Sex-independent senescence in a cooperatively breeding mammal

Abstract

1. Researchers studying mammals have frequently interpreted earlier or faster rates of ageing in males as resulting from polygyny and the associated higher costs of reproductive competition.
2. Yet, few studies conducted on wild populations have compared sex-specific senescence trajectories outside of polygynous species, making it difficult to make generalized inferences on the role of reproductive competition in driving senescence, particularly when other differences between males and females might also contribute to sex-specific changes in performance across lifespan.
3. Here, we examine age-related variation in body mass, reproductive output and survival in dominant male and female meerkats, Suricata suricatta. Meerkats are socially monogamous cooperative breeders where a single dominant pair virtually monopolizes reproduction in each group and subordinate group members help to rear offspring produced by breeders.
4. In contrast to many polygynous societies, we find that neither the onset nor the rate of senescence in body mass or reproductive output shows clear differences between males and females. Both sexes also display similar patterns of age-related survival across lifespan, but unlike most wild vertebrates, survival senescence (increases in annual mortality with rising age) was absent in dominants of both sexes, and as a result, the fitness costs of senescence were entirely attributable to declines in reproductive output from mid- to late-life.
5. We suggest that the potential for intrasexual competition to increase rates of senescence in females—who are hormonally masculinized and frequently aggressive—is offset by their ability to maintain longer tenures of dominance than males, and that these processes when combined lead to similar patterns of senescence in both sexes.
6. Our results stress the need to consider the form and intensity of sexual competition as well as other sex-specific features of life history when investigating the operation of senescence in wild populations.

1 INTRODUCTION

Across vertebrate and invertebrate taxa, there is striking variation in the extent to which rates of senescence differ between males and females (Austad, 2011; Barford & Dorling, 2006; Carroll & Sherratt, 2017). One of the more common arguments put forward to explain this diversity relates to variation in the intensity of reproductive competition. Life-history theories of senescence predict that any increase in the allocation of resources to reproduction in early life occurs to the detriment of reproductive performance and somatic maintenance in later life, translating into an earlier onset, and faster rate of senescence in the sex experiencing increased reproductive competition (as per the life-history theories of senescence—the antagonistic pleiotropy and disposable soma theories; Austad & Hoffman, 2018; Kirkwood, 2017; Williams, 1957). Because sex differences in reproductive competition are often related to the mating system of a species, the comparison of senesence trajectories across mating systems has provided a useful framework for exploring some of the key predictions of the life-history theories of senescence. For example, in polygynous, sexually dimorphic mammals, where males fight frequently and display energetically costly traits that improve fighting success or help to monopolize access to females, it is often males that have shorter lifespans and show higher rates of survival (‘actuarial’) senescence than females (Clutton-Brock & Isvaran, 2007; Loison, Festa-Bianchet, Jullien, Jorgenson, & Gaillard, 1999; Promislow, 1992; Toïgo & Gaillard, 2003). In contrast, in monogamous taxa, where levels of intrasexual competition are closer to parity, sexual dimorphism in longevity and rates of senescence are often absent, or much reduced (Allman, Rosin, Kumar, & Hasenstaub, 2002; Brownikowski et al., 2011; Clutton-Brock & Isvaran, 2007).

Yet, not all studies that have looked for sex differences in polygynous species have found them (Tidière et al., 2014, 2015; Toïgo & Gaillard, 2003), and more generally, the evidence for a direct role of sexual selection on rates of ageing is equivocal (Bonduriansky, Maklakov, Zajitschek, & Brooks, 2008; Graves, 2007; Maklakov, Bonduriansky, & Brooks, 2009). This is likely because males and females can compete in different ways, the traits involved can entail different costs, and might also be expressed at different stages of lifespan (Clutton-Brock, 1983; Ralls & Mesnick, 2009; Stockley & Bro-Jørgensen, 2011; Tompkins & Anderson, 2019). In this context, it is important to appreciate that sex differences in survival rates and senescence depend on the characteristics of the two sexes which, in turn, depend on the specific selection pressures generated by intraspecific competition in the two sexes rather than on sex differences in reproductive variance (Clutton-Brock, 1983). As a result, polygynous breeding systems where reproductive variance is higher in males than females will not necessarily generate higher rates of mortality and senescence in males in all species. Conversely, there are likely to be monogamous species where contrasts in the life histories of males and females generate sex differences in survival and rates of senescence even if there are no sex differences in variance in breeding success. Furthermore, there are other differences between the sexes that are not directly connected to the intensity of reproductive competition, such as sex differences in parental care (Allman et al., 2002), heterogamy (Marais et al., 2018), or maternal transmission of the mitochondrial genome (Beekman, Dowling, & Aanen, 2014; Zeh & Zeh, 2005), which may contribute to sex differences in senescence and further obscure patterns related to the mating system.

Existing data from alpine marmots Marmota marmota are particularly useful for highlighting the importance of life-history features for sex-specific patterns of senescence in the wild. Alpine marmots are socially monogamous, but males nonetheless face high costs of territoriality (Arnold, 1990) and are frequently challenged by out-of-group males, whereas females can largely suppress challenges from females within the group (Cohas, Yoccoz, Da Silva, Goossens, & Allainé, 2006). Consequently, male alpine marmots encounter greater costs of reproductive competition and this has been used to explain why male body mass deteriorates in later life, but female body mass shows no sign of senescing (Tafani et al., 2013). The fact that males also experience greater energetic costs of hibernation could also be a factor (Arnold, 1988). However, despite senescence in male body mass, there is no clear sex difference in the intensity of survival senescence in this species, raising the possibility that male body mass declines influence fitness by acting through reproduction rather than survival (Berger et al., 2016). How commonly other non-polygynous mammals in the wild display similar sex-specific patterns of senescence is unknown as most information on monogamous taxa comes from birds or uses data from captive populations. In addition, most studies of sex differences in senescence have focussed on mortality data (Gaillard, Garratt, & Lemaître, 2017), but to understand the mechanisms underlying age-related changes in fitness in naturally regulated populations, it is important to examine age-related changes in other biological parameters that are associated with individual performance and indicate why the sexes differ. This might include age-specific changes in reproductive effort (Lemaître & Gaillard, 2017), body mass and condition (Hämäläinen et al., 2014; Tafani et al., 2013), immune function (Beirne, Waring, McDonald, Delahay, & Young, 2016), and haematological parameters (Jégo et al., 2014), amongst others (Nussey, Froy, Lemaître, Gaillard, & Austad, 2013). From such work, it has become increasingly clear that different fitness-related traits can display divergent age-related trajectories within individuals (Evans, Gustafsson, & Sheldon, 2011; Hayward et al., 2015; Nussey et al., 2009), and are not necessarily closely related to each other (Bouwhuis, Choquet, Sheldon, & Verhulst, 2012). These observations contradict Williams’ prediction that senescence ‘should always be a generalized deterioration’ (1957; see Gaillard & Lemaître, 2017) and highlight the need to consider a wider range of life histories to better understand the operation of senescence in natural populations.

In this study, we examine sex-specific patterns of senescence across a variety of traits in a wild, naturally regulated population of meerkats, Suricata suricatta, a socially monogamous, cooperatively breeding mongoose. In meerkats, reproduction is monopolized by a dominant male and a dominant female in each group (Clutton-Brock & Manser, 2016) with subordinate individuals of both sexes helping to protect and feed juveniles born in the group. Only a small proportion of individuals ever acquire dominance (Duncan, Gaynor, & Clutton-Brock, 2018; Spong, Hodge, Young, & Clutton-Brock, 2008), with those that are unsuccessful in doing so experiencing an increasing mortality risk that is associated with extended periods of time spent away from the natal group beyond 1.5 years of age (Cram et al., 2018). As a result, subordinate individuals have shorter longevities than dominants so that the potential to detect senescence in subordinate individuals is limited (Sharp & Clutton-Brock, 2010).

While both sexes display similar high levels of reproductive monopoly, differences in the nature of intrasexual competition in meerkats generate a larger reproductive skew in females than in males (Clutton-Brock et al., 2006). Dominant females display higher circulating levels of testosterone (Davies et al., 2016; a trait associated with increased senescence rates elsewhere, Brooks & Garratt, 2016), and also show high levels of female–female aggression, often prompting dominant females to evict subordinates and commit infanticide of non-descendant kin (Clutton-Brock & Manser, 2016). Dominant females are also highly fecund and it is not unusual for them to produce three litters in a single calendar year (Clutton-Brock & Manser, 2016). Together, these characteristics are predicted to be associated with higher rates of senescence in females (see Sharp & Clutton-Brock, 2011a), reversing the sexual dimorphism typically observed in polygynous mammals. However, while vacant positions of dominance prompt intense female–female aggression, once dominance has been acquired, dominant females are unlikely to be displaced because they can evict subordinate females before they have reached an age and mass at which they become serious competitors (Duncan et al., 2018), and because the costs of challenging are prohibitively high (Sharp & Clutton-Brock, 2011b). By contrast, dominant males experience a consistent risk of displacement across their tenure from immigrant males (Spong et al., 2008) and disperse from their natal group should the incumbent dominant female die. Consequently, the duration of effective breeding (tenure length) is, on average, shorter in males and than females. In other mammals, sex differences in the duration of effective breeding are correlated with sex differences in life expectancy (Clutton-Brock & Isvaran, 2007), and so the the reduced duration of effective breeding in male meerkats could partly offset the costs of intrasexual competition on females and cause sex differences in senescence to be closer to parity.

To investigate sex-specific patterns of senescence in meerkats, we examined age-related changes in body mass, reproductive output and annual survival in individuals that acquired dominance within their lifetime. We then used the estimates for age-related changes in reproduction and survival to calculate the fitness costs of senescence for dominant meerkats (i.e. the difference between observed reproductive value and the hypothetical reproductive value if senescence were not occurring). In so doing, we add to the small number of mammal species in which such metrics have been calculated (Bouwhuis et al., 2012; Kowald & Kirkwood, 2015), and provide the first information from a cooperative breeder. Lastly, we examined whether the reproductive declines we detected in one sex contributed to the reproductive declines shown by the other sex (Fay, Barbraud, Delord, & Weimerskirch, 2016; Lemaître & Gaillard, 2017). Such contributions of partner age are likely to be particularly strong in monogamous species like meerkats where multiple mating is limited and where partners can be paired for long periods, potentially leading to correlated senescence in males and females. If, for example, males and females tend to pair up and then remain together into late life, then the poorer quality of both males and females in late life might amplify declines in reproductive performance. If, on the other hand, there is no underlying relationship between male and female age in later life, this would imply that the processes underlying reproductive senescence relate to intrinsic features of the individuals in question.

2 MATERIALS AND METHODS

2.1 Study population

Data were collected from a wild population of meerkats in the Kalahari region of South Africa (26°58′S, 21°49′E) between December 1996 and September 2018. Approximately 50 social groups were followed during this time, comprising over 2,500 individuals habituated to observation at short distance (<2 m). Individuals were tagged with a subcutaneous transponder chip at emergence and given a unique fur dye mark to allow identification. Groups were visited 3–4 days a week, enabling intensive sampling of life-history information (births, deaths, evictions, babysitters, changes in pregnancy or dominance status), behaviour and body mass. Body mass records were obtained early in the morning prior to foraging by enticing individuals onto an electronic balance with crumbs of boiled egg as a reward. As stated above, subordinates that never acquire a dominance position but remain in the study population until their death have usually died before they reach 3 years of age. Moreover, for a further sizeable proportion of subordinate individuals, we do not know their fate (i.e. death or emigration), meaning that we cannot control for possible effects of selective disappearance, an important source of between-individual variation that could lead to biased estimates of within-individual trait changes across lifespan (van de Pol & Verhulst, 2006). The same bias is not present in dominant individuals, whose fate is usually known. For these reasons, we limited our analyses to individuals that acquired a position of dominance during their lifetime, and of these individuals, we only considered those with a confirmable death. Confirmable deaths included individuals that were last seen in a state of terminal decline or were euthanized on site because they had developed clear outward signs of morbidity linked to advanced-stage tuberculosis.

2.2 Statistical analyses

Statistical analyses consisted of three steps. In the first step, we examined age-related variation in body mass, reproductive output and mortality, with models parameterized so that we could directly compare sex differences in ageing trajectories. In the second step, we extended the best supported models from step one to incorporate effects of partner age and thereby examine whether senescence declines were conflated by partner effects. In the third step, we used information on age-related changes in reproduction and survival to calculate the fitness costs of senescence (Bouwhuis et al., 2012; Kowald & Kirkwood, 2015). All analyses were undertaken in the r statistical environment v3.6.0 (R Core Team, 2019). Estimates present the mean ± 1 SE of the mean unless otherwise stated.

2.2.1 Age-related variation in body mass and reproductive output

The body mass dataset partitioned individual lifespans into 4-month periods, with body mass calculated as the mean daily morning mass within each period. This allowed us the highest possible resolution of sampling without losing full periods of weights due to pregnancy. Specifically, we excluded any weights taken during pregnancy by back-casting 70 days from the day of birth or litter loss, thus removing any weight increases due to gestational growth (Figure S1). On average, this resulted in 24.4 ± 0.6 mass records/female/period and 43.1 ± 0.7 mass records/male/period. In total, the body mass dataset comprised 83 females and 53 males (total n = 1,431 periods).

For the reproductive output dataset, we instead partitioned individual lifespans into 6-month periods. Preliminary analyses suggested that doing so reduced the number of zeroes in the dataset and therefore improved the fit of models compared to shorter time intervals. Reproductive output was defined as the number of offspring that were produced by a male or female within each 6-month period that survived to nutritional independence at 3 months of age. 78.0% of pups that survive to nutritional independence go on to reach adulthood at 1 year of age (n = 2,040 pups between January 1994 and July 2018). Parentage was assigned through genetic analysis of 18 microsatellites derived from tissue samples taken from the tip of individuals tails (Nielsen, 2012), and where genetic data were missing, maternity could be inferred from field observations where we were certain only a single female had given birth. The reproductive output dataset comprised 95 females and 67 males that produced an average of 1.67 ± 0.11 and 0.721 ± 0.10 pups/6-month period, respectively (total n = 980 periods).

Age-related variation in many vertebrate traits often takes the form an initial early-life increase, a mid-life plateau and a later-life decline. To capture this pattern for body mass and reproductive output in meerkats, we fitted a series of mixed effects models that included chronological age either as a quadratic function (a linear and quadratic age term) or as a threshold function (usually where linear slopes are estimated on either side of a fitted threshold age). Threshold functions are generally better equipped to reliably recover the full age-dependence of trait change but do so at the expense of additional parameters. We implemented our models in a Bayesian framework using the brms package (Bürkner, 2018). This offers a distinct advantage over a frequentist treatment, for while the former generates a posterior distribution for threshold parameters upon which other model terms are conditioned, frequentist analyses must fit multiple models and secondarily estimate the position of any thresholds (with associated confidence) through likelihood profiling (Ulm, 1989). We modelled body mass using a Gaussian error distribution, and reproductive output using a zero-inflated negative binomial distribution with a single zero-inflation parameter applied to all observations (zi ~ 1).

To test for sex differences in ageing patterns, we adopted a similar approach to Tompkins and Anderson (2019) and fitted six models for each trait (Table S1). In model 1, age was included as a quadratic function, and males and females were assumed to follow the same age trajectory. In model 2, the linear and quadratic age terms of model 1 were each interacted with a covariate for sex to allow for male and female age trajectories to differ. We then specified four forms of threshold model. For body mass, this included two thresholds for each sex, one in early life (first threshold age = T_SEX,1) and one in mid- to late-life (second threshold = T_SEX,2). For reproductive output, we only fit a single threshold in mid- to late-life as preliminary model fitting found no evidence for an additional threshold in early life. Threshold models then differed in the extent to which they forced males and females to have sex-specific slopes on age across lifespan, and/or sex-specific thresholds (Table S1). In the most advanced threshold model for body mass, model 6 (from which other models were derived), different threshold ages and different slopes across age were parameterized for males and females, such that the body mass of meerkat individual i at age j was parameterized as:

where for thresholds T_SEX,t, (Age_ij − T_SEX,t)₊ = (Age_ij − T_SEX,t)*I(Age_ij ≥ T_SEX,t). I(Age_ij ≥ T_SEX,t) is an indicator function equalling 1 when Age_ij ≥ T_SEX,t, and 0 otherwise. Thus, β₄ and β₅ are the difference in the slope of each response variable on age after the first threshold age relative to the slopes before the first threshold age (for females and males, respectively). The step function ‘switches’ β₄ and β₅ terms off for ages ≤ T_SEX,1 and on for ages > than T_SEX,1 in each sex, where sex is a dummy variable with females coded as ‘0’ and males as ‘1’. Additional population-level ‘fixed’ effects (β_kX_k…) included the age at first dominance (AFD), AFD:sex, longevity, longevity:sex, total rainfall, group size, season and dominance status. Total rainfall during each period was calculated from onsite rain gauge data. On days with missing information (10.3%) we imputed rainfall values from a remote-sensing dataset provided by the NASA GES DISC (Goddard Earth Sciences Data and Information Services Centre). Group size was taken as the average daily number of group members > 6 months old in each time period. Season was coded as a two-level factor (first = Oct–Mar, second = Apr–Sep). Dominance status indicated whether an individual was dominant within the period in question (recall that all individuals in the dataset do become dominant at some point in their lifetime). In the body mass models, we also included two predictors to examine possible sex differences in terminal decline (TD, TD:sex), where TD was a two-level factor noting whether it was the last period of an individual's life. Finally, u_i and u_group represent group-level (‘random’) effects of individual identity and group identity, and ε_ij is the residual error.

All Bayesian mixed effects models were fitted with four chains of 3,000 iterations, of which 2,000 were dedicated to the warm-up. We chose normal priors for all population-level and group-level effects. We also set upper and lower bounds on the prior for the threshold ages to more efficiently sample the posterior (T_SEX,1: lower bound = 1, upper bound = 4; T_SEX,2: lower bound = 4, upper bound = 8). Model diagnostics and posterior predictive checks highlighted adequate mixing of chains and appropriate choice of priors and error distributions. All continuous parameters were z-score transformed prior to model fitting, apart from age. Within each model, we used 95% Bayesian credible intervals (BCI) drawn from the posterior distribution as a measure of uncertainty, deeming as biologically important any effects where the credible intervals did not overlap zero. The predictive ability of candidate models was compared using k-fold cross-validation (‘k-Fold IC’) using subset number k = 10. This method divides the data into 10 subsets and validates the results of the nine subsets for each missing dataset. For each model, we also calculated the Bayesian equivalent of the R² using bayes_R2 function in brms (Gelman, Goodrich, Gabry, & Ali, 2017).

2.2.2 Partner age effects

To assess the extent to which reproductive declines in one sex might contribute to or partly explain reproductive declines in the other sex, we modelled the relationship between the age of a dominant female and the age of her male partner across her period of tenure, and vice versa. Partner age was fitted as the response variable in general additive models (gam), with the age of the focal individual included as a sex-specific smoother function in each model (6 knots). As this preliminary analysis hinted at a linear increase in partner age with the age of the focal dominant (Figure S2; female model edf = 1, F₁ = 31.93, p < .001; male model edf = 1, F₁ = 2.54, p = .114), we re-fitted the best supported reproductive output model from the above analysis (but this time excluding information from subordinates), and included a linear covariate for partner age, and an interaction between partner age and sex. To allow for a non-linear effect of partner age on reproductive output, we also fitted one further model with a quadratic partner age effect, and an interaction between the quadratic term and sex. We examined the influence of including partner age terms on the estimates of reproductive decline in male and female meerkats.

2.2.3 Age-related changes in mortality

To test for sex differences in longevity and survival across age, we performed both semi-parametric and parametric survival modelling with the survival and flexsurv packages (Jackson, 2016; Therneau, 2015). Our sample consisted of 98 females and 71 males with confirmed final fates, 63 females and 92 males who disappeared during the study with their fate being unknown, and 9 females and 7 males who were still alive at the end of the study. Individuals of unknown fate, who either disappeared during the study or were still alive at the end of study sampling period, can still be incorporated into the analysis through censoring. However, two key assumptions of censoring are that it is random with respect the individuals affected, and independent of the process of mortality such that individuals do not experience a change in mortality risk due to being censored. This is unlikely to be the case in meerkats, where censorship represents either unobserved mortalities or individual dispersal events, both of which will introduce positive bias and lead to overestimation of survival unless accounted for. To investigate the effect of censoring bias on our estimates of longevity and survival in males and females, we performed a sensitivity analysis using the InformativeCensoring package (Ruau et al., 2016) by re-running semi-parametric cox proportional hazard models while either increasing or decreasing the hazard that censored individuals are exposed to after censorship via the gamma imputation method (Jackson et al., 2014).

As the length of our study (21 years) is considerably longer than the oldest individual of known fate within our population (12.4 years), the possible bias introduced by excluding individuals censored during our study is expected to be negligible (and far less than the bias introduced by their inclusion). Therefore, when performing parametric survival models to characterize the pattern of survival senescence, our dataset only included individuals of known fate and individuals alive at the end of the study (who were censored). Null models were fitted with various error distributions (Gompertz, exponential, log-logistic, log-normal, gamma and Weibull) and model selection was guided by AIC and visual inspection of predicted survival and hazard plotted against the raw data. Males and females were first modelled independently to confirm their survival patterns could be best modelled with the same error distribution. We then included a model with both sexes, with a sex term fitted to all parameters of the error distribution. For comparison, these analyses were repeated with datasets without truncation of censored individuals and where all individuals censored prior to study end were considered as unobserved mortalities and thus modelled as known deaths.

2.2.4 Age-related reproductive value and the fitness costs of senescence

We calculated age-related variation in reproductive value (RV) using predicted changes in reproductive output and survival across lifespan. Reproductive value was calculated according to Stearns (1992):

where is the age for which reproductive value is being calculated, w is the age at last reproduction, l_x is survival at age x and m_x is reproductive output at age x. Further details are provided in the Supporting Information.

Finally, we used information on age-related changes in reproduction and survival to quantify the fitness costs of senescence (Bouwhuis et al., 2012; Kowald & Kirkwood, 2015). Because annual survival probability in meerkats was constant beyond the age of peak reproductive output (at age 5.4 for both males and females), the fitness costs of senescence could be entirely attributed to reproductive senescence. We calculated the costs of reproductive senescence, C_RS, as the difference between the estimated reproductive value at age 1 (to be consistent with Bouwhuis et al., 2012), and the hypothetical reproductive value at this age if reproductive declines were absent. For the latter, reproductive output was held constant from the age of peak reproductive output and RV was estimated as above. C_RS is then [(RV_observed – RV_{no RS})]/RV_{no RS} × 100%.

3 RESULTS

3.1 Age-related variation in body mass

Both males and females displayed a significant decline in body mass in later life (Figure 1a). Model comparisons highlighted that age-related variation in body mass was best described by threshold models that partitioned lifespan into three stages: an early-life increase, a mid-life plateau and later-life senescence (Table 1). In the best fitting model (model 3), males and females were parameterized to share common slopes and common threshold ages, indicating that the onset and rate of senescence were independent of sex (Figure 1a; Table 2). Specifically, the onset of senescence in body mass was estimated at 5.56 years for both sexes (95% BCI = 5.10–6.13), after which point males and females lost 19.35 grams per year (95% BCI = −26.36 to −12.35). The absence of sex differences in body mass senescence was reinforced by the most parameterized threshold model (model 6, Δk-Fold IC = 17.1), where early-life and late-life thresholds, and the slopes on age, all displayed similar estimates in males and females (Table S4). Aside from their age-related changes in body mass, both sexes displayed a terminal decline in their final period of life equating to 32.07 g in females (95% BCI = −40.95 to −23.08) and 30.85 g decline in males (95% BCI = −41.53 to −17.75, Figure 1b). At the between-individual level, in neither sex was there strong statistical support for the selective disappearance of lighter individuals (female estimate = 7.47, 95% BCI = −3.19 to 18.01; male estimate = 3.84, 95% BCI = −14.51 to 14.76), nor was there a clear influence of the age of dominance acquisition (Table S2).

Figure 1

(a) Age-related variation in body mass in female (red) and male (blue) meerkats after accounting for the effects of selective disappearance, terminal decline, social status, group size, rainfall and season. Solid circles represent the mean residual body mass per age with their associated standard error bar (with sex differences in the intercept removed from the residuals to allow visualization of sex differences in average body mass). Solid lines display predicted age-related changes in body mass according to the best-supported model. The upper and lower limits of the coloured shaded areas show the 95% Bayesian credible intervals (BCI) estimates of the chronological age effect based upon fixed effects uncertainty. The 95% BCI of the threshold estimates are shown by the horizontal error bars, with males and females sharing a single estimate for both thresholds. The population-level mean body mass is 710.5 g. (b) Terminal declines in body mass of female (red) and male (blue) meerkats in the last 3 months of life. Solid points display the predicted body mass decline, with the 95% BCI represented by the vertical lines

Table 1. Comparison of models investigating sex specificity of senescence in meerkats. Models are ranked according to k-fold IC, with the lowest k-fold IC taken as the best-supported model (bold). Threshold models differed in the extent to which they allowed males and females to have common or distinct sex-specific estimates for threshold ages (T) and/or slopes on age (as detailed in the main text). ‘params’ refers to the number of parameters estimated by each model. All models included additional population-level (‘fixed’) and group-level (‘random’) terms as described in the main text

Model	Sex- and age-related predictors	Body Mass				Reproductive Output
		params	k-fold IC (SE)	Δ k-fold IC (SE)	Bayesian R^2(95% BCI)	params	k-fold IC (SE)	Δ k-fold IC (SE)	Bayesian R^2(95% BCI)
Quadratic models
1	Age + Age^2 + Sex	16	14,857.4 (71.1)	93.8 (25.8)	.766 (0.753, 0.778)	16	2,351.2 (92.6)	31.2 (11.6)	.377 (0.293, 0.460)
2	Age + Age^2 + Sex + Sex:Age + Sex:Age^2	18	14,847.7 (71.3)	84.1 (25.0)	.768 (0.755, 0.779)	18	2,331.6 (92.6)	11.6 (12.0)	.372 (0.283, 0.454)
Threshold models
3	Common T, common slopes on age	19	14,763.6 (74.5)	0.0	.783 (0.771, 0.794)	17	2,338.3 (91.2)	18.3 (10.4)	.378 (0.292, 0.462)
4	Sex-specific T, common slopes on age	21	14,785.7 (75.0)	22.1 (13.6)	.784 (0.772, 0.794)	18	2,334.6 (90.5)	14.6 (9.0)	.380 (0.292, 0.462)
5	Common T, sex-specific slopes on age	22	14,766.6 (75.2)	3.0 (15.2)	.784 (0.771, 0.795)	19	2,320.0 (90.9)	0.0	.370 (0.284, 0.456)
6	Sex-specific T, sex-specific slopes on age	24	14,780.7 (76.0)	17.1 (14.4)	.784 (0.772, 0.794)	20	2,339.4 (92.0)	19.4 (8.2)	.374 (0.288, 0.457)

Table 2. Threshold age estimates and estimated slopes of age for body mass and reproductive output from the best-supported model in each case (models 3 and 5, respectively). 95% BCI shown in parentheses. Estimates for the slopes of reproductive output are on the link-scale (log link). Supporting information provides equivalent terms from the most heavily parameterized model for comparison

Trait	Model	Early-life slope	Threshold 1 (early-life to mid-life)	Mid-life slope	Threshold 2 (mid-life to late life)	Late-life slope
Female and male body mass	3	80.21 (69.81, 91.26)	2.20 (2.05, 2.42)	11.11 (0.96, 21.09)	5.81 (5.32, 6.24)	−19.35 (−26.25, −12.53)
Female reproductive output	5	NA	NA	0.23 (0.10, 0.36)	5.44 (4.87, 6.06)	−0.41 (−0.68, −0.18)
Male reproductive output	5	NA	NA	0.53 (0.31, 0.75)	5.44 (4.87, 6.06)	−0.45 (−0.80, −0.15)

3.2 Age-related variation in reproductive output

Males and females both experienced age-related declines in reproductive output (Table S3; Figure 2a,b). In the best supported model (model 5), males and females shared a common threshold age term, but were given separate slopes on age which took the form of a linear increase from mid-life to late life, followed a subsequent period of reproductive decline. Further examination of the model output highlighted that it was the initial mid-life increase in reproductive output that drove this trend, with estimates for the later-life slope showing no apparent difference between males and females (Table 2). As for body mass then, both the onset and rate of senescence in reproductive output were independent of sex (Figure 2). Based upon the best supported model, females experienced a 72.1% reduction in reproductive output between the ages of 5 and 9, and males a 74.3% decline in reproductive output over the same age period. The best supported model also highlighted the selective disappearance of females with lower reproductive output, with a 1 standard deviation increase in the longevity term (2.45 years) being associated with 1.55 more pups per 6-month period (mean estimate on log-scale = 0.44, 95% BCI = 0.11 ‒ 0.77, Table S3). A comparable trend was not present in males (mean estimate on log-scale = 0.29, 95% BCI = −0.22 to 0.84).

Figure 2

Age-related variation in reproductive output in female [red, (a)] and male [blue, (b)] meerkats. Solids lines display the predicted number of pups produced to emergence during each 6-month period for individuals who acquire dominance. The upper and lower limits of the coloured shaded areas show the 95% Bayesian credible intervals (BCI) estimates of the chronological age effect based upon fixed effects uncertainty. The 95% BCI of the threshold estimates are shown by the horizontal error bars. Predictions were made either side of the population mean age of dominance acquisition period (vertical striped line), with predictions representing subordinate reproductive output before this age and dominant reproductive output after. For predictions, rainfall and group size were set at the mean, longevity was set at 7 years, and season was set as ‘second’ (April–September). Points display the raw data with vertical error bars indicating ±1 SEM. Points are coloured to emphasize that, on average, most of the raw data prior to the vertical line come from individuals while they are subordinate (dark grey), whereas after this point, most data come from individuals that are dominant (red/blue)

Although analyses of the raw data provided some suggestion that older dominant individuals were more likely to be paired with an older partner (Figure S2), the effect was weak and the inclusion of partner age in a re-fitted reproductive output model (model 5) did not affect the estimated onset or rate of senescence when compared to a model where partner age terms were absent (Figure S3). Nor were the partner age terms themselves significant in the updated models (additional model one: partner age estimate = 0.03, 95% BCI = −0.16 to 0.26, partner age:sex estimate = −0.20, 95% BCI = −0.63 to 0.25; additional model two: partner age² estimate = −0.02, 95% BCI = −0.17 to 0.14, partner age²:sex estimate = −0.28, 95% BCI = −0.61 to 0.05).

3.3 Age-related changes in mortality

While a cox proportional hazard model assuming independent censoring suggests that males have marginally longer lifespans than females (estimate = −0.342, SE = 0.158, p = .030), this result is not robust to expected censoring bias. Our sensitivity analysis revealed that even under the conservative assumption that censored individuals are exposed to only a small increase in mortality risk compared to non-censored individuals, the sex difference in lifespan no longer held (Figure 3). Moreover, when we excise individuals that were censored before the end of the study, or if we treat them as having immediately died, no sex difference was apparent (Figure 3). In the scenario where censoring is associated with reduced risk, the effect of sex remains stable.

Figure 3

The change in the estimated effect size of the sex term (solid blue line gives the mean, blue shading the 95% confidence intervals) in relation to the hazard adjustment applied to censored individuals, as estimated by cox proportional hazard survival models. An increased log hazard of zero (vertical solid line) represents the standard model where censorship is assumed to be independent of mortality and individuals that are censored are expected to experience no change in mortality risk. Where confidence intervals cross zero (orange dashed line), the effect of sex is not significant. Point estimates and accompanying confidence intervals are plotted for the extreme scenarios where individuals that are censored become immune to mortality (Immune) and where censorship leads to instantaneous mortality (Death). Additionally, the point estimate and confidence intervals are plotted for a dataset where individuals that disappear during the study are truncated and individuals still alive at the end of study are censored with no adjustment to their log hazard (Truncate); this is the dataset used for down-stream parametric survival modelling

Parametric modelling of our survival data revealed that the pattern of survival in meerkats was best described by a log-normal distribution (Figure S4). The log-normal distribution model was in the top cohort of candidate models for both sexes and represented the model of best fit with both sexes modelled together (Table S5; Figure S4). The log-normal distribution allows mortality risk to initially increase with the risk reaching an asymptote later in life (Figure 4c). This suggests the absence of survival senescence in meerkats with the log-normal model providing a better fit than models with distributions that can capture survival senescence should it be present, such as the Gompertz and Weibull distributions (Table S5). However, as sample sizes decrease later in life, the power with which to detect senescence declines. Annual mortality derived from our log-normal survival model tracks mortality probabilities well with reasonable sample sizes up to around 8 years (Figure 4a,b), after which we are unlikely to be able to detect senescent trends. While the parametric models revealed no difference in the log mean parameter between the sexes (estimate = 0.015, 95% CI = −0.137 to 0.167), indicating no difference in mean longevity, there was a marginally significant difference between the sexes in the log standard deviation parameter (estimate = −0.264, CI = −0.481 to −0.047), reflecting the reduced variance in male lifespan.

Figure 4

(a, b) Annual mortality curves, (c) hazard rates and (d) reproductive value across lifespan for male (blue) and female (red) meerkats that acquired dominance at some point in their lives. (a, b) Solid lines represent annual mortality probabilities calculated for males (a) and females (b) using survival probabilities predicted from the parametric survival model of best fit (Figure S4), with points displaying raw values for annual survival probability—the size of point represents the total number of individuals observed across the year (log scaled) with exact values reported at the top of the figures. (c) Solid lines represent the estimated hazard rate from our model of best fit with the 95% confidence intervals plotted as shaded areas. Estimates are derived from a parametric survival model with a log-normal distribution and sex fitted as a covariate to the ancillary log standard deviation parameter but not the log mean parameter. (d) Reproductive values are calculated using predicted estimates of survival (Figure S4) and reproductive output (Figure 2)

3.4 Age-related reproductive value and the fitness costs of senescence

Prior to the acquisition of dominance reproductive value increased gradually, peaking at around 2.5 years in females and 4.5 years for males, before declining thereafter (Figure 4c). With no evidence of survival senescence in dominant individuals, these declines in reproductive value—the fitness costs of senescence—can be entirely attributed to reproductive senescence. Excluding reproductive senescence from the life history increases the reproductive value of females at age 1 from 12.68 to 14.49, and males at age 1 from 7.22 to 8.25. This entails a fitness cost of reproductive senescence, C_RS, of 12.5% in females, and 12.6% in males.

4 DISCUSSION

Our study finds that in meerkats, the form and rate of senescence across three components of life history are similar in males and females. We show that the onset and rate of senescence in body mass and reproductive output were largely independent of sex, with trait values peaking between 4 and 6 years of age and declining at similar rates thereafter. Age-related survival probability was also unaffected by sex, but unlike the former two traits, we found no evidence of survival senescence in dominant individuals as annual survival probability remained constant beyond the age of peak reproductive output. Consequently, the fitness costs of senescence in meerkats could be entirely attributed to reproductive deterioration in later life, contributing to an estimated reduction in reproductive value of approximately 12.5% when compared to the hypothetical reproductive value at 1 year of age if no senescence were to occur. These results agree with a wider body of literature which has highlighted that senescence in wild populations is often asynchronous across traits (Evans et al., 2011; Hayward et al., 2015; Nussey et al., 2009), though the extent to which this is the case varies widely across taxa and even among species with apparently similar ecologies (Bouwhuis et al., 2012).

Alongside research on grey mouse lemurs Microcebus murinus (Hämäläinen et al., 2014), alpine marmots (Tafani et al., 2013), and red Canis rufus and grey wolves Canis lupus (MacNulty, Smith, Mech, & Eberly, 2009; Sparkman et al., 2017), our study is one of only a handful to provide quantitative information on sex-specific patterns of senescence in a non-polygynous mammal in the wild, and is the first to combine information from multiple traits concurrently. Such a treatment is timely, as the predominance of long-term studies focussed on polygynous mammals has led to the conclusion that divergent ageing rates between the sexes are typically driven by sex differences in the intensity of intrasexual competition (Beirne, Delahay, & Young, 2015; Lemaitre, Gaillard, Pemberton, Clutton-Brock, & Nussey, 2014; Nussey et al., 2009). However, while this assertion might hold generally, we should not expect this to be the case in all species (Lemaître & Gaillard, 2012; Tidière et al., 2014) as there are likely to be other aspects of demography and life history that are also important in affecting how males and females allocate resources to somatic maintenance, survival and reproduction across the lifespan.

In meerkats, we suggest that the form of reproductive competition is as important as its intensity in affecting patterns of senescence. If intensity were more important, we would have expected dominant females to display an earlier onset and faster rate of senescence than dominant males, as dominant females are hormonally masculinized (Davies et al., 2016), show elevated parasite burdens (Smyth & Drea, 2016) and display regular bouts of aggression with subordinates to suppress their reproduction and prompt their eviction (Clutton-Brock et al., 2006). On the other hand, females can better manage the risk of usurpation and can maintain long tenures of dominance, whereas dominant males are exposed to the periodic threat from intruding males seeking to challenge their paternity share and their dominance. As a result, although the frequency with which males face reproductive competition is much lower than that faced by females, the implications for their tenure maintenance, and thus their continued survival are more severe, as reflected in their shortened tenures. Taken together, we suggest that the potential for intrasexual competition to increase rates of senescence in females is offset by their ability to maintain longer tenures of dominance than males, and that these processes combined lead to similar patterns of senescence in both sexes. Or, put differently, the realized costs of competition on fitness are not divergent enough to have led to the evolution of sex differences in senescence trajectories in meerkats.

We found that body mass and reproductive output senesced in parallel. For female meerkats, the fitness consequences of reduced body mass have already been well described (Ozgul, Bateman, English, Coulson, & Clutton-Brock, 2014), making it likely that the downturns in body mass are causally related to decreases in reproductive output through reductions in litter size and in the frequency of breeding (Sharp & Clutton-Brock, 2010). For males, the consequences of reduced body mass on fitness are less clear. We did not find any support for the selective disappearance of lighter males (that had acquired dominance), but it is possible that males in poorer condition in later life are less able to monopolize the paternity of the dominant female's pups and maintain their position of dominance. In making this suggestion, it must also be remembered that the reproductive output of any male is in part influenced by the quality of his female partner, and vice versa (Fay et al., 2016; Lemaître & Gaillard, 2017), such that the fitness declines in one sex could contribute heavily to fitness declines in the other sex. However, we do not find convincing support for partner age effects in our study, and by implication it is likely that intrinsic physiological declines in males and females are mostly responsible for the observed reproductive declines in either sex.

Our study also detected a strong age-independent contribution to body mass variation in the form of terminal declines. Terminal declines in condition are often thought to reflect age-independent deterioration due to disease (Coulson & Fairweather, 2001). In meerkats, tuberculosis represents a possible mechanism by which terminal declines might be mediated, and anecdotally, individuals reaching an advanced stage of morbidity through tuberculosis exposure display dramatic reductions in body mass, dying shortly afterwards. The telomeres of meerkats also shorten rapidly in the period immediately prior to death (Cram et al., 2018). A broader examination of physiological changes across the lifespan would no doubt identify further markers of bodily decline associated with age-dependent and age-independent mass variation (e.g. muscle wasting, Sierra et al., 2013), but it would be particularly useful to know whether reductions in body mass compromise foraging efficiency. A large proportion of the daily activity budget of meerkats is spent digging for subterranean invertebrates so any downturn in foraging ability is likely to be particularly damaging for individual condition.

Despite undergoing body mass senescence, neither male nor female meerkats showed evidence for increasing rates of mortality in later life. The absence of survival senescence contrasts with the general pattern seen in mammals and birds (Gaillard et al., 2017; Jones et al., 2014), though our result should be taken with the caveat that survival analyses were restricted to individuals that acquired dominance. That said, several recent studies of unusually long-lived species such as bats and seabirds have suggested that survival senescence in wild vertebrate populations is sometimes negligible (Coulson & Fairweather, 2001; Fleischer, Gampe, Scheuerlein, & Kerth, 2017). These species are also characterized by low fecundities imposed by energetic constraints (e.g. through flight: Jones & MacLarnon, 2001; or chick development: Lack, 1968), and with these constraints on reproductive output, lifespan extension might provide the main evolutionary route to maximizing fitness. Any survival senescence is therefore likely to carry heavy fitness costs in long-lived species, and this could provide strong selection against senescence. By comparison, meerkats are not particularly long-lived for their size, and their fecundity is high, so similar arguments are unlikely to explain the absence of survival senescence in dominant meerkats (or naked mole-rats Heterocephalus glaber: Ruby, Smith, & Buffenstein, 2018). Why then does mortality rate not increase in old age in meerkats? One possible explanation relates to group living. Incumbent dominants can maintain long tenures sheltered from extrinsic mortality in large groups with dedicated sentinels (Cram et al., 2018), and this buffering effect of group living might be enough to prevent age-dependent increases in mortality risk in dominants irrespective of individual declines in condition. Alternatively, the presence and intensity of survival senescence might vary over time as environmental and demographic conditions modify the likelihood of different forms of intrinsic and extrinsic mortality (Berger, Lemaître, Allainé, Gaillard, & Cohas, 2018; Hämäläinen et al., 2014; Ronget, Garratt, Lemaître, & Gaillard, 2017), and might go undetected when these sources of variation are not accounted for in survival analyses. A failure to detect survival senescence could also reflect sampling effort when longitudinal studies have not been run for long enough to capture its onset (Péron, Gimenez, Charmantier, Gaillard, & Crochet, 2010). With only modest amounts of data from very long-lived individuals, it is not currently feasible to investigate whether this is the case in meerkats with a high degree of confidence—but our data do provide reasonable evidence that general increases in mortality rate in the reproductive cohort of meerkats are absent several years after senescence has already begun in reproduction and body mass (see also Sharp & Clutton-Brock, 2010, 2011a).

The realization that males and females differed little in their annual survival probably was only arrived at through an appreciation of censoring. The independence of censoring is a critical assumption of most standard survival models but is often violated in studies of wild animals (Murray, 2006). Meerkats provide a case in point where censoring is not independent and is instead associated with missed or increased mortality, and models not taking this into account are subject to positive biases that tend to overestimate longevity. The level of bias produced by non-random censoring is related to the proportion of individuals censored (Campigotto & Weller, 2014), and as proportionally more males than females were censored in our dataset, we pick up a spuriously significant sex difference in longevity when censoring bias was not accounted for. While it is generally impossible to assess the extent to which the baseline hazards of censored individuals change, using sensitivity analyses one can examine how variation in baseline hazard influences specific covariates of interest (Jackson et al., 2014) and use this to guide interpretation of any results. In our case, even a minor increase in hazard for censored individuals (a reasonably conservative assumption) lead to a loss of significance and a reduction in the effect size of the sex term on longevity, a result that was also found when individuals that disappeared prior to study end were truncated rather than censored. Thus, with censoring-induced bias considered, we do not find strong support for differences in longevity between the sexes.

Irrespective of the causes of reproductive declines in dominant meerkats, direct fitness represents only one avenue of reproductive success in those individuals experiencing senescence: dominants can also accrue indirect fitness benefits through the reproduction of subordinates in their group. In stable groups, the costs of subordinate reproduction to dominants are high (Bell et al., 2014), and dominant females consequently employ behavioural tactics to limit the breeding opportunities of subordinates. Even so, these costs are likely to be lower when their own reproductive output is reduced, as in old age. Older, lighter dominants are presumably also less able to control subordinate reproduction should they attempt to do so. Either way, the reproductive output of subordinate group members may increase as dominant females' age and therefore cause indirect fitness to form a greater contribution to the reproductive success of dominants in later life. If indirect fitness were shown to increase in old-aged individuals, there are two important consequences. First, it would imply that the fitness costs of reproductive senescence are overestimated when only direct fitness is considered. Second, it might explain why dominant individuals appear to favour survival over reproduction in later life, as their own reproductive declines could be partly offset by the greater reproductive success of their close relatives.

In concert with other studies, our results emphasize that broad categorization into mating systems will likely only get us so far in understanding sex differences in ageing in wild vertebrates, for within mating systems and within species, sex differences in the degree to which males and females compete for reproductive opportunities, and the manner in which they do so, vary widely. To better understand why ageing rates differ so widely within and between species in the wild, and in particular, between the sexes, it will therefore be necessary to generate more targeted questions that place specific aspects of species’ life histories at the forefront of tests of evolutionary theories of senescence.

ACKNOWLEDGEMENTS

We thank Jean-Michel Gaillard, Jean-François Lemaître and one anonymous reviewer for previous comments on the manuscript. We are grateful to the Kalahari Research Trust and the Kalahari Meerkat Project for access to facilities and habituated animals in the Kuruman River Reserve, South Africa. The long-term research on meerkats is currently supported by a European Research Council Advanced Grant (no. 742808) to T.C.-B. and by grants from the University of Zurich to M.B.M. and the MAVA foundation. J.T. was funded by a Natural Environment Research Council (UK) Doctoral Training programme during the inception of the work. S.P.S. was funded by the Leverhulme Trust and the Isaac Newton Trust. We would like to thank the Northern Cape Department of Environment and Nature Conservation for permission to conduct the research (FAUNA 1020/2016), Dom Cram and Mark Dyble for comments on an earlier version of the manuscript, and numerous volunteers, for their historic and continued work with the meerkats in the Kalahari.

AUTHORS' CONTRIBUTIONS

J.T., C.D. and T.C.-B. conceived of the ideas for the work with earlier input from S.P.S., J.T. and C.D. collated and analysed the data. A first draft was put together by J.T. and T.C.-B., with all authors contributing to subsequent drafts. All authors gave their approval for publication. T.C.-B. and M.B.M. initiated and organized the long-term data collection.

Open Research

DATA AVAILABILITY STATEMENT

Data available from the Dryad Digital Repository https://doi.org/10.5061/dryad.n5tb2rbrv (Thorley et al., 2019).