Harem-holding males do not rise to the challenge: androgens respond to social but not to seasonal challenges in wild geladas

The challenge hypothesis has been enormously successful in predicting interspecific androgen profiles for vertebrate males. Nevertheless, in the absence of another theoretical framework, many researchers ‘retrofit’ the challenge hypothesis, so that its predictions also apply to intraspecific androgen comparisons. We use a wild primate, geladas (Theropithecus gelada), to illustrate several considerations for androgen research surrounding male contests that do not necessarily fit within the challenge hypothesis framework. Gelada society comprises harem-holding males (that can mate with females) and bachelor males (that cannot mate with females until they take over a harem). Using 6 years of data from known males, we measured androgens (i.e. faecal testosterone (fT) metabolites) both seasonally and across specific male contests. Seasonal androgen variation exhibited a very different pattern than variation resulting from male contests. Although harem-holding males had higher testosterone levels than bachelors across the year, bachelors had higher testosterone during the annual ‘takeover season’. Thus, harem-holding males did not ‘rise to the challenge’ exactly when needed most. Yet, androgen profiles across male contests indicated that both sets of males exhibit the expected fT rise in response to challenges. Results from male geladas also support the idea that the context before (e.g. male condition) and after (e.g. contest outcome) a contest are critical variables for predicting hormones and behaviour.

2. Introduction

For nearly a quarter of a century, the challenge hypothesis [1] has been the predominant paradigm for explaining short-term androgen responses to social challenges for male vertebrates [2,3]. This framework, originally developed for interspecific comparisons among temperate birds [1], predicts that social challenges associated with mating effort cause circulating androgens to rise to a physiological maximum—far surpassing levels necessary for homeostatic function or basal reproductive physiology (e.g. spermatogenesis, libido, mating behaviour). Yet, because high androgen levels come at the expense of paternal care [4,5] and survival [6,7], the challenge hypothesis proposes that (where necessary) androgen modulation is extremely flexible. Thus, androgen levels can be up- and downregulated by the hypothalamic–pituitary gonadal axis as necessitated by competing mating and parenting demands (‘androgen responsiveness’). It follows, then, that androgen responsiveness should track both the breeding season (of the individual) and the social system (of the species). For the most part, this broad pattern has been upheld using interspecific comparisons of seasonal androgen profiles in birds and fishes [2,3], but see [8], although support deriving from mammalian species has been less conclusive [3].

Perhaps because the field of behavioural endocrinology generally lacks unifying theoretical hypotheses (for a few elegant exceptions see [9–11]), the challenge hypothesis framework has been called upon to explain almost any androgen variation in males, even in situations only tangentially related to social challenges (e.g. life-history androgen variation in primates [12]). We argue that this one-size-fits-all application of the challenge hypothesis to androgen research has placed unnecessary (and sometimes restrictive) boundaries on how we think about androgen variation in competitive situations. To be clear, our purpose is not to challenge the challenge hypothesis. Rather, we seek to expand the hormonal and behavioural variables that researchers report in their androgen research. There are (at least) three ways that researchers can extend their analyses of androgens and behaviour surrounding male contests.

2.1 Seasonal versus social challenges

First, in its original form [1], the challenge hypothesis did not distinguish between elevations in androgens in preparation for challenges from those in response to challenges. The term ‘androgen responsiveness’ suggests that we should be concerned with only the immediate neuroendocrine changes that follow male contests (hereafter, ‘social challenges’). Yet, a majority of studies claiming to support the challenge hypothesis draw their evidence from seasonal changes in testosterone levels [13]. These studies demonstrate that seasonal periods of territory establishment or mate guarding are associated with higher androgens than other times of the year (hereafter, ‘seasonal challenges’). Yet, in the absence of detailed behavioural data across the breeding season, seasonal measures may not necessarily result from an increase in social challenges. Indeed, other factors could easily be responsible for the rise in androgens, such as increased food availability and improved body condition [14]. Thus, it is not clear whether both social and seasonal hormone changes represent the same flexible neuroendocrine response to male contests.

To address this problem, Goymann et al. [13] examined all avian studies that report changes in androgens in response to a social challenge (i.e. an experimental or observed male contest) or a seasonal challenge (i.e. a known breeding season) and found that many species demonstrated unexpected differences in these two measures. For example, male blue tits (Cyanistes caeruleus) demonstrated a dramatic seasonal increase in circulating testosterone during the breeding season. Yet, when these same males were challenged with a simulated territorial intrusion, they failed to show the expected rise in testosterone [15]. Indeed, more than half of the bird species investigated showed the expected increase in testosterone to seasonal challenges, but failed to uphold the predictions of the challenge hypothesis when faced with rival challenges via simulated intrusions [16]. Although the reasons for these discrepancies remain a puzzle, these examples highlight the importance of distinguishing these two different situations.

2.2 A participant's perspective of the challenge

Second, the predictions of the challenge hypothesis do not consider whether individual males anticipate the upcoming challenge or how they interpret its outcome (reviewed in [17]). Arguably, the most important contextual piece of information regarding a social challenge is whether a male wins or loses the contest (or, more importantly, whether he perceives the encounter as a win or a loss). Yet, because the challenge hypothesis was originally formulated for comparing androgen modulation across species (not individuals), its purpose was to identify species-typical neuroendocrine responses to challenges, not individual differences to context-specific challenges. Therefore, we argue that research on intraspecific responses to challenges needs to consider the context of the interaction, including contest-specific variables (e.g. whether a male is on his ‘home turf’, is playing offence or defence, or has recently won or lost a contest [18]) as well as the outcome of the contest for each contestant [11]. Research on the ‘winner/loser effect’ suggests that changes in androgens across a social challenge depend entirely on the outcome, with androgens typically rising in winners relative to losers [19]. Furthermore, the increase in androgens in winners enhances perceived fighting ability and promotes engagement in future dominance contests, whereas the decrease in losers promotes withdrawal from such contests [11]. This underscores the well-known, but surprisingly understudied, idea that hormones and behaviour interact and feedback on one another, such that any one-directional relationship is overly simplistic [20]. This leads to the third problem of limiting ourselves to a challenge hypothesis framework.

2.3 Not all males are equal or equally likely to engage in challenges

Although male contests necessarily involve two (or more) males, the majority of research on the challenge hypothesis ignores how androgen profiles may influence the likelihood of males engaging in aggressive contests in the first place (although bystander androgen levels have been examined [21]). It stands to reason that not all challengers represent the same degree of threat, and thus, a challenger's relative competitive ability may not only predict how a challenged male responds (physiologically or behaviourally), but also predict whether he responds at all. For example, in an experimental study on wild chacma baboons (Papio ursinus), Bergman et al. [22] found that subject males only moved away from a ‘rival’ male (simulated using vocalizations played from a hidden speaker) when both males had high androgen levels. Indeed, androgen levels trumped other variables known to predict aggression in chacma baboons, such as dominance rank and rank disparity [23]. These results suggest that not all males may perceive themselves to be rivals—and that the hormonal profiles of males prior to contests may indeed provide valuable insights into what may ensue.

Here, we test the predictions of the challenge hypothesis while addressing each of these shortcomings using a dataset from a wild primate, the gelada (Theropithecus gelada). Geladas are an ideal model species for several reasons. First, gelada males have conspicuous social challenges where contestant males have clear roles (offence or defence) and the outcome of contests is unambiguous. Geladas are large-bodied, terrestrial primates that live in an open grassland [24,25], making it relatively easy to observe fights between rival males. These male fights mediate access to reproductive females that reside in harems (‘reproductive units’ [26]). Reproductive units comprise one harem-holding male (‘leader male’), 1–12 related females and their offspring, and occasionally one or more subordinate males (‘follower males’). Leader males engage in little to no parental care and mate with females throughout the year. Follower males are generally recently deposed leader males, but can also be young adult males that enter a harem as subordinates [27]. The leader male accounts for all reproduction in reproductive units with one male (‘one-male units’); however, follower males sire approximately 17% of offspring in multi-male units [26]. Leader males (often joined by follower males) fiercely guard their harems from ‘bachelor males’ that reside in all-male groups [24]. Importantly, bachelor males gain reproductive access to females only if they challenge and defeat a leader male (‘takeover’), or submissively enter as a follower [27]. During male contests, bachelor males are always on the offensive (i.e. they are the challenger males), and leader males are always on the defensive (i.e. they are the challenged males). Moreover, the winner always assumes the dominant leader male position, and the loser (if he survives) remains in the harem as a subordinate follower male or returns to a bachelor group.

Second, geladas offer an excellent opportunity for examining both (i) broad androgen profiles surrounding seasonal challenges, as well as (ii) individual androgen profiles across specific social challenges. The frequency of gelada contests exhibits strong seasonality, with a clear ‘takeover season’ spanning from February to June each year (figure 1). However, in contrast with most avian studies where an increase in contests occurs simultaneously with an increase in food availability [28], the majority of gelada contests occur during a period when food availability is the lowest (the end of the dry season). As we will explain, this distinction is valuable for separating out condition-based increases in testosterone from other factors that elevate testosterone. Furthermore, observations of social challenges between males (i.e. fights and takeovers) are frequent occurrences in this population (n=77 total takeovers 2006–2011; 0.35 takeovers per unit per year for one-male units and 0.27 takeovers per unit per year for multi-male units) [26]. Specifically, we have hormone and behavioural data from 30 takeovers for the 6 year study period.

Third, in many cases, we have data for both males (the challenger and challenged male) across the contest. For some takeovers (n=7), we have data from both males both before and after the contest. For others (n=23), we have detailed data from one of the males before the contest and both of the males after it (unless one of the males is killed, n=6).

Using the challenge hypothesis framework, we made the general prediction that challenges will be associated with an elevation in testosterone. However, we extended this investigation to include several of the shortcomings outlined above. First, we predicted that (i) elevations in androgens will be associated with both seasonal challenges (i.e. the takeover season) and social challenges (i.e. specific contests between a bachelor and a leader male). Second, with respect to social challenges, we predicted that (ii) in the months before contests occur, bachelor males that go on to challenge leader males will have higher testosterone profiles than bachelor males that do not challenge; and similarly, leader males that are challenged by bachelors will have lower testosterone profiles than leader males that are not challenged. Although we have observed a few unsuccessful takeovers, they are extremely rare in this population. We therefore restricted our dataset to include only successful takeovers. Last, we predicted that (iii) after a contest, winners (i.e. bachelors that become leaders) will exhibit an increase in testosterone and losers (i.e. leaders that become followers) will exhibit a decrease.

3. Methods

3.1 Subjects and study site

Data were collected from a population of geladas living in the Simien Mountains National Park, Ethiopia as part of the long-term University of Michigan Gelada Research Project. Over the 6 year study period (2006–2011), we collected behavioural, demographic and hormonal data from all adult males from 21 reproductive units and 13 all-male groups (n=133 males; including leaders, followers and bachelors). Throughout, we refer to dominant leaders and subordinate followers as ‘unit males’ to distinguish them from ‘bachelor males’. This population has been under near-daily observation since January 2006; and all males are individually recognized and habituated to observers on foot.

The Simien Mountains National Park encompasses an area of Afroalpine habitat (150 km², 3200–4500 m above sea level (a.s.l.)), including open grassland plateau and a few remnant forests. The region experiences pronounced ‘wet’ and ‘dry’ seasons each year [29,30]. The wet and dry seasons are variable each year, but generally occur during June–September (wet season mean monthly rainfall=310.8±30.3 mm; 2006–2011) and October–May (dry season mean monthly rainfall=38.0±7.6 mm; 2006–2011), respectively [29] (figure 2). Temperatures can approach freezing at night, but daily means range from 7.99°C±0.04 (mean minimum temperature, n=1843 days) to 17.66°C±0.07 (mean maximum temperature, n=1843 days). Rainfall and maximum and minimum temperatures are recorded on a daily basis using a rain gauge and digital thermometer centrally located in the gelada's home range [29].

3.2 Behavioural data collection

Status categories (leader, follower or bachelor) were determined by observations of group membership and dyadic dominance interactions. Among unit males, follower males are always subordinate to leader males, and (to the best of our knowledge) no dominance relationships appear to exist among leader males across reproductive units [24,31]. Therefore, we denote all dominant males ‘leaders’ and all subordinate males ‘followers’. Takeovers are conspicuous and (for the most part) discrete events where a bachelor challenges and subsequently overthrows a dominant leader male. After takeovers, former leaders have been observed to: (i) disappear (and presumably die, because we were unable to find these males during censuses of all surrounding groups), (ii) remain in the reproductive unit as a subordinate follower male, or (in rare cases) (iii) return to a bachelor group. A successful takeover ensued if the former leader exhibited submissive behaviours (e.g. fear barking, crouching, displacement, lip flips) towards the new leader (i.e. the former bachelor) and if the new leader copulated with females after the takeover.

Although the majority of males occupied only a single status category across the study (n=44 bachelors, n=10 followers, n=21 leaders), we observed several transitions between categories. Sixteen previously known bachelors became leader males, 22 leaders were deposed and became followers within their harem and five males transitioned from bachelor, to leader and finally to follower.

3.3 Faecal hormone collection and analysis

We collected faecal hormone samples in a targeted fashion from unit males; faecal samples were collected from leader (n=1376 samples) and follower males (n=684 samples) once per month across the entire study period (2006–2011). Faecal samples were collected from bachelor males (n=676 samples) opportunistically between 2006 and 2009 and 1× per male per month from 2010 to 2011. In total, we collected 2730 samples from 133 known males (approx. 14 samples per male; range 2–91 samples per male). In 2009, the manufacturer (Diagnostics Systems Laboratory) discontinued production of the testosterone antibody that we were using. Therefore, we had to employ two separate methods over the course of the study. In all analyses, we controlled for this variation and refer to these variables as ‘methods-based fixed effects’. Extraction and analysis of faecal testosterone (fT) metabolites and the validation of the old and new testosterone antibodies for use in geladas are described in the electronic supplementary material as well as elsewhere [12,29,32].

3.4 Do seasonal challenges influence male testosterone levels?

Given the seasonal pattern in gelada male contests (figure 1), it stands to reason that testosterone levels should rise for both the challenging bachelors and the challenged leaders during this period of increased competition. However, we needed to control for several factors that are known to affect male testosterone in other primates, such as weather, age and social status [33–36]. For rainfall, we designated each sample as a ‘wet’ or ‘dry’ season sample if the cumulative rainfall for the previous month was above or below the median (53.7 mm), respectively (n=1365 wet season, 1365 dry season). For temperature, we grouped samples based on whether they were collected during ‘hot’ days (max temperature>median maximum temperature, 17.5°C, n=397), ‘cold’ days (minimum temperature<median minimum temperature, 7.9°C, n=214) or ‘average’ temperature days (all other samples, n=2119). For age, we do not have known dates of birth for adult males, because gelada males disperse from their natal groups [37,38]. Thus, adult male ages were estimated to the nearest half year using secondary sexual characteristics such as canine eruption, tooth wear, pelage coloration and cape length [12,31]. For status, we split males into their broadest status categories (leader, bachelor, follower), and, to gauge overall competition across the population, we included the number of takeovers across our known harems for each month (range 0–5 takeovers per month).

These statistical analyses (and all subsequent analyses) were conducted in R (v. 3.0.3) [39]. Using the function ‘lmer’ in the lme4 package [40], we ran a linear-mixed model (LMM) with fT as the outcome variable to determine the effect of weather and social factors on male testosterone. We log-transformed the outcome variable (testosterone in ng g⁻¹) in this and all subsequent analyses to approximate a normal distribution. Because each individual male had multiple data points (n=2730 samples from 64 leaders; 52 followers, 79 bachelors; range 2–91 samples per male), we included individual identity as a random effect. In addition to our methods-based fixed effects (see the electronic supplementary material) we included rainfall, temperature, male age, status and number of takeovers per month as additional fixed effects on testosterone levels. We compared univariate models that considered only a single fixed effect to multivariate models that considered a combination of fixed effects and/or interactions between fixed effects. We compared all candidate models using Akaike information criterion (AIC) and considered the model with the lowest AIC to be the best fit for our dataset [41,42]. If the difference in AIC was less than 2 for the lowest-ranked models, we considered both models to be equally good fits for the data [43]. For all LMMs, we visually inspected each model using a Q–Q plot, histogram of residuals and scatterplot of fitted versus residual values. For all models, residual values were normally distributed.