Birth timing generates reproductive trade-offs in a non-seasonal breeding primate

Abstract

The evolutionary benefits of reproductive seasonality are often measured by a single-fitness component, namely offspring survival. Yet different fitness components may be maximized by different birth timings. This may generate fitness trade-offs that could be critical to understanding variation in reproductive timing across individuals, populations and species. Here, we use long-term demographic and behavioural data from wild chacma baboons (Papio ursinus) living in a seasonal environment to test the adaptive significance of seasonal variation in birth frequencies. We identify two distinct optimal birth timings in the annual cycle, located four-month apart, which maximize offspring survival or minimize maternal interbirth intervals (IBIs), by respectively matching the annual food peak with late or early weaning. Observed births are the most frequent between these optima, supporting an adaptive trade-off between current and future reproduction. Furthermore, infants born closer to the optimal timing favouring maternal IBIs (instead of offspring survival) throw more tantrums, a typical manifestation of mother–offspring conflict. Maternal trade-offs over birth timing, which extend into mother–offspring conflict after birth, may commonly occur in long-lived species where development from birth to independence spans multiple seasons. Our findings therefore open new avenues to understanding the evolution of breeding phenology in long-lived animals, including humans.

1. Introduction

Empirical studies investigating variation in reproductive timing have mostly focused on fast-lived seasonal breeders, whose development from birth to independence generally occurs within the most productive season [1]. In long-lived mammals, the reproductive cycle from birth to weaning cannot similarly be squeezed into one annual food peak, and consequently, females must choose which stage(s) of the reproductive cycle to synchronize with one or more food peak(s). For example, female mammals could match the annual food peak to coincide with either late weaning or mid-lactation (two critical stages that require readily accessible food), but usually not both. The reproductive timing strategy is likely to depend on how females trade-off the survival of their offspring (mortality risks tend to peak at the end of weaning) [2–4] with their own reproductive costs (energetic demands tend to peak around mid-lactation) [5,6]. Whether such reproductive timing strategies can vary within populations is largely unknown. In addition, while evolutionary trade-offs between offspring quality and quantity have been described both within and across species through associations between birth spacing and infant growth and survival [7,8], the existence of maternal trade-offs over birth timing have only been suggested theoretically [3] and never tested empirically in mammals (but for a bird, Fulica atra, see [9]).

Here, we investigate variation in maternal reproductive success and mother–offspring relationships associated with variable birth timings in the annual cycle of wild chacma baboons (Papio ursinus) living in a seasonal semi-arid savannah (Tsaobis, Namibia). Baboons are African primates distributed across a wide latitudinal range and a classic model for understanding how early humans adapted to seasonal savannahs [10,11]. In particular, baboons typically breed year-round [12] and are therefore considered non-seasonal breeders, though the distribution of births shows moderate seasonality (i.e. varies along the annual cycle) in some species and populations [13–15]. In addition, infant baboons, like many young primates including human toddlers, commonly perform tantrums, a manifestation of mother–offspring conflict [16,17]. Using a combination of detailed long-term life-history and behavioural data collected over 15 years (2005–2019), we first characterize the reproductive and environmental seasonality of the Tsaobis baboons. Second, we quantify the consequences of birth timing on two components of female fitness: offspring survival and maternal interbirth intervals (IBIs) and seek to identify two distinct birth timing optima. We further test whether individual females may vary in their birth timing strategies, and specifically which individual traits predict whether a female is more likely to give birth around one or the other optimum. In particular, dominance rank and parity can affect various aspects of individual reproductive performance, including offspring survival and IBI [18–20], and may influence birth timing strategies accordingly. Similarly, mothers conceiving close to the optimal timing that alleviates the energetic costs of lactation may subsequently favour male over female embryos, which are more costly to produce in sexually dimorphic mammals [18,21]. Third, we investigate if maternal care can mitigate the costs of suboptimal birth timing for offspring, and whether infants born and weaned during suboptimal periods had higher tantrum frequencies.

2. Material and methods

(a) . Study population

Three habituated groups (named J, L and M) of wild chacma baboons were followed between 2005 and 2019: J and L since 2005, and M, a fission group from J, since 2016. They live in a desert-edge population at Tsaobis Nature Park (22°23 S, 15°44′50 E) in Namibia, in a seasonal and arid environment [22]. Water is always available through the presence of both natural seeps and artificial water points for wildlife and livestock. A field team was present each year, mainly during winter (between May to October), for a variable number of months (mean = 4.5, range: 1.9–7.9), that collected daily demographic and behavioural data, as well as GPS locations, while following the groups on foot. All individuals, including infants, are individually recognizable thanks to small ear markings performed during capture and/or other distinctive features.

(b) . Environmental data

In order to describe the relationship between reproductive and environmental seasonality, we characterize two aspects of environmental seasonality at Tsaobis: rainfall and vegetation cover (an index of food availability).

Daily rainfall in a 0.25 × 0.25 degree grid cell resolution (corresponding to 28 × 28 km at this latitude) was extracted using satellite data sensors from the Giovanni NASA website (product TRMM 3B42) [23]. We computed a rectangular geographic area that encompasses the global ranging area of the Tsaobis baboons, using GPS locations collected by observers every 30 min when following the study groups. We used the minimal and maximal latitude and longitude recorded between 2005 and 2019. Monthly cumulative rainfall (summed across daily values) was computed between 2005 and 2019.

We used the Normalized Difference Vegetation Index (NDVI) as an index of food availability. NDVI is obtained from the red : near-infrared reflectance ratio, with NDVI = (NIR − RED)/(NIR + RED), where NIR and RED are the amounts of respectively near-infrared and red light, reflected by the vegetation and captured by satellites [24]. NDVI thus produces a quantitative index of primary productivity with higher values corresponding to a higher degree of vegetation cover [25]. It has previously been used as an indicator of habitat quality for the Tsaobis baboons [26] and other baboon populations [27]. We further confirmed that temporal variation in NDVI reflected temporal variation in rainfall: mean cumulative rainfall over the past three months explained between 60 and 72% of the NDVI variation (electronic supplementary material, appendix S1). To index food availability using NDVI, we first computed 100% isopleth home ranges for each group using kernel density estimates with the adehabitatHR package (‘kernelUD’ function) [28], based on the daily 30 min GPS locations from 2005 to 2019 (from 2016 to 2019 for M group). We obtained one home range per group for the entire study period. We then extracted the mean NDVI per 16-day period on a 500 m × 500 m resolution (these 16-day windows are imposed by the resolution of the NASA datasets) across these periods using data provided by NASA (MODIS13A1 product) [25] within these home ranges. Daily NDVI was computed by linear interpolation and then averaged to obtain a monthly value. In contrast with rainfall, NDVI was measured with greater resolution and for each group separately because baboons finely adjust their ranging behaviour in relation to food availability [29].

(c) . Individual data

A female was considered an adult when she reached menarche. The reproductive state of each adult female was monitored daily. A female could be: (i) pregnant (assessed by the paracallosal skin turning red and absence of cycles over the following months), with the exact start date of pregnancy being determined post hoc following infant birth, and encompassing 190 days (mean gestation length in this population, n = 13 pregnancies where both conception and birth were observed, range: 181–200 days, s.d. = 5) between conception and birth; (ii) lactating, as long as the female did not resume cycling after an infant birth; and (iii) cycling, including both swollen females in oestrus (i.e. sexually receptive with a perineal swelling) and non-swollen females at other stages of their cycle. Conceptive cycles were established based on the beginning of a pregnancy and were usually confirmed by a birth. The first post-partum cycle (i.e. cycle resumption) is the first cycle following an infant's birth, when the female resumes cycling after lactation. The exact date of the cycle resumption corresponds to the first day of oestrus of the first post-partum cycle (i.e. the first day when a sexual swelling is recorded). The dates of these reproductive events (conceptions, births and cycling resumptions) were either known with accuracy when recorded by field observers or estimated in the absence of observers using the methods detailed in electronic supplementary material, appendix S2 and table S1.

Female parity was known from life-history records and defined as primiparous (between the birth of her first and second infant) or multiparous (after the birth of her second infant). The parity of adult females at the start of the study was assessed using both the presence of older offspring based on a combination of behavioural and genetic data [30], alongside female age estimated using teeth eruption patterns [31].

Female social rank was established annually for each group using ad libitum and focal observations of agonistic interactions between adult females: supplants, displacements, attacks, chases and threats [32]. We computed a linear hierarchy using Matman 1.1.4 (Noldus Information Technology, 2013) and then converted to a relative rank to control for group size (i.e. the number of adult females in the group). Each female was thus assigned one rank per year, ranging from 0 (lowest ranking) to 1 (highest ranking).

(d) . Fitness data

We tested the influence of birth timing in the annual cycle on two fitness measures, namely offspring mortality before weaning and the duration of the maternal IBI. For each infant born between 1 January 2005 and 1 August 2018, we investigated whether it died (yes/no) before weaning. The weaning age was identified as 550 days on the basis of the maximum length of post-partum anoestrus (n = 33 cases for which both birth and cycle resumption were known with accuracy; see also electronic supplementary material, appendix S3) and presumably reflected the upper threshold of weaning age in our population [33,34]. Death was recorded when a corpse was observed or when the infant had been missing in the group for five consecutive days. Infants born later than August 2018 were not considered as their survival outcome was unknown. Four infants that disappeared between consecutive field seasons were omitted because we could not establish whether the age of death was before or after 550 days.

We defined IBIs as the number of days between two consecutive live births of the same female. We only considered IBIs for which the first infant reached weaning [18] (i.e. survived until 550 days old). We discarded other IBIs as females resumed cycling rapidly after their infant's death when unweaned (median = 21 days, range = 9–51, n = 9 observed death), and their IBIs would have been shortened regardless of environmental seasonality.

(e) . Behavioural observations

In order to characterize variation in maternal care and in mother–offspring conflict, we used three behavioural indicators: suckling, infant carrying and tantrum frequencies. We also used these behavioural data, along with life-history data, to assign different developmental stages, including the different stages of weaning and the peak of lactation after an infant's birth (see electronic supplementary material, appendix S3). Field observers collected a total of 1185 h of focal observation [35] of 20 (in 2017, 2018 and 2019) or 60 (in 2006) minutes long on 69 infants (mean ± s.d. = 17.1 ± 7.8 h of observations per infants, range = 6.3–34.6) (see electronic supplementary material, appendix S4 for more details).

(i) . Maternal care during weaning

Maternal care was quantified through two measures: suckling frequency and infant carrying frequency, which represent the two main energetic costs of maternal care before weaning [5,36]. First, for each scan observation (taken every 5 min), we considered whether the infant was suckling (1) or not (0) to investigate the effect of birth timing on variation in suckling frequency. Here, suckling was recorded when the focal individual had its mouth on its mother's nipple, and we therefore could not distinguish comfort (when a juvenile suckles for reassurance, without any milk transfer [37]) from nutritive suckling. We considered only infants aged 2 months to 18 months old for this analysis (electronic supplementary material, figure S1), using 11 687 scans from 55 infants. The birth date uncertainty for these 55 infants ranged from 0 to 130 days (with a median birth date uncertainty of 16 days) and was taken into account in subsequent models (see electronic supplementary material, appendix S5).

Second, for each scan observation during which an infant was travelling, we determined whether the infant was carried by its mother (1) or travelled on its own (0). This variable allowed us to monitor the gradual decrease from full maternal dependence to full independence during travelling. We considered infants aged from 2 months to 12 months old for this analysis (electronic supplementary material, figure S1), using 924 scans from 35 infants.

(ii) . Mother–infant conflicts during weaning

We considered infant tantrums as a behavioural measure of mother–offspring conflict, reflecting when an infant's request to access resources from its mother was not initially satisfied [16]. We considered only infants aged 2–18 months for this analysis (electronic supplementary material, figure S1), using 2221 focal observations from 55 infants. During each focal observation, we determined if a tantrum occurred (1) or not (0), based on a range of distinctive offspring vocalizations (gecks, moans and loud screams) and behaviours (frenzied behaviour when infants hurl themselves to the ground, sometimes accompanied by rapidly rotating their tail) that were recorded on a continuous basis and are characteristic of baboon tantrums [38,39]. A tantrum was considered to occur when at least two of these behaviours or vocalizations were recorded, separated by at least 30 s (isolated complaints, and complaints that lasted fewer than 30 s, were thus not considered as tantrums here). Tantrums were usually caused by the maternal refusal of access to the nipple or to carrying and more rarely by maternal absence.

(f) . Statistical analysis

(i) . Characterization of the environmental and reproductive seasonality of the Tsaobis baboons

First, to assess the strength and direction of reproductive seasonality, we used a Rayleigh test, from circular statistics, which characterizes the deviation of circular data from a uniform distribution, via the direction (µ) and length (R) of the mean vector summing all observed events across the annual cycle (R = 0 when the event is evenly distributed, and R = 1 when all events are synchronized, i.e. occurs the same day) [40]. Our sample comprised 241 conceptions, 215 births and 171 cycle resumptions which occurred between 2005 and 2019. Uncertainties in those dates were taken into account in all subsequent analyses using 1000 randomized reproductive events for each variable (electronic supplementary material, appendix S5).

(ii) . Birth timing effects on two fitness traits and individual effects on birth timing

To quantify the effect of birth timing on the probability of offspring mortality before weaning (Model 1), we ran a generalized linear mixed model (GLMM) with a binomial error structure. We then ran a linear mixed model (LMM, Model 2), testing the effect of birth timing on IBIs.

In both models, we used a sine term to describe the timing of an infant's birth in the annual cycle (see electronic supplementary material, appendix S6 for more details on this procedure) [26]. We included as random effects the year of infant birth and identity of the mother to control for repeated observations. We also included maternal parity, rank (in the birth year of the focal infant) and infant sex as fixed effects, because maternal parity and rank often affect reproductive traits in primates, including baboons [18,41], while infant sex can affect both the mother's subsequent IBI [42] and the probability of infant survival in sexually dimorphic primates [43]. We also control for group identity as a fixed effect in both models, as data were collected from only three groups in this study [44].

We investigated the individual determinants of female reproductive decisions over birth timing, based on 215 births from 62 females. Our two response variables were the deviations, in days, from the birth timing that minimizes offspring survival (15 December) in Model 3 and maternal IBIs (1 September) in Model 4 (electronic supplementary material, table S2). For both Models 3 and 4, we tested the influence of infant sex, female parity and rank (as fixed effects) on the proximity of birth to the optimal timing for offspring survival (Model 3) or for maternal IBI (Model 4). We also controlled for the identity of the mother and birth year as random effects, in order to take into account the between-year environmental variation likely to affect birth timings. We included group identity as fixed effects (as there were only three levels for this factor [44]). We tested the significance of maternal identity using a likelihood-ratio test (LRT), comparing the model with and without this random effect.

(iii) . Birth timing effects on maternal care and tantrum probability

We ran three GLMMs with a binomial error structure to test the effect of birth timing on the probability of suckling (Model 5), infant carrying (Model 6) and tantrums (Model 7). Models 5 and 6 are scan-based data: during a scan observation, the infant is suckling (yes/no, Model 5), and during a travelling scan observation, the infant is carried by its mother (yes/no, Model 6). Model 7 is based on the entire focal observation as tantrum events are relatively rare: during a focal observation, the infant throws a tantrum (yes/no).

In order to investigate the potential effect of birth timing on maternal care and tantrum probability, we used a sine wave term for infant birth date as a fixed effect (electronic supplementary material, appendix S6). We included, as random effects, the identity of the infant (Models 5–7) to control for repeated observations. We also added the focal observation as a random effect for Models 5–6. We controlled for group identity and year of observation as fixed effects in all models, as there were less than five levels for both factors [44]. In all models, we included maternal parity, rank (in the year of birth of the focal infant) and infant sex as fixed effects. Such parameters are likely to affect reproductive performances as well as the probabilities of maternal care and mother–offspring conflict [39,43]. For Model 7, we also controlled for the duration of focal observation as a fixed effect.

For Models 5–7, we further controlled for the effects of infant age, which modulates the amount of maternal care and probability of tantrums throughout early development [16,43]. We considered four different possibilities for the form of the relationship between infant age and the response variable, using a regression thin-plate spline (general additive model), a simple linear effect and a polynomial regression (of 2 or 3 degrees), respectively [45]. To determine the best fit, we ran these different preliminary models with no other fixed effect but including all random effects (and the duration of focal observation for Model 7) and selected the model minimizing the AIC. The age effect was linear for suckling and infant carrying probabilities (Models 5 and 6) and a second-degree polynomial for tantrum probability (Model 7).

Lastly, mothers might be expected to invest more, and similarly infants might be expected to have more requests for maternal care, during the lean season, irrespective of the developmental trajectory of the infant (i.e. regardless of its age and birth timing, whether it was born in the optimal period or not). Therefore, we also investigated the potential effect of seasonality by assessing the influence of the observation date on suckling, infant carrying and tantrum probabilities (see electronic supplementary material, appendix S7 for more details). We did not include in the same model observation date and birth date, as they give redundant information (observation date is, by definition, the sum of birth date and infant's age, and infant's age is already included as a fixed effect). We present our models of birth date effects in the main text (Models 5–7; see also electronic supplementary material, table S3), and our models of observation date effects in the electronic supplementary material (Models 5–7; electronic supplementary material, table S4).

The structure of each model, with the different fixed and random effects included, alongside sample size, is summarized in electronic supplementary material, table S5.

(iv) . Statistical methods

All statistical analyses were conducted in R v. 3.5.0 [46]. For the Rayleigh test, we used the function ‘r.test’ from the R package ‘CircStats’ [47]. To run mixed models, we used ‘lmer’ (for LMMs) or ‘glmer’ (for binomial GLMMs) function on the lme4 package [48]. To run general additive mixed models (GAMMs) when investigating the best age effects on suckling, infant carrying and tantrum probabilities, we used the ‘gam’ function of the ‘mgcv’ package [45]. All quantitative fixed effects were z-transformed to facilitate model convergence. When we obtained singular fits, we confirmed the results by running the same models with a Bayesian approach, using the ‘bglmer’ and ‘blmer’ functions of the ‘blme’ package [49]. To diagnose the presence of multicollinearity, we calculated the variance inflation factor for each predictor in each full model using the ‘vif’ function of the R ‘car’ package [50]. These were lower than 2.5 in all cases. To assess the strength of the fixed effects in each model, we used the Wald chi-square tests with associated p-values computed with the ‘Anova’ function of the R package ‘car’ [50] and calculated the 95% Wald level confidence intervals. We further checked the distribution of residuals with ‘qqPlot’ function of the car package for LMMs [50] and with ‘simulateResiduals’ from DHARMa package for binomial GLMMs [51].

3. Results

(a) . Characterization of the environmental and reproductive seasonality of the Tsaobis baboons

Environmental seasonality was pronounced at Tsaobis (figure 1a). Mean annual rainfall was low and variable (mean ± s.d. = 192 ± 143 mm), falling mostly between January and April (figure 1a). Seasonal variation in NDVI, a satellite-based proxy of primary productivity, followed a similar, but slightly lagged pattern, to rainfall (figure 1a). The highest birth frequency occurred in October–November (i.e. 28.4% of annual births), preceding the peak in rainfall (February) and NDVI (March–April; figure 1a).

Figure 1.

Tsaobis baboons' reproductive timings in relation to environmental seasonality. In (a), we plotted the proportion of conceptions (n = 241, in light grey) and births (n = 215, in dark grey) recorded in 2005–2019 per month. We plotted the mean monthly cumulative rainfall (in mm) per month in blue and the mean NDVI value per month in green between 2005 and 2019. We represented the standard errors associated with vertical black segments. The pink and orange bars in the background represent, respectively, the maternal IBI and the offspring survival optimal birth timings. In (b), we aimed to represent the different phases of the female reproductive cycle, when the birth date occurs within the offspring survival (15 December) or maternal IBI (1 September) optimal timing, according to seasonal variation of NDVI. The green bar plot in the background indicates the mean NDVI per month (see y-axis). Pregnancy, indicated with grey arrows, occurs six months prior to a birth. Early weaning, indicated with salmon-colour arrows, occurs from 6 to 9 months after birth. Lactation peak, indicated with black stars, occurs around six months after birth. Weaning end, indicated with blue arrows, occurs from 12 to 18 months after birth (see electronic supplementary material, appendix S3 for the characterization of these different reproductive stages). (Online version in colour.)

Conceptions, births and cycle resumptions occurred throughout the year (electronic supplementary material, figure S2), indicating an absence of a strict breeding season. We used circular statistics to test whether moderate seasonality may still occur, computing, respectively, the mean annual angle (µ) and Rayleigh tests (R and p-values) for the annual distribution of 241 conceptions, 215 births and 171 cycle resumptions recorded between 2005 and 2019. The frequency of conceptions and births showed slight seasonal variations, which reached statistical significance for conceptions only (conceptions: µ = May 12, R = 0.13, p = 0.02; births: µ = 18 November, R = 0.09, p = 0.17; cycle resumptions: µ = 4 December, R = 0.08, p = 0.36; electronic supplementary material, figure S2).

(b) . Birth timing effects on two fitness traits and individual effects on birth timing

We considered two indicators of maternal fitness. First, we assessed whether or not infants survived until weaning (550 days). In our sample, 39 infants out of 195 (i.e. 20%) died before weaning, at a median age of 74 days (range 1–284 days, n = 17 known dates of death). Second, we gathered 120 IBIs from 43 adult females, ranging from 397 to 1132 days with a mean of 678 days (s.d. = 128).

Birth timing influenced these two indicators of maternal fitness. First, birth timing affected offspring survival (electronic supplementary material, table S6): infants born between 15 November and 1 January were the most likely to survive until weaning (electronic supplementary material, table S2), indicating an optimal birth timing for offspring survival in the annual cycle (figure 2a). Infants born on 15 July were 66% more likely to die before being weaned than the infants born on 15 December (electronic supplementary material, table S6). Second, the duration of maternal IBI was influenced by the timing of the birth opening the IBI (electronic supplementary material, table S6): females giving birth between 1 August and 15 September had the shortest IBIs (electronic supplementary material, table S2), indicating another different optimal birth timing for maternal reproductive pace in the annual cycle (figure 2b). Females giving birth on 1 September had IBIs 73 days shorter than females giving birth on 1 March.

Figure 2.

Distinct optimal birth timings for current and future reproduction. We plotted the predicted values of the full models (Model 1 looking at offspring mortality in (a), and Model 2 looking at IBIs in (b)) according to the month of infant birth. The number of births observed for each month is indicated in blue below the bar. The dots represent the mean values, while the vertical black bars represent its standard deviations. The offspring survival optimal birth timing is identified as the period minimizing offspring mortality, i.e. from November to February, and indicated with orange dots (a). The maternal IBI optimal birth timing is identified as the period minimizing maternal IBI, i.e. from July to September, and indicated with pink dots (b). (Online version in colour.)

We then asked whether some females might be more likely to time their births to maximize current over future reproduction, or vice versa. However, we failed to detect any significant variance associated with maternal identity on the deviation between observed birth and the optimal birth timing maximizing offspring survival (LRT = 0.66, p = 0.42) versus maternal IBI (LRT = 0.00, p = 0.98). This suggests that females did not consistently give birth at one timing over the other across successive births. Moreover, female parity, rank and infant sex did not influence the proximity of birth timing in relation to each optimum (table 1).

Table 1.

Predictors of female reproductive timing. Estimates, confidence intervals, X² statistics and p-values of the predictors of the two linear mixed models (Models 3 and 4). The response variables are, respectively, the deviation from the offspring survival optimal birth timing, i.e. from 15 December (Model 3), and the deviation from the maternal IBI optimal birth timing, i.e. from 1 September (Model 4), in days, based on 215 births from 62 females. Female identity and year of infant birth are included as random effects. For categorical predictors, the tested category is indicated between parentheses.

fixed effect		estimate	IC		χ2	p-value
			lower	upper
Model 3: deviation from the offspring survival optimal birth timing
infant sex	(male)	5.91	−7.42	19.23	0.76	0.385
female parity	(primiparous)	−12.77	−30.02	4.47	2.11	0.147
female rank		2.59	−4.82	10.00	0.47	0.493
group	(L)	5.16	−10.37	20.69	1.33	0.515
	(M)	−12.66	−44.70	19.39
Model 4: deviation from the maternal IBI optimal birth timing
infant sex	(male)	−3.19	−16.46	10.08	0.22	0.7637
female parity	(primiparous)	9.67	−7.49	26.82	1.22	0.269
female rank		−3.41	−10.75	3.92	0.83	0.362
group	(L)	10.67	−4.70	26.04	1.92	0.382
	(M)	0.93	−30.91	32.78

(c) . Birth timing effects on maternal care and tantrum probability

In order to test whether maternal care may compensate for the costs of suboptimal birth timings in offspring, we investigated the effects of birth timing on the frequency of suckling and infant carrying. We did not find any effect of infant birth date on patterns of maternal care (electronic supplementary material, table S3). Further analyses revealed that mothers increase maternal care in the dryer winter months, but such compensation occurs regardless of an infant's birth date (electronic supplementary material, appendix 7 and table S4).

We also investigated whether infants born in suboptimal timings may beg for maternal care more frequently, looking at tantrum frequencies. We found that infants born near the maternal IBI optimal timing (i.e. between 1 August and 1 October; electronic supplementary material, table S2), were more likely to exhibit tantrums than other infants (electronic supplementary material, table S3; figure 3). Observation date did not affect tantrum frequencies, meaning that such an effect was independent of the season of observation (electronic supplementary material, table S4).

Figure 3.

Influence of birth timing on tantrum probability. Predicted values of tantrum probability (Model 7) at weaning (age 12 months), according to infants’ birth month, based on 2221 focal observations from 55 infants. For graphical reasons, and given the low sample size of infants observed for some birth months, we pooled infants born in two consecutive months, so that Jan indicates infants born in both January and February, Mar indicates both March and April, etc. The brown horizontal bars indicate the median values of fitted values for each birth month category. (Online version in colour.)

4. Discussion

We identify two distinct optimal birth timings in the annual cycle, respectively favouring current reproduction (offspring survival) and future reproduction (maternal reproductive pace). These are separated by four months, and the highest birth frequency occurs between these optima, indicating that mothers balance current and future reproduction, though closer to the optimal birth timing favouring offspring survival. Several reasons might explain why offspring survival might be prioritized over maternal reproductive pace. First, lifespan and offspring survival are the primary components of female lifetime reproductive success in long-lived species such as baboons, while reproductive pace may be less important [52]. Second, shorter IBIs might compromise infant survival independently of the effect of birth timing and are thus not necessarily adaptive [7]. Finally, the effect size of birth timing is greater on offspring survival than on maternal IBIs in our population.

More broadly, these results further our understanding of the evolution of vertebrate reproductive timing in several ways. First, trade-offs over birth timing may be widespread in long-lived species with slow life histories, for which development from birth to independence spans several months, therefore exceeding the length of the most productive season. In such cases, different stage(s) of the reproductive cycle may be synchronized with one or more seasonal food peaks, with the specific pattern dependent on the trade-offs females make among different fitness components [53]. Such variation could account for empirical cases where the observed birth peak fails to coincide with the birth timing expected on the basis of a single-fitness measure. For example, in humans from pre-industrial Finland, births did not concentrate in the months with the highest infant survival expectations [54]. More generally, such trade-offs may contribute to explain the partial or total lack of breeding seasonality observed in some large mammals [55], such as social primates including apes [15] and humans [56,57].

Second, while different species synchronize different stages of their reproductive cycle with the seasonal food peak [1,2,58,59], this study reveals variations in breeding timing within the same population. However, while mothers experience a trade-off between reproductive pace and offspring survival in their birth timing, it is not clear if particular individuals consistently favour certain strategies, as we did not detect any inter-individual effects of female identity, parity or rank on parturition timing. Instead, intra-individual factors, such as maternal reproductive history, may constrain the evolution of such individually based specializations, if only because the duration of IBIs—22 months on average but with extensive variation—prevents females from giving birth every 2 years at the same period of the year. In addition, birth timings may be affected by many external factors beyond female control, such as male reproductive strategies. Overall, the costs of waiting for the next optimal timing may often outweigh the costs of giving birth at suboptimal timings.

Third, this study underlines the importance of weaning to understand the evolution of mammalian reproductive schedules. Late weaning is most critical for infants, as they must learn to ensure their own provisioning. Matching that stage, which occurs between 12 and 18 months of age in this population, with the most productive season may substantially enhance infant survival (figure 1b) [43,60,61]. Moreover, the peak of lactation, which is the most energetically costly reproductive stage for mothers [3,43], typically coincides with the onset of weaning, occurring around six months after birth in this population. Matching lactation peak with abundant resources can alleviate the costs of reproduction and help to accelerate the transition to feeding independence by granting infants access to a wealth of weaning foods (figure 1b) [43]. It may contribute to explain the shorter IBIs associated with this birth timing. Such patterns may be very general. In the lemur radiation, for instance, despite a variety of life-histories, ecologies and societies, and the fact that different species mate and give birth at different times of year, all species synchronize weaning with the food peak [60]. Our understanding of the ultimate causes of mammalian reproductive seasonality may gain from granting more consideration to the dynamics and consequences of weaning, which may have been underappreciated in comparison to the energetic costs of pregnancy and lactation [1,2,4].

Fourth, our results show that the trade-off over birth timing faced by mothers may subsequently translate into mother–offspring conflict after birth. Although mothers adjust maternal care seasonally, they do so regardless of the offspring's age. Offspring born at suboptimal periods face the dry season in a critical developmental window (i.e. the end of weaning), and maternal care is insufficient to buffer them entirely from the adverse consequences that lead to higher mortality. Consequently, baboon infants respond by throwing more tantrums, which may be an honest signal of need [38,62], just as children do in similar situations [63]. More generally, these results shed light on the potential influence of environmental fluctuations, and specifically seasonality, on mother–offspring conflicts over maternal care. While the literature focusing on optimal birth spacing has mainly examined trade-offs between current and future reproduction for an implicitly stable level of resources [7,64], such a stability may rarely be encountered by mothers in the wild, who typically face extensive, but partly predictable, fluctuations in food availability. Taking into account the intensity and predictability of resource fluctuations may largely re-draw the landscape of strategic decisions available to mothers confronted with trade-offs between current and future reproduction in natural environments [65,66].

Our findings open new perspectives to understand the evolutionary drivers of vertebrate reproductive seasonality, by revealing the occurrence of a maternal trade-off between current and future reproduction over birth timing, extended by mother–offspring conflict during weaning. Such a trade-off may commonly occur in organisms with a slow reproductive pace, and future studies on such taxa should investigate the consequences of reproductive timing on several fitness components. Indeed, multiple optimal birth timings in the annual cycle may generate a bimodal birth peak or an extended birth season. Evolutionary trade-offs over birth timing may therefore account for unexplained variation in the reproductive timing of long-lived vertebrates, including the evolution of non-seasonal breeding in humans and other species.

Ethics

Our research procedures were evaluated and approved by the Ethics Committee of the Zoological Society of London and the Ministry of Environment and Tourism, Namibia (MET Research/Collecting Permits 886/2005, 1039/2006, 1186/2007, 1302/2008, 1379/2009, 1486/2010, 1486/2011, 1696/2012, 1786/2013, 1892/2014, 2009/2015, 2147/2016, 2303/2017, RPIV00392018/2019) and adhered to the ASAB/ABS Guidelines for the Treatment of Animals in Behavioural Research and Teaching.

Data accessibility

The datasets necessary to run analyses included in this paper and the associated legends have been deposited in the public depository: https://github.com/JulesDezeure/Maternal-trade-off-over-birth-timing-in-baboon.

Authors' contributions

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

Competing interests

We declare we have no competing interests.

Funding

Data used in this study are part of long-term data collected within the framework of the Tsaobis Baboon Project, recently funded by a grant from the Agence Nationale de la Recherche (ANR ERS-17-CE02–0008, 2018–2021) awarded to EH.