Abstract Discerning the relationship between sociality and longevity would permit a deeper understanding of how animal life history evolved. Here, we perform a phylogenetic comparative analysis of ~1000 mammalian species on three states of social organization (solitary, pair-living, and group-living) and longevity. We show that group-living species generally live longer than solitary species, and that the transition rate from a short-lived state to a long-lived state is higher in group-living than non-group-living species, altogether supporting the correlated evolution of social organization and longevity. The comparative brain transcriptomes of 94 mammalian species identify 31 genes, hormones and immunity-related pathways broadly involved in the association between social organization and longevity. Further selection features reveal twenty overlapping pathways under selection for both social organization and longevity. These results underscore a molecular basis for the influence of the social organization on longevity. Subject terms: Gene expression, Social evolution, Transcriptomics, Social behaviour, Ageing __________________________________________________________________ To elucidate the relationship between sociality and longevity, the authors perform phylogenetic and transcriptomic comparative analysis of mammals. They find that group-living species lived longer than solitary species and identify 31 genes, hormones, and immunity-related pathways involved in this connection. Introduction Extant mammals exhibit a wide diversity of grouping arrangements or social organizations, including solitary living, pair-living, and various forms of group-living^[48]1, e.g., multilevel society (e.g., Rhinopithecus spp.)^[49]2 and eusociality (e.g., Heterocephalus glaber)^[50]3. Mammals also show an extreme 100-fold variation in maximum lifespan (or longevity), ranging from ~2 years in shrews (e.g., Sorex spp.) to more than 200 years in bowhead whales (Balaena mysticetus)^[51]4. The evolutionary relationships between sociality and longevity in mammals are complex^[52]5 yet important for understanding evolutionary strategies, i.e., life history diversity across organisms. In mammals, most of the evidence for links between sociality and longevity comes from single species. For example, affiliative social bonds, which are pervasive among group-living species, can extend a species’ lifespan by decreasing mortality and enhancing health and survival outcomes. In humans, strong social relationships can reduce the risk of physiological dysregulation^[53]6. With respect to other mammals, female chacma baboons (Papio ursinus) with strong and stable social bonds live longer than those with weak connections^[54]7,[55]8; similar results have been reported in rhesus macaques (Macaca mulatta)^[56]9. Conversely, a negative correlation between affiliative relationships and longevity has been reported in female yellow-bellied marmots (Marmota flaviventer)^[57]10. Even though a small number of cross-species studies have tested the association between sociality and aging or longevity, they primarily focused on eusocial species^[58]11–[59]13 and cooperatively breeding species^[60]14,[61]15. Therefore, it remains unclear whether associations between longevity and other types of social organization are a common feature across the mammalian phylogeny. In addition, the molecular mechanisms underlying the evolutionary association between social organization and longevity are not fully understood. Previous studies have suggested some possible processes, e.g., stress reduction, parasite infections, and pace of life (fast-slow continuum)^[62]16. According to the stress-buffering hypothesis, strong social bonds or social support can reduce adverse environmental stimuli or stress and enhance health and longevity in humans^[63]17. Social organization can also influence the spread of parasites in the population. For example, group-living species are vulnerable to infectious diseases because of the high social contact rates and close social interactions among individuals, but social species may have evolved a strong immune defense to minimize disease risk and protect themselves against pathogens^[64]18. One more possible link between social organization and longevity is the pace of life, which reflects an organism’s strategic allocation of resources between survival and reproduction. Species with a fast life history are characterized by rapid development, high reproductive rates, and short lifespan, whereas species with a slow life history are characterized by slow development, low reproductive rates, and a long lifespan^[65]19. Given that sociality and fitness are positively associated in some mammals^[66]20–[67]22 and social bonds require major time investments before they yield survival benefits, social bonds are expected to have evolved in species with a slow life history and a longer lifespan^[68]23. In summary, we are only beginning to understand the evolution and the molecular mechanisms underlying the diversity of social organizations^[69]24–[70]28 and longevity^[71]29,[72]30. In this work, we compare models with different evolutionary conjectures between social organization (i.e., solitary, pair-living, and group-living)^[73]31 and longevity across ~1000 mammals using a Bayesian framework. Moreover, we conduct a comparative brain transcriptomic analysis of 94 mammals to detect candidate genes and pathways associated with social organization and longevity, after controlling for body mass, ecological factors, life history traits, and phylogenetic relationships. We show that group-living species lived longer than solitary species and identify 31 genes, hormones, and immunity-related pathways involved in the correlated evolution of social organization and longevity. Results Evolutionary pathways for social organization and longevity To assess the evolutionary transitions among social states and evolutionary pathways for longevity, we collected data on the social organization for as many extant mammalian species as possible through a comprehensive literature survey. We assigned 974 species into three types of social organization: solitary (n = 497), pair-living (n = 115), and group-living (n = 412). Fifty species had more than one state (Fig. [74]1a, details of classification in “Methods”). Data on body mass and longevity (defined by the maximum lifespan of a given species) for these species were also collected (Supplementary Table [75]1 and Supplementary Data [76]1). We first used phylogenetic comparative methods to calculate the phylogenetic signal of social organization and longevity. The maximum likelihood estimates of Pagel’s λ for the three social states was 0.94 when taking into account social polymorphism for a given species (phylogenetic signal test: n = 974, log[λ] = −788.03, log[0] = −1344.67, P < 0.001) and was 0.94 when using the uni-state species subset (n = 924, log[λ] = −535.38, log[0] = −938.78, P < 0.001). Pagel’s λ for longevity was 0.97 (n = 974, log[λ] = 319.18, log[0] = 1475.26, P < 0.001), illustrating that closely-related taxa generally have similar social organizations and longevity. Fig. 1. Evolutionary analyses of social organization and longevity in 974 mammalian species. [77]Fig. 1 [78]Open in a new tab a Phylogenetic distribution of social organization, adult body mass, and longevity (n = 974). The inner circle represents a species’ social organization: solitary (blue), pair-living (orange), and group-living (red). The middle layer indicates the absolute adult body mass (g) and the outer layer indicates the longevity (years); both variables were log[10] transformed. Colors of shadings distinguish different mammals’ orders. b Difference in absolute longevity and c relative longevity (residuals of longevity, which was adjusted for body mass) across the three social organization states (solitary: n = 491, pair-living: n = 65, group-living: n = 368). We accounted for the effects of phylogenetic non-independence among species using a Phylogenetic ANOVA. Two-sided and Hommel method adjust P values are reported. The white dot represents the median in two violin plots, and the black box represents interquartile ranges (IQRs), i.e., the 25th and 75th percentiles. The whiskers extend up to the largest value within 1.5-fold IQR. Species numbers (n) are indicated in each social organization, respectively. Correlated evolution analysis for absolute short-lived (cyan) or long-lived state (purple): d non-solitary or solitary (blue); e non-pair-living or pair-living (orange); f non-group-living or group-living (red). d and f demonstrate correlated evolution. The number of species used in the analyses was n = 974. Arrows depict the likelihood of a transition between states, and their thickness corresponds to the magnitude of the various rates. Numbers indicate the transition rate across ten independent runs, and data are presented as mean ± SD. Silhouette images of animals are from PhyloPic database [[79]http://phylopic.org/]. Source data are provided as a Source data file. To determine the evolutionary pathways connecting the three states of social organization within mammals, we tested four models that allowed the transition rates between any of two states to vary differently. These four alternative models were: (a) the equal rates model (ER), in which all transition rates were the same; (b) the increasing complexity model (IC), which allowed transitions between solitary and pair-living, pair-living and group-living, but not solitary and group-living; (c) the all-rates-different model (ARD) or parameter-rich model^[80]24, in which all transition rates were different; and (d) the reversible-jump Markov chain Monte Carlo-derived model (RJ-MCMC), which is derived from the data using the reversible-jump procedure in Bayes Traits and has the highest posterior support^[81]32 (Supplementary Fig. [82]1a–d, Supplementary Table [83]2). Model comparisons showed that the ARD model was the best-supported model, which was significant against the RJ-MCMC model (Log BF = 9.24), the ER model (Log BF = 33.36), and the IC model (Log BF = 71.70) (Table [84]1). The ARD model showed that the transition rates varied across the three states of social organization (Supplementary Fig. [85]2a). For example, the transition rate from pair-living to solitary was 14 times higher than from solitary to pair-living (q[pair-living-solitary] = 4.00 ± 1.55 × 10^−3; q[solitary-pair-living] = 0.29 ± 1.50 × 10^−4), suggesting that the pair-living state was relatively unstable. Table 1. Comparison of evolutionary models for social organization and longevity Trait Model Rank Parameters Mean likelihood Log BF Social organization ARD 1 6 −555.27 – RJ-MCMC 2 5 −559.89 9.24 ER 3 1 −571.95 33.36 IC 4 4 −591.12 71.70 Absolute lifespan ARD 1 2 −346.26 – RJ-MCMC 1 2 −346.26 – ER 2 1 −361.43 30.34 Relative lifespan ARD 1 2 −329.61 – RJ-MCMC 1 2 −329.61 – ER 2 1 −339.16 19.10 [86]Open in a new tab n = 974. Log BF = 2 × (log marginal likelihood complex model – log marginal likelihood simple model). The simple model is favored when Log BF < 2, while there is positive evidence to support the complex model when Log BF > 2 (strong evidence, 5–10; very strong evidence, >10)^[87]32. ARD all-rates-different model, RJ-MCMC reversible-jump Markov chain Monte Carlo-derived model, ER equal rates model, IC increasing complexity model. We then reconstructed the evolutionary pathway of longevity by constructing three alternative models (i.e., ER model, ARD model, and RJ-MCMC model) (Supplementary Fig. [88]1e–g). Comparisons of these three models showed that the RJ-MCMC and ARD models were better supported, on the basis of Bayes factors (BF), than the ER model (Table [89]1 and Supplementary Table [90]3). The average transition rate from a long-lived state to a non-long-lived state (q[absolute] = 2.09 ± 3.85 × 10^−4; q[relative] = 1.71 ± 6.23 × 10^−4) was about four times greater than that from a non-long-lived to a long-lived state (q[absolute] = 0.54 ± 1.42 × 10^−4; q[relative] = 0.53 ± 1.13 × 10^−4) (Supplementary Fig. [91]2b, c). Correlated evolution of social organization and longevity We conducted phylogenetic ANOVA analyses to estimate differences in longevity among the three states of social organization while controlling for phylogenetic non-independence among species. Longevity was significantly different between the solitary state and the group-living state, with group-living species showing higher longevity than solitary species (phyloAVOVA: n[multi-states] = 974, t = 12.40, P-adjust = 0.04; n[uni-state] = 924, t = 12.28, P-adjust = 0.02, Fig. [92]1b). Since longevity is correlated with adult body mass (Spearman’s rank test: r = 0.71, P < 2.20 × 10^−16), we also measured relative longevity which was calculated using the body mass adjusted residuals with the equation from the AnAge database (“Methods”). Similar results were obtained for relative longevity (phyloAVOVA: n[multi-states] = 974, t = 12.01, P-adjust = 4.80 × 10^−2; n[uni-state] = 924, t = 11.94, P-adjust = 0.02, Fig. [93]1c). In addition, we conducted MCMCglmm models to control for phylogeny, body mass and factors related to external mortality: activity (diurnal, nocturnal or others), lifestyle (terrestrial, aerial, arboreal, semi-arboreal, freshwater, marine, or terrestrial-marine), and fossoriality (non-fossorial or subterranean). The results consistently showed that pair-living or/and group-living species lived longer than solitary species when using multi-states of the social organization dataset (MCMCglmm: n[multi-states] = 947, pair-living vs. solitary, post mean = 0.10, pMCMC = 1.11 × 10^−3; group-living vs. solitary, post mean = 0.06, pMCMC < 6.00 × 10^−4; pair-living and group-living, post mean = 0.06, pMCMC = 0.03) and uni-state of the social organization dataset (n[uni-state] = 897, pair-living vs. solitary, post mean = 0.10, pMCMC < 6.00 × 10^−4; group-living vs. solitary, post mean = 0.06, pMCMC = 1.11 × 10^−3). The results of activity, lifestyle and fossoriality are shown in Supplementary Table [94]4. To evaluate whether changes in longevity depended on social organization, we compared the independent and dependent RJ-MCMC models of three combinations of variables: non-solitary/solitary and absolute short-lived/long-lived (>26 years), non-pair-living/pair-living and absolute short-lived/long-lived as well as non-group-living/group-living and absolute short-lived/long-lived. The results favored the dependent model for both solitary (Log BF = 3.18, Table [95]2 and Supplementary Table [96]5) and group-living (Log BF = 9.58, Table [97]2 and Supplementary Table [98]6), suggesting the existence of correlated evolution between social organization and longevity across the mammalian phylogeny. We also considered the effect of taxonomic sampling and the different classifications of long-lived species on the correlated evolution analyses. Random taxon sampling (i.e., randomly selecting 50 to 95% of the total number of species at a 5% interval) and repeated comparisons of independent and dependent RJ-MCMC models provided further support for the correlated evolution of solitary living and longevity (>26 years, 51% of model comparisons), as well as group-living and longevity (>26 years, 63% of model comparisons, Supplementary Fig. [99]3a). Consistent with these findings, random taxon sampling and model comparisons with two different cut-offs for long-lived species (>17 or >35 years), suggested the correlation between social organization and longevity. There was a correlation between solitary living and longevity when using the 17-year cut-off (79% of model comparisons) and 35-year cut-off (52% of model comparisons); and also a correlation between group living and longevity when using the 35-year cut-off (98% of model comparisons, Supplementary Fig. [100]3b, c). In addition, when taking the uncertainty of phylogenetic relationships into account and using a different phylogenetic tree^[101]33, the correlated evolution between solitary and longevity was supported by the analyses with the 26-year and the 17-year cut-off; the correlated evolution between group-living and longevity was supported by the analyses with the 17-year and 35-year cut off (n[multi-states] = 969; Supplementary Table [102]7). Table 2. Likelihoods of dependent and independent models estimated for the correlated evolution of social organization and longevity Social states Longevity states Mean likelihood of model Log BF Correlated evolution (no/yes) (no/yes) Dependent Independent Solitary Absolute long-lived −795.02 −796.61 3.18 Yes Pair living Absolute long-lived −643.81 −637.58 −12.46 No Group living Absolute long-lived −775.78 −780.57 9.58 Yes Solitary Relative long-lived −768.18 −777.02 17.68 Yes Pair living Relative long-lived −627.74 −622.30 −10.88 No Group living Relative long-lived −756.35 −760.49 8.28 Yes [103]Open in a new tab n = 974. Absolute long-lived species: longevity >26 years. Relative long-lived species: the residual of longevity >1.38. The residual of longevity for each species was calculated using the body mass adjusted residuals using the equation from the AnAge. We then attempted to determine whether transitions to a long-lived state were more likely in group-living than solitary species. Model estimation revealed that the transition rate from a short-lived state to a long-lived state was higher for non-solitary than solitary species (q[non-solitary] = 12.44 ± 1.84 × 10^−3; q[solitary] = 2.77 ± 3.79 × 10^−3), and higher for group-living than non-group-living species (q[groupliving] = 11.86 ± 1.55 × 10^−3; q[non-group-living] = 2.86 ± 9.89 × 10^−4; Fig. [104]1d–f). This result is consistent with the prediction that group-living species are more likely to be long-lived. We then tested if transitions to a group-living state were different for long-lived and short-lived species; we found that the transition rate from a solitary to a non-solitary state was the same in long-lived species and short-lived species (q[long-lived] = 12.47 ± 2.52 × 10^−3; q[short-lived] = 12.44 ± 2.37 × 10^−3; Fig. [105]1d). The transition rate from a non-group-living state to a group-living state was also the same in long-lived species and short-lived species (q[long-lived] = 11.86 ± 1.53 × 10^−3; q[short-lived] = 11.86 ± 1.54 × 10^−3; Fig. [106]1f), suggesting that longer longevity does not promote the formation of group-living. In addition, the correlated evolution of social organization and longevity was also supported when body mass was taken into account (Table [107]2; residuals of longevity > 1.38, solitary: Log BF = 17.68; group-living: Log BF = 8.28, Supplementary Fig. [108]4a–c). The random taxon sampling of different classifications of relative long-lived species further supported the correlated evolution between solitary living and longevity (residuals of longevity >1.38, 92% of model comparisons; residuals of longevity >1.83, 98% of model comparisons; Supplementary Fig. [109]5a–c), and the correlated evolution between group living and longevity (residuals of longevity >1.38, 55% of model comparisons; residuals of longevity >1.83, 95% of model comparisons; Supplementary Fig. [110]5a–c). Similarly, when a different phylogenetic tree and different cut-offs of the residuals of longevity were used, the correlated evolution between social organization and relative longevity was also supported (n[multi-states] = 969; Supplementary Table [111]7). In addition, to investigate whether longevity favors any social organization transformation, we compared the independent and dependent RJ-MCMC models using species with a uni-state of social organization. The results supported that social organization transformation favors longer life during solitary transit to the group-living state rather than from solitary transit to pair-living, or from pair-living transit to the group-living state (Supplementary Table [112]8). The transition rate from short-lived to long-lived species was higher in group-living than solitary species (absolute longevity: q[solitary] = 2.98 ± 6.97 × 10^−3; q[group-living] = 10.45 ± 5.29 × 10^−3; relative longevity: q[solitary] = 3.37 ± 8.70 × 10^−3; q[group-living] = 10.30 ± 9.51 × 10^−3, Supplementary Fig. [113]6a–f). Gene expression of social organization and longevity To identify genes that could underpin the correlated evolution of social organization and longevity, we generated brain transcriptomics of 94 mammals belonging to 14 orders, 39 families, and 67 genera (Fig. [114]2a, Supplementary Data [115]2, “Methods”). Specifically, 57% of species and 62% of samples (166 samples of 54 species) were newly collected in this study. The sampled species were assigned to three states of social organization (solitary, n = 26; pair-living, n = 11; group-living, n = 65); eight species had more than one state. The sampled species also covered a longevity range from 3.2 years in Chinese mole shrew (Anourosorex squamipes) to 122.5 years in Homo sapiens (Fig. [116]2a, Supplementary Data [117]3, and Supplementary Table [118]9). Using the human coding sequences as a reference, we employed a reciprocal-blast approach to identify the orthologous gene set. The orthologous genes that were shared by >70% of the total number of species (i.e., 66 of 94 species) were selected for subsequent analyses (“Methods”). Finally, gene expression for 13,402 orthologous genes was measured across all brain samples. We then used MCMCglmm models to identify genes whose expression significantly correlated with any of the social organization states; these models also controlled for phylogenetic relationships and other confounding factors, including adult body mass, activity (nocturnal, diurnal, and other), diet (carnivore, herbivore, and omnivore), and lifestyle (non-aerial and aerial). Hundreds of genes were significantly associated with solitary living (up: 366 genes, down: 254 genes), pair-living (up: 393 genes, down: 66 genes), and group-living (up: 162 genes, down: 321 genes) (Supplementary Fig. [119]7a–f, Supplementary Data [120]4). There were three overlapping genes among the three states of social organization: ATP1A2, ALDH1L2, and WDFY1. We also detected genes that were shared by two states (solitary-pair-living: 21 genes; solitary-group-living: 284 genes; pair-living-group-living: 14 genes, Supplementary Fig. [121]7f). We detected 262 genes whose expression was significantly correlated with longevity; this was supported by all four different models in the MCMCglmm analyses (“Methods”, Supplementary Fig. [122]8, Supplementary Data [123]5; see Supplementary Data [124]6 for the results of each model). Fig. 2. Genes and pathways whose expression was correlated with social organization and longevity in 94 mammalian species. [125]Fig. 2 [126]Open in a new tab a Species (n = 94) with RNA-seq and six life history traits (social organization, activity, diet, lifestyle, adult body mass, and longevity) used in the MCMCglmm analyses. Colorful shadings display different mammals’ orders. Silhouette images of animals are from PhyloPic database [[127]http://phylopic.org/]. b Venn diagram showing the number of significant overlapping genes across solitary, pair-living, group-living, and longevity. c Clusters according to the function of 31 genes that were significantly associated with social organization and longevity. Each node represents a pathway from the Reactome database^[128]34, [129]35; pathways in which significant genes were involved are colored brown. Light green shading represents two clusters of gene function. d, e Example of a significant gene (XRCC6) that was downregulated in solitary species, upregulated in group-living species and also positively correlated with lifespan; the regression lines were generated from the linear regression model. Purple range display 95% confidence interval around the smooth line. Coefficients (post mean) and P-values (pMCMC) from the MCMCglmm analyses are also shown. The number of species used in the MCMCglmm was n = 94. f A heat map showing pathways that were significantly associated with social organization and longevity. S: solitary; PL: pair-living; GL: group-living; ML1–ML4: longevity in model 1 to model 4 (see “Methods”). Color code for social organization and longevity: blue = solitary; orange = pair-living; red = group-living; purple = lifespan. Source data are provided as a Source data file. In total, we found 31 genes whose expression levels were significantly associated with both social organizations and longevity (Fig. [130]2b, Supplementary Data [131]7). The pathway topology analysis of these genes using the Reactome platform^[132]34,[133]35 revealed two strong clusters. The first cluster included immune-related genes (Fig. [134]2c). Nine genes (i.e., UBL7, TNNT3, XRCC6, ATP2A2, NPHS1, KALRN, C1QC, MCL1, and ZFP36) that are involved in the innate immune response^[135]36. Gene C1QC (MCMCglmm: solitary, post mean = 1.16, pMCMC = 0.04; longevity, post mean = −2.33 ± 0.05, pMCMC = 0.03 ± 7.17 × 10^−3) participates in encoding a complex heterotrimer C1q that plays a vital recognition role in the complement pathway. C1q has diverse biological functions, including complement activation, innate immune defense, cellular regulation, reproduction, development and neurodegenerative disorders^[136]37,[137]38. The well-known immune gene ZFP36 (MCMCglmm: solitary, post mean = 1.07, pMCMC = 0.02; longevity, post mean = 2.21 ± 0.07, pMCMC = 0.02 ± 5.05 × 10^−3) modulates anti-viral immunity by controlling T cell activation^[138]39 and protects against inflammatory diseases through regulating inflammatory cytokines, such as TNF-α^[139]40,[140]41. ZFP36 also plays a role in neuroprotection and inhibits neuronal apoptosis^[141]42. Another gene of interest was XRCC6 (MCMCglmm: solitary, post mean = −1.50, pMCMC = 3.33 × 10^−3, Fig. [142]2d; group-living, post mean = 1.39, pMCMC = 4.44 × 10^−3, Fig. [143]2e; longevity, post mean = 2.27 ± 0.05, pMCMC = 0.01 ± 4.37 × 10^−3), which encodes subunit p70 of the p70/p80 autoantigen. A recent study has shown that a splicing variant in XRCC6 may cause autism, a disorder that causes significant social, communication, and behavioral challenges^[144]43. Knockout of XRCC6 decreases lifespan in mice^[145]44 and the high expression of XRCC6 leads to a longer average lifespan in humans^[146]45. Thus, this gene likely plays a role in both longevity and social organization. The second cluster of genes whose expression was correlated with longevity and social organization consisted of genes involved in the regulation of hormones, neural systems, and signal transduction (Fig. [147]2c), e.g., MTM1, SLC29A2, ATP2A2, KALRN, RHOBTB2, SLC6A19, and MCL1. Some of these genes are suggested to play a role in social behavior. For instance, the KALRN gene (MCMCglmm: group-living, post mean = −1.16, pMCMC = 0.02; longevity, post mean = −2.40 ± 0.11, pMCMC = 0.02 ± 6.93 × 10^−3) produces several alternatively spliced forms of kalirin, which is essential for synaptic connections, spine development, cognition, learning, fear conditioning and social behavior^[148]46. Knockout of this gene in mice caused working memory deficits, locomotor hyperactivity and reduced social behavior^[149]47,[150]48. Gene SLC29A2 (MCMCglmm: pair-living, post mean = −1.65, pMCMC = 0.02; longevity, post mean = 2.30 ± 0.05, pMCMC = 0.04 ± 3.18 × 10^−3) is linked to the development of depression. Knockout of the ATP2A2 (MCMCglmm: pair-living, post mean = −1.60, pMCMC = 0.02; longevity, post mean = 2.42 ± 0.21, pMCMC = 0.03 ± 0.02) impaired fear memory and changed behaviors in novel environments^[151]49. Nonetheless, the contribution of these genes to longevity is currently unknown and worthy of further exploration. To gain an overall view of gene expression related to social organization and longevity, we employed a modified summary statistic approach (i.e., the polysel method, “Methods”). This approach identifies pathways that show accumulated correlation rather than outlier genes^[152]50,[153]51. The sum of the posterior means (generated from MCMCglmm models) of the genes in each pathway was calculated as the SUMSTAT score and compared to a null distribution of random gene sets. We found 56, 56, and 45 pathways showing significant correlations with solitary, pair-living, and group-living species compared with non-solitary, non-pair-living, and non-group-living species, respectively (Supplementary Fig. [154]9, Supplementary Data [155]8). We also identified 14 longevity-associated pathways that occurred in four models (Supplementary Fig. [156]10, Supplementary Data [157]9; see Supplementary Fig. [158]11 and Supplementary Data [159]10 for the results of each model). A total of 10 pathways showed accumulated correlations with both social organization and longevity (Fig. [160]2f, Supplementary Data [161]11). Among them, the hormones-related pathway “G-protein-coupled receptors (GPCRs), class B secretin-like” was positively associated with both solitary living and longevity, but negatively associated with group living (polysel: solitary, score = 6.18, P = 4.71 × 10^−2; group-living, score = −6.04, P = 3.96 × 10^−2; longevity, score = 7.15, P = 3.83 × 10^−2). The secretin-like family of GPCRs include receptors for polypeptide hormones, such as secretin, parathyroid hormone and vasoactive intestinal peptide, which play vital roles in physiological homeostasis, nervous diseases, the stress response and longevity^[162]52–[163]54. Acting as a catalyst in steroid hormone synthesis^[164]55, the cytochrome P450-related pathway has been enriched (“drug metabolism - cytochrome P450”, polysel: pair-living, score = 18.52, P = 1.40 × 10^−4; longevity, score = 16.00 ± 0.04, P = 0.02 ± 1.06 × 10^−3). The mutation of cytochrome P450 has been shown to increase longevity in Caenorhabditis elegans^[165]56,[166]57. In addition, cytochromes P450 regulate inflammation and infection^[167]58 and the generation of eicosanoids (the “eicosanoid synthesis” pathway, polysel: solitary, score = 7.94, P = 7.71 × 10^−3; group-living, score = −9.07, P = 7.77 × 10^−4; longevity, score = 8.11, P = 4.68 × 10^−2). Eicosanoids have a broad range of functions, including reproduction, physiological homeostasis, and cell growth regulation; in particular, they play a role in regulating immune response and inflammatory processes in various diseases^[168]59–[169]62. Another immunity-related pathway “immunoregulatory interactions between a lymphoid and a non-lymphoid cell” was negatively correlated with longevity (polysel: longevity, score = −24.20 ± 0.28, P = 7.78 × 10^−3 ± 1.24 × 10^−3). Interestingly, this pathway is downregulated in solitary species, but upregulated in group-living species (solitary, score = −11.19, P = 0.01; group-living, score = 11.99, P = 5.52 × 10^−3), and may be an immune response to elevated pathogen transmission among hosts and infectious disease risks in a gregarious setting. Taken together, both the function annotation of overlapping genes and the gene set enrichment analysis of all genes identified the hormones and immunity processes underlying the association between social organization and longevity. Selection features of social organization and longevity Whether social organizations or longer lifespans are under selection remains controversial^[170]63,[171]64. To characterize the selection features of social organization and longevity, we used RELAX^[172]65 to estimate the selection coefficients (K) for orthologous genes under different states of social organization (solitary, pair-living, and group-living) and longevity (long-lived vs. short-lived) (Fig. [173]3a–d, Supplementary Data [174]12, Supplementary Data [175]13). In solitary branches, genes mostly experienced intensified selection (5448 genes, K > 1, likelihood ratio test (LRT) P < 0.05) rather than relaxed selection (3200 genes, K < 1, LRT P < 0.05, Fig. [176]3a). We identified 3747 genes that showed evidence of intensified selection and 4589 genes that showed evidence of relaxed selection in the pair-living state (Fig. [177]3b). A larger number of genes associated with group-living experienced relaxed selection (5570 genes, K < 1, LRT P < 0.05) than intensified selection (3170 genes, K > 1, LRT P < 0.05, Fig. [178]3c). Longevity appeared to be under intensified selection (Fig. [179]3d), as more genes were subjected to intensified selection (5364 genes, K > 1, LRT P < 0.05) than relaxed selection (3564 genes, K < 1, LRT P < 0.05) in the long-lived state. Moreover, a larger number of genes that experienced intensified selection for longevity were found under intensified selection in solitary rather than group-living species (Pearson’s chi-squared test: χ^2 = 527.96, df = 1, P < 0.001). By contrast, a greater number of genes under relaxed selection in the long-lived state also experienced more relaxed selection in group-living species than solitary species (Pearson’s chi-squared test: χ^2 = 430.42, df = 1, P < 0.001). These results suggest that the long-lived state in group-living mammals involves relaxation selection. Fig. 3. Cross-talk between expression and selection in social organization and longevity. [180]Fig. 3 [181]Open in a new tab The pattern of selection characterizing social organization identified with a RELAX analysis (species: n = 94): a solitary species, P = 1.95 × 10^−7, b pair-living species, P = 5.89 × 10^−6, c group-living species, P = 5.81 × 10^−7, and d longevity, P = 6.76 × 10^−7. P value was calculated using Likelihood-ratio test (LRT). K < 1 indicates relaxed selection and K > 1 indicates intensified selection. Genes were under purifying selection when d[N]/d[S] < 1 and positive selection when d[N]/d[S] > 1. Arrows represent the direction of change in d[N]/d[S]. The median value of the proportion of sites is shown with a bar plot. Social organization and longevity are colored as follows: solitary: blue; pair-living: orange; group-living: red; short-lived state: cyan; and long-lived state: purple. e A heat map of significant pathways that overlapped between social organization and longevity. S: solitary; PL: pair-living; GL: group-living; ML: maximum lifespan. f A Venn diagram showing the number of genes with changes in expression levels and selection among solitary, pair-living, group-living, and longevity. Examples of genes that were associated with expression and selection in both social organization and longevity: g gene XRCC6 was under relaxed selection in group-living species (P < 1.00 × 10^−17) and h long-lived species (P = 1.06 × 10^−5). P value was calculated using Likelihood-ratio test (LRT). The interpretation of K and d[N]/d[S] is the same as (a–d). The number of species used in RELAX analyses was n = 94. i Network of pathways showing correlations with expression (red round) and selection features (purple round) in both social organization and longevity. The circle size represents the number of genes in this pathway. The thickness of connective lines displays the number of shared genes between two pathways. Source data are provided as a Source data file. We then employed the same pathway enrichment approach as described above (polysel method) using the selection coefficient (K) of each gene as the statistic. For the trait social organization, we identified 132 pathways under significant intensified or relaxed selection (solitary: 50 pathways; pair-living: 56 pathways; group-living: 53 pathways; P < 0.05, Supplementary Data [182]14; Supplementary Fig. [183]12a). We detected more intensified pathways than relaxed pathways in solitary species (34 vs. 16) and pair-living (36 vs. 20), but fewer intensified than relaxed pathways in group-living species (21 vs. 32). One pathway was shared among all three states of social organization (“Spliceosome, U2-snRNP” pathway), and a few pathways were also identified in two states (i.e., solitary-pair-living: 2; solitary-group-living: 12; pair-living-group-living: 4). For long-lived species, we also detected signatures of intensified selection in 40 pathways and relaxed selection in 23 pathways (Supplementary Fig. [184]12b, Supplementary Data [185]15). In total, 20 overlapping pathways were detected under selection for both social organization and longevity (Fig. [186]3e, Supplementary Data [187]16); however, most of these pathways did not show an identical trend in the selection force. For example, “B cell receptor signaling pathway” showed accumulated relaxed selection in group-living species, but intensified selection in long-lived species (polysel: group-living, score = −21.03, P = 3.04 × 10^−3; longevity, score = 36.61, P = 2.42 × 10^−3); “glycosphingolipid biosynthesis-lacto and neolacto series” experienced accumulated intensified selection in group-living species, but relaxed selection in long-lived species (polysel: group-living, score = 5.37, P = 0.03; longevity, score = −3.07, P = 0.04). These findings suggest that even though common pathways can be utilized by natural selection for longevity and social organization, the underlying molecular mechanisms and regulatory approaches are different. Cross-talk between gene expression and selection We further discovered eight genes with changes in their expression levels and selection for both social organization (Supplementary Fig. [188]12c–e) and longevity (Supplementary Fig. [189]12f), i.e., SHKBP1, MTM1, XRCC6, UBL7, VWA5A, PUS3, MCL1, and COX7A1) (Fig. [190]3f, Supplementary Data [191]17). In particular, gene XRCC6 was not only identified in the gene expression analyses (see above), but also experienced selection in solitary, group-living and long-lived species (RELAX: solitary, K = 2.58, P = 4.34 × 10^−13; group-living, K = 0.31, P = 1.00 × 10^−17, Fig. [192]3g; longevity, K = 0.70, P = 1.06 × 10^−5, Fig. [193]3h). MTM1 was upregulated in both solitary and long-lived species (MCMCglmm: solitary, post mean = 1.38, pMCMC = 0.03; longevity, post mean = 2.55 ± 0.21, pMCMC = 0.03 ± 0.01). This gene was also under intensified selection in solitary, but relaxed selection in long-lived species (RELAX: solitary, K = 2.52, P = 1.00 × 10^−17; longevity, K = 0.52, P = 2.05 × 10^−7). Loss-of-function of MTM1 leads to a genetic neuromuscular disorder, X-linked centronuclear myopathy^[194]66, as evidenced by a decreased lifespan in knockout mice^[195]67. Another gene, MCL1, was upregulated (MCMCglmm: pair-living, post mean = 1.31, pMCMC = 0.04) and under relaxed selection in pair-living species (RELAX: K = 0.54, P = 2.97 × 10^−5). In addition, MCL1 was negatively associated with longevity (MCMCglmm: post mean = −2.33 ± 0.13, pMCMC = 0.04 ± 8.48 × 10^−3) and experienced intensified selection in long-lived species (RELAX: K = 2.28, P = 4.55 × 10^−6). As a notable member of the anti-apoptotic Bcl-2 family, MCL1 can regulate cell cycle, cell proliferation, and DNA damage repair, which may contribute to longevity^[196]68. MCL1 is critical for neuronal development^[197]69, where the loss of MCL1 leads to apoptosis of neuronal progenitors^[198]70. One pathway was correlated with the expression and selection features for both social organization and longevity: “sulfur relay system” (polysel: expression in solitary, score = −4.17, P = 0.02, in pair-living, score = 5.37, P = 0.03 and in longevity, score = 6.81 ± 0.20, P = 0.02 ± 3.41 × 10^−3; selection in solitary, score = 3.40, P = 0.03 and in longevity, score = 4.26, P = 0.03). The sulfur relay systems are involved in the complex process of trafficking and delivery of sulfur, which is an essential element for living organisms and a component of major biomolecules^[199]71,[200]72. For example, sulfur-containing nucleosides in tRNA molecules have diverse functions, including stabilization of tRNA structure, proper codon-anticodon base pairing, and insurance of accurate and efficient translation^[201]73,[202]74. Sulfur-containing modification at tRNA position 34 has revealed a biological role of sulfur in growth, oxidative stress, and metabolic cycles in yeast^[203]73. The lack of this modification can cause myoclonic epilepsy with ragged-red fibers (MERRF), which clinically manifests as cerebellar ataxia in humans^[204]75. Besides, only one pathway whose expression was associated with both social organization and longevity was under significant selection in social organization, but not longevity: “GPCRs, class B secretin-like.” In addition, five pathways that were under selection for both social organization and longevity also showed significant changes of expression levels in social organization (i.e., “tight junction interactions,” “axon guidance”) or longevity (i.e., “SRP-dependent cotranslational protein targeting to membrane,” “mitochondrial protein import,” “GPCRs, class A rhodopsin-like”). Among them, “GPCRs, class A rhodopsin-like” (polysel: expression in longevity, score = 42.45 ± 2.02, P = 4.86 × 10^−3 ± 1.25 × 10^−3; selection in solitary, score = −37.57, P = 2.50 × 10^−6 and in longevity, score = 29.16, P = 1.27 × 10^−3) is the largest group of GPCRs, representing members such as light, hormones and neurotransmitter receptors^[205]76,[206]77. These receptors are associated with regulation of neuroendocrine function, sleep-wake cycle, energy metabolism, feeding, anxiety, and stress responses^[207]78,[208]79. Since mutations in class A GPCRs can lead to a large number of diseases, including depressive disorders, schizophrenia, and bipolar disorder, they also serve as drug targets in humans^[209]80,[210]81. Another pathway, “axon guidance” (polysel: selection in pair-living, score = 33.48, P = 6.17 × 10^−4 and in longevity, score = 22.66, P = 0.04), is key to brain development and neural circuit formation^[211]82. In addition, the “tight junction interactions” pathway (polysel: selection in solitary, score = −7.03, P = 0.02; in group-living, score = 9.27, P = 1.25 × 10^−4 and in longevity, score = 8.20, P = 0.01) regulates cell-cell communication and cellular growth, development, differentiation, and pathogen infection^[212]83. Dysregulation of tight junctions not only increases the entry and spread of viruses or bacteria^[213]84, but also affects age-related neurodegenerative disorders^[214]85. Similarly, “SRP-dependent cotranslational protein targeting to membrane” (polysel: expression in longevity, score = 38.39, P = 0.02; selection in solitary, score = −26.09, P = 2.45 × 10^−4 and in longevity, score = 37.12, P = 2.50 × 10^−6) regulates viral infections^[215]86, but its function in longevity remains unclear. Very few pathways were shared by social organization and longevity in the gene expression and selection analyses. This finding points to a fine-tuned network (e.g., Fig. [216]3i), which involves beneficial mutations and changes of expression in different, but functionally connected genes or pathways, and is a favored approach to maintain the plasticity and stable evolution of social organization and longevity. Discussion In this study, we provide evidence for the correlated evolution of social organization and longevity across the mammalian phylogeny and show that group-living species lived longer than solitary species. There was no significant difference in longevity between pair-living and group-living species, or between pair-living and solitary species, suggesting that pair-living alone is unable to mediate lifespan extension despite it can generate an association between a pair of individuals. Long lifespan favored by group-living species may be because group living reduces extrinsic mortality by limiting the risks of predation and starvation, and the strong and stable social bonds formed among group members have the power to enhance longevity^[217]8,[218]9,[219]87. These benefits are expected to override the costs inherent in group living, such as competition for mating partners and food, stress from higher-ranking individuals, and the spread of infectious diseases via social contacts^[220]88. Another explanation for the correlated evolution is that kin selection may be a driver of longevity^[221]89. Group living leads to the co-residence of males or females and sex-specific philopatry. Preferential association among kin can influence coalition formation^[222]90, cooperative breeding^[223]91, parallel dispersal^[224]92, and the establishment of social hierarchies^[225]93, which further enhance individual health or offspring survival^[226]7,[227]93,[228]94 and ultimately increase an individual’s or their relatives’ evolutionary fitness^[229]95. The length of a lifespan may be affected by inclusive fitness benefits. For example, to maximize the rate of offspring survival, the lifespan of parents or grandparents may be extended to allow for the provision of parental care or even grandparental care to offspring^[230]16,[231]96. The transcriptomic features associated with the correlated evolution of social organization and longevity indicated that hormonal regulation and immunity constitute the mechanistic foundation for the association between social organization and longevity. Peptide hormones (e.g., growth hormone, insulin-like growth factor-1, and insulin) have also been shown to perform crucial roles in aging and longevity. For example, defects in growth hormone production extend longevity in Snell dwarf mice^[232]97. Reduced insulin-like growth factor-1 signaling has been shown to increase lifespan in mice, fruit flies, yeast, and worms^[233]98,[234]99. Other types of hormone, steroids (e.g., testosterone, estradiol, and progesterone), control a range of social behaviors, including copulatory behavior, aggression, grooming behavior, and paternal behavior^[235]100,[236]101. Specifically, neuroactive steroids produced in the nervous system and their receptors also play a role in the regulation of learning capacity, memory, decision-making, and depression^[237]102. A number of studies have demonstrated that steroid hormones also regulate lifespan; for example, the inhibition of sulfatase increases lifespan in C. elegans^[238]103. The immunity or inflammation pathways and genes identified in this study support the view that immunity is instrumental to the correlated evolution of sociality and longevity^[239]104. Social organizations can affect immune responses. For example, in captive group-living long-tailed macaques (Macaca fascicularis), a higher rate of affiliation enhances an individual’s immune response^[240]105. In contrast, social isolation or a limited number of social ties can activate neuroendocrine regulation, accumulated inflammation burden, and impair immune function^[241]106,[242]107. Moreover, several lines of evidence have demonstrated the effects of immunity and inflammation on social behavior, fitness, health and lifespan of social mammals. For example, interleukin-17a (IL-17a), a well-described mediator in inflammatory diseases, can rescue sociability deficits in offspring mice exposed to maternal immune activation by directly affecting their neuronal activity^[243]108. Immunity is also linked to reproductive behavior and thus indirectly affects the fitness of an individual^[244]109. A recent study has shown that male mice avoid mating when a female mouse is unhealthy^[245]110. Research has also shown that age-related changes in immunity, such as inflammatory markers interleukin-6 (IL-6) and TNF-alpha, increase with age in humans^[246]111 whereas the proportion of naive CD4 T cells in the blood declines with age in wild Soay sheep (Ovis aries)^[247]112. We assumed that solitary species are generally less social than pair-living species, and both are less social than group-living species. However, mammal societies vary enormously in individual composition, size, patterns of parental care, cooperation, social relationships, and spatiotemporal dynamics of group members. Although some studies have provided conceptual frameworks and indices to quantify sociality or social complexity^[248]31,[249]113, a consensus and more accurate measurements that could be used in large-scale comparative studies are needed. The accumulation of long-term field data on variables such as relatedness, affiliative relationships, social network cohesion, cooperation, and agonistic relationships among individuals will solidify our understanding of the evolutionary interplay between sociality and longevity. In summary, our study provides insights into the correlated evolution of social organization and longevity and serves as a basis for experimental validation and follow-up studies on the mechanistic drivers of this correlated evolution. Methods Data collection and compilation All animal care and research protocols of this study were approved by the Institute of Zoology, Chinese Academy of Sciences (No. IOZ-IACUC-2021-129). We compiled data on various mammalian traits, including life history, social system, behavior, and habitat. These data were obtained from the literature^[250]114–[251]118, reviews^[252]25,[253]26, and databases, such as PanTHERIA^[254]119, PHYLACINE^[255]120, and AnAge^[256]4. The last search date was August 5, 2022. The sources of each data are listed in Supplementary Data [257]1, Supplementary Data [258]3 and Supplementary References. A