Maternal effects on offspring growth indicate post-weaning juvenile dependence in chimpanzees (Pan troglodytes verus)

Background: In animals with altricial offspring, most growth occurs after birth and may be optimized by post-natal maternal care. Maternal effects on growth may be influenced by individual characteristics of the mothers, such as social status, individual investment strategies and the length of association with offspring. The prolonged juvenile dependence seen in humans is a distinctive life history adaptation, which may have evolved to facilitate sustained somatic and brain growth. In chimpanzees, offspring are typically weaned at approximately 4 years old, yet immature individuals continue to associate with their mothers for up to 10 years beyond weaning. Whether this lengthy association or the individual characteristics of mothers influences growth patterns in this species is not clear. The relationship between urinary creatinine and specific gravity is an established non-invasive measure of muscle mass in humans and chimpanzees. We analysed the urinary creatinine and specific gravity of 1318 urine samples from 70 wild chimpanzees from the Taï Forest, Ivory Coast aged 4 to 15 years. Results: We showed a clear increase in urinary creatinine levels with age in both males and females, replicating established growth curves in this species and reaffirming this measure as a reliable proxy for lean body mass. Comparing those who experience maternal loss (orphans) with non-orphan chimpanzees, maternal presence beyond weaning age and into late juvenility positively influenced offspring muscle mass throughout ontogeny such that orphans had significantly less muscle mass than age-matched non-orphans. In age-matched offspring with mothers, those with high-ranking mothers had greater muscle mass. Accounting for variation in muscle mass attributable to maternal presence, we found no effect of maternal investment (length of inter birth interval, from own birth to birth of following sibling) on offspring muscle mass. Conclusion: Chimpanzee mothers have an extended and multi-faceted influence on offspring phenotypes. Our results suggest that maternal investment extends beyond lactation and into early adulthood and has clear benefits to offspring physical development. Therefore, prolonged juvenile dependence, although unique in its form in human societies, may be a trait with deeper evolutionary origins. Data were collected at the Taï National Park, Côte d’Ivoire (5°45′N, 7°7′W [59]) on three different chimpanzee communities (i.e., North, South and East). Systematic observation effort, including collection of demography data in Taï, started with North group in 1982 ([53]; North group: 1982-present; South group: 1993-present; East group: 2000 – present). Behavioural data included nest to nest focal-follows [60] of individuals of the different social groups on a daily basis by trained local assistants and researchers. In addition to focal follows, observers regularly collected urine samples from young individuals (70 subjects, 4 to 15 years) between February 2000 and July 2018. Urine sample collection and demography data were used to investigate the effect of maternal presence and investment on offspring lean muscle mass. Weaning in chimpanzees is estimated to occur at around 4–5 years of age [61, 62]. Post-weaning, individuals typically stay in regular association with their mother up until the age of 10 years, around which time they gradually become fully independent: males begin integrating into the dominance hierarchy of the group and females are increasingly likely to disperse from their natal group [53, 63]. Accordingly, we defined maternal loss (i.e., orphans; n = 18) as individuals who lost their mother post-weaning and before 10 years of age [54], and investigated the influence of maternal dominance rank (see below) in individuals under 10 years, taking into account the period of mother-offspring association in chimpanzees. In accordance with published weaning age estimates [61, 62], in our study population, the youngest age a post-weaned orphan was sampled in our dataset was at 4.0 years old. Therefore, and in line with previous study [41], we restricted subsequent analyses to samples collected from individuals between the ages of 4 to 15 years. We could determine offspring IBIs (to subsequent births) for 54 individuals, using demography data. We had reliable information of the year and month of birth for all individuals (including the study subjects and their siblings for IBI calculations). In cases where the day of birth was unknown, we assigned the 15th of the respective birth month as the day of birth. The IBI for an individual was calculated as the exact time in years between their birth and the birth of the subsequent offspring of their mother. To determine the dominance relationships between adult females in each community, we used submissive uni-directional pant grunt vocalizations [57] (North: 966 vocalizations, South: 1302 vocalizations, and East: 207 vocalizations), and applied a likelihood-based adaptation of the Elo rating approach [64–66]. Within each social group, we then distinguished maternal ranks in two ways. First, for each offspring, we assigned a continuous Elo score of their mother’s rank (standardized between 0 and 1 in each group) calculated on the date of each offspring urine sample. Second, as rank effects may be non-linear, particularly in terms of resource acquisition in female chimpanzees, we also delineated between females of the highest dominance rank (alpha), and females with a rank other than alpha (subordinate). This is in accordance with other studies of female chimpanzee rank [30, 32, 58, 67], and follows previous findings of higher and more constant body mass in the highest ranking chimpanzee females [58]. Urine samples were collected in the field into 2 ml cryo vials from leaf litter using a plastic pipette. Upon arrival in camp, and within 12 h of collection, samples were transferred into liquid nitrogen. Samples were then shipped frozen on dry ice to the Laboratory of Endocrinology at the Max Plank Institute for Evolutionary Anthropology in Leipzig, Germany, where we stored them at − 80 °C until analysis. Creatinine levels were measured via colorimetric reaction of urine with picric acid. To account for the concentration of urine, we measured SG, which is independent of muscle mass, using a digital refractometer (TEC, Ober-Ramstadt, Germany). Following common practice, we exclude highly diluted urine samples. Thus, we excluded 137 samples with SG < 1.003 [41], and 16 additional samples with creatinine levels with values ≤0.05 mg/ml [68]. We estimate muscle mass of young individuals (4–15 years) by means of creatinine controlling for SG for each urine sample. This is a pre-established validated measure in chimpanzees [41] and humans [44]. To investigate the effects of age, sex, maternal presence, rank and investment on offspring muscle mass (log transformed creatinine mg/ml) we fitted Linear Mixed Models (LMM) [69] with Gaussian error structure and identity link function. In the first model (‘effects of maternal presence’), we investigated the effect of maternal presence or absence on offspring muscle mass during the development of immature male and female chimpanzees. Our test predictors for this model were the interaction between the age and sex of the individual, as well as whether the individual was an orphan or not at the time of urine sample collection. Furthermore, to account for urinary concentrations, we included SG (minus 1) as a control predictor [41]. As seasonal variation in rainfall, temperature and humidity may influence creatinine levels in Taï [70], we controlled for circannual variation in creatinine levels by converting Julian dates into a circular variable, and including its sine and cosine into the model [70, 71]. We included group membership as another control predictor (i.e., North, South, or East). Furthermore, as we use repeated measures from individuals, offspring with the same mother and days, we included the identity of the subject, its mother, the year (combination of group and year, termed “year id” hereafter) and day (combination of group and date, termed “day id” hereafter) as random effects. By this we account for non-independent sampling of certain subjects, their mothers, days or years disproportionally affecting urinary creatinine levels, and thus, avoid pseudo-replication [69]. In order to keep type I error rate at the nominal 5% and to account for potential non-uniform variation of our predictor variables within the random effects [72, 73], we included a maximal random slope structure, incorporating random slopes for the predictors with appropriate variation within the particular random effects. This resulted in random slopes for age, SG, and sine and cosine of date within subject, mother and year id. Our dataset for the ‘effects of maternal presence’ model included 1318 urine samples of 70 individuals (a mean + SD of 18.83 + 19.15 samples per subject) and 41 mothers. In the second LMM (‘maternal rank effects’), we investigated the effect of maternal dominance rank on the development of offspring muscle mass. We used submissive uni-directional pant grunt vocalizations [57] to calculate the dominance rank of mothers. Maternal dominance rank may affect offspring muscle mass through priority of access to high quality food sources of high-ranking mothers. Thus, in this model we included samples of offspring with known maternal rank and aged between 4 and 10 years to reflect the estimated period that young, weaned chimpanzees regularly associate with their mothers [53]. The ‘maternal rank effects’ model included all the predictors from the ‘effects of maternal presence’ model with the exception of the predictor ‘orphan (yes/no)’, as this model only included samples from individuals whose mother was alive. Our test predictors for this model were the dominance rank of the mother for each subject’s sample as both a continuous (linear) and a categorical term (alpha vs. subordinate). To evaluate reliably the effect of maternal rank on muscle mass, independently of maternal age, we included an additional control predictor of the age of the mother into the analysis (this showed no collinearity with maternal dominance rank: vif < 1.6). As per the ‘effects of maternal presence’ model we included the random effects of subject, mother, year id and day id, as well as random slopes for the age of the subject, age and rank of the mother, SG and seasonality variables within subject, mother and year identity. Our dataset for the ‘maternal rank effects’ model included 414 urine samples of 48 subjects (a mean + SD of 8.62 + 8.51 samples per subject) and 29 mothers. In our final analyses, to evaluate the effect of maternal investment on offspring development, we extracted the variance of the intercept of the random effect of individual identity (‘best linear unbiased predictors’ [74]) from the ‘effects of maternal presence’ model (this includes both orphans and non-orphans between 4 and 15 years of age). We only included subjects with known IBI and at least 2 urine samples (n = 45 subjects; a mean + SD of 19.20 + 20.68 samples per subject). We then fit a linear regression for each sex of these variance estimates against IBIs, our proxy measure for maternal investment [8, 41]. By taking this approach, we were able to investigate the independent effect of IBI on muscle mass while accounting for the effects of age, sex and maternal presence on muscle mass. This approach was preferred over including IBI in either the ‘effects of maternal presence’ or ‘maternal rank effects’ models as this would have limited the number of individuals that could be included given either IBI is unknown or absent (they are the sole dependent offspring of their mother during the sampling period) for many individuals included in these models. The best unbiased linear predictors were extracted from the ‘effects of maternal presence’ model rather that the ‘maternal rank effects’ model as the former contained a larger number of individuals and samples as well as greater variance attributable to the random effect of individual identity than observed in the latter model. For all statistical analyses, we used R (version 3.5.3 [75]) to process the data and fit the models. Prior to fitting the models, we checked the distribution of the response and all predictors. As a result, we log transformed creatinine levels to achieve a more symmetrical distribution. In addition, we z-transformed the covariates of IBI, SG, age of subject, and age and dominance rank of mother [76]. We verified the assumptions of normally distributed and homogeneous residuals by visual inspection of qq-plots and residuals plotted against fitted values. These evaluations did not reveal obvious deviations from model assumptions. We used the function vif of the R package ‘car’ [77] applied to a standard linear model lacking the random effects to derive Variance Inflation Factors (VIF), which did not reveal collinearity problems (largest VIF: ‘development model’ = 1.20; ‘maternal effect model’ = 1.89 [78]). We used the function lmer of the R package ‘lme4’ [79] to fit both models. We compared the fit of both full models with those of a respective null model lacking only the test predictors (thus the null model contains all control predictors, random effects and slopes), using a likelihood ratio test [80]. We obtained individual p-values for the fixed effects using the drop1 function in R, by systematically dropping each fixed effect from the model one at a time [72], and comparing the respective reduced model lacking the individual fixed effects with the full model. For both models, we assessed model stability by excluding the levels of the random effects (identities of subject, mother, day id, year id) one at a time and comparing the estimates derived for these subsets with those derived for the full data set. Stability assessment revealed no influential subjects, day or year id to exist. To obtain confidence intervals we conducted parametric bootstraps using the function bootMer of the package ‘lme4’. Finally, we evaluated the effect sizes (R2) using the function r.squaredGLMM of the R package ‘MuMIn’ [81]. We report the variance explained by the fixed effects (marginal-R2m), and the fixed and random effects (conditional-R2c) [82].