Androgens such as the male sex hormones testosterone (T) and 5αdihydrotestosterone (5α-DHT) are important mediators of competitive and reproductive behavior in a wide variety of taxa, including many mammals (Archer & Carré, 2016;Boonstra et al., 2017;Gettler et al., 2011;Schradin et al., 2009;Surbeck et al., 2012). These androgens have significant health and fitness consequences, both via behavioral pathways (e.g., by increasing the likelihood of engaging in risky or aggressive behaviors: reviewed in Muller, 2017) and physiological ones (e.g., via interference with immune or oxidative stress responses: Alonso-Alvarez et al., 2007;Foo et al., 2017;Muehlenbein & Watts, 2010;Salvador et al., 1996). Information about the relationship between androgens and socioecological factors can provide critical information on the physiological mechanisms that contribute to a wide variety of phenomena of interest to evolutionary biologists and ecologists, including dominance hierarchies (Prall & Muehlenbein, 2017;Rincon et al., 2017;von Engelhard et al., 2000), contest competition (Geniole et al., 2017;Wingfield et al., 1990), and cooperation (Trumble et al., 2015).
The challenge hypothesis, which predicts that males should upregulate testosterone production when competitive abilities are important, and downregulate it when they are not, is currently the dominant paradigm for understanding the observed variation in testosterone production across diverse vertebrate species (Wingfield et al., 1990). Originally developed in the context of understanding seasonal testosterone variation in birds who first compete for mates, then invest in parenting, it has subsequently been applied more broadly, including to many primates (Cavigelli & Pereira, 2000;Cristóbal-Azkarate et al., 2006;Girard-Buttoz et al., 2009;Marshall & Hohmann, 2005;Mendonça-Furtado et al., 2014;Morino, 2015;Ostner et al., 2002;Rincon et al., 2017;Rosenbaum et al., 2018).
Though specific findings vary considerably across species, past work has generally provided support for the broad applicability of the model. For example, Muller and Wrangham (2004) found that chimpanzees produced more testosterone when parous females (their most preferred sexual partners) had maximally tumescent sexual swellings, while Setchell et al. (2008) reported that male mandrills' androgen levels were higher when their dominance hierarchies were unstable.
However, even if theory predicts (and empirical evidence confirms) that greater male-male competition should correlate with higher androgen levels, intrasexual competition in primates is not a unitary phenomenon. In the many species in which males face both intra-and intergroup competition, it is not clear which should be more important, and thus to what degree group structure versus relative position within a group (i.e., dominance rank), might predict androgen concentrations. Males who live in multimale, multifemale groups face considerable intrasexual competition from in-group males. However, they also have the benefit of potential intragroup alliance partners when they face challengers (e.g., Berghänel et al., 2011;Dal Pesco, 2020;Duffy et al., 2007;Noë & Sluijter, 1995), and the outcome of any particular contest may be relatively low stakes, compared to their counterparts who live in single-male groups. Mating opportunities may not disappear entirely even if they are knocked further down the dominance hierarchy or forced to emigrate to a new social group. While there is certainly variation in the degree of reproductive skew from species to species, it is rarely the case that only dominant males reproduce (e.g., Alberts et al., 2003;Engelhardt et al., 2017;Minkner et al., 2018;Surbeck et al., 2017;Vigilant et al., 2015). Males in single-male groups do not face the continual intrasexual pressure that those in multimale groups do, but when they do have competitive encounters, the stakes are likely to be very high (Fernandez-Duque & Huck, 2013;Saj & Sicotte, 2005;Zhao et al., 2011). They have no potential male allies, and losing their position as the sole male in a group could easily spell the end of a reproductive career.
While androgens are expected to be important mediators of male competitive behavior in most if not all primate species (Muller, 2017;Prall & Muehlenbein, 2017), theory predicts that they will be particularly important in the subset of species among which there has been strong selection for male contest competition, that is, where males compete directly for exclusive control of reproductive access to groups of females (Plavcan, 2012). The genus Gorilla is an excellent example. Males invest heavily in morphological traits that improve their ability to compete with other males. They are more than twice the size of females and well-equipped for fighting, with large muscles, canines, and sagittal crests (Harcourt & Stewart, 2007;Plavcan, 2001). These morphological characteristics strongly suggest a long history of male contest competition, and thus of one-male, multifemale groups.
Western lowland gorillas (Gorilla gorilla gorilla) do indeed live almost exclusively in such groups, but closely-related mountain gorillas (Gorilla beringei beringei) exhibit a puzzling level of social flexibility given their morphology (Robbins & Robbins, 2018). The world's two remaining populations routinely contain both singlemale and multimale groups, as well as solitary males, and historically all-male groups have been observed as well (Caillaud et al., 2014;Gray et al., 2010;McNeilage et al., 2006;Robbins & Robbins, 2018;Robbins, 1995;Yamagiwa, 1986Yamagiwa, , 1987)). The drivers of this variation are poorly understood, but they make mountain gorillas attractive subjects for research questions about the relationship between androgens and various aspects of male competition.
There is considerable speculation in the literature about the costs and benefits of different group structures in mountain gorillas, as well as how such variability is maintained (Robbins & Robbins, 2018;Robbins et al., 2016). Variation in the androgen concentrations maintained by males in different types of social groups could provide important insights into both questions. For example, if males in multimale groups have the highest androgen concentrations, this would suggest that the sustained intragroup competition that these males face may have specific health (and ultimately fitness) costs that males in other group structures do not bear. In theory, this could at least partially offset the benefits of living in multimale groups. These benefits include better female retention and lower infant mortality, along with an upper hand in conflicts with other groups (Robbins et al., 2007(Robbins et al., , 2009)). Conversely, higher androgen concentrations could also be a mechanism by which males in single-male groups are able to gain and maintain exclusive access to females. Finally, while there should be clear advantages and few downsides to high androgen concentrations for solitary maleswho have strong incentives to compete and would pay no social costs for aggressive behavior-perhaps lower androgen concentrations would help explain why they are solitary in the first place.
Despite the long history of research on mountain gorillas, which have been studied nearly continuously since 1967 (Fossey, 1983;Robbins et al., 2005), prior research on androgens and social behavior in this species is quite limited. What research does exist has yielded inconsistent findings. Data gathered in the 1990s suggested that higher-ranking males in multimale groups had higher urinary androgen concentrations than lower-ranking ones (Czekala & Robbins, 2001;Robbins & Czekala, 1997). However, a replication of that analysis using double the number of urine samples (n = 670) failed to replicate this finding (Rosenbaum et al., 2020). Data from the more recent study also suggested that solitary males might have lower urinary androgen concentrations than males who live in social groups, but the sample size for solitary males was too small to draw meaningful conclusions. Small sample sizes are an inevitable byproduct of the use of urine samples for measuring hormones in mountain gorillas, since urine is difficult to collect in terrestrial primates.
The conflicting findings on the relationship between rank and urinary androgen metabolites highlight another issue: existing research has not conclusively established whether the androgen metabolites being measured are the ones most relevant to the questions under consideration. While it is likely that they are, androgens are a broad class of hormone produced by both the hypothalamicpituitary-adrenal and hypothalamic-pituitary-gonadal axes (Melmed et al., 2016), and only some are germane to the questions of interest.
Hormone metabolites excreted in bodily waste are not structurally identical to the biologically active, native hormone they are derived from, so it is important to establish which metabolite(s) a given assay captures (e.g., Monfort et al., 1997;Touma & Palme, 2005). The lack of validation in previous work has made it challenging to interpret past findings and the inconsistencies observed across studies.
Fecal samples are far easier to collect from gorillas than urine is, which permits much larger sample sizes (Eckardt et al., 2016). Before fecal sample analysis is adopted for use, however, we need to establish whether a given assay can detect fecal metabolite(s) derived from the androgens of interest (e.g., gonadally derived testosterone and 5a-DHT, which are highly anabolic and thus likely the androgens which are most relevant to questions about rank and competitive ability: Eckardt et al., 2016;e.g., Palme, 2005;Touma & Palme, 2005).
While total exclusion of other closely-related androgen metabolites (e.g., those that are derived from adrenal androgens) is unlikely, the goal is to find an assay that can detect the relevant metabolites with enough specificity to answer the kinds of questions that are typically posed in behavioral research.
In this study, we have three goals. First, we use high performance liquid chromatography (HPLC) to demonstrate that a commonly used testosterone immunoassay (T-EIA using antiserum R156/7, CJ Munro, UC-Davis) detects androgen metabolites of interest (e.g., testosterone, 5α-DHT) in mountain gorilla fecal samples.
Next, we evaluate whether this assay captures acute changes in FAM in response to social challenges related to competition and mating opportunities. Building on this validation work, we then use FAM measured in a large number of samples to examine the relationships between FAM and group structure, and FAM and dominance rank, in male mountain gorillas. Our study capitalizes on samples collected at different points in time. The first data set was collected between August 2003 and December 2004, when the gorillas lived in three large, multimale, multifemale groups (Table 1). One adult male in group SHI (age 18 as of the midpoint of the study) was not included in our analysis because he never developed secondary sex characteristics, which may have been an indicator of abnormal androgen levels. The 2003The -2004 data set also contained samples from one solitary male who dispersed from group SHI early in the study and was regularly reencountered over the following year. The second data set was collected between April 2011 and December 2012, and contains data from four multimale and three single-male groups (Table 2), as well as from five solitary males which the tracking and research staff encountered opportunistically in the forest. We knew the identities of three of these solitary males because they were habituated animals that had dispersed from KRC-monitored groups. In the other two other cases, their identities were unknown. One individual contributed 15 of the 19 total samples from solitary males, with a single sample from each of the other four.
Half (n = 11) of the animals in the 2003-2004 sample also contributed data to the 2011-2012 sample, so the total number of individuals included in our analyses is 32. Four of the animals who appear in both data sets successfully started their own mixed-sex groups in the gap between the two study periods. Two previously lower-ranking males became alphas in multimale groups after older males died, while two others became solitary males. The remaining three held dominance positions in multimale groups similar to the ones they held in 2003-2004. We determined males' dominance ranks in multimale groups using displacement patterns, as described in Stoinski et al. (2009).
There were not always a sufficient number of displacements to determine the specific dominance hierarchy position of some of the lower-ranking males. Therefore, we distinguish between the top three positions (alpha, beta, gamma), and place all males ranked fourth or lower into a single category called subordinate. Typically, older males are dominant over younger ones. During our initial data analysis, we also tried using Elo ratings instead of ordinal ranks (Neumann et al., 2011;Wright et al., 2019). However, since use of this variable decreased the R squared values of the models and would also limit comparability to previous research, Elo ratings were dropped from subsequent analyses.
Our analyses included males ages 10 and older, with the exception of an explicit comparison between an immature male and our adult male subjects (see Section 2.4 below). Though their secondary sex characteristics begin to visually distinguish males from females between eight and 10 years old and they can begin siring offspring as young as age nine (Rosenbaum et al., 2015;Vigilant et al., 2015),
1 Study group composition and sampling efforts in each group in 2003-2004 Group Number of fecal samples Total number of males ages 10+ (number included in current data) Group size a (range) Mean samples per male b BEE 283 7-8 c (8) 26.0 (26-26) 35.4 (SD = 17.6, range = 2-61) SHI 167 7-9 d (7) 25.0 (23-27) 23.9 (SD = 12.1, range = 2-42) PAB 139 6-8 e (6) 54.9 (52-58) 23.2 (SD = 7.9, range = 10-36) Solitary 42 1 f (1) 1 (NA) 42 Total 631 21 NA 30.0 (SD = 13.9, range = 2-61)
a Mean across entire period of data collection.
b In cases where the data set does not contain samples from all of the males in the group (i.e., SHI and PAB), means and SDs were calculated using the number of males included in the data. c Group BEE contained a male that turned 10 years old during data collection.
d Group SHI contained an 18-year-old male who never developed secondary sex characteristics (a possible sign of abnormal androgen production) and was thus excluded from our data. Two subordinate males dispersed from the group during data collection. This data set contains samples from one of them, but not the other.
e Two males in group PAB turned 10 during the final 2 months of data collection. Neither of these males are included in this data set. f The solitary male was initially a member of group SHI before dispersing. Because the same animal appears in both group SHI and as a solitary male, the total number of animals is 21. Because he was the only animal not living in a multimale group in this time period, samples collected while he was a solitary male are used strictly for a descriptive comparison.
T A B L E 2 Study group composition and sampling efforts in each group in 2011-2012 Group Number of fecal samples Total number of males ages 10+ (number included in current data) Group size a (range) Mean samples per male b BWE 122 1 (1) 9.6 (9-11) 122 INS 95 1 (1) 5.7 (4-8) 95 URU 79 1 (1) 6.2 (5-7) 79 ISA 159 2 (2) 11.7 (10-12) 79.5 (SD = 1.5, range = 78-81) KUY 229 2-3 c (3) 13.8 (13-15) 76.33 (SD = 48.61, range = 9-122) PAB 232 9-12 d (7) 45.1 (43-49) 33.14 (SD = 22.58, range = 2-60) TIT 44 4 e (2) 7.0 (7-7) 22.0 (SD = 0, range = 22-22) Solitary 19 5 (5) 1 (NA) 3.80 (SD = 5.60, range = 1-15) Total 979 22 NA 46.95 (SD = 40.06, range = 1-122)
a Mean across entire period of data collection.
b In cases where the data set does not contain samples from all of the males in the group (i.e., PAB and TIT), means and SDs were calculated using the number of males included in the data. c Group KUY contained three males for the first 3 months of data collection, and two males for the remainder. The beta-ranked male dispersed and was later found dead. The formerly gamma-ranked male therefore holds two ranks in our data (gamma before the dispersal/death, and beta after). d Group PAB contained between 9 and 12 males who were at least 10 years old over the course of data collection, representing 13 different individuals.
Eight were present for the whole study, two dispersed from the group 5 months into the study, 1 dispersed 12 months into the study, and 2 turned 10 years old during the study. Our data contains samples from 7 of the 13 males (four who were present the whole time, including the alpha-, beta-, and gamma-ranked males, and the three subordinates who dispersed).
e Our data contains samples from two of the four males (alpha and beta ranking).
male gorillas are not fully physically mature until ~15 years old (Galbany et al., 2017). We included males 10 and older to facilitate comparison with the few other existing papers on androgens in wild mountain gorillas (Robbins & Czekala, 1997;Rosenbaum et al., 2020).
Trained observers who were tested on their ability to identify individual animals collected all the fecal samples. When an animal was observed defecating, the collector waited for them to move away, then picked up the sample using a plastic bag or glove turned inside out over their hand. We only used samples that were not contaminated with rainwater, urine, or other gorillas' fecal material. All samples were collected between 7:00 a.m. and 4:00 p.m., though because fecal samples contain hormone "pools" that represent hours or even days, there is no relationship between collection time and fecal steroid metabolite concentration in mountain gorillas or other species with long gut passage times (Eckardt et al., 2016;Millspaugh & Washburn, 2004;Nizeyi et al., 2011).
The number of samples varies from animal to animal due to the many uncontrollable factors involved in sample collection from wild animals. In 2003 and 2004, samples were collected opportunistically.
In 2011 and 2012, samples were collected following a protocol for a study on physiological stress (Eckardt et al., 2016(Eckardt et al., , 2019)). As part of this protocol, we attempted to collect weekly "baseline" samples from each animal, as well as all possible samples in the 6 days following a designated event (e.g., an intergroup interaction, fight, female transfer, or other occurrence that theory would predict is related to changes in stress hormones and/or androgens; see Section 2.5 below). We refer to these samples as "event samples."
There were a small number of samples (n = 30) that we could not conclusively assign as baseline because of failure to contact the social group on a prior day, though we had no reason to believe that any event had occurred (e.g., no animals with visible injuries, excessive trampled vegetation, paths crossing those of other social groups, etc.). We tested our initial models with these samples assigned first as baseline samples, then as event samples. It made no difference to the results, so they are classified as baseline in these analyses. To improve the comparability of the 2003-2004 and 2011-2012 data sets, we retroactively retrieved information on the same set of events in 2003-2004, and thus were able to designate samples from 2003 to 2004 as baseline or event as well.
Sample collection from solitary males, who are not routinely monitored, always occurred opportunistically. For solitary males, samples were designated as event samples if we collected them on or in the 6 days after an event we knew they were involved in (e.g., an interaction with a study group). If they were encountered in the forest while we were searching for other animals, the sample was treated as a baseline. However, for these males it is very difficult to conclusively say whether they might have experienced an androgen-relevant event in the prior 6 days, since there is a much higher likelihood of unobserved events. with samples from this gorilla population (Eckardt et al., 2016;Santymire & Armstrong, 2010). After the ethanol-based extraction, samples were dried down, capped, and returned to the -20°C freezer until shipment to the Davee Center. At the Davee Center, all samples from both points in time were stored at -20°C until reconstitution and analysis.
We measured androgen concentrations in our samples with an inhouse testosterone enzyme immunoassay (EIA) at the Davee Center.
This assay is commonly used for androgen quantification in various species, and with various biological matrices (polyclonal antiserum R156/7, provided by CJ Munro, UC-Davis) (Edwards et al., 2015;Emery Thompson et al., 2012;Herrick et al., 2010;Jacobs et al., 2014;Jaeggi et al., 2015;Narayan et al., 2013). Biochemical validation revealed parallelism between the binding inhibition curves of fecal extract dilution (range, 1:80-1:10,240) and testosterone standard (difference between slopes in the 20%-80% binding range: t = 0.392, p = .708), as well as significant recovery (>90%) of exogenous testosterone added to the fecal extracts (2.3-300 pg/50 ml;
The T-EIA sensitivity was 46 pg/ml, intra-assay coefficients of variation (CV) were <10%, and the interassay CV was <15%. Loeding et al. (2011) and Santymire and Armstrong (2010) previously published cross-reactivities for this assay. Metabolite concentrations are expressed as ng/g of wet fecal material.
We validated our testosterone assay biologically and biochemically. For the biological validation, we evaluated the individual FAM responses of male mountain gorillas to specific events that theory predicts should lead to increased androgen levels-that is, events that involve male-male competition, reproductive opportunities, or both. We also biochemically validated the assay using HPLC. Data for the event analysis came from four males who lived in four different social groups in 2011 and 2012. Three lived in multimale, multifemale groups, and one lived in a single-male group. The hormone samples for the HPLC pool came from five different males living in four different groups. Four of the five were living in multimale groups, while the fifth was in a singlemale group. One of the males who was included in the biological validation analysis was included in this pool, while the other four were not. The males chosen for the HPLC pool were all animals whose FAM concentrations were known to be higher than average, since HPLC analysis requires high concentrations of the metabolite(s) in question (see Section 2.7 below).
As a secondary biological validation, we examined the difference in the FAM concentrations of an immature male for whom we had comparable results (age 4.4-4.8 when his fecal samples were collected)
versus the older males included in our other analyses. If the assay captures metabolites of interest, then the immature male should have lower FAM concentrations than older males. However, due to the complex nature of measuring hormone metabolites, we do not consider our study to be a validation of the use of this testosterone assay for juvenile male mountain gorillas. Most experts recommend separate validations for animals of different age/sex classes, since the EIA could be capturing different metabolites in different classes of animals (e.g., Goymann, 2012;Palme, 2005;Rincon et al., 2019).
We chose the samples for the primary biological validation from samples collected during the month before and the month after a behavioral event (e.g., a fight involving the study males or a female transfer; events are described in detail in Section 2.6). We only used 2011-2012 samples for the biological validation, since the collection protocol for this time period, which involved the intensive post-event sampling described in Section 2.2, was more appropriate for detecting post-event changes. Samples before an event were considered "baseline" samples, where no event that theory predicts might significantly affect fecal androgen concentrations was observed in the prior 6 days (e.g., no intergroup interactions, group fission events, evidence of females in estrous, or fights that resulted in wounding; minor disputes that do not involve injuries are part of gorillas' daily lives, and thus were not considered). Due to variation in lag time from hormone production in the body to metabolite excretion in fecal material, we would expect any event-related "spike"
in FAMs to appear in a sample or samples in a 20-140 h window after the event, based on our prior work with fecal glucocorticoid metabolites (Eckardt et al., 2016).
Starting the day after the event, we included all available samples, until the point at which the value of at least two postevent samples was no more than 1.5 times the value of the highest pre-event sample, with a minimum end point of 6 days post-event (Eckardt et al., 2016). Because animals are not monitored 24 h a day, we cannot be certain that unobserved androgen-relevant events did not occur. Due to variation in weather, vegetation, animal temperament, observer coverage, and the ability to find specific individuals, the number of samples both pre-and post-event varies from male to male, and event to event.
All of the samples used for HPLC were collected in 2015 and 2016. The biological and biochemical validations use data from two different points in time because we needed detailed information about behavioral events for the biological validation (available in 2011 and 2012), but samples that had never been reconstituted in buffers (whose salts interfere with HPLC) for the biochemical validation. These were only available for more recently collected (i.e., 2015 and onward) fecal samples. To ensure an adequate concentration of metabolite for HPLC, the samples included in the HPLC pool were the 20 highest-androgen samples (from n = 5 males, see Section 2.4 above) from an initial pool of 45, as measured using our T-EIA. The original pool of 45 contained samples from 14 different males who were selected because previous EIA analyses had shown that at least some of their samples had high FAM concentrations.
For comparison of the immature male to older males, we used all available analyzed immature male samples that were run on the same assay (immature male: n = 1 animal and 14 samples) and compared them to the largest of the adult male (10+ years) data sets used here (the group type/FAM analysis: n = 22 animals and 979 samples). This validation used data from 2011 to 2012 only, as we did not have samples from immatures in 2003-2004.
We used four discrete reproductive and/or competitive events that occurred in four different months across 2 years. Mountain gorillas breed year-round (Watts, 1998), and therefore there should be no seasonal differences in males' sex hormone levels or hormone responses to events. Our selected events are qualitatively representative of the types of competitive and mating events that male mountain gorillas experience, and cover both intragroup and extragroup social dynamics. We briefly describe each of the four events below.
2.6.
1 | Event one: Intragroup fight On February 22, 2012, a subordinate male in PAB group mated with an adult female. During the copulation, he was attacked by the betaranked male in his group, animal GIC. The conflict included contact aggression, though neither animal sustained any visible injuries. Our analysis includes eight samples collected between January 31 (22 days before the fight) and March 15 (22 days after the fight). All samples are from individual GIC, the beta-ranked male who instigated the fight. 2.6.2 | Event two: Intergroup interaction with female transfer On August 21, 2011, INS group interacted with PAB group. During the interaction, a nulliparous adult female transferred from PAB group to INS group. Our analysis includes 10 samples that were collected between August 4 (17 days before the event) and August 27 (6 days after the event). All samples are from individual INS, the only adult male in the group the female transferred to. 2.6.3 | Event three: Leading temporary subgroup On January 4, 2012, PAB group split into two subgroups. One
was led by the group's dominant male, while the other was led by the gamma-ranked male, MSK. The subgroup led by MSK contained another adult male, four blackbacks (males between the ages of 8 and 12 (Watts & Pusey, 1993)), seven adult females, three juveniles, and five infants. Our analysis includes nine samples that were collected between December 12th (23 days before the event) and January 24th (20 days after the event). All samples are from individual MSK, the adult male who led the temporary subgroup. We predicted that this subgrouping event would be associated with an androgenic spike for this male since it meant he temporarily rose to the top of a dominance hierarchy, and had unrestricted access to females without the usual threat of interference from the two higher-ranking males in his natal group.
On May 25th, 2011, two males from ISA group left the females and infants in their group behind and approached KUY group. These two males fought with three males from KUY group; all three males in KUY sustained superficial injuries. Our analysis includes seven samples collected between May 10th (15 days before the fight) and June 3rd (9 days after the fight). All samples are from individual KBH, the beta-ranked male from ISA group.
We used HPLC to isolate the fecal metabolite(s) captured by our T-EIA. The HPLC sample pool was created by reconstituting the 20 high-androgen samples described in Section 2.5 in methanol (MeOH), then combining 500 µl of neat sample (1:5 dilution) from each one into a single tube. This sample pool was dried down, reconstituted with half the volume (5 ml) of MeOH, vortexed for 30 s, then rotated on a tube spinner for 30 min to ensure that all the metabolites were released from the sides of the tube. The mixture was again dried down, and we repeated the same procedure first with 2.5 ml, then with 1 ml MeOH. At this point the sample was also shaken for 20 min at 60 rpm after the vortexing and spinning steps. The final 1 ml of MeOH was evaporated off, and the dried sample pool capped and sent to the Smithsonian Institute for Conservation Biology in Front Royal, VA.
Our Smithsonian colleagues performed the HPLC analyses with an elution method described in (Young et al., 2004). Briefly, a Novapak C-18 column was primed first with distilled H 2 O, then MeOH.
The sample pool was spiked with four radioactive hormone tracers (8000 counts/50 µl, see following two paragraphs), reconstituted using phosphate buffered sodium (PBS; pH 5), then filtered through the primed column to clean it. This was followed by citrate buffer, Since this T-EIA has a known high (57%) cross reactivity with 5α-DHT, in addition to a testosterone tracer, we also included a 5α-DHT tracer. 5α-DHT is a testosterone derivative that is a more potent androgen receptor agonist, and both are of high biological relevance.
Elution of the assay-captured fecal metabolites in the ranges of the testosterone and/or 5α-DHT tracer peaks are evidence that the metabolites in question are byproducts of these androgens.
In addition, we also included cortisol (cort) and corticosterone (cc) tracers. Previous work on various species has demonstrated correlations between changes in androgens and glucocorticoids (Girard-Buttoz et al., 2009;Hunt et al., 2006;Liening et al., 2010;Lynch et al., 2002;Suay et al., 1999). There are many situations in which they should theoretically rise simultaneously, since increased corticosteroid production would also be beneficial during the competitive events in which we expect androgen production to increase.
Our prior validation paper demonstrated glucocorticoid spikes after events similar to the ones used in this analysis (Eckardt et al., 2016).
However, there are concerns over whether cross-reactivity between corticosteroid and androgen assays might play a role in such findings (e.g., Pribbenow et al., 2017), and our interest here is in identifying androgen metabolites specifically. Elution of the assay-captured fecal metabolites in the ranges of the cort and/or cc tracer peaks would be evidence that the metabolites in question are byproducts of these hormones, rather than androgens. We refer to any immunoreactivity that occurred between the start of the authentic cort peak and the end of the authentic cc peak as corticosteroid-like, and any that occurred between the start of the authentic T peak and end of the authentic 5α-DHT peak as androgen-like.
To identify peaks in the HPLC data, we calculated the mean of all observed immunoreactivity that fell above the assay detection To answer the question "What is the relationship between dominance rank and FAM?," we used data from the three multimale groups available in 2003-2004 (Table 1), and the four multimale groups available in 2011-2012 (Table 2). In each case, we ran a linear regression model clustered on individual animal identification. The predictors in these models included rank (an ordinal variable where the reference category was subordinate males), a categorical age variable (coded so that males 10-12 years = 0, 12-14 years = 1, and 15+ = 2), the ratio of adult males (ages 10+) to adult females in the group (where < 1 indicates more females than males, and >1 indicates more males than females:
2003-2004 male:female ratios, mean = 0.80, SD = 0.29, range = 0.3-1.29;
2011-2012, mean = 0.74, SD = 0.71, range = 0.33-4.00), and the number of adult males in the group on the day the sample was collected (see Tables 1 and 2). We also included a dummy variable that indicated whether a sample was baseline or event, where 0 = baseline and 1 = event. The number of event samples varied from animal to animal Collinearity between age and rank-specifically driven by the large number of young, subordinate males (see Table 1)-was extremely high in 2003-2004 (correlation coefficient = 0.91 when age is treated as a continuous variable, 0.80 when categorical), as were male:female ratio and number of males per group (correlation coefficient = 0.83). In the 2011-2012 data, these correlations were much lower, especially between male:female ratio and number of males (age/rank correlation coefficient = 0.63 when age is treated as a continuous variable, 0.73 when categorical; male:female ratio/number of males = 0.16).
Such collinearity is a notoriously sticky problem, and there are conflicting opinions on how best to deal with colinear variables. The larger standard errors that collinearity introduces reduces the chances of identifying a real relationship between our variables of interest (in this case, rank and FAMs). To allow the reader to draw their own conclusions about the relative effects of these variables, we present the results of models that contain both variables (rank and age category) in the main text, and the results of models that exclude one or the other, as well as versions that limit the analysis to the oldest category of males (15+, when males have fully developed secondary sex characteristics), in the Supporting Information. Since age categorization increased the collinearity between age and rank in the 2011-2012 sample, we also report results from a model that used the continuous age variable instead of the categorical variable, for this subset of the data. In instances where the results of the models in the main text and the Supporting Information differ, we discuss these cases in the main text. In general, in places where delineating the effects of colinear variables is challenging, 2011-2012 data should be prioritized over 2003-2004, since it was less of an issue in this data set.
We experimented with using the percent of the group that was adult males in place of the male-to-female ratio, and with using total group size in place of total number of adult males. Both were highly correlated with the respective original variables, did not improve model fit, and did not result in any substantive changes to our results, so they were dropped from consideration. Including a categorical variable that identified social groups resulted in the model dropping the number of males in the group due to collinearity. We therefore excluded a categorical group identifier, since we were interested in the relationship between properties of the group (e.g., male female ratios, or number of males) and FAM, rather than in the identification of the group itself.
To answer the question "What is the relationship between group structure and FAM?" we used data from 2011-2012, when we had data available from multi-and single-male groups, and from multiple solitary males. We first used all available groups and males, then limited the sample strictly to "leader males" (i.e., the alpha males in multimale groups, males in single-male groups, and solitary males). We also provide a qualitative comparison of the only solitary male for whom we had data in 2003-2004 to the males living in multimale groups during that time period, but the lack of variation in group structure precludes formal analysis.
The group structure model that included all males was structured similarly to the rank models described above, except that we replaced the rank variable with a categorical variable that identified whether the male was solitary, lived in a single-male group, or lived in a multimale group, with multimale set as the reference category.
Rank was eliminated because solitary males and males in one-male groups do not have dominance ranks. Distributions of the values of the predictors were similar to those in the rank model (male:female ratio: mean = 0.63, SD = 0.62, range = 0-4; event samples per animal: mean = 20.14, SD = 21.05, range = 0-76; age: mean = 19.31, SD = 4.72, range = 11.99 = 33.58; correlation between male:female ratio and number of males per group = 0.27). For the group structure model that specifically compared leader males, we eliminated all variables except age and group type, since the other variables (number of males, male-female ratios, number of event samples)
were functionally uniquely identifying a single male. We did not include a group type × event sample interaction for solitary males, due to the tiny number of baseline samples (n = 3) and the uncertainty over whether they were, in fact, baseline samples. However, interaction terms are reported for other social group types. An interaction between the two would suggest that males in one particular group configuration have stronger FAM responses to events than males in another.
For the two unidentified solitary males included in the 2011-2012 sample, whose ages were unknown, we experimented with setting their ages to 12 (very young) and 30 (very old). This made no difference to our results, so their ages were set to the mean age for all males in the 2011-2012 sample. For all solitary males, male-female ratio was set to zero, and number of males was set to one. Our exploratory analyses indicated that neither date nor time of collection, nor the number of hours that elapsed between collection and freezing predicted FAM concentrations. Therefore, those variables were eliminated from the models presented here.
The modeling approach we used results in multiple tests on the same data set. To account for this, we generated a vector of the p values from our models and calculated sharpened false discovery rate q values for all results within a given data set (i.e., values were computed for 2003-2004and 2011-2012separately: Storey, 2003)). Results tables and the text report these q values instead of the original p values generated by the models.
We checked all models to confirm that they generally conformed to basic residual distribution assumptions, though the clustered, robust standard errors also have the advantage of relaxing homoscedasticity assumptions. All analyses were done using Stata 16.
Here we present a biological validation of the T-EIA. First, we examine the fecal androgen (FAM) responses of individual males to events that theory predicts should cause increased androgen production; next, we compare adult male FAM values to those of an immature male. Events, subjects, and sampling procedures are described in detail in Sections 2.4, 2.5, and 2.6.
In the 22 days before an intragroup fight in PAB group, the betaranked male had a mean FAM concentration of 197.5 ng/g (range:
153.15-237.25). The day after the intragroup fight that he instigated in response to a subordinate male mating with a female, his FAM concentration was 539.28 ng/g, a 2.7-fold increase over the prior days' mean value (Figure 1a). By day 5 postevent, his values were ≤1.5 times pre-event FAM concentrations.
In the 18 days before an intergroup interaction, the only male in INS group had a mean FAM concentration of 179.49 ng/g (range:
172. 50-192.38). The day after he interacted with another group, during which he gained a new female group member, his FAM concentration was 329.02 ng/g, and 2 days after, it was 465.21 ng/g. The later value is a 2.6-fold increase over the pre-event mean (Figure 1b). By day 4 postevent, his values were similar to pre-event FAM concentrations.
In the 23 days before splitting from his natal group and leading a subgroup, PAB group's gamma-ranked male had a mean FAM concentration of 87.78 ng/g (range: 74.25-96.59). The day after leading his subgroup, his FAM concentration was 571.19 ng/g, which is a 6.5fold increase over the pre-event sample mean (Figure 1c). Samples collected during post-event days 2-20 were similar to pre-event FAM concentrations.
In the 15 days before a physical altercation with three males from another group, ISA group's beta-ranked male had a mean FAM concentration of 218.74 ng/g (range: 88.1-401.74). The day after this altercation, his FAM concentration was 869.68 ng/g, which is a fourfold increase over the pre-event sample mean (Figure 1d). Samples collected on postevent days 5 and 9 were similar to pre-event FAM concentrations.
The immature male for whom we had comparable data had significantly lower FAM concentrations than the adult males (coef = -67.247, SE = 11.260, p < .001, n = 23 males and 993 samples). The difference was larger when the adult male samples were restricted to those from animals who were at least 15 years old (coef = -75.491, SD = 13.559, p < .001, n = 16 males and 755 samples).
Here we present a biochemical validation of the T-EIA, by using HPLC to determine how well the fecal metabolite(s) the assay captures match authentic testosterone and 5α-DHT, as well as determine whether there is corticosteroid crossreactivity. Sample selection and preparation, as well as the HPLC protocol, are described in detail in Sections 2.5 and 2.7).
There were six identifiable peaks in the tracers, the largest four of which corresponded to authentic cort, cc, T, and 5α-DHT (Table 3 and Figure 2).
A total of 8.16% of all observed immunoreactivity in our sample pool occurred in the corticosteroid-like range, that is, between the fractions where the authentic cort peak began eluting, and where the cc peak finished. 51.22% occurred in the androgen-like range, between the start of the authentic testosterone peak and the end of the 5α-DHT peak. Of this, just over half (25.57% of total) eluted simultaneously with authentic testosterone or 5α-DHT, while the remaining portion eluted at other fractions in the androgen-like range. Much of it (17.52%) was in the 6 fractions immediately following the authentic testosterone peak (Figure 2 and Table 4).
The largest single peak which fell outside either the corticosteroid or androgen-like ranges occurred on fractions 10 and 11, which also corresponded with an unidentified tracer peak (Tables 3
Fecal androgen metabolite (FAM) concentrations of an adult male mountain gorilla before and after an intragroup fight over a mating opportunity (indicated by the red dash line at day 0). One day after the fight, his FAM concentration rose to 2.7 times the mean value of the pre-event samples. By day five, his FAM concentrations were similar to what they were before the fight. (b) FAM concentrations of an adult male before and after an intergroup interaction, during which a female transferred into the male's group. Two days after the interaction, his FAM concentration was 2.6 times higher than the mean value of the pre-event samples. Samples collected on postinteraction days 3 through 6 had FAM concentrations that were similar to preinteraction values. (c) FAM concentrations of an adult male before and after he temporarily led a breakaway subgroup of animals. One day after the event, his FAM concentration was 6.5 times higher than the pre-event mean. On days 2-20 after the event, his FAM concentrations were similar to what they were before the subgrouping. (d) FAM concentrations of an adult male before and after he participated in a fight against extra-group males. One day after the fight, his FAM concentration was 4.0 times higher than the pre-event mean. On days 5 and 9 after the fight, his concentrations were similar to what they were before the fight
T A B L E 3 High-performance liquid chromotography tracer hormone peaks, by elution fraction Peak # Elution fraction Total DPM Hormone 1 1 0 -11 249.06 Unknown 2 6 2 -73 3314.64 Cort 3 8 5 -96 3360.20 cc 4 112-120 2849.50 T 5 130 103.87 Unknown 6 139-147 3395.97 5α-DHT Note: Breakdown of peaks (fractions with DPM > mean of total DPM) in the radioactive tracer hormones. Peaks 2, 3, 4, and 6 correspond to the identified authentic radiolabeled hormone the sample pool was spiked with. Abbreviations: DPM, disintegrations per minute.
and 4). This unknown metabolite accounted for 2.99% of total immunoreactivity, which was more than any individual peak that fell in the corticosteroid-like range.
3.7 | Relationship between dominance rank and fecal androgen metabolites (FAMs)
In 2003-2004, gamma males had higher FAM concentrations than any other rank category of male, followed by alphas, betas, and finally subordinate males (Figure 3, top panel and Table 5). Gamma males' values were not statistically significantly higher than alphas, though they were than betas and subordinates (Table 6, full model columns). Alphas trended toward having marginally higher FAM values than betas (where q = 0.093).
Visual inspection suggested that gamma males' high FAM concentrations were driven by the values of the gamma male in SHI group (Figure 3, top panel). When this male was excluded, all three categories of higher-ranking males had higher FAM concentrations than subordinates (where q = 0.076 for betas vs. subordinates), but alphas, betas, and gammas all had similar values to one another (Table 6, animal NTA excluded columns). There was no statistically significant interaction between rank and event samples, meaning there was no evidence that any particular rank of male had stronger FAM responses to events than any other (Table 5). Having larger numbers of males in the group was also associated with higher FAM concentrations, though this was a statistical trend rather than statistically significant (where q = 0.062). Each additional male predicted 0.354 (+/-0.145) SD greater FAM concentrations.
Versions of the model that exclude age, exclude rank, and limit the analysis to animals 15 and older are presented in the Supporting Information materials (Tables S1-S3). Running the model with the age category variable removed meant that both alphas and gammas had higher FAM concentrations than subordinates, while betas' values trended toward being higher (where q = 0.062), but that other relationships were unchanged (Tables S1 and 6, age category excluded). Males in the youngest age category (10-12 years) had statistically significantly higher FAM values than males in the 12-14 year old category in the model that included rank (Table 5), but when rank was removed, males in the 15+ years old category had higher FAMs than males in either of the two younger age categories (Table S2). Additionally, number of males no longer predicted FAM in the model limited to males 15+ years old (n = 9 males, 301 samples, Table S3).
While it is extremely difficult to disentangle the effects of age and rank when the collinearity is so high, in aggregate, the results suggest that age is probably a better explanation for subordinate males' lower FAM concentrations than rank is. Subordinates had statistically significantly lower FAM concentrations than other males when age was removed but not when it was included (Table 5 and S1), the relationship between rank and FAMs was otherwise nonlinear (i.e., gammas, not alphas, had the highest levels, and differences between other rank categories were small or nonexistent), and when the model was restricted to males who were 15 and over (where age should no longer be a factor, since they are fully adult), subordinates' FAM concentrations were indistinguishable from those of alphas and betas (Table S3).
Neither rank nor age predicted FAM concentrations in 2011-2012 if age was treated as a categorical variable (Table 7).
If age was treated as a continuous variable (thus lowering the rank/age collinearity), older males had higher FAM values than younger ones (coef = 0.028, SE = 0.006, p < .001). In this model, gamma males had significantly higher FAM concentrations than either alpha or beta males, but all other values were similar (Table 6, continuous age variable columns). There was a significant interaction between rank and event samples. Overall, post-event samples had lower FAM values than baseline samples, but this was reversed for gamma males specifically, who had higher FAM concentrations after events (Table 7). Neither malefemale ratios nor total number of males predicted FAM concentrations (Table 7). The overall R 2 of this model was considerably lower than the model for the 2003-2004 data. It explained only 2.2% of the variance (or 3.0%, if continuous
Graph of sample pool immunoreactivity on a testosterone enzyme immunoassay using antiserum R156/7, by highperformance liquid chromotography elution fraction. The majority of the observed immunoreactivity (51.22%) occurred in the androgenlike range between the start of the authentic testosterone and end of the authentic 5α-dihydrotestosterone (5α-DHT) peaks, while only 8.16% occurred in the corticosteroid-like range between authentic cort and cc. The peak at elution fractions 10-11 remains unidentified and accounted for 2.99% of total immunoreactivity. The solid black line is the mean value of all observed immunoreactivity that fell above the detection threshold of the assay; the dashed black line is the mean value of all observed disintegrations per minute of the radioactive tracers. Any values above these thresholds were identified as peaks in the sample or tracers, respectively (Table 3 and 4)
Testosterone enzyme immunoassay reactivity peaks by elution fraction Peak # Elution fraction Immunoreactivity % of total immunoreactivity Corresponds to authentic hormone 1 1 0 -11 89.31 2.99 2 50 20.02 0.67 3 54 21.56 0.72 4 62 20.30 0.68 cort-like 5 70-73 88.23 2.96 cort-like 6 75-76 4638 1.55 7 84 25.25 0.85 8 91-93 63.10 2.12 cc-like 9 100 18.57 0.62 10 105-106 43.02 1.44 11 111-112 51.59 1.73 T-like 12 115-127 866.50 29.05 T-like (through fraction 120) 13 129-133 124.86 4.19 14 135 22.27 0.75 15 137-139 103.66 3.48 16 141-147 394.21 13.22 5α-DHT-like
Note: Breakdown of observed immunoreactivity on a testosterone EIA using antiserum R156/7 by elution fraction, for fractions identified as peaks (values > mean of all observed immunoreactivity that fell above the assay detection threshold). Peak numbers in italics correspond to corticosteroid-like activity; peak numbers in bold correspond to androgen-like activity. Bolded values in "% of total immunoreactivity" column are peaks that accounted for ≥5% of all observed immunoreactivity.
F I G U R E 3 Relationship between rank and standardized fecal androgen metabolite concentrations by social group in 2003-2004 (top panel) and 2011-2012 (bottom panel). Other than standardizing within time period, values are unadjusted. Complementary model results can be found in Tables 567instead of categorical age was used), as opposed to 13.2% for the comparable 2003-2004 data.
Models that excluded age and excluded rank produced very similar results. When age was removed from the model, pairwise comparisons indicated that there were no differences in the FAM concentrations of different ranks of males (Table S4). When rank was removed, older males had higher FAM values than younger males, but only if age was treated as a continuous variable (Table S5). When the sample was restricted to males ages 15 and older, gamma males had statistically significantly higher FAMs than betas or alphas, but betas and alphas had very similar values (Table S6). There were no subordinate males included in this model, as all subordinates during 2011-2012 were <15 years old.
3.8 | Relationship between group structure and fecal androgen metabolites 3.8.1 | Descriptive comparison of a solitary male to males in multimale groups in [2003][2004] The one solitary male for whom we had data in 2003-2004 had FAM concentrations that were slightly lower than his peers living in multimale groups (solitary male: mean standardized FAM = -0.024, SD = 0.992, range = -0.989-2.780, n = 42 samples; males in multimale groups: mean = 0.047, SD = 1.026, range = -1.515-5.343, n = 589 samples). His values were also lower than the three available alpha males specifically, whose mean standardized FAM concentration was 0.351 (SD = 1.176, range = -1.318-5.342, n = 131 samples). Because this was a male that dispersed from a KRCmonitored group, we also had two samples from him while he was living in his natal (multimale) group. These two samples had a mean standardized FAM of 0.447 (SD = 0.161, range = 0.333-0.560).
T A B L E 5 Relationship between rank and fecal androgen metabolite concentrations in [2003][2004] (n = 589 samples, 21 males; R 2 = 0.132)
Coef. SE q a 95% confidence interval Dominance rank b Gamma 0.631 0.056 0.001 0.514 0.747 Beta 0.186 0.254 0.293 -0.343 0.716 Alpha 0.412 0.255 0.142 -0.119 0.943 Event sample 0.146 0.102 0.168 -0.068 0.359 Rank/event interaction Gamma × 1 0.040 0.111 0.401 -0.192 0.271 Beta x 1 -0.229 0.118 0.105 -0.474 0.017 Alpha x 1 -0.036 0.122 0.405 -0.290 0.218 Age category c 12-14 -0.364 0.085 0.007 -0.540 -0.187 15+ -0.064 0.239 0.413 -0.561 0.434 Male:female ratio -0.649 0.409 0.143 -1.502 0.204 Number of males 0.354 0.145 0.062 0.052 0.656 Constant -2.054 0.742 0.045 -3.602 -0.507 Abbreviations: FAM, fecal androgen metabolite. a Sharpened false discovery rate q values, computed from original model p values. b
Reference category was subordinate males (i.e., all ranks below gamma). See Table 6 for pairwise comparisons of all rank categories ((a) Full model). c Reference category was males ages 10-12. Males in the 12-14 and 15+
categories had FAM values that were statistically indistinguishable from one another. *q < 0.1; **q < 0.05; ***q < 0.01: sharpened false discovery rate q values, computed from original model p values.
Males in single-male groups had higher FAM concentrations than solitary males or males in multimale groups (Figure 4 and Table 8).
Solitary males and males in multimale groups were not statistically significantly different, but this seemed to be driven by solitary males' much larger standard errors; qualitatively, their FAM values appeared to be lower than males in multimale groups (Figure 4 and Table 8). When the sample is restricted to leader males, results are very similar. Males in single-male groups had higher concentrations than alpha males in multimale groups, and trended toward having higher FAMs than solitary males (alpha males in multimale groups vs.
males in single-male groups: coef = 0.367, SE = 0.116, q = 0.03; singlemale vs. solitary males: coef = -0.531, SE = 0.202, p = .069, n = 586 samples and 12 males).
Despite the long history of research on mountain gorillas and the species' obvious relevance to questions about androgens and behavior, our knowledge of the role that androgens play in their social dynamics lags far behind that of other well-studied primate species (reviewed in Muller, 2017). Validating an assay for use with fecal material-which is far easier to obtain from terrestrial primates than urine is-is a significant step toward rectifying this imbalance. Our HPLC validation confirmed that a T-EIA using antiserum R156/7
primarily detects fecal metabolites that are structurally similar to
T A B L E 7 Relationship between rank and fecal androgen metabolite concentrations in 2011-2012 (n = 664 samples, 14 males; R 2 = 0.022) Coef. SE q a 95% confidence interval Dominance rank b Gamma 0.315 0.142 0.113 0.007 0.622 Beta 0.024 0.144 0.638 -0.288 0.336 Alpha 0.181 0.222 0.375 -0.298 0.661 Event sample -0.148 0.036 0.006 -0.225 -0.070 Rank/event interaction Gamma × 1 0.240 0.039 0.001 0.154 0.325 Beta × 1 0.203 0.388 0.494 -0.635 1.040 Alpha × 1 0.027 0.215 0.638 -0.437 0.490 Age category c 0.082 0.116 0.420 -0.168 0.333 Male:female ratio -0.032 0.031 0.324 -0.098 0.034 Number of males 0.013 0.010 0.264 -0.008 0.035 Constant -0.293 0.148 0.145 -0.612 0.026 a Sharpened false discovery rate q values, computed from original model p values. b Reference category is subordinate males (i.e., all ranks below gamma). See Table 6 for pairwise comparisons of all rank categories ((b) Full model). c Reference category was males 12-14 years old. F I G U R E 4 Relationship between the type of social group males lived in (multimale, single-male, or solitary) and fecal androgen metabolite concentrations, after adjusting for post-event samples, age, the ratio of males to females in the group, and the total number of adult males in the group. For solitary males, male-female ratio was set to zero. Data are from 2011-2012 only, when there was variation in social group structure. Complementary model results can be found in Table 8 T A B L E 8 Relationship between group type and fecal androgen metabolite concentrations in 2011-2012 (n = 979 samples, 22 males; R 2 = 0.027) Coef. SE q a 95% confidence interval Group type b Single male 0.302 0.078 0.006 0.139 0.466 Solitary -0.200 0.177 0.312 -0.569 0.168 Event sample -0.038 0.163 0.625 -0.378 0.302 Group type/event interaction c Single male × 1 0.086 0.262 0.554 -0.461 0.633 Age category 0.115 0.068 0.178 -0.028 0.258 Male:female ratio -0.012 0.029 0.525 -0.071 0.048 Number of males 0.016 0.009 0.178 -0.004 0.036 Constant -0.225 0.063 0.008 -0.356 -0.094 a Sharpened false discovery rate q values, computed from original model p values. b Reference category is males living in multimale groups.
c Solitary male × event interaction was excluded due to an insufficient number of baseline samples.
testosterone and 5α-DHT, two bioactive androgens that are important for the maintenance of sexually selected anatomical and behavioral traits in male primates. Additionally, individual males demonstrated acute temporary spikes in FAM after a variety of competitive events that theory predicts would generate androgen increases, confirming that this assay is a useful tool to measure androgen dynamics relating to social challenges in mountain gorillas.
Data generated with this assay indicated that the structure of the social group males live in predicted FAM concentrations. Specifically, males in single-male groups had the highest concentrations, while solitary males had the lowest. However, we found no evidence of a clear relationship between dominance rank and FAM, consistent with the larger of the two prior studies on the relationship between urinary androgen metabolites and rank (Rosenbaum et al. 2020).
Four individual animals showed clear, detectable FAM responses to events that theory predicts would temporarily elevate androgens.
Though the results generally conformed to our expectations about males' responses to events, we were somewhat surprised to find that the male who had a female transfer into his group (Event 2) apparently experienced only a transient sex hormone increase, associated with the intergroup interaction/transfer event itself. The arrival of a new female typically means an immediate reproductive opportunity.
Males copulate frequently with newly arrived females for several weeks after their arrival; in this case, the male was first observed mating with the newly transferred female 3 days after the interaction (Karisoke long-term records). Additionally, there is often an uptick in female-female conflict, which males may attempt to control (Watts, 1991). Given these dynamics, a reasonable alternative possibility was that the male's sex hormone levels would remain elevated for some time post-event. Future work should explore whether this transient physiological response was specific to this male, or is typical of males living in one-male groups who face no intragroup, intrasex competition. High sex hormone levels may not be required to maintain high copulation rates, or to police the relatively mild aggression that is typical of conflict between female mountain gorillas (Watts, 1991(Watts, , 1994)).
Research on a variety of species has revealed complex relationships between social dynamics and androgens (Oliveira, 2004), and our event data suggest that this will be an important topic to explore in mountain gorillas. We can draw no firm conclusions without more information on a broader sample of relevant events, but anecdotally, it appears that the within-group conflict over a mating opportunity (Figure 1a) may have been associated with marginally elevated androgens for a longer period of time, relative to the other events. This is perhaps to be expected, since conflict with other group members is likely not limited to a single discrete interaction, especially in the presence of potentially fertile female(s). In bonobos, there is evidence that testosterone is mediated by males' social relationships with females more so than their competitive relationships with other males (Surbeck et al., 2012). Conversely, in chimpanzees, males experience testosterone responses to the presence of sexually receptive females, but data suggest it is likely related to increased male-male aggression rather than mating behavior itself (Muller & Wrangham, 2004). Mountain gorillas' flexible mating system will make them an interesting point of comparison for these other ape species in which hormones, including androgens, have been studied more extensively.
Our biochemical validation using HPLC confirmed that a large proportion of the observed immunoreactivity in our sample pool corresponded very closely to authentic testosterone and 5α-DHT.
However, some of the testosterone-like peak eluted in the six fractions that followed the end of the authentic testosterone peak.
Hormone metabolites are not identical to native hormone, and their molecular structure may be modified slightly for excretion (Möhle et al., 2002;Monfort et al., 1997;Palme, 2005). Notably, we found very little evidence of cross-reactivity with corticosteroid (cortisol and corticosterone) metabolites, which allays concerns over our ability to measure androgens independently whenever both are expected to be elevated. It is important to note that these results may not be generalizable to other species, even closely related ones, or to female or juvenile mountain gorillas. HPLC work on other anthropoid primates has shown that results can be highly species-and sexspecific, and assays must thus be validated individually for each species, sex, and even age class within species, as well as each metabolite matrix (Möhle et al., 2002;Preis et al., 2011;Pribbenow et al., 2017;Touma & Palme, 2005).
We found no clear relationship between males' FAM concentrations and their dominance ranks, a finding that runs counter to expectations based upon other primates with similar mating systems. However, it also confirms our recent findings in this population using urinary androgen metabolites (Rosenbaum et al. 2020). It is interesting that gamma males appeared to have higher FAM concentrations, including stronger responses to potentially androgen-elevating events. However, we are not able to draw firm conclusions about the biological importance of these findings in light of the small number of individuals of this rank. In 2003-2004 the result was driven by one male (out of three gammas total), and in 2011-2012 there were only two males who held the gamma rank. Future research will be necessary to clarify whether our findings of apparently elevated androgens and androgen reactivity in gamma males was an idiosyncratic finding, or a stable feature of mountain gorilla social systems.
Since we have now found no directionally consistent relationship between dominance and FAM in two studies, in two different time periods with substantially different social dynamics, and using two different biological sample types, we feel relatively confident that androgens are not a strong mediator of dominance ranks among male mountain gorillas in this population. This is surprising in light of male ROSENBAUM ET AL.
| 15 of 21 gorillas' marked sexual dimorphism in androgen-dependent behavioral and physical traits, but it is certainly not unprecedented. There are a variety of other primate species in which androgens do not appear to mediate male dominance ranks outside specific periods of rank instability or seasonal mating effort, including chacma baboons (Beehner et al., 2006), redfronted lemurs (Ostner et al., 2002(Ostner et al., , 2008)), Assamese macaques (Ostner et al., 2011), tufted capuchins (Lynch et al., 2002), bearded capuchins (Mendonça-Furtado et al., 2014), and vervet monkeys (Steklis et al., 1985).
One possibility that will need to be explored more carefully in future studies is whether higher-ranking males have a greater ability to facultatively upregulate androgens when needed (Muller & Wrangham, 2004;Wingfield et al., 1990). It is possible that androgens help mediate the outcome of specific competitive events that help establish dominance hierarchies, even if they do not contribute to their subsequent maintenance. Our event data clearly indicate that males respond to competitive events with spikes in androgens.
However, we will need a larger number of males and fecal samples, and more specific event targeting, to determine whether, for example, alpha males experience a more robust androgen response to particular types of competitive interactions than lower-ranking males do. As mentioned above, data from chimpanzees and bonobos indicate that there are multiple potential facets of competition that need to be explored (Muller & Wrangham, 2004;Surbeck et al., 2012).
Our 2011-2012 data were suited for looking at the relationship between group structure and FAMs, since the animals were living in three of four possible group configurations (only all-male bachelor groups were missing from our study), representing a wide variety of group sizes and male-female ratios. It is difficult to disentangle which specific group characteristics matter, since many components are often highly colinear (e.g., overall group size is highly correlated with number of males and number of females). However, after controlling for other factors, the males who lived in single male groups had FAM concentrations that were 33% of a SD higher than their counterparts in multimale groups, and 55% of a SD higher than our small sample of solitary males. This is a finding that will need to be explored in greater detail in future studies. One possible interpretation is that the males who are high-quality enough to maintain a single-male group in a population where multimale groups have a competitive advantage tend to be those males with higher androgen concentrations.
Data from the solitary males must be treated with caution. It is probable that the times when we are most likely to encounter solitary males are not a representative snapshot of their lives. When we observe them, it is often because they are close to a social group that KRC is monitoring, meaning they are disproportionately likely to be involved in competition in either the recent past or near future.
However, this theoretical over-representation of event-related samples in their data would mean that we should expect their FAMs to be higher than they would be at other times. Instead, they were lower than the concentrations observed in group-living males, which is consistent with what we previously reported based upon urinary androgen metabolite data collected in these animals (Rosenbaum et al., 2020). The male for whom we had data both while he lived in his natal group, and again once he was a solitary male, anecdotally had much higher FAM concentrations while living in the group. However, it is difficult to know whether this is due to the extreme disparity in sample sizes. We had only n = 2 samples from when he was a group-living male, and n = 42 samples from when he was solitary. Noninvasive hormone measurement includes enough noise that we hesitate to draw strong conclusions based on such a small sample from one of the two states. Additionally, this animal's group-living samples were collected shortly before he dispersed-a major life history event for a male gorilla-and thus may also not be representative of his typical FAM values over a longer period.
The finding that males in single-male groups had higher FAM concentrations than males in multimale groups and solitary males stands in contrast to what we found in our urinary androgen metabolite data (Rosenbaum et al., 2020), where we detected no statistically significant effect of group type. However, the current study availed of a much larger sample size for the group structure analysis specifically (979 fecal samples compared to 253 urine samples), and contained information on one additional social group along with three additional solitary males. This expanded sample allowed us more freedom to include other aspects of group structure, which may help explain the discrepancy. If this is the case, then it further underscores the challenges inherent in disentangling the specific features of groups that matter most to endocrine regulation. Social group features like size, structure, and male:female ratios are rarely independent of one another. Future analyses will need to verify these findings using a larger pool of males in single-male groups, and preferably include longitudinal data that capture the transition as males move from multi-to single-male groups.
While mountain gorillas may be unusually socially flexible, they are certainly not the only primate that occurs in multiple group configurations. Some langurs and colobus monkey species routinely organize into a variety of group structures (Newton, 1988;Stead & Teichroeb, 2019;Teichroeb et al., 2012), and there are interesting similarities to primates that live in multilevel societies (e.g., geladas or golden snub-nosed monkeys) where single-male units regularly interact with one another, and with all-male social units (Qi et al., 2009;Snyder-Mackler et al., 2012). More information about the relationship between social group structure and androgen concentrations in a wider variety of primates could help determine the particular aspect(s) of intrasexual competition that are most salient to males, and to refine our understanding of the diverse expressions of costs related to male-male competition. Having exclusive control over a group of females has obvious fitness benefits, but if sustained elevated androgen production is one of the costs males must pay to do so (e.g., in geladas, harem-holding males have higher testosterone than bachelor males across the year: Pappano & Beehner, 2014), this may help explain why multimale groups are so common. A recent study from the Amboseli Baboon Research Project found that male baboons appear epigenetically "old" for their age when they are high ranking (Anderson et al., 2021). It seems plausible that androgens may be one mechanism via which these biological changes occur. If so, then we might predict such effects would be even more dramatic among males who manage to maintain single-male groups.
The low R 2 values of our models came as a surprise to us, showing that the variables that we considered explained between 2% and 13% of the variation in FAM. This indicates that there are important predictors of FAM concentrations in male mountain gorillas that were not captured in our models, particularly in 2011-2012. This time period was characterized by a greater diversity of group types, more frequent intergroup interactions, and less stability in group composition and in male dominance hierarchies than 2003-2004(Caillaud et al., 2014, 2020;;Rosenbaum et al., 2020). Clearly, interand intragroup social dynamics will be an important avenue to explore in future work. Another class of steroid hormone, glucocorticoids, are predicted by home range sizes: mountain gorillas who live in groups with smaller exclusive-use areas (i.e., areas not used by other groups) have higher glucocorticoid levels than animals who can more easily avoid their neighbors (Eckardt et al., 2019), and it is not implausible that this may similarly impact FAM concentrations. Additionally, there is considerable variation in the social dynamics between males living in social groups with other males. In some, conflict is rare and males are highly tolerant of one another, while in others their relationships are characterized by avoidance and/or conflict (Harcourt & Stewart, 2007;Robbins, 1996;Stoinski et al., 2009). Our event data clearly indicate that males may experience variable FAM responses to events that strictly affect their own social groups, or those that involve other social units.
Future studies can examine the relationship between male mountain gorillas' sex hormones and multiple outcomes, including health measures, reproductive success, and social relationships. Since there are extensive longitudinal behavioral, demographic, and health data available for these animals, the ability to measure FAMs adds a powerful new tool for probing questions about mountain gorilla socioecology. Since mountain gorillas are among the most sexually dimorphic of the extant primates, complementary androgen and behavioral data from this species will provide important new data points for comparative analyses that address questions about the relationship between sex hormones and male behavior across the primate lineage.