Dynamics of genetic and somatic trade-offs in ageing and mortality

Main

We do not yet understand the genetic, molecular, cellular and organismal processes that shape the intrinsic variability in rates of ageing, lifespan and all-cause mortality in humans, mice or other model organisms^{1,2,3,4,5,6,7,8,9,10,11}. Although thousands of variants modulate risks of age-related diseases, the great majority influence proximate causes of death and not the core mechanisms that influence ageing rates per se^12,13. To disentangle causes from consequences and to define those variants that modulate mortality across entire lifespans, we developed actuarial methods that map variation in lifespan of progressively older survivorships—equivalent to a range from 12 to 94 years in humans¹⁴. This approach has been tested^4,15,16,17, but has not been applied systematically in any organism. The closest human parallels are biometric studies of age-dependent changes in heritability of lifespan in Scandinavian twin cohorts born between 1870 and 1910¹⁸. Our study is a complementary dissection resolved at the level of discrete genetic effects—59 well-defined loci that dynamically interact to influence mortality rates and lifespan. We address four sets of questions in geroscience:

1.
What DNA variants influence mortality rates, and how and when do they act? Which have persistent effects, which have transient effects, and which act only late in life?
2.
How do these loci align with sex differences in mortality during and after reproduction? Is there evidence of genotype-by-sex (G × S) interactions, genotype-by-genotype (G × G) interactions, or even genetic antagonism between the sexes?
3.
What loci account for the strong negative association between larger body size early in life and shorter lifespans versus larger body size late in life and longer lifespans? Are these loci discrete sets?
4.
How do the dynamics and action of loci and their many interactions support or refute predictions made by major evolutionary theories of ageing—the mutation accumulation theory¹⁹, the antagonistic pleiotropy theory²⁰ and the longevity-assurance/disposable soma theory²¹?

To dissect age-localized genetic effects on mortality and lifespan we relied on the largest study of mouse ageing—the National Institute on Aging’s Interventions Testing Program (ITP)²². For more than two decades, teams at three sites have used a population of genetic siblings called the University of Michigan Heterogeneous Cohort 3 (UM-HET3) to study effects of drug interventions on longevity^22,23. These mice are effectively one large sibship segregating for 11 million DNA variants. In earlier work we mapped lifespan using 3,055 mice and detected seven lifespan loci¹⁷. Here we have doubled the sample size to 6,438 mice, quadrupled numbers of genetic markers, and introduced a more powerful actuarial mapping method. The result has been a fourfold increase in numbers of lifespan loci (Vita type) and the confirmation of six of seven loci that we had previously mapped. In addition, we have mapped 30 loci of a new type that specifically influence correlations between body mass and life expectancies in males and females (Soma type loci) as a function of age. We also comprehensively analysed age-dependent heritability changes, and the strong sex- and age-dependent epistatic interactions among all loci.

The dynamics of Vita and Soma loci are highly complex. Some act almost exclusively at younger or older ages. A subset modulates life expectancies more uniformly and includes DNA variants that may act as pacemakers of ageing²⁴. Many have marked and even complementary or antagonistic sex differences, as do almost all epistatic partnerships. Some loci have effects that are predicted by the theory of antagonistic pleiotropy^19,25,26, whereas others have effects long after the reproductive phase of life and are consonant to different degrees with both the mutation accumulation theory and the longevity-assurance or disposable soma theory of ageing^2,27. It should now be practical to convert some of these loci into well-defined molecular and cellular mechanisms that modulate age-dependent rates of ageing and risks of mortality with potential translational relevance to human health and longevity.

The results are divided into six sections. The first section summarizes the actuarial mapping method and differences in survivorships by sex and genotype (Fig. 1 and Tables 1 and 2). The second partitions Vita loci by their dynamics and defines age-dependent shifts in heritable and non-heritable factors (Fig. 2). The third covers sex differences and antagonism (Fig. 3). The fourth introduces correlated trait mapping (Fig. 4) used to map 30 Soma loci (Tables 3 and 4) that modulate body mass–life expectancy trade-offs across age. The fifth is a systematic analysis of epistatic interactions among all loci (Fig. 5) and the discovery of a sharp split by sex. The last provides case studies of how genetic, bioinformatic and experimental data can be combined to move from maps towards genes and mechanisms of mortality and ageing (Fig. 6).

Fig. 1: Survivorship sizes, sex differences and Vita lifespan loci. — **Fig. 1: Survivorship sizes, sex differences and *Vita* lifespan loci.**

Fig. 2: Dynamics of Vita loci, their contributions to heritability, and source of variance. — **Fig. 2: Dynamics of *Vita* loci, their contributions to heritability, and source of variance.**

Fig. 3: Antagonistic sex effects within Vita loci. — **Fig. 3: Antagonistic sex effects within *Vita* loci.**

**Fig. 4: The genetic modulation of life expectancy by body mass is stronger in males than females.**

Fig. 5: Epistasis among Vita and Soma loci. — **Fig. 5: Epistasis among *Vita* and *Soma* loci.**

Fig. 6: Genetics of Vita9b and candidate gene analysis. — **Fig. 6: Genetics of *Vita9b* and candidate gene analysis.**

Table 1 Overview of Vita loci with effect sizes

Full size table

Table 2 Overview of Vita loci with sex and epistatic effects and mortality features

Full size table

Table 3 Summary of Soma loci linkage statistics, age ranges and positions

Full size table

Table 4 Summary of Soma loci effects and overlap with Mass and Vita loci

Full size table

29 Vita loci are defined by actuarial mapping

We analysed lifespans of 72 nested survivorships (Fig. 1a), each of which we generated by truncating from younger to older cut-offs in 15-day steps. The base survivorship includes mice that entered the study and survived to at least the first truncation age (T-age) of 42 days (Fig. 1a, T₄₂). The terminal survivorship includes only the oldest 559 mice (8.7%)—those that reached at least 1,100 days of age (T_1,100). At the T₄₂ T-age the sex difference of lifespan is 81 days (Fig. 1b): 806 ± 210 days (mean ± s.d.) for males and 887 ± 175 days for females. This difference is stable to T₂₁₅ because so few animals of either sex die before this age; only 20 males and 13 females. However, between 215 and 410 days 204 males die but only 18 females die, and by T₇₄₀, expectancies of males and females have converged at 946 days (Fig. 1b).

We mapped loci that modulate mean lifespans of survivorship at a false discovery rate (FDR) of P < 0.05 after applying a Bonferroni correction with an additional Cauchy correction for the actuarial analysis (Fig. 1c and Table 1). Mapping was stratified by sexes and in combination (Fig. 1d,e, Table 2, Extended Data Fig. 1 and Supplementary Tables 2–9 for trait and mapping data files). Loci have average effects of 36 ± 12 days (mean ± s.e.m.) on life expectancies that often depend strongly on sex and T-age (Fig. 1c–f, Table 1 and Extended Data Figs. 1–3). Peak effects explain 2.5 ± 0.9% of variance in females and 3.2 ± 0.5% in males. Linkage scores range from a low of 4.1 logarithm of the odds (LOD) (Vita5a) to a high of 9.0 LOD in males (Vita4a) and to a high of 8.1 LOD in females (Vita2c). Confidence intervals are 35 ± 19 Mb (mean ± s.d.) with the smallest two loci under 10 Mb (Fig. 1d-f and Table 1). All genes in loci were reviewed for variants that potentially affect function (Supplementary Table 1a,b). Vita loci have age-delimited effects that average 349 ± 229 days (mean ± s.d.) (Table 1, average of the ‘Duration’ column).

Dynamics of Vita loci

Effects of Vita loci are age-dependent (Fig. 1, Tables 1 and 2, Extended Data Figs. 1–4 and Supplementary Tables 2 and 3). Vita1a and Vita1b have stable actuarial effects from T₄₂ to T₈₉₀ (Figs. 1f and 2a) but strong late-age mortality effects. By contrast, Vita14b has a gradual reduction in effects (Fig. 2b) highlighted by the converging slopes of BD and CH genotypes from T₂₃₀ to T_1,100_. Other loci have transient effects. For example, Vita1c acts only in survivorships that include adults younger than 500 days (Fig. 2c), whereas Vita9c acts only in the oldest survivorships (Fig. 2d). Vita4a has marked effects that reverse between T₄₁₀ and T₈₀₀ in males (Fig. 2e,f). Inflection points in actuarial plots correspond to ages on Kaplan–Meier plots (Fig. 1g) at which mortality rates of genotype classes diverge or converge (Extended Data Figs. 2–5). In Figs. 1h and 2f there are, for example, strong age-dependent differences in mortality rates between maternal and paternal haplotypes by sex and age. The D haplotype contributes to higher mortality earlier in life; the H haplotype to higher mortality later in life. This explains the actuarial benefit even in the earliest survivorships of inheriting an H haplotype (Fig. 1f and Extended Data Fig. 4 for all other Vita loci). The impact of DNA variants within Vita loci on mortality is greatest at inflections in actuarial plots—at the peaks and troughs of mortality rate plots (Fig. 2a–f and Extended Data Figs. 2–4). These are the T-age ranges during which the mean lifespan of survivorships can swing by up to 15 days over just 45 days (Fig. 2e). By contrast, Vita loci that have nearly constant slopes from T₄₂ to T₇₅₀ (Fig. 2b) behave as expected of putative modulators of rates of ageing and of differences in all-cause mortality^5,24.

We have categorized the dynamics of Vita loci into four categories (Table 2, far right) using the plots reproduced in Extended Data Figs. 2–4.

1.
Loci with durable effects. Five loci have relatively constant effects on life expectancies of survivorships from as early as T₄₂ out to the T₇₅₀ survivorship in one or both sexes (Fig. 1f and Tables 1 and 2). Their actuarial durability makes them predictors of life expectancy even in adolescence, but this does not mean that they have age-independent effects on mortality. In fact, peak effects at a locus such as Vita1a are concentrated late in life (Fig. 1h,i).
2.
Loci with steadily diminishing effects have uniform actuarial slopes from as early as 215 days that extend to the T₇₅₀ survivorship. They are potential rate of ageing modulators (RAMs) using a liberal definition of ‘uniform slopes’. Seven have an almost age-independent hazard ratio (HR) over at least 500 days. Their effects converge towards minimal expectancy differences. Vita14b is an example in which the initially high difference in males between CH and BD genotypes fades from T₃₀₅ to T_1,100 (Fig. 2b). The effect is due to high mortality of B and D carriers up to 800 days followed by high mortality of C and H carriers thereafter (Extended Data Fig. 5).
3.
Loci with age-range restricted effects. This tripartite category includes loci with action limited to early, middle or late survivorships (Fig. 2c–f).
1. a.
  Early-age loci have effects before T₅₀₀. Loci of this type are three times more common in males than in females (Table 2). One of the best examples is Vita1c, for which the 50-day difference in life expectancies of carriers of BH and BD genotypes is restricted to males (Fig. 2c and Extended Data Figs. 2c and 3c1,c2). Exceptions are notable because polarities of genotype effects are flipped between sexes at four of these types of loci—Vita2b, Vita9a, Vita11b and Vita18a (Fig. 3c–f and Extended Data Fig. 2n,n1,s,s1,ab,ab1). The male bias in this type of locus is almost always linked to higher mortality rates in the T₂₀₀ to T₇₀₀ survivorships (Fig. 1a,b, Table 2 and Extended Data Fig. 4).
2. b.
  Mid-survivorship loci effects from T₅₀₀ to T₈₄₅. There are six loci of this type. Vita4a is an impressive example, with a biphasic fluctuation in mortality rates of BD and CH genotypes between T₅₀₀ and T₇₁₀ in males (Fig. 2e and Extended Data Figs. 2i and 3i1,3i2). This effect is caused by offset waves of mortality. A D haplotype wave of mortality starts at 400 days (vertical line in Fig. 2f) but is followed by C and H haplotype waves at 600 days. By 1,000 days both B and D haplotypes have high relative HR values approaching 2 (log₂(HR) ≈ 1). The offsets in peaks of effects in actuarial plots (Fig. 2e) versus mortality plots (Fig. 2f) is because survivorships integrate mortality out to the final death. These waves are more obvious in Fig. 2f, and for all loci in Extended Data Fig. 4.
3. c.
  Late-acting Vita loci effects after T₈₆₀. Five loci only act after T₈₆₀ (Fig. 2d and Tables 1 and 2). As mice die the ratios of genotypes do shift but this does not affect mapping until about T_1,100. Thereafter, genotype imbalance inhibited us from further mapping (Supplementary Table 2).
4.
Loci with reversals of genetic effects. Twelve loci have effects that reverse across survivorships (Table 2). Reversals can involve a single genotype (the red BD trace in Fig. 2a) or two or more, for example BD and CH effects of Vita4a (Fig. 2e,f). These patterns are associated with age-dependent and highly variable HRs.

Dynamics of heritability

The summed genetic variance explained by all 29 Vita loci on lifespan (V_g) is higher in males than in females from T₄₂ to T₃₂₀ (Fig. 2g,h and Table 1) but drops from a base at 40% to 30% by T₅₀₀ (Fig. 2g and Extended Data Fig. 5). This decline is due to the wave of early male mortality. By T₆₅₀ male Vita heritability rebounds and climbs to 50% in the oldest survivorships. Female Vita heritability is almost precisely 27% in all survivorships to T₆₂₀ but decreases to 20% by T₈₉₀, 450 days after the minimum for males. As in males, female heritability recovers in the oldest survivorships, although by 10% less than in males. Variance explained by two-locus epistatic interactions also differs markedly by sex (V_{g × g} in Fig. 2g,h). This value is twice as high in females than in males, 12% versus 6%, and this is true over the entire reproductive lifespan. Note that the V_g (Fig. 2) is not a conventional narrow-sense estimate of heritability, but rather is variance explained only by the Vita loci and also will include dominance effects (Extended Data Fig. 6). Our results are consonant with the strong age-dependent increase in heritability noted in humans¹⁸ and higher sex-averaged heritability estimates when adjusting for extrinsic causes of death²⁸.

We estimated the effect of three experimental sources of variance (V_exp): (1) treatment with or without a nominally ineffective drug; (2) cohort year; and (3) three sites. Collectively, V_exp differs by sex and age (Fig. 2g,h). The summed level of V_exp peaks at 47% in males in the T₄₄₀ survivorship but drops in rough synchrony with the end of reproductive life^29,30 (Fig. 2g). Less than 20% of variance in older survivorships is linked to these non-genetic experimental sources—a counter-intuitive reduction in sensitivity. Despite efforts by ITP teams to standardize protocols, variance attributable to site is up to 43% in males and 24% in females (Fig. 2i,j). In males, variance associated with an ineffective drug is much more modest (Fig. 2i), although in absolute terms these supplements do increase lifespan at T₄₂ by 40.2 ± 8.3 (mean ± s.e.m.) days. By contrast, in females, the effect of drugs is greater than that of site (Fig. 2j) and increases lifespan by 37 ± 6.7 (mean ± s.e.m.) days. This positive effect persists to T₆₈₀ in males and to the T₈₀₀ in females (Fig. 2i). However, it is negative by T₈₉₀ in males (–11.7 ± 5.3 (mean ± s.e.m.) days, P = 0.03) and by T_1,040 in females (–10.2 ± 5.3 (s.e.m.) days, P = 0.06). Other sources of variance (V_o) (orange lines in Fig. 2g,h) include stochastic causes of mortality, for example uncontrolled environmental and technical factors, and higher-order genetic and G × E effects that we have not modelled, including age-dependent changes in survivors per cage and unmeasured indirect social genetic effects³¹.

Vita loci have antagonistic sex interactions

The profound sex difference in life expectancies in early survivorships (Fig. 1b) can be decomposed into sets of loci that specifically influence G × S interactions and early male mortality. The likely cause is higher stress and aggression among males^32,33. High male mortality is visible in plots for Vita1c, Vita2a and Vita6b (Fig. 2c and Extended Data Figs. 2–4). These and other loci have strong but transient effects earlier in the lives of males. This youthful mortality risk is eliminated in survivorships above T₇₂₅ (Fig. 1b). From T₉₃₅ to T_1,100, males are typically housed solo and gain a small 8-day expectancy advantage over females (T_1,040, P = 0.036).

Fourteen Vita loci have strong G × S interactions (Fig. 3a and Table 2). The complex of loci on chromosome 2 is an extreme example (Fig. 3a–d). CH and BH genotypes at Vita2b confer a lifespan advantage to females up to T₇₀₀ (Fig. 3e), but a disadvantage to males through to T₃₆₅. Polarities of effects are also reversed, with CH and BH having longer lifespans in females but shorter lifespans in males (Fig. 3c,d). This is due to a sex inversion in timing of sequential waves of mortality of carriers of the paternal haplotypes (Fig. 3f). Vita2b also has a massive G × S effect with a LOD of 8.6 at T₁₄₀ that is entirely lost by T₈₆₀ (Fig. 3a). The reversal of male genotype effects at Vita2b fits models of antagonistic pleiotropy (Fig. 3d)—genotypes that reduce mortality in the first half of life increase mortality after 600 days. The Vita2 complex illustrates how mapping sexes together without fitting an interaction term is ill-advised. Although the sex-combined analysis defines highly significant peaks at T₈₀₀ for Vita2b and Vita2c (Fig. 3b), this is an artefact of not modelling the interaction (Extended Data Fig. 1e,f). Similarly, G × S mapping unmasks two sex interactions at the extreme distal end of chromosome 4 (154 Mb) at Vita4b, a locus that is nominally only detected in males (Fig. 3a).

The male-specific VitaXa locus encompasses the entire proximal 70 Mb of chromosome X (Fig. 3g and Tables 1 and 2) with effects that erode almost linearly with survivorship T-age. This locus is likely to integrate net effects of a set of recessive variants on this hemizygous chromosome in males. In females, the effects of C and B haplotypes do not differ (Extended Data Figs. 1ac and 2ac).

Soma loci balance body mass versus lifespan

Mice were weighed at 42, 183, 365, 548 and 730 days. Body masses at the first four ages correlate negatively with subsequent lifespans in both sexes (Fig. 4d–g) up until the oldest survivorships^34,35. However, correlations differ greatly by sex—rank order rho values of –0.28 for males but only –0.11 for females at T₁₈₅ (P < 0.001) (ref. ³⁶). This translates to a loss of 14.3 days g⁻¹ in males versus 3.7 days g⁻¹ in females at the peak of reproductive performance (Fig. 4e). The sex difference remains significant to T₈₀₀ (bold black linkage score trace in Fig. 4e). Female correlations are stable and remain close to –0.1 (Fig. 4e–h). By contrast, the negative correlation in males erodes with age and overlaps female values in older survivorships (T₈₉₀; Fig. 4g,h) and then shifts to positive values in both sexes for body mass at 730 days (Fig. 4h). We investigated what loci account for these dynamic shifts in relations between body mass and mortality. Our first step was to map body mass loci at all five ages and evaluate their roles in mortality. We detected 28 Mass loci (Fig. 4a and Extended Data Fig. 7), but these are almost entirely independent of the Soma loci (Fig. 4a,b, note faint vertical lines) that modulate trade-offs between body mass and life expectancies.

Using correlated trait mapping³⁷ we defined 30 Soma loci that modulate the negative and positive trade-offs between mass and mortality risk (Fig. 5b,c, Tables 3 and 4, Extended Data Fig. 8, Supplementary Tables 10 and 11). Fifteen Soma loci are detected only in males and four are detected only in females (Tables 3 and 4). Nineteen modulate the strength of negative correlations with body mass at 42 and 185 days. Eleven modulate the strength of positive correlations in the post-reproductive phase of life^30,38 (Tables 3 and 4 and Extended Data Fig. 8). The effects of Soma loci on expectancies range from 2 to 29 days g⁻¹. Table 4 presents these effects for each genotype both as day per gram and as differences in rank correlations. Soma3b has strong effects in males (Fig. 5i) but none in females (Fig. 5j). Males with the CD genotype lose 17.9 days g⁻¹; those with the CH lose only 8.4 days g⁻¹. In females, the differences in these negative correlations are insignificant (CH: –2.9 days g⁻¹ versus CD: –3.4 days g⁻¹). Soma11a has the strongest effect in females (Tables 3 and 4 and Extended Data Fig. 8), but effects are modest compared to those in males.

Soma loci overlap Vita and Mass loci (Fig. 4a–c and Table 4) at a chance level. The summed genome-wide coverage of Soma loci is 1,349 Mb (48% of the genome), and coverage of the Vita loci is 1,000 Mb (36% of the genome). Seven Soma loci have peaks within 10 Mb of Vita peaks (P = 0.19). Soma1a and Vita1a share top markers in both sexes, and both are modulated by the H and D haplotypes (Fig. 1f and Extended Data Figs. 3a and 8a). H lengthens life by 12 days, whereas D shortens life by 22 days. Surprisingly, effects of Soma1a oppose those of Vita1a, particularly in females. H decreases life expectancy by –4 days g⁻¹ (average ρ = –0.12), whereas D increases expectancy by 1 days g⁻¹ (rho = +0.02). All effect plots, for example those in Fig. 4m–p, are provided in Extended Data Fig. 8 for all loci, both sexes and T-ages.

Epistasis is segregated by sex

We tested for epistatic interactions among the 59 Vita and Soma loci at four ages by sex with cut-off thresholds close to a LOD of 4.0, a value significant at a Benjamini and Hochberg³⁹ FDR of 0.01. There are 41 significant Vita–Vita interactions among 387 Vita pairs that we tested in the base T₄₂ survivorship using an even more stringent Bonferroni correction at P < 0.05 (Fig. 5a,b and Table 2)—22 in males and 19 in females. Across all four survivorships the number of Vita–Vita interactions is 59 in males and 43 in females (Supplementary Table 12.4). Similarly, there are 57 Soma–Soma interactions in males and 35 in females (Fig. 5a,b and Table 2). Average LODs in the four survivorships and two sexes range from 4.4 to 4.6 ± 0.5 to 0.8 (mean ± s.d.). One of the stronger male interactions is between Vita1c and Vita3a (Fig. 5d, LOD 5.0) in which effects of BD and CH genotypes at Vita3a produce differences of 100 days in lifespan across the Vita1c genotypes, but there is no corresponding interaction in females (Fig. 5d). However, Vita3a has a strong interaction with Vita5a in females as late as T₇₄₀ but there is no corresponding interaction in males (Fig. 5f). The strongest female Vita–Vita interactions are Vita1a–Vita6b, Vita3a–Vita5a and Vita1b–Vita9a (Fig. 5a,b). The BD genotype at Vita1b masks any effects of Vita9a—the classic Mendelian definition of epistasis⁴⁰ (Fig. 5e). Finally, there are 197 Vita–Soma interactions—84 in females and 113 in males—a bias that is expected from the greater numbers of male Vita loci (20 versus 8 in females) and male Soma loci (18 versus 8 in females). Vita–Soma loci have LOD scores of 4.5 ± 0.6 (mean ± s.d.).

Epistatic interactions of Vita and Soma loci form segregated male and female networks (Fig. 5b,g,h and Extended Data Fig. 9). Even those few loci with minimal sex differences in their main effects do not share partnerships (Fig. 5a). For example, in females Vita1a pairs with Vita6b, Vita9a and Vita11a with LODs of 5.5, 4.6 and 4.7, respectively, but in males all of the corresponding LODs are marginal: 1.5, 3.2 and 3.5, respectively. No interactions are shared by Vita loci in the T₄₂ survivorship (Fig. 5b), and only 2 out of 61 possible interactions of all 3 types are shared at the most lenient threshold in any survivorship (Extended Data Fig. 9 and Supplementary Table 12). The same pattern is true of Soma–Soma and Vita–Soma interactions. Polarities of interactions can even display sexual antagonism^41,42,43,44; for example, the complementary effects of the Vita2b CD genotype (blue line) across Vita1c genotype columns (Fig. 5c). Although Vita1c does not have any main effect in females (Figs. 1e and 3a and Table 1), a test of epistasis exposes a strong interaction at this locus with Vita2b in females (Table 2).

Epistatic partnerships are stronger (Fig. 2g,h) and more stable in females than males as a function of age (Fig. 2g,h and Supplementary Table 12.3). For example, 21% and 17% of partnerships at T₄₂ are matched at T₇₄₀ and T₉₀₅, respectively, in females, but only 9% and 0% are matched in males. Essentially all of these findings are consistent with a strongly sex-dichotomized genetic architecture—a form of genetic diplomacy that we presume harmonizes male and female phenotypes with divergent life history styles but necessarily convergent reproductive goals^45,46.

From maps towards mechanisms

As a first step, we assembled data on variants within each locus that are predicted to have an effect on protein function or with potential roles in ageing⁴⁷ (Tables 1–4 and Supplementary Tables 1 and 13). We also conducted global enrichment analyses of loci with respect to Gene Ontology categories, KEGG (Kyoto Encyclopedia of Genes and Genomes) and Reactome networks (Supplementary Table 16). Although this process can help rank plausible candidate genes, there are hard limits. The two largest loci almost certainly encompass many variants. Vita17a extends from the centromere to 74 Mb (Table 1) and overlaps the major histocompatibility complex, which has been known since the 1970s to be linked to variation in lifespan⁴⁸. VitaXa covers the proximal 70 Mb of chromosome X and is therefore an intractable oligogenic male-specific locus.

By contrast, Vita1a and Vita9b are already tractable. Vita1a is compact, overlapping only 14 protein-coding genes, of which Mrpl15 (4.8 Mb), Atp6v1h (5.1 Mb) and Rb1cc1 (6.2 Mb) are strong candidates involved in mitochondrial function, autophagy and nutrient sensing, respectively. ATP6V1H is notable, given its ubiquitous activity as a modulator of lysosomal acidification⁴⁹ and mTOR function. Variants in this gene extend lifespan in Drosophila⁵⁰ and Caenorhabditis elegans⁵¹ and modulate insulin secretion and type-2 diabetes risk in humans⁵². The unique data that we generated on epistatic interactions is mechanistically relevant because we now know that Vita1a interacts non-linearly with Vita6b, Vita9a and Vita11a in females but not in males. These are all strong constraint when ranking candidate genes for testing at all three loci (Fig. 6g,h).

Vita9b is mechanistically intriguing because it modulates mortality in both sexes in older survivorships, and also because it is a replicated locus that was previously mapped in a different cohort⁵³ (Fig. 6a, inset). The D and H haplotypes at Vita9b reverse polarities of effects in males between T₅₀₀ and T₇₀₀, and there are strong mortality rate inflections in late survivorships (Fig. 6b and Extended Data Fig. 2o2). By contrast, C and B haplotypes have uniform effects that peak at T₇₈₅ (Fig. 6c). In females, the effects are modest and are delayed to T₈₀₀ (Extended Data Fig. 2o2). These age and sex differences are additional constraints in candidate ranking as shown below for alignment to human longevity genome-wide association studies (GWAS) data.

Vita9b spans 127 protein-coding genes with missense variants that are predicted to have moderate or high impact. We tested 11 C. elegans orthologues within Vita9b (Acad11, Hyal1–3, Col6a6, Lrrc2, Ip6k2, Trim71, Poc1a, Stt3b, Pcbp4, Acp3 and Apeh) and 3 candidates just distal to the locus (Slc22a13, Zfp445 and Csrnp1) using RNA interference (RNAi) knockdown (Fig. 6e–h). All but one of these genes have missense variants in UM-HET3 mice (Supplementary Table 14). Knockdowns of 4 candidate genes altered motility, a validated predictor of lifespan^54,55, in aged worms older than 14 days (Fig. 6f,h). For example, suppression of acds-10, an orthologue of the mouse Acad11 gene, boosted motility from 18 to 27 days with an effect comparable to the daf-2 (Igf1r in mouse) positive control⁵⁶ (Fig. 6f–h). In mice, Acad11 is trans-activated by p53, enhances fatty acid β-oxidation, and increases expression with age^57,58,59. This gene is a particularly high-rank candidate. Three other knockdowns reduced C. elegans motility after ten days (Fig. 6f–h): pho-6 (Acp3 in mouse), pes-4 (Pcbp4 in mouse) and dpf-5 (Apeh in mouse); APEH is also well supported by human GWAS data on longevity.

To bridge the translational gap between mouse and human ageing^60,61, we tested association between human orthologues of genes in Vita1a and Vita9b by Mendelian randomization^62,63. In the case of Vita1a, we did not detect significant effects of any of the candidates we were able to test—LYPLA1, MRPL15, PCMTD1 or TCEA1. In the case of Vita9b, we were able to test 7 out of 11 genes that we tested in C. elegans (Supplementary Table 14). APEH notably stood out. Variants in this gene are associated with parental longevity in humans (Benjamini–Hochberg –logP of 3.7 with paternal age at death) and with individual survival into the top 10% age group (Benjamini–Hochberg –logP of 1.37; Supplementary Table 15). Increased APEH expression in human blood is positively associated with male longevity, in line with the mouse genetics and our findings in C. elegans.

Discussion

Most previous work on the genetics of ageing has defaulted to a single metric—the duration of lifespan. The actuarial method that we use tracks age-dependent mortality from puberty to senescence as a function of genotype and sex. The 59 Vita and Soma loci shape age-dependent mortality rates. Loci such as Vita1c and Vita18a act only early in life and are much more common in males than in females. They are likely to be linked to male competition, hormone status and stress resilience^36,64. Other loci act almost exclusively from 500 to 700 days of age or affect only the oldest individuals. These late-acting loci are the prime target of longevity genetics using human centenarian cohorts^65,66,67. A final subset has persistent effects on mortality across survivorships and may influence rates of ageing. The Vita and Soma loci form hundreds of epistatic interactions that are strictly segregated by sex, and ignoring sex or mapping without modelling gene-by-sex interactions is problematic. Uncovering these networks of age-dependent loci and their interactions provides an empirical genetic context for molecular, cellular, organismal and life history studies of senescence in mice and humans^2,4,6,26. We are able to place loci into age-dependent mortality classes and to address their fit with three evolutionary explanations of ageing^19,20,21,27.

The first is that senescence is caused by late-acting variants that escape selection in natural populations owing to high extrinsic causes of mortality—the mutation accumulation theory¹⁹. This idea predicts post-reproductive effects of common genetic variants that are both good and bad with respect to longevity, but more neutral during the reproductive phase of life. Effects of this type should only emerge at the onset of senescence; gradually, like Vita5a in males, or abruptly, as thresholds of resilience are breached later in life. Examples include Vita1a and Vita11b in females and Vita5a and Vita10a in males. The second more physiological explanation is that senescence is a consequence of a compromise between reproductive investment and somatic self-repair^21,27. All early-acting Soma loci fit this disposable soma argument—they are genetic embodiments of a bioenergetic compromise between body mass at the height of reproduction and subsequent duration of life. The third explanation is that ageing represents a Faustian bargain that boosts fitness early in life, but calls for payment late in life—the antagonistic pleiotropy theory^2,20. Loci such as Vita2a, Vita2b, Vita2c and Vita9b (Fig. 6b) are vivid examples that are consistent with this idea, in that genotype effects of loci on mortality invert with age. A simple prediction is that alleles that improve fitness before 400 to 600 days are linked to higher reproductive success but also to accelerated mortality. The dynamics of Vita and Soma loci provide a bridge between evolutionary explanations of senescence and genetic and molecular processes that account for progressive functional decline^68,69.

We have defined 13 loci with strongest effects in males, 4 loci with strongest effects in females, and 12 loci with strongest effects when data from both sexes are pooled. This male bias is as pronounced among Soma loci as it is among Vita loci. Sex differences in the genetics of lifespan are amplified by their epistatic interactions, mirroring findings in Drosophila⁷⁰. The divergence is so large that we are tempted to call this pattern sexual antagonism, although sexual diplomacy or compromise may be better terms. This genetic imprint is highest during the reproductive crescendo³³, across the same age ranges that contribute to strong trade-offs between body mass and life expectancy that are accentuated by many male Soma loci³⁶. After reproduction, natural selection is necessarily mute, and male and female life expectancies converge.

Our analysis of sources of variability that contribute to age-dependent mortality differences between the sexes is incomplete. The higher genetic variance attributable to Vita loci in males than in females was unexpected, and is reflected in greater success in detecting male Vita and Soma loci. By contrast, the higher variance due to experimental factors such as site and dietary additives does not yet have any pathological explanation (no wounding or sudden weight loss). Male–male interactions are a likely but unproven culprit. The loss of cage mates in older survivorships may explain the marked reduction in V_exp in males after T₅₆₀ (Fig. 2g) and in females after T₈₀₀. The relentless rise of other sources of variability (V_o) probably reflects the increasing importance of hundreds of age-dependent frailty-associated loci that we have not modelled, as well as stochastic age-related diseases. This variance source must be the driver of the Gompertzian climb of mortality rates with age. The genetics of body mass coupling with lifespan is much stronger in males than in females, particularly in the younger survivorships. Interventions to blunt this coupling would therefore tend to be more effective if applied proactively early in life. By contrast, interventions that target geriatric cohorts—mice older than 800 days or humans older than 75 years—will probably have to contend with diverse drivers of escalating age-dependent diseases.

The ITP has tested pro-longevity effects of 62 drugs, about 25% of which have had positive effects. Less noted is a strong bias in favour of positive effects in males (updated to the 2021 cohort)⁷¹. Rapamycin is one welcome exception, with beneficial pro-longevity effects in both sexes even late in life. The loci that we have defined serve as anchor points to test compounds that modulate mortality and to determine whether they interact with known loci or define drug-specific loci as a function of sex and age. Our results can guide interventions to extend health and longevity^72,73,74.

Methods

The UM-HET3 sibship

UM-HET3 mice are progeny of a cross between two types of F₁ hybrids—female F₁ mice from matings of BALB/cByJ dams to C57BL/6J sires and male F₁ mice from matings of C3H/HeJ dams to DBA/2J sires. These four inbred progenitors are abbreviated CBy, B6, C3H and D2 when referring to mice and strains, and abbreviated C, B, H and D when referring to genotypes and haplotypes. These four fully inbred strains were selected to maximize phenotypic diversity. Young virgin F₁ males and females bred at the Jackson Laboratory (JL) JAX facility were transferred to ITP ageing colonies at JL in Bar Harbor Maine, the University of Michigan in Ann Arbor Michigan (UM) and the University of Texas Health Science Center in San Antonio (UT). Breeding cages were set up in spring. First litters were not used. All subsequent litters were used at UT, litters of 6 or more pups were used at JL, and those with 7 or more pups were used at UM. Weanlings were entered into the study over the next 7–8 months.

This study is based on tail samples acquired initially at the three ITP sites in accordance with standards of the Association for the Assessment and Accreditation of Laboratory Animal Care and recommendations of the National Institutes of Health Guide for the Care and Use of Laboratory Animals, including annual reviews and approvals of all protocols. Links to data for ITP papers and data are available at the Mouse Phenome Database⁷⁵.

All mice in the first four cohort years were born between April 2004 and January 2008. In these cohort years we have numerically well-balanced DNA samples from all sites. Almost no mice were generated in 2008 owing to a funding gap. All mice in the final cohort years used in this study were born between July 2009 and March 2013. The 2009 cohort includes mice from all sites, but 2010 and 2011 include mice only from UM and UT, and cohort years 2012 and 2013 include mice only from UM. The last UM-HET3 mouse in this study died in 21 December 2015. In all years, only a small percentage of mice (<10%) were born between January and April. Mice were weighed at 42 ± 2 days, and at 183 days (6 months), 365 days (12 months), 548 days (18 months) and at 730 days (24 months) with a timing error of ~7 days.

The UM-HET3 sibship segregates for ~10.6 million sequence variants (Supplementary Table 1). All UM-HET3 mice inherit C and B haplotypes from their F₁ mothers and H and D haplotypes from their fathers. As a result, the entire sibship segregates for four genotypes, the four two-way combinations of maternal and paternal haplotypes, on all autosomes—CH, CD, BH and BD. Females inherit one entire non-recombinant H-type chromosome X from their paternal grandmothers and a potentially recombined chromosome X from their mothers (recombinations between C and B haplotypes only; Fig. 3g). As a result, females have either CH or BH chromosome X genotypes. Hemizygous males inherit a potentially recombined C or B chromosome X from their mothers. All mitochondria and their genomes in all animals are derived from maternal grandmothers (C) and male chromosome Y is derived from paternal grandfathers (D).

We genotyped 6,872 UM-HET3 mice for which we had full lifespan estimates (n = 3,252 females, n = 3,620 males). Of these genetic siblings 6,438 passed all genotype quality control steps at 891 markers (n = 3401 females, n = 3037 males; see Fig. 1c and ‘Genotype quality control’). To ensure balanced numbers by sex later in life, every ITP cohort initially consists of 51 male and 44 female weanlings per year, per site and treatment category, but with twofold more common untreated controls at each site. This was done to enable overly aggressive males and any wounded mice to be removed while balancing male and female numbers later in life—almost always before 550 days. The earliest minimum inclusion age in this study is T₄₂, the pubescent age at which tails were docked for tissue acquisition, an age equivalent to about 12 years in humans. For numerical convenience we set the first T-age at T₃₅. To compensate for twofold higher male mortality in the first 2 years of life (33% versus 16%), we included 12% more males than females in the T₄₂ survivorship. The earliest death was at 46 days. This left us with a small surplus of 227 males at T₃₆₅, but by T₅₆₀, numerical balance was restored and there were 2,930 females and 2,929 males. This age corresponds to roughly 56 years of age in humans. All individual-level data, metadata, survivorships and mortalities per 15-day interval are provided in Supplementary Table 2.

We genotyped two major categories of mice—those not treated with any dietary intervention and mice treated with a dietary supplement that did not modify lifespan significantly on a per-drug basis using standard statistical criteria^76,77,78. These latter mice have been referred to as ‘no drug effect’ (NDE) cases. However, when data are combined across the entire class of these individually ‘ineffective’ agents from 2004 to 2013, there is a highly significant combined effect that is positive on lifespan (Fig. 2i,j).

Husbandry

Mice were weaned into same-sex cages—three males or four females per cage—at 20 ± 1 days (ref. ²²). They lived together from weaning to death without any replacements within cages. As a result, most mice lived alone after ~1,000 days, a factor to consider with respect to mortality risks, but not yet integrated into our analysis. From 2004 until 2013, all sites used NIH-31 standard diets. For breeding cages, UM used Purina 5008, UT used Teklad 7912, and JL used Purina 5K52. For mice up to ~122 days, UM used Purina 5001, UT used Teklad 7912, and JL used Purina 5LG6. After 2004, a single control diet—LabDiet 5LG6—was used by all sites. Mice were monitored daily for signs of ill health and aggression and euthanized if moribund. Tails were obtained at 42 ± 2 days for DNA extraction. Body mass at this age was acquired for 2,459 mice, and at half-year intervals for 4,688 mice at 183 days (6 months), and down to 2,208 mice at 730 days (24 months). Our analysis includes four experimental variables—sex, site, dietary drug treatment and cohort year (Supplementary Table 3). The class of nominally ineffective treatments included in this study are listed on Mouse Phenome Database ITP Portal under the acronyms 4OHPBN, CAPEhi, CAPElo, Cur, Enal, FOhi, FOlo, GTE, HBX, I767d, MB, MCTO, MET, NFP, OAA, Res07, Reshi3, Reslo3, Simhi, Simlo and UA.

DNA extraction, sequencing and sequence alignment

Tail tissue was processed using the MagMAX magnetic-bead extraction system and genomic DNA was quantified using a Qubit 4 fluorometer and a high-sensitivity DNA assay kit (Thermo Fisher Scientific). Locus-specific PCR primers were designed using BatchPrimer3⁷⁹ to generate 180-bp amplicons. They were mixed using an optimized Hi-Plex approach at Floodlight Genomics. Sample-specific 12-bp barcodes were added and amplicons were pooled⁸⁰. Primer sequences for markers with reference SNP rs identifiers are provided in Supplementary Table 4. The amplicons were sequenced to an average of ×1,000 per targeted DNA variant on an Illumina NovaSEQ (2× 150 bp reads). Data were demultiplexed and FASTQ files used for genotyping. FASTQ files were aligned to M. musculus GRCm38.p6/mm10 reference using Bowtie 2 (v2.3.4.1)⁸¹. BAM files were sorted and indexed using samtools (v1.6)⁸², and read group information was added using picard tools (v2.14.1).

Variant calling

We called short sequence variants using bcftools⁸² in three steps: (1) we created text pileup output for all BAM files using mpileup with mapping quality of 30 or greater; (2) we called using default settings; (3) we removed variants with a read depth of less than 100 across all samples or a QUAL score of less than 100. Sequencing depth per case was used as an additional criterion for filtering reads. Sequence for C57BL/6J, C3H/HeJ and DBA/2J was downloaded from the Wellcome Sanger Mouse Genome Project⁸³ (https://www.sanger.ac.uk/data/mouse-genomes-project) and that for BALB/cByJ was generated by Beierle and colleagues⁸⁴. From these files we extracted variants that differ among the four founders (Supplementary Table 1). Variants were called in three steps as above, the differences being that the minimum depth was reduced to ×10 and QUAL score to ≥30. Variants that were called reliably in all founders were advanced for genotype phasing and mapping. VCF files with variant data on UM-HET3 and their founders are provided in Supplementary Table 5.

Genotyping by Sequenom MassARRAY

Tails were placed in deep-well plates and sent in three batches to Neogen Corporation. Individuals were genotyped by SNP genotyping by Sequenom MassARRAY MALDI-TOF mass spectrometry⁸⁵ at a maximum of 270 markers. Fifty markers were removed after quality control, based on deviations from expected allele frequency (0.35 < frequency < 0.65). We chose pairs of linked markers to define maternal and paternal chromosome genotypes.

Phasing SNP genotypes

SNP genotypes were phased using R v4.0.2 (script at https://github.com/DannyArends/UM-HET3). We generated fully phased haplotypes for all sets of maternal and paternal autosomes and chromosome X. Markers fall into those that unambiguously define maternal haplotypes (n = 486, C versus B alleles and pink lines in Fig. 1c) and paternal haplotypes (n = 396, H versus D alleles and blue lines in Fig. 1c). We generated diploid genotype probabilities at all positions from phased chromosomes using the calc.genoprob() function in R/qtl. Markers were reviewed and those that had a call rate <30% were excluded. When we plot genotype or haplotype effects (Extended Data Figs. 2–5), we call genotypes and haplotypes when imputation certainty is greater than 80%. We used R/qtl to estimate number of recombinations per mouse. The mean ± s.d. value was 25.1 ± 8.6. Just over 370 animals had recombination numbers more than 2× s.d. above this value. We excluded 59 mice that had more than 80 recombinations, a value likely to be caused by sample cross-contamination. The final genotype files are available at GeneNetwork.org^86,87,88 by setting Species = Mouse, Group = Aging Mouse Lifespan Studies (NIA UM-HET3), and Type = DNA Markers and SNPs and pressing the Info button. Supplementary Table 6 lists all genotypes.

Genotype quality control

Markers and individuals with greater than 90% missing data were removed from further analysis (n = 333). We converted data from genotypes into parental haplotypes (see ‘Phasing SNP genotypes’). Markers were checked for Mendelian inheritance errors and removed when missing data after conversion exceeded 90%. We integrated MonsterPlex and Sequenom marker sets based on genomic positions. We inspected the marker map for linkage between neighbouring markers and removed a single marker (chr. 4_58784823) which did not show linkage to its neighbours. We also removed 20 individuals that died of unnatural causes and 22 that we inadvertently genotyped that had received an effective drug supplement. Finally, we constructed an R/qtl cross object for imputation using the remaining mice, imputed their haplotypes based on the full population (n = 6,438). Supplementary Table 7 shows the R/qtl cross object used for most analyses.

Inversions in the UM-HET3

There is a large inversion on chromosome 6 of the C3H/HeJ progenitor strain⁸⁹ between 51 Mb and 94 Mb which will affect regional map precision. The recombination fraction in this interval is 0.00025 versus an expectation of 0.21. At least 42 Mb of the paternal chromosome is locked in linkage. We detected a total of 13 potential inversions ranging in size of 3.5 to 54 Mb. Six suppress recombination on the maternal chromosome and seven suppress recombination on the paternal chromosome (Supplementary Table 8). There is overlap of maternal and paternal inversions on chromosomes 4, 7 and 11.

Lifespan estimates depend on age of entry

Based on work by König and colleagues^38,90, we know that in wild commensal populations of mice, 35–40% of pups do not survive beyond weaning. At birth, mean lifespan can be as low as 196 days. By 13 days, the mean expectancy has improved to 250 days for males and 500 days for females. We make these points to emphasize that lifespan estimates are made with reference to an operational starting point—a minimum ascertainment base age for a population. Researchers often incorrectly imply that mean lifespan is estimated from the date of birth, but this neglects earlier and often unknown numbers of deaths—referred to as left truncation. In our case we can only measure lifespan for UM-HET3 progeny that lived to be pubescent juveniles. When we say that the mean lifespan of the 6,438 sibs in this study is 844 ± 210 days (mean ± s.d., median of 870 days), we mean that this is the mean age at death conditioned on a mouse having lived to at least 42 ± 2 days—the age at which tail tips were taken for DNA extraction.

In a similar way we can determine the mean lifespan or life expectancy of the survivorship subsets of UM-HET3 mice that lived to be at least 185, 365, 545, 725 or 1,100 days (roughly 0.5, 1, 1.5, 2 and 3 years, respectively). The mean conditional lifespans of those five survivorships or age cohorts are 847 ± 206, 861 ± 188, 890 ± 160, 941 ± 126 and 1,162 ± 54 days (mean ± s.d.), respectively. The oldest mouse was a female that died at 1,456 days. We refer to these different ages at which lifespan can be computed as either survivorships or cohorts.

The actuarial procedure consists of sequentially studying progressive older and therefore smaller survivorships from the radix population of 6,438 mice in Fig. 1a. The minimum T-age is 42 days. Between T₄₂ and T₅₀ days a single male died. We truncated survivorships upward in 15-day steps from T₃₅ to T_1,100, a convenient step size for plots that can also be rationalized as roughly the range of survivorship T-ages (1,065 days) divided by the square root of the radix (n of 6,438). The first anchor point age was defined as 35 days in order to align plots at 365 days (1 year), 725 days (~2 years) and 1,100 days (~3 years) using 72 steps. We explored both right and bilateral truncation but do not cover the analyses here.

The actuarial mapping strategy is affected both by lower sample size and by right skew at higher truncations. To ensure that our results are not affected by this, we used three approaches. First, we use a non-parametric quantile regression in combination with standard parametric mapping to make sure associations are detected using both methods⁹¹. Second, we limited ourselves to a maximum T-age of T_1,100. Third, we used a time-dependent hazard function to validate Vita loci.

We tested a complementary approach for mapping under the assumption that loci contribute to variation in the time-dependent hazard function as in our previous work¹⁷ but with refinements. We carried out a test at each marker of the null hypothesis that the (time-dependent) hazard function of death does not depend on genotypes. To do this, we used a model where the hazard function is allowed to depend smoothly on age with haplotype and baseline covariates, using natural splines with three degrees of freedom. For computational tractability we created risk sets with 50-day windows⁹².

Implementation of actuarial mapping

We mapped in 15-day survivorship steps from T₄₂ to T_1,100. Mapping was performed using both conventional four-way mapping of genotypes at 891 markers (Fig. 1c) or by restricting analysis to each of the parental haplotypes—C versus B on the maternal chromosomes (n = 495 markers), and H versus D on the paternal chromosomes (n = 396 markers). Note that chromosome X only segregates for maternal C and B haplotypes in both sexes (Fig. 3g). In general, mapping all four genotypes provides better power. This method defined all loci in Tables 1 and 2 and full LOD scores across all survivorships are illustrated in Extended Data Figs. 1 and 2. The null model (model(H₀)) and alternative model (model(H_alt)) were fitted at each marker using the following linear models:

$${\rm{model}}({{\rm{H}}}_{0}):{\rm{lifespan}}={\rm{sex}}+{\rm{site}}+{\rm{cohort}}+{\rm{treatment}}+{\rm{error}}$$

$${\rm{model}}({{\rm{H}}}_{{\rm{alt}}}):{\rm{lifespan}}={\rm{sex}}+{\rm{site}}+{\rm{cohort}}+{\rm{treatment}}+{\rm{gtp}}+{\rm{error}}$$

Sex has two levels (M, F), site has three levels (JL, UT and UM), cohort year has nine levels (no animals were generated in 2008), treatment has two levels—untreated controls and cases that received a drug that had no significant effect on lifespan. The gtp (genotype probability) term is an n row by four-column matrix, where n is the number of individuals used in the mapping, and columns are genotype probabilities for markers computed by R/qtl (v1.48-1)^93,94. Sex was dropped as a co-factor from the regression model when mapping was stratified by sex. LOD scores were computed by comparing the fit of the null model with the fit of the alternative:

$$\mathrm{LOD}=(n/2)\times {\log }_{10}(\mathrm{sum}(\mathrm{residuals}{({{\rm{H}}}_{0})}^{2})/\mathrm{sum}(\mathrm{residuals}{({{\rm{H}}}_{\mathrm{alt}})}^{2}))$$

Mapping on chromosome X requires special care due to the four-way cross structure and greater difficulty defining highly reliable markers for haplotype, particularly close to both telomeres.

Actuarial effect size plots, their errors and interpretation

We use the four-way mapping algorithm implemented in R/qtl^94,95 while controlling for sex, site, cohort year and drug treatment. We compute maps for survivorships starting from T₄₂ up to T_1,100 (Fig. 1a,d–f). We introduce a new type of actuarial plots in Fig. 1b,d,e. The x axis defines the minimum inclusion age (T-age) of each survivorship. The expected increase in error terms in older survivorships with lower sample sizes are mitigated by the reduced range over which animals die within progressively older survivorships—from 1,410 days at T₄₂ to 356 days at T_1,100. Errors of lifespans of survivorships are relatively stable—from 844 ± 210 (mean ± s.d.) in the T₄₂ base population (s.d./range of 15%) to 1,162 ± 54.4 (mean ± s.d.) in the last T_1,100 survivorship (s.d./range of 15%). When interpreting effect size plots note that increases and decreases in mortality rates among subgroups are relative to the mean lifespan of that T-age.

The T-age of a survivorship always refers to an entire age range. Actuarial truncation in a forward direction means that the oldest-old mice are embedded in every survivorship. Reverse truncation flips the polarity by truncating from oldest to youngest and defines exitships. The _1,099T exitship consists of mice that died before 1,100 days and complements the T_1,100 survivorship. The combination of forward, reverse and even two-sided truncations is useful to dissect potential causes of mortality in age-restricted subpopulations. In this paper we restrict attention to forward truncations.

Significance testing for actuarial analyses

We evaluated the significance of mapping results across two axes: (1) the spatial location of markers and their independence across the genome; and (2) the temporal axis and the number of effectively independent actuarial tests using left truncation that progressively excludes mice that died before reaching the T-ages. We mapped using 891 markers but given the significant linkage between markers we effectively tested about 442 independent genome locations. This number of independent tests was estimated using the simpleM method with a block length ranging from 10 to 500 markers and a 10-marker step size⁹⁶. The threshold at a genome-wide type 1 Bonferroni error rate of 10% is 3.65 LOD (–log₁₀(0.1/442) at 5% it is 3.95, and at 1% it is 4.65. These are conservative thresholds. For maternal and paternal maps, the density of recombination is reduced by half, and corresponding thresholds are 3.44, 3.65 and 4.34.

Locus confidence intervals

We used the standard 1.5 LOD drop to estimate the confidence interval of linkage⁹⁷. We also estimated the temporal duration of action of loci, defined as the range of T-ages over which a locus has strong effects. The start and stop T-ages of these temporal confidence intervals are those that immediately precede or succeed survivorships that are genome-wide significant. For example, if linkage is above LOD of 3.95 from T₃₆₅ to T₇₂₅ survivorships, then the T-age confidence interval of the locus is from T₃₅₀ to T₇₄₀. This criterion is not conservative when only a few survivorships have significance. In this latter case, we defined the T-age range that brackets the peak T-age by at least a 1.5 LOD drop.

We evaluated the impact on significance testing of mapping multiple nested survivorships using a Cauchy combination test⁹⁸ at each marker across all survivorships (Table 1). Cauchy P values were adjusted for multiple testing using a Benjamini and Hochberg correction to account for multiple testing across markers, and the Cauchy –logP in Table 1 of ≥1.50 is significant at a P of less than 5%.

Locus names and ambiguities

Locus names are prefixed with Vita, Soma and Mass identifiers followed by chromosome and a letter suffix (for example, Vita1a, Soma2b and Mass11c). When the 95% confidence intervals of two adjacent QTLs overlap, and the genotype and haplotype effects appear to be the same, loci are generally assigned the same Vita identifier. For such ‘merged loci’, the 95% confidence interval is reduced to the overlapping interval. Determining whether a locus is detected in different subsets of the sibship is somewhat subjective, examples being Vita3a and Vita14b that could have been split into two loci by sex and T-age. We provide data for linkage scores in all three categories (sexes combined, females and males) for Vita loci (Supplementary Table 9), Soma loci (Supplementary Table 10) and Mass loci in (Supplementary Table 11).

Testing for sex interactions

There are significant differences between male and female stratified survivorship maps. To investigate this formally and to detect significant sex-marker interactions modulating age-specific lifespan we tested sex interactions of Vita loci at the T-age with the peak LOD (Fig. 3a). We computed the LOD score difference of two models against each other:

$${\rm{model}}({{\rm{H}}}_{0}):{\rm{longevity}}={\rm{sex}}+{\rm{site}}+{\rm{cohort}}\,{\rm{year}}+{\rm{treatment}}+{\rm{gtp}}+{\rm{error}}$$

$${\rm{model}}({{\rm{H}}}_{{\rm{alt}}}):{\rm{longevity}}={\rm{sex}}+{\rm{site}}+{\rm{cohort}}\,{\rm{year}}+{\rm{treatment}}+{\rm{gtp}}+{\rm{gtp}}:{\rm{sex}}+{\rm{error}}$$

Since we are only interested in significant sex interactions at previously detected Vita loci (excluding only VitaXa) we used a Bonferroni correction based on 28 markers. A LOD score above 2.75 is significant (–log₁₀(0.05/28)).

Testing pairwise epistatic interactions

A two-dimensional analysis of marker-marker interactions affecting life expectancies was performed by testing the following models against each other in four survivorships—T₄₂, T₃₆₅, T₇₄₀ and T₉₀₅.

$${\rm{model}}({{\rm{H}}}_{0}):{\rm{lifespan}}={\rm{sex}}+{\rm{site}}+{\rm{cohort}}\,{\rm{year}}+{\rm{treatment}}+{{\rm{gtp}}}_{{\rm{m}}1}+{{\rm{gtp}}}_{{\rm{m}}2}+{\rm{error}}$$

$${\rm{model}}({{\rm{H}}}_{{\rm{alt}}}):{\rm{lifespan}}={\rm{sex}}+{\rm{site}}+{\rm{cohort}}\,{\rm{year}}+{\rm{treatment}}+{{\rm{gtp}}}_{{\rm{m}}1}+{{\rm{gtp}}}_{{\rm{m}}2}+{{\rm{gtp}}}_{{\rm{m}}1}:{{\rm{gtp}}}_{{\rm{m}}2}+{\rm{error}}$$

We computed the difference in fit between H₀ and H_alt as explained in ‘Implementation of actuarial mapping’. To limit numbers of tests, we initially tested interactions between the 29 Vita loci (Table 2), excluding all pairs of loci on the same chromosome (Fig. 5a). This resulted in the matrix of 387 non-syntenic tests of pairs of loci per sex. We used Bonferroni corrections for most work. With 387 Vita–Vita tests, a LOD of approximately 4.0 is appropriate, but for more comprehensive tests of all loci in all 4 survivorships (11,768 tests) the corresponding P < 0.05 threshold is 5.37, a value that limits the yield of interactions to only 50 (Supplementary Table 12). By contrast, the Benjamini and Hochberg test suggests that 966 interactions have FDR of less than 0.005 (Supplementary Table 12.2). We opted for a compromise and analysed a total 289 unique epistatic interactions that are visualized in part in Fig. 5g,h for the T₄₂ survivorship, and much more comprehensively Extended Data Fig. 9 for all 4 survivorships. All interactions have LODs greater than 3.8. Of a total of 11,768 tests fewer than 4% have LODs ≥ 3.8 (Supplementary Table 12).

Visualizing genetic modulation of mortality rates

We plotted age-dependent mortality and the HRs of haplotypes to give more insight into the specific ages over which genetic differences operate with more or less force. We computed mortality rates of maternal and paternal haplotype pairs (C versus B, H versus D) for males and females separately. We used LOESS smoothers with span parameters of 0.2 (Extended Data Fig. 4), and adaptive LOESS with a span of 0.3 (refs. ^99,100,101) with shorter spans over ranges with high numbers of deaths (800–950 days) and longer spans over ranges with few deaths (Fig. 2f). log₂ HRs greater than 0.6 represent relatively strong genetic effects (HR > 1.5 or HR < 0.67 in Fig. 2f).

Body mass versus lifespan analysis—the Soma loci

Body masses were measured at 42 days (right before tails were biopsied for genotyping); and at 183, 365, 548 and 731 days (6, 12, 18 and 24 months). We used these data to map loci for body masses for both sexes so that we could align and compare body mass loci with Vita and Soma loci (for example, Fig. 5a–c). Sample sizes for mapping Mass loci range from a low of 2,459 at 42 days to a high of 4,688 at 183 days (Supplementary Table 11.1). Models used are like those used for mapping Vita loci, but the variable is body mass at each of five ages.

Actuarial analysis of Soma loci

To investigate the effects of body mass on lifespan, we adjusted body mass and lifespan using the model:

$$\begin{array}{l}{{\rm{Lifespan}}}_{({\rm{adjusted}})}={\rm{mean}}({\rm{Lifespan}})+{\rm{residuals}}({\rm{Lifespan}} \sim {\rm{Sex}}\\ \,+{\rm{Site}}+{\rm{Cohort}}+{\rm{Treatment}})\end{array}$$

$$\begin{array}{l}{{\rm{BW}}}_{({\rm{adjusted}})}={\rm{mean}}({\rm{BW}})+{\rm{residuals}}({\rm{BW}} \sim {\rm{Sex}}\\ \,+{\rm{Site}}+{\rm{Cohort}}+{\rm{Treatment}})\end{array}$$

Where BW is body mass. This model includes mice for which we have lifespan and body mass data at one or more of the five weight ages. We computed the sex-stratified Spearman rank order correlations between body mass at 42, 183, 365, 548 and 730 days with lifespan in 15-day increments (Fig. 4d–p). We compute the –logP of the difference in correlation between males and females using the cor.test function in R. The average correlation in the combined population between adjusted body mass at 183 days (~6 months) and adjusted lifespan is rho = –0.206. When computed for males, correlation is stronger (rho = –0.284), while it is weaker for females (rho = –0.110).

CTL mapping

We used correlated trait locus (CTL) mapping³⁷ to determine if a distinct set of loci modulate correlations between body mass and subsequent lifespan (Tables 3–4). Before using this procedure, we adjusted body mass and lifespan using sex, site, cohort year, and drug treatment as covariates. At each marker we stratified survivorships based on genotypes. Ideally, the sample size for each subgroup would be above 400, but in some cases, N was as low as 200 due to genotype uncertainty and to the lower sample sizes at T₄₂ (N = 2549) and T₇₃₀ (N = 2208). We computed rho correlations for each genotype with n > 100 and the z-scores associated with differences:

$$z=0.5\times \log ((1.0+\rho )/(1.0-\rho ))$$

Here cor is a four-value vector containing the observed correlations for CH, BH, CD and BD genotypes. We compute the sum of squares (sumOfSq) by multiplying the observed sample sizes (ss) of the CH, BH, CD and BD genotypes to the squared z-scores:

$${\rm{sumOfSq}}={\rm{sum}}({{\rm{ss}}}^{* }{{\rm{z}}}^{2})$$

We compute the squares of sums (sqOfSum):

$${\rm{sqOf}}{\rm{Sum}}={\rm{sum}}({{\rm{ss}}}^{* }{{\rm{z}}}^{2})$$

Using these values, and the sum of the sample sizes on which each correlation is based, we compute the critical value (Cv) which follows a chi-square distribution under the null hypothesis that all correlations (z-scores) are from the same distribution:

$${\rm{Cv}}={\rm{sumOfSq}}{\rm{\mbox{--}}}({\rm{sqOfSum}}/{\rm{sum}}({\rm{ss}}))$$

The Cv is converted to a P value using the chi-squared distribution using the pchisq() function in R, with P[X > x] and three degrees of freedom (number of genotypes – 1).

We adjust the significance threshold for multiple tests using the p.adjust function in R at a 5% FDR (refs. ^39,102). The 5% FDR threshold is approximately 2.75 –logP. While still stringent, this threshold is less harsh than that applied to Vita loci that use a highly conservative Bonferroni correction. Our rationale for this difference is our use of conservative non-parametric rho for comparing correlations of body mass to lifespan. Differences in correlation correspond to days gained or lost relative to an average individual that can be converted to effect sizes measured in days gained or lost per gram of body mass. We computed the linear regression coefficient for body mass on lifespan for the four genotypes as:

$${\mathrm{Lifespan}}_{(\mathrm{adjusted})}=\mathrm{mean}(\mathrm{Lifespan})+\beta \times {\mathrm{bodyweight}}_{(\mathrm{adjusted})}$$

where mean(Lifespan) is the intercept of the total population (all genotypes combined), while the β coefficient is the estimated effect size of body mass on lifespan based on the subpopulation defined by each genotype (Table 4). This leads to effect sizes relative to an averaged mouse. While there are caveats with respect to interpreting these effects because they are computed relative to this mean, these values give readers a sense of the impact of Soma loci on life expectancies.

Broad-sense haplotype-based heritability

We compute broad-sense genotype-based heritability by fitting a full model including the haplotype probabilities of the 29 top markers of all Vita loci:

$$\begin{array}{l}{{\rm{lifespan}}}_{({\rm{T}}-{\rm{age}})}={\rm{sex}}+{\rm{site}}+{\rm{cohort}}+{\rm{treatment}}+{{\rm{gtp}}}_{{Vita}1a}\\ \,+{{\rm{gtp}}}_{{Vita}1b}+\ldots +{{\rm{gtp}}}_{{VitaXa}}+{\rm{error}}\end{array}$$

Broad-sense haplotype-based heritability is estimated by fitting this model to all survivorships. When stratifying by sex, sex is dropped from the model as well as from the H²_e component described in the computation of environmental variance component below. Computing heritability is done by taking the following approach, adapted from Falconer¹⁰³ and Lynch and Walsh¹⁰⁴ using five steps:

Step 1: The vector of partial total variances (PTV):

PTV = (σ² – σ²_residuals)/r

Step 2: An adjustment factor adj—the sum of the mean sum of squares for the fixed effect including the residual variance:

adj = sum(σ²)/r

Step 3: The contribution of each parameter in PTV to the model:

C²_p = PTV/adj

Step 4: The total broad-sense genotype-based heritability:

H²_h = C²_Vita1a + C²_Vita1b + _… + C²_VitaXa

Step 5: The environmental variance:

H²_e = C²_sex + C²_site + C²_cohort + C²_treatment

where σ² is the vector containing all the means of sums of squares (including the residual) computed by the ANOVA model. σ²_residuals is the residual mean sum of squares. r is the average number of replicates for each genotype. H²_h can be interpreted as the broad-sense genotype-based heritability as a sum of all Vita loci. H²_e is the environmental variance estimate of the known environmental covariates. Unexplained variance H²_u can be computed as:

$${{H}^{2}}_{{\rm{u}}}=1-({{H}^{2}}_{{\rm{h}}}+{{H}^{2}}_{{\rm{e}}})$$

To obtain upper and lower bounds on H²_h and H²_e we generated 50 bootstrap resamples and fit the ANOVA model to each using a random subset of 90% of the survivorship. We computed median and standard deviations across all bootstraps to estimate errors of H²_h and H²_e as a function of survivorship T-age.

Candidate genes analysis

All annotated features located within the 95% confidence interval were downloaded using BioMart¹⁰⁵. We applied several criteria to prioritize positional candidate genes: (1) first, we considered only the protein-coding genes that reside in the 95% confidence intervals; (2) we then selected genes with annotated non-synonymous SNPs segregating in the population, and these were further ranked by the potential deleterious effect based on the ENSEMBL variant effect predictor¹⁰⁶; and (3) another priority score for the candidate genes was based on whether they were listed as ageing and longevity genes in GenAge¹⁰⁷. We also compared our positional candidate genes against known human, C. elegans, D. melanogaster and Saccharomyces cerevisiae genes reportedly associated with age in GenAge. To accomplish this, we use biomaRt to convert mouse gene symbols to the corresponding species-specific orthologous gene symbols. Supplementary Tables 13 and 14 provide access to all gene models and variant types segregating in Vita and Soma loci.

Gene set over-representation analysis

We analysed gene set enrichment in Vita and Soma loci using R. Gene annotations were retrieved from Ensembl via biomaRt, retaining only mouse protein-coding genes, but excluding olfactory receptor genes (Olf*, n = 1,427), vomeronasal receptor genes (Vmn*, n = 595), predicted genes (Gm*, n = 1,246) and RIKEN cDNA sequences (n = 490). For each locus we tested set over-representation against our filtered whole-genome background (n = 18,830) using clusterProfiler for Gene Ontology (Biological Process, Molecular Function, Cellular Component), KEGG pathways and Reactome pathways. An adjusted P cut-off of 0.2 was applied to capture marginally significant enrichments for subsequent analysis. Results are summarized in Supplementary Table 16.

C. elegans candidate gene screening of Vita9b

Out of 262 protein-coding genes in Vita9b, we highlighted 98 with missense variants and a subset with C. elegans orthologues using Ortholist2¹⁰⁸ (Supplementary Table 16). From these, we selected a subset of 15 genes (Fig. 6g) represented in the Ahringer C. elegans RNAi library¹⁰⁹ for testing: (1) acds-10 (orthologue of mouse Acad11); (2) chhy-1 (Hyal1, Hyal2 and Hyal3); (3) oct-1 (Slc22a13); (4) col-135 (Col6a6); (5) frm-8 (Frmpd1); (6) let-413 (Lrrc2); (7) F30A10.3 (Ip6k2); (8) lin-41 (Trim71); (9) lis-1 (Poc1a); (10) stt-3 (Stt3b); (11) C41D11.3 (Csrnp1); (12) znf-782 (Zfp445); (13) pes-4 (Pcbp4); (14) pho-6 (Acp3); and (15) dpf-5 (Apeh). We also screened using the L4440 empty vector control and daf-2 (Igf1r) as positive control. The sequence-verified single RNAi clones were inoculated in 10 ml lysogeny broth (LB) with ampicillin (100 μg ml⁻¹) and tetracycline (50 μg ml⁻¹) overnight at 37 °C. The next day, 5 ml of LB with 0.5 mM isopropyl β-D-1-thiogalactopyranoside was added and inoculants were incubated for 30 min at 37 °C with shaking for pre-induction. The bacterial inoculants were centrifuged (3,700 rpm for 15 min) and the pellet was resuspended in 10 ml liquid NGM with 50 µg ml⁻¹ IPTG, 100 µg ml⁻¹ ampicillin and 13.32 mg ml⁻¹ nystatin.

Gravid adult stage TJ1060 [spe-9(hc88);rrf-3(b26)] C. elegans were synchronized by bleaching. Resulting L₁ stage worms were suspended in concentrated heat-inactivated OP50 bacteria in liquid NGM and transferred into U-bottom 96-well plates in a final volume of 100 µl. The worms were grown on an orbital shaker (Heidolph Titramax 1000) at 600 rpm until they reached the L4 stage (2.0–2.5 days) at 25 °C to prevent progeny production. When worms reached L4 stage, 20 µl of bacterial inoculant of each RNAi clone were added into single wells of the U-bottom 96-well plates containing worms. At least 30 worms per well and 8 wells per clone were used for two independent screens. The MicroTracker (InVivo Biosystems) system was used to measure motility in liquid in 96-well plates by infrared beam breaks. Plates were measured daily for a period of at least 90 min. Significance of differences in activity were estimated using the AUC (Fig. 6f) for either the full lifespan (1–30 days relative lifelong motility, Fig. 6g) or only after 14 days (Fig. 6h) using a two-tailed t-test of control-normalized AUC values assuming unequal variance and using a Bonferroni correction for 15 tests.

Mendelian randomization analysis

We used Mendelian randomization methods to rank the translational relevance of genes within Vita loci against their human orthologues. We extracted lists of protein-coding genes within Vita1a and Vita9b with significant human cis-expression quantitative trait loci (cis-eQTLs) in blood from eQTLGen (31,684 individuals)¹¹⁰ as exposures (Supplementary Table 15). As outcome variables, we used summary statistics on parental longevity^111,112 from the IEU OpenGWAS project¹¹³ and on human longevity¹¹⁴ from the GWAS Catalog¹¹⁵. We tested for potential causal effects of variation in expression on: (1) extreme longevity of at least one parent (>95 years of age, IEU OpenGWAS.io trait ebi-a-GCST003395)¹¹¹; (2) both parents in the top 10% of longevity (IEU OpenGWAS.io trait ebi-a-GCST006698)¹¹²; (3) mother’s age at death (IEU OpenGWAS.io trait ebi-a-GCST006699)¹¹²; (4) father’s age at death IEU OpenGWAS.io trait (ebi-a-GCST006700)¹¹²; (5) mean parental age at death (IEU OpenGWAS.io trait ebi-a-GCST006702)¹¹²; (6) individual survival into the top 10% (GWAS Catalog: GCST008598)¹¹⁴; and (7) individual survival into the top 1% (GWAS Catalog: GCST008599)¹¹⁴. In many instances, multiple cis-eQTLs were detected in the GWAS summary statistics, allowing them to be used as instrumental variables. Because linkage disequilibrium pruning would leave only a small number of independent variants, we applied an established principal component analysis-based approach that aggregates correlated instrumental variables into linkage disequilibrium-based principal components (principal component instrumental variables)¹¹⁶ that were evaluated using the 1000 Genomes Project panel¹¹⁷ with PLINK¹¹⁸ v1.90b6.21. We retained top principal components explaining more than 99% of genetic variance and used the inverse-variance weighted method to test for potential causal effects. When only a single instrumental variable was available for both the exposure and the outcome, or when principal component analysis yielded a single component, causal effects were estimated using the Wald ratio method implemented in TwoSampleMR (v0.6.2)⁶³. For associations that were significant, we additionally performed linkage disequilibrium pruning and conducted a leave-one-out sensitivity analysis, sequentially removing each independent variant and estimating effects of the remainder. We evaluated heterogeneity of estimates using Cochran’s Q test, and horizontal pleiotropy was tested. In this analysis we were sensitive to key assumptions of Mendelian randomization analysis^62,119. The assumption of relevance is reasonably well satisfied but the assumptions of independence and exclusion are only partially satisfied. However, in the context of using Mendelian randomization methods to rank candidate genes in which negative findings are as informative as positive findings, this procedure improves bidirectional translation between mouse and human studies of mortality and longevity.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

All primary data used in this study are provided in the supplementary tables. We also provide all lifespan and body mass data for animals used in this study and for an additional 17,858 UM-HET3 mice (total of 24,296 ITP mice born up to 20 October 2020) at the URL: https://genenetwork.org/show_trait?trait_id=10001&dataset=HET3-ITPPublish along with tools for truncation, mapping and analysis of correlations between variables such as lifespan and body mass. All figures and supplementary tables are deposited at https://aging.genenetwork.org/UM-HET3 and https://files.genenetwork.org/current/umhet3_2025/. All published data of the ITP is also made openly available at the Mouse Phenome Database ITP Portal at https://phenome.jax.org/projects/ITP1/.

Code availability

Code used in mapping and analysis and scripts used to generate all figures and extended data figures are available at https://github.com/DannyArends/UM-HET3. Supplementary files and original figures in Adobe Illustrator format are available at https://files.genenetwork.org/current/umhet3_2025/.

References

Maynard Smith, J. Review lectures on senescence — I. The causes of ageing. Proc. R. Soc. Lond. B. 157, 115–127 (1962).
Article ADS Google Scholar
Rose, M. & Charlesworth, B. A test of evolutionary theories of senescence. Nature 287, 141–142 (1980).
Article ADS CAS PubMed Google Scholar
Gelman, R., Watson, A., Bronson, R. & Yunis, E. Murine chromosomal regions correlated with longevity. Genetics 118, 693–704 (1988).
Article CAS PubMed PubMed Central Google Scholar
Vaupel, J. W. et al. Biodemographic trajectories of longevity. Science 280, 855–860 (1998).
Article ADS CAS PubMed Google Scholar
Miller, R. A. Genes against aging. J. Gerontol. A 67, 495–502 (2012).
Article Google Scholar
Hook, M. et al. Genetic cartography of longevity in humans and mice: Current landscape and horizons. Biochim. Biophys. Acta 1864, 2718–2732 (2018).
Article CAS Google Scholar
Huang, W. et al. Context-dependent genetic architecture of Drosophila life span. PLoS Biol. 18, e3000645 (2020).
Article CAS PubMed PubMed Central Google Scholar
Di Francesco, A. et al. Dietary restriction impacts health and lifespan of genetically diverse mice. Nature 634, 684–692 (2024).
Article ADS PubMed PubMed Central Google Scholar
Yampolsky, L. Y., Pearse, L. E. & Promislow, D. E. Age-specific effects of novel mutations in Drosophila melanogaster I. Mortality. Genetica 110, 11–29 (2000).
Article Google Scholar
Glazier, D. S. Does death drive the scaling of life?. Biol. Rev. 100, 586–619 (2025).
Article PubMed Google Scholar
Flatt, T. & Partridge, L. Horizons in the evolution of aging. BMC Biol. 16, 93 (2018).
Article PubMed PubMed Central Google Scholar
Fortney, K. et al. Genome-wide scan informed by age-related disease identifies loci for exceptional human longevity. PLoS Genet. 11, e1005728 (2015).
Article PubMed PubMed Central Google Scholar
Timmers, P. R. et al. Genomics of 1 million parent lifespans implicates novel pathways and common diseases and distinguishes survival chances. eLife 8, e39856 (2019).
Article PubMed PubMed Central Google Scholar
Cottam, N. C. From circuits to lifespan: translating mouse and human timelines with neuroimaging-based tractography. J. Neurosci. 45, e1429242025 (2025).
Article CAS PubMed PubMed Central Google Scholar
Pletcher, S. D., Houle, D. & Curtsinger, J. W. Age-specific properties of spontaneous mutations affecting mortality in Drosophila melanogaster. Genetics 148, 287–303 (1998).
Article CAS PubMed PubMed Central Google Scholar
Spencer, C. C. & Promislow, D. E. L. Age-specific changes in epistatic effects on mortality rate in Drosophila melanogaster. J. Hered. 96, 513–521 (2005).
Article CAS PubMed Google Scholar
Bou Sleiman, M. et al. Sex- and age-dependent genetics of longevity in a heterogeneous mouse population. Science 377, eabo3191 (2022).
Article PubMed PubMed Central Google Scholar
Hjelmborg, J.vB. et al. Genetic influence on human lifespan and longevity. Hum. Genet. 119, 312–321 (2006).
Article Google Scholar
Medawar, P. B. The Uniqueness of the Individual (Basic Books, 1957).
Williams, G. C. Pleiotropy, natural selection, and the evolution of senescence. Evolution 11, 398–411 (1957).
Article Google Scholar
Kirkwood, T. B. L. Evolution of ageing. Nature 270, 301–304 (1977).
Article ADS CAS PubMed Google Scholar
Miller, R. A. et al. An Aging Interventions Testing Program: study design and interim report. Aging Cell 6, 565–575 (2007).
Article ADS CAS PubMed Google Scholar
Harrison, D. E. et al. Rapamycin fed late in life extends lifespan in genetically heterogeneous mice. Nature 460, 392–395 (2009).
Article ADS CAS PubMed PubMed Central Google Scholar
Miller, R. A. Biomarkers of aging. Sci. Aging Knowledge Environ. 2001, pe2 (2001).
Article CAS PubMed Google Scholar
Williams, G. C. & Taylor, P. D. in Evolution of Longevity in Animals: A Comparative Approach (eds Woodhead, A. D. & Thompson, K. H.) 235–245 (Springer, 1987).
Hamilton, W. D. The moulding of senescence by natural selection. J. Theor. Biol. 12, 12–45 (1966).
Article ADS CAS PubMed Google Scholar
Kirkwood, T. B. L., Rose, M. R. & Harvey, P. H. Evolution of senescence: late survival sacrificed for reproduction. Phil. Trans. R. Soc. Lond. B 332, 15–24 (1991).
Article ADS CAS Google Scholar
Shenhar, B. et al. Heritability of intrinsic human life span is about 50% when confounding factors are addressed. Science 391, 504–510 (2026).
Article CAS PubMed Google Scholar
Schmidt, J. A., Oatley, J. M. & Brinster, R. L. Female mice delay reproductive aging in males. Biol. Reprod. 80, 1009–1014 (2009).
Article CAS PubMed PubMed Central Google Scholar
Suntzeff, V., Cowdry, E. V. & Hixon, B. B. Influence of maternal age on offspring in mice. J. Gerontol. 17, 2–7 (1962).
Article CAS PubMed Google Scholar
Baud, A. et al. Genetic variation in the social environment contributes to health and disease. PLoS Genet. 13, e1006498 (2017).
Article PubMed PubMed Central Google Scholar
Clutton-Brock, T. H. & Isvaran, K. Sex differences in ageing in natural populations of vertebrates. Proc. Biol. Sci. 274, 3097–3104 (2007).
CAS PubMed PubMed Central Google Scholar
Razzoli, M. et al. Social stress shortens lifespan in mice. Aging Cell 17, e12778 (2018).
Article PubMed PubMed Central Google Scholar
Ingram, D. K., Reynolds, M. A. & Les, E. P. The relationship of genotype, sex, body weight, and growth parameters to lifespan in inbred and hybrid mice. Mech. Ageing Dev. 20, 253–266 (1982).
Article CAS PubMed Google Scholar
Miller, R. A., Harper, J. M., Galecki, A. & Burke, D. T. Big mice die young: early life body weight predicts longevity in genetically heterogeneous mice. Aging Cell 1, 22–29 (2002).
Article CAS PubMed Google Scholar
Cheng, C. J., Gelfond, J. A. L., Strong, R. & Nelson, J. F. Genetically heterogeneous mice exhibit a female survival advantage that is age- and site-specific: results from a large multi-site study. Aging Cell 18, e12905 (2019).
Article PubMed PubMed Central Google Scholar
Arends, D. et al. Correlation trait loci (CTL) mapping: Phenotype network inference subject to genotype. J. Open Source Softw. 1, 87 (2016).
Article ADS Google Scholar
Manser, A., Lindholm, A. K., König, B. & Bagheri, H. C. Polyandry and the decrease of a selfish genetic element in a wild house mouse population. Evolution 65, 2435–2447 (2011).
Article PubMed PubMed Central Google Scholar
Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. B 57, 289–300 (1995).
Article Google Scholar
Mackay, T. F. C. Epistasis and quantitative traits: Using model organisms to study gene–gene interactions. Nat. Rev. Genet. 15, 22–33 (2014).
Article CAS PubMed Google Scholar
Chippindale, A. K., Gibson, J. R. & Rice, W. R. Negative genetic correlation for adult fitness between sexes reveals ontogenetic conflict in Drosophila. Proc. Natl Acad. Sci. USA 98, 1671–1675 (2001).
Article ADS CAS PubMed PubMed Central Google Scholar
Zhu, C. et al. Amplification is the primary mode of gene-by-sex interaction in complex human traits. Cell Genomics 3, 100297 (2023).
Article CAS PubMed PubMed Central Google Scholar
Cole, J. M. et al. The battle of the sexes in humans is highly polygenic. Proc. Natl Acad. Sci. USA 121, e2412315121 (2024).
Article CAS PubMed PubMed Central Google Scholar
Vieira, C. et al. Genotype-environment interaction for quantitative trait loci affecting life span in Drosophila melanogaster. Genetics 154, 213–227 (2000).
Article CAS PubMed PubMed Central Google Scholar
Arnqvist, G., Vellnow, N. & Rowe, L. The effect of epistasis on sexually antagonistic genetic variation. Proc. Biol. Sci .281, 20140489 (2014).
PubMed PubMed Central Google Scholar
Ruzicka, F. et al. Genome-wide sexually antagonistic variants reveal long-standing constraints on sexual dimorphism in fruit flies. PLoS Biol. 17, e3000244 (2019).
Article CAS PubMed PubMed Central Google Scholar
de Magalhães, J. P. et al. Human ageing genomic resources: Updates on key databases in ageing research. Nucleic Acids Res. 52, D900–D908 (2024).
Article PubMed PubMed Central Google Scholar
Smith, G. S. & Walford, R. L. Influence of the main histocompatibility complex on ageing in mice. Nature 270, 727–729 (1977).
Article ADS CAS PubMed Google Scholar
Stevens, T. H. & Forgac, M. Structure, function and regulation of the vacuolar (H⁺)-ATPase. Annu. Rev. Cell Dev. Biol. 13, 779–808 (1997).
Article CAS PubMed Google Scholar
Landis, G. N., Bhole, D. & Tower, J. A search for doxycycline-dependent mutations that increase Drosophila melanogaster life span identifies the VhaSFD, Sugar baby, filamin, fwd and Cctl genes. Genome Biol. 4, R8 (2003).
Article PubMed PubMed Central Google Scholar
Li, T. Y. et al. A lysosomal surveillance response to stress extends healthspan. Nat. Cell Biol. 27, 1083–1097 (2025).
Article CAS PubMed PubMed Central Google Scholar
Yang, S. et al. ATP6V1H deficiency impairs glucose tolerance by augmenting endoplasmic reticulum stress in high fat diet fed mice. Arch. Biochem. Biophys. 716, 109116 (2022).
Article CAS PubMed Google Scholar
Jackson, A. U., Galecki, A. T., Burke, D. T. & Miller, R. A. Mouse loci associated with life span exhibit sex-specific and epistatic effects. J. Gerontol. A 57, B9–B15 (2002).
Article Google Scholar
Hsu, A.-L., Feng, Z., Hsieh, M.-Y. & Xu, X. Z. S. Identification by machine vision of the rate of motor activity decline as a lifespan predictor in C. elegans. Neurobiol. Aging 30, 1498–1503 (2009).
Article PubMed Google Scholar
Newell Stamper, B. L. et al. Movement decline across lifespan of Caenorhabditis elegans mutants in the insulin/insulin-like signaling pathway. Aging Cell 17, e12704 (2018).
Article PubMed Google Scholar
Kenyon, C., Chang, J., Gensch, E., Rudner, A. & Tabtiang, R. A C. elegans mutant that lives twice as long as wild type. Nature 366, 461–464 (1993).
Article ADS CAS PubMed Google Scholar
Zhu, J. & Chen, X. MCG10, a novel p53 target gene that encodes a KH domain RNA-binding protein, is capable of inducing apoptosis and cell cycle arrest in G₂-M. Mol. Cell. Biol. 20, 5602–5618 (2000).
Article CAS PubMed PubMed Central Google Scholar
Ito, Y. et al. Suppression of poly(rC)-binding protein 4 (PCBP4) reduced cisplatin resistance in human maxillary cancer cells. Sci. Rep. 5, 12360 (2015).
Article ADS PubMed PubMed Central Google Scholar
Yan, W. et al. Mice deficient in poly(C)-binding protein 4 are susceptible to spontaneous tumors through increased expression of ZFP871 that targets p53 for degradation. Genes Dev. 30, 522–534 (2016).
Article CAS PubMed PubMed Central Google Scholar
Wang, X. et al. Joint mouse–human phenome-wide association to test gene function and disease risk. Nat. Commun. 7, 10464 (2016).
Article ADS CAS PubMed PubMed Central Google Scholar
Wang, J. et al. MARRVEL: integration of human and model organism genetic resources to facilitate functional annotation of the human genome. Am. J. Hum. Genet. 100, 843–853 (2017).
Article CAS PubMed PubMed Central Google Scholar
Sanderson, E. et al. Mendelian randomization. Nat. Rev. Methods Primers 2, 6 (2022).
Article CAS PubMed PubMed Central Google Scholar
Hemani, G. et al. The MR-Base platform supports systematic causal inference across the human phenome. eLife 7, e34408 (2018).
Article PubMed PubMed Central Google Scholar
Jiang, N. & Nelson, J. F. Sex differences in mouse longevity and responses to geroprotective drugs: Implications for human intervention. Public Policy Aging Rep. 33, 120–124 (2023).
Article PubMed PubMed Central Google Scholar
Garagnani, P. et al. Whole-genome sequencing analysis of semi-supercentenarians. eLife 10, e57849 (2021).
Article CAS PubMed PubMed Central Google Scholar
Sebastiani, P. & Perls, T. T. The genetics of extreme longevity: lessons from the new England centenarian study. Front. Genet. 3, 277 (2012).
Article PubMed PubMed Central Google Scholar
Soerensen, M. et al. No association between variation in longevity candidate genes and aging-related phenotypes in oldest-old Danes. Exp. Gerontol. 78, 57–61 (2016).
Article CAS PubMed PubMed Central Google Scholar
Kirkwood, T. B. in Genetic Effects on Aging II (ed. Harrison, D. E.) 9–19 (Telford Press, 1990).
Partridge, L. & Harvey, P. H. The ecological context of life history evolution. Science 241, 1449–1455 (1988).
Article ADS CAS PubMed Google Scholar
Leips, J. & Mackay, T. F. Quantitative trait loci for life span in Drosophila melanogaster: interactions with genetic background and larval density. Genetics 155, 1773–1788 (2000).
Article CAS PubMed PubMed Central Google Scholar
Strong, R. et al. Extension of lifespan by epicatechin, halofuginone and mitoglitazone in male but not female genetically heterogeneous mice. Geroscience https://doi.org/10.1007/s11357-025-01881-6 (2025).
Fontana, L., Partridge, L. & Longo, V. D. Extending healthy life span–from yeast to humans. Science 328, 321–326 (2010).
Article ADS CAS PubMed PubMed Central Google Scholar
Ziehm, M. et al. Drug repurposing for aging research using model organisms. Aging Cell 16, 1006–1015 (2017).
Article CAS PubMed PubMed Central Google Scholar
Schork, N. J. & Goetz, L. H. From precision interventions to precision health. Nat. Commun. 16, 5024 (2025).
Article ADS CAS PubMed PubMed Central Google Scholar
Bogue, M. A. et al. Mouse Phenome Database: a data repository and analysis suite for curated primary mouse phenotype data. Nucleic Acids Res. 48, D716–D723 (2019).
Google Scholar
Miller, R. A. et al. Rapamycin, but not resveratrol or simvastatin, extends life span of genetically heterogeneous mice. J. Gerontol. A 66, 191–201 (2011).
Article Google Scholar
Gonzalez-Freire, M. et al. The road ahead for health and lifespan interventions. Ageing Res. Rev. 59, 101037 (2020).
Article CAS PubMed PubMed Central Google Scholar
Strong, R. et al. Rapamycin-mediated mouse lifespan extension: Late-life dosage regimes with sex-specific effects. Aging Cell 19, e13269 (2020).
Article CAS PubMed PubMed Central Google Scholar
You, F. M. et al. BatchPrimer3: a high throughput web application for PCR and sequencing primer design. BMC Bioinformatics 9, 253 (2008).
Article PubMed PubMed Central Google Scholar
Nguyen-Dumont, T. et al. Hi-Plex for high-throughput mutation screening: application to the breast cancer susceptibility gene PALB2. BMC Med. Genomics 6, 48 (2013).
Article PubMed PubMed Central Google Scholar
Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359 (2012).
Article CAS PubMed PubMed Central Google Scholar
Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079 (2009).
Article PubMed PubMed Central Google Scholar
Keane, T. M. et al. Mouse genomic variation and its effect on phenotypes and gene regulation. Nature 477, 289–294 (2011).
Article ADS CAS PubMed PubMed Central Google Scholar
Beierle, J. A. et al. Genetic basis of thermal nociceptive sensitivity and brain weight in a BALB/c reduced complexity cross. Mol. Pain 18, 17448069221079540 (2022).
Article CAS PubMed PubMed Central Google Scholar
Gabrial, S., Ziaugra, L. & Tabbaa, D. SNP genotyping using the Sequenom MassARRAY iPLEX platform. Curr. Protoc. Hum. Genet. 60, 2.12.1–2.12.18 (2009).
Google Scholar
Williams, R. W. The Portable Dictionary of the Mouse Genome: a personal database for gene mapping and molecular biology. Mamm. Genome 5, 372–375 (1994).
Article CAS PubMed Google Scholar
Sloan, Z. et al. GeneNetwork: framework for web-based genetics. J. Open Source Softw. 1, 25 (2016).
Article ADS Google Scholar
Mulligan, M. K., Mozhui, K., Prins, P. & Williams, R. W. GeneNetwork: a toolbox for systems genetics. Methods Mol. Biol. 1488, 75–120 (2017).
Article CAS PubMed PubMed Central Google Scholar
Akeson, E. C. et al. Chromosomal inversion discovered in C3H/HeJ mice. Genomics 87, 311–313 (2006).
Article CAS PubMed Google Scholar
König, B. & Lindholm, A. K. in Evolution of the House Mouse (eds Piálek, J. et al.) 114–134 (Cambridge Univ. Press, 2012).
Koenker, R., Chernozhukov, V., He, X. & Peng, L. Handbook on Quantile Regression (Chapman and Hall/CRC, 2017).
Therneau, T. M. Modeling Survival Data: Extending the Cox Model (Springer, 2000).
Broman, K. W. & Sen, S. A Guide to QTL Mapping with R/Qtl (Springer, 2009).
Arends, D., Prins, P., Jansen, R. C. & Broman, K. W. R/qtl: high-throughput multiple QTL mapping. Bioinformatics 26, 2990–2992 (2010).
Article CAS PubMed PubMed Central Google Scholar
Broman, K. W., Wu, H., Sen, Ś & Churchill, G. A. R/qtl: QTL mapping in experimental crosses. Bioinformatics 19, 889–890 (2003).
Article CAS PubMed Google Scholar
Gao, X., Starmer, J. & Martin, E. R. A multiple testing correction method for genetic association studies using correlated single nucleotide polymorphisms. Genet. Epidemiol. 32, 361–369 (2008).
Article PubMed Google Scholar
Manichaikul, A. & Dupuis, J. Sen, Ś. & Broman, K. W. Poor performance of bootstrap confidence intervals for the location of a quantitative trait locus. Genetics 174, 481–489 (2006).
Article PubMed PubMed Central Google Scholar
Liu, Y. & Xie, J. Cauchy combination test: a powerful test with analytic p-value calculation under arbitrary dependency structures. J. Am. Stat. Assoc. 115, 393–402 (2020).
Article CAS PubMed Google Scholar
Cleveland, W. S. & Devlin, S. J. Locally weighted regression: an approach to regression analysis by local fitting. J. Am. Stat. Assoc. 83, 596–610 (1988).
Article Google Scholar
Buja, A., Hastie, T. & Tibshirani, R. Linear smoothers and additive models. Ann. Stat. 17, 453–510 (1989).
Google Scholar
Hastie, T., Tibshirani, R. & Friedman, J. in The Elements of Statistical Learning: Data Mining, Inference, and Prediction (eds Hastie, T. et al.) 191–218 (Springer, 2009).
Benjamini, Y., Drai, D., Elmer, G., Kafkafi, N. & Golani, I. Controlling the false discovery rate in behavior genetics research. Behav. Brain Res. 125, 279–284 (2001).
Article CAS PubMed Google Scholar
Falconer, D. S. Introduction To Quantitative Genetics, 4th Edn (1996).
Lynch, M. & Walsh, B. Genetics and Analysis of Quantitative Traits (Sinauer, 1998).
Kasprzyk, A. BioMart: driving a paradigm change in biological data management. Database 2011, bar049 (2011).
Article PubMed PubMed Central Google Scholar
McLaren, W. et al. The Ensembl variant effect predictor. Genome Biol. 17, 122 (2016).
Article PubMed PubMed Central Google Scholar
de Magalhães, J. P. et al. The Human Ageing Genomic Resources: online databases and tools for biogerontologists. Aging Cell 8, 65–72 (2009).
Article PubMed Google Scholar
Kim, W., Underwood, R. S., Greenwald, I. & Shaye, D. D. OrthoList 2: a new comparative genomic analysis of human and Caenorhabditis elegans genes. Genetics 210, 445–461 (2018).
Article CAS PubMed PubMed Central Google Scholar
Fraser, A. G. et al. Functional genomic analysis of C. elegans chromosome I by systematic RNA interference. Nature 408, 325–330 (2000).
Article ADS CAS PubMed Google Scholar
Võsa, U. et al. Large-scale cis- and trans-eQTL analyses identify thousands of genetic loci and polygenic scores that regulate blood gene expression. Nat. Genet. 53, 1300–1310 (2021).
Article PubMed PubMed Central Google Scholar
Pilling, L. C. et al. Human longevity is influenced by many genetic variants: evidence from 75,000 UK Biobank participants. Aging 8, 547–560 (2016).
Article CAS PubMed PubMed Central Google Scholar
Pilling, L. C. et al. Human longevity: 25 genetic loci associated in 389,166 UK biobank participants. Aging 9, 2504–2520 (2017).
Article CAS PubMed PubMed Central Google Scholar
Elsworth, B. et al. The MRC IEU OpenGWAS data infrastructure. Preprint at bioRxiv https://doi.org/10.1101/2020.08.10.244293 (2020).
Deelen, J. et al. A meta-analysis of genome-wide association studies identifies multiple longevity genes. Nat. Commun. 10, 3669 (2019).
Article ADS PubMed PubMed Central Google Scholar
Cerezo, M. et al. The NHGRI-EBI GWAS Catalog: standards for reusability, sustainability and diversity. Nucleic Acids Res. 53, D998–D1005 (2025).
Article CAS PubMed PubMed Central Google Scholar
Burgess, S., Zuber, V., Valdes-Marquez, E., Sun, B. B. & Hopewell, J. C. Mendelian randomization with fine-mapped genetic data: choosing from large numbers of correlated instrumental variables. Genet. Epidemiol. 41, 714–725 (2017).
Article PubMed PubMed Central Google Scholar
Auton, A. et al. A global reference for human genetic variation. Nature 526, 68–74 (2015).
Article ADS PubMed PubMed Central Google Scholar
Purcell, S. et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet. 81, 559–575 (2007).
Article CAS PubMed PubMed Central Google Scholar
Williams, S. M. et al. Expectations for papers performing Mendelian randomization analyses. PLoS Genet. 21, e1011767 (2025).
Article CAS PubMed PubMed Central Google Scholar

Download references

Acknowledgements

We thank B. Charlesworth for comments on this work while in preparation; and A. Finn, M. Abdul Wahid, S. Hasib, T. Y. Gracen, A. Fong, M. Rawet, P. Walker, S. Wijekantha, L. Beck, P. Livingstone, H. Rankin and W. Bardsley for help in QTL mapping, coding and quality control. This work was supported by grants from the NIH (R01AG043930 and R01AG070913) (R.W.W.); NIH R01AG075813 (D.G.A.); the University of Tennessee-Oak Ridge National Laboratory Governor’s Chair in Computational Genomics for genotyping, and the University of Tennessee Center for Integrative and Translational Genomics for data processing (R.W.W., D.G.A., D.A., L.L., S.R., P.P., A.G.C. and Z.S.); NIH grants NIA U01AG025707 and U01AG022308 (D.E.H.); U01AG022303 and OAIC P30AG024824 (R.A.M.); U01AG022307, P30-AG13319 and U01AG025707 (R.S.); R01GM123489 (P.P., S.S., K.W.B. and R.W.W.); École Polytechnique Fédérale de Lausanne, European Research Council (AdG-787702), Swiss National Science Foundation (310030B160318), AgingX Swiss Initiative for Systems Biology RTD 2013/153 (J.A.); NIH NIDA P30DA044223 (S.S. and R.W.W.); R01AG043930, NIH R21AG055841 and R56AG066625 (K.M.).

Author information

Authors and Affiliations

Department of Applied Sciences, Northumbria University, Newcastle upon Tyne, UK
Danny Arends
Department of Genetics, Genomics and Informatics, University of Tennessee Health Science Center, Memphis, TN, USA
David G. Ashbrook, Suheeta Roy, Lu Lu, Zachary Sloan, Arthur G. Centeno, Pjotr Prins, Amandeep Bajwa, Khyobeni Mozhui & Robert W. Williams
Institute for Neuroscience and Cardiovascular Research, The University of Edinburgh, Edinburgh, UK
David G. Ashbrook
Department of Entomology and Plant Pathology, University of Tennessee Knoxville, Knoxville, TN, USA
Kurt H. Lamour
Department of Inflammation and Ageing, University of Birmingham, Birmingham, UK
João Pedro de Magalhães
Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison, Madison, WI, USA
Karl W. Broman
Department of Preventive Medicine, University of Tennessee Health Science Center, Memphis, TN, USA
Saunak Sen & Khyobeni Mozhui
Ludwig Princeton Branch, Princeton University, Princeton, NJ, USA
Sarah J. Mitchell
Lewis-Sigler Institute for Integrative Genomics, Princeton University, Princeton, NJ, USA
Michael R. MacArthur
Department of Health Sciences and Technology, ETH Zurich, Zurich, Switzerland
Özlem Altintas Akin
Laboratory of Integrative Systems Physiology, Institute of Bioengineering, École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland
Xiaoxu Li & Johan Auwerx
Department of Surgery, University of Tennessee Health Science Center, Memphis, TN, USA
Amandeep Bajwa
Barshop Institute for Longevity and Aging Studies, University of Texas Health San Antonio, San Antonio, TX, USA
Vivian Diaz, Randy Strong & James F. Nelson
The Jackson Laboratory, Bar Harbor, ME, USA
David E. Harrison
Luxembourg Centre for Systems Biomedicine, University of Luxembourg, Esch-sur-Alzette, Luxembourg
Evan G. Williams
Department of Pathology and Geriatrics Center, University of Michigan, Ann Arbor, MI, USA
Richard A. Miller

Authors

Danny Arends
David G. Ashbrook
Suheeta Roy
Lu Lu
Zachary Sloan
Arthur G. Centeno
Kurt H. Lamour
João Pedro de Magalhães
Pjotr Prins
Karl W. Broman
Saunak Sen
Sarah J. Mitchell
Michael R. MacArthur
Özlem Altintas Akin
Xiaoxu Li
Amandeep Bajwa
Vivian Diaz
David E. Harrison
Randy Strong
James F. Nelson
Khyobeni Mozhui
Johan Auwerx
Evan G. Williams
Richard A. Miller
Robert W. Williams

Contributions

D.A., D.G.A., J.A., R.A.M., K.M., S.R., S.S., E.G.W. and R.W.W. were responsible for conceptualization. ITP colony management and informatics were directed by V.D., D.E.H., R.A.M. and R.S. Tissue and DNA sample processing and data curation were managed by D.A., D.G.A., A.B., A.G.C., V.D., K.H.L., L.L., R.A.M., J.F.N. and S.R. Statistics, bioinformatics and assembly of web services and databases were conducted by D.A., D.G.A., K.W.B., A.G.C., X.L., J.P.d.M., P.P., S.R., S.S., Z.S., E.G.W. and R.W.W. Code for mapping of UM-HET3 was developed by D.A., K.W.B., P.P., S.S. and Z.S. Mendelian randomization analysis was performed by J.A. and X.L. C. elegans analysis was performed by Ö.A.A., M.R.M. and S.J.M. Proofing was performed by J.A., R.A.M. and R.W.W. Manuscript, illustrations, tables and supplementary materials were prepared by D.A., J.A., M.R.M., E.G.W. and R.W.W. All authors reviewed the manuscript.

Corresponding authors

Correspondence to Danny Arends or Robert W. Williams.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature thanks Gary Churchill, Joris Deelen and Shripad Tuljapurkar for their contribution to the peer review of this work. Peer review reports are available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Sex-stratified and combined maps of mean lifespans of survivorships.

a,b,c, This extends the content of Fig. 1d to include female and male survivorship maps across the entire range of survivorships. Other conventions as in Fig. 2d. The LOD threshold is defined by the horizontal dashed lines for each survivorship. Note that only every third survivorship is plotted, accounting for some Vita loci that do not reach significance here.

Extended Data Fig. 2 Vita loci genotype effect plots.

Trios of effect size plots with standard error bands as described in detail in the legend to Figs. 1 and 2. Each trio for the 29 Vita loci includes combined data for both sexes (left), for females only (middle), and for males only (right). To understand the genetic sources that account for timing changes in mortality rates it is helpful to compare these more complex genotype plots to pairs of haplotype plots in Extended Data Figs. 3, 4.

Extended Data Fig. 3 Vita loci haplotype effect plots.

There are two trios of effect size plots for each Vita locus with standard error bands. The upper trio (blue and purple shades) give the difference in actuarial effects of the maternal haplotypes—C and B. The lower trio (pink and beige shades) give the difference in effects of the paternal haplotypes—H and D. These plots are symmetric and easier to interpret than those in Extended Data Fig. 2. However, they generally have lower LOD scores.

Extended Data Fig. 4 Mortality rates of maternal and paternal haplotypes of Vita loci.

Relative mortality rate difference between the maternal pair of haplotypes (C and B on the top of each panel) and between the paternal pair of haplotypes (H and D on the bottom of each panel). Here we have used a LOESS smoother with a relatively short α span 0.2. At any one age only the haplotype with the higher rate of mortality is shown, accounting for abrupt reversals.

Extended Data Fig. 5 Direct comparisons of all methods of display of survivorship effect sizes of genotypes, haplotypes, and mortality rates of a Vita14b in females (left) and males (right).

a,b, The actuarial genotype effect plots with standard error bands of Vita14b (see Fig. 2b) for both sexes. c,d, Corresponding haplotype-specific plots. g,h, Age-dependent relative mortality rate differences using a LOESS smoother with a span over the entire range of ages (α of 1) that averages mortality difference at a high level. While effects are similar between sexes, only the much stronger male effect in b reaches significance—the sum of the two reinforcing haplotype effects. i,j. An analysis of age-dependent differences in mortality rates using a LOESS with an α span of 0.2 that exposes much finer details of both age-dependent and haplotype-dependent differences in mortality. Extended Data Fig. 4 uses only this finer-grained smoother.

Extended Data Fig. 6 Dynamics and variance of 29 Vita loci as a function of sex and survivorship.

a,b, Variance contributions of Vita loci were estimated for each survivorship and sex: males in a and females in b with intensity of colors indicating the approximate fraction of variance explained, where white = 0–1%, faint = 1–2%, light = 2–4%, medium 4–8%, strong 8–16%, dark 16–32%. Non-genetic experimental sources of variance (Ve) are provided in green for males and purple for females. Drug is variance attributable to multiple nominally ineffective supplementary drug treatments versus the standard chow diet. Cohort is variance attributable to the nine annual cycles of production of UM-HET3 mice from late spring through late fall between 2004 and 2013 (2008 was a hiatus year). Site is variance associated with the three ITP sites: The Jackson Laboratory, The University of Michigan, and University of Texas Health San Antonio.

Extended Data Fig. 7 Body Mass loci at five ages.

a, Female maps. b, Male maps. c, Combined maps in which all significant Mass loci have been labeled. Note that three of the 28 Mass loci are more distinct in the female map (Mass6a, Mass6b, and Mass11a) than in combined or male maps. Mass11b is more distinct in the male map. At all ages the mapping model compensates for drug treatment, cohort year, and site, and in the case of the combined sexes also adjusts for sex. Compare (c) with Fig. 5a, but note that color assignments are different here. d, Table of sample sizes for each sex and for the sites. Animals at 42 days were only weighed for the first four cohorts, accounting for the lower sample size. The lower sample size at 730 days is due in part to mortality.

Extended Data Fig. 8 Soma actuarial effect size plots.

a, An overlay of all 30 Soma loci and their full actuarial effect size plots at five ages at which body weights were measured as in Fig. 4i,j. This PDF should be opened in Adobe Illustrator or Inkscape to preserve the separation of the 30 layers (file also available at files.genenetwork.org/current/umhet3_2025/). With control of layers users can examine one or more of the 30 Soma loci, in the same way as in Fig. 4i–l but at all five ages at which animals were weighed. Soma1a is displayed in panel a and in the matched layer (upper left). Soma19b is displayed in panel ad and the matched layer. Actuarial correlation values are given in 15-day steps from the five ages at which mice were weighed. All technical details are provide in Fig. 4i–l.

Extended Data Fig. 9 Epistatic interaction plots for the T₄₂, T₃₆₅, T₇₄₀ and T₉₀₅ survivorships.

a,g, Overview of all epistatic interactions in survivorship using the same conventions. Chromosomes are labeled with abbreviated Vita and Soma symbols. Color and type of lines define partnership types (orange lines are Vita-Soma pairs). i,j, An overlay of all four survivorships that mainly highlights the greater cumulative numbers of male than female interactions. This pair of circle plots is a set of overlapping layers useful to directly compare different survivorships and the Adobe Illustrator format is available upon request. Number of epistatic links for both males and females is given in the lower right corner of each female plot. Few epistatic interactions overlap in both sexes. All values are given in Supplementary Table 12.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Arends, D., Ashbrook, D.G., Roy, S. et al. Dynamics of genetic and somatic trade-offs in ageing and mortality. Nature (2026). https://doi.org/10.1038/s41586-026-10407-9

Download citation

Received: 06 August 2025
Accepted: 12 March 2026
Published: 22 April 2026
Version of record: 22 April 2026
DOI: https://doi.org/10.1038/s41586-026-10407-9