Gene expressions associated with longer lifespan and aging exhibit similarity in mammals

Abstract Although molecular features underlying aging and species maximum lifespan (MLS) have been comprehensively studied by transcriptome analyses, the actual impact of transcriptome on aging and MLS remains elusive. Here, we found that transcriptional signatures that are associated with mammalian MLS exhibited significant similarity to those of aging. Moreover, transcriptional signatures of longer MLS and aging both exhibited significant similarity to that of longer-lived mouse strains, suggesting that gene expression patterns associated with species MLS contribute to extended lifespan even within a species and that aging-related gene expression changes overall represent adaptations that extend lifespan rather than deterioration. Finally, we found evidence of co-evolution of MLS and promoter sequences of MLS-associated genes, highlighting the evolutionary contribution of specific transcription factor binding motifs such as that of E2F1 in shaping MLS-associated gene expression signature. Our results highlight the importance of focusing on adaptive aspects of aging transcriptome and demonstrate that cross-species genomics can be a powerful approach for understanding adaptive aging transcriptome.


INTRODUCTION
Owing to the recent advances in the understanding of the molecular mechanisms of aging, it is becoming clear that aging itself is a treatable condition ( 1 ). Since aging is the largest risk factor for physical functional decline and numerous diseases, de v elopment of its treatment has a potential to vastly improve the quality of life of elderly people ( 2 ). Ther efor e, further understanding of the mechanisms of aging is of crucial importance. In order to gain a systematic understanding of fundamental molecular processes of aging, much effort has been devoted to clarify agingrelated gene expression changes ( 3 ). Microarray and sequencing technologies enab led comprehensi v e e valuation of aging-related changes in transcriptome (4)(5)(6)(7)(8), which basically r epr esents the overall state of cells / tissues. To date, Tabula Muris Senis ( 7 , 8 ) and the Genotype-Tissue Expression (GTEx) consortium ( 6 ) provided the most comprehensi v e information of aging-related transcriptomic changes in mouse and human, respecti v ely. These up-to-date largescale anal yses, to gether with numerous preceding studies, identified some common transcriptional features of aging, such as upregulation of inflammatory genes ( 3 , 9 ), which are belie v ed to contribute to chronic inflamma tory sta te called inflammaging. In addition to the overr epr esentation of some gene ontologies, it has been discovered that the majority of aging-related gene expression changes represent re v ersals or e xtensions of de v elopmental gene e xpression changes ( 10 ), suggesting that aging-related gene expr ession changes ar e not just the result of damage accumulation but are associated with the de v elopmental program ( 11 , 12 ). A link between de v elopment and aging has been also suggested from a physiological point of view (12)(13)(14). Gene expression changes driven by such developmental drift and damage accumulation may contribute to agingrela ted deteriora tion ( 11 , 15 ). Howe v er, although agingrelated transcriptomic changes have been described in great detail, their functional consequences still remain largely unproven.
Cross-species analysis is another approach to study genetic programs involved in lifespan regulation. For example, it has been reported that genes whose manipulation affects organismal lifespan are more conserved during evolution, and studies of such so-called longevity-associated genes (LAGs) have illuminated species-dependent and -independent longevity-associa ted pa thways ( 16 ), novel pr otein-pr otein interaction networks that may contribute to longevity, and genetic links between aging and age-related diseases ( 17 ). In addition, investigation of transcriptional features of long-li v ed species would help to re v eal prolongevity genetic progr ams. Tr anscriptomic analysis of e xceptionally long-li v ed species, such as the naked mole-rat ( 18 , 19 ), blind mole-rat ( 20 , 21 ), bat ( 22 ), gray whale ( 23 ) and bow-head whale ( 24 , 25 ) provided insights into the pathways that might potentially support their long lifespan. In a milestone study published in 2015, a group led by Vadim N Gladyshev reported a transcriptome of liver, kidney, and brain of 33 mammalian species (4 Carnovora, 2 Chiropetra, 1 Didelphimorphia, 1 Diprotodoncia, 1 Erinaceomorpha, 4 Euungulata, 1 Lagomorpha, 1 Monotremata, 8 Primates, 9 Rodentia and 1 Soricomorpha) and uncovered parallel evolution of gene expression and lifespan ( 26 ). They found that the expressions of DNA repair-and detoxificationrelated genes positi v el y correlate with maxim um lifespan (MLS), whereas the expressions of genes involved in central metabolism and growth hormone signaling negati v ely correlate with MLS. One year later the same group reported transcriptomic study of cultured skin fibroblasts of 16 mammalian species (2 Chiroptera, 1 Eulipotyphla and 13 Rodentia) and confirmed the positi v e correlation between MLS and the expressions of genes involved in DNA repair ( 27 ). Recently in 2022, a group led by Vera Gorbunova and Andr ei Seluanov r eported transcriptomic study of li v er, kidney, brain, heart, lung and skin of 26 mammals (3 Eulipotyphla and 23 Rodentia) ( 28 ). While these mammals are highly di v erse in their MLS, they are mostly rodents and are closel y related phylo geneticall y. Anal yses using this dataset would thus have high power to detect MLS-associated signatures, although it may be unclear whether the findings a ppl y onl y to Rodentia and Eulipotyphla or also to other orders of mammals. They found that genes related to DNA repair, microtubule organization, and RNA transport are mor e expr essed in long-li v ed Rodentia / Eulipotyphla while genes related to energy metabolism and in-flammation are less expressed in long-li v ed Rodentia / Eulipotyphla.
To date, although cross-species transcriptomic studies re v ealed many features of MLS-associated gene expression patterns, how expression of these genes relates to aging still remains largely unexplored. Here, we showed that gene expression patterns associated with species MLS and aging exhibited significant similarity. Moreover, transcriptional signatures of longer MLS and aging were all associated with longer lifespan in BXD ( 29 , 30 ) mouse strains. Our results suggest that gene expression patterns associated with species MLS contribute to extended lifespan e v en within a species (i.e. BXD mouse strains), and that aging-related gene expression changes overall r epr esent adapta tions tha t extend lifespan ra ther than deterioration.

Identification of genes associated with MLS in mammals
Genes associated with MLS in mammals were identified as described below using amalgamated cross-species RNA-Seq dataset compiled by Fukushima and Pollock ( 31 ). This dataset includes transcriptomic data of 52 vertebrates that are normalized by TPM and corrected by surrogate variable analysis (SVA) to remove biases likely arising from heterogeneous sampling / sequencing procedures within the dataset. For the current study, we used transcriptomic data of li v er, kidney, brain, heart, ovary, and testis of 29 mammals (3 Artiodactyla, 3 Carnovora, 1 Cingulata, 1 Didelphimorphia, 1 Eulipotyphla, 1 Lagomorpha, 1 Monotremata, 1 Perissodactyla, 10 Primates, 6 Rodentia and 1 Scandentia) of which their MLS is r egister ed in AnAge database ( 32 ). Detailed information of these species, namely species name, common name, order, maximum lifespan, body weight, and age of sexual maturity are gi v en in Supplementary Table S1. Expression data of different species were joined based on human Ensembl gene name. Expression data of non-human species that do not have 1:1 human orthologs were not joined with those of other species. For the analysis, we only used genes whose expression data were available in 10 or more species in a gi v en tissue. A total of 16 616, 16 590, 16 175, 15 934, 14 677 and 16 001 genes passed these criteria in li v er, kidney, brain, heart, ovary and testis, respecti v ely (hereafter referred to as 'all investigated genes'). Association between gene expression and longevity was analyzed by Spearman test or by PGLS analysis using SVA-corrected log 2 -transformed TPM values and log 2transformed MLS values. PGLS analysis was conducted using R libraries 'ape" and 'nlme'. To take phylogenetic uncertainty into account, we performed PGLS analyses on 100 phylogenetic trees downloaded from vertlife.org ( 33 ) that were generated using randomized parameters and the maximum P -value was used to test the association between MLS and gene expression level. Note that a set of 100 sampled trees has been suggested to capture uncertainty of parameters used for generating phylogenetic trees ( 33 ). Benjamini, Krieger and Yekutieli method ( 34 ) was used to control the FDR and a FDR of < 10% was considered to be statistically significant.
Nucleic Acids Research, 2023, Vol. 51, No. 14 7207 Enrichment analyses of GO terms and transcription factor targets.
Go enrichment analyses were conducted using GOrilla ( 35 ) and Enrichr ( 36 ). Enrichment of transcription factor targets was analyzed by the ChEA3 ( 37 ) or by Fisher's exact test using the Enrichr gene set library EN-CODE and ChEA Consensus TFs from ChIP-X. All investigated genes as defined above were used as the background except for the analyses using the ChEA3 and Enrichr w e b tool.

Analysis of evolutionary changes in promoter sequences of MLS-associated genes.
Sequences of 1000 bp promoter r egions wer e r etrie v ed using BioMart ( 38 ). The genome assembly of Damaraland mole-rat was on the Scaffold le v el and was not available in BioMart, and thus was excluded from the analysis. Nonredundant vertebrate JASPAR CORE motifs ( 39 ) were scanned using FIMO ( 40 ) with default settings.

Retrieval of aging-associated genes.
The lists of genes that are differentially expressed between young and old mice tissues, and the bulk ( 7 ) and single-cell RNA-Seq data ( 8 ) of young and old mouse tissues were retrie v ed from the Tabula Muris Senis dataset. The lists of genes that are differ entially expr essed between young and old human tissues were retrie v ed from the GTEx dataset ( 6 ) via Enrichr library GTEx Aging Signatures 2021 ( 36 ). These signatures were generated by Evangelista et al. using limma-voom R package ( 41 ) as described in detail in their paper under the section 'Gene expression signatures of aging from GTE' ( 42 ). Genes differ entially expr essed between 20-29 years age group and either 30-39, 40-49, 50-59, 60-69, 70-79 years old group were considered to be associated with aging. Genes upregulated in some age group and downregulated in some other age group were not included in aging-associated genes. The lists of genes that are differ entially expr essed between 20-59 and 60-99 years old human brain were retrie v ed from the dataset generated by Carl W. Cotman and colleagues ( 4 ). The numbers of these aging-associated genes are provided in Supplementary Tables S2. Expression levels shown in the figures are gi v en in TPMs or DESeq2 normalized counts ( 43 ).

Cross-species comparison of the ratio of T cells in kidney and liver immune cells.
For mouse tissue, processed droplet scRNA-Seq data and the results of clusterization were retrie v ed from the Tabula Muris Senis dataset ( 8 ). For human tissues, count data were retrie v ed from the original studies ( 44 , 45 ) and processed using Seurat version 2.3.4 and then clustered by shared nearest neighbor method using 29 and 30 principal components for li v er and kidney, respecti v ely (resolution = 0.8). Cell clusters with high CD45 le v els were e xtracted as immune cells and further classified by shared nearest neighbor method using 30 principal components (resolution = 0.1 for mouse kidney, 0.2 for mouse li v er, 0.24 for human kidney and 0.2 for human li v er). Cell clusters were then classified based on the expression levels of CD11c, CD11b, CD19 and CD3 ε . A cell cluster in human kidney that could not be classified using the above markers was found to express high levels of kidney epithelial markers and was not considered as a part of immune cell. A cell cluster expressing high levels of HBA1 or HBA2 (red blood cell markers) were identified in mouse kidney, human kidney, and human li v er, and were not considered as a part of immune cells.
Retrieval of genes associated with longevity of rodentia / eulipothyphla and BXD strains.
The lists of genes that are correlated with MLS of Rodentia / Eulipothyphla (Spearman test, BH-adjusted Pvalue < 0.05 and | | > 0.4) were retrie v ed from the study by Lu et al. ( 28 ). The lists of genes whose expression in li v er and kidney correlate with the median lifespan of female mice of BXD mouse strains (| | > 0.4) wer e r etrie v ed from the study by Vitiello et al. ( 30 ). Identifica tion of genes associa ted with the median lifespan of BXD mouse strains were conducted using microarray that contains one or more probes for each gene. Genes identified to be positi v ely associated with lifespan by one probe but also identified to be negati v ely associated with lifespan by another probe were not included in longevity-associated genes.

Statistical significance test of the o verlap betw een tw o gene sets.
Statistical significance of the overlap between two gene sets was tested by binominal test under null hypothesis that the two gene sets overlap only by chance. All statistical tests were two-sided.

Aging-like GO enrichment of mammalian MLS-associated genes.
In order to investigate the characteristics of MLSassociated genes, we first e xtracted e xpression data of li v er, kidney, brain, heart, ovary, and testis of 29 mammals (Figure 1 A, Supplementary Table S1) with known maximum lifespan ( 32 ) from amalgamated cross-species RNA-Seq dataset compiled by Fukushima and Pollock in 2020 ( 31 ). We then identified MLS-associated genes using Spearman correlation analysis as described in Figure 1 B and in Materials and Methods (Supplementary Tables S2, S3). Correlation with FDR less than 10% was considered to be statistically significant. In the li v er and kidney, 637 and 470 genes correlated with MLS, respecti v ely (Figure 1 C), and fewer genes correlated with age of sexual maturity or body weight (Supplementary Table S3). In brain, 68 genes correlated with MLS and about twice the number of genes correlated with age of sexual maturity and body weight. In heart, only eight genes correlated with MLS and e v en fewer genes correlated with age of sexual maturity or body weight. In ovary, none of the genes correlated with MLS and only one gene correlated with body weight, while as many as 127 genes correlated with age of sexual maturity. In testis, 58, 265 and 10 genes correlated with MLS, age of sexual maturity, and body weight, respecti v ely, suggesting The names of 29 mammalian species that were used to extract mammalian MLSassociated genes and a phylogenetic tree drawn by R package 'ape' using phylogenetic data downloaded from VertLife.or g. A heatmap sho ws the availability of transcriptomic data and each box is colored blue if transcriptomic data was available in the corresponding tissue of corresponding species. Right bar graphs show MLS, age of female sexual maturity, and body weight of each species. ( B ) Procedures for the identification of mammalian MLS-associated genes. ( C ) The bar graph shows the number of MLS-associated genes identified in this study. ( D ) The correlation plot r epr esents the overlap among mammalian MLS-associated genes identified in li v er, kidney, brain, heart, ovary and testis by Spearman and PGLS methods. Each pie r epr esents the percentage of overlapping genes in the gene set shown on the left side. tha t cross-species dif fer ential gene expr essions in r eproducti v e organs more reflect age of sexual maturity than MLS. MLS-associated genes found in these tissues had absolute correla tion coef ficient values grea ter than 0.53 (Supplementary Table S3) and were largely tissue-specific (Figure 1 D).
We next examined the enrichment of gene ontology (GO) terms relati v e to all inv estiga ted genes, and found tha t genes whose expression in the kidney was positi v ely and negati v ely correlated with mammalian MLS were enriched with inflammatory and mitochondrial genes, respecti v ely (Supplementary Figure S1). Considering that high expression of inflammatory genes and low expression of mitochondrial genes are primary features of aging ( 3 , 9 ), these results may suggest some similarity between MLS-associated genes and aging-associate genes.

MLS-and mouse aging-associated genes overlap significantly.
Although our MLS-associated genes exhibited enrichment of GO terms that is reminiscent of aging transcriptome, it is not clear at this point whether it is just tha t dif ferent genes within the same GO term is associated with mammalian MLS and aging, or whether MLS-associated genes significantl y overla p with aging-associated genes (i.e. genes whose expression was correlated with aging). To address this question, we compared MLS-associated genes with the genes whose expression changes during aging in mouse li v er, kidney, brain and heart, which were identified by Wyss-Coray and Quake labs and Tabula Muris consortium ( 7 ) (Supplementary Table S2, S4).
Surprisingly, we found a significant overlap between MLS-and aging-associated genes in li v er, kidney, and brain (Figure 2 A). That is, genes positi v ely correlated with both MLS and aging, and genes negati v ely correlated with both MLS and aging, which we hereafter collecti v ely refer to as 'concordant genes', appeared significantly more often than expected under the assumption of independence of MLS-and aging-associated genes (Figure 2 B). In addition, in the kidney, genes positi v ely correlated with MLS and sim ultaneousl y negati v ely correlated with aging, and genes negati v el y correlated with MLS and sim ultaneousl y positi v ely correlated with aging, which we hereafter collecti v ely refer to as 'discordant genes', appeared significantly less often than expected under the assumption of independence of MLS-and aging-associated genes (Figure  2 B). This indicates that the discordance between MLS-and aging-associated gene expression patterns were smaller than expected by chance. These results collectively support the similarity between the transcriptional signatures of aging and longer MLS. This similarity was maintained in the kidney e v en after r emoving genes corr elated with age of sexual maturity or body weight (Supplementary Figure S2). Thus, although these factors are known to be associated with species MLS, they are not the primary determinant of the similarity between the transcriptional signatures of aging and longer MLS.
Genes positi v ely correlated with both mammalian MLS and mouse aging in the kidney were enriched with inflamma tion-rela ted genes (Supplementary Figure S1). These inflammatory genes included se v eral T cell markers, such as CD2, CD4 and CD6, suggesting that the enrichment of inflamma tion-rela ted genes a t least partially reflects an increase in T cells in longer-li v ed mammals (Figure 2 C). Indeed, re-analysis of Tabula Muris Senis ( 8 ) and public human scRNA-Seq datasets ( 44 , 45 ) showed that the percentages of T cells in immune cells in kidney and li v er were higher in human than in mouse (Figure 2 D). Genes negati v ely correlated with both mammalian MLS and mouse aging in the kidney were enriched with mitochondrial genes (Figure 2 C, Supplementary Figure S1). Re-analysis of Tabula Muris Senis scRNA-Seq dataset ( 8 ) showed that these mitochondrial genes tend to be downregulated during aging in kidney parenchymal cells, such as proximal convoluted tubule epithelial cells (Supplementary Figure  S3). Ne xt, in or der to corroborate the similarity between MLS-and aging-associated genes, we extracted MLSassociated genes using another method, namely phylogenetic generalized least squares (PGLS) regression analysis ( 46 ) (Supplementary Table S2, S3). Unlike Spearman test, PGLS is a parametric test but does take the phylogenetic non-independence of species into consideration. As a result, we found 201, 832, and 114 MLS-associated genes in li v er, kidney, and brain, respecti v ely (FDR < 0.1). Ovary and testis had only small numbers of MLS-associated genes (23 and 22 genes, respecti v ely). No MLS-associated gene was identified in the heart (Supplementary Table S2, S3). MLS-associated genes identified by PGLS method were largel y non-overla pping with those identified by Spearman method (Figure 1 D). The discrepancy is likely due at least partially to the limited sensitivity of statistical analyses but it should be noted that the overlap between the lists of MLSassociated genes identified by these methods was still statistically significant (Supplementary Figure S4A). Nonetheless, MLS-associated gene expression patterns identified by PGLS method also exhibited significant similarity with gene expression patterns associated with mouse aging (Supplementary Figure S4B and C, Tables S2 and S4). The fact that Spearman and PGLS methods pointed to the same conclusion indicates its robustness. In order to enhance the comprehensi v eness of the following analyses, we hereafter combined the lists of MLS-associated genes obtained by these two methods.
In order to further verify our finding, we next investigated the similarity between MLS-and aging-associated gene expression signatures using MLS-associated genes of li v er, kidney, brain, heart, lung, and skin of Rodentia and Eulipotyphla that have been identified by Lu et al. ( 28 ). Since 23 out of 26 mammals used in their study were rodents, species-dependent gene expression would be less affected by mammalian order-dependent traits irrelevant to MLS. Rodentia / Eulipotyphla MLS-associated genes identified by Lu et al . were largely non-overlapping with mammalian MLS-associated genes identified by us (Figure 3 A). Nonetheless, Rodentia / Eulipotyphla MLS-associated gene e xpression patterns e xhibited significant similarity with gene expression patterns associated with mouse aging in li v er, kidney, brain, heart, and lung (Figure 3 B and C, Supplementary Table S5, S6). Genes correlated with mouse skin aging was the exception and their overlap with MLSassociated genes was significantly smaller than expected by  chance. This could be explained by our observation that aging-related gene expression changes were highly fluctuating in mouse skin (Supplementary Figure S5). Thus, unlike other tissues, aging-related genes in mouse skin are less likely to r epr esent cumulati v e effects of aging, which could be due to the fact that external factors play major roles in skin aging.

MLS-and human aging-associated genes overlap significantly.
Although we established the similarity between gene expression pa tterns associa ted with mouse aging and the MLS of Rodentia / Eulipotyphla or a wider range of mammals (hereafter collecti v ely refer to as species MLS), it remains to be tested whether aging-associated gene expression patterns of other species, most importantly human, also exhibit similarity to MLS-associated gene expression patterns. We ther efor e r etrie v ed genes that are associated with human aging from the GTEx dataset ( 6 ) via Enrichr library ( 36 ) and tested their relationship with MLS-associated genes (Supplementary Table S2 and S6). Odds ratios of concordant genes were higher than 1 in mammalian MLS-associated genes in two out of three tissues (li v er and brain) and in Rodentia / Eulipotyphla MLS-associated genes in fiv e out of six tissues (li v er, kidney, brain, heart, and skin) (Figure 4 A-D). On the other hand, odds ratios of discordant genes w ere low er than 1 in mammalian MLS-associated genes in two out of three tissues (li v er and kidney) and in Rodentia / Eulipotyphla MLS-associated genes in all six tissues. Importantly, odds ratios of concordant genes were higher than those of discordant genes in all cases. The similarity between species MLS-and human aging-associated gene expression patterns was statistically significant in kidney, brain, and skin (Figure 4 B and D). In order to corroborate this finding, we retrie v ed genes associated with human brain aging from another independent dataset generated by Carl W. Cotman and colleagues ( 4 ). Using this dataset, we confirmed that gene expression pattern associated with MLS in the brain exhibited a significant similarity to gene expression pattern of human brain aging ( Figure 4 E and F). Taken together, our results indicate the similarity between the transcriptional signatures of longer MLS and aging of mouse and human. Genes positi v ely and negati v ely correlated with species MLS both contributed to the similarity between MLS-and aging-associated gene expression patterns (Supplementary Figure S6).

MLS-and aging-associated expressions are both linked to longer lifespan.
Similarity between gene expression patterns associated with MLS and aging might suggest that a significant portion of aging-related gene expression changes is actually beneficial rather than harmful for longevity. In order to investigate how MLS-and aging-associated genes relate to the variation in lifespan within a mammalian species, we next compared MLS-and aging-associated gene expression patterns with those associated with longer median lifespan in geneticall y hetero geneous BXD mouse strains ( 29 , 30 ) that were generated by crossing C57BL / 6J and DBA / 2J mice. Gene expression patterns associated with longer median lifespan in these strains were identified in li v er and kidney (and also in eye, bone and adrenal glands) in a previous study by Ewald et al. ( 30 ). In li v er and kidney, genes positi v ely correlated with both species MLS and mouse lifespan, and genes negati v ely correlated with both species MLS and mouse lifespan, appeared significantly more of-ten than expected by chance. On the other hand, genes positi v ely correlated with species MLS and simultaneously negati v ely correlated with mouse lifespan, and genes negati v ely correlated with species MLS and simultaneously positi v ely correlated with mouse lifespan, appeared significantly less often than expected by chance ( Figure 5 ). These results suggest that gene expression pa tterns associa ted with species MLS contribute to extended longevity even within a species. Importantly, the same significant similarities were observed between gene expression patterns associated with mouse aging and mouse lifespan. Gene expressions commonly associated with species MLS and mouse aging also exhibited similarity to those associated with mouse lifespan ( Figure 5 , Supplementary Figure S7). These results suggest tha t aging-rela ted gene e xpression changes ov erall r epr esent adapta tions tha t extend lifespan rather than deterioration.
Although both mammalian and Rodentia / Eulipotyphla MLS-associated genes strongly supported our conclusion, it should be noted that these two gene sets have very different relationships with immune regula tion. Tha t is, while we found that inflammatory genes were overr epr esented in genes that are positi v ely correlated with mammalian MLS, Lu et al. ( 28 ) reported such overrepresentation in genes that are negati v ely correlated with Rodentia / Eulipotyphla MLS. To better understand this difference, we extracted immune / inflammatory response genes whose Spearman coefficients of correlation between species MLS and expressions in liver, kidney, brain, and heart were significantly different between mammals and Rodentia / Eulipotyphla (Figure 6 A). We found that genes involv ed in positi v e regulation of cytokine production and NF B signaling tend to be positi v ely and negati v ely correlated with mammalian and Rodentia / Eulipotyphla MLS, respecti v ely. On the other hand, genes involved in negati v e regulation of cytokine production and NF B signaling tend to be negati v ely and positi v ely correlated with mammalian and Rodentia / Eulipotyphla MLS, respecti v ely (Figure 6 B). Importantl y, imm une / inflammatory response genes whose correla tion coef ficients with mammalian MLS were significantly higher than with Rodentia / Eulipotyphla MLS showed a weak but significant positi v e correlation with median lifespan of BXD mouse strains (Figure 6 C). Immune / inflammatory response genes that were upregulated during aging in mouse li v er and kidney were also positi v ely correlated with mouse lifespan. On the other hand, immune / inflammatory response genes whose correla tion coef ficients with Rodentia / Eulipotyphla MLS were significantly higher than with mammalian MLS did not correlated with mouse lifespan. Thus, adaptation of immune regulation to longer lifespan e volv ed in different ways in Rodentia / Eulipotyphla and other mammals.
Our finding that gene expressions associated with aging and longer lifespan exhibit similarity aligns well with a previous argument by de Magalh ˜ aes et al. ( 9 ) that many of aging-related gene expression changes could represent adapta tions to aging. Aging-rela ted increase in the expression of transcriptional repressor REST (RE1-Silencing Transcription Factor) is a proven example of such adaptations. In human brain, age-related upregulation of REST is beneficial for longevity and cognitive performance ( 47 , 48 ).
Ther efor e, we tested whether such adaptation also occurred during the evolution of longer lifespan. Enrichment analysis of transcription factor (TF) targets using ChEA3 ( 37 ) and ENCODE ChIP-seq library showed that REST targets were the most enriched TF targets in genes whose expression in the brain negati v ely correlated with mammalian MLS (Figure 7 A). In genes whose expression in the brain negati v ely correlated with Rodentia / Eulipotyphla MLS, the top 6th enriched TF targets were those of REST. Moreover, consensus REST targets (i.e. genes that were commonly identified to be bound by REST in multiple experiments) were significantly over-and underr epr esented in the genes whose expression in the brain was negati v ely and positi v ely corr elated with species MLS, r especti v ely (Figur e 7 B). Thus, suppr ession of REST targets that is protecti v e a gainst a ging-rela ted brain d ysfunction seems to occur both during aging and the evolution of longer lifespan.

TFs regulating the evolutionary changes in the expression of MLS-associated genes.
It is important to note that the aforementioned results do not necessarily indica te tha t REST activity itself correlates with MLS in the brain. Rather, REST binding sites might have been increased selectively in the promoter regions of genes associated with MLS in the brain. It is also very possible that the evolutionary changes in the expression of these REST-and MLS-associated genes were dri v en by other TFs. Understanding of the regulatory mechanisms that had dri v en the evolutionary changes in the expression of MLS-associated genes is of great importance, since the effects of MLS-associated gene expression on longevity could depend on the biological conte xt. In or der to investigate such regulatory mechanisms, we analyzed the evolu-Nucleic Acids Research, 2023, Vol. 51, No. 14 7215  tionary changes in the presence of TF binding sequences in the 1000 bp upstr eam r egions of transcriptional start sites (TSS) of mammalian MLS-associated genes using FIMO ( 40 ) and vertebrate JASPAR CORE position weight matrices ( 39 ). We limited our TFBS analysis to 1000 bp upstream regions of TSS due to technical reasons, although it should be kept in mind that other regions also contain many TFBS. Rodentia / Eulipotyphla MLS-associated genes were not investigated because promoter sequences were not available for the majority of species used in the comparati v e study ( 28 ). For each motif and each species, we calculated ( 1 ) the ratio of genes containing at least one motif in their pro-moters and ( 2 ) the average number of motifs per promoter for MLS-associated genes and background genes (i.e. all investigated genes). We then calculated the odds ratios of these two values between MLS-associated genes and background genes and tested their correlation with MLS (Spearman test). Motif was considered to be associated with MLS genes if the odds ratios of both of these values were significantly correlated with MLS (FDR < 5%). As a result, we found se v en TF binding motifs whose presence in the promoter regions of mammalian MLS-associated genes corr elated with MLS (Figur e 7 C, Supplementary Table S7), suggesting that MLS e volv ed in parallel with promoter sequences of MLS-associated genes. Importantly, evolutionary changes in the presence of these transcriptional activator-and r epr essor-binding motifs wer e all consistent with the evolutionary changes in the expression of MLSassociated genes (Figure 7 C). For example, in genes whose expression in the brain positively correlated with MLS, the presence of the binding motif of transcriptional activator LHX2, which has been reported to play major roles in forebrain de v elopment ( 49 ), was positi v ely correlated with MLS. On the other hand, in genes whose expression in the brain negati v ely corr elated with MLS, the pr esence of the binding motifs of transcriptional activator E2F1 ( 50 ) and putati v e transcriptional r epr essor TGIF2LY ( 51 ) was negati v ely and positi v ely correlated with MLS, respecti v ely. Notabl y, genes w hose expression in the brain negatively correlated with MLS not only lost binding motifs of an activator E2F (E2F1) during the evolution of longevity (Figure 7 C), but were also enriched with genes targeted by a r epr essor E2F (E2F6) ( 50 ) in the longest-li v ed primate (human) (Figure 7 A). These consistent results suggest that E2F transcription factors rather than REST had shaped MLS-associated gene expr ession signatur e in the brain during the evolution of longevity.

DISCUSSION
Our conclusions are based on the analyses of two different types of MLS-associated genes, namely those identified by Lu et al. ( 28 ) in Rodentia / Eulipotyphla and those identified by us in a wider range of mammals. Identification of MLS-associated genes in Rodentia / Eulipotyphla would be less affected by mammalian order-dependent traits irrelevant to MLS. On the other hand, findings obtained from the investigations of genes associated with longevity in a wider range of mammals including primates and human would more likely to be relevant to human biology. Importantly, MLS-associated genes identified using Spearman and PGLS methods both lead to the same conclusions. The complementarity of these analyses ther efor e enhances the robustness and generality of our conclusions (Figure 8 ). Regarding the identification of MLS-associated genes, two novel sophisticated methods using ranking-based method ( 23 ) and machine learning analysis ( 52 ) have been recently de v eloped by groups led by Alexey A. Moskalev, Vadim E. Fraifeld and Robi Tacutu, and their utilization may allow identification of additional overlaps between MLS-and aging-associated genes.
In this study, we clearly showed that genes correlated with mouse / human aging and species MLS overlap significantly. MLS-associated gene expression signature is likely contributing to longer lifespan and thus our results suggest that a significant portion of aging-related gene expression changes could be beneficial for longevity. Although dysregula ted inflamma tory responses underlie pa thogenesis of many diseases, their physiological functions are also crucial for tumor suppression ( 53 ). High expression of inflam-matory genes in long-li v ed mammals may r epr esent better immunity. Indeed, we found that gene expression patterns associated with species MLS and aging were all associated with longer lifespan in genetically heterogeneous mouse strains. This strongly supports the notion that gene expression pa tterns associa ted with species MLS contribute to extended lifespan even within a species, and that agingrelated gene expression changes overall represent adaptations that extend lifespan.
Ther e ar e some caveats r egarding the interpretation of our results. First, considering that loss of proteostasis is a major hallmark of a ging ( 54 ), a ge-related changes in mRNA le v els may tend not to be reflected in protein levels. In the current study, we only analyzed mRNA le v els and it is not clear if the same conclusion could be drawn from the same analysis using the le v els of functional proteins instead of mRNA le v els. Second, e xpression data used in this study was quantified at the gene le v el and not at the transcript le v el. Thus, we cannot e xclude the possibility that age-or species-dependent splicing has some influence on our results. Third, changes in mRNA levels during aging and in the evolution of longevity may be the result rather than the cause of aging / longevity. For example, age-r elated decr eases in mRNA le v els may reflect DNA damage blocking their transcription, and longer-lived species may downregulate damage-vulnerable gene transcripts. Fourth, molecular changes observed during aging are also often observed during development ( 10 , 11 , 55-57 ). Many de v elopment-associated genes are subsumed within aging-associated genes, especially when the latter were identified by comparing young adult and old animals. Ther efor e, the similarity between gene expressions associated with aging and longer lifespan may be at least partially due to a link between genetic programs regulating de v elopment and longevity. Finally, it should be noted that although genes associated with aging and longer lifespan exhibited statistically significant similarity, these two gene sets were still largel y non-overla pping with each other. One of the major reasons for this is that genes identified to be associated with aging or longer lifespan highly depend on species, strains, sex, conditions of animals, and methods used for the analysis. For example, as described above for Spearman and PGLS methods, each statistical method has strengths and weaknesses. In this regard it should be also noted that DESeq2 ( 43 ) that was used to identify mouse aging-related genes has been shown to se v erely understate FDR values when the sample size is large ( 58 ). Also, it has been reported that genes associated with aging greatly vary among 129sv, B ALB / c, CB A, DB A, C57BL / 6, C3H and B6C3F mice ( 59 ). If the majority of genes associated with longer lifespan in a gi v en set of species ov erlapped with genes associated with aging in a gi v en strain of animal, these MLS-associated genes would not overlap to the same extent with genes associated with aging in other strains or species. Therefore, if genes associated with longer lifespan and aging have similarity in mammals regardless of species , strains , sex, and other conditions, the overlap between these two gene sets, e v en if statistically significant, theoretically has to be weak.

DA T A A V AILABILITY
All the data analyzed in this study were retrie v ed from freely accessib le pub lic data bases as described a bove. Source data underlying the figures are provided in the supplementary tables.

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.