Early life adversity has sex-dependent effects on survival across the lifespan in rhesus macaques

Abstract

Exposure to early life adversity is linked to detrimental fitness outcomes across taxa. Owing to the challenges of collecting longitudinal data, direct evidence for long-term fitness effects of early life adversity from long-lived species remains relatively scarce. Here, we test the effects of early life adversity on male and female longevity in a free-ranging population of rhesus macaques (Macaca mulatta) on Cayo Santiago, Puerto Rico. We leveraged six decades of data to quantify the relative importance of 10 forms of early life adversity for 6599 macaques. Individuals that experienced more early life adversity died earlier than those that experienced less adversity. Mortality risk was highest during early life, defined as birth to 4 years old, but heightened mortality risk was also present in macaques that survived to adulthood. Females and males were affected differently by some forms of adversity, and these differences might be driven by varying energetic demands and dispersal patterns. Our results show that the fitness consequences of early life adversity are not uniform across individuals but vary as a function of the type of adversity, timing and social context, and thus contribute to our limited but growing understanding of the evolution of early life sensitivities.

This article is part of the discussion meeting issue ‘Understanding age and society using natural populations’

1. Introduction

Adversity such as food shortages and social isolation experienced prior to adulthood can result in long-term health and evolutionary fitness consequences in a wide range of insects, birds, fish, reptiles and mammals [1–4]. These detrimental outcomes in adulthood are hypothesized to arise from organisms adjusting their developmental trajectories in response to adversity in order to improve immediate survival [3,5,6]. To date, research on non-human animals has borrowed hypotheses and methodologies from the extensive literature on early life adversity in humans. Early life adversity in humans is associated with poorer health and reduced longevity [7–9]. A common and growing approach in the study of humans is to use cumulative indices, which measure the total amount of adversities experienced by an individual in their early life, rather than focus on different forms of adversity separately [10–13]. Empirical evidence suggests that the accumulation of multiple adversities is a better predictor of adult outcomes than any particular form of adversity, but there is also evidence that specific forms of adversity lead to different outcomes [11,14,15]. A small but growing number of studies have tested the long-term impacts of early life adversity, but further research on how cumulative early life adversity and different forms of adversity shape the timing of fitness consequences in a variety of species, populations and contexts is needed to better understand the evolution of early life sensitivities to adversity.

In long-lived species, females that live longer have a longer reproductive span and are able to produce more offspring, such that longevity or survival can be used as proxies of fitness [16–21]. Long-term studies of female yellow baboons (Papio cynocephalus) and spotted hyenas (Crocuta crocuta) have shown that exposure to greater amounts of cumulative early life adversity is associated with reduced survival in adulthood [14,15,22,23]. In addition to the adverse effects experienced in adulthood, mortality prior to reproductive maturity results in a fitness of zero as organisms fail to reproduce. Accordingly, in populations characterized by high mortality rates prior to reproductive maturity, the total number of offspring reaching reproductive maturity is the best proxy for fitness [24]. Male and female mountain gorillas (Gorilla beringei beringei) exposed to more cumulative early life adversity experience reduced survival prior to reproductive maturity but do not experience survival costs after maturity [25]. As such, adult survival patterns as a function of cumulative early life adversity in gorillas differ from those in yellow baboons and spotted hyenas, but pre-reproductive survival patterns are not yet available across species for comparison. More studies across different species are thus needed to draw comparisons and better understand the evolutionary pressures that shape early life sensitivities to cumulative adversity and different forms of putative adversity, and to test the relationships between developmental responses to adversity, the timing of fitness consequences across the lifespan and detrimental adult outcomes.

The fitness consequences of early life adversity might vary in a sex-dependent manner owing to differences in life history strategies [26]. During adverse early life conditions, the sex with more energetically demanding traits is predicted to be more susceptible to nutritional constraints [27,28]. In many species, male life histories are considered more energetically costly given faster growth and larger body size compared to females. When male fitness is largely determined by access to mates via competitive ability, males should also invest in costly developmental processes such as play and motor skill development [29]. Support for these predicted differences comes from a study focusing on one form of adversity: maternal death prior to weaning in red deer (Cervus elaphus) was linked to higher mortality risk among male compared to female offspring [30]. Sex-dependent effects of early life adversity are challenging to study because many species are characterized by sex-biased dispersal such that pre- or post-dispersal data are typically missing for individuals of the dispersing sex (e.g. [15]). More studies are thus needed that can follow both males and females from birth until death to investigate how life history strategies shape sex-specific fitness consequences and developmental responses to early life adversity.

Here, we study the free-ranging rhesus macaques (Macaca mulatta) of Cayo Santiago to advance our understanding of the magnitude, form, timing and sex dependence of early life adversity effects. To date, direct evidence of the lifelong effects of early life adversity is relatively scarce in long-lived species owing to the difficulties in collecting data from early life until death. In this study, we leveraged complete demographic records that extend back to the late 1950s for thousands of male and female macaques at the Cayo Santiago field site. We add to the growing body of early life adversity research in long-lived species [14,15,22,23,25,31,32]. Following these previous studies, we examined both the effects of cumulative early life adversity and the effects of specific forms of early life adversity to identify which forms best predict survival. We build upon this previous work by incorporating more forms of adversity and leveraging an exceptionally large sample size. Further, this analysis incorporates data from females and males from birth until death, which has only been previously examined in this population and one study of wild mountain gorillas [25,32]. Specifically, we examined the effects of 10 forms of potential early life adversity on sex-specific mortality risk across early life and, separately, across adulthood. By examining mortality across the lifespan, we can identify the full extent of variation in the fitness consequences of early life adversity, including how survivorship biases in early life may influence survival patterns observed in adulthood.

Previous research has demonstrated that some forms of early life adversity can shape the behaviour and health of macaques. In captivity, infant macaques experimentally exposed to abusive mothers or reared with peers show decreased immune function, more self-injury, more impulsivity, differential DNA methylation in brain cells and a suite of physical health anomalies in adulthood [33]. Such experimental studies are complemented by targeted observational studies that have examined the effects of one or two forms of naturally occurring adversity. In the Cayo Santiago population, abusive maternal care behaviour is linked to differences in hypothalamic–pituitary–adrenal (HPA) axis function in juveniles, the presence of a competing younger sibling is linked to reduced survival during juvenility, and exposure to hurricanes and high population density during early life are linked to life history trade-offs in adulthood [34–36].

As for wild rhesus macaques, the Cayo Santiago monkeys live in naturally forming multi-male, multi-female social groups characterized by dominance hierarchies and male dispersal. This species is sexually dimorphic, with males exhibiting larger body mass and canine length than females [37,38]. Males queue for dominance rank, have large testes and experience strong indirect male–male competition, including sperm competition [39–41]. We predicted that rhesus macaques exposed to greater amounts of early life adversity would have increased mortality risk. We predicted mortality risks to be more severe during the first 4 years of life when individuals are still growing and adversity is more recent. However, we also predicted that heightened mortality risk would persist into adulthood among those surviving past 4 years of age. Given that male life history strategies in this species [42] prioritize costly traits like faster growth, larger body size and motor skill development [37,38,43], we predicted that early life adversity would exert larger effects on males than females. We also predicted that some forms of early life adversity, such as maternal loss, would have larger effects on mortality risk than others.

2. Methods

(a). Study site and population

We studied a free-ranging population of rhesus macaques living on Cayo Santiago, a 15.2 ha island off the southeastern coast of Puerto Rico. The current population of ~1700 individually recognized rhesus macaques living in 12 social groups is descended from 409 monkeys that were transported from India to the island in 1938. This population is managed by the Caribbean Primate Research Center (CPRC) of the University of Puerto Rico. Monkeys are fed monkey chow daily, and water catchments provide ad libitum access to drinking water.

During the study period (1960–2021), observers monitored and recorded demographic events daily. These records included births, deaths, sex, matriline, matriline rank, maternal identification, sires when genetic data were available and group emigration and immigration events. The island is free of predators and, as approved by the Institutional Animal Care and Use Committee (IACUC), there is no regular veterinary intervention, such that the primary causes of death are injury and illness [44]. Injured individuals are three times more likely to die than uninjured individuals in the population [44]. We have had few obvious outbreaks of infectious diseases, an exception being the 1940 and 2010 outbreaks of Shigella [45,46]. If a carcass is found, then animals are confirmed as dead and removed as per the date the carcass is retrieved. The daily census notes when animals were last seen. If animals have not been seen by anyone for 6 months then they are assumed dead, with the death date recorded as the last date of observation plus 1 day. A genetic pedigree is available for much of the population [47]. Daily total rainfall and mean maximum temperature data were obtained from the NOAA station in Rio Piedras, Puerto Rico. Over a 61-year period (1959–2020), data were not recorded by this NOAA station for 21% of days. Rather than removing this large portion of data, we imputed missing rainfall and temperature data using the ‘mice’ package (v. 3.14.0) in R [48]. This study included 6599 individuals for which there were complete data available, including data covering the entire lifespan—birth to death—for 2513 macaques. The remaining 4086 macaques were either alive (n = 173) at the time of this study or were removed (n = 3913) from the island prior to natural death as a result of population management (i.e. constituted right-censored samples). Over the course of the study, three forms of selection criteria for animal removal have been used: removal of entire social groups, removal based on sex (e.g. removal of adult males to maintain appropriate sex ratios) and removal based on age (e.g. removal of immature individuals) [49]. Note that these selection criteria are independent of the early life adversity variables used in our analysis.

(b). Early life adversities

We used historical demographic records to assess individual exposure to early life adversity. We considered 10 forms of potential early life adversity. Here, we aim to select experiences or inputs that limit resources available to an individual or otherwise lead an individual to allocate their resources across life history traits in a suboptimal manner. In other words, early life adversity is an experience that diverts an individual away from their optimal developmental trajectory. Eight of these forms of adversity have been used in previous studies of primates [15,35,50], while two additional forms of adversity were chosen based on the natural history of the present study system (hurricanes and high temperatures). Each form of adversity is further justified below, with previous research demonstrating its deleterious consequences (see electronic supplementary material, table S1 for further details regarding empirical support). In choosing time periods of exposure for each form of adversity, we followed the methods presented by Tung & colleagues [15] in regard to wild baboons, which wean and reach reproductive maturity at similar ages to rhesus macaques. Each of these 10 adversities could be considered an early life advantage from the flipped perspective because the measures can be considered on an axis that ranges from ‘adversity’ to ‘advantage’ [51]. For example, monkeys born into small maternal social networks are predicted to experience adversity, while monkeys born into large maternal networks are predicted to experience advantage. For consistency with previous work, we recorded when individuals experienced the ‘adversity’ end of the axis for each variable.

(i). Maternal loss

Maternal death increases offspring mortality in humans and other mammals [15,30,52–55]. We considered an individual to experience maternal loss if their mother died [including natural death owing to causes such as disease or injury (n = 1165) and permanent removal from the population (n = 299)] before the individual reached 4 years of age [15,56]. This 4-year window includes the period during which young macaques are nutritionally dependent on their mothers, and the period during which young macaques are weaned but still socially dependent on their mothers. While a mother’s removal from the population had to occur while the offspring was alive (i.e. prior to offspring death if they have died) to be considered an adversity, this was not a requirement for natural maternal death because an impending maternal death is linked to offspring mortality risk—an association likely explained by poor maternal condition [56]. We do not know the cause of death in most cases, but most deaths on the island are owing to illness, injury and old age, and we would consider all or most natural deaths to be condition dependent. Our decision to include cases in which the mother died within 4 years after the offspring’s birth was based on previously established methodology used by Zipple & colleagues [56]. In their study of several primate species (chimpanzees (Pan troglodytes), northern muriquis (Brachyteles hypoxanthus), blue monkeys (Cercopithecus mitis), mountain gorillas, yellow baboons, capuchins (Cebus capucinus) and Verreaux’s sifakas (Propithecus verreauxi)], offspring death was significantly associated with an impending maternal death, even if the mother’s death did not occur until 3–4 years after the offspring’s birth [56]. Further, offspring did not have a higher mortality risk if their mother’s death was more imminent (within 1 year of birth) versus more delayed (3.5–4 years after birth) [56]. In our study, there were 220 maternal death cases in which the focal individual died before their mother died. The delay in death was as short as 2 days and as long as 3.8 years (median = 358.5 days; mean = 470 days). About 73% of the delayed maternal deaths occurred within 2 years of the infant death. Maternal loss was measured as a binary variable: experienced maternal loss or did not experience loss.

(ii). Competing sibling

The presence of a younger sibling that is close in age represents a source of competition over maternal resources and is associated with higher mortality risk [15,34]. We considered a sibling to be a competitor if the sibling was born within 355 days of a subject, which represented the bottom quartile of interbirth intervals (IBIs) in our sample. While short IBIs have been linked to high juvenile mortality risk in the study population [34], in other primates, short IBIs can also be an indicator of high maternal quality [50,57], complicating the interpretation of results. The presence of a older sibling that is close in age could also constitute a cost for the younger sibling, although there are further complexities for this metric. In this study population, short and long preceding IBIs are predictive of higher infant mortality [34], and the presence of older siblings could represent an important source of social support [58]. Here, we followed previous work that focused on the presence of younger siblings that were close in age (e.g. [15,25]). Last-born offspring and individuals that died before their sibling was born could not experience this adversity. The presence of competing siblings was measured as a binary variable.

(iii). Group size

High group size and high population density are indicative of more competition and are associated with reductions in fecundity [17,35,59,60]. We used group size as a proxy for within-group competition. Demographic records were used to construct group composition over the study period. Group size was defined as the number of adults (≥4 years of age) of both sexes in an individual’s social group on the day that individual was born [15] and it was included in our models as a continuous variable. Group size varied across the study period (range: 2–222 individuals) but was fairly consistent across individuals’ early lives (group size at birth and 4 years of age: Pearson’s r = 0.75, p‐value < 0.0005).

(iv). Primiparity

The high energetic demands on first-time mothers can result in negative outcomes for offspring such as increased mortality risk [52,61–65]. First-time mothers might struggle to provide care, social support and energetic resources for their offspring. Being born to a first-time mother might limit the energetic resources available for the developing individual and how those resources are allocated, which meets our definition of adversity. We used a binary measurement for primiparity: first born or not first born.

(v). Matriline rank

Dominance rank mediates access to food and is linked to survival, fecundity and offspring growth in primates [66–70]. Matrilineal dominance hierarchies for a given social group and year are recorded by the CPRC as categorical—high, middle and low—based on dyadic agonistic interactions recorded over the course of each observation year (e.g. threats, displacements and submissive behaviours) [34,71]. As such, this measure represents the matrilineal rank for an individual’s birth year. We follow previously established methods for characterizing rank [67,72]. Female macaques acquire the rank adjacent to their mothers, so individuals that are members of the same matriline (i.e. descendants of a shared crown female ancestor) tend to be adjacent in rank. The rank of a matriline can thus serve as a proxy for individual rank. Because matriline ranks are stable over time [73], if the rank of a matriline is not measured in some years, it can be extrapolated back or forward in time based on known ranks. As a point of methodological comparison, we also treated matrilineal rank as an ordinal variable (i.e. ‘0, 1, 2’ rather than ‘high, middle, low’). Since females inherit rank from their mothers, matriline rank might be considered a measure of experiences across the lifespan for females and a measure of early life experience for males. However, given previous research treating maternal rank as a form of early life adversity in females and males in similar social systems, we included it here for comparative purposes.

(vi). Kin network

Among prime-aged adult females at Cayo Santiago, the presence of more maternal kin is linked to better survival in any given year [72]. We measured an individual’s maternal kin network size at birth as the number of living females over 4 years of age with a relatedness coefficient of at least 0.063. This relatedness coefficient was chosen because it represents the threshold at which macaques in this population can recognize kin via vocalizations [74], and this threshold was used in previous work showing a positive association between the number of relatives present and adult survival [72]. Kin network size was included as a continuous variable (range: 1–21 individuals). The size of kin networks was fairly stable across individuals’ early lives (size at birth and 4 years: Pearson’s r = 0.75, p‐value < 0.0005).

(vii). Maternal social isolation

Greater social connectedness is associated with longer adult lifespans, lower age-specific mortality risks, the production of a greater number of offspring and longer offspring survival [75–80]. We used behavioural data collected during 10-min focal animal samples on adults in several social groups from 2010 to 2017 [details provided in the electronic supplementary material]. To measure maternal social isolation, we calculated a composite sociality index (CSI) using the affiliative social behaviours, approaches and grooming. For each mother in each year, we tabulated the rate of approaches (approaches to and from other adult females/hours observed) and the rate of grooming bouts (number of grooming bouts given and received during the hours of observation). A mother’s approach and grooming rates were divided by the mean rate for all adult females in each social group in each year. These standardized approach and grooming rates were added together and divided by 2 (the number of behaviours) to create the CSI for each mother. Here, we followed Tung and colleagues: for each offspring in our analyses, we averaged their mother’s CSI for the first 2 years of life [15].

(viii). Rainfall

More rainfall is indicative of greater food and water availability and is linked to greater fecundity and better survival in primates [81–83]. Because food and water are provisioned, low rainfall might not be as impactful in this population compared to wild primate populations in marginal environments, but we still predict low rainfall to be predictive of higher mortality risk. We also predict that negative outcomes could be associated with high rainfall, given the association between high rainfall and tropical storms. Here, we used total rainfall across the first year of life (range: 1021.4–3157.1 mm).

(ix). Temperature

Higher temperatures are linked to reduced cognitive performance in Western Australian magpies (Cracticus tibicen doralis), poorer health and welfare in dairy cattle (Bos taurus) and higher mortality risk in southern pied babblers (Turdoides bicolor) and geladas (Theropithecus gelada, but also increased fecundity in geladas [83–86]. Thus, we predicted negative outcomes associated with high temperatures. Here, we averaged mean maximum daily temperatures across the first year of life (range: 85.12–89.89 F).

(x). Hurricanes

In this population, exposure to major hurricanes is associated with accelerated age-related immune changes [87]. Among females on Cayo Santiago, hurricane exposure during early life is associated with delayed reproductive maturation but higher fertility during reproductive prime; also, during hurricane years, adult females are less likely to produce offspring that survive to 1 year old [35,88]. We recorded individual exposure to any of the three major hurricanes that had major impacts on Cayo Santiago (Hugo on 18 September 1989, Georges on 21 September 1998 and Maria on 20 September 2017) during the first year of life. We treated this as a categorical variable to assess the effects of each major storm. Hurricane exposure was not included in previous studies of early life adversity, so we chose the first year of life as our window of exposure to align with our other weather variables—rainfall and temperature. If individuals were exposed to hurricanes when they were over 1 year old, we did not consider them to have experienced this adversity as early life adversity.

Here, following previous theoretical and empirical work [11,13–15], we examined cumulative adversity measures and individual forms of adversity separately. To construct a cumulative early life adversity index, we summed individuals’ exposure to different forms of adversity. Previous studies typically relied on binary scores for each form of adversity. Following Patterson et al. [50], to avoid arbitrary categorization and to consider the severity of exposure, we used continuous measures of adversity when feasible. For purposes of the cumulative index, continuous measures (i.e. group size, kin network size, temperature and rainfall) were normalized so that values ranged from zero to one. Where lower variable values indicate greater adversity, such as for kin network size and rainfall, the normalized variables were reversed so that values closer to one (i.e. higher values) always indicate greater adversity. For binary and categorical measures (i.e. maternal loss, being a first born, presence of a competing sibling, matriline rank and hurricane exposure), individuals were assigned a value of one if they experienced a given form of adversity and a value of zero if they did not experience it. Those born into high-ranking matrilines were assigned a zero, mid-ranking matrilines were assigned 0.5 and low-ranking matrilines were assigned a value of one. Those exposed to any of the three major hurricanes during the first year of life were assigned a one, and those that were not exposed to any of these major hurricanes during the first year of life were assigned a zero. As such, each variable ranged from zero to one, and for each individual the variables were summed together into a cumulative index to represent the total exposure to early life adversity. Our main cumulative early life adversity index could range from 0 to 9 because it included nine variables: maternal loss, the presence of a competing younger sibling, high group size, primiparity, low matrilineal dominance rank, small kin network, hurricane exposure, high temperature and low rainfall. Maternal social isolation was not included in the cumulative adversity index because it relied on behavioural observations of individuals and was thus derived only for a subset of our data (n = 299 early life survival; n = 101 adult survival).

(c). Data analysis

To determine whether early life adversity predicts survival, we used survival models. The outcome variable was age at death. Individuals that were either still alive at the end of the study or removed from the island for population control were right-censored. We ran models on the full sample of all ages (n = 6599) but right-censored to 4 years old to examine early life mortality, and we ran models on a subsample of individuals that survived beyond 4 years of age (n = 2866) to examine mortality across adulthood. Early life adversity predictor variables were modelled in two ways: (i) cumulative index model, which included all forms of adversity summed together into one variable; and (ii) multivariate model, which included all nine forms of adversity (maternal social isolation was run separately), modelled as separate predictor variables in the same model. We assessed the efficacy of these two approaches by comparing the fit of the cumulative index models versus the multivariate models using information on the difference in the expected predictive accuracy. Models included early life adversity index or all individual adversity variables, sex and an interaction term between sex and early life adversity. Models also included a varying intercept for birth year and maternal identification. The different forms of adversity that we examined were not correlated, but temperatures during the first year of life were highly correlated with birth year (electronic supplementary material, table S3).

Analyses were run in R (v. 4.1.2 [89]) and RStudio (v. 1.4.1106 [90]). We first used Cox survival models, but the proportional hazards assumption in the Cox model was violated [cox.zph function in R package, ‘survival’ (v. 3.2.13): p < 0.05; see electronic supplementary material for specifics]. Instead, we fit accelerated failure time (AFT) survival models with a Weibull distribution. The presence of a competing younger sibling is time-dependent since individuals cannot experience this exposure unless they survive until a given age, i.e. until it is biologically possible for the mother to give birth again, which involves weaning the current offspring, cycle resumption, conception, and gestation. To include this variable, we would need to include it as a time-varying variable in a Cox proportional hazard model. Alternatively, one could subset the data to examine survival during periods in which it is biologically possible for a younger sibling to be born (e.g. 2–4 years); however, this could create biases in the dataset since it involves removing individuals based on their age at death. As such, we were unable to test how the presence of a competing younger sibling affects survival in early life, and we excluded this variable from the cumulative index for the early life survival model. We could, however, examine this in adulthood since all individuals in the sample survived to adulthood, and the presence of a competing younger sibling is not time-varying.

Genetics can contribute to the effects of early life adversity. For example, individuals experiencing maternal loss might have shorter lifespans owing to genes shared by both the mother and offspring. To estimate to what extent variance in survival is explained by genetics, we accounted for pedigree in a subsample of the data for which we had complete pedigree information (i.e., for 923 individuals during their early life and for 307 adults). To do so, we used the animal model [91] and incorporated the genetic relationship covariance matrix as a random effect. The effects of pedigree on survival were substantial, but accounting for genetic relatedness in the model did not diminish the effects of early life adversity on survival in early life (with pedigree: β = −0.33 ± 0.12; without pedigree: β = −0.31 ± 0.11; electronic supplementary material, table S4) or adulthood (with pedigree: β = −0.12 ± 0.04; without pedigree: β = −0.12 ± 0.04; electronic supplementary material, table S4). Some forms of adversity are probably dependent on genetics (i.e. maternal loss, competing sibling, matriline rank and kin network), whereas other forms of adversity are genetics-independent (i.e. rainfall, hurricanes, temperature, primarity and group size), although there is probably a genetic basis to how one responds to adversity in all cases. Because pedigree might differentially impact the effect of these genetics-dependent and genetics-independent adversities, we ran our pedigree models with cumulative adversity indices based only on genetics-dependent and genetics-independent adversities. The effects of pedigree on survival remained substantial, but including genetic relatedness in these models did not reduce the effects of adversity on survival during early life (genetics-independent: with pedigree β = −0.19 ± 0.08; without pedigree β = −0.11 ± 0.07; genetics-dependent with pedigree β = −0.40 ± 0.09; without pedigree β = −0.35 ± 0.08; electronic supplementary material, table S4) or adulthood (genetics-independent: with pedigree β = 0.03 ± 0.03; without pedigree β = 0.01 ± 0.02; genetics-dependent with pedigree β = −0.07 ± 0.03; without pedigree β = −0.08 ± 0.02; electronic supplementary material, table S4). Because the effects of early life adversity were unaffected by pedigree inclusion and because the sample size for pedigree inclusion was much smaller (paternity is unknown for many animals earlier in the study), we have presented the larger set of data without pedigree in the main text.

Models were run with the brms package (v. 2.16.3) [92]. All continuous predictor variables were standardized to a mean of 0 and a standard deviation of 1 (note that this differs from how we normalized the continuous adversities to range from 0 to 1 for addition into the cumulative early life adversity index). We ran our analyses with continuous variables modelled linearly and quadratically and compared the fit of each modelling approach using the ‘loo’ criterion in the brms package in R [92]. The nonlinear models provided no more explanatory power compared to the linear models (see electronic supplementary material for results of model comparisons). These results suggest that the linear models with fewer parameters should be used, so we reported the linear results here. All models were Bayesian and we used weakly informative priors for fixed effects, setting the mean to zero and the standard deviation to one. To produce more accurate predictions for age at death, we used more regularizing priors for the intercept (a mean of 1 and standard deviation of 0.1 for the early life survival models, and a mean of 12 and standard deviation of 0.4 for the adult survival models). Specifically, our analyses contained a high proportion of right-censored cases, which can lead to model predictions that overestimate life expectancy [93]. We used credible intervals to determine whether the effect of a variable was substantial or not. If the 85% credible interval for an effect did not overlap with zero, the effect was considered substantial. When the vast majority of the 85% credible interval did not span zero, but there was some overlap, we described the model as being ‘uncertain’ about the effect. To compare how the cumulative index and multivariate models fit the data, we used the ‘loo’ model fit criterion in the brms package [92]. The code used can be found here: https://github.com/skpatter/ELA_Survival_Macaques.

3. Results

(a). Cumulative early life adversity is associated with reduced survival during early life and during adulthood

Individuals that experienced more cumulative early life adversity had higher mortality during early life (β = −0.29 ± 0.07; figure 1, electronic supplementary material, table S5). There were no clear differences in mortality for males versus females during early life (β = −0.05 ± 0.10) and there was no evidence that early life adversity differentially affected mortality risk as a function of sex during early life (β = −0.07 ± 0.09; figure 1, electronic supplementary material, table S5). Adults that experienced more cumulative early life adversity had shorter lives than adults that experienced less early life adversity (β = −0.04 ± 0.02; figure 1, electronic supplementary material, table S5). Among adults, females lived longer than males (β = −0.13 ± 0.02) and there was no evidence that cumulative early life adversity affected mortality risk differentially between males and females (β = 0.00 ± 0.02; figure 1; electronic supplementary material, table S5). Between those that experienced the least and the most amount of cumulative early life adversity in our sample, these effects translate to a 4.78-year difference in average life expectancy among adult females and a 3.94-year difference in average life expectancy among adult males. The maximum ages observed were 31 for females and 29 for males on Cayo, and the average lifespans for those that survived to adulthood were approximately 18 and 15 years for females and males, respectively ([70]; CPRC data). Given that we defined adulthood as beginning at 4 years of age, this indicates an average adult lifespan of 14 years for females and 11 years for males. The average differences in life expectancy between the most adversity-exposed and the least adversity-exposed female and male adults (4.78 and 3.94 years, respectively) represent substantial portions of the adult lifespan: approximately 34% of the lifespan for females and 36% for males.

Figure 1.

Model effects of cumulative early life adversity and sex on survival during early life (a) and adulthood (c). The outer bars show the 85% credible intervals, the inner box shows the 50% credible intervals and the black circle in the middle shows the median of the posterior distribution. Model predictions are shown for the effect of cumulative early life adversity on lifespan in early life (b) and adulthood (d). Cumulative early life adversity on the x-axis is standardized such that an adversity value of zero represents the mean amount of adversity experienced in the sample. Green predictions represent females, and blue predictions represent males. The solid lines show the median estimates, and the shaded region shows the 85% credible intervals.

(b). Several forms of early life adversity are associated with reduced survival, and some effects are sex dependent

Individuals that lost their mother during the first 4 years of life had a higher mortality risk during early life (β = −0.33 ± 0.14) and adulthood (β = −0.06 ± 0.04) than those that did not lose their mother (figure 2, electronic supplementary material, table S6). Maternal loss had a larger negative effect on sons than daughters in both early life (β = −0.32 ± 0.20) and adulthood (β = −0.05 ± 0.06; figure 3, electronic supplementary material, table S6). First-born offspring also had elevated mortality risk during early life (β = −0.36 ± 0.16). In contrast, individuals born to primiparous mothers had better survival in adulthood than those born to multiparous mothers, although the model was uncertain about this effect (β = 0.06 ± 0.05). Effects of maternal primiparity were not moderated by sex (early life: β = 0.09 ± 0.22; adulthood: β = −0.04 ± 0.07). Macaques born into low-ranking matrilines had a higher mortality risk during early life (β = −0.44 ± 0.17) and adulthood (β = −0.16 ± 0.05) than those born into high-ranking matrilines. Matrilineal rank was more strongly associated with female survival than male survival during both periods of life (early life: β = 0.47 ± 0.22; adulthood: β = 0.04 ± 0.06; figure 3; electronic supplementary material, table S6). We also treated matrilineal rank as an ordinal variable and found similar results (electronic supplementary material, table S5). Smaller maternal kin networks at birth were associated with higher early life mortality risk, especially for males (β = 0.05 ± 0.07; sex interaction: β = 0.14 ± 0.10). Smaller maternal kin networks at birth were associated with better survival for adult males but reduced survival for adult females (β = 0.01 ± 0.02; sex interaction: β = −0.05 ± 0.03; figure 3; electronic supplementary material, table S6). The presence of competing younger siblings was associated with higher mortality risk in adulthood, but the model was uncertain about this effect (β = −0.03 ± 0.04). Although the model was uncertain, a competing sibling had a slightly larger effect on females (β = 0.06 ± 0.06). We were unable to examine survival effects during early life given time-varying issues. The model was uncertain about the effects of maternal social isolation; individuals born to socially isolated mothers seemed to have higher mortality during early life than those born to more socially connected mothers (β = 0.15 ± 0.17; electronic supplementary material, table S7), and males were more affected by this than females (β = 0.29 ± 0.25). No effect of maternal social isolation was observed among adults (β = −0.02 ± 0.12; electronic supplementary material, table S7).

Figure 2.

Model effects of sex and the forms of early life adversity on survival during early life (a) and adulthood (b). The outer bars show the 85% credible intervals, the inner boxes show the 50% credible intervals and the black circles in the middle show the medians of the posterior distributions. Green shading represents negative effect sizes, meaning that the variable is associated with shorter lifespans, and purple shading represents positive effect sizes, meaning that the variable is associated with longer lifespans. Note that we do not include the interaction effects between sex and early life adversity in this figure. Because interaction effects influence the interpretation of the sex and adversity effects, this figure should be interpreted with caution as it does not illustrate the complete picture. The full model results, including interaction effects, are presented in electronic supplementary material, Figure S1.

Figure 3.

Interactions between sex and three forms of early life adversity affecting early life survival and adult survival. (a,e) Predicted relationship between maternal kin network size at birth and survival for females (green) and males (blue). (b,f) Predicted relationship between high temperatures during the first year of life and survival for females (green) and males (blue). (c,g) Predicted relationship between maternal loss during the first 4 years of life and survival for males versus females. The circles show the median estimate and the bars show the 85% credible intervals. (d,h) Predicted relationship between matrilineal rank and survival for males versus females. The circles show the median estimate and the bars show the 85% credible intervals.

Higher temperatures during the first year of life were associated with higher mortality risks in early life (β = −0.06 ± 0.24) and adulthood (β = −0.02 ± 0.02), but the models were uncertain about these effects (figure 2, electronic supplementary material, table S6). The effect of high temperatures on survival was moderated by sex (figure 3; electronic supplementary material, table S6). Higher temperatures during the first year of life were more strongly associated with reduced survival among males than females during early life (β = −0.23 ± 0.09), but in adulthood, only females experienced this survival cost (β = 0.02 ± 0.03). Low rainfall was associated with reduced survival during early life, but the model was uncertain about this estimate (β = 0.23 ± 0.18). No effect of rainfall was found during adulthood (β = −0.01 ± 0.02; figure 2; electronic supplementary material, table S6). No effect of group size was observed during early life (β = 0.01 ± 0.07), and while the model was uncertain, it seems that adults born into larger groups exhibited reduced survival (β = −0.02 ± 0.02; figure 2; electronic supplementary material, table S6). No effect of major hurricanes was observed during early life (Georges: β = −0.04 ± 0.50; Hugo: β = −0.08 ± 0.50; Maria: β = 0.08 ± 0.44) or adulthood (Georges: β = −0.02 ± 0.10; Hugo: β = −0.04 ± 0.10; figure 2; electronic supplementary material, table S6).

Model comparisons revealed no substantial difference between models constructed with the cumulative early life adversity index versus those constructed with each form of early life adversity separately (electronic supplementary material, table S8). We ran additional models with a cumulative adversity index that excluded maternal loss and matriline rank—the two strongest predictors of mortality—to see whether the accumulation of adversities with smaller effect sizes impacted mortality (electronic supplementary material, table S9). Individuals that experienced more adversity, as measured by the reduced index, had a higher mortality risk during early life but the effect size was smaller than in the model with the full cumulative index (β = −0.13 ± 0.08). In the reduced index model, an interaction with sex arose such that male survival was more negatively affected by cumulative adversity than female survival during early life (β = −0.12 ± 0.09). Although the effect of cumulative early life adversity on adult survival disappears with the reduced adversity index (β = 0.00 ± 0.02), this is at least partially explained by an interaction with sex: more cumulative adversity, as measured with the reduced index, was associated with poorer adult female survival but better adult male survival (β = 0.02 ± 0.02). The sex-dependent multivariate results above provide more nuanced, informative findings than the cumulative and reduced cumulative index models.

4. Discussion

Our findings indicate that early life adversity shapes both early life survival and adult survival in free-ranging rhesus macaques. Individuals that experienced more cumulative early life adversity lived shorter lives than those that experienced less adversity. The effect size of early life adversity on mortality risk was larger in the first 4 years of life than adulthood, but risks were also elevated in adulthood. Strong effects on early life mortality risk are consistent with the notion of an overall greater vulnerability during development [94,95]. Given the fitness costs of dying prior to reproduction, our results demonstrate that the effects of early life adversity prior to maturity have major fitness ramifications and the full consequences of early life adversity are likely to be larger than predicted in previous studies focused on adult fitness.

The accumulation of multiple adversities did not predict mortality risk better than individual forms of adversity, and assessing the various forms of potential adversity revealed that maternal-related adversities exhibited the largest effects on survival. Maternal death and low matrilineal rank were associated with higher mortality risk in early life and adulthood. The lasting effects of these maternal-related adversities are unsurprising given similar consequences in other mammalian species [22,55,56], as well as consequences of parental-related hardships in humans [96–100]. Survival advantages were also observed among offspring born to more socially connected mothers, but there was considerable variation in this effect and it did not persist into adulthood. From previous analyses, we know that the presence of a competing younger sibling increases mortality risk during early life [34] and these effects appear to persist into adulthood, at least for females. We found that first-born offspring were more likely to die during early life than those that were not the first born. However, while our model estimates were uncertain, among those that survived into adulthood, survival odds were better for those born to primiparous than multiparous mothers. Given the strong negative effects of primiparity on offspring survival in the first 4 years of life, the higher survival odds of adults that were first borns could reflect survivorship bias. Alternatively, because primiparous mothers are young, first-born daughters might have more older kin present (e.g. grandmother, older aunts and older cousins) across a larger span of their life than daughters born to older, multiparous mothers. This can create trade-offs such that first borns face elevated infant mortality risks owing to constraints on maternal care and investment, but if they survive, they might experience reduced mortality risks stemming from support provided by older maternal kin (e.g. [101] ).

Consistent with previous analyses in this population [32,35,102], we did not find substantial impacts of early life hurricane exposure on survival. This is surprising given that macaques in this population exposed to Hurricane Maria showed divergent immune cell gene regulation, suggestive of accelerated ageing [87]. Exposure to major hurricanes also led to greater heterogeneity in reproductive strategies and longevity, and macaques might reduce fertility as a strategy to prioritize survival odds [88,102]. Given our results showing heat effects on mortality and the fact that temperatures increase following hurricanes on Cayo Santiago [45,87], hurricanes might affect macaques indirectly via factors such as deforestation, shade scarcity and heat. Further, given the recency of Hurricane Maria and our small sample of individuals exposed to Maria in this study’s dataset, we are currently limited in our ability to analyse survival outcomes for this most recent hurricane event. Potential impacts of Hurricane Maria may also have been buffered socially—macaques on Cayo Santiago adjusted their social networks after Hurricane Maria [45] and built new social connections, which may buffer negative impacts.

The survival effects of some forms of early life adversity were sex dependent. During early life, male survival was more negatively affected by three forms of adversity: small maternal kin networks, high temperatures and maternal loss. In adulthood, males continued to suffer greater costs of early maternal loss, perhaps reflecting the long-term costs of severe energetic constraints during early life. Males might be more affected by these adversities than females prior to reproductive maturity owing to their energetically costly developmental trajectories and/or owing to maternal decisions to reduce investment in energetically costly offspring during harsh environments [27,28,32,103]. In adulthood, females were more affected by several forms of early life adversity than males: matriline rank, maternal kin network and temperature. Adult females were more affected by matriline rank than adult males, probably because males disperse [70], female dominance hierarchies are fairly stable across time [67] and females typically inherit dominance rank via their matriline. Being born into large maternal kin networks had a positive effect on adult female survival but a negative effect on adult male survival. Given dispersal, males might not receive any immediate benefits of kin support in adulthood and thus only experience the long-term costs associated with earlier competition, consistent with the idea that individuals face trade-offs between benefits of kin support and costs of competition with kin [104]. Males were more susceptible than females to high temperatures during early life, but females were more susceptible in adulthood. In humans, findings have suggested both greater and lesser susceptibility to heat stroke in women versus men [105]. Experimental studies with mice showed that females but not males exhibited delayed myocardial dysfunction following exertional heat stroke [106]. Future work is needed to explore these temperature effects and potential underlying or mediating factors such as how body size, physiology, cardiovascular health and energetic expense patterns are linked to temperature fluctuations, hurricane exposures and mortality across ages in this population.

We faced several limitations in this study. Our results could be shaped by the nature, structure and characteristics of the data. Specifically, the data contain a large proportion of right-censored observations, which could affect the accuracy of lifespan estimates. We were also unable to use time-varying variables because the data violated the Cox model assumptions. This led to the exclusion of the competing sibling variable from our early life survival model and led us to treat maternal loss as a binary variable rather than a continuous variable (i.e. age at maternal loss). Some forms of potential early life adversity had minimal to no effect on mortality risk, which is not entirely unexpected given similar null findings for variables such as drought, group size and maternal social isolation in other primate species [15,25]. However, it is also plausible that some variables like rainfall and group size might have limited effects on survival because drinking water and food are provisioned in the study population. While the macaques still compete over access to food and water resources, competition is probably reduced compared with wild populations. The consequences of adversity might further be hampered in this population because the macaques are not exposed to predators. Our study population might navigate an environment in which the consequences of early life adversity are relaxed, but this represents natural variation that exists across species and populations.

The results of this study open the door for future lines of inquiry. Our results illustrate clear fitness consequences of early life adversity in the form of increased mortality risk, but further research into the biological mechanisms underlying these survival patterns is needed to better understand how early life adversity impacts fitness and health. It was beyond the scope of this analysis to investigate how the timing and length of the windows of exposure to adversity might impact mortality risk. Sensitive windows are periods during which individuals are especially sensitive to inputs and exhibit heightened plasticity [107]. The timing of these windows is hypothesized to vary across forms of adversity, biological systems, individuals and populations [94]. Additional analyses are also needed to investigate more nuanced aspects of early life adversity such as the severity, duration, frequency and predictability of exposure to different forms of adversity [6]. Another angle to investigate, especially given the sex-dependent mortality patterns, is how effects of early life adversity might be moderated or mediated by developmental trajectories and parental investment strategies. Analyses are also needed that examine whether individuals adjust other aspects of their life history strategies (e.g. pace of reproduction, age at maturity) to compensate for reduced life expectancy [35,88,108]. Importantly, variation in model estimates and predictions conveys that while early life adversity can have negative consequences, such effects are not definitive. Social connections and behavioural adjustments [45,109] should be investigated as potential contributors to resilience. Further age-related changes in social behaviour and age-related changes in the benefits and costs of sociality should be investigated as a function of early life adversity and adult health [110–114] .

In sum, our study demonstrates that exposure to early life adversity increases mortality risk in male and female rhesus macaques. Lower odds of surviving to reproductive age indicate that early life adversity can have major fitness ramifications for both an organism and their parents. Reduced life expectancy among those that survive to adulthood suggests that early life adversity can have persistent fitness costs and long-term health consequences. Adversities related to the maternal environment generally had the largest impacts on offspring survival. Sex-dependent effects of early life adversity in rhesus macaques are probably driven by social system characteristics (i.e. female philopatry) and sex-based variation in energetic demands (but see [32]). We were able to show that the form of adversity, socio-sexual context and other biological factors interact to shape the timing and severity of consequences. Natural populations of non-human animals can prove valuable not only for improving our understanding of the evolutionary pressures that shape developmental plasticity, life history strategies, ageing and early life sensitivities but also for better contextualizing findings in humans and informing future research in humans [115,116].

Ethics

This animal study was reviewed and approved by University of Puerto Rico, Institutional Animal Care and Use Committee (IACUC).

Data accessibility

The data used can be found at [117]. Supplementary material is available online [118]. Code is available at [119].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

S.K.P.: conceptualization, data curation, formal analysis, funding acquisition, methodology, visualization, writing—original draft, writing—review and editing; E.A.: investigation, writing—review and editing; A.M.A.: data curation, investigation, project administration; M.I.M.: funding acquisition, project administration; J.E.N.-D.V.: investigation; R.M.P.: methodology, writing—review and editing; D.P.: investigation; A.R.: investigation, writing—review and editing; A.R.-L.: data curation, investigation, project administration; I.V.: investigation, writing—review and editing; A.J.L.: funding acquisition, methodology, supervision, writing—review and editing; N.S.-M.: conceptualization, data curation, funding acquisition, methodology, project administration, supervision, writing—review and editing; L.J.N.B.: conceptualization, data curation, funding acquisition, methodology, project administration, supervision, writing—review and editing; J.P.H.: conceptualization, data curation, funding acquisition, methodology, project administration, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by the National Institutes of Health (R01-AG060931; R01-AG084706; R21AG078554; R56-AG071023; ORIP-P40OD012217), the National Science Foundation (SMA-2105307), a collaborative initiative between the National Science Foundation and the European Research Council (supplemental funding for NSF-SMA-2105307 and ERC-864461), the Pinkerton Foundation and the NYU Tandon School of Engineering’s Applied Research Innovations in Science and Engineering program.