends Genet., 29, 559-560. J De Magalhães, R Tacutu, Handbook of the Biology of Aging. Kaeberlein, M. and Martin, G.London, UKAcademic Press8th editionde Magalhães, J. and Tacutu, R. (2016) Kaeberlein, M. and Martin, G. (eds.), In Handbook of the Biology of Aging, 8th edi- tion. Academic Press, London, UK, pp. 263-285. GeneFriends: a human RNA-seq-based gene and transcript co-expression database. S Van Dam, T Craig, J P De Magalhaes, Nucleic Acids Res. 43van Dam, S., Craig, T. and de Magalhaes, J.P. (2015) GeneFriends: a human RNA-seq-based gene and transcript co-expression database. Nucleic Acids Res., 43, D1124-D1132. Genome-environment interactions that modulate aging: powerful targets for drug discovery. J P De Magalhaes, D Wuttke, S H Wood, M Plank, C Vora, Pharmacol. Rev. 64de Magalhaes, J.P., Wuttke, D., Wood, S.H., Plank, M. and Vora, C. (2012) Genome-environment interactions that mod- ulate aging: powerful targets for drug discovery. Pharmacol. Rev., 64, 88-101. Cell divisions and mammalian aging: integrative biology insights from genes that regulate longevity. J P De Magalhaes, R G Faragher, Bioessays. 30de Magalhaes, J.P. and Faragher, R.G. (2008) Cell divisions and mammalian aging: integrative biology insights from genes that regulate longevity. Bioessays, 30, 567-578. Genetic evidence for common pathways in human age-related diseases. S C Johnson, X Dong, J Vijg, Y Suh, Aging Cell. 14Johnson, S.C., Dong, X., Vijg, J. and Suh, Y. (2015) Genetic evi- dence for common pathways in human age-related dis- eases. Aging Cell, 14, 809-817. Age-dependent paraoxonase 1 (PON1) activity and LDL oxidation in Wistar rats during their entire lifespan. D Kumar, S I Rizvi, Scientific World Journal. 538049Kumar, D. and Rizvi, S.I. (2014) Age-dependent paraoxonase 1 (PON1) activity and LDL oxidation in Wistar rats during their entire lifespan. Scientific World Journal, 2014, 538049. Vascular endothelial growth factor (VEGF) plasma concentrations in coronary artery disease. H F Alber, M Frick, J Dulak, J Dorler, R H Zwick, W Dichtl, O Pachinger, F Weidinger, Heart. 91Alber, H.F., Frick, M., Dulak, J., Dorler, J., Zwick, R.H., Dichtl, W., Pachinger, O. and Weidinger, F. (2005) Vascular endothe- lial growth factor (VEGF) plasma concentrations in coronary artery disease. Heart, 91, 365-366. Analyses of human-chimpanzee orthologous gene pairs to explore evolutionary hypotheses of aging. J P De Magalhaes, G M Church, Mech Ageing Dev. 128de Magalhaes, J.P. and Church, G.M. (2007) Analyses of human-chimpanzee orthologous gene pairs to explore evolutionary hypotheses of aging. Mech Ageing Dev., 128, 355-364. Human disease genes: patterns and predictions. N G C Smith, A Eyre-Walker, Gene. 318Smith, N.G.C. and Eyre-Walker, A. (2003) Human disease genes: patterns and predictions. Gene, 318, 169-175. Systematic analysis, comparison, and integration of disease based human genetic association data and mouse genetic phenotypic information. Y Zhang, S De, J R Garner, K Smith, S A Wang, K G Becker, BMC Med Genomics. 31Zhang, Y., De, S., Garner, J.R., Smith, K., Wang, S.A. and Becker, K.G. (2010) Systematic analysis, comparison, and integration of disease based human genetic association data and mouse ge- netic phenotypic information. BMC Med Genomics, 3, 1. Age-related changes in the musculoskeletal system and the development of osteoarthritis. R F Loeser, Clin. Geriatr. Med. 26Loeser, R.F. (2010) Age-related changes in the musculoskele- tal system and the development of osteoarthritis. Clin. Geriatr. Med., 26, 371-386. Differences in incidence of reported asthma related to age in men and women -A retrospective analysis of the data of the European Respiratory Health Survey. R De Marco, F Locatelli, J Sunyer, P Burney, E C R H Surve, Am. J. Respir. Crit. Care Med. 162de Marco, R., Locatelli, F., Sunyer, J., Burney, P. and Surve, E.C.R.H. (2000) Differences in incidence of reported asthma related to age in men and women -A retrospective analysis of the data of the European Respiratory Health Survey. Am. J. Respir. Crit. Care Med., 162, 68-74. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. D W Huang, B T Sherman, R A Lempicki, Nat. Protoc. 4Huang, D.W., Sherman, B.T. and Lempicki, R.A. (2009) Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc., 4, 44-57. DGIdb: mining the druggable genome. M Griffith, O L Griffith, A C Coffman, J V Weible, J F Mcmichael, N C Spies, J Koval, I Das, M B Callaway, J M Eldred, Nature Methods. 10Griffith, M., Griffith, O.L., Coffman, A.C., Weible, J.V., McMichael, J.F., Spies, N.C., Koval, J., Das, I., Callaway, M.B., Eldred, J.M., et al. (2013) DGIdb: mining the druggable ge- nome. Nature Methods, 10, 1209-1210.

2023-04-05T05:20:17Z

CCBY

https://www.aging-us.com/article/203791/pdf

GOLD

827848750e52042b9707f71a789b39eb217c827c

10.18632/aging.203791

34919532

8751603

Small molecules for cell reprogramming: a systems biology analysis Published: December 17, 2021 Anna Knyazer The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Gabriela Bunu Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Dmitri Toren Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Teodora Bucaciuc Mracica Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Yael Segev The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Marina Wolfson The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael Khachik K Muradian D.F. Chebotarev Institute of Gerontology of National Academy of Medical Sciences of Ukraine KievUkraine Robi Tacutu [email protected]@0000-0001-5853-2470 Systems Biology of Aging Group Institute of Biochemistry of the Romanian Academy BucharestRomania Vadim E Fraifeld [email protected]@orcid.org The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer-ShevaIsrael D.F. Chebotarev Institute of Gerontology of National Academy of Medical Sciences of Ukraine KievUkraine Small molecules for cell reprogramming: a systems biology analysis 1324Published: December 17, 2021Received: August 6, 2021 Accepted: November 24, 202125739 AGING Research Paper *Equal contribution Correspondence to: Robi Tacutu, Vadim E. Fraifeld;cell reprogrammingchemically-induced pluripotencychemical-protein interactionsprotein-protein interaction networkslongevity pathways If somatic stem cells would be able to maintain their regenerative capacity over time, this might, to a great extent, resolve rejuvenation issues. Unfortunately, the pool of somatic stem cells is limited, and they undergo cell aging with a consequent loss of functionality. During the last decade, low molecular weight compounds that are able to induce or enhance cell reprogramming have been reported. They were named "Small Molecules" (SMs) and might present definite advantages compared to the exogenous introduction of stemness-related transcription factors (e.g. Yamanaka's factors). Here, we undertook a systemic analysis of SMs and their potential gene targets. Data mining and curation lead to the identification of 92 SMs. The SM targets fall into three major functional categories: epigenetics, cell signaling, and metabolic "switchers". All these categories appear to be required in each SM cocktail to induce cell reprogramming. Remarkably, many enriched pathways of SM targets are related to aging, longevity, and age-related diseases, thus connecting them with cell reprogramming. The network analysis indicates that SM targets are highly interconnected and form protein-protein networks of a scale-free topology. The extremely high contribution of hubs to network connectivity suggests that (i) cell reprogramming may require SM targets to act cooperatively, and (ii) their network organization might ensure robustness by resistance to random failures. All in all, further investigation of SMs and their relationship with longevity regulators will be helpful for developing optimal SM cocktails for cell reprogramming with a perspective for rejuvenation and life span extension. INTRODUCTION The pool of adult stem cells is limited, and they undergo cell aging with a consequent loss of functionality [1][2][3]. This limits the application of adult stem cells for cell replacement therapy. Induced pluripotency (iP), a state where somatic differentiated cells become functionally similar to embryonic stem cells (ESC), may serve as an alternative solution. The breakthrough findings of iP, first discovered by Takahashi and Yamanaka in 2006, by ectopic overexpression of four stemness-related transcription factors (TFs: Oct3/4, Sox2, Klf4, and c-Myc; OSKM in short), in mouse fibroblasts [4], and then repeated in human fibroblasts [5], proved the plasticity potential of differentiated cells to rejuvenate back to the ESC-AGING like state. Since then, various combinations of transcription factors for iP have been proposed [6][7][8]. Still, the exogenous introduction of transgenes provides a low yield, both in vitro and in vivo, and may have undesirable complications, including tumorigenicity (reviewed by [3]). Recently, a number of small molecules (SMs) that are able to induce or enhance pluripotency have been discovered [9][10][11]. They have definite advantages and could be used for iP as a much safer alternative [12]. First of all, cell dedifferentiation activity could be finetuned by varying the concentrations of SM. When needed, the application of lineage-alternating SMs could induce cell differentiation and inhibit cell proliferation. Moreover, SMs are distinguished by nonimmunogenicity, cost-efficiency, minimal residual effects on the genome, and feasibility of in vivo application [13,14]. Consequently, this strategy may have great potential in clinical practice. With this in mind, the major goal of this study was to provide a systems biology view of the SMs, thus supporting researchers with a potential basis for the optimal selection of drugs for cell reprogramming. In this in silico study we performed: (i) a comprehensive data mining of SMs; (ii) the characterization of SMs and SM cocktails, including assessing their protein targets and possible interactions between them; (iii) the analysis of pathways targeted by SMs, (iv) the comparison of targets and pathways of SM cocktails with those of the OSKM TFs, and (v) screening for SMs as human metabolites. RESULTS General characterization of SMs and SM cocktails for cell reprogramming We first compiled a full list of SMs established thus far, based on a keyword meta-analysis of the literature. Comprehensive data mining with subsequent curation (see Methods) resulted in a total of 92 chemical compounds (Supplementary Table 1) that can either induce or enhance pluripotency, alone or in combination with TFs. These compounds for chemical reprogramming were named "Small Molecules" (SMs) because of their relatively low molecular weight [9], which ranges from 42.4 g/mol (LiCl) to 914.2 g/mol (Rapamycin). The vast majority of SMs represent organic compounds belonging to various chemical classes; however, among SMs were also several inorganic compounds (e.g., Lithium salts). The analysis of the basic biological activities of the collected SMs revealed that they fall into three major categories ( Figure 1 and Supplementary Tables 2-5): (i) signaling modifiers, (ii) epigenetic modifiers, and (iii) metabolic modifiers. It should also be mentioned that some SMs do not fall into definite categories or belong to more than one functional category. Table 4). The most "popular" AGING (i.e., most frequently used in SM cocktails) signaling modifiers include inhibitors of TGFβ and Hedgehog signaling, both involved in cell differentiation [15,16]. In the epigenetic category, most SMs inhibit either methyltransferases (HMTs and DNMTs, 9 and 6, respectively) or HDACs (n = 4). Other molecules possess either dual activity (HDAC inducers and/or inhibitors, n = 3) or combined (inhibition of HMT+DNMT or DNMT+HDAC) activities. This, respectively, shifts the condensed form of chromatin (heterochromatin) towards a relaxed state (euchromatin) or decreases the level of DNA methylation, thereby ensuring more DNA to be available for transcription. Lastly, metabolic modifiers switch the metabolism from oxidative phosphorylation towards glycolysis, mostly through the inhibition of the GSK3 enzyme [17]. Other SMs (n = 8; 8.7%; Supplementary Table 5) include antioxidants, regulators of calcium transport, autophagy, etc. To date, several combinations of SMs have been tested for cell reprogramming activity. Of them, 10 SM cocktails have been established. Their compositions, which vary from three [10] to ten [18] compounds, are presented in Supplementary Table 6. The common denominator for all these cocktails is that they are able to induce cell reprogramming, either full (pluripotent state) or partial (multipotent/progenitor cells), without transfection of stemness-related TFs. A comparison between the cocktails revealed 22 nonredundant chemicals, presented in Table 1. It should be emphasized that each cocktail contains at least one SM from each of the epigenetic, signaling or metabolic activity categories, which coincide well with the results presented above. Of note, TGFβ inhibitors are presented in all cocktails. In particular, RepSox, which can replace Sox2 [19], is included in 7 of the 10 cocktails, and in the other three, the TGFβ inhibitors are replaced by SB431542 or Tranilast, both able to replace Sox2 [10,19], or by A-83-01 [20]. Another frequently-used signaling modifier included Forskolin (found in six cocktails) or BrdU (in Cocktail 5). The mentioned compounds can replace Oct4 [9,21] (see Supplementary Table 1). The nuclear RARα selective agonist AM 580 and the synthetic retinoic acid receptor ligand TTNPB affecting the retinoic acid signaling pathway are used in four cocktails. As seen in Table 1, the GSK3 inhibitors (CHIR99021, LiCl or Li2CO3) which promote glycolysis are mandatory components of each reprogramming cocktail. Finally, all the cocktails include one or more epigenetic modifiers: HDAC inhibitors (VPA, NaB, Trichostatin A), DNMT inhibitors (5-aza-dC), the inhibitor of LSD1 acting on histone H3 (Parnate), and the inhibitors of histone methyltransferases (DZNep, EPZ004777, SGC0946). The common SMs are presented in the reprogramming cocktails in descending order: CHIR99021 = RepSox (n = 7), VPA = Forskolin (n = 6), Parnate (n = 5), DZNep (n = 4), AM 580 (n = 3), EPZ004777 (n = 2); other SMs are found only in one cocktail (see Table 1). KEGG pathways enrichment analysis of SM targets To get further insight into the mechanisms of chemically-induced reprogramming, we carried out an enrichment analysis for SM protein targets. For that purpose, we first used the STITCH database (https://pubmed.ncbi.nlm.nih.gov/26590256/) for extracting the chemical-protein interactions. Then, using the DAVID bioinformatics tools [22], we determined the enriched KEGG pathways of the found SM protein targets (in total, 1023). Figure 2 depicts the most enriched KEGG categories (p < 0.001 after Benjamini correction, with at least two-fold enrichment) among SM targets (for a full list of the enriched pathways, see Supplementary Table 7). The most significantly enriched KEGG pathways include pathways associated with regulation of longevity such as mTOR signaling (p = 9.1E-18), AMPK signaling (p = 5.7E-17), Insulin signaling (p = 2.3E-13), FoxO signaling (p = 4.2E-23), and pathways involved in cell-cell and cell-extracellular matrix interactions (Focal adhesion, p = 3.7E-13, Adherens junction, p = 5.4E-06). Also, SM targets are overpresented in the signaling pathways associated with age-related diseases, including different types of cancer, type II diabetes mellitus (p = 7.7E-08), amyotrophic lateral sclerosis (p = 2.3E-04), and Alzheimer's disease (p = 2.5E-03). Among the enriched pathways are numerous growth-promoting pathways, cell survival (PI3K-Akt, p = 3.9E-19) or cell death (Apoptosis, p = 1.7E-18) signaling. Many enriched pathways are related to immune and inflammatory responses. Among them are the pathways related to innate immunity (Toll-like receptor signaling pathway, p = 1.8E-09; NK-cell mediated cytotoxicity, p = 2.6E-06), specific immune responses (T cell receptor signaling pathway, p = 5.3E-15; B cell receptor signaling pathway, p = 2.1E-07), and inflammatory signaling (Chemokine signaling pathway, p = 1.5E-13; Adipocytokine signaling pathway, p = 2.1E-11), etc. Not surprisingly, the enriched pathways include regulation of cell cycle (p = 3.5E-09), cell differentiation (Neurotrophins, p = 3.3E-18; TGFβ signaling, p = 9.4E-06), and Signaling pathways regulating pluripotency of stem cells (p = 2.8E-06). Network analysis of SM targets To further evaluate to what extent the SM targets interact between themselves, we determined their AGING protein-protein interactions (PPIs), annotated in the BioGRID database [23]. These data are currently available for 991 out of 1023 SM target proteins. The analysis revealed that many of these targets interact with each other and exhibit multiple PPIs (in total, 6072 interactions). Remarkably, a significant fraction of the interacting SM targets (851 out of 991 proteins; 85.8%) forms a continuous network between themselves AGING ( Figure 3A). This fraction is significantly higher than expected by chance, i.e., higher than for the same number of randomly selected proteins with annotated PPIs ( Figure 3B) (random sampling, mean ± SD: 52.8 ± 3.5%; z-score for observed value: 9.37). Next, we aimed to understand the topology of the constructed network. To address this point, we calculated the distribution of node connectivity. The regression equation in Figure 3C (P(k) = 221 x k -1. 16 ) follows a power-law distribution of connectivity and AGING indicates that the PPI network of SM targets has a scalefree topology, with an extremely high contribution of hubs to the average network connectivity. Using the same approach, we built the chemical-protein interaction and PPI networks for the ten SM cocktails used thus far for chemical reprogramming (see Supplementary Table 6). As seen in Figure 4 and Supplementary Figures 1-9, the total number of annotated protein targets in SM cocktails varied from 6 (Cocktail 10) to 174 (Cocktail 7), mostly falling around 50. In all cases, the fraction of proteins forming a continuous PPI network was extremely high (from 25% to 75.9%) for such small sizes of protein sets ( Figure 3B), z-scores computed after random sampling being between 5.33 and 30. Collectively, the results obtained indicate that the SM targets are highly interconnected. Comparison of targets and pathways of SM cocktails with Yamanaka's factors It seems plausible that the cocktails for chemical cell reprogramming and TFs for iP, specifically Yamanaka's factors (OSKM), have common targets ( Figure 5A). However, their comparison showed that only the gene targets of Cocktail #7 (15 targets; p = 0.0033) overlap significantly with the targets of a "classical" combination of iP transcription factors ( Figure 5A). Other cocktails overlap insignificantly (p > 0.05) with OSKM. Of note, Cocktail #7 has much more targets than any other cocktail for chemical reprogramming. In contrast to specific targets, several cocktails (#2, 3, 4 and 7) have significantly overlapped pathways with OSKM ( Figure 5B). As seen in Table 2, most common pathways are cancer-related. Though not reaching the level of significance, the common pathways of other cocktails (#1, 5, 6, 8, 9 and 10) are also cancer-related. SMs as human metabolites Most SMs are artificially synthesized chemicals. Of special interest is whether among the SMs are compounds that are natural (human) metabolites or their analogs. Overlapping the 92 SMs with the molecules found in the Human Metabolome Database -HMDB [24] gives a positive answer to this question: 28 compounds from the SM list are also found in HMDB ( Table 3). The overlap is statistically extremely significant AGING (p = 9.7E-83). For example, among SMs are essential natural metabolites (n = 8) including several vitamins (A, C, D), molecules belonging to fatty acids and their derivatives (NaB, PGE2), organooxygen (Fru-2,6-P2) and organonitrogen (Spermidine) compounds, and prenol lipids (Retinoic acid). Other "natural" SMs represent nutrients that integrate into the human body when consuming products of plant metabolism (n = 11). Interestingly, several of these compounds (e.g. EGCG, 7-hydroxyflavone, apigenin, curcumin, quercetin, resveratrol) are components of plant extracts that have been already shown to improve healthspan, in particular stress resistance and cognitive abilities [25]. Several SMs are medications, which under specific conditions can be found in the human body. Although they are not the products of human metabolism or essential nutrients, most of them are analogs of natural metabolites. For example, 5'-azaC or 5'-Aza-2'-deoxycytidine are analogs of the nucleoside cytidine; N-acetyl-cysteine is metabolized into L-cysteine, a precursor to the biologic antioxidant glutathione; Valproic acid (VPA) is a branched shortchain fatty acid derived from the naturally occurring Valeric acid [26]. Furthermore, using STITCH tools [27], we found another 963 molecules that are similar (based on the STITCH drug similarity score) to the SMs that induce or enhance pluripotency, of them, 210 compounds (data not shown) are present in the Human Metabolome Database [24]. Among these compounds are neurotransmitters (serotonin, dopamine and GABA), fatty acids, and their derivatives involved in energy metabolism, such as citric acid, succinate and lactate. We determined the targets of these 210 chemicals, of the abovementioned eight human essential natural metabolites, and then compared them with the targets of all collected SMs (n = 1,023) and SM cocktails (n = 204) (Supplementary Table 10). As seen in the Supplementary Table 10, there is an extremely significant (p < E-25, Fisher test) overlap between the targets of the 210 SM-like chemicals (n = 4,614) and the targets of all SMs or the targets of SM cocktails. The common targets cover more than 76% (782 of 1023 targets) and 65% (132 out of 204 targets), respectively. Also, an extremely significant overlap was found for the targets of the abovementioned 8 human natural metabolites (n = 318) and the targets of SM cocktails (21%, 43 of 204). AGING DISCUSSION Until now, the selection of SMs for chemically-induced pluripotency or cell reprogramming was done mainly on an empirical basis, and no analysis of SMs and their targets has been undertaken. Several reviews published in the past [28][29][30][31][32] focused on specific aspects of SMs but none of them provided a "systemic" view. Our comprehensive data mining with subsequent data curation revealed 92 SMs that have been reported in connection to cell reprogramming. Most of the SMs were primarily used as enhancers of iP, i.e., for increasing the efficiency of cocktails containing TFs (e.g., Yamanaka's factors) [30,33,34]. Of note, to a lesser degree, SMs were also used as enhancers of cell reprogramming in SM cocktails without TFs. Apart from cell dedifferentiation, in the last years, SMs have also been used for cell transdifferentiation (for a review see Xie et al., 2017 [13]). Still, we found among the studied SMs many that could be classified as stand-alone inducers of cell reprogramming. These SMs were able to induce cellular reprogramming by themselves, thus either fully replacing the essential TFs [9,10] or by increasing their expression [35,36]. For example, Forskolin can replace Oct4, while RepSox can substitute Sox2 (see Supplementary Table 1). Besides the classical iP by means of the combinations of overexpressed TFs (e.g., Yamanaka's factors, OSKM), a total of ten cocktails that contain SMs only with cell reprogramming activity have been established and tested thus far. Functional analysis of SMs and their targets revealed that they are distributed between three major categories: epigenetics, intra-and inter-cellular signaling, and metabolic "switchers". All these categories appear to be mandatorily presented in each SM cocktail to induce cell reprogramming. Specifically, it seems that sufficient components for a "minimal reprogramming" cocktail have to include an inhibitor of HDAC (e.g. VPA or NaB), an inhibitor of TGFβ signaling (e.g. RepSox), and GSK3-inhibiting SMs (e.g. CHIR99021 or LiCl). This assumption was further confirmed by the KEGG pathways enrichment analysis. The unusually significant enrichment of epigenetic and signaling pathways highlights their importance in chemical iP. Remarkably, many enriched pathways were related to aging, longevity and age-related diseases, thus presumably connecting them with the processes of cell reprogramming. This notion has recently been supported experimentally by demonstrating induction of cellular senescence by activation of OSKM, in vitro [37] and also in vivo on i4F reprogrammable mice [38][39][40]. Yet, this does not minimize the potential importance of pathways that are only slightly enriched or are not enriched at all. For example, Glycolysis/Gluconeogenesis pathway appears in our analysis as a marginally significantly enriched pathway (p = 0.051), although it is a well-recognized metabolic pathway for cell reprogramming; moreover, it is well known that the pluripotent stem cells rely on glycolysis rather than OXPHOS (reviewed by [3]). The possible explanation for this result is most likely related to the small number of glycolytic enzymes among the SM targets, relative to the total number of targets. Further strengthening the importance of metabolic components of iP is the observation that the HIF-1 signaling pathway is among the most significantly enriched pathways (fold AGING 20). Indeed, the hypoxiainducible factor 1 alpha (Hif1alpha) activates glycolysis and concomitantly promotes telomerase expression and enhances self-renewal of stem cells [41]. Another important observation is that the main transcription factors of pluripotency, Oct4 and Nanog, can directly induce expression of the key glycolytic enzymes hexokinase 2 and pyruvate kinase M2, thus delaying differentiation and preserving pluripotency of ESCs [42]. In turn, the genes involved in the control of glucose uptake (GLUT3) and metabolism (PKM2) are also involved in the regulation of Oct4 expression [43]. For unclear reasons, some promising SMs have not been used in reprogramming cocktails developed thus far. For example, vitamin C (see Table 3 and Supplementary Tables 1, 5) was shown to modulate the TET enzymes, which promotes demethylation of histones and DNA, with subsequent enhancing cell reprogramming induced AGING by OSKM [44][45][46], however it was not yet evaluated in combination with any SM cocktail. It is still a matter of debate whether SMs act independently of each other in triggering cell reprogramming, or if they act in a cooperative, epistatic manner. The latter suggests the interactions between their targets, including direct (physical) interactions. With this in mind, we analyzed the connectivity and interconnectivity of targets of SMs and SM cocktails. The network analysis indicates that their targets are highly interconnected and form PPI networks with a scale-free topology that confers robustness and persistent connectivity. This means that: (i) the SM targets probably act in a cooperative manner to induce cell reprogramming; (ii) a scale-free topology of SM targets ensures higher integrity of the network and its resistance to random attacks [47,48], thus making the cell reprogramming process highly reliable. Recently, we hypothesized that cell reprogramming is a natural process that is triggered and regulated via two major networksa genetic one (triggered by transcription factors, e.g. OSKM) and a chemical one (controlled by metabolites, e.g. similar to SMs) [3,49]. In line with this hypothesis are our data demonstrating that: (i) a large number of SMs (28 of 92; Table 3) used for cell reprogramming are found in the human metabolome (derivatives of nucleotides, fatty acids, etc.), and (ii) many more metabolites (over 200) are functionally similar to SMs, thus offering the potential of being cell reprogramming agents. In addition to the chemical factors, environmental factors such as hypoxia and/or hypercapnia (which eventually act as chemical factors, namely through low concentrations of oxygen and high concentrations of carbon dioxide) may greatly influence the cell dedifferentiation process [3,50]. It should be mentioned again (see above) that hypoxic/hypercapnic microenvironment associated with a low reactive oxygen species (ROS) generation and activation of glycolysis, is essential for maintenance and proper functioning of dedifferentiated cells. Further supporting our hypothesis are the data on the common targets of SM cocktails and Yamanaka's factors. This comparison revealed an insignificant overlap between the SM cocktails' targets and OSKM, except for Cocktail #7. The lack of common targets between the cocktails and Yamanaka factors was quite a surprising observation. More prominent overlap was however observed between pathways, meaning that despite different targets, both SM cocktails and Yamanaka's factors "use" more or less the same pathways. Altogether, this suggests that the two systems, chemical (SMs) and genetic (TFs), might cooperate to increase the efficiency of cell reprogramming. Interestingly, the overlapping pathways for SM cocktails and OSKM targets are mainly cancer-or virus-related but not related to key reprogramming processes, such as demethylation and chromatin decondensation or pluripotency pathways, as it might be expected. One of the reasons could be rooted in statistical issues. In Table 2, only the pathways significantly overlapping with at least one SM cocktail, are presented. Another important point is that cancer-related pathways are not "purely" cancer pathways, but include many components related to cell division and reprogramming. For example, Wnt/β-catenin and MAPK signaling pathways are known for their role in cell dedifferentiation [51,52]. These pathways are also well known for their involvement in carcinogenesis [53]. Although beyond the scope of the present study, it is worth mentioning that there is a significant overlap between the collected 92 SMs and the compounds found in the DrugAge database [54] (n = 20 drugs; p = 4.95E-15). Among the common drugs are Rapamycin, Valproic acid, Caffeic acid, and Lithium chloride. Similarly, there is a large overlap between the SM targets and the longevity-associated genes (LAGs) hosted in the GenAge database [55] (n = 132, p = 3E-88 for human LAGs and n = 136, p = 5E-24 for human orthologs of model organism LAGs). Lastly, SM targets also overlap with the list of genes related to cellular senescence (CS) from the CellAge database [2] (n = 85, p = 1E-42). As a point for further investigation is testing the established or newly constructed SM cocktails in vivo. In this regard, testing SM cocktails in the naked mole-rat model could be of particular interest as induction of pluripotency in the cells of this animal requires special conditions and is not always achievable [56][57][58]. All in all, SMs and their relationship with TFs definitely warrants further investigation which could probably shed more light on the mechanisms of cell reprogramming and will be helpful for developing the most optimal SM cocktails with effects on CS, aging and longevity. MATERIALS AND METHODS Data sources Data AGING HMDB contains the collection of small molecules found in the human body, including nucleic acids, carbohydrates, lipids, peptides, amino acids, organic acids, biogenic amines, vitamins, minerals, food additives, drugs, cosmetics, contaminants, pollutants, and other chemicals that enter the human body [24]. Data mining and organization The papers were searched using the following keywords: "induced pluripotency", "chemically induced pluripotency", "chemical reprogramming", "chemically induced dedifferentiation", "induction of pluripotency by small molecules". In order to be included in the analysis, each article had to contain data: (i) on SM(s) or their cocktail(s) that either induced or enhanced cellular reprogramming; (ii) on the bioactivity of the SMs; and (iii) on the SM dosage and cell type. According to their role in cell reprogramming, the compounds found were divided into two major groups of molecules: iP inducers and iP enhancers. Since it was not always possible to definitely link the compounds to one of the groups, as in some cases a given compound was considered an inducer and in other cases an enhancer, these entities were marked as "inducer and/or enhancer". From each paper the following data were collected and manually curated: (i) the name(s) of SM(s) that either induce or enhance pluripotency, with or without TFs; (ii) the effect of SM(s) on the iP efficiency; and (iii) whether a given SM can substitute the pluripotency-associated TFs. The collected SMs were organized in a table as shown in Supplementary Tables 1, 6. The data regarding each compound included: common name, formula, molecular weight (MW), main bioactivity/target(s), comments relevant to cellular reprogramming, link to PubChem references, PMID. Only the SM cocktails which induced cell reprogramming (not necessary to the stage of iPSCs) without TFs were included in the analysis. Drug-protein interaction network To determine the protein targets of the collected SMs, we used the STITCH database (version 5.0), http://stitch.embl.de/, one of the largest repositories of chemical-protein interactions [27], which include direct (physical) and indirect (functional) interactions. For the scope of the analyses in this study, text-mining and predicted interactions were excluded. If not indicated otherwise, a confidence score of medium stringency (0.4) was used for including interaction in the analysis. Drug similarity analysis was performed using the STITCH tool as described by Kuhn et al. [60]. Gene targets overlap To obtain the list of OSKM transcription factors the TRRUST database [61], https://www.grnpedia.org/tr rust/, was used. The overlaps between gene targets of drug cocktails and OSKM transcription factors were calculated using only the genes that are present in both STITCH and TRRUST databases. In order to compute the overlap between gene targets of SMs and GenAge [55], https://genomics.senescence.info/genes/index.html, two lists of longevity-associated genes (LAGs) were used: i) the manually curated list of human LAGs from GenAge, build 20 and ii) the human orthologs of model organisms LAGs from GenAge, build 20. Orthologs of genes were computed using a script developed in our lab, that queries the database InParanoid 8 [62], https://in paranoid.sbc.su.se/cgi-bin/index.cgi. For stringency, we selected for each gene only inparalogs with scores of 1.0. The significance of the overlaps with GenAge [55] and CellAge [2] -https://genomics.senescence.info/cells/, was computed using Fisher's exact test. SMs overlap with chemical databases The overlaps between: i) the list of SMs and HMDB, and ii) the list of SMs and DrugAge [54] were calculated using the PubChem IDs of the compounds as identifiers. The significance of the overlap was computed using Fisher's exact test and considering all PubChem and all DrugBank compounds, respectively, as background. KEGG pathways and gene ontology enrichment analysis Functional and pathway enrichment analyses were performed with the DAVID Bioinformatics Resources tool, version 6.8 [22], https://david.ncifcrf.gov. Statistical significance of enrichment was evaluated using default parameters set in DAVID. A threshold of 0.001 was used for the adjusted P-value. Protein-protein interaction networks Protein-protein interaction (PPI) data were taken from the BioGRID database [23], http://thebiogrid.org, human interactome, Build 3.5.177. The PPI network construction and analyses were performed using Cytoscape [63], http://www.cytoscape.org, version 3.7.1. Prior to any network analyses, genetic interactions, self-loops, duplicate edges and interactions with proteins from other species were removed from the interactome, and the remaining network was used as a control. The interconnectivity was computed as the fraction of nodes in the largest connected component out of the input gene set, by using the breadth-first AGING search algorithm. Modeling the relationship between node subset size and interconnectivity in the human interactome was carried out by randomly sampling subsets of nodes in the interactome, with a sample size varying from 50 to 17,600 nodes (step of 50). In this case, sampling was performed 100 times for each subset size. In order to evaluate the statistical significance of the observed network interconnectivity for cocktails and SMs gene targets, random sampling from the BioGRID network was performed 1000 times, for a subset of nodes of equal size to each evaluated network. For each set of random samplings, average interconnectivity, standard deviation and z-score of the observed interconnectivity were computed. For a joint protein-drug network, the protein targets of the collected SMs, determined from the STITCH database, were used together with PPIs from BioGRID. Abbreviations for SM cocktails Cocktail 3 (VCR) Valproic acid CHIR99021 RepSox Cocktail 4 (TLT) Trichostatin A (TSA) Li2CO3 Tranilast AUTHOR CONTRIBUTIONS This study was carried out by the VEF and RT research groups. Data collection, processing, analysis of the result and their description were done by AK and GB. Interpretation of the results was done by all authors. VEF and RT coordinated and supervised the project. All authors have participated in the writing of the manuscript. All authors reviewed the manuscript. CONFLICTS OF INTEREST The authors declare that they have no conflicts of interest. Figure 1 . 1Distribution of SMs by functional categories. The basic biological activities of all SMs that induce or enhance pluripotency (n = 92) were extracted from the STITCH online tool, PubChem database and scientific literature. Functional categories of SMs were based on Gene Ontology Resource. Figure 2 . 2Top enriched KEGG pathways of SM protein targets. Enriched pathways at high confidence (p < 0.001 after Benjamini correction, with at least two-fold enrichment) are presented. Because of visualization limitations, only the top-most enriched 50 pathways are included in the figure. For a full list of the enriched pathways, see Supplementary AGINGFigure 3 . 3(A) Graphical output of the PPI network of the entire set of SMs' targets. (B) Simulation of expected interconnectivity given the size of a random sample. The observed interconnectivity of SMs' gene targets in the interactome, depicted by the red dot in the scatter plot and the observed interconnectivity of cocktails' gene targets, depicted by the orange dots, can be compared to the percentage of interconnected nodes (on the Y-axis), found in the largest continuous component of the network, for randomly sampled node sets. The plot shows the sampling of subsets of random interactome nodes, of various sizes (represented in a log10 scale on the X-axis, from 50 to 17,600 nodes). For each step, the interconnectivity was computed 100 times. Simulations were performed only for samples larger than 50 nodes, because of the increased variability of very small node sets. (C) The log-log plot of P(k) against k, illustrating scale-free topology of the network (for details, see the text and Methods). For all the nodes and edges in the network seeSupplementary Table 9. (A, C) The construction and display of the network and the degree distribution regression were performed using Cytoscape, which pulls physical PPIs data determined in vitro and in vivo from the BioGRID database. Figure 4 . 4The network with the highest interconnectivity (corresponding to the TLT cocktail). In total, 58 protein targets are in the network. Continuous network without taking into account drug connectivity (chemical-protein interactions) includes 44 genes/proteins (75.9%; values for random sampling (mean ± SD): 4.5 ± 2.4; z-score for observed value: 30.03). Figure 5 . 5(A) Venn diagram of the gene targets of OSKM significantly overlapping with gene targets of cocktails. (B) Venn diagram of significantly overlapping enriched pathways for gene targets of SM cocktails and of OSKM. In order to simplify the figure, only statistically significant overlaps between OSKM and cocktails are displayed. Overlaps between pairs of cocktails are not shown. Gallate; ESCs: Embryonic stem cells; EZH: Enhancer of Zeste Homologue; Fru-2,6-P2: Fructose 2,6bisphosphate; GSK3: Glycogen synthase kinase 3; HDAC: Histone deacetylase; HIF: Hypoxia-inducible factor-1; Hif1alpha: Hypoxia-inducible factor 1 alpha; HMDB: Human Metabolome Database; HMT: Histone methyltransferase; IBMX: 3-Isobutyl-1-Methylxanthine; iP: Induced pluripotency; LAGs: Longevity-associated genes; MW: Molecular weight; O4I3: OCT4-inducing compound 3; OSKM: Oct3/4, Sox2, Klf4, and c-Myc (Yamanaka's factors); PDK1: 3′-phosphoinositidedependent kinase-1; PFK-1: Phosphofructokinase 1; PI3K: Phosphoinositide 3-kinase; PPIs: Protein-protein interactions; ROS: Reactive oxygen species; SAH: S-Adenosyl-l-homocysteine; SAHA: Suberoylanilide hydroxamic acid; SMs: Small molecules; TFs: Transcription factors. FUNDING This work was supported by the National Authority for Scientific Research and Innovation, and by the Ministry of European Funds, Romania, through the Competitiveness Operational Programme 2014-2020, POC-A.1-A.1.1.4-E-2015 [Grant number: 40/02.09.2016, ID: P_37_778, to RT] and by the Romanian Ministry of Education and Research, CCCDI -UEFISCDI, through PNCDI III [Grant number: PN-III-P2-2.1-PED-2019-2593 to RT]. We are also grateful for the funding received from the Dr. Amir Abramovich Research Fund [granted to VEF]. The SMs with signaling activity represent the largest group(51 out of 92 compounds; 55.4%; Supplementary Table 2), followed by epigenetic (n = 26; 28.3%; Supplementary Table 3) and metabolic modifiers (n = 7; 7.6%; Supplementary Table 1 . 1Non-redundant SMs for reprogramming cocktails and their main bioactivities.SM Main bioactivity Cocktail 1 2 3 4 5 6 7 8 9 10 CHIR99021 GSK3 inhibitor RepSox TGFβ inhibitor [can replace Sox2] VPA HDAC inhibitor Forskolin cAMP activator [can replace Oct4] Parnate Inhibitor of LSD1 acting on histone H3 DZNep Inhibitor of HMT EZH and SAH synthesis AM 580 Nuclear RARα selective agonist EPZ004777 DOT1L histone (H3K79) methyltransferase inhibitor NaB HDAC inhibitor TTNPB Synthetic retinoic acid receptor ligand BrdU Synthetic analog of thymidine [can replace Oct4] LiCl GSK3 inhibitor SB431542 TGFβ inhibitor [can replace RepSox] Tranilast TGFβ inhibitor [can replace RepSox] Trichostatin A HDAC inhibitor Li2CO3 GSK3 inhibitor 5'-aza-dC DNMT inhibitor SGC0946 DOT1L histone (H3K79) methyltransferase inhibitor Cyclic pifithrin-a p53 inhibitor A-83-01 TGF-beta receptor inhibitor Thiazovivin Rho Kinase (ROCK) inhibitor PD0325901 Potent MKK1 (MEK1) and MKK2 (MEK2) inhibitor Table 7 , 7and for the enriched pathways for each SM cocktail, seeSupplementary Table 8. Table 2 . 2Overlapping pathways for targets of SM cocktails and OSKM.Pathways Cocktails #1 #2 #3 #4 #5 #6 #7 #8 #9 #10 Pathways in cancer * * * * Chronic myeloid leukemia * * * * Prostate cancer * * * Bladder cancer * Small cell lung cancer * Viral carcinogenesis * * * HTLV-I infection * * * Hepatitis B * * * Epstein-Barr virus infection * p53 signaling pathway * Dark gray color with (*) depicts overlaps with p < 0.05. Light gray depicts the pathways with insignificant overlaps (p > 0.05). Table 3 . 3SMs as human metabolites.Name Role in induced cell reprogramming (inducer/enhancer) Chemical class 5'-Azacytidine (5'-azaC) Enhancer Nucleotides and nucleotide derivatives 5'-Aza-2'-deoxycytidine Enhancer Nucleotides and nucleotide derivatives 7-hydroxyflavone Enhancer Flavonoids 90-D3 (Vitamin D3) Enhancer Steroids and steroid derivatives Apigenin Enhancer Flavonoids Caffeic acid Putative enhancer or inducer Cinnamic acids and derivatives Chlorogenic acid Putative enhancer or inducer Fatty acids and derivatives Curcumin Enhancer Diarylheptanoids Dasatinib Inducer Benzene and derivatives Dexamethasone Enhancer Steroids and steroid derivatives EGCG Enhancer Flavonoids Fisetin Enhancer Flavonoids Forskolin Inducer Benzofurans Fru-2,6-P2 Enhancer Organooxygen compounds Luteolin Enhancer Flavonoids N-acetyl-cysteine Enhancer Amino acids and derivatives Sodium Butyrate (NaB) Inducer and enhancer Fatty acids and derivatives Prostaglandin E2 Enhancer Fatty acids and derivatives Quercetin Enhancer Flavonoids Rapamycin Enhancer Macrolide lactams Resveratrol Enhancer Stilbenes Retinoic acid Enhancer Prenol lipids SAHA Enhancer Benzene and derivatives Spermidine Enhancer Organonitrogen compounds Valproic acid Inducer Fatty acids and derivatives Vitamin A (Retinol acetate) Enhancer Prenol lipids Vitamin C (Ascorbic acid; Ascorbate) Enhancer Dihydrofurans Zolpidem Enhancer Azoles enrichment = 5, p < 2.0E- Cocktail 5 (BrdUC6F)Cocktail 6 (VC6TF + AM 580 + EPZ004777) Cocktail 7 (VC6TF + AM580 + DZNep + 5-aza-dC + SGC0946 + EPZ004777) SMs and their protein targets.Cocktail 9 (VC6TF + AM 580 + DZNep)BrdU CHIR99021 RepSox Forskolin VPA CHIR99021 RepSox Parnate Forskolin AM 580 EPZ004777 VPA CHIR99021 RepSox Parnate Forskolin AM 580 DZNep 5-aza-dC SGC0946 EPZ004777 Cocktail 8 (VC6TF + DZNep) VPA CHIR99021 AGING RepSOX Parnate Forskolin DZNep VPA CHIR99021 RepSox Parnate Forskolin AM 580 DZNep Cocktail 10 (CNɑATP) CHIR99021 NaB cyclic pifithrin-a (ɑ) A-83-01 Thiazovivin PD0325901 The role of DNA damage and repair in aging through the prism of Koch-like criteria. A A Moskalev, M V Shaposhnikov, E N Plyusnina, A Zhavoronkov, A Budovsky, H Yanai, V E Fraifeld, 10.1016/j.arr.2012.02.00122353384Ageing Res Rev. 12Moskalev AA, Shaposhnikov MV, Plyusnina EN, Zhavoronkov A, Budovsky A, Yanai H, Fraifeld VE. The role of DNA damage and repair in aging through the prism of Koch-like criteria. Ageing Res Rev. 2013; 12:661-84. https://doi.org/10.1016/j.arr.2012.02.001 PMID:22353384 A multidimensional systems biology analysis of cellular senescence in aging and disease. R A Avelar, J G Ortega, R Tacutu, E J Tyler, D Bennett, P Binetti, A Budovsky, K Chatsirisupachai, Johnson E Murray, A Shields, S Tejada-Martinez, D Thornton, D , 10.1186/s13059-020-01990-9Genome Biol. 2132264951Avelar RA, Ortega JG, Tacutu R, Tyler EJ, Bennett D, Binetti P, Budovsky A, Chatsirisupachai K, Johnson E, Murray A, Shields S, Tejada-Martinez D, Thornton D, et al. A multidimensional systems biology analysis of cellular senescence in aging and disease. Genome Biol. 2020; 21:91. https://doi.org/10.1186/s13059-020-01990-9 PMID:32264951 iPSCs-Induced Cellular Reprogramming. Encyclopedia of Biomedical Gerontology. K K Muradian, V E Fraifeld, Academic PressMuradian KK, Fraifeld VE. iPSCs-Induced Cellular Reprogramming. Encyclopedia of Biomedical Gerontology. Academic Press. 2020; 291-8. . 10.1016/B978-0-12-801238-3.11299-1https://doi.org/10.1016/B978-0-12-801238-3.11299-1 Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors. K Takahashi, S Yamanaka, 10.1016/j.cell.2006.07.02416904174Cell. 126Takahashi K, Yamanaka S. Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors. Cell. 2006; 126:663-76. https://doi.org/10.1016/j.cell.2006.07.024 PMID:16904174 Induction of pluripotent stem cells from adult human fibroblasts by defined factors. K Takahashi, K Tanabe, M Ohnuki, M Narita, T Ichisaka, K Tomoda, S Yamanaka, 10.1016/j.cell.2007.11.01918035408Cell. 131Takahashi K, Tanabe K, Ohnuki M, Narita M, Ichisaka T, Tomoda K, Yamanaka S. Induction of pluripotent stem cells from adult human fibroblasts by defined factors. Cell. 2007; 131:861-72. https://doi.org/10.1016/j.cell.2007.11.019 PMID:18035408 The developmental potential of iPSCs is greatly influenced by reprogramming factor selection. Y Buganim, S Markoulaki, N Van Wietmarschen, H Hoke, T Wu, K Ganz, B Akhtar-Zaidi, Y He, B J Abraham, D Porubsky, E Kulenkampff, D A Faddah, L Shi, Cell Stem Cell. 15Buganim Y, Markoulaki S, van Wietmarschen N, Hoke H, Wu T, Ganz K, Akhtar-Zaidi B, He Y, Abraham BJ, Porubsky D, Kulenkampff E, Faddah DA, Shi L, et al. The developmental potential of iPSCs is greatly influenced by reprogramming factor selection. Cell Stem Cell. 2014; 15:295-309. . 10.1016/j.stem.2014.07.00325192464https://doi.org/10.1016/j.stem.2014.07.003 PMID:25192464 Induced pluripotent stem cell lines derived from human somatic cells. J Yu, M A Vodyanik, K Smuga-Otto, Antosiewicz- Bourget, J Frane, J L Tian, S Nie, J Jonsdottir, G A Ruotti, V Stewart, R Slukvin, I I Thomson, J A , 10.1126/science.115152618029452Science. 318Yu J, Vodyanik MA, Smuga-Otto K, Antosiewicz-Bourget J, Frane JL, Tian S, Nie J, Jonsdottir GA, Ruotti V, Stewart R, Slukvin II, Thomson JA. Induced pluripotent stem cell lines derived from human somatic cells. Science. 2007; 318:1917-20. https://doi.org/10.1126/science.1151526 PMID:18029452 Enhanced efficiency of generating induced pluripotent stem (iPS) cells from human somatic cells by a combination of six transcription factors. J Liao, Z Wu, Y Wang, L Cheng, C Cui, Y Gao, T Chen, L Rao, S Chen, N Jia, H Dai, S Xin, J Kang, 10.1038/cr.2008.5118414447Cell Res. 18Liao J, Wu Z, Wang Y, Cheng L, Cui C, Gao Y, Chen T, Rao L, Chen S, Jia N, Dai H, Xin S, Kang J, et al. Enhanced efficiency of generating induced pluripotent stem (iPS) cells from human somatic cells by a combination of six transcription factors. Cell Res. 2008; 18:600-3. https://doi.org/10.1038/cr.2008.51 PMID:18414447 . P Hou, Y Li, X Zhang, C Liu, J Guan, H Li, T Zhao, J Ye, Hou P, Li Y, Zhang X, Liu C, Guan J, Li H, Zhao T, Ye J, www.aging-us.com AGING Pluripotent stem cells induced from mouse somatic cells by smallmolecule compounds. W Yang, K Liu, J Ge, J Xu, Q Zhang, 10.1126/science.123927823868920Science. 341Yang W, Liu K, Ge J, Xu J, Zhang Q, et al. Pluripotent stem cells induced from mouse somatic cells by small- molecule compounds. Science. 2013; 341:651-4. https://doi.org/10.1126/science.1239278 PMID:23868920 Generation of neural progenitor cells by chemical cocktails and hypoxia. L Cheng, W Hu, B Qiu, J Zhao, Y Yu, W Guan, M Wang, W Yang, G Pei, 10.1038/cr.2014.3224638034Cell Res. 24Cheng L, Hu W, Qiu B, Zhao J, Yu Y, Guan W, Wang M, Yang W, Pei G. Generation of neural progenitor cells by chemical cocktails and hypoxia. Cell Res. 2014; 24:665-79. https://doi.org/10.1038/cr.2014.32 PMID:24638034 Pluripotent stem cells induced from mouse neural stem cells and small intestinal epithelial cells by small molecule compounds. J Ye, J Ge, X Zhang, L Cheng, Z Zhang, S He, Y Wang, H Lin, W Yang, J Liu, Y Zhao, H Deng, Cell Res. 26Ye J, Ge J, Zhang X, Cheng L, Zhang Z, He S, Wang Y, Lin H, Yang W, Liu J, Zhao Y, Deng H. Pluripotent stem cells induced from mouse neural stem cells and small intestinal epithelial cells by small molecule compounds. Cell Res. 2016; 26:34-45. . 10.1038/cr.2015.14226704449https://doi.org/10.1038/cr.2015.142 PMID:26704449 The triumph of chemically enhanced cellular reprogramming: a patent review. M A Anwar, S Kim, S Choi, 10.1517/13543776.2016.111805826593376Expert Opin Ther Pat. 26Anwar MA, Kim S, Choi S. The triumph of chemically enhanced cellular reprogramming: a patent review. Expert Opin Ther Pat. 2016; 26:265-80. https://doi.org/10.1517/13543776.2016.1118058 PMID:26593376 Chemical reprogramming and transdifferentiation. X Xie, Y Fu, J Liu, 10.1016/j.gde.2017.07.00328755566Curr Opin Genet Dev. 46Xie X, Fu Y, Liu J. Chemical reprogramming and transdifferentiation. Curr Opin Genet Dev. 2017; 46:104-13. https://doi.org/10.1016/j.gde.2017.07.003 PMID:28755566 Chemical-induced cardiac reprogramming in vivo. C Huang, W Tu, Y Fu, J Wang, X Xie, 10.1038/s41422-018-0036-4Cell Res. 2829670223Huang C, Tu W, Fu Y, Wang J, Xie X. Chemical-induced cardiac reprogramming in vivo. Cell Res. 2018; 28:686-9. https://doi.org/10.1038/s41422-018-0036-4 PMID:29670223 The role of the Hedgehog signaling pathway in cancer: A comprehensive review. A M Skoda, D Simovic, V Karin, V Kardum, S Vranic, L Serman, 10.17305/bjbms.2018.275629274272Bosn J Basic Med Sci. 18Skoda AM, Simovic D, Karin V, Kardum V, Vranic S, Serman L. The role of the Hedgehog signaling pathway in cancer: A comprehensive review. Bosn J Basic Med Sci. 2018; 18:8-20. https://doi.org/10.17305/bjbms.2018.2756 PMID:29274272 Interplay between BMPs and Reactive Oxygen Species in Cell Signaling and Pathology. C Sánchez-De-Diego, J A Valer, C Pimenta-Lopes, J L Rosa, F Ventura, 10.3390/biom910053431561501Biomolecules. 9534Sánchez-de-Diego C, Valer JA, Pimenta-Lopes C, Rosa JL, Ventura F. Interplay between BMPs and Reactive Oxygen Species in Cell Signaling and Pathology. Biomolecules. 2019; 9:534. https://doi.org/10.3390/biom9100534 PMID:31561501 Small Molecule Inhibition of Glycogen Synthase Kinase-3 in Cancer Immunotherapy. A Taylor, C E Rudd, 10.1007/978-3-030-22254-3_1731576552Adv Exp Med Biol. 1164Taylor A, Rudd CE. Small Molecule Inhibition of Glycogen Synthase Kinase-3 in Cancer Immunotherapy. Adv Exp Med Biol. 2019; 1164:225-33. https://doi.org/10.1007/978-3-030-22254-3_17 PMID:31576552 A XEN-like State Bridges Somatic Cells to Pluripotency during Chemical Reprogramming. Y Zhao, T Zhao, J Guan, X Zhang, Y Fu, J Ye, J Zhu, G Meng, J Ge, S Yang, L Cheng, Y Du, C Zhao, 10.1016/j.cell.2015.11.01726686652Cell. 163Zhao Y, Zhao T, Guan J, Zhang X, Fu Y, Ye J, Zhu J, Meng G, Ge J, Yang S, Cheng L, Du Y, Zhao C, et al. A XEN-like State Bridges Somatic Cells to Pluripotency during Chemical Reprogramming. Cell. 2015; 163:1678-91. https://doi.org/10.1016/j.cell.2015.11.017 PMID:26686652 A small-molecule inhibitor of Tgf-Beta signaling replaces sox2 in reprogramming by inducing nanog. J K Ichida, J Blanchard, K Lam, E Y Son, J E Chung, D Egli, K M Loh, A C Carter, Di Giorgio, F P Koszka, K Huangfu, D Akutsu, H Liu, D R , 10.1016/j.stem.2009.09.01219818703Cell Stem Cell. 5Ichida JK, Blanchard J, Lam K, Son EY, Chung JE, Egli D, Loh KM, Carter AC, Di Giorgio FP, Koszka K, Huangfu D, Akutsu H, Liu DR, et al. A small-molecule inhibitor of Tgf-Beta signaling replaces sox2 in reprogramming by inducing nanog. Cell Stem Cell. 2009; 5:491-503. https://doi.org/10.1016/j.stem.2009.09.012 PMID:19818703 Optimized Approaches for Generation of Integration-free iPSCs from Human Urine-Derived Cells with Small Molecules and Autologous Feeder. D Li, L Wang, J Hou, Q Shen, Q Chen, X Wang, J Du, X Cai, Y Shan, T Zhang, T Zhou, X Shi, Y Li, 10.1016/j.stemcr.2016.04.00127132887Stem Cell Reports. 6Li D, Wang L, Hou J, Shen Q, Chen Q, Wang X, Du J, Cai X, Shan Y, Zhang T, Zhou T, Shi X, Li Y, et al. Optimized Approaches for Generation of Integration-free iPSCs from Human Urine-Derived Cells with Small Molecules and Autologous Feeder. Stem Cell Reports. 2016; 6:717-28. https://doi.org/10.1016/j.stemcr.2016.04.001 PMID:27132887 Bromodeoxyuridine promotes full-chemical induction of mouse pluripotent stem cells. Y Long, M Wang, H Gu, X Xie, 10.1038/cr.2015.9626251165Cell Res. 25Long Y, Wang M, Gu H, Xie X. Bromodeoxyuridine promotes full-chemical induction of mouse pluripotent stem cells. Cell Res. 2015; 25:1171-4. https://doi.org/10.1038/cr.2015.96 PMID:26251165 Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. D W Huang, B T Sherman, R A Lempicki, 10.1038/nprot.2008.21119131956Nat Protoc. 4Huang DW, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. 2009; 4:44-57. https://doi.org/10.1038/nprot.2008.211 PMID:19131956 The BioGRID interaction database: 2019 update. R Oughtred, C Stark, B J Breitkreutz, J Rust, L Boucher, C Chang, N Kolas, O&apos; Donnell, L Leung, G Mcadam, R Zhang, F Dolma, S Willems, A , 10.1093/nar/gky107930476227Nucleic Acids Res. 47Oughtred R, Stark C, Breitkreutz BJ, Rust J, Boucher L, Chang C, Kolas N, O'Donnell L, Leung G, McAdam R, Zhang F, Dolma S, Willems A, et al. The BioGRID interaction database: 2019 update. Nucleic Acids Res. 2019; 47:D529-41. https://doi.org/10.1093/nar/gky1079 PMID:30476227 . D S Wishart, Y D Feunang, A Marcu, A C Guo, K Liang, R Vázquez-Fresno, T Sajed, D Johnson, C Li, N Karu, Z Sayeeda, E Lo, N Assempour, 10.1093/nar/gkx108929140435Nucleic Acids Res. 46HMDB 4.0: the human metabolome databaseWishart DS, Feunang YD, Marcu A, Guo AC, Liang K, Vázquez-Fresno R, Sajed T, Johnson D, Li C, Karu N, Sayeeda Z, Lo E, Assempour N, et al. HMDB 4.0: the human metabolome database for 2018. Nucleic Acids Res. 2018; 46:D608-17. https://doi.org/10.1093/nar/gkx1089 PMID:29140435 Natural products improve healthspan in aged mice and rats: A systematic review and meta-analysis. C Musillo, M Borgi, N Saul, S Möller, W Luyten, A Berry, F Cirulli, 10.1016/j.neubiorev.2020.12.00133309907Neurosci Biobehav Rev. 121Musillo C, Borgi M, Saul N, Möller S, Luyten W, Berry A, Cirulli F. Natural products improve healthspan in aged mice and rats: A systematic review and meta-analysis. Neurosci Biobehav Rev. 2021; 121:89-105. https://doi.org/10.1016/j.neubiorev.2020.12.001 PMID:33309907 Valproic acid pathway: pharmacokinetics and pharmacodynamics. Y Ghodke-Puranik, C F Thorn, J K Lamba, J S Leeder, W Song, A K Birnbaum, R B Altman, T E Klein, 10.1097/FPC.0b013e32835ea0b223407051Pharmacogenet Genomics. 23Ghodke-Puranik Y, Thorn CF, Lamba JK, Leeder JS, Song W, Birnbaum AK, Altman RB, Klein TE. Valproic acid pathway: pharmacokinetics and pharmacodynamics. Pharmacogenet Genomics. 2013; 23:236-41. https://doi.org/10.1097/FPC.0b013e32835ea0b2 PMID:23407051 STITCH 5: augmenting protein-chemical interaction networks with tissue and affinity data. D Szklarczyk, A Santos, C Von Mering, L J Jensen, P Bork, M Kuhn, Nucleic Acids Res. 44Szklarczyk D, Santos A, von Mering C, Jensen LJ, Bork P, Kuhn M. STITCH 5: augmenting protein-chemical interaction networks with tissue and affinity data. Nucleic Acids Res. 2016; 44:D380-4. . 10.1093/nar/gkv127726590256https://doi.org/10.1093/nar/gkv1277 PMID:26590256 Small molecules for cell reprogramming and heart repair: progress and perspective. M Xie, N Cao, S Ding, 10.1021/cb400865w24372513ACS Chem Biol. 9Xie M, Cao N, Ding S. Small molecules for cell reprogramming and heart repair: progress and perspective. ACS Chem Biol. 2014; 9:34-44. https://doi.org/10.1021/cb400865w PMID:24372513 Selected small molecules as inducers of pluripotency. M Baranek, W T Markiewicz, J Barciszewski, 10.18388/abp.2016_136327785481Acta Biochim Pol. 63Baranek M, Markiewicz WT, Barciszewski J. Selected small molecules as inducers of pluripotency. Acta Biochim Pol. 2016; 63:709-16. https://doi.org/10.18388/abp.2016_1363 PMID:27785481 Small molecules for reprogramming and transdifferentiation. H Qin, A Zhao, X Fu, 10.1007/s00018-017-2586-xCell Mol Life Sci. 7428698932Qin H, Zhao A, Fu X. Small molecules for reprogramming and transdifferentiation. Cell Mol Life Sci. 2017; 74:3553-75. https://doi.org/10.1007/s00018-017-2586-x PMID:28698932 Chemical modulation of cell fates: in situ regeneration. H Qin, A Zhao, X Fu, 10.1007/s11427-018-9349-5Sci China Life Sci. 6130099708Qin H, Zhao A, Fu X. Chemical modulation of cell fates: in situ regeneration. Sci China Life Sci. 2018; 61:1137-50. https://doi.org/10.1007/s11427-018-9349-5 PMID:30099708 Small-molecule-mediated reprogramming: a silver lining for regenerative medicine. Y Kim, J Jeong, D Choi, 10.1038/s12276-020-0383-3Exp Mol Med. 5232080339Kim Y, Jeong J, Choi D. Small-molecule-mediated reprogramming: a silver lining for regenerative medicine. Exp Mol Med. 2020; 52:213-26. https://doi.org/10.1038/s12276-020-0383-3 PMID:32080339 The use of small molecules in somatic-cell reprogramming. A J Federation, J E Bradner, A Meissner, 10.1016/j.tcb.2013.09.01124183602Trends Cell Biol. 24Federation AJ, Bradner JE, Meissner A. The use of small molecules in somatic-cell reprogramming. Trends Cell Biol. 2014; 24:179-87. https://doi.org/10.1016/j.tcb.2013.09.011 PMID:24183602 Effect of small molecules on cell reprogramming. M Baranek, A Belter, M Z Naskręt-Barciszewska, M Stobiecki, W T Markiewicz, J Barciszewski, 10.1039/c6mb00595k27918060Mol Biosyst. 13Baranek M, Belter A, Naskręt-Barciszewska MZ, Stobiecki M, Markiewicz WT, Barciszewski J. Effect of small molecules on cell reprogramming. Mol Biosyst. 2017; 13:277-313. https://doi.org/10.1039/c6mb00595k PMID:27918060 Generation of iPSCs from mouse fibroblasts with a single gene, Oct4, and small molecules. Y Li, Q Zhang, X Yin, W Yang, Y Du, P Hou, J Ge, C Liu, W Zhang, X Zhang, Y Wu, H Li, K Liu, 10.1038/cr.2010.14220956998Cell Res. 21Li Y, Zhang Q, Yin X, Yang W, Du Y, Hou P, Ge J, Liu C, Zhang W, Zhang X, Wu Y, Li H, Liu K, et al. Generation of iPSCs from mouse fibroblasts with a single gene, Oct4, and small molecules. Cell Res. 2011; 21:196-204. https://doi.org/10.1038/cr.2010.142 PMID:20956998 Small molecule compound induces chromatin de-condensation and facilitates induced pluripotent stem cell generation. X Wei, Y Chen, Y Xu, Y Zhan, R Zhang, M Wang, Q Hua, H Gu, F Nan, X Xie, 10.1093/jmcb/mju02424838272J Mol Cell Biol. 6Wei X, Chen Y, Xu Y, Zhan Y, Zhang R, Wang M, Hua Q, Gu H, Nan F, Xie X. Small molecule compound induces chromatin de-condensation and facilitates induced pluripotent stem cell generation. J Mol Cell Biol. 2014; 6:409-20. https://doi.org/10.1093/jmcb/mju024 PMID:24838272 Senescence impairs successful reprogramming to pluripotent stem cells. A Banito, S T Rashid, J C Acosta, S Li, C F Pereira, I Geti, S Pinho, J C Silva, V Azuara, M Walsh, L Vallier, J Gil, 10.1101/gad.181160919696146Genes Dev. 23Banito A, Rashid ST, Acosta JC, Li S, Pereira CF, Geti I, Pinho S, Silva JC, Azuara V, Walsh M, Vallier L, Gil J. Senescence impairs successful reprogramming to pluripotent stem cells. Genes Dev. 2009; 23:2134-9. https://doi.org/10.1101/gad.1811609 PMID:19696146 Tissue damage and senescence provide critical signals for cellular reprogramming in vivo. L Mosteiro, C Pantoja, N Alcazar, R M Marión, D Chondronasiou, M Rovira, P J Fernandez-Marcos, M Muñoz-Martin, C Blanco-Aparicio, J Pastor, G Gómez-López, De Martino, A Blasco, M A , 10.1126/science.aaf444527884981Science. 354Mosteiro L, Pantoja C, Alcazar N, Marión RM, Chondronasiou D, Rovira M, Fernandez-Marcos PJ, Muñoz-Martin M, Blanco-Aparicio C, Pastor J, Gómez- López G, De Martino A, Blasco MA, et al. Tissue damage and senescence provide critical signals for cellular reprogramming in vivo. Science. 2016; 354:aaf4445. https://doi.org/10.1126/science.aaf4445 PMID:27884981 Senescence promotes in vivo reprogramming through p16 INK4a and IL-6. L Mosteiro, C Pantoja, A De Martino, M Serrano, 10.1111/acel.1271129280266Aging Cell. 1712711Mosteiro L, Pantoja C, de Martino A, Serrano M. Senescence promotes in vivo reprogramming through p16 INK4a and IL-6. Aging Cell. 2018; 17:e12711. https://doi.org/10.1111/acel.12711 PMID:29280266 The senescence-associated secretory phenotype induces cellular plasticity and tissue regeneration. B Ritschka, M Storer, A Mas, F Heinzmann, M C Ortells, J P Morton, O J Sansom, L Zender, W M Keyes, Genes Dev. 31Ritschka B, Storer M, Mas A, Heinzmann F, Ortells MC, Morton JP, Sansom OJ, Zender L, Keyes WM. The senescence-associated secretory phenotype induces cellular plasticity and tissue regeneration. Genes Dev. 2017; 31:172-83. . 10.1101/gad.290635.11628143833https://doi.org/10.1101/gad.290635.116 PMID:28143833 Stem cells, telomerase regulation and the hypoxic state. J Mathews, P M Davy, L H Gardner, R C Allsopp, 10.2741/438926709774Front Biosci. 21Mathews J, Davy PM, Gardner LH, Allsopp RC. Stem cells, telomerase regulation and the hypoxic state. Front Biosci (Landmark Ed). 2016; 21:303-15. https://doi.org/10.2741/4389 PMID:26709774 Core Pluripotency Factors Directly Regulate Metabolism in Embryonic Stem Cell to Maintain Pluripotency. H Kim, H Jang, T W Kim, B H Kang, S E Lee, Y K Jeon, D H Chung, J Choi, J Shin, E J Cho, H D Youn, 10.1002/stem.207326059508Stem Cells. 33Kim H, Jang H, Kim TW, Kang BH, Lee SE, Jeon YK, Chung DH, Choi J, Shin J, Cho EJ, Youn HD. Core Pluripotency Factors Directly Regulate Metabolism in Embryonic Stem Cell to Maintain Pluripotency. Stem Cells. 2015; 33:2699-711. https://doi.org/10.1002/stem.2073 PMID:26059508 GLUT3 and PKM2 regulate OCT4 expression and support the hypoxic culture of human embryonic stem cells. D R Christensen, P C Calder, F D Houghton, 10.1038/srep1750026639784Sci Rep. 517500Christensen DR, Calder PC, Houghton FD. GLUT3 and PKM2 regulate OCT4 expression and support the hypoxic culture of human embryonic stem cells. Sci Rep. 2015; 5:17500. https://doi.org/10.1038/srep17500 PMID:26639784 Vitamin C enhances the generation of mouse and human induced pluripotent stem cells. M A Esteban, T Wang, B Qin, J Yang, D Qin, J Cai, W Li, Z Weng, J Chen, S Ni, K Chen, Y Li, X Liu, 10.1016/j.stem.2009.12.00120036631Cell Stem Cell. 6Esteban MA, Wang T, Qin B, Yang J, Qin D, Cai J, Li W, Weng Z, Chen J, Ni S, Chen K, Li Y, Liu X, et al. Vitamin C enhances the generation of mouse and human induced pluripotent stem cells. Cell Stem Cell. 2010; 6:71-9. https://doi.org/10.1016/j.stem.2009.12.001 PMID:20036631 Vitamin C modulates TET1 function during somatic cell reprogramming. J Chen, L Guo, L Zhang, H Wu, J Yang, H Liu, X Wang, X Hu, T Gu, Z Zhou, J Liu, J Liu, H Wu, 10.1038/ng.280724162740Nat Genet. 45Chen J, Guo L, Zhang L, Wu H, Yang J, Liu H, Wang X, Hu X, Gu T, Zhou Z, Liu J, Liu J, Wu H, et al. Vitamin C modulates TET1 function during somatic cell reprogramming. Nat Genet. 2013; 45:1504-9. https://doi.org/10.1038/ng.2807 PMID:24162740 Reprogramming the Epigenome With Vitamin C. Lee Chong, T Ahearn, E L Cimmino, L , 10.3389/fcell.2019.0012831380368Front Cell Dev Biol. 7Lee Chong T, Ahearn EL, Cimmino L. Reprogramming the Epigenome With Vitamin C. Front Cell Dev Biol. 2019; 7:128. https://doi.org/10.3389/fcell.2019.00128 PMID:31380368 Network biology: understanding the cell's functional organization. A L Barabási, Z N Oltvai, 10.1038/nrg127214735121Nat Rev Genet. 5Barabási AL, Oltvai ZN. Network biology: understanding the cell's functional organization. Nat Rev Genet. 2004; 5:101-13. https://doi.org/10.1038/nrg1272 PMID:14735121 MicroRNA-regulated protein-protein interaction networks: how could they help in searching for prolongevity targets?. R Tacutu, A Budovsky, M Wolfson, V E Fraifeld, 10.1089/rej.2009.098020367577Rejuvenation Res. 13Tacutu R, Budovsky A, Wolfson M, Fraifeld VE. MicroRNA-regulated protein-protein interaction networks: how could they help in searching for pro- longevity targets? Rejuvenation Res. 2010; 13:373-7. https://doi.org/10.1089/rej.2009.0980 PMID:20367577 Embryonic Stem Cells, Telomeres and Aging. K K Muradian, D A Tolstun, A G Paier, A Popa-Wagner, V E Fraifeld, J Ageing Restor Med. 2Muradian KK, Tolstun DA, Paier AG, Popa-Wagner A, Fraifeld VE. Embryonic Stem Cells, Telomeres and Aging. J Ageing Restor Med. 2019; 2:115-22. Metabolic remodelling of mice by hypoxic-hypercapnic environment: imitating the naked mole-rat. D A Tolstun, A Knyazer, T V Tushynska, T A Dubiley, V V Bezrukov, V E Fraifeld, K K Muradian, 10.1007/s10522-019-09848-9Biogerontology. 2131667660Tolstun DA, Knyazer A, Tushynska TV, Dubiley TA, Bezrukov VV, Fraifeld VE, Muradian KK. Metabolic remodelling of mice by hypoxic-hypercapnic environment: imitating the naked mole-rat. Biogerontology. 2020; 21:143-53. https://doi.org/10.1007/s10522-019-09848-9 PMID:31667660 Wnt/β-catenin signaling is critical for dedifferentiation of aged epidermal cells in vivo and in vitro. C Zhang, P Chen, Y Fei, B Liu, K Ma, X Fu, Z Zhao, T Sun, Z Sheng, 10.1111/j.1474-9726.2011.00753.xAging Cell. 1121967252Zhang C, Chen P, Fei Y, Liu B, Ma K, Fu X, Zhao Z, Sun T, Sheng Z. Wnt/β-catenin signaling is critical for dedifferentiation of aged epidermal cells in vivo and in vitro. Aging Cell. 2012; 11:14-23. https://doi.org/10.1111/j.1474-9726.2011.00753.x PMID:21967252 Dedifferentiation: A New Approach in Stem Cell Research. S A Cai, X Fu, Z Sheng, 10.1641/B570805Bioscience. 57655Cai SA, Fu X, Sheng Z. Dedifferentiation: A New Approach in Stem Cell Research. Bioscience. 2007; 57:655. https://doi.org/10.1641/B570805 Wnt/catenin signaling in adult stem cell physiology and disease. A Ring, Y M Kim, M Kahn, Stem Cell Rev Rep. 10Ring A, Kim YM, Kahn M. Wnt/catenin signaling in adult stem cell physiology and disease. Stem Cell Rev Rep. 2014; 10:512-25. . 10.1007/s12015-014-9515-224825509https://doi.org/10.1007/s12015-014-9515-2 PMID:24825509 The DrugAge database of aging-related drugs. D Barardo, D Thornton, H Thoppil, M Walsh, S Sharifi, S Ferreira, A Anžič, M Fernandes, P Monteiro, T Grum, R Cordeiro, E A De-Souza, A Budovsky, Aging Cell. 16Barardo D, Thornton D, Thoppil H, Walsh M, Sharifi S, Ferreira S, Anžič A, Fernandes M, Monteiro P, Grum T, Cordeiro R, De-Souza EA, Budovsky A, et al. The DrugAge database of aging-related drugs. Aging Cell. 2017; 16:594-7. . 10.1111/acel.1258528299908https://doi.org/10.1111/acel.12585 PMID:28299908 Human Ageing Genomic Resources: new and updated databases. R Tacutu, D Thornton, Johnson E Budovsky, A Barardo, D Craig, T , Diana E Lehmann, G Toren, D Wang, J Fraifeld, V E De Magalhães, J P , Nucleic Acids Res. 46Tacutu R, Thornton D, Johnson E, Budovsky A, Barardo D, Craig T, Diana E, Lehmann G, Toren D, Wang J, Fraifeld VE, de Magalhães JP. Human Ageing Genomic Resources: new and updated databases. Nucleic Acids Res. 2018; 46:D1083-90. . 10.1093/nar/gkx104229121237https://doi.org/10.1093/nar/gkx1042 PMID:29121237 Naked Mole Rat Induced Pluripotent Stem Cells and Their Contribution to Interspecific Chimera. S G Lee, A E Mikhalchenko, S H Yim, A V Lobanov, J K Park, K H Choi, R T Bronson, C K Lee, T J Park, V N Gladyshev, 10.1016/j.stemcr.2017.09.01329107591Stem Cell Reports. 9Lee SG, Mikhalchenko AE, Yim SH, Lobanov AV, Park JK, Choi KH, Bronson RT, Lee CK, Park TJ, Gladyshev VN. Naked Mole Rat Induced Pluripotent Stem Cells and Their Contribution to Interspecific Chimera. Stem Cell Reports. 2017; 9:1706-20. https://doi.org/10.1016/j.stemcr.2017.09.013 PMID:29107591 A naked mole rat iPSC line expressing drug-inducible mouse pluripotency factors developed from embryonic fibroblasts. S G Lee, A E Mikhalchenko, S H Yim, V N Gladyshev, 10.1016/j.scr.2018.06.01030107334Stem Cell Res. 31Lee SG, Mikhalchenko AE, Yim SH, Gladyshev VN. A naked mole rat iPSC line expressing drug-inducible mouse pluripotency factors developed from embryonic fibroblasts. Stem Cell Res. 2018; 31:197-200. https://doi.org/10.1016/j.scr.2018.06.010 PMID:30107334 Naked Mole Rat Cells Have a Stable Epigenome that Resists iPSC Reprogramming. L Tan, Z Ke, G Tombline, N Macoretta, K Hayes, X Tian, R Lv, J Ablaeva, M Gilbert, N V Bhanu, Z F Yuan, B A Garcia, Y G Shi, 10.1016/j.stemcr.2017.10.00129107597Stem Cell Reports. 9Tan L, Ke Z, Tombline G, Macoretta N, Hayes K, Tian X, Lv R, Ablaeva J, Gilbert M, Bhanu NV, Yuan ZF, Garcia BA, Shi YG, et al. Naked Mole Rat Cells Have a Stable Epigenome that Resists iPSC Reprogramming. Stem Cell Reports. 2017; 9:1721-34. https://doi.org/10.1016/j.stemcr.2017.10.001 PMID:29107597 PubChem 2019 update: improved access to chemical data. S Kim, J Chen, T Cheng, A Gindulyte, J He, S He, Q Li, B A Shoemaker, P A Thiessen, B Yu, L Zaslavsky, J Zhang, E E Bolton, 10.1093/nar/gky103330371825Nucleic Acids Res. 47Kim S, Chen J, Cheng T, Gindulyte A, He J, He S, Li Q, Shoemaker BA, Thiessen PA, Yu B, Zaslavsky L, Zhang J, Bolton EE. PubChem 2019 update: improved access to chemical data. Nucleic Acids Res. 2019; 47:D1102-9. https://doi.org/10.1093/nar/gky1033 PMID:30371825 STITCH 4: integration of protein-chemical interactions with user data. M Kuhn, D Szklarczyk, S Pletscher-Frankild, T H Blicher, C Von Mering, L J Jensen, P Bork, 10.1093/nar/gkt120724293645Nucleic Acids Res. 42Kuhn M, Szklarczyk D, Pletscher-Frankild S, Blicher TH, von Mering C, Jensen LJ, Bork P. STITCH 4: integration of protein-chemical interactions with user data. Nucleic Acids Res. 2014; 42:D401-7. https://doi.org/10.1093/nar/gkt1207 PMID:24293645 TRRUST v2: an expanded reference database of human and mouse transcriptional regulatory interactions. H Han, J W Cho, S Lee, A Yun, H Kim, D Bae, S Yang, C Y Kim, M Lee, E Kim, S Lee, B Kang, D Jeong, Nucleic Acids Res. 46Han H, Cho JW, Lee S, Yun A, Kim H, Bae D, Yang S, Kim CY, Lee M, Kim E, Lee S, Kang B, Jeong D, et al. TRRUST v2: an expanded reference database of human and mouse transcriptional regulatory interactions. Nucleic Acids Res. 2018; 46:D380-6. . 10.1093/nar/gkx101329087512https://doi.org/10.1093/nar/gkx1013 PMID:29087512 InParanoid 8: orthology analysis between 273 proteomes, mostly eukaryotic. E L Sonnhammer, G Östlund, 10.1093/nar/gku1203Nucleic Acids Res. 43Sonnhammer EL, Östlund G. InParanoid 8: orthology analysis between 273 proteomes, mostly eukaryotic. Nucleic Acids Res. 2015; 43:D234-9. https://doi.org/10.1093/nar/gku1203 Cytoscape: a software environment for integrated models of biomolecular interaction networks. P Shannon, A Markiel, O Ozier, N S Baliga, J T Wang, D Ramage, N Amin, B Schwikowski, T Ideker, 10.1101/gr.123930314597658Genome Res. 13Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N, Schwikowski B, Ideker T. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003; 13:2498-504. https://doi.org/10.1101/gr.1239303 PMID:14597658

2022-03-16T09:34:26Z

CCBY

http://www.oncotarget.com/index.php?journal=oncotarget&op=download&page=article&path[]=25257&path[]=8501

GOLD

0710ad8191756ad4bab52d214e8ce8e72fb563fa

10.18632/oncotarget.8501

2334927608

27049721

5008274

Tissue repair genes: the TiRe database and its implication for skin wound healing Hagai Yanai The Shraga Segal Department of Microbiology, Immunology and Genetics, Center for Multidisciplinary Research on Aging, Ben-Gurion University of the Negev Beer ShevaIsrael Arie Budovsky The Shraga Segal Department of Microbiology, Immunology and Genetics, Center for Multidisciplinary Research on Aging, Ben-Gurion University of the Negev Beer ShevaIsrael Judea Regional Research and Development Center CarmelIsrael Robi Tacutu The Shraga Segal Department of Microbiology, Immunology and Genetics, Center for Multidisciplinary Research on Aging, Ben-Gurion University of the Negev Beer ShevaIsrael Thomer Barzilay The Shraga Segal Department of Microbiology, Immunology and Genetics, Center for Multidisciplinary Research on Aging, Ben-Gurion University of the Negev Beer ShevaIsrael Amir Abramovich The Shraga Segal Department of Microbiology, Immunology and Genetics, Center for Multidisciplinary Research on Aging, Ben-Gurion University of the Negev Beer ShevaIsrael Rolf Ziesche Division of Pulmonary Medicine Department of Internal Medicine II Medical University of Vienna Waehringer GuertelViennaAustria Vadim E Fraifeld [email protected] The Shraga Segal Department of Microbiology, Immunology and Genetics, Center for Multidisciplinary Research on Aging, Ben-Gurion University of the Negev Beer ShevaIsrael Oncotarget Tissue repair genes: the TiRe database and its implication for skin wound healing 716Received: February 23, 2016 Accepted: March 18, 2016 Published: March 31, 2016Oncotarget 21145 Correspondence to: Vadim E. Fraifeld,wound healinggenesdatabaseskinagingGerotarget Wound healing is an inherent feature of any multicellular organism and recent years have brought about a huge amount of data regarding regular and abnormal tissue repair. Despite the accumulated knowledge, modulation of wound healing is still a major biomedical challenge, especially in advanced ages. In order to collect and systematically organize what we know about the key players in wound healing, we created the TiRe (Tissue Repair) database, an online collection of genes and proteins that were shown to directly affect skin wound healing. To date, TiRe contains 397 entries for four organisms: Mus musculus, Rattus norvegicus, Sus domesticus, and Homo sapiens. Analysis of the TiRe dataset of skin wound healing-associated genes showed that skin wound healing genes are (i) over-conserved among vertebrates, but are under-conserved in invertebrates; (ii) enriched in extracellular and immunoinflammatory genes; and display (iii) high interconnectivity and connectivity to other proteins. The latter may provide potential therapeutic targets. In addition, a slower or faster skin wound healing is indicative of an aging or longevity phenotype only when assessed in advanced ages, but not in the young. In the long run, we aim for TiRe to be a one-station resource that provides researchers and clinicians with the essential data needed for a better understanding of the mechanisms of wound healing, designing new experiments, and the development of new therapeutic strategies. TiRe is freely available online at http://www.tiredb.org INTRODUCTION Tissue repair (often referred to as wound healing [WH]) is an inherent feature of any multicellular organism. Its major goal is to restore the integrity (and ideally function) of a damaged tissue. Some species from diverse taxa (such as salamander, axolotle, hydra, and several others [1]) and early mammalian embryos are able to fully regenerate damaged tissues/organs [2]. In mammals, however, this ability is drastically reduced after birth and continues to decline with age [2,3]. For most organs, this reduced regenerative capacity is in fact a normative response, favoring speed over functional restoration, so that regular tissue repair results in scar formation [2]. Deviations from regular tissue repair may lead to diverse pathological conditions, from slow or ineffective wound healing to hyper-fibroproliferative responses [4,5], both of which are often observed in advanced ages. Thus, factors that govern tissue repair are strongly associated with aging and age-related pathologies, and as such are potential gerotargets. Recent years have brought about a huge amount of data regarding regular and abnormal wound healing. However, despite the accumulated knowledge, modulation of wound healing is still a major biomedical challenge [6]. This problem is expected to become even more challenging considering the phenomenon of population www.impactjournals.com/oncotarget aging. Therefore, there is an essential need to collect and systematically organize what we know about tissue repair and, in particular, what we know about its key genetic and molecular players. With this in mind, we have created TiRe (Tissue Repair), a publicly available and manually curated database of factors that were identified as having a role in the wound healing process. An attempt to create a database on this subject, the "Compendium of Genetically Modified Mouse Wound Healing Studies", was undertaken in the past [4] but is unfortunately no longer available. Here, we have revived this important initiative, and updated and extended the data by including additional model organisms and humans. The current build of the database is focused on skin wound healing, based on the following considerations: (i) the skin is the most frequently injured tissue, and its quick repair is vital for the organism [7,8]; (ii) the basic events during skin repair have much in common across a variety of wounded organs [9]; (iii) due to its accessibility, the skin is more suitable for experimentation than other organs; (iv) the rate of skin wound healing is often used as a biomarker of mammalian aging [10, and references therein]. Altogether, these make the skin a widely used model system for studying the intricate process of wound healing [11,12]. Not surprisingly, the amount of data on wound healing in the skin is superior to most organs, and is constantly increasing. In the long run, we aim for TiRe to be a onestation resource that provides researchers and clinicians with the essential data needed for a better understanding of the mechanisms of wound healing, designing new experiments, and the development of new therapeutic strategies. RESULTS AND DISCUSSION Overview of experimental models used to establish wound healing-associated genes (WHAGs) There is a great variety of methods available for the study of skin wound healing (WH), both with regard to types of genetic interventions and wounding assays [11]. In our dataset, the dominant interventions used in the mouse model are genetic (i.e. knockout or overexpression), whereas other interventions, such as protein administration, are more common in the other species (Table 1). As seen in Table 2, the most common wounding method by far is the dorsal full-thickness excision model. Characterization of WHAGs The TiRe data collection offers an opportunity to gain insight into the features of WHAGs. Most of these genes were identified in the mouse model. Notably, genes that were studied in rats, swine and humans were also studied in mice, and some in more than two species (Table 3). Despite the differences in intervention and wounding methods, targeting the common genes across the species mostly led to consistent results, i.e. to concordant effects. This suggests that WH across these species has much in common. WHAGs are differentially conserved across vertebrates and invertebrates To broaden this perspective, we further investigated the evolutionary conservation of WHAGs. For that purpose, we extracted the WHAG orthologs for all species available in the InParanoid database [13]. As seen in Figure 1, WHAGs are over-conserved among vertebrates, but are under-conserved in invertebrates (for specific details, see Suppl. Table 1). This implies that (i) many of the skin WHAGs are a relatively recent acquisition in the course of evolution; and (ii) despite the significant differences in the anatomy and physiology of the skin between vertebrate species and the resulting differences in wound healing [14], the genetic basis of WH is conserved among vertebrates. WHAGs are enriched in extracellular and immuno-inflammatory pathways Complementary to the results above, our enrichment analysis on WHAGs sheds further light on this vertebratespecific evolutionary conservation. As seen in Figure 2A, WHAGs predominantly encode for extracellular proteins and those involved in cell-cell/cell-ECM interactions. Furthermore, KEGG pathway enrichment analysis highlights a particular role for the focal adhesion pathway as well as for the ECM receptor interaction, regulation of actin cytoskeleton pathways, and various immune/inflammatory-related pathways ( Figure 2B). Remarkably, the pathways involved in immune and inflammatory responses are even more over-represented when considering the genes that are conserved only in vertebrates ( Figure 2C). This was especially noted for the cytokinecytokine receptor interaction and the associated JAK-STAT signaling pathways, hematopoietic cell lineage, complement and coagulation cascade pathways, and the adipocytokine signaling pathway which are enriched only in genes unique to vertebrates. Altogether, the results point to the importance of immuno-inflammatory reactions in wound healing, in vertebrates in particular. This is in line with numerous studies showing the importance of the inflammatory phase and with the unique immunity profile and functionality of vertebrates [15,16]. WHAGs are highly interactive and form a protein-protein interaction network Further supporting the notion that WH is a highly orchestrated and coordinated process [17] is the observation that WHAGs are greatly interconnected and more than two thirds of the WHAGs from the interactome (204/311) can be organized as a continuous protein-protein interaction (PPI) network (clustering coefficient of the entire set = 0.127). Moreover, as seen in Figure 3, many of Table 1). Black triangle -WHAGs (n = 329); grey circle -entire proteome control (n = 20,834). Chi square (χ 2 ) goodness of fit is significant (p < 0.05) for all but 3 of 205 species (see Suppl. Table 1). Evaluation was performed for a score of 1.0. www.impactjournals.com/oncotarget KEGG pathway enrichment for WHAGs that are evolutionary conserved only in vertebrates (see Suppl. Table 1). Enrichment analysis was performed with Enrichr [33] against the Gene Ontology and the KEGG databases. All enrichments presented were statistically significant (adjusted p < 0.05). The presented combined score is the multiplication of the p-value (Fisher exact test) and the z-score of the deviation from the expected rank (for more details see: http://amp.pharm.mssm.edu/Enrichr/). www.impactjournals.com/oncotarget 16.4 for the entire interactome) and together with their first-order interaction partners they would form a huge PPI network of 6,109 proteins, i.e., almost a third of the entire interactome. This incredible connectivity indicates that WHAGs are also in the "epicenter" of many other processes. It has been previously shown that the most critical proteins in a given dataset have more connections within the network than is expected by chance [18][19][20]. Therefore, it is reasonable to assume that first-order partners of WHAGs could also be screened for their importance in WH using their connectivity (both total or with other WHAGs only) for prioritization. The power of such an approach has recently been proven useful for searching new longevity regulators of C. elegans among partners of longevity-associated genes [21]. As an example are Table 3: Summary of genes (human orthologs) tested for their effect on skin wound healing in more than one species. Filled box indicates an examined species. Note: the majority of effects on wound healing were concordant across species 14 selected candidate genes that are not in our original WHAG list, but are highly enriched in connectivity to WHAGs and therefore have a high chance to be valuable for WH (Table 4). Of note, most of them participate in at least one of the WHAGs-enriched signaling pathways ( Figure 2B). For example, SRC which interacts with 39 WHAGs, is at the crossroad of 5 WHAGs-enriched signaling pathways, and has been shown to be an important modulator of cell migration during WH after electric stimulation [22]. Another interesting example is EP300, which is connected to 40 WHAGs, and is involved in both adherens junctions, and the Jak-STAT signaling pathways. EP300 has been suggested to mediate the stimulatory effect of mechanical stress on WH [23]. Of course, further work is required to validate the significance of these candidates, yet this approach demonstrates well how the TiRe dataset can be used to find new wound healing targets. Is accelerated wound healing "good" for longevity? In an attempt to address this question, we have extended our previous analysis [10] by comparing the list of WHAGs with those reported as being involved in Table 4: Selected partners of WHAGs with a strong potential to modulate wound healing. Gene selection was based on candidate's connectivity with other WHAGs (Figure 3), using the Hypergeometric Distribution test (p < 1E-9 for all presented genes). For more details see Suppl. Table 2. the control of lifespan [24]. The comparison yielded 17 genetic mouse models of extended lifespan (longevity phenotype), or reduced lifespan (premature aging phenotype), which were also tested for skin WH. The results are summarized in Table 5. It is important to note that many studies used the rate of skin wound closure as a biomarker, assuming a priori that slower skin WH is indicative of an aging phenotype. Yet, our analysis shows that a slower or faster skin WH is indicative of an aging or longevity phenotype, respectively, only when assessed in advanced ages ( Table 5), but not in the young. For example, Agtr1a knockout resulted in slower wound healing in young mice but also in an extended lifespan [25]. In contrast, Cav1 knockout, which accelerated wound closure, was accompanied by reduced longevity [26]. This means that pro-or anti-longevity effects of genetic interventions manifest in accelerated or delayed skin WH only in advanced ages, but not in young animals. Moreover, it seems that the association between the rate of WH and longevity is primarily attributed to an overall effect of the target gene on organismal aging rather than to its skin-specific action. This assumption is strongly exemplified by our study on the long-lived αMUPA mice, which preserve their skin WH capacity up to an old age (at least 25 months) [10,27]. In this unique model [28], the uPa transgene is expressed in the ocular lens and the brain stem but not in the skin, thus excluding the gene-specific effects on WH. Overall, the results emphasize that the age factor should be taken into account when evaluating the links between skin WH, aging and longevity. To better understand these links, including older animals in the analysis is encouraged while using only young animals might yield confusing or misleading results. In particular, the opposite effect between the rate of skin WH in young age and the effect on life span could be explained by the links between WH and cancer, and the role of cancer in the determination of mouse longevity. Indeed, Schäfer and Werner [29] consider "cancer as an overhealing wound". This could be especially relevant to mice as cancer is the main cause of death for a variety of murine strains [30,31]. For example, Tert overexpression in the young leads to accelerated WH, a high incidence of cancer, and increased mortality [32]. Another example is the tumor suppressor gene Pten, known to negatively regulate the activity of the PI3K/mTOR pathway, which is involved in various cancers [33,34]. Knockout of this gene resulted in accelerated WH in young age but a decreased lifespan [35], which is most likely associated with increased tumorigenesis. CONCLUDING REMARKS This first build of the TiRe database is devoted to skin wound healing genes. It is simple to use, yet an effective source of information. TiRe has a friendly interface that allows researchers and clinicians in the field to easily obtain relevant data, facilitates a view of the "bigger picture", and assists in designing new experiments, especially in the selection of new therapeutic targets. It is important to note that while a gene that is in TiRe is undoubtedly involved in WH, there are genes that are not yet in our database, since their involvement in WH has thus far been established only by expression profile, or in vitro assays. We have taken this gap into account, and intend to expand our database accordingly in our future builds. Yet, the merit of using criteria based only on direct interventions, has been previously shown for the analysis of complex phenomena such as aging, age-related diseases, and cellular senescence [19,24,36]. Surprisingly, despite the rapidly increasing number of skin WHAGs established in model organisms, only a few of them have been tested in human studies (Table 1). Considering the concordant effects observed for model species and humans as well as the evolutionary conservation of WHAGs across mammals, the TiRe gene list could be utilized for the selection of potential targets in future human trials. TiRe is continuously updated and developed. In the next build we aim to include: (i) gene/protein expression data from skin wound healing experiments; (ii) genes associated with tissue repair pathologies (e.g. hypertrophic scars, keloids, scleroderma); (iii) pharmacological interventions (including medicinal plants [37,38]), and (iv) other organs such as lungs, liver, kidney, etc. Of particular interest would be a comparison between WHmodulating drugs and geroprotectors [39]. In perspective, TiRe will serve as a platform for a comprehensive compendium on many aspects of tissue repair, wound healing, and tissue fibrosis. METHODS OF DATABASE CONSTRUCTION AND ANALYSIS Database content Skin wound healing-associated genes (WHAGs) were determined based on genetic studies (knockout, knockdown, overexpression), or interventions that directly influence the level and/or activity of the protein product (antibody treatment, protein administration, etc.). A summary of the types of interventions is listed in Table 1. For all WHAGs included in the database, a given intervention has been observed to cause a marked change in the skin wound healing phenotype (such as accelerated or delayed wound closure, or alterations in the quality of repair In addition to information about the WHAGs, the database also includes the genetic background of the animal model, the type of genetic/protein intervention, the wound model used, wound dimensions and location, and a brief description of the wound healing outcome, with a reference to the original research. Interface TiRe has a user-friendly website interface, with simple and intuitive navigation tools. Searching can be done either by gene symbol, its full name, or gene aliases. Alternatively, the data can be reached by species browsing. The website also allows for downloading the entire dataset from the download page, in order to carry out more extensive analyses offline. A build counter and a build release date are provided to keep track of different database versions. Availability The TiRe database is available at http://www.tiredb. org, with the data made available under the permissive Creative Commons license, allowing data to be used in other analyses. There are options to either download the entire database or its parts. Feedback is welcome. www.impactjournals.com/oncotarget Data analysis Evolutionary conservation The analysis was performed using a software package developed in our lab, which automatically extracts and analyses data from the InParanoid database (http://inparanoid.sbc.su.se/cgi-bin/index.cgi [13]). For each gene, the presence or absence of orthologs across 205 proteomes (all species available excluding parasites) was defined and the evolutionary conservation was expressed as percentage of orthologs. The evaluation was performed for an inparalog score of 1.0. All comparisons were statistically significant unless otherwise mentioned (Chisquared χ 2 test; p < 0.05). Enrichment analysis Enrichment analysis of WHAGs was performed using the EnrichR toolset () [40]. As the data on human genes and proteins is the most complete among the tested species, the human orthologs of WHAGs defined in model organisms were used for the analysis. Statistical significance of enrichment was evaluated with the Fisher's exact t-test and the EnrichR combined score. Longevity-associated genes (LAGs) Longevity-associated genes were extracted from The Human Ageing Genomic Resources (HAGR) -GenAge Database of Ageing-Related Genes, build 17 [24]. Protein-protein interaction network Protein-protein interaction (PPI) data from the BioGRID database ( [41], http://thebiogrid.org), human interactome, release 3.4.129, was used for the analysis of connectivity and interconnectivity. The entire human interactome was used as control. Network construction and analysis was performed using Cytoscape ( [42], http://www.cytoscape.org), version 3.3.0. Prediction of important network interactors was performed using the hypergeometric distribution test for relative connectivity [18]. CONFLICTS OF INTEREST The authors declare that they have no conflict of interest. GRANT SUPPORT This study was funded by the European Union FP7 Health Research Grant number HEALTH-F4-2008-202047, and by the Israel Ministry of Science and Technology. This work was also supported by the Fund in Memory of Dr. Amir Abramovich. Figure 1 : 1Evolutionary conservation of skin wound healing-associated genes. Each dot represents the percentage of orthologs between humans and a given species (in descending order by % of orthology). A total of 205 species from all kingdoms of life are presented (for a full list of species and conservation data see Suppl. Figure 2 : 2Enrichment analysis. A. Cellular component enrichment for all WHAGs. B. KEGG pathway enrichment for all WHAGs. C. Figure 3 : 3Protein-protein interaction network of skin wound healing genes. Depicted in the figure is the largest continuous component of the wound healing network with the most enriched signaling pathways (See Figure 2). Included are also 27 WHAGs connected to the WH network only through the enriched pathways. Genes (N = 231) are depicted with green circles and KEGG pathways (N = 9) with yellow rectangles. The enriched pathways include a total of 114 WHAGs, with several genes belonging to multiple pathways. Depicted are 188 Gene-pathways connections and 566 gene-gene interactions. 5 : 5Comparison of the effects of genetic interventions on skin wound healing and longevity in mice. a Loss-of-function b Enhanced function c Ectopic expression For full gene names please refer to the database. Table 1 : 1Summary of interventions included in the TiRe database. Each entry represents a single study.Intervention Number of entries per species Mus musculus Rattus norvegicus Sus domesticus Homo sapiens Total Knockout 260 0 0 0 260 Overexpression 47 8 3 1 59 Mutation 13 0 0 0 13 siRNA 5 0 0 0 5 Protein administration 15 27 8 12 62 Antibody treatment 8 6 0 0 14 Agonist/antagonist/inhibitor administration 8 2 0 0 10 Other 2 1 1 1 5 Table 2 : 2Summary of wound healing assays included in the TiRe database. Intervention Number of entries per species Mus musculus Rattus norvegicus Sus domesticus Homo sapiens Total Note: over 95% of the indicated studies were performed on the dorsum. www.impactjournals.com/oncotargetFull-thickness excisional punch 264 16 2 0 282 Full-thickness incision 36 13 1 0 50 Flap 5 15 0 0 20 Clinical trial/case report 0 0 0 14 14 Skin graft 4 3 4 0 11 Partial-Thickness wound 4 0 6 0 10 Burn wound 2 2 2 0 6 Embryonic skin wound 6 0 0 0 6 Ear hole 5 0 0 0 5 Other 26 6 0 0 32 Table ). The list of WHAGs established thus far in model organisms and humans was compiled from scientific literature and manually curated. To date, the list contains 397 entries for four organisms: Mus musculus, Rattus norvegicus, Sus domesticus, and Homo sapiens (330, 40, 12 and 14 entries, respectively). www.impactjournals.com/oncotarget Ontogenetic decline of regenerative ability and the stimulation of human regeneration. D M Gardiner, Rejuvenation Res. 8Gardiner DM. Ontogenetic decline of regenerative ability and the stimulation of human regeneration. Rejuvenation Res. 2005; 8:141-153. Scar-free healing: From embryonic mechanisms to adult therapeutic intervention. M W Ferguson, S O&apos;kane, Philos Trans R Soc Lond B Biol Sci. 359Ferguson MW, O'Kane S. Scar-free healing: From embryonic mechanisms to adult therapeutic intervention. Philos Trans R Soc Lond B Biol Sci. 2004; 359:839-850. Mitochondria-targeted antioxidant SkQ1 improves impaired dermal wound healing in old mice. I A Demyanenko, E N Popova, V V Zakharova, O P Ilyinskaya, T V Vasilieva, V P Romashchenko, A V Fedorov, V N Manskikh, M V Skulachev, R A Zinovkin, O Y Pletjushkina, V P Skulachev, B V Chernyak, 10.18632/aging.100772Aging (Albany NY). 7Demyanenko IA, Popova EN, Zakharova VV, Ilyinskaya OP, Vasilieva TV, Romashchenko VP, Fedorov AV, Manskikh VN, Skulachev MV, Zinovkin RA, Pletjushkina OY, Skulachev VP, Chernyak BV. Mitochondria-targeted antioxidant SkQ1 improves impaired dermal wound healing in old mice. Aging (Albany NY). 2015; 7:475-485. doi: 10.18632/aging.100772. Wound healing studies in transgenic and knockout mice. A review. R Grose, S Werner, Methods Mol Med. 78Grose R, Werner S. Wound healing studies in transgenic and knockout mice. A review. Methods Mol Med. 2003; 78:191-216. Fibrosis-a common pathway to organ injury and failure. D C Rockey, P D Bell, J A Hill, N Engl J Med. 372Rockey DC, Bell PD, Hill JA. Fibrosis-a common pathway to organ injury and failure. N Engl J Med. 2015; 372:1138-1149. Cellular and molecular mechanisms of repair in acute and chronic wound healing. P Martin, R Nunan, Br J Dermatol. 173Martin P, Nunan R. Cellular and molecular mechanisms of repair in acute and chronic wound healing. Br J Dermatol. 2015; 173:370-378. Cutaneous wound healing. A J Singer, R A Clark, N Engl J Med. 341Singer AJ, Clark RA. Cutaneous wound healing. N Engl J Med. 1999; 341:738-746. Advances in skin grafting and treatment of cutaneous wounds. B K Sun, Z Siprashvili, P A Khavari, Science. 346Sun BK, Siprashvili Z, Khavari PA. Advances in skin grafting and treatment of cutaneous wounds. Science. 2014; 346:941-945. Wound repair and regeneration. G C Gurtner, S Werner, Y Barrandon, M T Longaker, Nature. 453Gurtner GC, Werner S, Barrandon Y, Longaker MT. Wound repair and regeneration. Nature. 2008; 453:314-321. Is rate of skin wound healing associated with aging or longevity phenotype?. H Yanai, A Budovsky, R Tacutu, V E Fraifeld, Biogerontology. 12Yanai H, Budovsky A, Tacutu R, Fraifeld VE. Is rate of skin wound healing associated with aging or longevity phenotype? Biogerontology. 2011; 12:591-597. The future of wound healing: Pursuing surgical models in transgenic and knockout mice. R R Reid, H K Said, J E Mogford, T A Mustoe, J Am Coll Surg. 199Reid RR, Said HK, Mogford JE, Mustoe TA. The future of wound healing: Pursuing surgical models in transgenic and knockout mice. J Am Coll Surg. 2004; 199:578-585. Wound repair at a glance. T J Shaw, P Martin, J Cell Sci. 122Shaw TJ, Martin P. Wound repair at a glance. J Cell Sci. 2009; 122:3209-3213. InParanoid 8: Orthology analysis between 273 proteomes, mostly eukaryotic. E L Sonnhammer, G Ostlund, Nucleic Acids Res. 43Sonnhammer EL, Ostlund G. InParanoid 8: Orthology analysis between 273 proteomes, mostly eukaryotic. Nucleic Acids Res. 2015; 43:D234-9. Regenerative skin wound healing in mammals: State-of-the-art on growth factor and stem cell based treatments. B M Borena, A Martens, S Y Broeckx, E Meyer, K Chiers, L Duchateau, J H Spaas, Cell Physiol Biochem. 36Borena BM, Martens A, Broeckx SY, Meyer E, Chiers K, Duchateau L, Spaas JH. Regenerative skin wound healing in mammals: State-of-the-art on growth factor and stem cell based treatments. Cell Physiol Biochem. 2015; 36:1-23. Evolution of vertebrate immunity. T Boehm, Curr Biol. 22Boehm T. Evolution of vertebrate immunity. Curr Biol. 2012; 22:R722-32. The acquired immune system: A vantage from beneath. S M Hedrick, Immunity. 21Hedrick SM. The acquired immune system: A vantage from beneath. Immunity. 2004; 21:607-615. Wound healing-aiming for perfect skin regeneration. P Martin, Science. 276Martin P. Wound healing-aiming for perfect skin regeneration. Science. 1997; 276:75-81. . M Bessarabova, A Ishkin, L Jebailey, T Nikolskaya, Bessarabova M, Ishkin A, JeBailey L, Nikolskaya T, www.impactjournals.com/oncotarget Knowledge-based analysis of proteomics data. Y Nikolsky, BMC Bioinformatics. 13Suppl 16:S13-2105-13-S16-S13. EpubNikolsky Y. Knowledge-based analysis of proteomics data. BMC Bioinformatics. 2012; 13 Suppl 16:S13-2105-13- S16-S13. Epub 2012 Nov 5. Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. R Tacutu, A Budovsky, H Yanai, V E Fraifeld, 10.18632/aging.100413Aging (Albany NY). 3Tacutu R, Budovsky A, Yanai H, Fraifeld VE. Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. Aging (Albany NY). 2011; 3:1178-1191. doi: 10.18632/aging.100413. The signaling hubs at the crossroad of longevity and age-related disease networks. M Wolfson, A Budovsky, R Tacutu, V Fraifeld, Int J Biochem Cell Biol. 41Wolfson M, Budovsky A, Tacutu R, Fraifeld V. The signaling hubs at the crossroad of longevity and age-related disease networks. Int J Biochem Cell Biol. 2009; 41:516- 520. Prediction of C. elegans longevity genes by human and worm longevity networks. R Tacutu, D E Shore, A Budovsky, J P De Magalhaes, G Ruvkun, V E Fraifeld, S P Curran, PLoS One. 748282Tacutu R, Shore DE, Budovsky A, de Magalhaes JP, Ruvkun G, Fraifeld VE, Curran SP. Prediction of C. elegans longevity genes by human and worm longevity networks. PLoS One. 2012; 7:e48282. Electrical signals control wound healing through phosphatidylinositol-3-OH kinase-gamma and PTEN. M Zhao, B Song, J Pu, T Wada, B Reid, G Tai, F Wang, A Guo, P Walczysko, Y Gu, T Sasaki, A Suzuki, J V Forrester, Nature. 442Zhao M, Song B, Pu J, Wada T, Reid B, Tai G, Wang F, Guo A, Walczysko P, Gu Y, Sasaki T, Suzuki A, Forrester JV et al. Electrical signals control wound healing through phosphatidylinositol-3-OH kinase-gamma and PTEN. Nature. 2006; 442:457-460. A set of genes previously implicated in the hypoxia response might be an important modulator in the rat ear tissue response to mechanical stretch. V Saxena, D Orgill, I Kohane, BMC Genomics. 8430Saxena V, Orgill D, Kohane I. A set of genes previously implicated in the hypoxia response might be an important modulator in the rat ear tissue response to mechanical stretch. BMC Genomics. 2007; 8:430. Human ageing genomic resources: Integrated databases and tools for the biology and genetics of ageing. R Tacutu, T Craig, A Budovsky, D Wuttke, G Lehmann, D Taranukha, J Costa, V E Fraifeld, J P De Magalhaes, Nucleic Acids Res. 41Tacutu R, Craig T, Budovsky A, Wuttke D, Lehmann G, Taranukha D, Costa J, Fraifeld VE, de Magalhaes JP. Human ageing genomic resources: Integrated databases and tools for the biology and genetics of ageing. Nucleic Acids Res. 2013; 41:D1027-33. Reduced angiogenesis and delay in wound healing in angiotensin II type 1a receptordeficient mice. M Kurosaka, T Suzuki, K Hosono, Y Kamata, A Fukamizu, H Kitasato, Y Fujita, M Majima, Biomed Pharmacother. 63Kurosaka M, Suzuki T, Hosono K, Kamata Y, Fukamizu A, Kitasato H, Fujita Y, Majima M. Reduced angiogenesis and delay in wound healing in angiotensin II type 1a receptor- deficient mice. Biomed Pharmacother. 2009; 63:627-634. Nitric oxide elicits functional MMP-13 protein-tyrosine nitration during wound repair. T R Lizarbe, C Garcia-Rama, C Tarin, M Saura, E Calvo, J A Lopez, C Lopez-Otin, A R Folgueras, S Lamas, C Zaragoza, Faseb j. 22Lizarbe TR, Garcia-Rama C, Tarin C, Saura M, Calvo E, Lopez JA, Lopez-Otin C, Folgueras AR, Lamas S, Zaragoza C. Nitric oxide elicits functional MMP-13 protein-tyrosine nitration during wound repair. Faseb j. 2008; 22:3207-3215. Wound healing and longevity: Lessons from long-lived alphaMUPA mice. H Yanai, D Toren, K Vierlinger, M Hofner, C Nohammer, M Chilosi, A Budovsky, V E Fraifeld, doi: 10.18632/ aging.100726Aging (Albany NY). 7Yanai H, Toren D, Vierlinger K, Hofner M, Nohammer C, Chilosi M, Budovsky A, Fraifeld VE. Wound healing and longevity: Lessons from long-lived alphaMUPA mice. Aging (Albany NY). 2015; 7:167-176. doi: 10.18632/ aging.100726. AlphaMUPA mice: A transgenic model for longevity induced by caloric restriction. R Miskin, O Tirosh, M Pardo, I Zusman, B Schwartz, S Yahav, G Dubnov, R Kohen, Mech Ageing Dev. 126Miskin R, Tirosh O, Pardo M, Zusman I, Schwartz B, Yahav S, Dubnov G, Kohen R. AlphaMUPA mice: A transgenic model for longevity induced by caloric restriction. Mech Ageing Dev. 2005; 126:255-261. Cancer as an overhealing wound: An old hypothesis revisited. M Schafer, S Werner, Nat Rev Mol Cell Biol. 9Schafer M, Werner S. Cancer as an overhealing wound: An old hypothesis revisited. Nat Rev Mol Cell Biol. 2008; 9:628-638. Mutant and genetically modified mice as models for studying the relationship between aging and carcinogenesis. V N Anisimov, Mech Ageing Dev. 122Anisimov VN. Mutant and genetically modified mice as models for studying the relationship between aging and carcinogenesis. Mech Ageing Dev. 2001; 122:1221-1255. Pathobiology of aging mice and GEM: Background strains and experimental design. C F Brayton, P M Treuting, J M Ward, Vet Pathol. 49Brayton CF, Treuting PM, Ward JM. Pathobiology of aging mice and GEM: Background strains and experimental design. Vet Pathol. 2012; 49:85-105. Antagonistic effects of telomerase on cancer and aging in K5-mTert transgenic mice. E Gonzalez-Suarez, C Geserick, J M Flores, M A Blasco, Oncogene. 24Gonzalez-Suarez E, Geserick C, Flores JM, Blasco MA. Antagonistic effects of telomerase on cancer and aging in K5-mTert transgenic mice. Oncogene. 2005; 24:2256-2270. Molecular damage in cancer: An argument for mTOR-driven aging. M V Blagosklonny, 10.18632/aging.100422Aging (Albany NY). 3Blagosklonny MV. Molecular damage in cancer: An argument for mTOR-driven aging. Aging (Albany NY). 2011; 3:1130-1141. doi: 10.18632/aging.100422. Cell cycle arrest is not yet senescence, which is not just cell cycle arrest: Terminology for TORdriven aging. M V Blagosklonny, 10.18632/aging.100443Aging (Albany NY). 4Blagosklonny MV. Cell cycle arrest is not yet senescence, which is not just cell cycle arrest: Terminology for TOR- driven aging. Aging (Albany NY). 2012; 4:159-165. doi: 10.18632/aging.100443. Accelerated wound healing by mTOR activation in genetically defined mouse models. C H Squarize, R M Castilho, T H Bugge, J S Gutkind, PLoS One. 510643Squarize CH, Castilho RM, Bugge TH, Gutkind JS. Accelerated wound healing by mTOR activation in genetically defined mouse models. PLoS One. 2010; 5:e10643. Common gene signature of cancer and longevity. A Budovsky, R Tacutu, H Yanai, A Abramovich, M Wolfson, V Fraifeld, Mech Ageing Dev. 130Budovsky A, Tacutu R, Yanai H, Abramovich A, Wolfson M, Fraifeld V. Common gene signature of cancer and longevity. Mech Ageing Dev. 2009; 130:33-39. Uncovering the geroprotective potential of medicinal plants from the judea region of israel. A Budovsky, A Shteinberg, H Maor, O Duman, H Yanai, M Wolfson, V E Fraifeld, Rejuvenation Res. 17Budovsky A, Shteinberg A, Maor H, Duman O, Yanai H, Wolfson M, Fraifeld VE. Uncovering the geroprotective potential of medicinal plants from the judea region of israel. Rejuvenation Res. 2014; 17:134-139. Effect of medicinal plants on wound healing. A Budovsky, L Yarmolinsky, S Ben-Shabat, Wound Repair Regen. 23Budovsky A, Yarmolinsky L, Ben-Shabat S. Effect of medicinal plants on wound healing. Wound Repair Regen. 2015; 23:171-183. org: A new, structured and curated database of current therapeutic interventions in aging and age-related disease. A Moskalev, E Chernyagina, J P De Magalhaes, D Barardo, H Thoppil, M Shaposhnikov, A Budovsky, V E Fraifeld, A Garazha, V Tsvetkov, E Bronovitsky, V Bogomolov, A Scerbacov, 10.18632/aging.100799Aging (Albany NY). 7Moskalev A, Chernyagina E, de Magalhaes JP, Barardo D, Thoppil H, Shaposhnikov M, Budovsky A, Fraifeld VE, Garazha A, Tsvetkov V, Bronovitsky E, Bogomolov V, Scerbacov A et al. Geroprotectors.org: A new, structured and curated database of current therapeutic interventions in aging and age-related disease. Aging (Albany NY). 2015; 7:616-628. doi: 10.18632/aging.100799. Interactive and collaborative HTML5 gene list enrichment analysis tool. E Y Chen, C M Tan, Y Kou, Q Duan, Z Wang, G V Meirelles, N R Clark, A Ma&apos;ayan, Enrichr, BMC Bioinformatics. 14Chen EY, Tan CM, Kou Y, Duan Q, Wang Z, Meirelles GV, Clark NR, Ma'ayan A. Enrichr: Interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinformatics. 2013; 14:128-2105-14-128. The BioGRID interaction database: 2015 update. A Chatr-Aryamontri, B J Breitkreutz, R Oughtred, L Boucher, S Heinicke, D Chen, C Stark, A Breitkreutz, N Kolas, O&apos; Donnell, L Reguly, T Nixon, J Ramage, L , Nucleic Acids Res. 43Chatr-Aryamontri A, Breitkreutz BJ, Oughtred R, Boucher L, Heinicke S, Chen D, Stark C, Breitkreutz A, Kolas N, O'Donnell L, Reguly T, Nixon J, Ramage L et al. The BioGRID interaction database: 2015 update. Nucleic Acids Res. 2015; 43:D470-8. A travel guide to cytoscape plugins. R Saito, M E Smoot, K Ono, J Ruscheinski, P L Wang, S Lotia, A R Pico, G D Bader, T Ideker, Nat Methods. 9Saito R, Smoot ME, Ono K, Ruscheinski J, Wang PL, Lotia S, Pico AR, Bader GD, Ideker T. A travel guide to cytoscape plugins. Nat Methods. 2012; 9:1069-1076.

2022-03-04T08:54:38Z

CCBYNC

https://academic.oup.com/nar/article-pdf/41/D1/D1027/3653409/gks1155.pdf

GOLD

e0a0b2806d53b071c546a01709b5531be8c11f13

journals/nar/TacutuCBWLTCFM13

10.1093/nar/gks1155

2123703699

23193293

3531213

Human Ageing Genomic Resources: Integrated databases and tools for the biology and genetics of ageing Robi Tacutu Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool L69 7ZBLiverpoolUK Thomas Craig Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool L69 7ZBLiverpoolUK Arie Budovsky The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 84105Beer-ShevaIsrael Judea Regional Research & Development Center 90404CarmelIsrael Daniel Wuttke Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool L69 7ZBLiverpoolUK Gilad Lehmann The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 84105Beer-ShevaIsrael Dmitri Taranukha The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 84105Beer-ShevaIsrael Joana Costa Department of Genetics Liverpool Women's Hospital L8 7SSLiverpoolUK Vadim E Fraifeld The Shraga Segal Department of Microbiology, Immunology and Genetics Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev 84105Beer-ShevaIsrael Joã Pedro De Magalhã Es Integrative Genomics of Ageing Group Institute of Integrative Biology University of Liverpool L69 7ZBLiverpoolUK Human Ageing Genomic Resources: Integrated databases and tools for the biology and genetics of ageing 10.1093/nar/gks1155Received September 19, 2012; Revised October 24, 2012; Accepted October 25, 2012 The Human Ageing Genomic Resources (HAGR, http://genomics.senescence.info) is a freely available online collection of research databases and tools for the biology and genetics of ageing. HAGR features now several databases with high-quality manually curated data: (i) GenAge, a database of genes associated with ageing in humans and model organisms; (ii) AnAge, an extensive collection of longevity records and complementary traits for >4000 vertebrate species; and (iii) GenDR, a newly incorporated database, containing both gene mutations that interfere with dietary restrictionmediated lifespan extension and consistent gene expression changes induced by dietary restriction. Since its creation about 10 years ago, major efforts have been undertaken to maintain the quality of data in HAGR, while further continuing to develop, improve and extend it. This article briefly describes the content of HAGR and details the major updates since its previous publications, in terms of both structure and content. The completely redesigned interface, more intuitive and more integrative of HAGR resources, is also presented. Altogether, we hope that through its improvements, the current version of HAGR will continue to provide users with the most comprehensive and accessible resources available today in the field of biogerontology. INTRODUCTION The Human Ageing Genomic Resources (HAGR, http:// genomics.senescence.info) is a web portal encompassing several online databases and tools, aiming to organize the increasing amount of data and information relevant to the biology of ageing, and make them accessible to the research community. Since its first publication in 2005 (1), HAGR has been the leading online resource for biogerontologists, acting as a reference point for various studies and in particular for genetic and evolutionary studies of longevity and ageing. Initially, HAGR was established around two manually curated databases: GenAge, a database of genes potentially associated with human ageing, and AnAge, a database of ageing, longevity and life history traits in animals. While the goal of GenAge is to function as a compilation of genetic observations reflecting our current knowledge about human ageing, AnAge contains an extensive collection of longevity records, developmental, reproductive and metabolic traits and other key observations related to ageing in >4000 vertebrate species. Since then, the GenAge and AnAge databases have been continuously curated, and new data have been incorporated. Moreover, entire new data sets have also been included. In the update published in 2009 (2), a new list of human genes tested for their possible association with human longevity and a data set of genes associated with longevity and/or ageing in the most studied model organisms were added to GenAge. Although the latter does not include human genes, it was included to serve as a tool for researchers studying ageing in model organisms and because many ageing-related discoveries in model organisms could provide important insights into human ageing (3,4). With the emergence of new high-throughput technologies, many genes associated with ageing and longevity are still being identified, in particular in model organisms (5)(6)(7)(8)(9)(10). Although the field of biogerontology is rapidly evolving, we are still relatively far from having a complete picture of the human ageing process (11), and the need to continue collecting and systematically organizing what we know about the genetics and comparative biology of ageing is more important than ever. In particular, extracting quantitative data from the scientific literature is especially important for performing bioinformatics and systems biology analyses and for guiding experiments. Of note, some genes related to ageing are already being targeted for drug discovery, and translating findings from the genetics of ageing into clinical interventions is an emerging prospect (12). This article describes the major updates in HAGR since its previous publication in 2009. After a brief introduction to each database, general statistics on content and descriptions of the new types of annotations introduced in HAGR are presented, and a newly added database-GenDR, which focuses on dietary restriction (DR)-associated genes-is then described. Finally, the new HAGR interface, completely redesigned to be more intuitive and user friendly, with an improved cross-integration of our different databases and tools, is presented. DATABASE CONTENT New content and features in GenAge The main aim of the GenAge database (http://genomics. senescence.info/genes/) is to host high-quality curated gene-centric information relevant to human ageing. Although initially GenAge was designed to include only human genes potentially associated with ageing, the database has significantly grown since, and several new gene sets have been added to it. For example, GenAge includes, since 2008, a list of genes from model organisms based on genetic manipulation experiments (2). Currently, the database is divided into three main sections intertwined through weblinks and crossreferences: (i) the set of human ageing-associated genes, which includes the few genes directly related to ageing in humans, and the best candidate human genes, supported by evidence from model organisms, cellular experiments and functional analyses; (ii) the set of longevity-associated genes in model organisms, based on lifespan-modulating genetic interventions; and (iii) a list of genes whose expression is commonly altered during ageing in multiple tissues of mammalian species, inferred from microarray data (Table 1). In addition to these data sets, a workin-progress list of genes analysed for their possible association with human longevity in population studies is also available in GenAge (http://genomics.senescence.info/ genes/longevity.html). Since our last update in 2009, we have made numerous improvements in terms of both the quantity and the quality of the content in GenAge. The data set of human genes in GenAge has grown only slightly, accounting now for 288 genes, an increase of 27 genes. As mentioned before, the data set contains a list of genes potentially associated with human ageing. For each gene, a description compiled from the studies that link the gene to ageing is provided. It should be noted that our focus is on genes that might affect the ageing process, rather than individual age-related pathologies; genes affecting multiple, even if not all, age-related processes or pathologies may be selected. Besides containing genes directly linked to human ageing (mostly those genes in which mutations result in segmental progeroid syndromes), the homologous genes with the strongest evidence from model organisms, especially from mammals, will typically be found in the human data set. More formally, genes were included in the human gene data set, and annotated accordingly, if one or more of the following criteria were met, the gene was directly linked: (i) to ageing in humans; (ii) to ageing in a mammalian model organism; (iii) to human longevity and/or multiple age-related phenotypes (the only new criteria since our last update); (iv) to ageing in a non-mammalian model organism; (v) to ageing in a cellular model system; (vi) to the regulation or control of genes previously linked to ageing; (vii) to a pathway or mechanism linked to ageing; or (viii) if the gene was acting downstream of a pathway, mechanism or other gene product linked to ageing. As these data are under continuous curation, and new observations are being actively added to existing entries, the quality of the data set has also been improved since our previous update; for example, the bibliography for this data set has increased significantly, and it currently comprises >2200 references. In contrast to the human data set, whose improvements were mostly qualitative, the latest GenAge build hosts now >1700 genes associated with longevity in model organisms, a drastic boost compared with the 2009 update (a 2.17-fold increase, in fact, in the total number of entries). This rapid increase in the volume of data in the model organisms data set could be mainly attributed to the significant advancements in high-throughput technologies in recent years. For example, in yeast, >750 of the total number of longevity observations (73%) come from several recent large-scale screens (6-10). All the model organism gene entries in GenAge are based on experimentally validated results from the peer-reviewed scientific literature and are manually extracted by our database curators. Genes are considered for inclusion if genetic manipulations (including knockout, mutations, overexpression or RNA interference) result in noticeable changes in the ageing phenotype and/or lifespan. In the cases where a reduction in lifespan is observed, we include only studies in which the authors are linking the gene interventions to ageing (either mechanistically or by checking various signs and markers of ageing). Exceptions are genes from large-scale experiments in yeast, though as detailed below, these have a different classification. Although we try to be as objective as possible, our selection process is still to a certain degree subjective. Our longestablished policy is to have an inclusive rather than exclusive policy, but providing the evidence and links to the relevant literature to allow users to reach their own opinions about each featured gene. In addition to the increase in the number of entries, in the current update, we have altered the database schema allowing us to host for each longevity-associated gene multiple observations from the same or different studies. By doing this, we are aligning our policy for the data set of genes in model organisms with that for the human data set, namely, reporting all supporting and/or conflicting results instead of only the first study. As of writing (build 16), GenAge provides 1708 genes and 2121 lifespan observations (on average 1.24 observations per gene), of which only a small number (4% of the genes) represent conflicting results (Table 2). Moreover, most (83%) of these 'conflicting' results were found for yeast genes. One of the reasons GenAge has been so popular since its creation is because it has allowed computational gerontologists to access and directly use its data. In this update, we have gone one step further and acknowledged that GenAge can be even more useful if more quantitative data is included. As such, the structure and content of the database have been updated, and the new version of GenAge includes extensive quantitative data. Namely, we have extracted for each experiment, where data was available, the effect (relative change) that a certain genetic intervention has on the mean and/or maximal lifespan. In total, >1250 observations, accounting for 1057 genes for model organisms, have been annotated this way. Additionally, we have recorded the type of intervention and categorized longevity-associated genes either as proor anti-longevity. The criteria for the division of longevity-associated genes into pro-and anti-longevity are based on the type of intervention (loss-of-function or gain-of-function) and its impact on lifespan, and were described previously (3). Briefly, pro-longevity genes are defined as the genes whose overexpression extends lifespan, or whose decreased activity (e.g. because of knockout or RNA interference) reduces lifespan. As aforementioned, in the latter case, we focus on genes linked to ageing processes, such as genes in which mutations result in signs of premature ageing. As this can be problematic for yeast large-scale lifespan screens, yeast genes derived from screens or for which a link to ageing processes has not been observed have been annotated as 'necessary for fitness' genes instead of 'pro-longevity' genes. Anti-longevity genes are those for which the aforementioned interventions have the opposite effects. In cases where conflicting results were observed, or the data were not sufficient to draw a definite conclusion, our policy has been to keep all observation and annotate the genes as 'Unclear' and 'Unannotated', respectively. Not taking into account genes with conflicting results, GenAge catalogues a total of 721 anti-longevity genes, 413 prolongevity genes and 485 genes necessary for fitness ( Table 2). It should be, however, noted that this does not necessarily reflect the real genome-wide distribution of pro-and anti-longevity genes, as our data does not account for any biases introduced by the type of experimental design most commonly used by gerontologists. As in previous versions (1,2), additional external information (including homologues, cytogenetic information, gene ontology annotation, protein-protein interaction data and sequence information) and links are also incorporated in GenAge. GenDR-genomics of DR DR, of which caloric restriction is the most widely studied regimen, is the most robust non-genetic intervention shown to extend lifespan in a multitude of species, from yeast to mammals (12,14). However, the exact mechanisms of how DR extends lifespan remain unknown. To decipher the mechanisms of DR in a systematic fashion, we established GenDR (http://genomics.senescence.info/ diet/), the first database of DR-associated genes. Because GenDR and related analysis of DR networks have been recently described elsewhere (15), they will only be briefly To create GenDR, we compiled from the literature a list of DR-essential genes from model organisms. DR-essential genes were defined as those which, if genetically modified, interfere with DR-mediated lifespan extension and, ideally, do not affect the lifespan of animals on an ad libitum diet (or at least do not appear to be merely causing disease). A subset of these genes act as genetic DR mimetics, as their manipulation leads to an increased lifespan for ad libitum fed animals, which is not further extended by DR. One such example is the growth hormone receptor gene in mice (16), in fact the only mouse gene currently in GenDR. In GenDR, the respective homologues of DR-essential genes are included for all the common model organisms, as well as for humans (15). A complementary data set in GenDR is a list of genes consistently differentially expressed in mammals under DR. In a recent meta-analysis, a common signature of genes differentially expressed in DR across different mammalian species, strains, tissues and experiments was derived. This signature provides a set of genes that are most robustly responding to DR (17). Presently, build 1 of GenDR features 158 DR-essential genes plus 173 genes that are part of the conserved molecular signature of DR in mammals. We hope that GenDR will help decipher DR-mediated life extension and promote the development of pharmacological DR mimetics with clinical applications (12). Importantly, GenDR has been fully integrated in HAGR with abundant cross-links between GenDR and GenAge. Enhancing data quality in AnAge, the database of animal ageing and longevity As previously mentioned, the AnAge database (http:// genomics.senescence.info/species/) hosts ageing-related observations, longevity records and a multitude of additional data (including developmental and reproductive traits, taxonomic information and basic metabolic characteristics) for >4000 animal species. Since its inception (1), the main focus of AnAge has been on longevity data and in particular on the quality of data on maximum longevity, rather than merely taking the highest value that is error prone (18). AnAge includes a qualifier of confidence in the data, an estimate of sample size to aid the use of longevity data in comparative studies of ageing, whether maximum longevity comes from a specimen kept in captivity or from the wild and the specific source of the longevity record. Longevity records are manually curated and, if necessary, evaluated by contacting experts (veterinarians, zoologists, etc.) with first-hand experience in a given taxa to assess the reliability of the data, as previously described (18). Anecdotes, unverified longevity claims and controversial issues are mentioned as additional observations. As such, AnAge is arguably the 'gold standard' longevity data in animals. To assist and facilitate comparative studies of ageing, quantitative data on other traits that often correlate with longevity are also included in AnAge, but these are often taken from other data compilations (19)(20)(21)(22)(23)(24)(25). Nonetheless, a variety of automated quality control procedures are in place to detect potential errors in all data in AnAge (18). The focus of AnAge has always been on vertebrates and on mammals in particular. AnAge features higher quality data from mammals, also because they tend to be subjected to more studies than other taxa. At present (build 12), AnAge features data for 4205 species, of which >1000 species are mammals, but also, >1000 species of birds, >500 reptiles, nearly 200 amphibians and nearly 1000 fish species (Table 3) are present. Seventeen other classes, each represented by only a few species in AnAge (maximum lifespan ranging from 0.16 to 15 000 years) have been excluded from Table 3. As AnAge also serves as an informational website for research on ageing, traditional biomedical model organisms, including yeast and invertebrates, are featured. In fact, all species for which there are ageing-or DR-associated genes are present in AnAge with various cross-links between AnAge and HAGR's gene-centric databases. Although in terms of content, AnAge has grown only slightly, we have consistently continued to curate its data and enhance its quality. Additionally, new information and references have also been added to AnAge entries. Specifically, we added a modest 109 new species (mostly birds: 91), suggesting that AnAge may be reaching saturation in terms of reflecting the available quality data on longevity of vertebrates and of mammals in particular. Indeed, of the hundreds of longevity records and entries updated, the vast majority (87%) were from birds. This is mostly because of the continually breaking of longevity records in the wild from various banding studies (26,27). In addition, ample information and >200 references have been added, and >1000 references are now cited in AnAge. The information provided for the most important organisms for research on ageing has also been expanded. For example, in the case of the traditional biomedical model organisms, additional information on ageing phenotypes and/or age-related changes and pathologies has been added, and AnAge thus serves as a reference for the organisms mentioned in GenAge and GenDR. Overall, AnAge continues to be the reference longevity database for comparative and evolutionary studies of ageing. Although its coverage appears to be reaching saturation, at least for mammals, data quality continues to be improved with more information added and more studies cited (in total, HAGR now includes observations for >1950 species). Like for all HAGR databases, AnAge's interface has been greatly improved, while maintaining a familiar functionality, and cross-links with other databases has been expanded. Lastly, AnAge has also proven valuable for a variety of studies in other areas, from conservation studies to various evolutionary analyses that use the comparative method and/or benefit from longevity data. Other informational resources Associated with HAGR is senescence.info (http://www. senescence.info), which aims to provide an informational repository on the science of ageing. Since its creation, senescence.info has continually grown and has been recently updated. We hope that it provides a comprehensive introduction to biogerontology, for both scientists and non-scientists alike, covering a wide range of aspects of ageing research and its social implications. The senescence.info website is also an important resource for students and educators and is a source of teaching materials for various courses on ageing. However, contrary to HAGR, which involves several curators and experts, senescence.info is developed by only one of us (J.P.M.). A complementary resource to HAGR is the Who's Who in Gerontology website (http://whoswho.senescence.info/), based around the WhosAge database, which contains information on individuals, mainly researchers, and companies working on ageing and lifespan extension. At the time of writing, WhosAge contains information for 262 researchers and 22 companies working on ageing and related fields. During the development of HAGR, a series of software programs (mostly Perl and SPSS scripts) have been also created to help our team in a variety of bioinformatic analyses, such as demographic analysis of age-related mortality (28), a repository of which is available to users as well (http://genomics.senescence.info/software/). NEW AND IMPROVED INTERFACE HAGR has grown extensively since it was first published, encompassing more data on ageing and incorporating new databases. One of the key problems that sites encounter when they have grown so rapidly over time is a steady decline in their usability and navigation: sections become harder to discover and navigational features become overloaded. To address and significantly improve these issues, we have completely redesigned the HAGR interface and added many new features. The redesign has overhauled the interface, giving it a new visual look that is more consistent and helps tie each of the resources into a single cohesive piece, as well as improving the navigation to enhance the discoverability of sections. There are now two parts to the navigation, the first being the global navigation bar, which is present across the top of each page in HAGR. This provides quick and consistent navigation to each section, as well as integration for the database-specific searches into a single location. It also provides links to external ageing resources that are relevant, including the Digital Ageing Atlas-a portal of ageing-related changes, which is currently under development (http://ageing-map.org)-senescence.info and the WhosAge database. Finally, the new design provides a global search function that can query each database in HAGR, providing a simple way of searching the data in HAGR without knowing exactly which database is relevant. To complement the global navigation bar, each page contains a left navigation bar that contains context-specific links and information related to the current page. For instance, while using the model organisms' gene section in GenAge, the navigation contains search functions and links to tools related to this data set ( Figure 1). This, combined with the greater clarity and structure that the visual refresh provides, makes it much easier for users to find the relevant section and information. All data sets in HAGR have been better integrated and linked to each other, and each gene entry now contains direct links to all other relevant entries in HAGR's data sets. For example, researchers can quickly see from GenAge if homologues, from the InParanoid database (29), for a model organism gene are present in the human ageing gene set as well or if the gene is present in GenDR. Species information can also be easily accessed through links to AnAge. While each database was adapted to the new visual style, GenAge was given some further enhancements to help improve the quality of the information within, as mentioned earlier. To support these changes, the search function has gained the ability to filter and sort by organisms and gene annotations (observations of 'increased' or 'decreased' lifespan effect, and pro-and anti-longevity designations), allowing for a quicker identification of the gene(s) of interest. When presenting search results, if a gene has multiple observations, all suggesting the same effect on lifespan, that effect and the highest observed value will be shown (and implicitly used for filtering). Another resource that has been completely redesigned is Who's Who in Gerontology, which has seen changes both to the interface and to its available features. The section focused on individuals has gained a density map indicating the number of researchers working on ageing in each country, as well as a better indication for each person of which country they are currently working in. Each company also now lists their approximate location on a map. These new features combined with a new cleaner design make the Who's Who resource much easier to use, as well as providing more information than previously. AVAILABILITY Same as with our previous access policy, the collection of databases in HAGR is freely available at http://genomics. senescence.info. For all databases, we also provide users with the possibility to export, download and reuse the data for their own analyses, under a Creative Commons Attribution license. Feedback via email is heartily welcome, and we encourage the subscription to the HAGR mailing list to be informed of major updates and changes in HAGR. CONCLUDING REMARKS Since its creation, HAGR has proved to be a widely used science of ageing portal with much needed resources for biogerontologists. Both GenAge and AnAge have been featured in a number of other databases and resources, and HAGR has been cited >200 times (Table 4). For example, GenAge has been used as the basis for all the longevity networks hosted in the NetAge database [(30); http://netage-project.org], is featured in JenaLib [(31); http://www.fli-leibniz.de/IMAGE.html] and participates in LinkOut from NCBI resources like OMIM and Entrez Gene (32). AnAge is a content partner of the Encyclopedia of Life (http://eol.org), and its data have also been incorporated into the Biology of Ageing Overall, the current version of the HAGR provides users with a new intuitive interface, significantly augmented information content in the rapidly evolving field of ageing and a better integration of its resources. It is fitting that as a resource for the systems biology of ageing, HAGR has become a resource in which the whole is greater than the sum of its parts. Developments in high-throughput approaches, such as recent advances in next-generation sequencing (34), promise to continue generating ever greater amounts of data in biogerontology that we expect HAGR to continue to accommodate, and thus, we anticipate HAGR to steadily grow. We hope that these and other ongoing improvements will further enhance the use of HAGR's collection of databases and allow HAGR to continue to be the leading online resource for biogerontology. ( http://biologyofaging.org) portal, the Animal Diversity Web (http://animaldiversity.ummz.umich.edu) and the Comparative Cellular and Molecular Biology of Longevity Database [(33); http://genomics.brocku.ca/ ccmbl/]. Table 1 . 1Gene data sets in GenAgeGene set Description Human A comprehensive set of genes potentially associated with human ageing. The list contains genes that have been directly linked to ageing in humans, as well as the best candidate genes, supported by different types of evidence in model organisms, cells and/or functional analyses. Models A set of genes in model organisms (predominantly from Mus musculus, Drosophila melanogaster, Caenorhabditis elegans and Saccharomyces cerevisiae) shown to significantly affect lifespan through genetic manipulations. Microarray A list of genes commonly altered during mammalian ageing from a meta-analysis of microarray studies (13). Table 2 . 2Summary of the genes and longevity observations for model organismsEntries in GenAge M. musculus D. melanogaster C. elegans S. cerevisiae Total Number of genes 91 128 680 809 1708 Pro-longevity 64 85 221 43 413 Anti-longevity 24 41 439 217 721 Necessary for fitness - - - 485 485 Number of observations 119 151 801 1050 2121 Observations per gene 1.31 1.17 1.18 1.30 1.24 Greatest lifespan increase 50% 92% 10-fold 6-fold - Table 3 . 3Summary of AnAge species according to taxonomic classClass Number of species a Range of MLS (years) Average and STDEV of MLS (years) Aves 1088 (1186) 3-79 19.2 ± 16.4 Mammalia 989 (1330) 2.1-211 19.0 ± 15.6 Actinopterygii 811 (822) 0.16-205 17.9 ± 22.8 Reptilia 508 (542) 0.4-177 21.3 ± 17.0 Amphibia 149 (173) 4.1-102 15.6 ± 10.8 Chondrichthyes 115 (116) 6-75 22.3 ± 14.3 a Included in the table are the species with data quality annotated as acceptable or above. In brackets the total number of species present in AnAge is given. MLS, maximum lifespan; STDEV, standard deviation. User statistics include the number of unique visitors per month (in thousands) of all resources combined.HAGR citations 1 6 7 11 25 33 44 46 23 User statistics c 3 6.6 10.6 11.9 15.7 16.1 14.8 16.5 15.1 a Starting in mid-2004. b By mid-2012. c ACKNOWLEDGEMENTSThe authors thank the numerous contributors and experts who have provided valuable advice and suggestionsConflict of interest statement. None declared. . Wellcome Trust, ME050495MES to J.P.M.Wellcome Trust [ME050495MES to J.P.M.]; . Ellison Medical Foundation (to J.P.M. Ellison Medical Foundation (to J.P.M.); European Commission FP7 Health Research. HEALTH-F4-2008-202047 to V.E.F., in part. Funding for open access charge: Wellcome Trust [ME050495MESEuropean Commission FP7 Health Research [HEALTH-F4-2008-202047 to V.E.F., in part]. Funding for open access charge: Wellcome Trust [ME050495MES]. HAGR: the Human Ageing Genomic Resources. J P References 1. De Magalha˜es, J Costa, O Toussaint, Nucleic Acids Res. 33REFERENCES 1. de Magalha˜es,J.P., Costa,J. and Toussaint,O. (2005) HAGR: the Human Ageing Genomic Resources. Nucleic Acids Res., 33, D537-D543. The Human Ageing Genomic Resources: online databases and tools for biogerontologists. J P De Magalha˜es, A Budovsky, G Lehmann, J Costa, Y Li, V Fraifeld, G M Church, Aging Cell. 8de Magalha˜es,J.P., Budovsky,A., Lehmann,G., Costa,J., Li,Y., Fraifeld,V. and Church,G.M. (2009) The Human Ageing Genomic Resources: online databases and tools for biogerontologists. Aging Cell, 8, 65-72. Common gene signature of cancer and longevity. A Budovsky, R Tacutu, H Yanai, A Abramovich, M Wolfson, V Fraifeld, Mech. Ageing Dev. 130Budovsky,A., Tacutu,R., Yanai,H., Abramovich,A., Wolfson,M. and Fraifeld,V. (2009) Common gene signature of cancer and longevity. Mech. Ageing Dev., 130, 33-39. The signaling hubs at the crossroad of longevity and age-related disease networks. M Wolfson, A Budovsky, R Tacutu, V Fraifeld, Int. J. Biochem. Cell Biol. 41Wolfson,M., Budovsky,A., Tacutu,R. and Fraifeld,V. (2009) The signaling hubs at the crossroad of longevity and age-related disease networks. Int. J. Biochem. Cell Biol., 41, 516-520. Lifespan regulation by evolutionarily conserved genes essential for viability. S Curran, G Ruvkun, PLoS Genet. 356Curran,S. and Ruvkun,G. (2007) Lifespan regulation by evolutionarily conserved genes essential for viability. PLoS Genet., 3, e56. Shortest-path network analysis is a useful approach toward identifying genetic determinants of longevity. J R Managbanag, T M Witten, D Bonchev, L A Fox, M Tsuchiya, B K Kennedy, M Kaeberlein, PLoS One. 33802Managbanag,J.R., Witten,T.M., Bonchev,D., Fox,L.A., Tsuchiya,M., Kennedy,B.K. and Kaeberlein,M. (2008) Shortest-path network analysis is a useful approach toward identifying genetic determinants of longevity. PLoS One, 3, e3802. Quantitative evidence for conserved longevity pathways between divergent eukaryotic species. E D Smith, M Tsuchiya, L A Fox, N Dang, D Hu, E O Kerr, E D Johnston, B N Tchao, D N Pak, K L Welton, Genome Res. 18Smith,E.D., Tsuchiya,M., Fox,L.A., Dang,N., Hu,D., Kerr,E.O., Johnston,E.D., Tchao,B.N., Pak,D.N., Welton,K.L. et al. (2008) Quantitative evidence for conserved longevity pathways between divergent eukaryotic species. Genome Res., 18, 564-570. Genome-wide screen in Saccharomyces cerevisiae identifies vacuolar protein sorting, autophagy, biosynthetic, and tRNA methylation genes involved in life span regulation. P Fabrizio, S Hoon, M Shamalnasab, A Galbani, M Wei, G Giaever, C Nislow, V D Longo, PLoS Genet. 61001024Fabrizio,P., Hoon,S., Shamalnasab,M., Galbani,A., Wei,M., Giaever,G., Nislow,C. and Longo,V.D. (2010) Genome-wide screen in Saccharomyces cerevisiae identifies vacuolar protein sorting, autophagy, biosynthetic, and tRNA methylation genes involved in life span regulation. PLoS Genet., 6, e1001024. A microarray-based genetic screen for yeast chronological aging factors. M Matecic, D L Smith, X Pan, N Maqani, S Bekiranov, J D Boeke, J S Smith, PLoS Genet. 61000921Matecic,M., Smith,D.L., Pan,X., Maqani,N., Bekiranov,S., Boeke,J.D. and Smith,J.S. (2010) A microarray-based genetic screen for yeast chronological aging factors. PLoS Genet., 6, e1000921. Identification of evolutionarily conserved genetic regulators of cellular aging. G T Laschober, D Ruli, E Hofer, C Muck, D Carmona-Gutierrez, J Ring, E Hutter, C Ruckenstuhl, L Micutkova, R Brunauer, Aging Cell. 9Laschober,G.T., Ruli,D., Hofer,E., Muck,C., Carmona- Gutierrez,D., Ring,J., Hutter,E., Ruckenstuhl,C., Micutkova,L., Brunauer,R. et al. (2010) Identification of evolutionarily conserved genetic regulators of cellular aging. Aging Cell, 9, 1084-1097. The biology of ageing: a primer. J P De Magalha˜es, An Introduction to Gerontology. Stuart-Hamilton,I.Cambridge, UKCambridge University Pressde Magalha˜es,J.P. (2011) The biology of ageing: a primer. In: Stuart-Hamilton,I. (ed.), An Introduction to Gerontology. Cambridge University Press, Cambridge, UK, pp. 21-47. Genome-environment interactions that modulate aging: powerful targets for drug discovery. J P De Magalha˜es, D Wuttke, S H Wood, M Plank, C Vora, Pharmacol. Rev. 64de Magalha˜es,J.P., Wuttke,D., Wood,S.H., Plank,M. and Vora,C. (2012) Genome-environment interactions that modulate aging: powerful targets for drug discovery. Pharmacol. Rev., 64, 88-101. . J P De Magalha˜es, J Curado, G M Church, de Magalha˜es,J.P., Curado,J. and Church,G.M. (2009) Meta-analysis of age-related gene expression profiles identifies common signatures of aging. Bioinformatics. 25Meta-analysis of age-related gene expression profiles identifies common signatures of aging. Bioinformatics, 25, 875-881. Extending healthy life span-from yeast to humans. L Fontana, L Partridge, V D Longo, Science. 328Fontana,L., Partridge,L. and Longo,V.D. (2010) Extending healthy life span-from yeast to humans. Science, 328, 321-326. Dissecting the gene network of dietary restriction to identify evolutionarily conserved pathways and new functional genes. D Wuttke, R Connor, C Vora, T Craig, Y Li, S Wood, O Vasieva, R S Reis, F Tang, J P De Magalha˜es, PLoS Genet. 81002834Wuttke,D., Connor,R., Vora,C., Craig,T., Li,Y., Wood,S., Vasieva,O., Reis,R.S., Tang,F. and de Magalha˜es,J.P. (2012) Dissecting the gene network of dietary restriction to identify evolutionarily conserved pathways and new functional genes. PLoS Genet., 8, e1002834. Targeted disruption of growth hormone receptor interferes with the beneficial actions of calorie restriction. M S Bonkowski, J S Rocha, M M Masternak, K A Regaiey, A Bartke, Proc. Natl Acad. Sci. USA. 103Bonkowski,M.S., Rocha,J.S., Masternak,M.M., Al Regaiey,K.A. and Bartke,A. (2006) Targeted disruption of growth hormone receptor interferes with the beneficial actions of calorie restriction. Proc. Natl Acad. Sci. USA, 103, 7901-7905. A meta-analysis of caloric restriction gene expression profiles to infer common signatures and regulatory mechanisms. M Plank, D Wuttke, S Van Dam, S A Clarke, J P De Magalha˜es, Mol. Biosyst. 8Plank,M., Wuttke,D., van Dam,S., Clarke,S.A. and de Magalha˜es,J.P. (2012) A meta-analysis of caloric restriction gene expression profiles to infer common signatures and regulatory mechanisms. Mol. Biosyst., 8, 1339-1349. A database of vertebrate longevity records and their relation to other life-history traits. J P De Magalha˜es, J Costa, J. Evol. Biol. 22de Magalha˜es,J.P. and Costa,J. (2009) A database of vertebrate longevity records and their relation to other life-history traits. J. Evol. Biol., 22, 1770-1774. Asdell's Patterns of Mammalian Reproduction: A Compendium of Species-Specific Data. V D Hayssen, A Van Tienhoven, A Van Tienhoven, Comstock Publishing AssociatesIthaca, NYHayssen,V.D., Van Tienhoven,A. and Van Tienhoven,A. (1993) Asdell's Patterns of Mammalian Reproduction: A Compendium of Species-Specific Data. Comstock Publishing Associates, Ithaca, NY. CRC Handbook of Avian Body Masses. J Dunning, CRC PressBoca RatonDunning,J. (2008) CRC Handbook of Avian Body Masses. CRC Press, Boca Raton. Walker's Mammals of the World. R Nowak, Johns Hopkins University PressBaltimoreNowak,R. (1999) Walker's Mammals of the World. Johns Hopkins University Press, Baltimore. PanTHERIA: a species-level database of life history, ecology, and geography of extant and recently extinct mammals. K E Jones, J Bielby, M Cardillo, S A Fritz, J O&apos;dell, C D L Orme, K Safi, W Sechrest, E H Boakes, C Carbone, Ecology. 2648Jones,K.E., Bielby,J., Cardillo,M., Fritz,S.A., O'Dell,J., Orme,C.D.L., Safi,K., Sechrest,W., Boakes,E.H., Carbone,C. et al. (2009) PanTHERIA: a species-level database of life history, ecology, and geography of extant and recently extinct mammals. Ecology, 90, 2648. FishBase. World Wide Web electronic publication. www.fishbase.org, version. R Froese, D Pauly, date last accessedFroese,R. and Pauly,D. (eds), (2012) FishBase. World Wide Web electronic publication. www.fishbase.org, version (August 2012, date last accessed). Life history characteristics of placental non-volant mammals. S K M Ernest, Ecology. 84Ernest,S.K.M. (2003) Life history characteristics of placental non-volant mammals. Ecology, 84, 3402-3402. The Birds of North America Online. A Poole, date last accessedPoole,A. (ed.), (2005) The Birds of North America Online. http:// bna.birds.cornell.edu/BNA/ (August 2012, date last accessed). . T Fransson, T Kolehmainen, C Kroon, L Jansson, T Wenninger, EURING List of Longevity Records for European BirdsFransson,T., Kolehmainen,T., Kroon,C., Jansson,L. and Wenninger,T. (2010) EURING List of Longevity Records for European Birds. Longevity Records of North American Birds, Version 2012.2. Bird Banding Laboratory. J A Lutmerding, A S Love, Patuxent Wildlife Research Center, Laurel, MDLutmerding,J.A. and Love,A.S. (2012) Longevity Records of North American Birds, Version 2012.2. Bird Banding Laboratory, Patuxent Wildlife Research Center, Laurel, MD. The influence of genes on the aging process of mice: a statistical assessment of the genetics of aging. J P De Magalha˜es, J A Cabral, D Magalha˜es, Genetics. 169de Magalha˜es,J.P., Cabral,J.A. and Magalha˜es,D. (2005) The influence of genes on the aging process of mice: a statistical assessment of the genetics of aging. Genetics, 169, 265-274. InParanoid 7: new algorithms and tools for eukaryotic orthology analysis. G Ostlund, T Schmitt, K Forslund, T Kostler, D N Messina, S Roopra, O Frings, E L L Sonnhammer, Nucleic Acids Res. 38Ostlund,G., Schmitt,T., Forslund,K., Kostler,T., Messina,D.N., Roopra,S., Frings,O. and Sonnhammer,E.L.L. (2009) InParanoid 7: new algorithms and tools for eukaryotic orthology analysis. Nucleic Acids Res., 38, D196-D203. The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. R Tacutu, A Budovsky, V E Fraifeld, Biogerontology. 11Tacutu,R., Budovsky,A. and Fraifeld,V.E. (2010) The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. Biogerontology, 11, 513-522. A comparative view at comprehensive information resources on three-dimensional structures of biological macro-molecules. R Hu¨hne, F T Koch, J Su¨hnel, Brief. Funct. Genomic. Proteomic. 6Hu¨hne,R., Koch,F.T. and Su¨hnel,J. (2007) A comparative view at comprehensive information resources on three-dimensional structures of biological macro-molecules. Brief. Funct. Genomic. Proteomic., 6, 220-239. Database resources of the National Center for Biotechnology Information. E W Sayers, T Barrett, D A Benson, E Bolton, S H Bryant, K Canese, V Chetvernin, D M Church, M Dicuccio, S Federhen, Nucleic Acids Res. 40Sayers,E.W., Barrett,T., Benson,D.A., Bolton,E., Bryant,S.H., Canese,K., Chetvernin,V., Church,D.M., Dicuccio,M., Federhen,S. et al. (2012) Database resources of the National Center for Biotechnology Information. Nucleic Acids Res., 40, D13-D25. A comparative cellular and molecular biology of longevity database. J A Stuart, P Liang, X Luo, M M Page, E J Gallagher, C A Christoff, E L Robb, 10.1007/s11357-012-9458-yAge (Dordr). 27epub ahead of printStuart,J.A., Liang,P., Luo,X., Page,M.M., Gallagher,E.J., Christoff,C.A. and Robb,E.L. (2012) A comparative cellular and molecular biology of longevity database. Age (Dordr), 27 July (doi:10.1007/s11357-012-9458-y; epub ahead of print). Next-generation sequencing in aging research: emerging applications, problems, pitfalls and possible solutions. J P De Magalha˜es, C E Finch, G Janssens, Ageing Res. Rev. 9de Magalha˜es,J.P., Finch,C.E. and Janssens,G. (2010) Next-generation sequencing in aging research: emerging applications, problems, pitfalls and possible solutions. Ageing Res. Rev., 9, 315-323.

2022-01-27T06:41:21Z

CCBY

https://genomebiology.biomedcentral.com/track/pdf/10.1186/s13059-020-01990-9

GOLD

8fa95a5b1ed993a8c5291d8346a1d026ca2950d4

10.1186/s13059-020-01990-9

3014402470

32264951

7333371

A multidimensional systems biology analysis of cellular senescence in aging and disease Roberto A Avelar † Javier Gómez Ortega Robi Tacutu Eleanor J Tyler Dominic Bennett Paolo Binetti Arie Budovsky Kasit Chatsirisupachai Emily Johnson Alex Murray Samuel Shields Daniela Tejada-Martinez Daniel Thornton Vadim E Fraifeld Cleo L Bishop João Pedro De Magalhães A multidimensional systems biology analysis of cellular senescence in aging and disease 10.1186/s13059-020-01990-9R E S E A R C H Open Access Background: Cellular senescence, a permanent state of replicative arrest in otherwise proliferating cells, is a hallmark of aging and has been linked to aging-related diseases. Many genes play a role in cellular senescence, yet a comprehensive understanding of its pathways is still lacking.Results: We develop CellAge (http://genomics.senescence.info/cells), a manually curated database of 279 human genes driving cellular senescence, and perform various integrative analyses. Genes inducing cellular senescence tend to be overexpressed with age in human tissues and are significantly overrepresented in anti-longevity and tumor-suppressor genes, while genes inhibiting cellular senescence overlap with pro-longevity and oncogenes. Furthermore, cellular senescence genes are strongly conserved in mammals but not in invertebrates. We also build cellular senescence protein-protein interaction and co-expression networks. Clusters in the networks are enriched for cell cycle and immunological processes. Network topological parameters also reveal novel potential cellular senescence regulators. Using siRNAs, we observe that all 26 candidates tested induce at least one marker of senescence with 13 genes (C9orf40, CDC25A, CDCA4, CKAP2, GTF3C4, HAUS4, IMMT, MCM7, MTHFD2, MYBL2, NEK2, NIPA2, and TCEB3) decreasing cell number, activating p16/p21, and undergoing morphological changes that resemble cellular senescence. Conclusions: Overall, our work provides a benchmark resource for researchers to study cellular senescence, and our systems biology analyses reveal new insights and gene regulators of cellular senescence. Background In the 1960s, Leonard Hayflick and Paul Moorhead demonstrated that human fibroblasts reached a stable proliferative growth arrest between their fortieth and sixtieth divisions [1]. Such cells would enter an altered state of "replicative senescence," subsisting in a nonproliferating, metabolically active phase with a distinct vacuolated morphology [2]. This intrinsic form of senescence is driven by gradual replicative telomere erosion, eventually exposing an uncapped free double-stranded chromosome end and triggering a permanent DNA damage response [3,4]. Additionally, acute premature senescence can occur as an antagonistic consequence of genomic, epigenomic, or proteomic damage, driven by oncogenic factors, oxidative stress, or radiation [5]. Initially considered an evolutionary response to reduce mutation accrual and subsequent tumorigenesis, the pleiotropic nature of senescence has also been positively implicated in processes including embryogenesis [6,7], wound healing [8], and immune clearance [9,10]. By contrast, the gradual accumulation and chronic persistence of senescent cells with time promotes deleterious effects that are considered to accelerate deterioration and hyperplasia in aging [11]. Senescent cells secrete a cocktail of inflammatory and stromal regulators-denoted as the senescence-associated secretory phenotype, or SASP-which adversely impact neighboring cells, the surrounding extracellular matrix, and other structural components, resulting in chronic inflammation, the induction of senescence in healthy cells, and vulnerable tissue [12,13]. Mice expressing transgenic INK-ATTAC, which induces apoptosis of p16-positive senescent cells, also have increased lifespan and improved healthspan [14]. It is, therefore, no surprise that in recent years gerontology has heavily focused on the prevention or removal of senescent cells as a means to slow or stop aging and related pathologies [15][16][17]. Research has sought to ascertain the genetic program and prodrome underlying the development and phenotype of senescent cells [18]. Expedited by recent advances in genomic and transcriptomic sequencing, alongside highthroughput genetic screens, a wealth of publicly available data now exists which has furthered the understanding of senescence regulation [19,20]. Unfortunately, despite our increasing knowledge of cellular senescence (CS), determining whether a cell has senesced is not clearcut. Common senescence markers used to identify CS in vitro and in vivo include senescence-associated βgalactosidase (SA-β-gal) and p16 INK4A (p16) [21][22][23]. However, β-galactosidase activity has been detected in other cell types such as macrophages, osteoclasts, and cells undergoing autophagy [24][25][26]. Furthermore, some forms of senescence are not associated with p16 expression, while p16 has been detected in non-senescent cells [3,27]. As such, there are now over 200 genes implicated in CS in humans alone. Therefore, it is necessary to conglomerate this data into a purposefully designed database. Gene databases are highly useful for genomic computational analyses, as exemplified by the Human Ageing Genomic Resources (HAGR) [28]. HAGR provides databases related to the study of aging, including the GenAge database of aging-related genes, which contains genes related to longevity and aging in model organisms and humans, and DrugAge, which includes a compilation of drugs, compounds, and supplements that extend lifespan in model organisms. CellAge builds on these HAGR facilities to provide a means of studying CS in the context of aging or as a standalone resource; the expectation is that CellAge will now provide the basis for processing the discrete complexities of cellular senescence on a systematic scale. Our recent understanding of biological networks has led to new fields, like network medicine [29]. Biological networks can be built using protein interaction and gene co-expression data. A previous paper used proteinprotein interactions to build genetic networks identifying potential longevity genes along with links between genes and aging-related diseases [30]. Here, we present the network of proteins and genes co-expressed with the CellAge senescence genes. Assaying the networks, we find links between senescence and immune system functions and find genes highly connected to CellAge genes under the assumption that a guilt-by-association approach will reveal genes with similar functions [31]. In this study, we look at the broad context of CS genes-their association with aging and aging-related diseases, functional enrichment, evolutionary conservation, and topological parameters within biological networks-to further our understanding of the impact of CS in aging and diseases. Using our networks, we generate a list of potential novel CS regulators and experimentally validate 26 genes using siRNAs, identifying 13 new senescence inhibitors. Results The CellAge database The CellAge website can be accessed at http://genomics. senescence.info/cells/. Figure 1a presents the main CellAge data browser, which allows users to surf through the available data. The browser includes several columns with information that can be searched and filtered efficiently. Users can search for a commaseparated gene list or for individual genes. Once selected, a gene entry page with more detailed description of the experimental context will open. CellAge was compiled following a scientific literature search of gene manipulation experiments in primary, immortalized, or cancer human cell lines that caused cells to induce or inhibit CS. The first CellAge build comprises 279 distinct CS genes, of which 232 genes affect replicative CS, 34 genes affect stress-induced CS, and 28 genes affect oncogene-induced CS. Of the 279 total genes, 153 genes induce CS (~54.8%), 121 inhibit it (~43.4%), and five genes have unclear effects, both inducing and inhibiting CS depending on experimental conditions (~1.8%) (Fig. 1b). The genes in the dataset are also classified according to the experimental context used to determine these associations. We have also performed a meta-analysis to derive a molecular signature of replicative CS and found 526 overexpressed and 734 underexpressed genes [32]. These gene signatures are also available on the CellAge website. Of the 279 CellAge genes, 44 genes were present in the signatures of CS (15.8%). This overlap was significant (p value = 1.62e−08, Fisher's exact test). While 13 of the CellAge inducers of CS significantly overlapped with the overexpressed signatures of CS (8.5%, p = 2.06e−06, Fisher's exact test), only 7 overlapped with the underexpressed signatures (4.6%, p = 5.13e−01, Fisher's exact test). The CellAge inhibitors of CS significantly overlapped with both the overexpressed signatures of CS (n = 7, 5.8%, p = 4.08e−02, Fisher's exact test) and underexpressed signatures of CS (n = 17, 14%, p = 2.06e−06, Fisher's exact test). CellAge gene functions High-quality curated datasets enable systematic computational analyses [33,34]. Since we are interested in learning more about the underlying processes and Fig. 1 a The CellAge database of CS genes. The main data browser provides functionality to filter by multiple parameters like cell line and senescence type, and select genes to view details and links with other aging-related genes on the HAGR website. b Breakdown of the effects all 279 CellAge genes have on CS, and the types of CS the CellAge genes are involved in. Genes marked as "Unclear" both induce and inhibit CS depending on biological context. Numbers above bars denote the total number of genes inhibiting, inducing, or having unclear effects on CS. c Functional enrichment of the nonredundant biological processes involving the CellAge genes (p < 0.05, Fisher's exact test with BH correction) (Additional file 1: Table S3). GO terms were clustered based on semantic similarities functionality shared by human CS genes, we started by exploring functional enrichment within the CellAge dataset. Using the database for annotation, visualization and integrated discovery-DAVID Version 6.8 [35,36], we found that genes in CellAge are enriched with several clusters associated with Protein Kinase Activity, Transcription Regulation, DNA-binding, DNA damage repair, and Cell cycle regulation in cancer. In particular, genes that induce senescence were more associated with promoting transcription, while genes that inhibit senescence were more associated with repressing transcription. Furthermore, we found that inducers of senescence were significantly associated with VEGF and TNF signalling pathways (p < 0.01, Fisher's exact test with Benjamini-Hochberg correction) (Additional file 1: Table S1 and S2). WebGestalt 2019 was used to determine which nonredundant biological processes the CellAge genes are involved in, and REVIGO was used to cluster related processes (p < 0.05, Fisher's exact test with BH correction) [37,38]. A total of 298 categories were significantly enriched and clustered: Signal transduction by p53 class mediator; Aging; Protein localization to nucleus; DNAtemplated transcription, initiation; Epithelial cell proliferation; Cell growth; Rhythmic process; Cellular carbohydrate metabolism; Reactive oxygen species metabolism; Cytokine metabolism; Adaptive thermogenesis; Organic hydroxy compound metabolism; Methylation; Generation of precursor metabolites and energy ( Fig. 1c; Additional file 1: Table S3). Evolutionary conservation of CellAge genes in model organisms Next, we looked at the conservation of CellAge genes across a number of mammalian and non-mammalian model organisms with orthologues to human CellAge genes using Ensembl BioMart (Version 96) [39] in order to understand the genetic conservation of CS processes. There was a significantly higher number of human orthologues for CellAge genes than for other proteincoding genes in mouse, rat, and monkey, while nonmammalian species did not show significant conservation of CellAge genes (two-tailed z-test with BH correction) (Additional file 1: Table S4; Additional file 2: Fig. S1A). Interestingly, previous studies have found that longevityassociated genes (LAGs) are substantially overrepresented from bacteria to mammals and that the effect of LAG overexpression in different model organisms was mostly the same [40]. It remains unclear what the evolutionary origin of most of the CellAge genes is or why they are not present in more evolutionarily distant organisms. Unique evolutionary pressures could have played an important role in the evolution of CellAge genes in mammals. However, somatic cells in C. elegans and Drosophila are post mitotic and lack an equivalent CS process, which could explain why the CellAge genes are not conserved. We further compared the conservation of CellAge inducers and inhibitors of CS and found that while the inducers were significantly conserved in the mammal model organisms, the inhibitors were not (Additional file 2: Fig. S1B). We also report the number of orthologous CellAge genes present in 24 mammal species using the OMA standalone software v. 2.3.1 algorithm [41] (Additional file 2: Fig. S1C). From 279 CellAge genes, we report 271 orthogroups (OGs) (Additional file 3). Twenty-two OGs were conserved in the 24 mammals, including the following genes: DEK, BRD7, NEK4, POT1, SGK1, TLR3, CHEK1, CIP2A, EWSR1, HDAC1, HMGB1, KDM4A, KDM5B, LATS1, MORC3, NR2E1, PTTG1, RAD21, NFE2L2, PDCD10, PIK3C2A, and SLC16A7 (Additional file 1: Table S5). Within the long-lived mammalian genomes analyzed (human, elephant, naked mole rat, bowhead whale, and little brown bat), we found 128 OG CellAge genes (Additional file 3; genomes available in Additional file 1: Table S6). However, finding OGs is dependent on genome quality and annotations, and higher-quality genomes would likely yield more OGs. For the evolutionary distances, we found that the longlived species had similar distances to the other species, meaning the branch lengths for long-lived species are distributed throughout the phylogeny as expected in a random distribution (Additional file 2: Fig. S1D). This was the case when we analyzed the concatenated tree for the 271 CellAge OGs as well as when we analyzed the 22 individual CellAge genes conserved among all 24 mammalian species (Additional file 4). CellAge vs human orthologues of longevity-associated model organism genes To understand how senescence is linked to the genetics of aging processes, we looked at the intersection of CellAge genes and the 869 genes in the human orthologues of model organisms' longevity-associated genes (LAGs) dataset, collected based on quantitative changes in lifespan [34]. Like CellAge, where genes are classified based on whether their upregulation induces, inhibits, or has an unknown impact on CS, the longevity orthologues dataset also provides information on the effect of upregulation of its genes, namely whether it promotes (pro, 421) or inhibits (anti, 448) longevity (Additional file 1: Table S7; Additional file 2: Fig. S2). The CS inducers statistically overlapped with the antilongevity genes and not with the pro-longevity genes (anti: n = 9,~6%, p = 1.42e−02; pro: n = 6,~4%, p = 1.40e−01, Fisher's exact test with BH correction). We noted an inverse result with the inhibitors of CS, where there was a much greater overlap between the CellAge inhibitors and the pro-longevity genes, resulting in the smallest p value of all the overlaps (n = 18,~15%, p = 2.61e−10, Fisher's exact test with BH correction). However, there was also a significant overrepresentation of genes inhibiting the CS process within the anti-longevity genes (n = 7,~6%, p = 2.41e−02, Fisher's exact test with BH correction). It is possible that some of the pathways the CS inhibitors are associated with increase longevity, whereas other pathways have anti-longevity effects. Overall, these results highlight a statistically significant association between CS and the aging process and suggest a potential inverse relationship between CS and longevity, at least for some pathways. Gene overlaps are available in Additional file 1: Table S8. CellAge genes differentially expressed with age In another work, we performed a meta-analysis to find molecular signatures of aging derived from humans, rats, and mice [42]. To investigate how the expression of CellAge genes changes with age, we looked for CellAge genes which either induce (153) or inhibit (121) senescence within the list of aging signatures. The genes overexpressed with age (449) had a significant overlap with the CellAge genes (CS inducers: n = 17,~11%, p = 6.58e−07; CS inhibitors: n = 9,~7%, p = 6.35e−03, two-tailed Fisher's exact test with BH correction) while the genes underexpressed with age (162) did not (CS inducers: n = 0, p = 8.57e−01; CS inhibitors: n = 3,~3%, p = 1.64e−01). The overexpressed genetic signatures of replicative CS (526) also significantly overlapped with the overexpressed signatures of aging (n = 60,~11%, p = 1.18e−23), but not the underexpressed signatures of aging (n = 3,~1%, p = 8.79e−01). Finally, the underexpressed signatures of replicative CS (734) did not significantly overlap with the overexpressed (n = 18,~3%, p = 8.79e−01) or underexpressed (n = 9,~1%, p = 3.26e−01) signatures of aging. Given that 112 (40%) of CellAge genes have only been confirmed to control CS in fibroblasts, we repeated the above analyses using a subgroup of CellAge genes that have been shown to affect CS in other cell types. A total of 91 CellAge inducers of CS and 72 inhibitors were overlapped with the signatures of aging. The same overlaps were still significant after FDR correction, indicating that the differential expression of CellAge genes with age cannot exclusively be attributed to fibroblast idiosyncrasies (CS inducers overexpressed: n = 10,~11%, p = 1.50e−04; underexpressed: n = 0, p = 1. CS inhibitors overexpressed: n = 6,~8%, 1.34e−02; underexpressed: n = 2,~3%, p = 1.98e−01). Using all protein-coding genes from the meta-analysis as a background list [42], we further examined the CS inducers overexpressed with age for functional enrichment using WebGestalt 2019 to determine if specific biological processes were enriched [38]. In parallel, we performed this analysis using the genes which overlapped between CellAge inhibitors and genes overexpressed with age. In total, 71 GO terms were significantly enriched for the overlap between CellAge senescence inducers and age upregulated genes (p < 0.05 Fisher's exact test with BH correction) (Additional file 1: Table S9). Because many of the enriched GO terms were redundant (e.g., wound healing and response to wound healing, regulation of cytokine production and cytokine production), they were clustered based on semantic similarity scores using REVIGO [37]. We found groups enriched for regulation of apoptotic processes, response to lipid, epithelium development, rhythmic process, circadian rhythm, cytokine metabolism, and cell-substrate adhesion (Additional file 2: Fig. S3A). A total of 71 enriched GO terms for the overexpressed signatures of CS overexpressed with age were clustered using REVIGO, resulting in enriched terms relating to regulated exocytosis, aging, response to beta-amyloid, and cell proliferation (Additional file 1: Table S10; Additional file 2: Fig. S3B). No GO terms were significantly enriched for the inducers of CS underexpressed with age, the inhibitors of CS differentially expressed with age, the underexpressed signatures of CS differentially expressed with age, or the overexpressed signatures of CS underexpressed with age. Tissue-specific CS gene expression and differential expression of CS genes in human tissues with age The Genotype-Tissue Expression (GTEx) project contains expression data from 53 different tissue sites collected from 714 donors ranging from 20 to 79 years of age, grouped into 26 tissue classes [43]. We asked if CellAge genes and differentially expressed signatures of CS were expressed in a tissue-specific manner [42] and determined how CS gene expression changes across different tissues with age [32]. We first examined tissue-specific CS expression and found that CellAge genes were either expressed in a tissue-specific manner less than expected by chance, or in line with expectations; in other words, the majority of CellAge genes tended to be expressed across multiple tissues (Additional file 1: Table S11; Additional file 2: Fig. S4A). Testis was the only tissue with significant differences between the actual and expected number of tissue-specific CellAge genes expressed (less tissuespecific genes than expected by chance, p < 0.05, Fisher's exact test with BH correction). The underexpressed signatures of CS were significantly less tissue-specific in the testis and liver, while the overexpressed signatures of CS were significantly less tissue-specific in the brain, liver, pituitary, and skin, and more tissue-specific in blood. We also compared the ratio of tissue-specific to nontissue-specific genes in the CS datasets to all protein-coding genes. While~25% of all protein-coding genes are expressed in a tissue-specific manner, only~10% of CellAge genes and~11% of signatures of CS are expressed in a tissue-specific manner (Additional file 2: Fig. S4B), significantly less than expected by chance (p = 2.52e−12 and 3.93e−48 respectively, Fisher's exact test with BH correction). Then, we examined the differential expression of CS genes with age in different tissues. Using a previously generated gene set of differentially expressed genes (DEGs) with age in 26 tissues on GTEx [32,43], we found overlaps with 268 CellAge inducers and inhibitors of CS present in the gene expression data (Fig. 2a). The process of finding DEGs with age filters out lowly expressed genes, which explains the 11 missing CellAge CS regulators. Overall, senescence inducers were overexpressed across different tissues with age, although none of the overlaps were significant after FDR correction (Fisher's exact test with BH correction, p < 0.05) (Additional file 1: Table S12). There was the opposite trend in the inhibitors of CS, where there was noticeably less overexpression of CS inhibitors with age, although these overlaps were also not significant after FDR correction. A total of 1240 differentially expressed signatures of CS were also overlapped with the GTEx aging DEGs in 26 human tissues, including 9 tissues previously analyzed ( Fig. 2b) [32]. The overexpressed signatures of CS were significantly overexpressed across multiple tissues with age, and only significantly underexpressed with age in the brain and uterus (p < 0.05, Fisher's exact test with BH correction) (Additional file 1: Table S13). Furthermore, the underexpressed signatures of CS trended towards being overexpressed less than expected by chance across multiple tissues with age, although these overlaps were only significant after FDR adjustment in the colon and nerve, while the underexpressed signatures of CS were significantly overexpressed more than expected in the uterus. Finally, the underexpressed signatures of CS were underexpressed with age more than expected by chance in the colon, lung, and ovary, and underexpressed with age less than expected by chance in the brain. We also compared the ratio of differentially expressed to non-differentially expressed CS genes in at least one tissue with age to the equivalent ratio in all protein-coding genes (Additional file 2: Fig. S5A and S5B) (see Overlap Analysis in Methods). We found that 64% of all protein-coding genes did not significantly change expression with age in any human tissues, whilẽ 19% were overexpressed and~17% were underexpressed (~7% were both overexpressed and underexpressed across multiple tissues) (Additional file 1: Table S14 and S15). For the CellAge genes, the number of inducers of CS significantly overexpressed with age in at least one tissue was significantly higher than the genome average (n = 50,~30%, p = 1.5e−3, Fisher's exact test with BH correction). The inducers of CS underexpressed with age and the inhibitors of CS differentially expressed with age were not significantly different from the protein-coding average. We also compared the number of signatures of CS differentially expressed with age in at least one tissue to the proteincoding genome average. The overexpressed signatures of CS were significantly differentially expressed with age compared to all protein-coding genes, whereas the number of underexpressed signatures of CS was underexpressed with age more than expected by chance. The overall fold change (FC) with age of the CS genes was also compared to the FC with age of all proteincoding genes for each tissue in GTEx ( Fig. 2c; Additional file 1: Table S16). The median log 2 FC with age of the CellAge CS inducers and the overexpressed signatures of CS was greater than the genome median for the majority of tissues on GTEx, although the difference in log 2 FC distribution with age between the inducers of CS and all protein-coding genes was only significant in seven tissues (Wilcoxon rank sum test with BH correction, p < 0.05). The median log 2 FC with age of the CellAge inhibitors of CS and the underexpressed signatures of aging was smaller than the genome median in the majority of tissues, showcasing the opposite trend to the inducers of CS and overexpressed signatures of CS. However, the only tissues with significantly different distributions of log 2 FC with age for the inhibitors of CS were the skin and esophagus, where the median log 2 FC distribution was significantly less than the genome average, and the salivary gland, where the median log 2 FC distribution was significantly more than the genome average. We also found that the distribution of log 2 FC with age of the differentially expressed signatures of CS significantly changed in opposite directions with age in 14 tissues. Interestingly, this trend was present even in the adrenal gland and uterus, where the signatures of CS changed with age in the opposite direction to the majority of other tissues. The expression of the majority of CS genes does not change with age (Additional file 2: Fig. S5A), yet a significant number of CS genes trend towards differential expression with age across multiple tissues in humans (Fig. 2). We ran 10,000 simulations on the GTEx RNAseq data to determine the likelihood of a CS gene being differentially expressed with age in more than one tissue by chance (see Simulation of CS Gene Expression in Human Aging in Methods) (Additional file 2: Fig. S5C; Additional file 5). The likelihood of a CellAge gene being overexpressed with age in more than three tissues and underexpressed with age in more than two tissues by chance was less than 5% (CS gene expression simulations) ( Fig. 2d; Additional file 1: Table S17; Additional file 2: Fig. S5C). CS inducers overexpressed in significantly more tissues with age than expected by chance included CDKN2A, NOX4, CPEB1, IGFBP3. Red values indicate that there were more genes differentially expressed with age than expected by chance (−log 2 (p-val)). Blue values indicate that there were less genes differentially expressed with age than expected by chance (log 2 (p-val)). Asterisks (*) denote tissues with significantly more CS genes differentially expressed with age (p < 0.05, Fisher's exact test with BH correction, abs(50*log 2 FC) > log 2 (1.5)) (Additional file 1: Table S12 and S13). c Comparison of the median log 2 FC and distribution of log 2 FC with age between the CS genes and all protein-coding genes in human tissues. Red tiles indicate that the median log 2 FC of the CellAge and CS genes is higher than the median log 2 FC of all protein-coding genes for that tissue, while blue tiles indicate that the median log 2 FC of the CS genes is lower than the median genome log 2 FC. Asterisks (*) indicate significant differences between the log 2 FC distribution with age of CS genes and the log 2 FC distribution with age of all protein-coding genes for that tissue (p < 0.05, Wilcoxon rank sum test with BH correction) (Additional file 1: Table S16). d CellAge genes differentially expressed in at least two tissues with age. Gray tiles are genes which had low basal expression levels in the given tissue and were filtered out before the differential gene expression analysis was carried out [32]. Colored tiles indicate significant differential expression with age (p < 0.05, moderated ttest with BH correction, abs(50*log 2 FC) > log 2 (1.5)). Numbers by gene names in brackets denote the number of tissues differentially expressing the CellAge gene with age. Red gene names specify that the CellAge gene was significantly overexpressed with age in more tissues than expected by chance, while blue gene names show the CellAge genes significantly underexpressed with age in more tissues than expected by chance (p < 0.05, random gene expression tissue overlap simulations) (Additional file 1: Table S17 -S20). Liver, pancreas, pituitary, spleen, small intestine, and vagina did not have any significant CS DEGs with age ABI3, CDKN1A, CYR61, DDB2, MATK, PIK3R5, VENTX, HK3, SIK1, and SOX2, while PTTG1, DHCR24, IL8, and PIM1 were underexpressed in significantly more tissues (Additional file 1: Table S18; Additional file 2: Fig. S5D). ZMAT3 and EPHA3 were the two CS inhibitors overexpressed in significantly more tissues with age than expected by chance, while CDK1, AURKA, BMI1, BRCA1, EZH2, FOXM1, HJURP, MAD2L1, SNAI1, and VEGFA were underexpressed in significantly more tissues. We also performed simulations to determine the likelihood of gene expression signatures of CS being differentially expressed with age in multiple human tissues by chance (Additional file 1: Table S19): less than 5% of the genes in the CS signatures are expected by chance to be overexpressed with age in more than three tissues or underexpressed with age in more than two tissues. A total of 46 CS signature genes (29 overexpressed, 17 underexpressed) were overexpressed with age in significantly more tissues than expected by chance, and 139 CS signature genes were underexpressed in more tissues than expected by chance (26 overexpressed genes in CS, 113 underexpressed genes in CS) (Additional file 1: Table S20). Do CS and longevity genes associate with aging-related disease genes? A previous paper [34] grouped 769 aging-related diseases (ARDs) into 6 NIH Medical Subject Heading (MeSH) classes [44] based on data from the Genetic Association Database [45]: cardiovascular diseases (CVD), immune system diseases (ISD), musculoskeletal diseases (MSD), nutritional and metabolic diseases (NMD), neoplastic diseases (NPD), and nervous system diseases (NSD). The same approach was used to build the HAGR aging-related disease gene selection tool (http://genomics.senescence.info/diseases/ gene_set.php), which we used to obtain the ARD genes for each disease class and overlap with the CellAge genes. There were links between the CellAge genes and NPD genes, which is expected given the anti-tumor role of senescence (Additional file 1: Table S21). Without accounting for publication bias (i.e., some genes being more studied than others), all ARD classes are significantly associated with CellAge genes, with lower commonalities with diseases affecting mostly non-proliferating tissue such as NSD. NPD genes are even more overrepresented in the GenAge human dataset, which could suggest commonality between aging and senescence through cancer-related pathways. Both the strong association of NPD genes with Gen-Age and senescence, and the strong link between GenAge and all ARD classes is interesting. Indeed, longevityassociated genes have been linked to cancer-associated genes in previous papers [46]. Considering age is the leading risk factor for ARD [47,48], the results from GenAge support the previously tested conjecture that there are (i) at least a few genes shared by all or most ARD classes; and (ii) those genes are also related to aging in general [34]. We also looked for genes that are shared across multiple disease classes and are also recorded as CS genes. CellAge genes shared across multiple ARD classes included VEGFA and IFNG (5 ARD classes), SERPINE1, MMP9, and AR (4 ARD classes), and CDKN2A (3 ARD classes). Results are summarized in Additional file 2: Fig. S6. Are CS genes associated with cancer genes? Cellular senescence is widely thought to be an anti-cancer mechanism [49]. Therefore, the CellAge senescence inducers and inhibitors of senescence were overlapped with oncogenes from the tumor-suppressor gene (TSG) database (TSGene 2.0) (n = 1018) [50] and the ONGene database (n = 698) [51] (Additional file 1: Table S22 -S27). The number of significant genes overlapping are shown in Fig. 3a, while the significant p values from the overlap analysis are shown in Fig. 3b (p < 0.05, Fisher's exact test with BH correction). The significant overlap between CellAge genes and cancer indicates a close relationship between both processes. Specifically, the overlap between CellAge inhibitors and oncogenes, and the overlap between CellAge inducers and TSGs were more significant, with lower p values and larger odds ratios (Fig. 3) [52]. This analysis was repeated after filtering out CellAge genes that were only shown to induce senescence in fibroblasts. The overlaps were still significant after FDR correction, indicating that the overlap between CellAge and cancer genes is not specific to genes controlling CS in fibroblasts (CS inducers with oncogenes: n = 10, p = 9e−05; with TSGs: n = 23, p = 4e−12. CS inhibitors with oncogenes: n = 17, 1e−12; with TSGs: n = 8, p = 9e−04, p < 0.05, Fisher's exact test with BH correction) (Additional file 2: Fig. S7). Gene ontology (GO) enrichment analyses were performed using WebGestalt to identify the function of the overlapping genes [38]. Overlapping genes between CellAge senescence inducers and TSGs were enriched in GO terms related to p53 signalling and cell cycle phase transition (Additional file 2: Fig. S8A). The enriched functions of overlapping genes between CellAge senescence inducers and oncogenes were mainly related to immune system processes and response to stress (Additional file 2: Fig. S8B). Overlapping genes between CellAge senescence inhibitors and TSGs were enriched in only 5 terms, which are cellular response to oxygen-containing compound, positive regulation of chromatin organization, and terms relating to female sex differentiation (Additional file 2: Fig. S8C). Finally, overlapping genes between CellAge senescence inhibitors and oncogenes were related to processes such as negative regulation of nucleic acid-templated transcription, cellular response to stress, and cell proliferation (Additional file 2: Fig. S8D). All of the functional enrichment data can be found in Additional file 1: Table S28 -S31. Network analyses The CellAge genes form both protein-protein and gene co-expression networks. The formation of a proteinprotein interaction (PPI) network is significant in itself given that only~4% of the genes in a randomly chosen gene dataset of similar size are interconnected [53]. In order to have a more holistic view of CS, we were interested in the topological parameters of the networks that CS genes form. For this, several types of networks were constructed using the CellAge genes as seeds: the CS PPI network, along with two CS gene co-expression networks built using RNA-seq and microarray data. Biological networks generally have a scale-free topology in which the majority of genes (nodes) have few interactions (edges), while some have many more interactions, resulting in a power law distribution of the node degree (the number of interactions per node) [31,54]. As expected, the node-degree distribution of the above networks does confirm a scale-free structure (Additional file 2: Fig. S9). Additional file 1: Table S32 presents the network summary statistics for the resulting networks. The network parameters we looked at were as follows: Degree, Betweenness Centrality (BC), Closeness Centrality (CC), and Increased Connectivity (IC). The degree is the number of interactions per node and nodes with high degree scores are termed network hubs. BC is a measure of the proportion of shortest paths between all node pairs in the network that cross the node in question. The nodes with high BC are network bottlenecks and may connect large portions of the network which would not otherwise communicate effectively or may monitor information flow from disparate regions in the network [31]. CC is a measure of how close a certain node is to all other nodes and is calculated with the inverse of the sum of the shortest paths to all other nodes. Lower CC scores indicate that nodes are more central to the network, while high CC scores indicate the node may be on the periphery of the network and thus less central. The IC for each node measures the statistical significance for any overrepresentation of interactions between a given node and a specific subset of nodes (in our case CellAge proteins) when compared to what is expected by chance. Taken together, genes that score highly for degree, BC, CC, and IC within the senescence networks are likely important regulators of CS even if up until now they have not been identified as CS genes. Looking at the topology of CS networks, the PPI network, microarray-based co-expression network, and RNA-seq coexpression network all possess comparable scale-free structures. However, gene co-expression data is less influenced by publication bias. This is particularly important considering published literature often reports positive proteinprotein interactions over protein interactions that do not exist [55]. The lack of negative results for protein interaction publications complicates the interpretation of PPI networks even more, as the absence of edges in networks does not necessarily mean they do not exist. On the other hand, RNA-seq and microarray co-expression data, while not influenced by publication bias, does not give indications of actual experimentally demonstrated interactions (physical or genetic). Furthermore, RNA read counts do not directly correlate to protein numbers, with previous studies reporting that only 40% of the variation in protein concentration can be attributed to mRNA levels, an important aspect to consider when interpreting RNA-seq data [56]. Finally, the microarray network was constructed using the COXPRESdb (V6), which contains 73,083 human samples and offered another degree of validation [57]. Although RNA-seq reportedly detects more DEGs including ncRNAs [58], GeneFriends [59] contains 4133 human samples, far less than the microarray database from COXPRESdb. The protein-protein interaction network associated with CS We only used interactions from human proteins to build the CellAge PPI network. The network was built by taking the CellAge genes, their first-order partners and the interactions between them from the BioGrid database. The CellAge PPI network comprised of 2487 nodes across four disjointed components, three of which only comprised of two nodes each, and the main component containing 2481 nodes. The genes with the highest degree scores were TP53, HDAC1, BRCA1, EP300, and MDM2. These same genes also ranked in the top five CC. Expectedly, several of these genes also possessed the highest BC: TP53, BRCA1, HDAC1, and MDM2 (with BAG3, a gene with a slightly smaller degree also within the top 5). On the other hand, the genes ranked by top 5 IC were CCND1, CCND2, CDKN2A, SP1, and EGR1. Of note among these nodes, EP300, MDM2, CCND2, and EGR1 were not already present in CellAge. Additional file 2: Fig. S10 summarizes the gene intersection across the computed network parameters, while Additional file 1: Table S33 identifies potential senescence regulators not already present in CellAge from the PPI network. We found that from the top 12 PPI candidates, 11 have been recently shown to regulate senescence in human cell lines and will be added to CellAge build 2. Within the main PPI network component, a large portion of CS genes and their partners formed a single large module with 1595 nodes. Using DAVID version 6.8, we found the terms enriched within the module; the top five are: Transcription, DNA damage & repair, cell cycle, Proteasome & ubiquitin, and ATP pathway [35,36] (Additional file 1: Table S34). These results are all in line with previously described hallmarks of cellular senescence [60]. It is prudent to note that centrality measures in PPI networks must be interpreted with caution due to publication bias that can be an inherent part of the network [61,62]. The top network genes identified from the PPI network are likely to be heavily influenced by publication bias [63]. Looking at the average PubMed hits of the gene symbol in the title or abstract revealed a mean result count of approximately 2897 per gene, far higher than the genome average (136) or existing CellAge genes (712) (Additional file 2: Fig. S11). Unweighted RNA-Seq co-expression network We used CellAge genes that induce and inhibit CS and their co-expressing partners to build a cellular senescence co-expression network. The network consists of a main connected network with 3198 nodes, and a number of smaller "islands" that are not connected to the main network (Fig. 4a). The main interconnected network included 130 Cel-lAge genes. Among these, we also found that 14% of them are also human aging-related genes, reported in GenAge -Human dataset, whereas the remainder of the smaller networks only comprised of 1.6% longevity genes [64]. Next, we looked at a number of centrality parameters to see how CellAge genes are characterized compared to the entire network. CellAge genes had a mean BC of 0.00363, whereas the remainder of the genes had a BC of 0.00178, revealing that if CellAge genes are removed, modules within the network may become disconnected more easily. While nodes scoring highly for BC in PPI networks are likely bottleneck regulators of gene expression, this is not necessarily true for coexpression networks. In this case, nodes can also have high BC scores if they are co-activated via various signalling pathways. Although BC alone is not enough to determine which genes are regulating CS, taking BC into account with other network topological parameters can be a good indicator of gene function. Aside from high BC, CellAge genes also had a lower local clustering coefficient of 0.58, compared to a mean of 0.76 across non-CellAge genes, indicating that locally, CellAge genes connect to other genes less than the average for the network. This can also be seen at the degree level, where CellAge genes averaged only 53 connections, compared to an average of 103 connections in non-CellAge genes. Finally, the mean CC score was not significantly different between CellAge nodes and other genes in the network (0.148 in CellAge vs 0.158). CellAge genes were therefore more likely to be bottlenecks in signalling across different modules and occupy localized areas with lower network redundancy, suggesting that perturbations in their expression might have a greater impact on linking different underlying cellular processes. The topological analysis of the main network component as a whole revealed a more modular topology than the PPI network, resulting in genes tending not to appear in multiple measures of centrality. There were 23 nodes with significant IC with senescence-related genes, including PTPN6, LAPTM5, CORO1A, CCNB2 and HPF1. No node from the top 5 IC was present in the top 5 genes with high BC, CC, or Degree. Overall, the primary candidates of interest included KDM4C, which had a significant IC and was at the top 1% of CC and top 5% of BC, along with PTPN6, SASH3 and ARHGAP30, which all had significant IC values and were at the top Table S35, genes in Additional file 1: Table S36). b RNA-Seq Unweighted Co-expression Network, local clustering. Red/Orange represents nodes with high clustering coefficient, whereas pale green represents nodes with lower clustering coefficient. Degree is also weighted using node size. CellAge nodes are colored purple, and GenAge Human nodes are also shown and highlighted in bright green. The right-hand panel is an enlarged view of the left-hand panel 5% of BC. We found that KDM4C and PTPN6 have been shown to regulate CS in human cell lines, and will be added to build 2 of CellAge [65,66]. Previous studies have advocated that measures of centrality are generally important to identify key network components, with BC being one of the most common measures. However, it has also been postulated mathematically that intra-modular BC is more important than intermodular BC [67]. Therefore, by isolating network clusters of interest and identifying genes with high BC or centrality within submodules, we propose to identify new senescence regulators from the co-expression network. Using the CytoCluster app (see Networks in Methods) [68], we found 54 clusters in the network, of which we represent the top clusters colored according to modularity (Module 1-16) or size (Module 17-19) (Fig. 4a). Reactome pathway enrichment for all main clusters highlighted cell cycle and immune system terms in the two largest clusters [35,36]. The largest cluster of 460 nodes (17 CellAge nodes, Module 4), possessed a high modularity score and was strongly associated with cell cycle genes, including the following general terms: Cell Cycle; Cell Cycle, Mitotic; Mitotic Prometaphase; Resolution of Sister Chromatid Cohesion; and DNA Repair. The second largest cluster (Module 16), however, had weak modularity (ranking 26); it comprised of 450 nodes (19 CellAge nodes) and was enriched for immunerelated pathways including: Adaptive Immune System; Innate Immune System; Immunoregulatory interactions between a Lymphoid and a non-Lymphoid cell; Neutrophil degranulation; and Cytokine Signaling in Immune system. Cluster 4 and Cluster 5 were not enriched for Reactome Pathways. A visual inspection showed a number of bottleneck genes between Module 1 and Module 16, consistent with the role of the immune system in clearance and surveillance of senescence cells and the secretion of immunomodulators by senescent cells [69] (Additional file 1: Table S35). We were also interested in visualizing areas in the network with a high local clustering coefficient, as this parameter represents areas with many neighborhood interactions and, therefore, more robust areas in the network. It was found that the two clusters of interest, enriched for cell cycle terms and immune system terms, overlapped with regions of lower clustering coefficient, potentially implying parts of the biological system with less redundancy in the underlying process. Figure 4b depicts regions of high local clustering coefficient in the network (orange) and regions less well connected locally (green). Unweighted microarray co-expression network We also made an unweighted microarray co-expression network built from the COXPRESdb database of microarray gene co-expression (V6) [57] (Additional file 2: Fig. S12). Compared with the RNA-seq co-expression network, the microarray network is significantly smaller, and only included 34% of the CellAge genes (Additional file 1: Table S32). However, we found that SMC4 was an important bottleneck in the microarray network, being in the top 5% CC and IC (Additional file 2: Fig. S12D and S12E). SMC4 was not independently associated with senescence despite being part of the condensing II complex, which is related to cell senescence [70]. Furthermore, SMC4 is associated with cell cycle progression and DNA repair, two key antagonist mechanisms of cell senescence development [71,72]. SMC4 has been linked to cell cycle progression, proliferation regulation, and DNA damage repair, in accordance to the most significantly highlighted functional clusters in the module 2 and in the whole Microarray network (Additional file 1: Table S39 and S40; Additional file 2: Fig. S13) [73,74]. There was limited overlap between the microarray co-expression network and the RNA-seq co-expression network, although this is not surprising considering the higher specificity and sensitivity, and ability to detect low-abundance transcripts of RNAseq [75]. Experimental validation of senescence candidates We set out to test if candidate genes from our network analyses are indeed senescence inhibitors using a siRNA-based approach, whereby knockdowns enable the p16 and/or the p21 senescence pathway to be induced, leading to senescence [76]. We tested 26 potential senescence inhibitor candidates, 20 of which were chosen using GeneFriends, a guilt-by-association database to find co-expressed genes [59]. For this, we used the CellAge CS inhibitors as seed genes, with the assumption that genes co-expressed with senescence inhibitors would also inhibit senescence, and generated a list of the top co-expressed genes with CS inhibitors based on RNA-seq data (Additional file 1: Table S41). Furthermore, CellAge has multiple ways of partitioning genes, including the type of senescence the genes are involved in (Fig. 1b). We decided to look for genes co-expressed with stress-induced premature senescence (SIPS) inhibitors. We generated a list of genes that are co-expressed with the CellAge SIPS genes (Additional file 1: Table S42). We chose to validate five additional genes that were both co-expressed with the CellAge SIPS and are present as underexpressed in our signature of CS [32]. Finally, we chose SMC4 from the microarray network due to its interaction with other senescence genes within the network, its association with cell cycle progression, and the fact that it is underexpressed in senescent cells, indicating it may be inhibiting senescence in replicating cells. The genes chosen, along with experimental validation results are shown in Fig. 5, while the justification for our validation and Z-scores are shown in Additional file 1: Table S43 and S44 respectively. Next, we performed transient siRNA transfections of normal human fibroblasts using the 26 candidates and identified those siRNAs that generated the induction of a senescence phenotype, using multiparameter analysis of morphological measures and a panel of senescence markers. Senescence induction is associated with a loss of proliferation, as measured by a decrease in Ki67 index and cell number, and changes in cellular morphology, as measured by an increase in cell and nuclear area. We also quantitated changes in p16 and p21 (key senescence effectors [76]), interleukin 6 (IL-6, a common SASP marker) and SA-β-galactosidase. Knockdown of cyclophilin B, a housekeeper, acted as a negative control [2], while knockdown of CBX7, a potent senescence inhibitor, was included as a positive control for senescence induction [77]. Of the 26 genes tested, 80.7% (21/26) . a DAPI (blue) and Ki67 (green). b DAPI (blue) and Cell Mask (red). c DAPI (blue), p16 (green) and p21 (red). d DAPI (blue) and IL-6 (red). e Brightfield images following staining for SA-β-galactosidase. Size bar, 100 μm. f Heatmap of multiparameter analysis of proliferation markers (cell number and % Ki67 positive), senescence-associated morphology (cellular and nuclear area) and senescence markers (% p16 positive, p21 intensity, perinuclear IL-6 and perinuclear SA-β-galactosidase). Colors illustrate the number of Z-scores the experimental siRNA is from the cyclophilin B (cycloB) negative control mean. Data are ranked by whether or not the siRNA is a top hit (siRNAs between the thick horizontal lines), and then by the cell number Z-score. Red values indicate Z-scores that are "senescence-associated measures." The CBX7 positive control is also shown for comparison. Data presented are from at least two independent experiments each performed with a minimum of three replicates. All Z-scores are available in Additional file 1: Table S44 resulted in a decrease in Ki67 positive nuclei greater than 1 Z-score (i.e., direction of change also observed for the CBX7 siRNA positive control, Fig. 5; Additional file 1: Table S44) Discussion CellAge aims to be the benchmark database of genes controlling cellular senescence and we expect it to be an important new resource for the scientific community. The development of CellAge has also provided us with the means to perform systematic analyses of CS. While showcasing the functionality of CellAge in this manuscript, we have also explored the links between CS and aging, ARDs, and cancer. At the same time, we have aimed to expand the knowledge on both the evolution and function of senescence genes, and on how CS genes interact and form genetic networks. We showed that the use of CellAge may help in identifying new senescencerelated genes and we have validated several such genes experimentally. As the body of knowledge around senescence grows, it is our aim to maintain a quality resource to allow integrative analyses and guide future experiments. We began our CellAge analysis by gaining further insight into the function of CellAge genes (Additional file 2: Fig. S3). Unsurprisingly, inducers of CS were enriched for both VEGF and TNF signalling (Additional file 1: Table S1 and S2). Secretion of VEGF is a component of the senescence phenotype and has been shown to contribute towards cancer progression [78]. Interestingly, the CellAge genes are more strongly conserved in mammals compared to other protein-coding genes, an effect not seen in worms, yeast, or flies (Additional file 1: Table S4; Additional file 2: Fig. S1A and S1B). Given the role that many of the senescence genes in CellAge play in regulating the cell cycle, it makes sense that they are evolutionarily conserved; it is not entirely surprising that there is a greater evolutionary pressure towards conserving cell cycle tumor-suppressor genes than there is towards conserving other genes. Notably, the pattern of evolutionary conservation of CS genes was found to be almost identical to that of cancer-associated genes, apparently reflecting the co-evolution between these two phenomena [53]. Nonetheless, evolutionary genomics in a comparative context allows us to have a more comprehensive understanding of the genetic bases in important phenotypic traits, like longevity [79]. During their evolutionary history, it is possible that longlived species found ways to more efficiently solve problems related to the aging process [80,81]. Lineages where naturally important gene regulators (e.g., TP53) have alternative molecular variants or have been lost from their genomes [82,83] can be investigated as natural knockouts [84], since they have found a different way to solve agingrelated diseases like cancer [85,86]. We also found that the evolutionary distance between long-lived species is randomly distributed (Additional file 2: Fig. S1D; Additional file 4). Since longevity is a plastic trait that is related to multiple factors in the evolutionary history of the organisms (e.g., reproduction, body mass, habitat, metabolism, risk of predation), the way in which these genes evolved could be independent in the long-lived species analyzed. The relationship between CS and longevity was highlighted across various sections of this manuscript. The inducers of senescence were significantly overrepresented in the anti-longevity human orthologues, while the inhibitors of senescence were even more overrepresented in the pro-longevity human orthologues (Additional file 1: Table S7) [34]. Furthermore, both the CellAge regulators of CS and the overexpressed signatures of CS were significantly overrepresented in the overexpressed aging signatures from the human, rat, and mouse aging signature meta-analysis [42]. Interestingly, we found that the overexpressed signatures of replicative CS overexpressed with age were significantly enriched for regulated exocytosis (including leukocyte activation), cell proliferation, and aging (Additional file 1: Table S10; Additional file 2: Fig. S3B). The SASP is a known inducer of chronic inflammation in aged tissue [12,13], and the enrichment of terms relating to leukocyte activation highlights the role CS plays in activating the immune system via inflammatory factors with age. One tissue that consistently showed different CS expression patterns with age was the uterus. This observations was already noted in a previous study which also observed that DEGs downregulated in cancer were upregulated with age and DEGs upregulated in cancer were downregulated with age in six tissues, but not in the uterus [32]. CS genes are not expressed in a tissue-specific manner (Additional file 1: Table S11; Additional file 2: Fig. S4) and less than half of the CS genes undergo a significant change in expression with age ( Fig. 2; Additional file 2: Fig. S5A), suggesting that the pathways triggering differential expression of CS genes with age are shared between cells across tissues. Indeed, we found that CDKN2A was overexpressed in 19 human tissues with age, albeit only significantly so in 10 of the tissues (Additional file 1: Table S18) [32]. Nonetheless, across all simulations, CS genes significantly overexpressed across multiple tissues with age by chance never exceeded seven tissues ( Fig. 2d; Additional file 1: Table S17 and S19). The significant increase in CDKN2A expression across a significant number of human tissues with age is an indicator that at least some cell types are undergoing CS with age. ZMAT3, an inhibitor of CS, was also significantly overexpressed with age in seven tissues, including blood vessel, lung, and prostate, which also had significant increases in CDKN2A expression. Indeed, both ZMAT3 and CDKN2A were overexpressed across the majority of GTEx tissues with age (Additional file 2: Fig. S5D). Furthermore,~40% of the CellAge database was compiled using experiments exclusively in human fibroblast cell lines. Of the 20 studies used to compile the signatures of CS, 10 also performed gene manipulation experiments on fibroblasts [32]. Fibroblasts are present in connective tissues found between other tissue types across the human body, and the tissue samples analyzed to compile GTEx likely contained fibroblast gene expression. This may partially explain the lack of tissue-specific CellAge genes. It is further unclear whether the trends in differential expression of the Cel-lAge genes we see across aged human tissue samples is a result of fibroblast senescence, or if heterogenous gene populations are undergoing CS. We have partially addressed this issue by doing subgroup analysis of CellAge genes confirmed to control senescence outside of fibroblast cell lines and found that the overlap between these genes and both the signatures of aging and cancer genes is still significant. We found a strong association between senescence and neoplastic diseases (Additional file 1: Table S21). This is not surprising given the known role of senescence in tumor suppression. Some CS genes were also shared between many of the ARD classes. These results are in line with a previous analysis investigating the relationship between CS and ARD genes carried out using different datasets [53]. Tacutu et al reported significant overlaps (i.e., 138 genes -53%in common between CS and cancer vs 21-8%between CellAge and neoplasms); many more than we did. The study found that many genes shared between CS and several non-cancer ARDs are also involved in cancer. While removing cancer genes from our ARD dataset did not result in such a striking effect, it nonetheless substantially cut the number of overlaps to a statistically insignificant level, adding weight to the hypothesis that cancer genes have a bridging role between CS and ARDs. Furthermore, we found a significant overlap between both the CellAge inhibitors and inducers of senescence, and oncogenes and TSG (Fig. 3). Genes that induce senescence, however, tended to be tumor suppressors, while genes that inhibit senescence tended to be oncogenes, a finding that is consistent with the classical view of cellular senescence as a tumor-suppressor mechanism. We next explored what information could be obtained by applying a network analysis to CellAge. From the list of CellAge genes, three networks of CS were generated: a PPI network and two co-expression networks, with the aim of identifying new senescence regulators based primarily on network centrality of the genes. The examination of the PPI network to identify possible regulators based on centrality revealed 25 central genes in the network, ranking in the top 1% in at least two network topological parameters (degree, BC, CC, or IC) (Additional file 1: Table S33). However, 13 of these genes are already in the CellAge database, and we found 11 of these genes have already been shown to drive CS in human cell lines and will be added into build 2 of CellAge. We looked at the RNA-Seq co-expression network in detail, using the main connected component of 3198 genes to find highly central genes to the network as a whole, and those occupying subnetworks of interest. The RNA-Seq was a highly modular network, separated into some subnetworks of distinct functions (Fig. 4). The two largest and more central networks contained a number of known senescence genes. We expanded the analysis of these networks in particular, identifying a number of bottleneck nodes. Cluster 1 was enriched for cell cycle processes, which is not overly surprising given that senescence involves changes in cell cycle progression. However, cluster 2 comprised of enriched terms relating to immune system function. One of the aims in biogerontology is to understand and reverse the effects of aging on the immune system. Additional file 1: Table S38 highlights the genes in both clusters that are potential CS bottlenecks within the network and may warrant further study. Using siRNAs, we were able to test the potential role of 26 gene candidates in inhibiting senescence (Fig. 5). The list of candidates was primarily compiled using CellAge inhibitors as seeds to generate co-expressed genes in Gen-eFriends, a collection of RNA-seq co-expression data [59] (Additional file 1: Table S43). Of the 26 genes, 13 were top hits, decreasing cell number, altering at least one morphological measure, and activating the p16 and/or p21 pathway. Additional file 1: Table S45 highlights the four CS candidates we found that have not yet been associated with senescence. We have showcased how co-expression networks can be used to accurately infer senescence gene candidates, which can then be experimentally verified. Conclusion Overall, our CellAge database is the first comprehensive cellular senescence database, which will be a major resource for researchers to understand the role of senescence in aging and disease. Besides, we found that CS genes are conserved in vertebrates but not invertebrates and that genes related to the CS tend not to be tissuespecific. We observed that genes inducing CS trended towards upregulation with age across most human tissues, and these genes are overrepresented in both antilongevity and tumor-suppressing gene datasets, while genes inhibiting senescence were not overexpressed with age and were overrepresented in pro-longevity and oncogene datasets. CS genes were also overrepresented in genes linked to aging-related diseases, primarily in neoplasms. Using network biology, we implicated the CellAge genes in various processes, particularly cell division and immune system processes. We used network topology to identify potential regulators of CS and bottlenecks that could impact various downstream processes if deregulated. Indeed, we identified 11 genes that have already been shown to contribute towards CS, which will be added to future versions of CellAge. Finally, we experimentally verified 26 genes that induce CS morphology or biomarkers when knocked down in human mammary fibroblasts. Of these, 13 genes (C9orf40, CDC25A, CDCA4, CKAP2, GTF3C4, HAUS4, IMMT, MCM7, MTHFD2, MYBL2, NEK2, NIPA2, and TCEB3) were strong hits in inducing a senescent phenotype. Cellular senescence is one of the hallmarks of aging [87] and the accumulation of senescent cells in human tissues with age has been implicated as a driver of agingrelated diseases. Indeed, pharmacological approaches targeting senescent cells, like senolytics, are a major and timely area of research that could result in human clinical applications [5,88]. It is imperative that we fully understand and deconstruct cellular senescence in order to target aging-related diseases. We hope that CellAge will help researchers understand the role that CS plays in aging and aging-related diseases and contributes to the development of drugs and strategies to ameliorate the detrimental effects of senescent cells. Methods CellAge compilation CellAge was compiled following a scientific literature search, manual curation, and annotation, with genes being appended to the database if they met the following criteria: Only gene manipulation experiments (gene knockout, gene knockdown, partial or full loss-of-function mutations, overexpression or drug-modulation) were used to identify the role of the genes in cellular senescence. The search focussed on genes from genetic manipulation experiments to ensure objectivity in the selection process. The genetic manipulation caused cells to induce or inhibit the CS process in the lab. Cellular senescence was detected by growth arrest, increased SA-βgalactosidase activity, SA-heterochromatin foci, a decrease in BrdU incorporation, changes in morphology, and/or specific gene expression signatures. The experiments were performed in primary, immortalized, or cancer human cell lines. 40% of the experiments were conducted exclusively in fibroblasts. The data was compiled from 230 references. The curated database comprises cell senescence genes together with a number of additional annotations useful in understanding the context of each identified CS gene (Additional file 1: Table S46). We categorized genes according to three types of senescence: replicative, oncogene-induced or stress-induced. Replicative senescence was the default category, while genes were listed as oncogene-induced if the reference explicitly mentioned the gene induced or delayed oncogene-induced senescence. Finally, stress-induced senescence was used to indicate that the gene was necessary to induce or inhibit senescence caused by external stressors like drugs/chemicals, serum deprivation, or radiation. We also recorded whether a gene induces or inhibits CS. For example, a gene whose overexpression is associated with increased senescence is classified with the "induces" tag, whereas if the overexpression of a gene inhibits senescence, then it is classified with the "inhibits" tag. Similarly, if the knockout or knockdown of a gene induces senescence, then it is recorded with the "inhibits" tag. Together with the annotations identified in Additional file 1: Table S46, we also incorporated a number of secondary annotations into the database such as various gene identifiers, the gene description, gene interaction(s), and quick links to each senescence gene. The CellAge database also provides crosslinks to genes in other HAGR resources, i.e., GenAge, GenDR and Longevity-Map, which we hope will enable inferences to be made regarding the link between human aging and CS. CellAge data sources Build 1 of CellAge resulted in a total of 279 curated cell senescence genes which we have incorporated into the HAGR suite of aging resources. The HAGR platform comprises a suite of aging databases and analysis scripts. The CellAge interface has been designed with the help of JavaScript libraries to enable more efficient retrieval and combinatorial searches of genes. As with the other HAGR databases, we have used PHP to serve the data via an Apache web server. The raw data can be downloaded via the main HAGR downloads page in CSV format or filtered and downloaded from the main search page. The first part of our work consisted in finding which genes driving CS are also associated with ARDs or with longevity, using the following data sources: Human genes associated with CS: CellAge build 1. Human genes associated with human aging: GenAge human build 19. Human orthologues of model organisms' genes associated with longevity: proOrthologuesPub.tsv and antiOrthologuesPub.tsv file (https://github.com/ maglab/genage-analysis/blob/master/Dataset_4_ aging_genes.zip) [34]. Human oncogenes: Oncogene database (http:// ongene.bioinfo-minzhao.org/index.html). Human tumor suppressor gene database: TSGene 2.0 (https://bioinfo.uth.edu/TSGene/index.html). Human genes associated with ARDs (https://github. com/maglab/genage-analysis/blob/master/ Dataset_5_disease_genes.zip) [34]. This data concerns the 21 diseases with the highest number of gene associations, plus asthma, a non-aging-related respiratory system disease used as a control. Human genes differentially expressed with age from the GTEx project (v7, January 2015 release) [32,43]. CellAge data analysis Statistical significance was determined by comparing the p-value of overlapping CellAge gene symbols with the different data sources, computed via a hypergeometric distribution and Fisher's exact test. We used PubMed to understand the relative research focus across the protein-coding genome and incorporate this into the analysis to account for publication bias. We used Bio-Mart to obtain approximately 19,310 protein-coding genes, then using an R script we queried NCBI for the publication results based on the gene symbol using the following query [89,90]: ("GENE_SYMBOL"[Title/Abstract] AND Homo [ORGN]) NOT Review [PTYP] The GENE_SYMBOL was replaced in the above query by each of the genes in turn. Certain genes were removed as they matched common words and, therefore, skewed the results: SET, SHE, PIP, KIT, CAMP, NODAL, GC, SDS, CA2, COPE, TH, CS, TG, ACE, CAD, REST, HR, and MET. The result was a dataframe in R comprising variables for the "gene" and the "hits." We used the R package called "rentrez" to query PubMed for the result count [91]. Evolution of CellAge genes The percentage of CellAge genes with orthologues in Rhesus macaque, Rattus norvegicus, Mus musculus, Saccharomyces cerevisiae, Caenorhabditis elegans, and Drosophila melanogaster were found using Biomart version 88 by filtering for genes with "one2one" homology and an orthology confidence score of one [89]. We also found the total number of human genes with orthologues in the above species using Biomart. Significance was assessed using a two-tailed z-test with BH correction. The phylogenetic arrangement included twenty-four species representative of major mammalian groups. The genomes were downloaded in CDS FASTA format from Ensembl (http://www.ensembl.org/) and NCBI (https:// www.ncbi.nlm.nih.gov/) (Additional file 1: Table S6). To remove low quality sequences we used the clustering algorithm of CD-HITest version 4.6 [92] with a sequence identity threshold of 90% and an alignment coverage control of 80%. The longest transcript per gene was kept using TransDecoder.LongOrfs and TransDecoder. Predict (https://transdecoder.github.io) with default criteria [93]. In order to identify the orthologs of the 279 CellAge human genes in the other 23 mammalian species, the orthology identification analysis was done using OMA standalone software v. 2.3.1 [41]. This analysis makes strict pairwise sequence comparisons "all-againstall," minimizing the error in orthology assignment. The orthologous pairs (homologous genes related by speciation events) are clustered into OrthoGroups (OG) [94]; this was done at the Centre for Genomic Research computing cluster (Linux-based) at the University of Liverpool. The time calibrated tree was obtained from TimeTree (http://www.timetree.org/) and the images were downloaded from PhyloPic (http://phylopic.org/). In order to structure the evolutionary distance for the CellAge genes between the five long-lived mammals and the others 19 mammalian species, the amino acid sequences from the 271 CellAge OrthoGroups were aligned using the L-INS-i algorithm from MAFFT v.7 [95]. Ambiguous and missing sites were removed from the alignments using the pxclsq function from phyx [96]. We concatenated the amino acid alignments using the concat function from AMAS [97] for the 271 CellAge genes. To analyze the variation of the CellAge genes in mammals, we obtained the branch lengths using loglikelihood for a fixed tree through IQ-TREE [98] for (a) the concatenated alignment (271 CellAge genes) and (b) the 22 CellAge genes conserved among the 24 mammalian species in order to understand the individual gene evolution. The topology of reference was the phylogenetic tree from TimeTree. We used the Faith's phylogenetic diversity index (PD) [99] through the "picante" R package [100] to calculate the evolutionary distances. The Faith's PD index was used to calculate the sum of the total phylogenetic branch length for one or multiples species. We calculated the observed Faith's PD from our data and we compared the results with the expected Faith's PD (expected.pd) using a binomial sampling with a fixed probability of each tip being sampled. Overlap analysis We conducted overlap analysis using R to understand how the CellAge genes and signatures of CS were differentially expressed with GenAge, ARD, and cancer genes. We also examined the overlap between CS genes and differentially expressed signatures of aging [42], and genes differentially expressed in various human tissues with age. Fisher's exact test was used on the contingency tables and significance was assessed by p values adjusted via Benjamini-Hochberg (BH) correction. For the comparison of genes differentially expressed in at least one tissue with age between the CS genes and the genome, some genes were differentially expressed in opposite directions across numerous tissues (Additional file 2: Fig. S5A). Genes differentially expressed in both directions were added to the overexpressed and underexpressed DEGs in each CS gene list, and to the total number of genes in the genome to compensate for the duplicate gene count (Additional file 1: Table S14 and S15). Fisher's exact test was also used to test for significance of tissue-specific CellAge gene expression. Significance of overlap analysis between CellAge and LAGs was computed using a hypergeometric distribution and FDR was corrected using Bonferroni correction. The GeneOverlap package in R was used to test for overlaps between the CellAge inducers and inhibitors of senescence, and the oncogenes and TSGs [101]. Results for all overlap analyses were plotted using the ggplot2 library [90,102]. Simulation of CS gene expression in human aging The RNA-seq gene expression data on GTEx was scrambled in such a way that all protein-coding genes in each tissue were assigned a random paired p and log 2 FC value from the original gene expression data of each respective tissue. The randomly sorted gene expression data was then filtered for significance (p < 0.05, moderated t-test with BH correction, absolute log 2 FC > log 2 (1.5)) [32,103], and the CellAge accessions were extracted and overlapped across all the simulated expression data in 26 tissues from GTEx. The probability of a CS gene being overexpressed or underexpressed across multiple tissues by chance was calculated across 10,000 simulations. Functional enrichment The analysis of CellAge included gene functional enrichment of the database. We used DAVID functional clustering (https://david.ncifcrf.gov/) to identify functional categories associated with CellAge [35,36]. The Overrepresentation Enrichment Analysis (ORA) of biological processes (Gene Ontology database) was done via the WEB-based Gene SeT AnaLysis Toolkit (WebGestalt) for the analysis of all CellAge genes, CellAge CS regulators and overexpressed signatures of CS overexpressed in the meta-analysis of aging signatures, and for the CellAge genes overlapping with tumor suppressor and oncogenes [38]. A p value cutoff of 0.05 was used, and p values were adjusted using BH correction. Redundant GO terms were removed and the remaining GO terms were grouped into categories based on their function using the default parameters on Reduce + Visualize Gene Ontology (REVIGO) [37]. Results were then visualized using and the R package treemap [104] (Fig. 1c; Additional file 2: Fig. S8A -S8D). Venn diagrams to represent gene overlaps were created using Venny [52] and the ggplot2 library [90,102]. Networks We used Cytoscape version 3.6.1 to generate networks and R version 3.3.1 to perform aspects of the statistical analysis [90,105]. The networks were built starting from a list of seed nodes-all genes included in build 1 of Cel-lAge, part of the Human Ageing Genomic Resources [28]. Network propagation was measured using the Cytoscape plugin Diffusion [106]. The analysis of the fit to the scale-free structure was calculated by the Network Analyzer tool of Cytoscape 3.2.1 [105]. Network analyzer is a Cytoscape plugin which performs topological analysis on the network and reports the pillar nodes on the network structure based on a series of mathematical parameters. Network analyzer also calculates the fit of the distribution of the number of edges per node to the power law distribution. A significant fit to the power law indicates the presence of a scale-free structure in the network [61,107]. The analysis was applied to the PPI network, the RNA-seq Unweighted Co-expression network, and the Microarray Unweighted Co-expression network of cellular senescence (Additional file 2: Fig. S9). The Network Analyzer tool was also used to calculate BC, CC, and IC in the networks. Protein-protein interaction network The protein-protein interaction network was built from the BioGrid database of physical multi-validated protein interactions (Biology General Repository for Interaction Datasets) version 3.4.160, using CellAge proteins as seed nodes and extracting the proteins encoded by CellAge genes as well as the first-order interactors of CellAge proteins [108]. After removing duplicated edges and self-loops, the network consisted of 2643 nodes and 16, 930 edges. The network was constructed and visualized in Cytoscape version 3.6.1. The "CytoCluster" App in Cytoscape was used to identify modules in the network with the following parameters: HC-PIN algorithm; Weak, Threshold = 2.0; ComplexSize Threshold = 1% [68]. Unweighted RNA-Seq co-expression network The RNA-seq co-expression network was built using CellAge data and RNA-Seq co-expression data taken from Genefriends (http://genefriends.org/RNAseq) [59]. The unweighted co-expression network was built applying the method of correlation threshold selection described by Aoki to the GeneFriends database of RNA-Seq co-expression version 3.1 [109]. Aoki initially designed this methodology for plant co-expression network analysis, but it has been successfully applied to build human networks [110]. The Pearson Correlation Coefficient (PCC) threshold which generated the database of edges with the lowest network density was selected. The network density is the proportion of existing edges out of all possible edges between all nodes. The lower the network density is the more nodes and fewer edges are included in the network. The lower the number of edges, the higher the minimum correlation in expression between each pair of genes represented by the edges. The higher the number of nodes, the higher the portion of nodes from CellAge included, and, therefore, the more representative the network is of the CellAge database. The PCC threshold of 0.65 generated the database of interactions of RNA-Seq co-expression with the lowest network density, 0.01482 (Additional file 2: Fig. S14A). The unweighted RNA-Seq network was generated and visualized in Cytoscape 3.6.1. Microarray co-expression network The microarray co-expression network was generated using the CellAge genes as seed nodes and their direct interactions and edges, derived using the COXPRESdb database of Microarray co-expression (version Hsa-m2.c2-0) [57]. PCC threshold of 0.53 created the Microarray database with the lowest network density, 1.006 × 10 − 2 (Additional file 2: Fig. S14B). The adjustment of the node-degree distribution to the power law distribution had a correlation of 0.900 and an R-squared of 0.456 (Additional file 2: Fig. S9C). The fit to the power law distribution confirmed the scale-free structure of the network. Experimental validation of new CS genes We used normal human mammary fibroblasts (HMFs) and siRNAs to find new CS regulators based on highranking co-expressed inhibitors of CS and SIPS inhibitors. We also tested SMC4 due to its high-scoring topological parameters within the microarray co-expression network (see Experimental Validation of Senescence Candidates in Results). Cell culture and reagents Fibroblasts were obtained from reduction mammoplasty tissue of a 16-year-old individual, donor 48 [111]. The cells were seeded at 7500 cells/cm 2 and maintained in Dulbecco's modified Eagle's medium (DMEM) (Life Technologies, UK) supplemented with 10% fetal bovine serum (FBS) (Labtech.com, UK), 2 mM L-glutamine (Life Technologies, UK) and 10 μg/mL insulin from bovine pancreas (Sigma). All cells were maintained at 37°C/5% CO 2 . All cells were routinely tested for mycoplasma and shown to be negative. siRNA knockdown experiments For high-content analysis (HCA), cells were forward transfected with 30 nM siRNA pools at a 1:1:1 ratio (Ambion) using Dharmafect 1 (Dharmacon) in 384-well format. Control siRNA targeting cyclophilin B (Dharmacon) or Chromobox homolog 7 (CBX7, Ambion) were also included as indicated. Cells were incubated at 37°C/ 5% CO 2 and medium changed after 24 h. Cells were then fixed/stained 96 h later and imaged as described below. The siRNA sequences are provided in Additional file 1: Table S47A and S47B. Z-score generation For each of the parameters analyzed, significance was defined as one Z-score from the negative control mean and average Z-scores from at least two independent experiments performed in at least triplicate are presented. Z-scores were initially generated on a per experiment basis according to the formula below: Z−score ¼ mean value of target siRNA À mean value for cyclophilin B siRNA ð Þ =standard deviation SD ð Þfor cyclophilin B siRNA: Immunofluorescence microscopy and high-content analysis Cells were fixed with 3.7% paraformaldehyde, permeabilized for 15 min using 0.1% Triton X and blocked in 0.25% BSA before primary antibody incubations. Primary antibodies used are listed in Additional file 1: Table S48. Cells were incubated for 2 h at room temperature with the appropriate AlexaFluor-488 or AlexaFluor-546 conjugated antibody (1:500, Invitrogen), DAPI, and CellMask Deep Red (Invitrogen). Images were acquired using the IN Cell 2200 automated microscope (GE), and HCA was performed using the IN Cell Developer software (GE). Supplementary information Supplementary information accompanies this paper at https://doi.org/10. 1186/s13059-020-01990-9. Additional file 1: Supplementary Tables. Excel file containing Supplementary Tables S1-S48. Fig. 2 2Differential expression of a CellAge inducers and inhibitors of CS and b differentially expressed signatures of CS in human tissues with age. Fig. 3 a 3Overlap between CellAge inducers and inhibitors, and oncogenes and tumor-suppressing genes. b Adjusted p value and odds ratio of the overlap analysis. The number of overlapping genes in each category was significant (p < 0.05, Fisher's exact test with BH correction). p values are shown in gray writing for each comparison. Data available in Additional file 1:Table S22-S27 Fig. 4 a 4Cluster analysis of the RNA-Seq Unweighted Co-expression Network. The 171 seed nodes obtained from CellAge and their first order interactors. The colours represent the breakdown of the network into clusters. The algorithm revealed 52 distinct clusters, of which we color and order the 19 clusters with the best rankings for modularity, or in the case of module 17-19, size. The CellAge nodes are colored in dark purple, appearing throughout the network. Larger nodes have higher betweenness centrality. In order of decreasing modularity, the main function clusters of the modules were related to; Spermatogenesis (Module 1), Synapse (Module 2), Cardiac muscle contraction (Module 3), Cell Cycle (Module 4), Secreted (Module 5), Tudor domain (Module 6), ATP-binding (Module 7), Symport (Sodium ion transport) (Module 8), DNA damage and repair (Module 9), transit peptide: Mitochondrion (Module 10), Steroid metabolism (Module 11), Transcription regulation (Module 12), Protein transport (Module 13), Mitochondrion (Module 14), Heme biosynthesis (Module 15), Innate immunity (Module 16), Signal peptide (Module 17), Keratinocyte (Module 18), and Transcription repression (Module 19) (Enrichment results in Additional file 1: Fig. 5 5Experimental validation of 26 senescence candidates. a-e Representative images of fibroblasts following transfection with cyclophilin B siRNA (top row), CBX7 siRNA (middle row), or GFT3C4 siRNA (bottom row) ; 80.7% (21/26) increased p16; 96.2% increased p21 (25/26); 65.4% increase IL-6; and 65.4% (17/ 26) increase SA-β-galactosidase. Of the siRNAs that resulted in a decrease in Ki67 index, 61.9% (13/21) were classified as top hits as they concomitantly decreased cell number and altered at least one morphological measure. 92.3% (12/13) of the top hits activated both the p16 and p21 pathway, 84.6% (11/13) upregulated the SASP factor IL-6, while 61.5% (8/13) generated an increase in the percentage of SA-β-galactosidase positive cells. In general, we have shown the power of networks in predicting gene function, with 13 "top hits" (GTF3C4, C9orf40, HAUS4, MCM7, TCEB3, CDC25A, CDCA4, CKAP2, MTHFD2, NEK2, IMMT, MYBL2, and NIPA2). Additional file 5. GTEx simulated expression script. R script to find expected number of overlaps between GTEx tissue DEGs with age when gene names are scrambled.Additional file 6. Review history.Peer review informationYixin Yao was the primary editor on this article and managed its editorial process and peer review in collaboration with the rest of the editorial team.Review historyThe review history is available as Additional file 6.Availability of data and materialsSupplementary figures and citations, tables, and files are available on the Integrative Genomics of Ageing Group CellAge_supplementary GitHub repository, along with an R script to recreate the scrambled GTEx gene expression data (Additional file 5) (https://github.com/maglab/CellAge_ supplementary)[112]. CellAge, GenAge, and disease genes are available on HAGR (https://genomics.senescence.info/)[28,34]. Tissue-specific differentially expressed genes with age and signatures of cellular senescence are from[32].Ethics approval and consent to participate Not applicable.Consent for publicationNot applicable.Competing interestsThe authors declare that they have no competing interests.Author details The serial cultivation of human diploid cell strains. L Hayflick, P S Moorhead, Exp Cell Res. 25Hayflick L, Moorhead PS. The serial cultivation of human diploid cell strains. Exp Cell Res. 1961;25:585-621. The essence of senescence. T Kuilman, C Michaloglou, W J Mooi, D S Peeper, Genes Dev. 2422Kuilman T, Michaloglou C, Mooi WJ, Peeper DS. The essence of senescence. Genes Dev. 2010;24(22):2463-79. Telomere shortening triggers senescence of human cells through a pathway involving ATM, p53, and p21(CIP1), but not p16(INK4a). U Herbig, W A Jobling, B P Chen, D J Chen, J M Sedivy, Mol Cell. 144Herbig U, Jobling WA, Chen BP, Chen DJ, Sedivy JM. Telomere shortening triggers senescence of human cells through a pathway involving ATM, p53, and p21(CIP1), but not p16(INK4a). Mol Cell. 2004;14(4):501-13. A DNA damage checkpoint response in telomere-initiated senescence. F Di Fagagna, P M Reaper, L Clay-Farrace, H Fiegler, P Carr, Von Zglinicki, T , Nature. 4266963d'Adda di Fagagna F, Reaper PM, Clay-Farrace L, Fiegler H, Carr P, Von Zglinicki T, et al. A DNA damage checkpoint response in telomere-initiated senescence. Nature. 2003;426(6963):194-8. Stress, cell senescence and organismal ageing. J P De Magalhaes, J F Passos, Mech Ageing Dev. 170de Magalhaes JP, Passos JF. Stress, cell senescence and organismal ageing. Mech Ageing Dev. 2018;170:2-9. Senescence is a developmental mechanism that contributes to embryonic growth and patterning. M Storer, A Mas, A Robert-Moreno, M Pecoraro, M C Ortells, Di Giacomo, V , Cell. 1555Storer M, Mas A, Robert-Moreno A, Pecoraro M, Ortells MC, Di Giacomo V, et al. Senescence is a developmental mechanism that contributes to embryonic growth and patterning. Cell. 2013;155(5):1119-30. Programmed cell senescence during mammalian embryonic development. D Munoz-Espin, M Canamero, A Maraver, G Gomez-Lopez, J Contreras, S Murillo-Cuesta, Cell. 1555Munoz-Espin D, Canamero M, Maraver A, Gomez-Lopez G, Contreras J, Murillo-Cuesta S, et al. Programmed cell senescence during mammalian embryonic development. Cell. 2013;155(5):1104-18. An essential role for senescent cells in optimal wound healing through secretion of PDGF-AA. M Demaria, N Ohtani, S A Youssef, F Rodier, W Toussaint, J R Mitchell, Dev Cell. 316Demaria M, Ohtani N, Youssef SA, Rodier F, Toussaint W, Mitchell JR, et al. An essential role for senescent cells in optimal wound healing through secretion of PDGF-AA. Dev Cell. 2014;31(6):722-33. Senescence surveillance of pre-malignant hepatocytes limits liver cancer development. T W Kang, T Yevsa, N Woller, L Hoenicke, T Wuestefeld, D Dauch, Nature. 4797374Kang TW, Yevsa T, Woller N, Hoenicke L, Wuestefeld T, Dauch D, et al. Senescence surveillance of pre-malignant hepatocytes limits liver cancer development. Nature. 2011;479(7374):547-51. Cellular senescence: Immunosurveillance and future immunotherapy. Dga Burton, A Stolzing, Ageing Res Rev. 43Burton DGA, Stolzing A. Cellular senescence: Immunosurveillance and future immunotherapy. Ageing Res Rev. 2018;43:17-25. Aging, cellular senescence, and cancer. J Campisi, Annu Rev Physiol. 75Campisi J. Aging, cellular senescence, and cancer. Annu Rev Physiol. 2013; 75:685-705. The role of senescent cells in ageing. J M Van Deursen, Nature. 5097501van Deursen JM. The role of senescent cells in ageing. Nature. 2014; 509(7501):439-46. A complex secretory program orchestrated by the inflammasome controls paracrine senescence. J C Acosta, A Banito, T Wuestefeld, A Georgilis, P Janich, J P Morton, Nat Cell Biol. 158Acosta JC, Banito A, Wuestefeld T, Georgilis A, Janich P, Morton JP, et al. A complex secretory program orchestrated by the inflammasome controls paracrine senescence. Nat Cell Biol. 2013;15(8):978-90. Naturally occurring p16(Ink4a)-positive cells shorten healthy lifespan. D J Baker, B G Childs, M Durik, M E Wijers, C J Sieben, J Zhong, Nature. 5307589Baker DJ, Childs BG, Durik M, Wijers ME, Sieben CJ, Zhong J, et al. Naturally occurring p16(Ink4a)-positive cells shorten healthy lifespan. Nature. 2016; 530(7589):184-9. Clearance of p16Ink4a-positive senescent cells delays ageingassociated disorders. D J Baker, T Wijshake, T Tchkonia, N K Lebrasseur, B G Childs, B Van De Sluis, Nature. 4797372Baker DJ, Wijshake T, Tchkonia T, LeBrasseur NK, Childs BG, van de Sluis B, et al. Clearance of p16Ink4a-positive senescent cells delays ageing- associated disorders. Nature. 2011;479(7372):232-6. Directed elimination of senescent cells by inhibition of BCL-W and BCL-XL. R Yosef, N Pilpel, R Tokarsky-Amiel, A Biran, Y Ovadya, S Cohen, Nat Commun. 711190Yosef R, Pilpel N, Tokarsky-Amiel R, Biran A, Ovadya Y, Cohen S, et al. Directed elimination of senescent cells by inhibition of BCL-W and BCL-XL. Nat Commun. 2016;7:11190. Targeted apoptosis of senescent cells restores tissue homeostasis in response to chemotoxicity and aging. M P Baar, Rmc Brandt, D A Putavet, Jdd Klein, Kwj Derks, Brm Bourgeois, Cell. 1691e16Baar MP, Brandt RMC, Putavet DA, Klein JDD, Derks KWJ, Bourgeois BRM, et al. Targeted apoptosis of senescent cells restores tissue homeostasis in response to chemotoxicity and aging. Cell. 2017;169(1): 132-47. e16. Reconstitution of telomerase activity in normal human cells leads to elongation of telomeres and extended replicative life span. H Vaziri, S Benchimol, Curr Biol. 85Vaziri H, Benchimol S. Reconstitution of telomerase activity in normal human cells leads to elongation of telomeres and extended replicative life span. Curr Biol. 1998;8(5):279-82. Unmasking transcriptional heterogeneity in senescent cells. A Hernandez-Segura, T V De Jong, S Melov, V Guryev, J Campisi, M Demaria, Curr Biol. 2717e4Hernandez-Segura A, de Jong TV, Melov S, Guryev V, Campisi J, Demaria M. Unmasking transcriptional heterogeneity in senescent cells. Curr Biol. 2017; 27(17):2652-60. e4. Scoring of senescence signalling in multiple human tumour gene expression datasets, identification of a correlation between senescence score and drug toxicity in the NCI60 panel and a pro-inflammatory signature correlating with survival advantage in peritoneal mesothelioma. K Lafferty-Whyte, A Bilsland, C J Cairney, L Hanley, N B Jamieson, N Zaffaroni, BMC Genomics. 11532Lafferty-Whyte K, Bilsland A, Cairney CJ, Hanley L, Jamieson NB, Zaffaroni N, et al. Scoring of senescence signalling in multiple human tumour gene expression datasets, identification of a correlation between senescence score and drug toxicity in the NCI60 panel and a pro-inflammatory signature correlating with survival advantage in peritoneal mesothelioma. BMC Genomics. 2010;11:532. Stressing the cell cycle in senescence and aging. H Chandler, G Peters, Curr Opin Cell Biol. 256Chandler H, Peters G. Stressing the cell cycle in senescence and aging. Curr Opin Cell Biol. 2013;25(6):765-71. A biomarker that identifies senescent human cells in culture and in aging skin in vivo. G P Dimri, X Lee, G Basile, M Acosta, G Scott, C Roskelley, Proc Natl Acad Sci. 9220Dimri GP, Lee X, Basile G, Acosta M, Scott G, Roskelley C, et al. A biomarker that identifies senescent human cells in culture and in aging skin in vivo. Proc Natl Acad Sci U S A. 1995;92(20):9363-7. Forging a signature of in vivo senescence. N E Sharpless, C J Sherr, Nat Rev Cancer. 157Sharpless NE, Sherr CJ. Forging a signature of in vivo senescence. Nat Rev Cancer. 2015;15(7):397-408. Beta-galactosidase--an indicator of the maturational stage of mouse and human mononuclear phagocytes. I Bursuker, J M Rhodes, R Goldman, J Cell Physiol. 1123Bursuker I, Rhodes JM, Goldman R. Beta-galactosidase--an indicator of the maturational stage of mouse and human mononuclear phagocytes. J Cell Physiol. 1982;112(3):385-90. Beta-galactosidase staining on bone marrow. The osteoclast pitfall. H G Kopp, A T Hooper, S V Shmelkov, S Rafii, Histol Histopathol. 229Kopp HG, Hooper AT, Shmelkov SV, Rafii S. Beta-galactosidase staining on bone marrow. The osteoclast pitfall. Histol Histopathol. 2007;22(9):971-6. Connecting autophagy to senescence in pathophysiology. A R Young, M Narita, Curr Opin Cell Biol. 222Young AR, Narita M. Connecting autophagy to senescence in pathophysiology. Curr Opin Cell Biol. 2010;22(2):234-40. The meaning of p16(ink4a) expression in tumors: functional significance, clinical associations and future developments. A K Witkiewicz, K E Knudsen, A P Dicker, E S Knudsen, Cell Cycle. 1015Witkiewicz AK, Knudsen KE, Dicker AP, Knudsen ES. The meaning of p16(ink4a) expression in tumors: functional significance, clinical associations and future developments. Cell Cycle. 2011;10(15):2497-503. Human ageing genomic resources: new and updated databases. R Tacutu, D Thornton, Johnson E Budovsky, A Barardo, D Craig, T , Nucleic Acids Res. 46D1Tacutu R, Thornton D, Johnson E, Budovsky A, Barardo D, Craig T, et al. Human ageing genomic resources: new and updated databases. Nucleic Acids Res. 2018;46(D1):D1083-D90. Network medicine: a network-based approach to human disease. A L Barabasi, N Gulbahce, J Loscalzo, Nat Rev Genet. 121Barabasi AL, Gulbahce N, Loscalzo J. Network medicine: a network-based approach to human disease. Nat Rev Genet. 2011;12(1):56-68. Longevity network: construction and implications. A Budovsky, A Abramovich, R Cohen, V Chalifa-Caspi, V Fraifeld, Mech Ageing Dev. 1281Budovsky A, Abramovich A, Cohen R, Chalifa-Caspi V, Fraifeld V. Longevity network: construction and implications. Mech Ageing Dev. 2007;128(1):117-24. Interactome networks and human disease. M Vidal, M E Cusick, A L Barabasi, Cell. 1446Vidal M, Cusick ME, Barabasi AL. Interactome networks and human disease. Cell. 2011;144(6):986-98. A human tissuespecific transcriptomic analysis reveals a complex relationship between aging, cancer, and cellular senescence. K Chatsirisupachai, D Palmer, S Ferreira, J P De Magalhaes, Aging Cell. 18613041Chatsirisupachai K, Palmer D, Ferreira S, de Magalhaes JP. A human tissue- specific transcriptomic analysis reveals a complex relationship between aging, cancer, and cellular senescence. Aging Cell. 2019;18(6):e13041. The DrugAge database of aging-related drugs. D Barardo, D Thornton, H Thoppil, M Walsh, S Sharifi, S Ferreira, Aging Cell. 163Barardo D, Thornton D, Thoppil H, Walsh M, Sharifi S, Ferreira S, et al. The DrugAge database of aging-related drugs. Aging Cell. 2017;16(3):594-7. Systematic analysis of the gerontome reveals links between aging and agerelated diseases. M Fernandes, C Wan, R Tacutu, D Barardo, A Rajput, J Wang, Hum Mol Genet. 2521Fernandes M, Wan C, Tacutu R, Barardo D, Rajput A, Wang J, et al. Systematic analysis of the gerontome reveals links between aging and age- related diseases. Hum Mol Genet. 2016;25(21):4804-18. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. W Huang D, B T Sherman, R A Lempicki, Nat Protoc. 41Huang d W, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. 2009;4(1):44-57. Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. W Huang D, B T Sherman, R A Lempicki, Nucleic Acids Res. 371Huang d W, Sherman BT, Lempicki RA. Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists. Nucleic Acids Res. 2009;37(1):1-13. REVIGO summarizes and visualizes long lists of gene ontology terms. F Supek, M Bosnjak, N Skunca, T Smuc, PLoS One. 6721800Supek F, Bosnjak M, Skunca N, Smuc T. REVIGO summarizes and visualizes long lists of gene ontology terms. PLoS One. 2011;6(7):e21800. WebGestalt 2019: gene set analysis toolkit with revamped UIs and APIs. Y Liao, J Wang, E J Jaehnig, Z Shi, B Zhang, Nucleic Acids Res. 47W1Liao Y, Wang J, Jaehnig EJ, Shi Z, Zhang B. WebGestalt 2019: gene set analysis toolkit with revamped UIs and APIs. Nucleic Acids Res. 2019;47(W1): W199-205. The BioMart community portal: an innovative alternative to large, centralized data repositories. D Smedley, S Haider, S Durinck, L Pandini, P Provero, J Allen, Nucleic Acids Res. 43W1Smedley D, Haider S, Durinck S, Pandini L, Provero P, Allen J, et al. The BioMart community portal: an innovative alternative to large, centralized data repositories. Nucleic Acids Res. 2015;43(W1):W589-98. Wide-scale comparative analysis of longevity genes and interventions. H Yanai, A Budovsky, T Barzilay, R Tacutu, V E Fraifeld, Aging Cell. 166Yanai H, Budovsky A, Barzilay T, Tacutu R, Fraifeld VE. Wide-scale comparative analysis of longevity genes and interventions. Aging Cell. 2017;16(6):1267-75. The OMA orthology database in 2015: function predictions, better plant support, synteny view and other improvements. A M Altenhoff, N Skunca, N Glover, C M Train, A Sueki, I Pilizota, Nucleic Acids Res. 43Altenhoff AM, Skunca N, Glover N, Train CM, Sueki A, Pilizota I, et al. The OMA orthology database in 2015: function predictions, better plant support, synteny view and other improvements. Nucleic Acids Res. 2015;43(Database issue):D240-9. Ageing Transcriptome Meta-Analysis Reveals Similarities Between Key Mammalian Tissues. bioRxiv. D Palmer, F Fabris, D Aoife, A Freitas, J P De Magalhaes, Palmer D, Fabris F, Aoife D, Freitas A, de Magalhaes JP. Ageing Transcriptome Meta-Analysis Reveals Similarities Between Key Mammalian Tissues. bioRxiv. 2019. The Genotype-Tissue Expression (GTEx) project. G T Consortium, Nat Genet. 456Consortium GT. The Genotype-Tissue Expression (GTEx) project. Nat Genet. 2013;45(6):580-5. Medical subject headings (MeSH) terms. I K Dhammi, S Kumar, Indian J Orthop. 485Dhammi IK, Kumar S. Medical subject headings (MeSH) terms. Indian J Orthop. 2014;48(5):443-4. The genetic association database. K G Becker, K C Barnes, T J Bright, S A Wang, Nat Genet. 365Becker KG, Barnes KC, Bright TJ, Wang SA. The genetic association database. Nat Genet. 2004;36(5):431-2. Common gene signature of cancer and longevity. A Budovsky, R Tacutu, H Yanai, A Abramovich, M Wolfson, V Fraifeld, Mech Ageing Dev. 1301-2Budovsky A, Tacutu R, Yanai H, Abramovich A, Wolfson M, Fraifeld V. Common gene signature of cancer and longevity. Mech Ageing Dev. 2009; 130(1-2):33-9. Ageing as a risk factor for disease. T Niccoli, L Partridge, Curr Biol. 2217Niccoli T, Partridge L. Ageing as a risk factor for disease. Curr Biol. 2012; 22(17):R741-52. Geroscience: linking aging to chronic disease. B K Kennedy, S L Berger, A Brunet, J Campisi, A M Cuervo, E S Epel, Cell. 1594Kennedy BK, Berger SL, Brunet A, Campisi J, Cuervo AM, Epel ES, et al. Geroscience: linking aging to chronic disease. Cell. 2014;159(4):709-13. How ageing processes influence cancer. J P De Magalhaes, Nat Rev Cancer. 135de Magalhaes JP. How ageing processes influence cancer. Nat Rev Cancer. 2013;13(5):357-65. TSGene: a web resource for tumor suppressor genes. M Zhao, J Sun, Z Zhao, Nucleic Acids Res. 41Zhao M, Sun J, Zhao Z. TSGene: a web resource for tumor suppressor genes. Nucleic Acids Res. 2013;41(Database issue):D970-6. ONGene: a literature-based database for human oncogenes. Y Liu, J Sun, M Zhao, J Genet Genomics. 442Liu Y, Sun J, Zhao M. ONGene: a literature-based database for human oncogenes. J Genet Genomics. 2017;44(2):119-21. An interactive tool for comparing lists with Venn's diagrams. J C Oliveros, Venny, Oliveros JC. Venny. An interactive tool for comparing lists with Venn's diagrams. 2015 [Available from: http://bioinfogp.cnb.csic.es/tools/venny/ index.html]. Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. R Tacutu, A Budovsky, H Yanai, V E Fraifeld, Aging (Albany NY). 312Tacutu R, Budovsky A, Yanai H, Fraifeld VE. Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. Aging (Albany NY). 2011;3(12):1178-91. The signaling hubs at the crossroad of longevity and age-related disease networks. M Wolfson, A Budovsky, R Tacutu, V Fraifeld, Int J Biochem Cell Biol. 413Wolfson M, Budovsky A, Tacutu R, Fraifeld V. The signaling hubs at the crossroad of longevity and age-related disease networks. Int J Biochem Cell Biol. 2009;41(3):516-20. Bias tradeoffs in the creation and analysis of protein-protein interaction networks. J Gillis, S Ballouz, P Pavlidis, J Proteome. 100Gillis J, Ballouz S, Pavlidis P. Bias tradeoffs in the creation and analysis of protein-protein interaction networks. J Proteome. 2014;100:44-54. Insights into the regulation of protein abundance from proteomic and transcriptomic analyses. C Vogel, E M Marcotte, Nat Rev Genet. 134Vogel C, Marcotte EM. Insights into the regulation of protein abundance from proteomic and transcriptomic analyses. Nat Rev Genet. 2012;13(4):227-32. COXPRESdb in 2015: coexpression database for animal species by DNA-microarray and RNAseq-based expression data with multiple quality assessment systems. Y Okamura, Y Aoki, T Obayashi, S Tadaka, S Ito, T Narise, Nucleic Acids Res. 43Okamura Y, Aoki Y, Obayashi T, Tadaka S, Ito S, Narise T, et al. COXPRESdb in 2015: coexpression database for animal species by DNA-microarray and RNAseq-based expression data with multiple quality assessment systems. Nucleic Acids Res. 2015;43(Database issue):D82-6. Comparison of RNA-Seq and microarray gene expression platforms for the toxicogenomic evaluation of liver from short-term rat toxicity studies. M S Rao, Van Vleet, T R Ciurlionis, R Buck, W R Mittelstadt, S W Blomme, Eag , Front Genet. 9636Rao MS, Van Vleet TR, Ciurlionis R, Buck WR, Mittelstadt SW, Blomme EAG, et al. Comparison of RNA-Seq and microarray gene expression platforms for the toxicogenomic evaluation of liver from short-term rat toxicity studies. Front Genet. 2018;9:636. GeneFriends: a human RNA-seq-based gene and transcript co-expression database. S Van Dam, T Craig, J P De Magalhaes, Nucleic Acids Res. 43van Dam S, Craig T, de Magalhaes JP. GeneFriends: a human RNA-seq-based gene and transcript co-expression database. Nucleic Acids Res. 2015; 43(Database issue):D1124-32. Hallmarks of cellular senescence. A Hernandez-Segura, J Nehme, M Demaria, Trends Cell Biol. 286Hernandez-Segura A, Nehme J, Demaria M. Hallmarks of cellular senescence. Trends Cell Biol. 2018;28(6):436-53. Network-based analysis of differentially expressed genes in cerebrospinal fluid (CSF) and blood reveals new candidate genes for multiple sclerosis. N Safari-Alighiarloo, M Rezaei-Tavirani, M Taghizadeh, S M Tabatabaei, S Namaki, PeerJ. 42775Safari-Alighiarloo N, Rezaei-Tavirani M, Taghizadeh M, Tabatabaei SM, Namaki S. Network-based analysis of differentially expressed genes in cerebrospinal fluid (CSF) and blood reveals new candidate genes for multiple sclerosis. PeerJ. 2016;4:e2775. Tools for protein-protein interaction network analysis in cancer research. R Sanz-Pamplona, A Berenguer, X Sole, D Cordero, M Crous-Bou, J Serra-Musach, Clin Transl Oncol. 141Sanz-Pamplona R, Berenguer A, Sole X, Cordero D, Crous-Bou M, Serra- Musach J, et al. Tools for protein-protein interaction network analysis in cancer research. Clin Transl Oncol. 2012;14(1):3-14. Comprehensive curation and analysis of global interaction networks in Saccharomyces cerevisiae. T Reguly, A Breitkreutz, L Boucher, B J Breitkreutz, G C Hon, C L Myers, J Biol. 5411Reguly T, Breitkreutz A, Boucher L, Breitkreutz BJ, Hon GC, Myers CL, et al. Comprehensive curation and analysis of global interaction networks in Saccharomyces cerevisiae. J Biol. 2006;5(4):11. Meta-analysis of age-related gene expression profiles identifies common signatures of aging. J P De Magalhaes, J Curado, G M Church, Bioinformatics. 257de Magalhaes JP, Curado J, Church GM. Meta-analysis of age-related gene expression profiles identifies common signatures of aging. Bioinformatics. 2009;25(7):875-81. Increased SHP-1 expression results in radioresistance, inhibition of cellular senescence, and cell cycle redistribution in nasopharyngeal carcinoma cells. Z Sun, X Pan, Z Zou, Q Ding, G Wu, G Peng, Radiat Oncol. 10152Sun Z, Pan X, Zou Z, Ding Q, Wu G, Peng G. Increased SHP-1 expression results in radioresistance, inhibition of cellular senescence, and cell cycle redistribution in nasopharyngeal carcinoma cells. Radiat Oncol. 2015;10:152. Targeting the senescence-overriding cooperative activity of structurally unrelated H3K9 demethylases in melanoma. Y Yu, K Schleich, B Yue, Ji S Lohneis, P Kemper, K , Cancer Cell. 332e8Yu Y, Schleich K, Yue B, Ji S, Lohneis P, Kemper K, et al. Targeting the senescence-overriding cooperative activity of structurally unrelated H3K9 demethylases in melanoma. Cancer Cell. 2018;33(2):322-36. e8. When is hub gene selection better than standard meta-analysis?. P Langfelder, P S Mischel, S Horvath, PLoS One. 8461505Langfelder P, Mischel PS, Horvath S. When is hub gene selection better than standard meta-analysis? PLoS One. 2013;8(4):e61505. CytoCluster: a Cytoscape plugin for cluster analysis and visualization of biological networks. M Li, D Li, Y Tang, F Wu, J Wang, Int J Mol Sci. 189Li M, Li D, Tang Y, Wu F, Wang J. CytoCluster: a Cytoscape plugin for cluster analysis and visualization of biological networks. Int J Mol Sci. 2017;18(9). https://www.mdpi.com/about/announcements/784. Immune surveillance of senescent cells--biological significance in cancer-and non-cancer pathologies. L Hoenicke, L Zender, Carcinogenesis. 336Hoenicke L, Zender L. Immune surveillance of senescent cells--biological significance in cancer-and non-cancer pathologies. Carcinogenesis. 2012; 33(6):1123-6. A novel role for the condensin II complex in cellular senescence. Y Yokoyama, H Zhu, R Zhang, K Noma, Cell Cycle. 1413Yokoyama Y, Zhu H, Zhang R, Noma K. A novel role for the condensin II complex in cellular senescence. Cell Cycle. 2015;14(13):2160-70. Living on a break: cellular senescence as a DNAdamage response. Adda Di Fagagna, F , Nat Rev Cancer. 87d'Adda di Fagagna F. Living on a break: cellular senescence as a DNA- damage response. Nat Rev Cancer. 2008;8(7):512-22. SMC4, which is essentially involved in lung development, is associated with lung adenocarcinoma progression. C Zhang, M Kuang, M Li, L Feng, K Zhang, S Cheng, Sci Rep. 634508Zhang C, Kuang M, Li M, Feng L, Zhang K, Cheng S. SMC4, which is essentially involved in lung development, is associated with lung adenocarcinoma progression. Sci Rep. 2016;6:34508. RNaseH2A is involved in human gliomagenesis through the regulation of cell proliferation and apoptosis. B Dai, P Zhang, Y Zhang, C Pan, G Meng, X Xiao, Oncol Rep. 361Dai B, Zhang P, Zhang Y, Pan C, Meng G, Xiao X, et al. RNaseH2A is involved in human gliomagenesis through the regulation of cell proliferation and apoptosis. Oncol Rep. 2016;36(1):173-80. The hypusine cascade promotes cancer progression and metastasis through the regulation of RhoA in squamous cell carcinoma. T Muramatsu, K I Kozaki, S Imoto, R Yamaguchi, H Tsuda, T Kawano, Oncogene. 3540Muramatsu T, Kozaki KI, Imoto S, Yamaguchi R, Tsuda H, Kawano T, et al. The hypusine cascade promotes cancer progression and metastasis through the regulation of RhoA in squamous cell carcinoma. Oncogene. 2016;35(40):5304-16. RNA-Seq: a revolutionary tool for transcriptomics. Z Wang, M Gerstein, M Snyder, Nat Rev Genet. 101Wang Z, Gerstein M, Snyder M. RNA-Seq: a revolutionary tool for transcriptomics. Nat Rev Genet. 2009;10(1):57-63. Differential roles for cyclindependent kinase inhibitors p21 and p16 in the mechanisms of senescence and differentiation in human fibroblasts. G H Stein, L F Drullinger, A Soulard, V Dulic, Mol Cell Biol. 193Stein GH, Drullinger LF, Soulard A, Dulic V. Differential roles for cyclin- dependent kinase inhibitors p21 and p16 in the mechanisms of senescence and differentiation in human fibroblasts. Mol Cell Biol. 1999;19(3):2109-17. Polycomb CBX7 has a unifying role in cellular lifespan. J Gil, D Bernard, D Martinez, D Beach, Nat Cell Biol. 61Gil J, Bernard D, Martinez D, Beach D. Polycomb CBX7 has a unifying role in cellular lifespan. Nat Cell Biol. 2004;6(1):67-72. Secretion of vascular endothelial growth factor by primary human fibroblasts at senescence. J P Coppe, K Kauser, J Campisi, C M Beausejour, J Biol Chem. 28140Coppe JP, Kauser K, Campisi J, Beausejour CM. Secretion of vascular endothelial growth factor by primary human fibroblasts at senescence. J Biol Chem. 2006;281(40):29568-74. Has gene duplication impacted the evolution of Eutherian longevity?. A Doherty, J P De Magalhaes, Aging Cell. 155Doherty A, de Magalhaes JP. Has gene duplication impacted the evolution of Eutherian longevity? Aging Cell. 2016;15(5):978-80. A proposal to sequence genomes of unique interest for research on aging. J P De Magalhaes, J M Sedivy, C E Finch, S N Austad, G M Church, J Gerontol A Biol Sci Med Sci. 626de Magalhaes JP, Sedivy JM, Finch CE, Austad SN, Church GM. A proposal to sequence genomes of unique interest for research on aging. J Gerontol A Biol Sci Med Sci. 2007;62(6):583-4. Comparative genetics of longevity and cancer: insights from long-lived rodents. V Gorbunova, A Seluanov, Z Zhang, V N Gladyshev, J Vijg, Nat Rev Genet. 158Gorbunova V, Seluanov A, Zhang Z, Gladyshev VN, Vijg J. Comparative genetics of longevity and cancer: insights from long-lived rodents. Nat Rev Genet. 2014;15(8):531-40. The origins and evolution of the p53 family of genes. V A Belyi, P Ak, E Markert, H Wang, W Hu, A Puzio-Kuter, Cold Spring Harb Perspect Biol. 261198Belyi VA, Ak P, Markert E, Wang H, Hu W, Puzio-Kuter A, et al. The origins and evolution of the p53 family of genes. Cold Spring Harb Perspect Biol. 2010;2(6):a001198. Evolutionary history of the reprimo tumor suppressor gene family in vertebrates with a description of a new reprimo gene lineage. I A Wichmann, K Zavala, F G Hoffmann, M W Vandewege, A H Corvalan, J D Amigo, Gene. 5911Wichmann IA, Zavala K, Hoffmann FG, Vandewege MW, Corvalan AH, Amigo JD, et al. Evolutionary history of the reprimo tumor suppressor gene family in vertebrates with a description of a new reprimo gene lineage. Gene. 2016;591(1):245-54. Evolutionary mutant models for human disease. R C Albertson, W Cresko, H W Detrich, J H Postlethwait, Trends Genet. 252Albertson RC, Cresko W, Detrich HW 3rd, Postlethwait JH. Evolutionary mutant models for human disease. Trends Genet. 2009;25(2):74-81. Hallmarks of cancer: the next generation. D Hanahan, R A Weinberg, Cell. 1445Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell. 2011;144(5):646-74. Evolution in health and medicine Sackler colloquium: evolutionary perspectives on health and medicine. S C Stearns, R M Nesse, D R Govindaraju, P T Ellison, Proc Natl Acad Sci U S A. 1071SupplStearns SC, Nesse RM, Govindaraju DR, Ellison PT. Evolution in health and medicine Sackler colloquium: evolutionary perspectives on health and medicine. Proc Natl Acad Sci U S A. 2010;107(Suppl 1):1691-5. The hallmarks of aging. C Lopez-Otin, M A Blasco, L Partridge, M Serrano, G Kroemer, Cell. 1536Lopez-Otin C, Blasco MA, Partridge L, Serrano M, Kroemer G. The hallmarks of aging. Cell. 2013;153(6):1194-217. The business of anti-aging science. J P De Magalhaes, M Stevens, D Thornton, Trends Biotechnol. 3511de Magalhaes JP, Stevens M, Thornton D. The business of anti-aging science. Trends Biotechnol. 2017;35(11):1062-73. Ensembl BioMarts: a hub for data retrieval across taxonomic space. R J Kinsella, A Kahari, S Haider, J Zamora, G Proctor, G Spudich, Database (Oxford). 30Kinsella RJ, Kahari A, Haider S, Zamora J, Proctor G, Spudich G, et al. Ensembl BioMarts: a hub for data retrieval across taxonomic space. Database (Oxford). 2011;2011:bar030. R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing. R Core Team, R Core Team. R: A language and environment for statistical computing. Vienna: R Foundation for Statistical Computing; 2018. Rentrez: an R package for the NCBI eUtils API. D J Winter, R J. 92Winter DJ. Rentrez: an R package for the NCBI eUtils API. R J. 2018;9(2):520-6. CD-HIT: accelerated for clustering the nextgeneration sequencing data. L Fu, B Niu, Z Zhu, S Wu, W Li, Bioinformatics. 2823Fu L, Niu B, Zhu Z, Wu S, Li W. CD-HIT: accelerated for clustering the next- generation sequencing data. Bioinformatics. 2012;28(23):3150-2. . B Haas, A Papanicolaou, Transdecoder, Haas B, Papanicolaou A. Transdecoder [Available from: https://transdecoder. github.io]. Phylogenetic and functional assessment of orthologs inference projects and methods. A M Altenhoff, C Dessimoz, PLoS Comput Biol. 511000262Altenhoff AM, Dessimoz C. Phylogenetic and functional assessment of orthologs inference projects and methods. PLoS Comput Biol. 2009;5(1):e1000262. MAFFT multiple sequence alignment software version 7: improvements in performance and usability. K Katoh, D M Standley, Mol Biol Evol. 304Katoh K, Standley DM. MAFFT multiple sequence alignment software version 7: improvements in performance and usability. Mol Biol Evol. 2013; 30(4):772-80. Phyx: phylogenetic tools for unix. J W Brown, J F Walker, S A Smith, Bioinformatics. 3312Brown JW, Walker JF, Smith SA. Phyx: phylogenetic tools for unix. Bioinformatics. 2017;33(12):1886-8. AMAS: a fast tool for alignment manipulation and computing of summary statistics. M L Borowiec, PeerJ. 41660Borowiec ML. AMAS: a fast tool for alignment manipulation and computing of summary statistics. PeerJ. 2016;4:e1660. IQ-TREE: a fast and effective stochastic algorithm for estimating maximum-likelihood phylogenies. L T Nguyen, H A Schmidt, Von Haeseler, A Minh, B Q , Mol Biol Evol. 321Nguyen LT, Schmidt HA, von Haeseler A, Minh BQ. IQ-TREE: a fast and effective stochastic algorithm for estimating maximum-likelihood phylogenies. Mol Biol Evol. 2015;32(1):268-74. Conservation evaluation and phylogenetic diversity. D P Faith, Biol Conserv. 611Faith DP. Conservation evaluation and phylogenetic diversity. Biol Conserv. 1992;61(1):1-10. Picante: R tools for integrating phylogenies and ecology. S W Kembel, P D Cowan, M R Helmus, W K Cornwell, H Morlon, D D Ackerly, Bioinformatics. 2611Kembel SW, Cowan PD, Helmus MR, Cornwell WK, Morlon H, Ackerly DD, et al. Picante: R tools for integrating phylogenies and ecology. Bioinformatics. 2010;26(11):1463-4. Test and visualize gene overlaps. R package version 1. L Shen, M Sinai, Geneoverlap, 16.0. 2013Shen L, Sinai M. GeneOverlap: Test and visualize gene overlaps. R package version 1.16.0. 2013 [Available from: http://shenlab-sinai.github.io/shenlab-sinai/]. ggplot2: elegant graphics for data analysis. H Wickham, Springer-VerlagNew YorkWickham H. ggplot2: elegant graphics for data analysis. New York: Springer- Verlag; 2016. limma powers differential expression analyses for RNA-sequencing and microarray studies. M E Ritchie, B Phipson, D Wu, Y Hu, C W Law, W Shi, Nucleic Acids Res. 43747Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015;43(7):e47. M Tennekes, Treemap, Treemap Visualization. R package version. 22017Tennekes M. treemap: Treemap Visualization. R package version 2.4-2 2017 [Available from: https://CRAN.R-project.org/package=treemap]. Cytoscape: a software environment for integrated models of biomolecular interaction networks. P Shannon, A Markiel, O Ozier, N S Baliga, J T Wang, D Ramage, Genome Res. 1311Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003;13(11):2498-504. Network propagation in the cytoscape cyberinfrastructure. D E Carlin, B Demchak, D Pratt, E Sage, T Ideker, PLoS Comput Biol. 13101005598Carlin DE, Demchak B, Pratt D, Sage E, Ideker T. Network propagation in the cytoscape cyberinfrastructure. PLoS Comput Biol. 2017;13(10):e1005598. Error and attack tolerance of complex networks. R Albert, H Jeong, A L Barabasi, Nature. 4066794Albert R, Jeong H, Barabasi AL. Error and attack tolerance of complex networks. Nature. 2000;406(6794):378-82. The BioGRID interaction database: 2017 update. A Chatr-Aryamontri, R Oughtred, L Boucher, J Rust, C Chang, N K Kolas, Nucleic Acids Res. 45D1Chatr-Aryamontri A, Oughtred R, Boucher L, Rust J, Chang C, Kolas NK, et al. The BioGRID interaction database: 2017 update. Nucleic Acids Res. 2017; 45(D1):D369-D79. Approaches for extracting practical information from gene co-expression networks in plant biology. K Aoki, Y Ogata, D Shibata, Plant Cell Physiol. 483Aoki K, Ogata Y, Shibata D. Approaches for extracting practical information from gene co-expression networks in plant biology. Plant Cell Physiol. 2007;48(3):381-90. The human blood metabolome-transcriptome interface. J Bartel, J Krumsiek, K Schramm, J Adamski, C Gieger, C Herder, PLoS Genet. 1161005274Bartel J, Krumsiek J, Schramm K, Adamski J, Gieger C, Herder C, et al. The human blood metabolome-transcriptome interface. PLoS Genet. 2015;11(6): e1005274. Metabolism of benzo [a] pyrene by human mammary epithelial cells: toxicity and DNA adduct formation. M R Stampfer, J C Bartholomew, H S Smith, J C Bartley, Proc Natl Acad Sci. 7810Stampfer MR, Bartholomew JC, Smith HS, Bartley JC. Metabolism of benzo [a] pyrene by human mammary epithelial cells: toxicity and DNA adduct formation. Proc Natl Acad Sci U S A. 1981;78(10):6251-5. CellAge_ supplementary: GitHub. R A Avelar, J G Ortega, R Tacutu, E Tyler, D Bennett, P Binetti, Avelar RA, Ortega JG, Tacutu R, Tyler E, Bennett D, Binetti P, et al. CellAge_ supplementary: GitHub; 2019 [Available from: https://github.com/maglab/ CellAge_supplementary]. Publisher's Note. Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

2022-03-15T13:18:35Z

CCBY

https://doi.org/10.18632/aging.100413

GOLD

2d3414df04526fdcfae99c0cdcb380d29c6ee325

10.18632/aging.100413

2131391897

22184282

3273898

The complex nature of aging and aging-associated phenomena including CS requires a holistic view with a Molecular links between cellular senescence, longevity and age- related diseases -a systems biology perspective December 2011 Robi Tacutu The Shraga Segal Department of Microbiology and Immunology Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer ShevaIsrael Arie Budovsky The Shraga Segal Department of Microbiology and Immunology Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer ShevaIsrael The Judea Regional R&D Center Moshav CarmelIsrael Hagai Yanai The Shraga Segal Department of Microbiology and Immunology Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer ShevaIsrael Vadim E Fraifeld [email protected] The Shraga Segal Department of Microbiology and Immunology Center for Multidisciplinary Research on Aging Ben-Gurion University of the Negev Beer ShevaIsrael The complex nature of aging and aging-associated phenomena including CS requires a holistic view with a Molecular links between cellular senescence, longevity and age- related diseases -a systems biology perspective www.impactaging.com AGING 312December 2011Δ Correspondence to: Vadim E. Fraifeld, MD, PhD; Copyright: © Tacutu et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are creditedcellular senescenceage-related diseasesgenesmicroRNAspathwaysnetworks The role of cellular senescence (CS) in age-related diseases (ARDs) is a quickly emerging topic in aging research.Our comprehensive data mining revealed over 250 genes tightly associated with CS. Using systems biology tools, we found that CS is closely interconnected with aging, longevity and ARDs, either by sharing common genes and regulators or by protein-protein interactions and eventually by common signaling pathways. The most enriched pathways across CS, ARDs and aging-associated conditions (oxidative stress and chronic inflammation) are growth-promoting pathways and the pathways responsible for cell-extracellular matrix interactions and stress response. Of note, the patterns of evolutionary conservation of CS and cancer genes showed a high degree of similarity, suggesting the co-evolution of these two phenomena. Moreover, cancer genes and microRNAs seem to stand at the crossroad between CS and ARDs. Our analysis also provides the basis for new predictions: the genes common to both cancer and other ARD(s) are highly likely candidates to be involved in CS and vice versa. Altogether, this study shows that there are multiple links between CS, aging, longevity and ARDs, suggesting a common molecular basis for all these conditions. Modulating CS may represent a potential prolongevity and anti-ARDs therapeutic strategy. INTRODUCTION Since Hayflick's discovery of the phenomenon of cellular (replicative) senescence [1], the contribution or even relevance of this phenomenon to organismal aging has been a subject for continuous debates [2][3][4][5]. Although the question still remains open, an increasing amount of evidence, especially from recent years, indicates that cellular senescence (CS) could have a role in aging and age-related diseases (ARDs), rather than being just a laboratory phenomenon [3,[6][7][8][9][10][11]. In fact, the current situation in the field could be defined as an attempt to understand to what extent and how is CS involved in aging and ARDs. Apart from an irreversible growth arrest ("Hayflick's limit" -a finite number of cell divisions), the CS phenotype is characterized by cell hypertrophy, an increased metabolic activity including synthesis of Research Paper macromolecules (RNA, protein, lipid) and organelles [12,13], increased secretion of pro-inflammatory substances and resistance to apoptosis [7,8,11]. After being initially discovered in primary cultures of human fibroblasts, CS has also been found in other cell types such as keratinocytes, endothelial cells, lymphocytes, adrenocortical cells, vascular smooth muscle cells, chondrocytes, etc., both for in vitro and in vivo conditions [3,11,14,15], and in cell cultures derived from many other organisms examined thus far (e.g., mice, monkeys, chickens, Galapagos tortoise, etc.) [16][17][18]. Moreover, it appears that CS is not restricted only to dividing cells. At least some features of CS were also found in classical post-mitotic cells such as neurons, myocardiocytes and adipocytes (reviewed by Tchkonia et al. [19]). focus on the interplay between their components [11,12,20,21]. Here we consider the potential molecular links between CS, longevity, ARDs, oxidative stress, and chronic inflammation from a systems biology perspective. Highlighting the common genes, interactions, regulatory molecules (miRNAs) and common pathways may help in understanding how CS interplays with and contributes to other aging-associated conditions. RESULTS AND DISCUSSION CS genes share common features with LAGs and ARD genes A comprehensive data mining of scientific literature brought about a list of 262 human genes identified as being associated with CS (see Suppl. Table 1). These genes possess diverse functions, with the majority falling into three categories: regulation of cell cycle and proliferation, biosynthesis and programmed cell death (for GO functional analysis, see Suppl. Table 2). We have previously shown that longevity-associated (LAGs) and ARD genes also show functional diversity. Besides that, they display a number of distinct features including higher connectivity and interconnectivity, evolutionary conservation, and essentiality to growth and development [22][23][24]. This combination makes many of them putative candidates as antagonistic pleiotropy genes, i.e., genes which may have undesirable effects later in life, potentially linking aging, longevity and ARDs [25,26]. Therefore, one of the first questions that arise in this context is whether CS genes share any common features with LAGs and ARD genes. Connectivity and interconnectivity: miRNAregulated PPI networks To what extent are the CS genes/proteins working in a cooperative manner? In most cases, proteins do not act on their own but rather together with their partners through protein-protein interactions (PPIs). Currently, the human interactome includes approximately 10,000 genes with more than 35,000 physical PPIs ( [27], http://thebiogrid.org). Most of CS genes (231 of the 262) as well as LAGs and ARD genes can be found in the human interactome [24]. As shown in Fig. 1 (insert), they have a much higher average connectivity (number of first-order protein partners) compared to all interactome proteins. This is in accordance with observations demonstrating that disease proteins have higher average connectivity than other proteins, and that highly connected proteins are more likely to be diseaseassociated [28]. It was particularly evident for cancer genes [22] and for genes common to major human ARDs [23,25]. simulations with sets of randomly selected proteins are presented as dots. For all the sets of interest, the fraction of interconnected proteins was significantly higher than expected by chance (p < E-25). Insert: average connectivity (number of first-order protein partners) of the sets analyzed in this study. For more details, see Materials and Methods. www.impactaging.com Not only are CS genes, LAGs, and ARD genes more connected, but they are also highly interconnected. Indeed, when compared with randomly generated sets, the above genes display a significantly higher interconnectivity (the fraction of genes that form a continuous network) (Fig. 1). For example, 59% of the CS genes are connected between themselves and eventually form a continuous network (Fig. 2), whereas only 4 ± 2% (mean ± SD) genes form a network by chance (p < E−25). The percent of interconnected CS genes would be even higher if other (regulatory) interactions are considered. In particular, if we also take into account posttranscriptional regulation of gene expression by microRNAs (miRNAs), almost two thirds (64%) of the CS genes become connected, either through PPIs or common miRNAs (Fig. 2). Thus, CS genes together with their regulatory miRNAs might work in a cooperative manner by forming a miRNA-regulated PPI network. Such networks are also formed by LAGs and ARD genes (currently available in the NetAge database: [24], http://www.netage-project.org). www.impactaging.com Essentiality Genes with multiple PPIs have a higher probability to be essential, just because the deletions of these genes may result in the disruption of function of a larger number of proteins [28][29][30]. In line with this assumption, the portion of essential genes among the CS genes, LAGs and ARD genes is much higher than that in the whole genome or interactome (Fig. 3). Moreover, there is a significant correlation between connectivity and essentiality of different sets examined in this study (Fig. 3, insert). It is important to stress that many genes essential for development and growth tend to have detrimental effects at the later stages of life as suggested by the theory of antagonistic pleiotropy [31]. Remarkably, the percent of essential CS genes (42%) is considerably higher than that for all genes (p < E-25) and even higher than the percent of essential LAGs (p = 2.5E-10 and p = 6.2E-06 for entire genome and interactome, respectively), genes involved in Alzheimer's disease (p = 1.5E-04 and p = 0.001) and aging-associated processes (p < 0.002) (Fig. 3). This could at least in part be explained by the fact that the CS genes are highly enriched with genes involved in the regulation of basic, housekeeping processes such as cell cycle, cell growth, programmed cell death, DNA repair, and cellular response to stress (Suppl. Table 2). Evolutionary conservation Essential genes are generally more evolutionary conserved than non-essential ones [32]. In support of this notion are recent findings of Waterhouse et al. [33] who demonstrated that the essential genes from model organisms are significantly enriched in orthologs across the vertebrate, arthropod and fungal lineages. The high percentage of essential CS genes led us to explore the possibility that CS genes are highly evolutionary conserved as a whole. With this in mind, we have examined the frequency of orthologs for the human CS genes in over 100 species found currently in the InParanoid database ( [34], http://inparanoid.sbc.su.se). We found that CS genes are significantly more conserved than are the genes in the whole human genome. This is clearly noted across the vertebrate species but the difference is insignificant in lower organisms. In Fig. 4, this is shown for a selected set of well-studied model organisms, from yeast to mouse. The same observation was true for the whole InParanoid set (data not shown). Notably, the pattern of evolutionary conservation of CS genes is almost identical to that of cancer-associated genes (see Fig. 1B in [22]). It would be then tempting to speculate that this similarity is a result of the co-evolution between these two phenomena. Indirectly supporting this assumption are observations on the naked mole rat, a species with an extraordinary resistance to cancer [35], whose fibroblasts do not undergo CS. Instead, they display early contact inhibition, an anti-cancer mechanism based on cell division arrest before reaching a high cell density [36]. Another example is chinchilla, a rodent with a low cancer incidence [37] that also does not develop CS, but evolved other anticancer adaptations such as continuous slow cell proliferation [18]. Molecular links between CS genes, LAGs and ARD genes For a more specific analysis, we further addressed the following questions: (i) Are there genes common for CS and other aging/longevity-related categories? (ii) Could genes involved in CS, ARDs and in the control of life span interact via direct PPIs or their common partners? (iii) Is the set of CS genes enriched in genes associated with ARDs and age-related conditions? (iv) Could the expression of these genes be under the control of common regulatory molecules, more specifically, miRNAs? (v) Are there common pathways for CS, ARDs and aging-associated processes? Common genes The analysis revealed that 19% of the CS genes are also orthologous to LAGs from model organisms, and 53% of the CS genes are involved in at least one ARD ( Table 1). The highest overlap was observed for cancer (53%); lesser values were observed for atherosclerosis (20%), Alzheimer's disease (9%) and type 2 diabetes (9%). The overlap of CS genes with oxidative stress and chronic inflammation associated genes reached 21% and 8%, respectively. In all the above cases, the overlap was significantly greater than expected by chance (p < E-25). Notably, among overlapping genes there are many which are essential for growth and development (from 38% to 55%, depending on the gene set). This high percentage of essential genes is comparable to that found in the Common Gene Signature of longevity and major ARD networks [25]. Protein-protein interactions and common protein partners Apart from common genes, a great number of CS proteins directly interact with LAGs and ARD proteins through PPIs (Table 1). As such, the majority of CS genes fall either in the category of common genes or in that of genes directly connected to LAGs or ARD genes. In total, the genes in these two categories exceed 80% of the entire CS set. In addition, there are many common external protein partners, the number of which is more than one order of magnitude higher than that of common genes. As a result, almost all CS genes are linked to longevity and/or ARDs in one of the following ways -as common molecules, by forming protein complexes via PPIs, or through common partners. AS -Atherosclerosis, AD -Alzheimer's disease Common miRNA regulators Another important possibility by which CS genes, LAGs and ARD genes could be linked is the posttranscriptional co-regulation of their expression through common miRNAs. Among the CS genes, 40 have thus far been experimentally validated as being the targets of 39 miRNAs. Of these miRNAs, almost all have targets reported to be involved in cancer and atherosclerosis and many have targets associated with other age-related conditions and longevity (Table 1). Notably, a large number of these miRNAs were indeed found to be directly involved in ARDs (Table 2). For example, out of the 38 miRNAs regulating the expression of both CS and cancer genes, miRNAs belonging to the miR-17, miR-19, miR-21, miR-24, miR-155, miR-214, miR-221, miR-372, and miR-373 families are oncogenic (oncomirs), while miRNAs of the let-7, miR-1, miR-8, miR-15, miR-29, miR-34, miR-101, miR-124, miR-125, miR-127, miR-145 families have a tumor suppressor activity [38,39,40,41]. In fact, as seen in Table 2, all the miRNAs common to CS and atherosclerosis, AD or type 2 diabetes are also cancer-associated. Cancer genes and miRNAs bridge CS with other ARDs There are many common genes for the major human ARDs [22,23,25]. What are these genes and how do they contribute to the overlaps between CS and ARD genes? As already mentioned (see section 2.1), the CS genes are highly over-represented among LAGs and in all major ARDs and aging-associated conditions ( Table 1). This also follows from the enrichment analysis as shown in Fig. 5. The highest fold-increase (19.3-fold vs. the expected value; p < E-25) was found for oxidative stress. This was quite expected as oxidative stress is one of the major CS inducers [42,43]. The unexpected observation, however, was that the second highest foldincrease value was found for atherosclerosis (18.7-fold; p < E-25), which is almost twice as high as that for cancer (10.9-fold; p<E-25) and other ARDs (10-12fold; p < E-25). However, further analysis revealed that cancer genes are the primary determinants of the links between CS and ARDs. Indeed, when the cancer genes were removed from the other ARD sets, no significant enrichment was found for any of these sets in CS. In contrast, after the removal of atherosclerosis, diabetes or Alzheimer's disease genes from the cancer set, the enrichment value for cancer genes in CS remained almost unchanged. Thus, the cancer genes are central in linking CS with other ARDs. Of note, the enrichment for CS genes with cancer genes increases when they are also represented in another or several other ARDs (Table 3). In particular, this can explain the high foldincrease in the case of atherosclerosis, since almost all atherosclerosis genes in CS are also cancer-associated (51 of 52). Though the impact of CS on the developwww.impactaging.com ment of other ARDs is only beginning to be unveiled, accumulating evidence suggests a role of vascular cell senescence in atherosclerosis [44] and a clear cellular senescence component in type 2 diabetes [45] and Alzheimer's disease [46]. Altogether, our findings indicate that (i) cancer genes (together with miRNAs) determine the links between CS and other major human ARDs, and (ii) CS is particularly enriched in genes which are common to both cancer and other ARD(s). Common signaling pathways While analyzing the PPI networks of longevity and major human ARDs, we found that about half of the common proteins are related to signal transduction [23]. Moreover, we showed that the vast majority of these proteins are hubs, thus playing a central role in linking different ARDs. Therefore, our next question was whether there are common signaling pathways to CS and www.impactaging.com other sets examined in this study. Enrichment analysis could serve as a tool for answering this question. We found that several pathways are particularly enriched across CS, ARDs and aging-associated processes (Suppl . Table 3). Surprisingly, among them are many growth-promoting pathways such as the MAPK signaling, insulin signaling, mTOR signaling, ErbB signaling, neurotrophin signalling, etc (Fig. 6). This might seem paradoxical, since an irreversible growth arrest is the major (though not the only) feature of CS. However, recent findings shed light on this apparent discrepancy, clearly demonstrating that stimulation of cells, which have ceased proliferation, with growthpromoting mediators induces CS. In other words, growth-promoting pathways convert reversible quiescence into senescence [12,47,48]. Such an activation eventually leads to an enhanced secretion of cytokines, chemokines, proteases and ROS by senescent cells [49], collectively termed as the senescenceassociated secretory phenotype (SASP) [9,50]. This could be especially relevant to aging organisms since they display a persistent activation of growth-promoting pathways [13,51]. As such, senescent cells are likely to Table 3. www.impactaging.com create a pro-inflammatory/tumorigenic microenvironment which promotes the development of agingassociated conditions (chronic inflammation and oxidative stress) and ARDs [6,51,53]. In addition, the formation of the "senescent" microenvironment could be greatly attributed to the pathways ensuring cell-cell and cell-ECM (extracellular matrix) interactions, such as "Focal adhesion" and "Regulation of actin cytoskeleton", which we also found to be significantly enriched across CS and ARDs (Fig. 6, Suppl. Table 3). In particular, the importance of these pathways in CS is highlighted by the fact that the downregulation of caveolin-1, a central regulator of focal adhesion kinase activity and actin stress fiber formation [54] resulted in the re-entry of senescent human fibroblasts into cell cycle and restoration of their clonogenic potential [55]. Remarkably, focal adhesion is among the most enriched pathways in the common signaling network for the major human ARDs and longevity [23]. Given the over-representation of growth-promoting and cell-cell/ECM contact pathways in CS, it was expected to see cancer-associated pathways in the enrichment list, since cell proliferation and growth promotion are intrinsic properties of the cancer cells. Indeed, the most overrepresented category in CS is "Pathways in cancer" followed by the pathways for specific forms of cancer, including "Bladder cancer", "Prostate cancer", "Colorectal cancer", "Chronic myeloid leukemia", "Glioma", "Melanoma", and others and tumor suppressor "p53 signaling" (Suppl. Table 3). It should however be kept in mind that the final outcome (CS or cancer) of activation or inhibition of these pathways depends on many additional factors, discussed in details elsewhere [21]. www.impactaging.com All the above mentioned signaling pathways are also involved in cellular response to stress, further linking CS with aging and ARDs. For example, the CSassociated Gadd45 proteins, prominent stress sensors, are involved in the determination of the stressed cell fate via interactions with p53, MAPK, mTOR and other growth-promoting signaling pathways (reviewed by Moskalev et al. [56]). An age-related decrease in Gadd45 inducibility could promote tumorigenesis, immune disorders and insulin resistance [56]. Depending on the severity of stress and activated gene modules, the CS-associated proteins could either mediate DNA repair with subsequent cell survival (quiescence or re-entry to cell cycle), or induce CS or apoptosis (for details see: [57]). A common event in the stressed cells is a down-regulation of insulin/IGF1-Akt-mTOR axis by up-regulated p53 [58][59][60][61][62][63][64]. The resultant temporary quiescence period allows the repair of the DNA damage and the return to the routine cell cycle. However, if the level of stress is too high and/or its duration is too long, the cell cycle arrest turns into CS or apoptosis [64][65][66][67]. The induction of CS is mediated by the reactivation of the initially repressed PI3K/Akt-mTOR pathway [68,69]. Of note, its activation is also considered a hallmark of organismal aging [8,70,71]. Accordingly, the inhibition of mTOR signaling with rapamycin decreased the hypertrophic phenotype of senescent cells in vitro [47,72], extended the lifespan and delayed cancer in mice, even when the treatment was initiated later in life [73]. Thus, the common signaling pathways and the very mechanisms of CS induction link it to aging and ARDs. CONCLUDING REMARKS Our study shows that CS is tightly interconnected to aging, longevity and ARDs, either by sharing common genes and regulators or by PPIs and eventually by common pathways. The identification of a common molecular basis is an important step towards understanding the relationships between all these conditions. The next natural step would be the integration of these data with gene/miRNA expression profiles. Such integration could further highlight the key players in linking CS and ARDs. However, this is not a trivial task as the vast majority of data concerning CS derives from in vitro studies on fibroblasts while ARDs are well studied in a variety of cells and in vivo systems. Broadening the CS investigation by including more cell types, 3D in vitro models and in vivo studies will help in developing a more holistic view on the CS phenomenon. Our analysis also provides the basis for new predictions since the genes common to both cancer and other ARD(s) are highly likely candidates to be involved in CS and vice versa. In addition, a higher connectivity, evolutionary conservation among vertebrates and essentiality may increase the probability for CS genes to be found as being involved in ARD(s). Another interesting finding is the similarity between the patterns of evolutionary conservation of CS and cancer genes, which suggests the co-evolution of these two phenomena and calls for a wide comparative study. Of special interest would be the investigation of CS in species with exceptional longevity and resistance to tumors, such as bowhead whales. The comparative studies could shed more light on the links between cancer and CS, and may help in understanding why cancer genes stand at the crossroad of CS and ARDs. An important point for future investigations is examining the role of CS in the formation of an aging microenvironment and its impact on the pathogenesis of ARDs. In turn, the CS phenotype could be modulated by microenvironment. As demonstrated by Choi et al. [74], the interaction of senescent cells with ECM from young cells is sufficient to restore their replicative potential and youthful morphotype. Our present and previous [23] in silico analyses together with experimental data [55] indicate that a special emphasis should be put on focal adhesion and its interactions with growth-promoting pathways. The principal questions remain as to when it is worthwhile to induce and when to inhibit CS. What strategy is preferable, anti-or pro-CS? These questions stem, in particular, from the suggested antagonistic nature of CS in cancer [6,75]. The same dual role of CS could also be true for other ARDs but yet has to be established. Whatever the case, it seems quite plausible that the links between CS and ARDs may vary at different stages of disease and life. Induction of CS was proposed as a potential anti-cancer therapy [76,77]. Alternatively, the recent study of Baker et al. [10] showed that the drug-induced clearance of senescent cells from an early age delayed the onset of several agerelated conditions such as sarcopenia, cataracts and loss of adipose tissue in progeroid mice, and these beneficial effects were also pronounced when the elimination of the senescent cells was initiated in the adults. As a feasible tool for enhancing the clearance of senescent cells, Krizhanovsky et al. [78] suggested immunostimulatory therapy. To some extent, in favor of the anti-CS strategy are the results of our preliminary analysis hinting that the pro-longevity genes are rather anti-CS while the pro-CS genes dominate among the anti-longevity genes (unpublished data). Given the potential benefits of the anti-CS approach, an intriguing www.impactaging.com possibility could be based on iPS technology. Using this technology, Lapasset et al. [79] demonstrated that senescent cells and the cells derived from centenarians could be reprogrammed and eventually rejuvenated. As a perspective for an in vivo application of this method, "a possible scenario may be that after several rounds of iP, the microenvironment itself would also assume a younger phenotype" [80]. Future studies of the different aspects of the links between CS and ARDs will help in selecting the most adequate therapeutic strategy. MATERIALS AND METHODS A list of genes that have been established as being involved in CS was compiled from scientific literature and manually curated. The selection of genes was based on two lines of evidence: 1) genetic or RNA interference (RNAi) interventions (gene knockout, partial or full loss-of-function mutations, RNAi-induced gene silencing, overexpression) which reportedly cause cells to either induce, inhibit or reverse CS, and 2) genes shown to be markers of CS. The lists of LAGs and the genes involved in ARDs and aging-associated processes (oxidative stress and chronic inflammation) were obtained from databases and scientific literature as described in detail elsewhere [23,24,81]. The list of differentially expressed miRNAs in CS and of those which have been shown to affect CS was gathered and manually curated from the scientific literature [82][83][84][85][86]. Annotations regarding the involvement of miRNAs in different ARDs were taken from the Human MicroRNA Disease Database ( [87], http://cmbi.bjmu.cn/hmdd). Oncogene/Tumor Suppressor classification of cancerassociated miRNAs was done according to Wang et al. [38]. Experimentally validated targets of miRNAs have been obtained from the TarBase database ( [88], http://diana.cslab.ece.ntua.gr/tarbase/) and updated by data mining from the scientific literature. Evolutionary conservation of CS genes was analyzed using the InParanoid7 -Eukaryotic Ortholog Groups database ( [34], http://inparanoid.sbc.su.se). To exclude inparalogs, we have used the default threshold score of 0.05. Essentiality of human genes was evaluated based on the data on mouse lethal phenotypes, which were retrieved from the Mouse Genome Informatics database ( [89], www.informatics.jax.org). Functional and pathway analyses were performed with the tools provided by DAVID Bioinformatics Resources 6.7 ( [90,91], http://david.abcc.ncifcrf.gov) using data from the KEGG database ( [92], http://www.genome.jp/kegg/ pathway.html) and Gene Ontology ( [93], http://www.geneontology.org). Protein-protein interacttion (PPI) data from the BioGRID database ( [27], http://thebiogrid.org), human interactome release 3.1.71, was used for the analysis of connectivity and interconnectivity. The largest fraction of proteins which forms a continuous network was used as a measure of interconnectivity. To generate control data, simulations with sets of randomly selected proteins from the interactome were performed. The size of protein sets ranged from 50 to 8000 genes, with a step of 50. In each case, one hundred simulations were run and the relation between the size of the set and the fraction of genes interconnected by chance was quantified. The simulations and the creation of the microRNA-regulated PPI CS network were done using YABNA (Yet Another Biological Networks Analyzer). The YABNA software program and the algorithm for the construction of miRNA-regulated PPI networks have been previously described in detail ( [24,25], http://www.netageproject.org). The graphical output of the CS miRNAregulated PPI network was generated using Cytoscape 2.8.0 ( [94], http://www.cytoscape.org/). The statistical package for the social sciences (SPSS, Inc., Chicago, IL) software was used for the statistical evaluation of the results. Significance of the difference between mean values was calculated using the Student's t-test. The difference between observed and expected values was evaluated using the chi-square test. Pearson's coefficient of correlation was used for pairwise correlative analysis. Differences were considered significant at p-value less than 0.05. Statistical analysis of the enrichment of GO categories and KEGG pathways was carried out using DAVID Bioinformatics Resources 6.7 ( [90,91], http://david.abcc.ncifcrf.gov), with Bonferroni correction. Figure 1 . 1Fraction of CS, longevity and ARD proteins forming a continuous PPI network. Values obtained from Figure 2 . 2MicroRNA-regulated cellular senescence PPI network. Genes are depicted as red circles and miRNAs as green squares. Figure 3 . 3Fraction of genes which are essential to growth and development in each of the gene sets under analysis. The difference between each set and all genes (control) was highly significant (p < E-25). Insert: The correlation between essentiality and average connectivity (R -Pearson's coefficient of correlation; p = 0.004). Figure 4 . 4Evolutionary conservation of human CS genes. The difference between CS genes and all genes in InParanoid was significant for D. rerio (p = 0.0001), X. tropicalis (p = 0.003), G. gallus (p = 0.002), R. norvegicus (p = 0.006) and M. musculus (p = 0.02). Figure 5 . 5Enrichment of genes involved in ARDs and agingassociated conditions among CS genes. The fold-increase was computed as the ratio between the number of observed genes vs. the expected value. In all cases, the fold-increase was highly significant (p < E-25). Figure 6 . 6Common pathways enriched across CS and ARDs. Pathways directly involved in specific pathologies were excluded in order to remove bias. See also Suppl. AGING, December 2011, Vol.3 No.12 ACKNOWLEDGEMENTSCONFLICT OF INTERESTS STATEMENTThe authors of this manuscript have no conflict of interest to declare.SUPPLEMENTARY TABLESTo see the Supplementary Tables please follow the links in Full Text version of this manuscript.www.impactaging.com The serial cultivation of human diploid cell strains. L Hayflick, P S Moorhead, Exp Cell Res. 25Hayflick L, Moorhead PS. The serial cultivation of human diploid cell strains. Exp Cell Res. 1961; 25: 585-621. Paradigms in aging research: a critical review and assessment. H Gershon, Gershon D , Mech Ageing Dev. 117Gershon H, Gershon D. Paradigms in aging research: a critical review and assessment. Mech Ageing Dev. 2000; 117: 21- 28. Cellular senescence and organismal aging. J C Jeyapalan, J M Sedivy, Mech Ageing Dev. 129Jeyapalan JC, Sedivy JM. Cellular senescence and organismal aging. Mech Ageing Dev. 2008; 129: 467-474. Cell aging in vivo and in vitro. H Rubin, Mech. Ageing Dev. 98Rubin H. Cell aging in vivo and in vitro. Mech. Ageing Dev. 1997; 98: 1-35. Aging cell culture: methods and observations. S M Phipps, J B Berletch, L G Andrews, T O Tollefsbol, Methods Mol Biol. 371Phipps SM, Berletch JB, Andrews LG, Tollefsbol TO. Aging cell culture: methods and observations. Methods Mol Biol. 2007; 371: 9-19. Senescent cells, tumor suppression, and organismal aging: good citizens, bad neighbors. J Campisi, Cell. 120Campisi J. Senescent cells, tumor suppression, and organismal aging: good citizens, bad neighbors. Cell. 2005; 120: 513-522. Cellular senescence: putting the paradoxes in perspective. J Campisi, Curr Opin Genet Dev. 21Campisi J. Cellular senescence: putting the paradoxes in perspective. Curr Opin Genet Dev. 2011; 21: 107-112. Aging and immortality: quasiprogrammed senescence and its pharmacologic inhibition. M V Blagosklonny, Cell Cycle. 5Blagosklonny MV. Aging and immortality: quasi- programmed senescence and its pharmacologic inhibition. Cell Cycle. 2006; 5: 2087-2102. Persistent DNA damage signalling triggers senescence-associated inflammatory cytokine secretion. F Rodier, J P Coppé, C K Patil, W A Hoeijmakers, D P Muñoz, S R Raza, A Freund, E Campeau, A R Davalos, J Campisi, Nat Cell Biol. 11Rodier F, Coppé JP, Patil CK, Hoeijmakers WA, Muñoz DP, Raza SR, Freund A, Campeau E, Davalos AR, Campisi J. Persistent DNA damage signalling triggers senescence-associated inflammatory cytokine secretion. Nat Cell Biol. 2009; 11: 973-979. Clearance of p16(Ink4a)-positive senescent cells delays ageing-associated disorders. D J Baker, T Wijshake, T Tchkonia, N K Lebrasseur, B G Childs, B Van De Sluis, J L Kirkland, J M Van Deursen, 10.1038/nature10600Nature. Baker DJ, Wijshake T, Tchkonia T, Lebrasseur NK, Childs BG, van de Sluis B, Kirkland JL, van Deursen JM. Clearance of p16(Ink4a)-positive senescent cells delays ageing-associated disorders. Nature. 2011; doi: 10.1038/nature10600. Impact of cellular senescence signature on ageing research. E Sikora, T Arendt, M Bennett, M Narita, Ageing Res Rev. 10Sikora E, Arendt T, Bennett M, Narita M. Impact of cellular senescence signature on ageing research. Ageing Res Rev. 2011; 10: 146-52. Cell cycle arrest is not senescence. M V Blagosklonny, Aging (Albany NY). 3Blagosklonny MV. Cell cycle arrest is not senescence. Aging (Albany NY). 2011; 3: 94-101. Growth and aging: a common molecular mechanism. M V Blagosklonny, M N Hall, Aging (Albany NY). 1Blagosklonny MV, Hall MN. Growth and aging: a common molecular mechanism. Aging (Albany NY). 2009; 1: 357-362. Cellular senescence: molecular mechanisms, in vivo redox considerations. M Muller, Antioxid Redox Signal. 11Muller M. Cellular senescence: molecular mechanisms, in vivo redox considerations. Antioxid Redox Signal. 2009; 11: 59- 98. Cellular senescence controls fibrosis in wound healing. J I Jun, L F Lau, Aging (Albany NY). 2Jun JI, Lau LF. Cellular senescence controls fibrosis in wound healing. Aging (Albany NY). 2010; 2: 627-631. How and Why We Age. L Hayflick, Ballantine BooksNew YorkHayflick L. (1994) How and Why We Age (Ballantine Books, New York). Cellular senescence in aging primates. U Herbig, M Ferreira, L Condel, D Carey, J M Sedivy, Science. 3111257Herbig U, Ferreira M, Condel L, Carey D, Sedivy JM. Cellular senescence in aging primates. Science. 2006; 311: 1257. Distinct tumor suppressor mechanisms evolve in rodent species that differ in size and lifespan. A Seluanov, C Hine, M Bozzella, A Hall, T H Sasahara, A A Ribeiro, K C Catania, D C Presgraves, V Gorbunova, Aging Cell. 7Seluanov A, Hine C, Bozzella M, Hall A, Sasahara TH, Ribeiro AA, Catania KC, Presgraves DC, Gorbunova V. Distinct tumor suppressor mechanisms evolve in rodent species that differ in size and lifespan. Aging Cell. 2008; 7: 813-823. Fat tissue, aging, and cellular senescence. T Tchkonia, D E Morbeck, Von Zglinicki, T Van Deursen, J Lustgarten, J Scrable, H Khosla, S Jensen, M D Kirkland, J L , Aging Cell. 9Tchkonia T, Morbeck DE, Von Zglinicki T, Van Deursen J, Lustgarten J, Scrable H, Khosla S, Jensen MD, Kirkland JL. Fat tissue, aging, and cellular senescence. Aging Cell. 2010; 9: 667- 684. Hormetic prevention of molecular damage during cellular aging of human skin fibroblasts and keratinocytes. S I Rattan, R E Ali, Ann N Y Acad Sci. 1100Rattan SI, Ali RE. Hormetic prevention of molecular damage during cellular aging of human skin fibroblasts and keratinocytes. Ann N Y Acad Sci. 2007; 1100: 424-430. Four faces of cellular senescence. F Rodier, J Campisi, J Cell Biol. 192Rodier F, Campisi J. Four faces of cellular senescence. J Cell Biol. 2011; 192: 547-556. Common gene signature of cancer and longevity. A Budovsky, R Tacutu, H Yanai, A Abramovich, M Wolfson, V Fraifeld, Mech Ageing Dev. 130Budovsky A, Tacutu R, Yanai H, Abramovich A, Wolfson M, Fraifeld V. Common gene signature of cancer and longevity. Mech Ageing Dev. 2009; 130: 33-39. The signaling hubs at the crossroad of longevity and age-related disease networks. M Wolfson, A Budovsky, R Tacutu, V Fraifeld, Int J Biochem Cell Biol. 41Wolfson M, Budovsky A, Tacutu R, Fraifeld V. The signaling hubs at the crossroad of longevity and age-related disease networks. Int J Biochem Cell Biol. 2009; 41: 516-520. The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. R Tacutu, A Budovsky, V E Fraifeld, Biogerontology. 11Tacutu R, Budovsky A, Fraifeld VE. The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. Biogerontology. 2010; 11: 513-522. MicroRNAregulated protein-protein interaction networks: how could they help in searching for pro-longevity targets?. R Tacutu, A Budovsky, M Wolfson, V E Fraifeld, Rejuvenation Res. 13Tacutu R, Budovsky A, Wolfson M, Fraifeld VE. MicroRNA- regulated protein-protein interaction networks: how could they help in searching for pro-longevity targets? Rejuvenation Res. 2010; 13: 373-377. Lifespan regulation by evolutionarily conserved genes essential for viability. S P Curran, G Ruvkun, PLoS Genet. 356Curran SP, Ruvkun G. Lifespan regulation by evolutionarily conserved genes essential for viability. PLoS Genet. 2007; 3: e56. The BioGRID Interaction Database: 2011 update. C Stark, B J Breitkreutz, A Chatr-Aryamontri, L Boucher, R Oughtred, M S Livstone, J Nixon, K Van Auken, X Wang, X Shi, T Reguly, J M Rust, A Winter, K Dolinski, M Tyers, Nucleic Acids Res. 39Database issueStark C, Breitkreutz BJ, Chatr-Aryamontri A, Boucher L, Oughtred R, Livstone MS, Nixon J, Van Auken K, Wang X, Shi X, Reguly T, Rust JM, Winter A, Dolinski K, Tyers M. The BioGRID Interaction Database: 2011 update. Nucleic Acids Res. 2011; 39(Database issue): D698-704. The human disease network. K I Goh, M E Cusick, D Valle, B Childs, M Vidal, A L Barabási, Proc Natl Acad Sci U S A. 104Goh KI, Cusick ME, Valle D, Childs B, Vidal M, Barabási AL. The human disease network. Proc Natl Acad Sci U S A. 2007; 104: 8685-8690. Lethality and centrality in protein networks. H Jeong, S P Mason, A L Barabási, Z N Oltvai, Nature. 411Jeong H, Mason SP, Barabási AL, Oltvai ZN. Lethality and centrality in protein networks. Nature. 2001; 411: 41-42. Why do hubs in the yeast protein interaction network tend to be essential: reexamining the connection between the network topology and essentiality. E Zotenko, J Mestre, D P O&apos;leary, T M Przytycka, PLoS Comput Biol. 4Evolution.Zotenko E, Mestre J, O'Leary DP, Przytycka TM. Why do hubs in the yeast protein interaction network tend to be essential: reexamining the connection between the network topology and essentiality. PLoS Comput Biol. 2008; 4: e1000140. 31. Williams GC. Pleiotropy, natural selection, and the evolution of senescence. Evolution. 1957; 11: 398-411. Essential genes are more evolutionarily conserved than are nonessential genes in bacteria. I K Jordan, I B Rogozin, Y I Wolf, E V Koonin, Genome Res. 12Jordan IK, Rogozin IB, Wolf YI, Koonin EV. Essential genes are more evolutionarily conserved than are nonessential genes in bacteria. Genome Res. 2002; 12: 962-968. Correlating traits of gene retention, sequence divergence, duplicability and essentiality in vertebrates, arthropods, and fungi. R M Waterhouse, E M Zdobnov, E V Kriventseva, Genome Biol Evol. 3Waterhouse RM, Zdobnov EM, Kriventseva EV. Correlating traits of gene retention, sequence divergence, duplicability and essentiality in vertebrates, arthropods, and fungi. Genome Biol Evol. 2011; 3: 75-86. InParanoid 7: new algorithms and tools for eukaryotic orthology analysis. G Ostlund, T Schmitt, K Forslund, T Köstler, D N Messina, S Roopra, O Frings, E L Sonnhammer, Nucleic Acids Res. 38Ostlund G, Schmitt T, Forslund K, Köstler T, Messina DN, Roopra S, Frings O, Sonnhammer EL. InParanoid 7: new algorithms and tools for eukaryotic orthology analysis. Nucleic Acids Res. 2010; 38(Database issue): D196-203. The naked mole-rat: a new long-living model for human aging research. R Buffenstein, J Gerontol A Biol Sci Med Sci. 60Buffenstein R. The naked mole-rat: a new long-living model for human aging research. J Gerontol A Biol Sci Med Sci. 2005; 60: 1369-1377. Hypersensitivity to contact inhibition provides a clue to cancer resistance of naked mole-rat. A Seluanov, C Hine, J Azpurua, M Feigenson, M Bozzella, Z Mao, K C Catania, V Gorbunova, Proc Natl Acad Sci. 106Seluanov A, Hine C, Azpurua J, Feigenson M, Bozzella M, Mao Z, Catania KC, Gorbunova V. Hypersensitivity to contact inhibition provides a clue to cancer resistance of naked mole-rat. Proc Natl Acad Sci U S A. 2009; 106: 19352-19357. Spontaneous tumors of small mammals. C B Greenacre, Vet Clin North Am Exot Anim Pract. 7Greenacre CB. Spontaneous tumors of small mammals. Vet Clin North Am Exot Anim Pract. 2004; 7: 627-651. Human microRNA oncogenes and tumor suppressors show significantly different biological patterns: from functions to targets. D Wang, C Qiu, H Zhang, J Wang, Q Cui, Y Yin, PLoS One. 513067Wang D, Qiu C, Zhang H, Wang J, Cui Q, Yin Y. Human microRNA oncogenes and tumor suppressors show significantly different biological patterns: from functions to targets. PLoS One. 2010; 5. pii: e13067. The oncogenic RNA-binding protein Musashi1 is regulated by tumor suppressor miRNAs. D T Vo, M Qiao, A D Smith, S C Burns, A J Brenner, L O Penalva, RNA Biol. 8Vo DT, Qiao M, Smith AD, Burns SC, Brenner AJ, Penalva LO. The oncogenic RNA-binding protein Musashi1 is regulated by tumor suppressor miRNAs. RNA Biol. 2011; 8: 817-828. Regulatory mechanisms of microRNAs involvement in cancer. M Fabbri, M Ivan, A Cimmino, M Negrini, G A Calin, Expert Opin Biol Ther. 7Fabbri M, Ivan M, Cimmino A, Negrini M, Calin GA. Regulatory mechanisms of microRNAs involvement in cancer. Expert Opin Biol Ther. 2007; 7: 1009-1019. The guardians of the genome (p53, TA-p73, and TA-p63) are regulators of tumor suppressor miRNAs network. L Boominathan, Cancer Metastasis Rev. 29Boominathan L. The guardians of the genome (p53, TA-p73, and TA-p63) are regulators of tumor suppressor miRNAs network. Cancer Metastasis Rev. 2010; 29: 613-639. Stress-induced premature senescence and tissue ageing. O Toussaint, V Royer, M Salmon, J Remacle, Biochem Pharmacol. 64Toussaint O, Royer V, Salmon M, Remacle J. Stress-induced premature senescence and tissue ageing. Biochem Pharmacol. 2002; 64: 1007-1009. The signals and pathways activating cellular senescence. I Ben-Porath, R A Weinberg, Int J Biochem Cell Biol. 37Ben-Porath I, Weinberg RA. The signals and pathways activating cellular senescence. Int J Biochem Cell Biol. 2005; 37: 961-976. Vascular cell senescence: contribution to atherosclerosis. T Minamino, I Komuro, Circ Res. 100Minamino T, Komuro I. Vascular cell senescence: contribution to atherosclerosis. Circ Res. 2007; 100: 15-26. Vascular endothelial senescence: from mechanisms to pathophysiology. J D Erusalimsky, J Appl Physiol. 106Erusalimsky JD. Vascular endothelial senescence: from mechanisms to pathophysiology. J Appl Physiol. 2009; 106: 326- 332. Evidence that aging and amyloid promote microglial cell senescence. B E Flanary, N W Sammons, C Nguyen, D Walker, W J Streit, Rejuvenation Res. 10Flanary BE, Sammons NW, Nguyen C, Walker D, Streit WJ. Evidence that aging and amyloid promote microglial cell senescence. Rejuvenation Res. 2007; 10: 61-74. Growth stimulation leads to cellular senescence when the cell cycle is blocked. Z N Demidenko, M V Blagosklonny, Cell Cycle. 7Demidenko ZN, Blagosklonny MV. Growth stimulation leads to cellular senescence when the cell cycle is blocked. Cell Cycle. 2008; 7: 3355-3361. mTOR: from growth signal integration to cancer, diabetes and ageing. R Zoncu, A Efeyan, D M Sabatini, Nat Rev Mol Cell Biol. 12Zoncu R, Efeyan A, Sabatini DM. mTOR: from growth signal integration to cancer, diabetes and ageing. Nat Rev Mol Cell Biol. 2011; 12: 21-35. Senescent fibroblasts promote epithelial cell growth and tumorigenesis: A link between cancer and aging. A Krtolica, S Parrinello, S Lockett, P Desprez, J Campisi, Proc Natl Acad Sci U S A. 98Krtolica A, Parrinello S, Lockett S, Desprez P, Campisi J. Senescent fibroblasts promote epithelial cell growth and tumorigenesis: A link between cancer and aging. Proc Natl Acad Sci U S A. 2001; 98: 12072-12077. Senescence-associated secretory phenotypes reveal cell-nonautonomous functions of oncogenic RAS and the p53 tumor suppressor. J P Coppé, C K Patil, F Rodier, Y Sun, D P Muñoz, J Goldstein, P S Nelson, P Y Desprez, J Campisi, PLoS Biol. 6Coppé JP, Patil CK, Rodier F, Sun Y, Muñoz DP, Goldstein J, Nelson PS, Desprez PY, Campisi J. Senescence-associated secretory phenotypes reveal cell-nonautonomous functions of oncogenic RAS and the p53 tumor suppressor. PLoS Biol. 2008; 6: 2853-2868. Separating cause from effect: how does insulin/IGF signalling control lifespan in worms, flies and mice?. M D Piper, C Selman, J J Mcelwee, L Partridge, J Intern Med. 263Piper MD, Selman C, McElwee JJ, Partridge L. Separating cause from effect: how does insulin/IGF signalling control lifespan in worms, flies and mice? J Intern Med. 2008; 263: 179- 191. Differential expression of thymosin beta-10 by early passage and senescent vascular endothelium is modulated by VPF/VEGF: evidence for senescent endothelial cells in vivo at sites of atherosclerosis. E Vasile, Y Tomita, L F Brown, O Kocher, H F Dvorak, FASEB J. 15Vasile E, Tomita Y, Brown LF, Kocher O, Dvorak HF. Differential expression of thymosin beta-10 by early passage and senescent vascular endothelium is modulated by VPF/VEGF: evidence for senescent endothelial cells in vivo at sites of atherosclerosis. FASEB J. 2001; 15: 458-466. The role of chondrocyte senescence in osteoarthritis. J S Price, J G Waters, C Darrah, C Pennington, D R Edwards, S T Donell, I M Clark, Aging Cell. 1Price JS, Waters JG, Darrah C, Pennington C, Edwards DR, Donell ST, Clark IM. The role of chondrocyte senescence in osteoarthritis. Aging Cell. 2002; 1: 57-65. Morphological adjustment of senescent cells by modulating caveolin-1 status. K A Cho, S J Ryu, Y S Oh, J H Park, J W Lee, H P Kim, K T Kim, I S Jang, S C Park, J Biol Chem. 279Cho KA, Ryu SJ, Oh YS, Park JH, Lee JW, Kim HP, Kim KT, Jang IS, Park SC. Morphological adjustment of senescent cells by modulating caveolin-1 status. J Biol Chem. 2004; 279: 42270- 42278. Senescent phenotype can be reversed by reduction of caveolin status. K A Cho, S J Ryu, J S Park, I S Jang, J S Ahn, K T Kim, S C Park, J Biol Chem. 278Cho KA, Ryu SJ, Park JS, Jang IS, Ahn JS, Kim KT, Park SC. Senescent phenotype can be reversed by reduction of caveolin status. J Biol Chem. 2003; 278: 27789-27795. Gadd45 proteins: Relevance to aging, longevity and age-related pathologies. A A Moskalev, Z Smit-Mcbride, M V Shaposhnikov, E N Plyusnina, A Zhavoronkov, A Budovsky, R Tacutu, V E Fraifeld, 10.1016/j.arr.2011.09.003Ageing Res Rev. Moskalev AA, Smit-McBride Z, Shaposhnikov MV, Plyusnina EN, Zhavoronkov A, Budovsky A, Tacutu R, Fraifeld VE. Gadd45 proteins: Relevance to aging, longevity and age-related pathologies. Ageing Res Rev. 2011; doi:10.1016/j.arr.2011.09.003. Deciphering the intricate regulatory mechanisms for the cellular choice between cell repair, apoptosis or senescence in response to damaging signals. A Erol, Cell Signal. 23Erol A. Deciphering the intricate regulatory mechanisms for the cellular choice between cell repair, apoptosis or senescence in response to damaging signals. Cell Signal. 2011; 23: 1076- 1081. Coordination and communication between the p53 and IGF-1-AKT-TOR signal transduction pathways. A J Levine, Z Feng, T W Mak, H You, Jin S , Genes Dev. 20Levine AJ, Feng Z, Mak TW, You H, Jin S. Coordination and communication between the p53 and IGF-1-AKT-TOR signal transduction pathways. Genes Dev. 2006; 20: 267-275. The coordinate regulation of the p53 and mTOR pathways in cells. Z Feng, H Zhang, A J Levine, Jin S , Proc Natl Acad Sci U S A. 102Feng Z, Zhang H, Levine AJ, Jin S. The coordinate regulation of the p53 and mTOR pathways in cells. Proc Natl Acad Sci U S A. 2005; 102: 8204-8209. p53 target genes sestrin1 and sestrin2 connect genotoxic stress and mTOR signaling. A V Budanov, M Karin, Cell. 134Budanov AV, Karin M. p53 target genes sestrin1 and sestrin2 connect genotoxic stress and mTOR signaling. Cell. 2008; 134: 451-460. Paradoxical suppression of cellular senescence by p53. Z N Demidenko, L G Korotchkina, A V Gudkov, M V Blagosklonny, Proc Natl Acad Sci U S A. 107Demidenko ZN, Korotchkina LG, Gudkov AV, Blagosklonny MV. Paradoxical suppression of cellular senescence by p53. Proc Natl Acad Sci U S A. 2010; 107: 9660-9664. The choice between p53-induced senescence and quiescence is determined in part by the mTOR pathway. L G Korotchkina, O V Leontieva, E I Bukreeva, Z N Demidenko, A V Gudkov, M V Blagosklonny, Aging (Albany NY). 2Korotchkina LG, Leontieva OV, Bukreeva EI, Demidenko ZN, Gudkov AV, Blagosklonny MV. The choice between p53-induced senescence and quiescence is determined in part by the mTOR pathway. Aging (Albany NY). 2010; 2: 344-352. Decision-making by p53 and mTOR. C G Maki, Aging (Albany NY). 2Maki CG. Decision-making by p53 and mTOR. Aging (Albany NY). 2010; 2: 324-326. Shifting senescence into quiescence by turning up p53. M Serrano, Cell Cycle. 9Serrano M. Shifting senescence into quiescence by turning up p53. Cell Cycle. 2010; 9: 4256-4257. Weak p53 permits senescence during cell cycle arrest. O V Leontieva, A V Gudkov, M V Blagosklonny, Cell Cycle. 9Leontieva OV, Gudkov AV, Blagosklonny MV. Weak p53 permits senescence during cell cycle arrest. Cell Cycle. 2010; 9: 4323-4327. p53 and senescence: a little goes a long way. J S Long, K M Ryan, Cell Cycle. 9Long JS, Ryan KM. p53 and senescence: a little goes a long way. Cell Cycle. 2010; 9: 4050-4051. TP53 and MTOR crosstalk to regulate cellular senescence. L Galluzzi, O Kepp, G Kroemer, Aging (Albany NY). 2Galluzzi L, Kepp O, Kroemer G. TP53 and MTOR crosstalk to regulate cellular senescence. Aging (Albany NY). 2010; 2: 535- 537. DNA damaging agents and p53 do not cause senescence in quiescent cells, while consecutive re-activation of mTOR is associated with conversion to senescence. O V Leontieva, M V Blagosklonny, Aging (Albany NY). 2Leontieva OV, Blagosklonny MV. DNA damaging agents and p53 do not cause senescence in quiescent cells, while consecutive re-activation of mTOR is associated with conversion to senescence. Aging (Albany NY). 2010; 2: 924-935. mTOR and its link to the picture of Dorian Gray -re-activation of mTOR promotes aging. J Wesierska-Gadek, Aging (Albany NY). 2Wesierska-Gadek J. mTOR and its link to the picture of Dorian Gray -re-activation of mTOR promotes aging. Aging (Albany NY). 2010; 2: 892-893. Aging: ROS or TOR. M V Blagosklonny, Cell Cycle. 7Blagosklonny MV. Aging: ROS or TOR. Cell Cycle. 2008; 7: 3344-3354. TOR and ageing: a complex pathway for a complex process. M A Mccormick, S Y Tsai, B K Kennedy, Philos Trans R Soc Lond B Biol Sci. 366McCormick MA, Tsai SY, Kennedy BK. TOR and ageing: a complex pathway for a complex process. Philos Trans R Soc Lond B Biol Sci. 2011; 366: 17-27. Rapamycin decelerates cellular senescence. Z N Demidenko, S G Zubova, E I Bukreeva, V A Pospelov, T V Pospelova, M V Blagosklonny, Cell Cycle. 8Demidenko ZN, Zubova SG, Bukreeva EI, Pospelov VA, Pospelova TV, Blagosklonny MV. Rapamycin decelerates cellular senescence. Cell Cycle. 2009; 8: 1888-1895. Rapamycin fed late in life extends lifespan in genetically heterogeneous mice. D E Harrison, R Strong, Z D Sharp, J F Nelson, C M Astle, K Flurkey, N L Nadon, J E Wilkinson, K Frenkel, C S Carter, M Pahor, M A Javors, E Fernandez, R A Miller, Nature. 460Harrison DE, Strong R, Sharp ZD, Nelson JF, Astle CM, Flurkey K, Nadon NL, Wilkinson JE, Frenkel K, Carter CS, Pahor M, Javors MA, Fernandez E, Miller RA. Rapamycin fed late in life extends lifespan in genetically heterogeneous mice. Nature. 2009; 460: 392-395. Restoration of senescent human diploid fibroblasts by modulation of the extracellular matrix. H R Choi, K A Cho, H T Kang, J B Lee, M Kaeberlein, Y Suh, I K Chung, S C Park, Aging Cell. 10Choi HR, Cho KA, Kang HT, Lee JB, Kaeberlein M, Suh Y, Chung IK, Park SC. Restoration of senescent human diploid fibroblasts by modulation of the extracellular matrix. Aging Cell. 2011; 10: 148-157. The senescence-associated secretory phenotype: the dark side of tumor suppression. J P Coppé, P Y Desprez, A Krtolica, J Campisi, Annu Rev Pathol. 5Coppé JP, Desprez PY, Krtolica A, Campisi J. The senescence-associated secretory phenotype: the dark side of tumor suppression. Annu Rev Pathol. 2010; 5: 99-118. Senescence and tumor clearance is triggered by p53 restoration in murine liver carcinomas. J Xue, L Zender, C Miething, Hernando E Krizhanovsky, V Cordon-Cardo, C Lowe, S W , Nature. 445Xue J, Zender L, Miething C, Hernando E, Krizhanovsky V, Cordon-Cardo C, Lowe SW. Senescence and tumor clearance is triggered by p53 restoration in murine liver carcinomas. Nature. 2007; 445: 656-660. Prosenescence therapy for cancer treatment. C Nardella, J G Clohessy, A Alimonti, P P Pandolfi, Nat Rev Cancer. 11Nardella C, Clohessy JG, Alimonti A, Pandolfi PP. Pro- senescence therapy for cancer treatment. Nat Rev Cancer. 2011; 11: 503-511. Senescence of activated stellate cells limits liver fibrosis. V Krizhanovsky, Yon M Dickins, R A Hearn, S Simon, J Miething, C Yee, H Zender, L Lowe, S W , Cell. 134Krizhanovsky V, Yon M, Dickins RA, Hearn S, Simon J, Miething C, Yee H, Zender L, Lowe SW. Senescence of activated stellate cells limits liver fibrosis. Cell. 2008; 134: 657-667. Rejuvenating senescent and centenarian human cells by reprogramming through the pluripotent state. L Lapasset, O Milhavet, A Prieur, E Besnard, A Babled, N Aït-Hamou, J Leschik, F Pellestor, J M Ramirez, De Vos, J Lehmann, S Lemaitre, J M , Genes Dev. 25Lapasset L, Milhavet O, Prieur A, Besnard E, Babled A, Aït- Hamou N, Leschik J, Pellestor F, Ramirez JM, De Vos J, Lehmann S, Lemaitre JM. Rejuvenating senescent and centenarian human cells by reprogramming through the pluripotent state. Genes Dev. 2011; 25: 2248-2253. Have we reached the point for in vivo rejuvenation?. A Abramovich, K K Muradian, V E Fraifeld, Rejuvenation Res. 11Abramovich A, Muradian KK, Fraifeld VE. Have we reached the point for in vivo rejuvenation? Rejuvenation Res. 2008; 11: 489-492. The Human Ageing Genomic Resources: online databases and tools for biogerontologists. J P De Magalhaes, A Budovsky, G Lehmann, J Costa, Y Li, V Fraifeld, G M Church, Aging Cell. 8de Magalhaes JP, Budovsky A, Lehmann G, Costa J, Li Y, Fraifeld V, Church GM The Human Ageing Genomic Resources: online databases and tools for biogerontologists. Aging Cell. 2009; 8: 65-72. Stepwise up-regulation of microRNA expression levels from replicating to reversible and irreversible growth arrest states in WI-38 human fibroblasts. O C Maes, H Sarojini, E Wang, J Cell Physiol. 221Maes OC, Sarojini H, Wang E. Stepwise up-regulation of microRNA expression levels from replicating to reversible and irreversible growth arrest states in WI-38 human fibroblasts. J Cell Physiol. 2009; 221: 109-119. Alterations in microRNA expression in stress-induced cellular senescence. G Li, C Luna, J Qiu, D L Epstein, P Gonzalez, Mech Ageing Dev. 130Li G, Luna C, Qiu J, Epstein DL, Gonzalez P. Alterations in microRNA expression in stress-induced cellular senescence. Mech Ageing Dev. 2009; 130: 731-741. MiRNA profile associated with replicative senescence, extended cell culture, and ectopic telomerase expression in human foreskin fibroblasts. L N Bonifacio, M B Jarstfer, PLoS One. 512519Bonifacio LN, Jarstfer MB. MiRNA profile associated with replicative senescence, extended cell culture, and ectopic telomerase expression in human foreskin fibroblasts. PLoS One. 2010; 5. pii: e12519. Multiple microRNAs rescue from Rasinduced senescence by inhibiting p21(Waf1/Cip1). V Borgdorff, M E Lleonart, C L Bishop, D Fessart, A H Bergin, M G Overhoff, D H Beach, Oncogene. 29Borgdorff V, Lleonart ME, Bishop CL, Fessart D, Bergin AH, Overhoff MG, Beach DH. Multiple microRNAs rescue from Ras- induced senescence by inhibiting p21(Waf1/Cip1). Oncogene. 2010; 29: 2262-2271. MicroRNA profiling in human diploid fibroblasts uncovers miR-519 role in replicative senescence. B S Marasa, S Srikantan, J L Martindale, M M Kim, E K Lee, M Gorospe, K Abdelmohsen, Aging (Albany NY). 2Marasa BS, Srikantan S, Martindale JL, Kim MM, Lee EK, Gorospe M, Abdelmohsen K. MicroRNA profiling in human diploid fibroblasts uncovers miR-519 role in replicative senescence. Aging (Albany NY). 2010; 2: 333-343. An analysis of human microRNA and disease associations. M Lu, Q Zhang, M Deng, J Miao, Y Guo, W Gao, Q Cui, PLoS One. 33420Lu M, Zhang Q, Deng M, Miao J, Guo Y, Gao W, Cui Q. An analysis of human microRNA and disease associations. PLoS One. 2008; 3: e3420. Hatzigeorgiou AG The database of experimentally supported targets: a functional update of TarBase. G L Papadopoulos, M Reczko, V A Simossis, P Sethupathy, Nucleic Acids Res. 37Papadopoulos GL, Reczko M, Simossis VA, Sethupathy P, Hatzigeorgiou AG The database of experimentally supported targets: a functional update of TarBase. Nucleic Acids Res. 2009; 37: D155-158. Mouse mutants and phenotypes: Accessing information for the study of mammalian gene function. M Ringwald, J T Eppig, Methods. 5Ringwald M, Eppig JT. Mouse mutants and phenotypes: Accessing information for the study of mammalian gene function. Methods. 2011; 5 3: 405-410. G DennisJr, B T Sherman, D A Hosack, J Yang, W Gao, H C Lane, R A Lempicki, David, Database for Annotation, Visualization, and Integrated Discovery. 43Dennis G Jr, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki RA. DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol. 2003; 4: P3. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. W Huang Da, B T Sherman, R A Lempicki, Nat Protoc. 4Huang da W, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. 2009; 4: 44-57. KEGG for representation and analysis of molecular networks involving diseases and drugs. M Kanehisa, S Goto, M Furumichi, M Tanabe, M Hirakawa, Nucleic Acids Res. 38Kanehisa M, Goto S, Furumichi M, Tanabe M, Hirakawa M. KEGG for representation and analysis of molecular networks involving diseases and drugs. Nucleic Acids Res. 2010; 38(Database issue): D355-360. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. M Ashburner, C A Ball, J A Blake, D Botstein, H Butler, J M Cherry, A P Davis, K Dolinski, S S Dwight, J T Eppig, M A Harris, D P Hill, L Issel-Tarver, A Kasarskis, S Lewis, J C Matese, J E Richardson, M Ringwald, G M Rubin, G Sherlock, Nat Genet. 25Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT, Harris MA, Hill DP, Issel-Tarver L, Kasarskis A, Lewis S, Matese JC, Richardson JE, Ringwald M, Rubin GM, Sherlock G. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet. 2000; 25: 25-29. Cytoscape: a software environment for integrated models of biomolecular interaction networks. P Shannon, A Markiel, O Ozier, N S Baliga, J T Wang, D Ramage, N Amin, B Schwikowski, T Ideker, Genome Res. 13Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N, Schwikowski B, Ideker T. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003; 13: 2498-2504.

2022-03-25T21:31:33Z

CCBY

https://www.frontiersin.org/articles/10.3389/fnagi.2018.00055/pdf

GOLD

0add33d12480596b2a4125c2270056c076972d00

10.3389/fnagi.2018.00055

2791835398

29563869

5845755

Altered Brain Functional Hubs and Connectivity in Type 2 Diabetes Mellitus Patients: A Resting-State fMRI Study March 2018 Nicola Canessa Robi Tacutu Jiuquan Zhang Jian Wang [email protected] Liu D Duan S Zhou C Wei P Chen L Yin X Zhang J Wang J Daihong Liu † Shanshan Duan † Chaoyang Zhou Department of Radiology, Southwest Hospital Third Military Medical University (Army Medical University) ChongqingChina Ping Wei Department of Radiology, Southwest Hospital Third Military Medical University (Army Medical University) ChongqingChina Lihua Chen Department of Radiology, Southwest Hospital Third Military Medical University (Army Medical University) ChongqingChina Xuntao Yin Department of Radiology, Southwest Hospital Third Military Medical University (Army Medical University) ChongqingChina Jiuquan Zhang Department of Radiology, Southwest Hospital Third Military Medical University (Army Medical University) ChongqingChina Jian Wang Department of Radiology, Southwest Hospital Third Military Medical University (Army Medical University) ChongqingChina Department of Endocrinology The Third Affiliation Hospital of Chongqing Medical University ChongqingChina Department of Endocrinology, Southwest Hospital Third Military Medical University (Army Medical University) ChongqingChina Istituto Universitario di Studi Superiori di Pavia (IUSS) Institute of Biochemistry of the Romanian Academy Arie Budovsky, Judea R & D Center Italy, Romania, Israel Altered Brain Functional Hubs and Connectivity in Type 2 Diabetes Mellitus Patients: A Resting-State fMRI Study Frontiers in Aging Neuroscience | www.frontiersin.org 155March 201810.3389/fnagi.2018.00055Received: 19 June 2017 Accepted: 19 February 2018 Published: 06 March 2018 Citation:ORIGINAL RESEARCH Edited by: Vadim Fraifeld, Ben-Gurion University of the Negev, Israel Reviewed by: *Correspondence: † These authors have contributed equally to this work. Altered Brain Functional Hubs and Connectivity in Type 2 Diabetes Mellitus Patients: A Resting-State fMRI Study. Front. Aging Neurosci. 10:55. Liu et al. Altered Brain Network in T2DMtype 2 diabetes mellitus, resting-state functional MRI, degree centrality, Granger causality analysis, functional connectivity, cognitive impairment Abbreviations: ACG, anterior cingulate gyrusAVLT, Auditory Verbal Learning TestBMI, body mass indexDC, degree centralityDST, Digit Span TestEPI, echo planar imagingFC, functional connectivityFLAIR, fluid-attenuated inversion recoveryfMRI, functional magnetic resonance imagingFT3, free triiodothyronineFT4, free thyroxineGCA, Granger causality analysisHAMD, Hamilton Depression Rating ScaleHbA 1c , glycosylated hemoglobinHC, healthy controlHDL, high-density lipoproteinHOMA2-IR, updated homeostasis model assessment of insulin resistanceLDL, low-density lipoproteinLOC, lateral occipital corticesMMSE, Mini-Mental State ExaminationMNI, Montreal Neurological InstituteMoCA, Montreal Cognitive AssessmentMP-RAGE, magnetization prepared by rapid-acquisition gradient-echoPreCG, precentral gyrusROI, region of interestT2DM, Type 2 diabetes mellitusTMT, Trail Making TestTSH, thyroid-stimulating hormoneVFT, Verbal Fluency Test Type 2 diabetes mellitus (T2DM) affects a vast population and is closely associated with cognitive impairment. However, the mechanisms of cognitive impairment in T2DM patients have not been unraveled. Research on the basic units (nodes or hubs and edges) of the brain functional network on the basis of neuroimaging may advance our understanding of the network change pattern in T2DM patients. This study investigated the change patterns of brain functional hubs using degree centrality (DC) analysis and the connectivity among these hubs using functional connectivity and Granger causality analysis. Compared to healthy controls, the DC values were higher in the left anterior cingulate gyrus (ACG) and lower in the bilateral lateral occipital cortices (LOC) and right precentral gyrus (PreCG) in T2DM patients. The functional connectivity between the left ACG and the right PreCG was stronger in T2DM patients, whereas the functional connectivity among the right PreCG and bilateral LOC was weaker. A negative causal effect from the left ACG to left LOC and a positive effect from the left ACG to right LOC were observed in T2DM patients, while in healthy controls, the opposite occurred. Additionally, the reserve of normal brain function in T2DM patients was negatively associated with the elevated glycemic parameters. This study demonstrates that there are brain functional hubs and connectivity alterations that may reflect the aberrant information communication in the brain of T2DM patients. The findings may advance our understanding of the mechanisms of T2DM-related cognitive impairment. Abbreviations: ACG, anterior cingulate gyrus; AVLT, Auditory Verbal Learning Test; BMI, body mass index; DC, degree centrality; DST, Digit Span Test; EPI, echo planar imaging; FC, functional connectivity; FLAIR, fluid-attenuated inversion recovery; fMRI, functional magnetic resonance imaging; FT3, free triiodothyronine; FT4, free thyroxine; GCA, Granger causality analysis; HAMD, Hamilton Depression Rating Scale; HbA 1c , glycosylated hemoglobin; HC, healthy control; HDL, high-density lipoprotein; HOMA2-IR, updated homeostasis model assessment of insulin resistance; LDL, low-density lipoprotein; LOC, lateral occipital cortices; MMSE, Mini-Mental State Examination; MNI, Montreal Neurological Institute; MoCA, Montreal Cognitive Assessment; MP-RAGE, magnetization prepared by rapid-acquisition gradient-echo; PreCG, precentral gyrus; ROI, region of interest; T2DM, Type 2 diabetes mellitus; TMT, Trail Making Test; TSH, thyroid-stimulating hormone; VFT, Verbal Fluency Test. INTRODUCTION Type 2 diabetes mellitus (T2DM) affects 415 million individuals and is predicted to increase to 642 million in 2040, according to the Diabetes Atlas 7th Edition published by the International Diabetes Federation (http://www.idf.org/about-diabetes/factsfigures). Numerous studies have suggested that T2DM is closely associated with cognitive impairment, including the domains of motor function, executive function, processing speed and memory (Palta et al., 2014). Clarification of the underlying mechanism of cognitive impairment in T2DM patients for diagnosis and therapeutic effect estimation is essential before these patients develop dementia. As a proven informative neuroimaging method, functional magnetic resonance imaging (fMRI) has been extensively applied to investigate alterations of brain function in T2DM patients. In fMRI studies, T2DM patients manifest functional changes in certain brain regions, and these changes are different from those associated with normal aging. For instance, the abnormal amplitude of low-frequency fluctuation, regional homogeneity, and functional connectivity (FC) in T2DM patients have been associated with poor performance in cognitive tests (Xia et al., 2013;Chen et al., 2014;Cui et al., 2014Cui et al., , 2015Moheet et al., 2015). These studies have focused on local spontaneous brain activity or have analyzed the FC or network within the selected brain regions based on a priori assumption. According to the graph theory, a network is defined as a set of pairwise relationships between the elements of a system, which formally consists of a set of edges that link a set of nodes (Barabasi and Oltvai, 2004;Petersen and Sporns, 2015). Network analysis offers a new conceptual framework to investigate the network biology of aging (Wolfson et al., 2009;Tacutu et al., 2011), T2DM (Sandor et al., 2017) and neurodegenerative diseases (de Haan et al., 2017) at a variety of levels of scale including genes, proteins, synapses, neurons, neuronal circuits, neuronal populations, and systems (Petersen and Sporns, 2015). Therefore, the aforementioned fMRI findings provide a clue to explore T2DM-related brain dysfunction from the perspective of macroscopic nodes and connectivity at the whole brain level, which may share universal laws of network. Degree centrality (DC), a measure based on graph theory, provides an approach for identifying the candidate functional hubs in the diabetic brain. DC is defined as the number of links that are strongly correlated to a given voxel or node for a binary graph and enables whole brain analysis at the voxel level, which may avoid the bias caused by selecting brain regions according to a priori assumption (Buckner et al., 2009;Zuo et al., 2012). It also considers the weights of these links for a weighted graph, and the weighted version of DC is more robust against confounding factors (Zuo et al., 2012). Thus, it can quantify the importance of a node to the rest of the brain, and nodes with high DC are defined as hubs (Zuo et al., 2012). Therefore, DC enables an investigation of the complexity and patterns of the brain functional connectome in diseases, including the subsequent analysis of FC between the hubs and the rest of the brain (Cui et al., 2016), or the interactions among these functional hubs in the present study. Connectivity depicts the relationships among functionally segregated brain systems and can be classified into undirected and directed functional connectivity. Traditional FC is often used to investigate the statistical dependencies among several given seed regions in an undirected manner (Biswal et al., 1995). Directed FC is explicitly used to explore the functional interactions in a directed manner and can be estimated using the Granger causality model (Seth et al., 2015). The positive casual effect and negative causal effect can be quantified with positive and negative signed-path coefficients and interpreted as activation and inhibition, respectively (Hamilton et al., 2011). Both undirected and directed FC are the components of functional integration that are used to analyze functional brain architectures (Friston et al., 2013). By combining the traditional FC and Granger causality analysis (GCA) approaches, more information can be obtained from different perspectives to map the connectivity pattern among the brain functional hubs that are probably affected by T2DM. In this study, we hypothesized that the aberrant brain function of hubs and their connectivity may contribute to brain dysfunction in T2DM patients. The DC method was applied to identify the candidate functional hubs. Next, the FC and GCA methods were applied to investigate the connectivity among these functional hubs. We also investigated the relationships of brain functional alterations with clinical data and neuropsychological performance. Our study provides evidence to further understand the neurological mechanisms that underlie T2DM-associated cognitive impairment. MATERIALS AND METHODS Subjects T2DM patients were recruited from inpatients and the community and healthy controls (HC) were recruited from the community between December 2013 and December 2015. Forty-seven T2DM patients and 47 healthy controls, who were matched with regard to age, sex, education and body mass index (BMI), were enrolled in the study. Both the T2DM patients and the healthy controls met the following inclusion criteria: (1) between 45 and 70 years old; 2) at least six years of education; and (3) right-handedness. T2DM was diagnosed by endocrinologists according to the criteria published by the World Health Organization in 1999 (Alberti and Zimmet, 1998). The T2DM patients had at least 1 year of disease duration. The exclusion criteria for all participants were as follows: (1) organic disease in the brain, such as stroke, tumor, or a white matter lesion rating score ≥ 2 (Wahlund et al., 2001); (2) physical disability; (3) pregnancy, thyroid dysfunction; (4) signs of dementia (Mini-Mental State Examination [MMSE] score ≤ 24) (Galea and Woodward, 2005), major depression > 20) (Hamilton, 1960), or other psychiatric disorders; (5) severe hearing or visual impairment; and (6) contraindications to MRI. Patients with T2DM-related complications were also excluded, including diabetic foot, retinopathy, and nephropathy. The study protocol was approved by the Medical Research Ethics Committee of the Southwest Hospital (Chongqing, China). Written informed consent was obtained from participants after they were informed of the study details. Additionally, the study process strictly obeyed the protocol. Clinical Data The following information about the subjects was recorded using a standardized protocol: handedness, height, weight, BMI [weight in kg]/[height in m] 2 ), resting arm arterial blood pressure, medical history and current medications. For T2DM patients, we also recorded the date of diagnosis to calculate the disease duration. Venous blood samples were collected by venipuncture after overnight fasting for the biometric measurements, including glucose-related parameters (fasting plasma glucose, fasting insulin, fasting C-peptide, and glycosylated hemoglobin [ Neuropsychological Tests All participants underwent cognitive status assessments with a battery of neuropsychological tests in a fixed order that assessed their global cognitive level and major cognitive subdomains. The global level of cognition was evaluated with the MMSE and Montreal Cognitive Assessment (MoCA) tests. The depressive state was evaluated with HAMD to exclude cases with major depression. The major cognitive subdomains were evaluated with the following tests: (1) the Trail Making Test (TMT, parts A and B) for executive function and psychomotor speed (Bowie and Harvey, 2006); (2) the Verbal Fluency Test (VFT) for mental flexibility (Diamond, 2013); (3) the Digit Span Test (DST, forwards and backwards) for working memory (Diamond, 2013); and (4) the Auditory Verbal Learning Test (AVLT, including immediate recall, short-term delayed recall, longterm delayed recall, long-term delayed recognition and total score) for episodic memory (Zhao et al., 2015). A trained neuropsychologist administered the battery of tests and was blinded to the group status. Each participant completed the assessment in approximately 60 min. MRI Data Acquisition MRI scan was performed on the same day that the clinical data were obtained and the cognitive test was performed. MRI data were acquired using a 3.0-T MR scanner (Trio, Siemens Medical, Erlangen, Germany) and a 12-channel head coil. Subjects were instructed to close their eyes, stay awake and avoid thinking about any topics. Earplugs and cushions were used to alleviate noise influence and restrict head motion, respectively. T2-weighted images: repetition time = 6,000 ms, echo time = 89 ms, flip angle = 120 • , field of view = 230 × 230 mm 2 , slices = 20, thickness = 5.0 mm, matrix = 448 × 448 and voxel size = 0.5 × 0.5 × 5.0 mm 3 . Fluid-attenuated inversion recovery (FLAIR) images: repetition time = 9,000 ms, echo time = 93 ms, flip angle = 130 • , field of view = 220 × 220 mm 2 , slices = 25, thickness = 4.0 mm, matrix = 256 × 256 and voxel size = 0.9 × 0.9 × 4.0 mm 3 . Resting-state functional images were collected using an echo planar imaging (EPI) sequence: repetition time = 2,000 ms, echo time = 30 ms, flip angle = 90 • , field of view = 192 × 192 mm 2 , slices = 36, thickness = 3 mm, matrix = 64 × 64 and voxel size = 3 × 3 × 3 mm 3 ; 240 volumes were transversely acquired. T1-weighted structural images were collected using a volumetric 3D magnetization prepared by rapid-acquisition gradient-echo (MP-RAGE) sequence: repetition time = 1,900 ms, echo time = 2.52 ms, flip angle = 9 • , field of view = 256 × 256 mm 2 , slices = 176, thickness = 1 mm, matrix = 256 × 256 and voxel size = 1 × 1 × 1 mm 3 , sagittally scanned. MRI Data Analysis No subjects were excluded after the T2-weighted and FLAIR images were reviewed by two radiologists with at least five years of work experience. Structural and functional data analyses were performed using toolkits based on Statistical Parametric Mapping 8 software (SPM 8, http://www.fil.ion.ucl.ac.uk/spm). Intracranial tissue segmentation was performed with Voxel Based Morphometry Toolbox 8 software (VBM8, version 435) according to a previously described protocol (Whitwell, 2009). The main steps include spatial normalization to match every subject's T1weighed images to the template image, and segmentation of the intracranial tissue into gray matter, white matter and cerebrospinal fluid, which automatically produces information about the volumes of each part of the brain tissue. Gray matter was smoothed with 4 mm full-width half-maximum. Functional data were preprocessed using Data Processing Assistant for Resting-State fMRI (DPARSF module v4.1 of Data Processing & Analysis of Brain Imaging v2.1, http://rfmri.org/ dpabi) according to the standard procedure (Yan and Zang, 2010). All DICOM files were converted to NifTI files. Next, the first 10 volumes were removed to enable the subjects to adapt to the scanning environment, especially the noise. Slice-timing was performed to correct the time differences between slices. Realignment was performed to correct head motion, and a report of head motion was created. Any subjects with head motion > 2.0 mm in any direction of x, y, and x or > 2.0 • at any angle were excluded from the subsequent statistical analyses. Friston 24-parameter model was applied to regress out head motion effects (Yan et al., 2013). Other nuisance variables including white matter signal and cerebrospinal fluid signal were regressed out. Individual functional images were normalized into the Montreal Neurological Institute (MNI) space for intersubject comparison. The resulting images were smoothed with 4 mm full-width halfmaximum. Detrending was applied to remove the systematic drift of the baseline signal. The data were bandpass filtered (0.01-0.08 Hz) to reduce physiological noise at other bands of frequency. Based on preprocessing, DC calculations were performed using DPARSF in a voxel-wise manner with a threshold r > 0.25 in accordance with previous studies (Beucke et al., 2013;Mueller et al., 2013). Peak MNI coordinates of the candidate brain functional hubs were obtained via group comparison of DC maps and considered to be the center of the spherical region of interest (ROI) with a 6-mm radius. Connectivity among the ROIs was analyzed using rs-fMRI data analysis toolkits (REST v1.8, http://www.restfmri.net) in a ROI-wise manner, including FC and GCA analyses. For FC, the correlation coefficients between the ROIs were computed and normalized with Fisher's r-to-z transformation. For GCA, signed-path coefficients between ROIs were computed in a multivariate mode for the subsequent parametric statistical analyses (Hamilton et al., 2011). Statistical Analysis Inter-group comparisons of numeric data were conducted using SPSS software (version 20.0; IBM Corp., Armonk, NY) and included demographic data, clinical parameters, and neuropsychological test scores. First, the data distribution was verified using the Kolmogorov-Smirnov test. Second, independent samples of the t-test and Mann-Whitney U test were applied to normally distributed continuous data and to non-normally distributed data, respectively. The sex proportion was examined with the χ 2 test. p < 0.05 indicated statistical significance. Intra-and inter-group analyses of DC maps were conducted using REST software. A one-sample t-test was applied to investigate the DC pattern with the base of "0" in the intragroup. An independent t-test was applied to investigate the DC differences between the groups with age, sex, education, BMI, blood pressure, blood biometric parameters (except fasting glucose, fasting insulin, fasting C-peptide and HbA 1c ) entered as covariates. Gray matter maps were also included as covariates to control the influence of structural changes in the T2DM patients. The resulting maps were corrected for multiple comparisons using AlphaSim (p < 0.001, cluster size > 20 voxels, number of Monte Carlo simulations = 10,000, cluster connection radius: rmm = 5.0). For the z scores of FC and signed-path coefficients, statistical analyses were performed using SPSS software. A one-sample t-test was applied to investigate the patterns of FC and GCA in each group. An independent sample t-test was applied to investigate the differences between T2DM patients and healthy controls in terms of FC and GCA. The relationships of DC and connectivity with clinical parameters and neuropsychological test scores were explored through partial correlation analyses in SPSS software within the T2DM group. Both the independent sample t-test and the correlation analysis were adjusted with gray matter volumes, and the same covariates that employed in the inter-group analyses of the DC maps were applied to control their possible effect on the results. The independent sample t-test was also controlled for multiple comparisons with Bonferroni correction (p < 0.05/6 for FC and p < 0.05/12 for GCA). RESULTS Demographic and Clinical Data Comparison No significant inter-group differences were found in terms of age, sex, education, BMI, blood pressure, total cholesterol, LDL cholesterol, blood urea nitrogen, cystatin C, uric acid and TSH. The T2DM patients had significantly higher levels of glucoserelated parameters, including fasting plasma glucose, fasting insulin, HbA 1c and HOMA2-IR index (all p < 0.05). The T2DM patients had significantly higher levels of triglycerides and homocysteine but lower fasting C-peptide and serum creatine levels (all p < 0.05). The inter-group differences in FT3 and FT4 were significant (all p < 0.05); however, the levels were within the normal range. The details are presented in Table 1. Neuropsychological Test Comparison The T2DM patients exhibited poorer performance on the tests of MoCA, TMT-B, DST forwards and long-term delayed recall of AVLT (all p < 0.05). No significant inter-group differences were observed in the other tests. The details are presented in Table 2. Brain Volume Comparison The general volumes of gray matter, white matter and brain parenchyma (the sum of gray and white matter) revealed no significant differences between the two groups ( Table 1). DC Analysis Compared to the global mean value, both groups displayed higher DC values in the posterior cingulate cortex, cuneus, angular gyrus, occipital cortex, superior frontal cortex, precentral gyrus and postcentral gyrus (Figures 1A,B). In the T2DM patients, significantly higher DC values were observed in the left anterior cingulate gyrus (ACG), and significantly lower DC values were observed in the bilateral lateral occipital cortices (LOC) and right precentral gyrus (PreCG). Details of the changed brain areas are presented in Table 3 ( Figures 1C,D). Connectivity Analysis For the FC analyses, the one-sample t-test suggested that FC existed in both groups among the ROIs, including the left ACG, right PreCG, and bilateral LOC (all p < 0.05; Figures 2A,B). The independent t-test suggested that the FC of the left ACG-right PreCG (p = 0.015) in T2DM patients was significantly stronger than that in the healthy controls, and the FC values of the left LOC-right PreCG (p = 0.011), right LOC-right PreCG (p = 0.019), and left LOC-right LOC (p = 0.001) were significantly weaker than those in the healthy controls (Figures 2C,D). After multiple comparisons correction, the FC of the left LOC-right LOC in T2DM patients was significantly weaker than that in the healthy controls (p < 0.05/6). With respect to the GCA analyses, the one-sample t-test showed a positive causal effect from the left ACG to the left LOC and a negative causal effect from the left ACG to the right LOC in the healthy controls (Figures 3A,C). However, the conditions were inverted such that a negative causal effect from the left ACG to the left LOC and a positive causal effect from the left ACG to the right LOC were observed in the T2DM patients (Figures 3B,C). Significant differences between the two groups in terms of the two directed functional edges were observed (p < 0.05; Figure 3C). Unfortunately, the independent t-test results of GCA could not bear the multiple comparisons correction (p > 0.05/12). Correlation Analysis Signed-path coefficients from the left ACG to the left LOC were negatively correlated with fasting C-peptide levels (ρ = −0.386, p = 0.007; Figure 4A) in T2DM patients. Poor performance on the DST forwards was associated with elevated HbA 1c levels in T2DM patients (HbA 1c [%], ρ = −0.301, p = 0.040; HbA 1c [mmol/mol], ρ = −0.301, p = 0.040; Figure 4B). However, no correlations were observed between disrupted DC or connectivity and lower scores on the neuropsychological tests or disease duration. DISCUSSION To explore the possible neurological mechanisms underlying cognitive dysfunction in T2DM patients, we combined the DC, FC and GCA approaches to investigate the changes in the candidate brain functional hubs and their connectivity. The increased DC in left ACG and decreased DC in the occipital areas were consistent with a previous study (Cui et al., 2016). A novel finding was the increased DC in the right PreCG. The changed FC pattern of right PreCG may be associated with impaired preparation and execution of goal-directed actions. The GCA approach further revealed the disordered direct connectivity from the left ACG to bilateral occipital areas. The brain regions where identified hubs are located have been reported to be abnormal in previous studies using other neuroimaging metrics in T2DM patients, which suggests that these brain regions are susceptible to T2DM. The ACG is usually identified as the disturbed brain region across experiments and appeared with increased amplitude of low-frequency fluctuation, regional homogeneity Liu D. et al., 2016), and cerebral blood flow (Cui et al., 2017). The aforementioned hyperactivity of ACG has usually been regarded as a means to compensate for the cognitive loss and maintain normal cognition (He et al., 2015;Liu D. et al., 2016;Cui et al., 2017). However, the mechanism of the compensation requires further research. In contrast, a decreased amplitude of low-frequency fluctuation and regional homogeneity were observed in the occipital lobe (Xia et al., 2013;Cui et al., 2014;Peng et al., 2016). Decreased connectivity of the PreCG was found with the posterior cingulate cortex and the thalamic . These brain regions also exhibited gray matter atrophy in T2DM patients (Chen et al., 2012;Garcia-Casares et al., 2014;Peng et al., 2015). We discovered a change in the number of edges that connect the bilateral LOC and right PreCG, which rendered these brain regions candidate functional hubs to be damaged in T2DM patients (Zuo et al., 2012). These findings provided information about the reorganized layout of the brain network with a graph theory-based approach in addition to previous studies. Placing the identified brain functional hubs into the context of brain local networks, the changed FC of the right PreCG may suggest impaired preparation and execution of goal-directed actions. The motor system participates in the constitution of internal representations of sensory information (Crowe et al., 2004) and obtains visually perceived information to prepare a motor response (Merchant and Georgopoulos, 2006). However, the decreased DC and FC values of the right PreCG indicated its decline in communication efficiency with the visual cortex and other brain regions. Additionally, the motor system is closely connected with the ACG which mediates focusing attention on targets to select appropriate actions (Isomura et al., 2003). The increased FC between the ACG and PreCG may be interpreted as brain efforts to guide motor behavior. Notably, the process of goal-directed action is a main constituent of executive function. Moreover, for its motor speed component, the poor executive performance in TMT-B may be the behavioral level evidence of cognitive and motor system dysfunction. The GCA results may further suggest that the reorganization of the brain network in T2DM patients may be associated with the impaired visual information acquisition for executive function. The ACG serves an important role in the top-down control network (Petersen and Posner, 2012). The occipital cortex is an important brain area of vision-related information encoding for working memory which is the primary component of executive function (Bentley et al., 2004;Diamond, 2013). Furthermore, visual cortex can be mediated by the ACG which monitors and resolves conflict by enhancing the to-be-attended information and suppressing distracter stimuli (Petersen and Posner, 2012;Liu Y. et al., 2016). In this study, the disordered excitatory and inhibitory effects from the ACG to the visual cortex suggest that T2DM patients may experience a cognitive impairment in visual information acquisition that is indirectly related to the deficit in executive function. In addition, unlike the scale free organization of molecular networks (Budovsky et al., 2007;Tacutu et al., 2010), the human brain network was proposed to exhibit prominent small-world architecture which facilitates efficient information communication (Liao et al., 2017). The impacts of brain functional hubs and connectivity alterations on the small-world property remains to be further explored. GCA connections appear at where FC connections exist; however, we observed that there was no overlap of significant alterations between undirected and directed connectivity in the present study. On one hand, undirected FC is usually assessed with the correlation coefficient, which can be regarded as descriptive (Biswal et al., 1995). On the other hand, the directed FC depicted by GCA rests explicitly on the linear vector autoregressive models (Seth et al., 2015). It is difficult to make a direct comparison between correlation data and a model. The non-overlapping alterations between the undirected and directed connectivity may result from the distinction of mathematical theories. It is also reported that internet gaming disorder individuals without FC alterations exhibited impaired GCA connections (Guller et al., 2012). This phenomenon requires a further investigation. Our finding also suggested that the level of glycemic parameters may be linked to the reserve of normal brain function in T2DM patients. Elevated HbA 1c was negatively correlated with spontaneous brain activity in the middle temporal gyrus (Xia et al., 2013) and with cognitive performance such as VFT . The higher insulin resistance has been associated with decreased FC between the posterior cingulate cortex and middle temporal gyrus Yang et al., 2016). With the higher elevation of C-peptide in this study, a greater deviation of directed FC was observed in patients than in the healthy controls. Additionally, the patients with lower HbA 1c levels obtained higher scores on the DST forwards. The results were similar to those of several previous studies, which indicated that T2DM patients with elevated glycemic parameters exhibited decreased/inverted brain functional activity and poor cognitive performance (Xia et al., 2013;Chen et al., 2014;Liu D. et al., 2016;Yang et al., 2016). However, the findings of this study did not provide sufficient information about the relationships between the altered brain function and cognitive performance. Additional studies are required to investigate these mysteries. The main limitations of the present investigation are as follows: First, the study comprised a relatively small population. The inter-group comparisons of FC and GCA were unable to bear the multiple corrections. These factors may restrict its statistical power and the explanation of the results. Second, we deduced the visually related cognition impairment in T2DM patients according to fMRI metrics. However, the lack of assessment on this cognitive domain weakened the basis of our inference. Third, the approach of GCA is controversial due to the poor temporal resolution in fMRI studies. However, the GCA method remains a powerful tool to explore the directed connectivity and helps elucidate the complexity of the brain (Seth et al., 2015). A large-sample study with comprehensive cognitive assessments and the application of new technology, such as compressed sensing (Lustig et al., 2008), to accelerate the sampling may solve these problems in the future. We are also considering the application of other network measurements, for instance, betweenness, closeness, path length (Rubinov and Sporns, 2010), which could be combined with the molecular networks to map the complex biological systems of T2DMrelated cognitive impairment in depth Vidal et al., 2011). CONCLUSIONS Our findings suggest that the DC of the left ACG, bilateral LOC and right PreCG, as well as the connectivity among them reflected by FC and GCA in different perspectives, are altered in T2DM patients. These alterations may be associated with the disruption of visual information acquisition and goal-directed action execution, both of which have previously proven to be related to executive function. Patients with lower glycemic parameters may reserve more normal brain functions. This study provides insight into the neurological underpinnings of T2DMrelated cognitive impairment using neuroimaging. AUTHOR CONTRIBUTIONS DL contributed to the experiments, data analysis and writing of the manuscript. SD contributed to performing the experiments and writing and revising the manuscript. CZ contributed to the data collection. PW designed the experiment and revised the manuscript. LC contributed to the data analysis and manuscript revision. XY contributed to the manuscript revision. JZ and JW are the guarantors of this study and had complete access to all data in the study. They accept responsibility for the integrity of the data and the accuracy of the data analysis. FUNDING The study was supported by the National Natural Science Foundation of China (81471647, 81771814) and the Innovation Fund for Younger Investigators of Southwest Hospital of the Third Military Medical University (SWH2013QN09). FIGURE 1 | 1DC value distribution of intra-group and inter-group comparisons. (A,B) The spatial distribution of the DC value in the HC group and T2DM group. (C) The significantly altered DC map in the T2DM group. (D) Comparison of DC value between the two groups. AlphaSim corrected (p < 0.001, cluster size > 20 voxels). The color bar denotes the t-value. Error bars define the SEM. ACG, anterior cingulate gyrus; LOC, lateral occipital cortices; PreCG, precentral gyrus; R, right; L, left. FIGURE 2 | 2FC pattern of intra-group and inter-group comparisons. (A,B) The FC pattern in the HC group and T2DM group. (C) The significantly altered FC in the T2DM group. (D) Comparison of FC z scores between the two groups. *p < 0.05, # p < 0.05/6 (Bonferroni correction). Error bars define the SEM. ACG, anterior cingulate gyrus; LOC, lateral occipital cortices; PreCG, precentral gyrus; R, right; L, left. FIGURE 3 | 3GCA pattern of intra-group and inter-group comparisons. (A,B) Causal effect patterns of the HC group and T2DM group. The red arrow indicates a positive casual effect and its direction. The blue arrow indicates a negative casual effect and its direction. (C) Comparison of signed-path coefficients between the two groups. *p < 0.05. Error bars define the SEM. FIGURE 4 | 4Correlations among the connectivity, neuropsychological performance and diabetes-related parameters. (A) Signed-path coefficients of the left ACG to the left LOC vs. C-peptide (ng/mL). (B) DST forwards vs. HbA 1c (%). The updated homeostasis model assessment of the insulin resistance (HOMA2-IR) index was calculated using the HOMA2 Calculator v2.2.3 software (http://www.dtu.ox.ac. uk/homacalculator/) to evaluate insulin resistance in all subjects.HbA 1c ]), lipoid parameters (total cholesterol, triglyceride, high-density lipoprotein [HDL] cholesterol, low-density lipoprotein [LDL] cholesterol), renal function parameters (blood urea nitrogen, serum creatine, uric acid, and cystatin C), thyroid function parameters (free triiodothyronine [FT3], free thyroxine [FT4] and thyroid-stimulating hormone [TSH]), and homocysteine. TABLE 1 | 1Demographic and clinical data of the subjects.T2DM HC p-value Age (years) 58.66 ± 6.87 57.36 ± 5.42 0.312 Sex (male:female) 28/19 25/22 0.533 a Education (years) 10.60 ± 3.06 10.77 ± 2.65 0.773 T2DM duration (years) 8.87 ± 6.61 - - BMI (kg/m 2 ) 25.46 ± 5.11 23.94 ± 3.87 0.109 Gray matter (cm 3 ) 608.15 ± 58.00 615.16 ± 51.67 0.538 White matter (cm 3 ) 534.91 ± 64.85 524.16 ± 66.56 0.430 Brain parenchyma (cm 3 ) 1143.06 ± 114.78 1139.32 ± 107.80 0.871 Systolic blood pressure (mmHg) 128.91 ± 16.81 134.55 ± 17.88 0.119 Diastolic blood pressure (mmHg) 79.49 ± 10.01 80.26 ± 10.25 0.715 HbA 1c (%) 8.26 ± 2.08 5.63 ± 0.39 < 0.001 HbA 1c (mmol/mol) 66.76 ± 22.80 38.15 ± 4.33 < 0.001 Fasting plasma glucose (mmol/L) 7.47 ± 2.70 5.22 ± 0.46 < 0.001 Fasting insulin (mIU/L) 14.91 (9.58, 24.54) 11.41 (8.39, 17.10) 0.035 b Fasting C-peptide (ng/mL) 1.81 ± 1.05 2.22 ± 0.90 0.047 HOMA2-IR 0.28 (0.19, 0.52) 0.21 (0.16, 0.32) 0.015 b Total cholesterol (mmol/L) 4.89 ± 1.04 5.10 ± 1.02 0.325 Triglyceride (mmol/L) 1.45 (1.05, 1.98) 1.23 (0.91, 1.48) 0.023 b HDL cholesterol (mmol/L) 1.06 ± 0.23 1.41 ± 0.36 < 0.001 LDL cholesterol (mmol/L) 3.16 ± 0.83 3.27 ± 0.76 0.489 Homocysteine (µmol/L) 16.62 ± 10.03 10.70 ± 4.10 < 0.001 Blood urea nitrogen (mmol/L) 6.36 ± 2.45 5.72 ± 1.23 0.113 Serum creatine (µmol/L) 65.00 (57.00, 78.00) 78.00 (66.00, 85.00) 0.007 b Cystatin C (mg/L) 0.72 (0.64, 0.91) 0.76 (0.69, 0.86) 0.639 b Uric acid (µmol/L) 316.67 ± 79.72 328.34 ± 69.09 0.450 Free triiodothyronine, FT3 (pmol/L) 4.31 ± 0.91 5.13 ± 0.72 < 0.001 Free thyroxine, FT4 (pmol/L) 15.52 ± 2.34 16.48 ± 1.89 0.032 Thyroid-stimulating hormone, TSH (mIU/L) 2.12 ± 0.95 2.57 ± 1.54 0.090 p < 0.05 indicates statistical significance. a χ 2 test for sex (n). b Mann-Whitney U-test for non-normally distributed data [median (QR)]. Independent t-test for the other normally distributed continuous data (means ± SD). T2DM, type 2 diabetes mellitus; HC, healthy control; BMI, body mass index; HDL, high-density lipoprotein; LDL, low-density lipoprotein. TABLE 2 | 2Comparison of the neuropsychological test results between the two groups.T2DM HC p-value GENERAL COGNITION MMSE 29.00 (27.00, 29.00) 29.00 (28.00, 29.00) 0.203 a MoCA 23.17 ± 2.88 24.51 ± 2.31 0.015 EXECUTIVE FUNCTION AND PSYCHOMOTOR SPEED TMT-A 74.67 ± 41.11 64.17 ± 32.59 0.173 TMT-B 183.19 ± 89.90 142.00 ± 58.35 0.010 MENTAL FLEXIBILITY VFT 42.17 ± 7.33 40.55 ± 7.05 0.279 WORKING MEMORY DST forwards 8.83 ± 1.37 9.60 ± 1.50 0.011 DST backwards 4.00 (3.00, 5.00) 4.00 (3.00, 4.00) 0.911 a EPISODIC MEMORY AVLT immediate recall 6.46 ± 1.69 6.86 ± 1.49 0.226 AVLT short-term delayed recall 6.77 ± 3.18 7.74 ± 2.41 0.096 AVLT long-term delayed recall 5.40 ± 3.77 6.72 ± 2.43 0.047 AVLT long-term delayed recognition 11.00 (8.00, 13.00) 12.00 (9.00, 13.00) 0.327 a AVLT total score 28.97 ± 9.94 32.34 ± 7.45 0.065 p < 0.05 was considered statistically significant. a Mann-Whitney U-test for non-normally distributed data [median (QR)]. Independent t-test for the other normally distributed continuous data (means ± SD). MMSE, Mini-Mental State Examination; MoCA, Montreal Cognitive Assessment; TMT, Trail Making Test; VFT, Verbal Fluency Test; DST, Digital Span Test; AVLT, Auditory Verbal Learning Test. TABLE 3 | 3Brain regions with significant DC differences between the two groups.Brain regions BA Peak MNI t-value Cluster (voxels) X Y Z 1 Left anterior cingulate gyrus 32 −9 42 9 3.1455 28 2 Right lateral occipital cortex 19 27 −78 24 −3.1327 20 3 Left lateral occipital cortex 19 −24 −84 21 −3.9907 147 4 Right precentral gyrus 43 63 −3 30 −3.4576 55 MNI, Montreal Neurological Institute; BA, Broadmann Area. R, Right; L, left; AlphaSim corrected (p < 0.001, cluster > 20 voxels). Frontiers in Aging Neuroscience | www.frontiersin.org March 2018 | Volume 10 | Article 55 ACKNOWLEDGMENTSThe authors thank all volunteers who participated in the study and the staff of the Department of Laboratory Medicine and Department of Nuclear Medicine at the Southwest Hospital of the Third Military Medical University (Army Medical University) in Chongqing, China for their selfless and valuable assistance.Conflict of Interest Statement:The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.Copyright © 2018 Liu, Duan, Zhou, Wei, Chen, Yin, Zhang and Wang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classification of diabetes mellitus provisional report of a WHO consultation. K G Alberti, P Z Zimmet, 10.1002/(SICI)1096-9136(199807)15:7<539::AID-DIA668>3.0.CO;2-SDiabet. Med. 15Alberti, K. G., and Zimmet, P. Z. (1998). Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classification of diabetes mellitus provisional report of a WHO consultation. Diabet. Med. 15, 539-553. doi: 10.1002/(SICI)1096-9136(199807)15:7<539::AID-DIA668>3.0.CO;2-S Network medicine: a network-based approach to human disease. A L Barabasi, N Gulbahce, J Loscalzo, 10.1038/nrg2918Nat. Rev. Genet. 12Barabasi, A. L., Gulbahce, N., and Loscalzo, J. (2011). Network medicine: a network-based approach to human disease. Nat. Rev. Genet. 12, 56-68. doi: 10.1038/nrg2918 Network biology: understanding the cell's functional organization. A L Barabasi, Z N Oltvai, 10.1038/nrg1272Nat. Rev. Genet. 5Barabasi, A. L., and Oltvai, Z. N. (2004). Network biology: understanding the cell's functional organization. Nat. Rev. Genet. 5, 101-113. doi: 10.1038/nrg1272 Effects of cholinergic enhancement on visual stimulation, spatial attention, and spatial working memory. P Bentley, M Husain, R J Dolan, 10.1016/S0896-6273(04)00145-XNeuron. 41Bentley, P., Husain, M., and Dolan, R. J. (2004). Effects of cholinergic enhancement on visual stimulation, spatial attention, and spatial working memory. Neuron 41, 969-982. doi: 10.1016/S0896-6273(04)00145-X Abnormally high degree connectivity of the orbitofrontal cortex in obsessive-compulsive disorder. J C Beucke, J Sepulcre, T Talukdar, C Linnman, K Zschenderlein, T Endrass, 10.1001/jamapsychiatry.2013.173JAMA Psychiatry. 70Beucke, J. C., Sepulcre, J., Talukdar, T., Linnman, C., Zschenderlein, K., Endrass, T., et al. (2013). Abnormally high degree connectivity of the orbitofrontal cortex in obsessive-compulsive disorder. JAMA Psychiatry 70, 619-629. doi: 10.1001/jamapsychiatry.2013.173 Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. B Biswal, F Z Yetkin, V M Haughton, J S Hyde, 10.1002/mrm.1910340409Magn. Reson. Med. 34Biswal, B., Yetkin, F. Z., Haughton, V. M., and Hyde, J. S. (1995). Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn. Reson. Med. 34, 537-541. doi: 10.1002/mrm.1910340409 Administration and interpretation of the trail making test. C R Bowie, P D Harvey, 10.1038/nprot.2006.390Nat. Protoc. 1Bowie, C. R., and Harvey, P. D. (2006). Administration and interpretation of the trail making test. Nat. Protoc. 1, 2277-2281. doi: 10.1038/nprot.2006.390 Cortical hubs revealed by intrinsic functional connectivity: mapping, assessment of stability, and relation to Alzheimer's disease. R L Buckner, J Sepulcre, T Talukdar, F M Krienen, H Liu, T Hedden, 10.1523/JNEUROSCI.5062-08.2009J. Neurosci. 29Buckner, R. L., Sepulcre, J., Talukdar, T., Krienen, F. M., Liu, H., Hedden, T., et al. (2009). Cortical hubs revealed by intrinsic functional connectivity: mapping, assessment of stability, and relation to Alzheimer's disease. J. Neurosci. 29, 1860-1873. doi: 10.1523/JNEUROSCI.5062-08.2009 Longevity network: construction and implications. A Budovsky, A Abramovich, R Cohen, V Chalifa-Caspi, V Fraifeld, 10.1016/j.mad.2006.11.018Mech. Ageing Dev. 128Budovsky, A., Abramovich, A., Cohen, R., Chalifa-Caspi, V., and Fraifeld, V. (2007). Longevity network: construction and implications. Mech. Ageing Dev. 128, 117-124. doi: 10.1016/j.mad.2006.11.018 Aberrant brain functional connectivity related to insulin resistance in type 2 diabetes: a resting-state fMRI study. Y C Chen, Y Jiao, Y Cui, S A Shang, J Ding, Y Feng, 10.2337/dc13-2127Diabetes Care. 37Chen, Y. C., Jiao, Y., Cui, Y., Shang, S. A., Ding, J., Feng, Y., et al. (2014). Aberrant brain functional connectivity related to insulin resistance in type 2 diabetes: a resting-state fMRI study. Diabetes Care 37, 1689-1696. doi: 10.2337/dc13-2127 Thalamic resting-state functional connectivity: disruption in patients with type 2 diabetes. Metab. Brain Dis. Y C Chen, W Xia, C Qian, J Ding, S Ju, G J Teng, 10.1007/s11011-015-9700-230Chen, Y. C., Xia, W., Qian, C., Ding, J., Ju, S., and Teng, G. J. (2015). Thalamic resting-state functional connectivity: disruption in patients with type 2 diabetes. Metab. Brain Dis. 30, 1227-1236. doi: 10.1007/s11011-015-9700-2 Mapping the brain in type II diabetes: voxel-based morphometry using DARTEL. Z Chen, L Li, J Sun, L Ma, 10.1016/j.ejrad.2011.04.025Eur. J. Radiol. 81Chen, Z., Li, L., Sun, J., and Ma, L. (2012). Mapping the brain in type II diabetes: voxel-based morphometry using DARTEL. Eur. J. Radiol. 81, 1870-1876. doi: 10.1016/j.ejrad.2011.04.025 Participation of primary motor cortical neurons in a distributed network during maze solution: representation of spatial parameters and time-course comparison with parietal area 7a. D A Crowe, M V Chafee, B B Averbeck, A P Georgopoulos, 10.1007/s00221-004-1876-3Exp. Brain Res. 158Crowe, D. A., Chafee, M. V., Averbeck, B. B., and Georgopoulos, A. P. (2004). Participation of primary motor cortical neurons in a distributed network during maze solution: representation of spatial parameters and time-course comparison with parietal area 7a. Exp. Brain Res. 158, 28-34. doi: 10.1007/s00221-004-1876-3 Aberrant functional connectivity of default-mode network in type 2 diabetes patients. Y Cui, Y Jiao, H J Chen, J Ding, B Luo, C Y Peng, 10.1007/s00330-015-3746-8Eur. Radiol. 25Cui, Y., Jiao, Y., Chen, H. J., Ding, J., Luo, B., Peng, C. Y., et al. (2015). Aberrant functional connectivity of default-mode network in type 2 diabetes patients. Eur. Radiol. 25, 3238-3246. doi: 10.1007/s00330-015-3746-8 Altered spontaneous brain activity in type 2 diabetes: a resting-state functional MRI study. Y Cui, Y Jiao, Y C Chen, K Wang, B Gao, S Wen, 10.2337/db13-0519Diabetes. 63Cui, Y., Jiao, Y., Chen, Y. C., Wang, K., Gao, B., Wen, S., et al. (2014). Altered spontaneous brain activity in type 2 diabetes: a resting-state functional MRI study. Diabetes 63, 749-760. doi: 10.2337/db13-0519 Disrupted brain connectivity patterns in patients with type 2 diabetes. Y Cui, S F Li, H Gu, Y Z Hu, X Liang, C Q Lu, 10.3174/ajnr.A4858AJNR. Am. J. Neuroradiol. 37Cui, Y., Li, S. F., Gu, H., Hu, Y. Z., Liang, X., Lu, C. Q., et al. (2016). Disrupted brain connectivity patterns in patients with type 2 diabetes. AJNR. Am. J. Neuroradiol. 37, 2115-2122. doi: 10.3174/ajnr.A4858 Cerebral perfusion alterations in type 2 diabetes and its relation to insulin resistance and cognitive dysfunction. Y Cui, X Liang, H Gu, Y Hu, Z Zhao, X Y Yang, 10.1007/s11682-016-9583-9Brain Imaging Behav. 11Cui, Y., Liang, X., Gu, H., Hu, Y., Zhao, Z., Yang, X. Y., et al. (2017). Cerebral perfusion alterations in type 2 diabetes and its relation to insulin resistance and cognitive dysfunction. Brain Imaging Behav. 11, 1248-1257. doi: 10.1007/s11682-016-9583-9 Altering neuronal excitability to preserve network connectivity in a computational model of Alzheimer's disease. W De Haan, E C W Van Straaten, A A Gouw, C J Stam, 10.1371/journal.pcbi.1005707PLoS Comput. Biol. 131005707de Haan, W., van Straaten, E. C. W., Gouw, A. A., and Stam, C. J. (2017). Altering neuronal excitability to preserve network connectivity in a computational model of Alzheimer's disease. PLoS Comput. Biol. 13:e1005707. doi: 10.1371/journal.pcbi.1005707 Executive functions. A Diamond, 10.1146/annurev-psych-113011-143750Annu. Rev. Psychol. 64Diamond, A. (2013). Executive functions. Annu. Rev. Psychol. 64, 135-168. doi: 10.1146/annurev-psych-113011-143750 Analysing connectivity with Granger causality and dynamic causal modelling. K Friston, R Moran, A K Seth, 10.1016/j.conb.2012.11.010Curr. Opin. Neurobiol. 23Friston, K., Moran, R., and Seth, A. K. (2013). Analysing connectivity with Granger causality and dynamic causal modelling. Curr. Opin. Neurobiol. 23, 172-178. doi: 10.1016/j.conb.2012.11.010 Mini-Mental State Examination (MMSE). M Galea, M Woodward, Galea, M., and Woodward, M. (2005). Mini-Mental State Examination (MMSE). . 10.1016/S0004-9514(05)70034-9Aust. J. Physiother. 51198Aust. J. Physiother. 51:198. doi: 10.1016/S0004-9514(05)70034-9 Structural and functional brain changes in middle-aged type 2 diabetic patients: a cross-sectional study. N Garcia-Casares, M L Berthier, R E Jorge, P Gonzalez-Alegre, A Gutierrez Cardo, J Rioja Villodres, 10.3233/JAD-131736J. Alzheimers. Dis. 40Garcia-Casares, N., Berthier, M. L., Jorge, R. E., Gonzalez-Alegre, P., Gutierrez Cardo, A., Rioja Villodres, J., et al. (2014). Structural and functional brain changes in middle-aged type 2 diabetic patients: a cross-sectional study. J. Alzheimers. Dis. 40, 375-386. doi: 10.3233/JAD-131736 Conserved functional connectivity but impaired effective connectivity of thalamocortical circuitry in schizophrenia. Y Guller, G Tononi, B R Postle, 10.1089/brain.2012.0100Brain Connect. 2Guller, Y., Tononi, G., and Postle, B. R. (2012). Conserved functional connectivity but impaired effective connectivity of thalamocortical circuitry in schizophrenia. Brain Connect. 2, 311-319. doi: 10.1089/brain.20 12.0100 Investigating neural primacy in major depressive disorder: multivariate Granger causality analysis of resting-state fMRI time-series data. J P Hamilton, G Chen, M E Thomason, M E Schwartz, I H Gotlib, 10.1038/mp.2010.46Mol. Psychiatry. 16Hamilton, J. P., Chen, G., Thomason, M. E., Schwartz, M. E., and Gotlib, I. H. (2011). Investigating neural primacy in major depressive disorder: multivariate Granger causality analysis of resting-state fMRI time-series data. Mol. Psychiatry 16, 763-772. doi: 10.1038/mp.2010.46 A rating scale for depression. M Hamilton, 10.1136/jnnp.23.1.56J. Neurol. Neurosurg. Psychiatr. 23Hamilton, M. (1960). A rating scale for depression. J. Neurol. Neurosurg. Psychiatr. 23, 56-62. doi: 10.1136/jnnp.23.1.56 Hyperactivation of working memory-related brain circuits in newly diagnosed middle-aged type 2 diabetics. X S He, Z X Wang, Y Z Zhu, N Wang, X Hu, D R Zhang, 10.1007/s00592-014-0618-7Acta Diabetol. 52He, X. S., Wang, Z. X., Zhu, Y. Z., Wang, N., Hu, X., Zhang, D. R., et al. (2015). Hyperactivation of working memory-related brain circuits in newly diagnosed middle-aged type 2 diabetics. Acta Diabetol. 52, 133-142. doi: 10.1007/s00592-014-0618-7 Neural coding of "attention for action" and "response selection" in primate anterior cingulate cortex. Y Isomura, Y Ito, T Akazawa, A Nambu, M Takada, J. Neurosci. 23Isomura, Y., Ito, Y., Akazawa, T., Nambu, A., and Takada, M. (2003). Neural coding of "attention for action" and "response selection" in primate anterior cingulate cortex. J. Neurosci. 23, 8002-8012. Small-world human brain networks: perspectives and challenges. X Liao, A V Vasilakos, Y He, 10.1016/j.neubiorev.2017.03.018Neurosci. Biobehav. Rev. 77Liao, X., Vasilakos, A. V., and He, Y. (2017). Small-world human brain networks: perspectives and challenges. Neurosci. Biobehav. Rev. 77, 286-300. doi: 10.1016/j.neubiorev.2017.03.018 Aberrant brain regional homogeneity and functional connectivity in middle-aged T2DM patients: a resting-state functional MRI study. D Liu, S Duan, J Zhang, C Zhou, M Liang, X Yin, 10.3389/fnhum.2016.00490Front. Hum. Neurosci. 10490Liu, D., Duan, S., Zhang, J., Zhou, C., Liang, M., Yin, X., et al. (2016). Aberrant brain regional homogeneity and functional connectivity in middle-aged T2DM patients: a resting-state functional MRI study. Front. Hum. Neurosci. 10:490. doi: 10.3389/fnhum.2016.00490 Topdown modulation of neural activity in anticipatory visual attention: control mechanisms revealed by simultaneous EEG-fMRI. Y Liu, J Bengson, H Huang, G R Mangun, M Ding, 10.1093/cercor/bhu204Cereb. Cortex. 26Liu, Y., Bengson, J., Huang, H., Mangun, G. R., and Ding, M. (2016). Top- down modulation of neural activity in anticipatory visual attention: control mechanisms revealed by simultaneous EEG-fMRI. Cereb. Cortex 26, 517-529. doi: 10.1093/cercor/bhu204 . M Lustig, D L Donoho, J M Santos, Pauly , J M , Lustig, M., Donoho, D. L., Santos, J. M., and Pauly, J. M. (2008). Compressed Sensing MRI. 10.1109/MSP.2007.914728IEEE Signal Process. Mag. 25Compressed Sensing MRI. IEEE Signal Process. Mag. 25, 72-82. doi: 10.1109/MSP.2007.914728 Neurophysiology of perceptual and motor aspects of interception. H Merchant, A P Georgopoulos, 10.1152/jn.00422.2005J. Neurophysiol. 95Merchant, H., and Georgopoulos, A. P. (2006). Neurophysiology of perceptual and motor aspects of interception. J. Neurophysiol. 95, 1-13. doi: 10.1152/jn.00422.2005 Impact of diabetes on cognitive function and brain structure. A Moheet, S Mangia, E R Seaquist, 10.1111/nyas.12807Ann. N.Y. Acad. Sci. 1353Moheet, A., Mangia, S., and Seaquist, E. R. (2015). Impact of diabetes on cognitive function and brain structure. Ann. N.Y. Acad. Sci. 1353, 60-71. doi: 10.1111/nyas.12807 Individual variability in functional connectivity architecture of the human brain. S Mueller, D Wang, M D Fox, B T Yeo, J Sepulcre, M R Sabuncu, 10.1016/j.neuron.2012.12.028Neuron. 77Mueller, S., Wang, D., Fox, M. D., Yeo, B. T., Sepulcre, J., Sabuncu, M. R., et al. (2013). Individual variability in functional connectivity architecture of the human brain. Neuron 77, 586-595. doi: 10.1016/j.neuron.2012. 12.028 Magnitude of cognitive dysfunction in adults with type 2 diabetes: a meta-analysis of six cognitive domains and the most frequently reported neuropsychological tests within domains. P Palta, A L Schneider, G J Biessels, P Touradji, F Hill-Briggs, 10.1017/S1355617713001483J. Int. Neuropsychol. Soc. 20Palta, P., Schneider, A. L., Biessels, G. J., Touradji, P., and Hill-Briggs, F. (2014). Magnitude of cognitive dysfunction in adults with type 2 diabetes: a meta-analysis of six cognitive domains and the most frequently reported neuropsychological tests within domains. J. Int. Neuropsychol. Soc. 20, 278-291. doi: 10.1017/S1355617713001483 Cerebral alterations of type 2 diabetes mellitus on MRI: a pilot study. B Peng, Z Chen, L Ma, Y Dai, 10.1016/j.neulet.2015.08.030Neurosci. Lett. 606Peng, B., Chen, Z., Ma, L., and Dai, Y. (2015). Cerebral alterations of type 2 diabetes mellitus on MRI: a pilot study. Neurosci. Lett. 606, 100-105. doi: 10.1016/j.neulet.2015.08.030 Abnormal spontaneous brain activity in type 2 diabetes with and without microangiopathy revealed by regional homogeneity. J Peng, H Qu, J Peng, T Y Luo, F J Lv, L Chen, 10.1016/j.ejrad.2015.12.024Eur. J. Radiol. 85Peng, J., Qu, H., Peng, J., Luo, T. Y., Lv, F. J., Chen, L., et al. (2016). Abnormal spontaneous brain activity in type 2 diabetes with and without microangiopathy revealed by regional homogeneity. Eur. J. Radiol. 85, 607-615. doi: 10.1016/j.ejrad.2015.12.024 The attention system of the human brain: 20 years after. S E Petersen, M I Posner, 10.1146/annurev-neuro-062111-150525Annu. Rev. Neurosci. 35Petersen, S. E., and Posner, M. I. (2012). The attention system of the human brain: 20 years after. Annu. Rev. Neurosci. 35, 73-89. doi: 10.1146/annurev-neuro-062111-150525 Brain networks and cognitive architectures. S E Petersen, O Sporns, 10.1016/j.neuron.2015.09.027Neuron. 88Petersen, S. E., and Sporns, O. (2015). Brain networks and cognitive architectures. Neuron 88, 207-219. doi: 10.1016/j.neuron.2015.09.027 Complex network measures of brain connectivity: uses and interpretations. M Rubinov, O Sporns, 10.1016/j.neuroimage.2009.10.003Neuroimage. 52Rubinov, M., and Sporns, O. (2010). Complex network measures of brain connectivity: uses and interpretations. Neuroimage 52, 1059-1069. doi: 10.1016/j.neuroimage.2009.10.003 Diverse type 2 diabetes genetic risk factors functionally converge in a phenotype-focused gene network. C Sandor, N L Beer, C Webber, 10.1371/journal.pcbi.1005816PLoS Comput. Biol. 131005816Sandor, C., Beer, N. L., and Webber, C. (2017). Diverse type 2 diabetes genetic risk factors functionally converge in a phenotype-focused gene network. PLoS Comput. Biol. 13:e1005816. doi: 10.1371/journal.pcbi.1005816 Granger causality analysis in neuroscience and neuroimaging. A K Seth, A B Barrett, L Barnett, 10.1523/JNEUROSCI.4399-14.2015J. Neurosci. 35Seth, A. K., Barrett, A. B., and Barnett, L. (2015). Granger causality analysis in neuroscience and neuroimaging. J. Neurosci. 35, 3293-3297. doi: 10.1523/JNEUROSCI.4399-14.2015 The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. R Tacutu, A Budovsky, V E Fraifeld, 10.1007/s10522-010-9265-8Biogerontology. 11Tacutu, R., Budovsky, A., and Fraifeld, V. E. (2010). The NetAge database: a compendium of networks for longevity, age-related diseases and associated processes. Biogerontology 11, 513-522. doi: 10.1007/s10522-010-9265-8 Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. R Tacutu, A Budovsky, H Yanai, V E Fraifeld, 10.18632/aging.100413Aging. 3Tacutu, R., Budovsky, A., Yanai, H., and Fraifeld, V. E. (2011). Molecular links between cellular senescence, longevity and age-related diseases -a systems biology perspective. Aging 3, 1178-1191. doi: 10.18632/aging.100413 Interactome networks and human disease. M Vidal, M E Cusick, A L Barabasi, 10.1016/j.cell.2011.02.016Cell. 144Vidal, M., Cusick, M. E., and Barabasi, A. L. (2011). Interactome networks and human disease. Cell 144, 986-998. doi: 10.1016/j.cell.2011.02.016 A new rating scale for age-related white matter changes applicable to MRI and CT. L O Wahlund, F Barkhof, F Fazekas, L Bronge, M Augustin, M Sjogren, 10.1161/01.STR.32.6.1318Stroke. 32Wahlund, L. O., Barkhof, F., Fazekas, F., Bronge, L., Augustin, M., Sjogren, M., et al. (2001). A new rating scale for age-related white matter changes applicable to MRI and CT. Stroke 32, 1318-1322. doi: 10.1161/01.STR.32. 6.1318 Voxel-based morphometry: an automated technique for assessing structural changes in the brain. J L Whitwell, 10.1523/JNEUROSCI.2160-09.2009J. Neurosci. 29Whitwell, J. L. (2009). Voxel-based morphometry: an automated technique for assessing structural changes in the brain. J. Neurosci. 29, 9661-9664. doi: 10.1523/JNEUROSCI.2160-09.2009 The signaling hubs at the crossroad of longevity and age-related disease networks. M Wolfson, A Budovsky, R Tacutu, V Fraifeld, 10.1016/j.biocel.2008.08.026Int. J. Biochem. Cell Biol. 41Wolfson, M., Budovsky, A., Tacutu, R., and Fraifeld, V. (2009). The signaling hubs at the crossroad of longevity and age-related disease networks. Int. J. Biochem. Cell Biol. 41, 516-520. doi: 10.1016/j.biocel.2008.08.026 Altered baseline brain activity in type 2 diabetes: a resting-state fMRI study. W Xia, S Wang, Z Sun, F Bai, Y Zhou, Y Yang, 10.1016/j.psyneuen.2013.05.012Psychoneuroendocrinology. 38Xia, W., Wang, S., Sun, Z., Bai, F., Zhou, Y., Yang, Y., et al. (2013). Altered baseline brain activity in type 2 diabetes: a resting-state fMRI study. Psychoneuroendocrinology 38, 2493-2501. doi: 10.1016/j.psyneuen.2013.05.012 A comprehensive assessment of regional variation in the impact of head micromovements on functional connectomics. C G Yan, B Cheung, C Kelly, S Colcombe, R C Craddock, A Di Martino, 10.1016/j.neuroimage.2013.03.004Neuroimage. 76Yan, C. G., Cheung, B., Kelly, C., Colcombe, S., Craddock, R. C., Di Martino, A., et al. (2013). A comprehensive assessment of regional variation in the impact of head micromovements on functional connectomics. Neuroimage 76, 183-201. doi: 10.1016/j.neuroimage.2013.03.004 DPARSF: a Matlab toolbox for "pipeline" data analysis of resting-state fMRI. C G Yan, Y F Zang, 10.3389/fnsys.2010.00013Front. Syst. Neurosci. 413Yan, C. G., and Zang, Y. F. (2010). DPARSF: a Matlab toolbox for "pipeline" data analysis of resting-state fMRI. Front. Syst. Neurosci. 4:13. doi: 10.3389/fnsys.2010.00013 Altered intranetwork and internetwork functional connectivity in type 2 diabetes mellitus with and without cognitive impairment. S Q Yang, Z P Xu, Y Xiong, Y F Zhan, L Y Guo, S Zhang, 10.1038/srep32980Sci. Rep. 632980Yang, S. Q., Xu, Z. P., Xiong, Y., Zhan, Y. F., Guo, L. Y., Zhang, S., et al. (2016). Altered intranetwork and internetwork functional connectivity in type 2 diabetes mellitus with and without cognitive impairment. Sci. Rep. 6:32980. doi: 10.1038/srep32980 Auditory verbal learning test is superior to rey-osterrieth complex figure memory for predicting mild cognitive impairment to Alzheimer's disease. Q Zhao, Q Guo, X Liang, M Chen, Y Zhou, D Ding, 10.2174/1567205012666150530202729Curr. Alzheimer Res. 12Zhao, Q., Guo, Q., Liang, X., Chen, M., Zhou, Y., Ding, D., et al. (2015). Auditory verbal learning test is superior to rey-osterrieth complex figure memory for predicting mild cognitive impairment to Alzheimer's disease. Curr. Alzheimer Res. 12, 520-526. doi: 10.2174/1567205012666150530202729 Network centrality in the human functional connectome. X N Zuo, R Ehmke, M Mennes, D Imperati, F X Castellanos, O Sporns, 10.1093/cercor/bhr269Cereb. Cortex. 22Zuo, X. N., Ehmke, R., Mennes, M., Imperati, D., Castellanos, F. X., Sporns, O., et al. (2012). Network centrality in the human functional connectome. Cereb. Cortex 22, 1862-1875. doi: 10.1093/cercor/bhr269