Study population
We studied wild meerkats at the Kuruman River Reserve (a ~63 km2 area comprising dry riverbeds, herbaceous flats and grassy dunes) in the Kalahari region of South Africa (26°58′S, 21°49′E)28,48. Our study period (Nov 2011–Apr 2015) included an extended drought, during which female reproductive success tracked rainfall22 (Supplementary Fig. 1). The annual mean population size was 270 animals, in 22 established clans of 4–39 animals15,22. Habituated to close observation (<2 m), handling and routine weighing28, the animals are microchipped and individually identifiable via unique dye marks, and matriarchs are radio-collared to allow tracking clans. Matriarchs have a distinct morphotype and are distinguishable by weight advantage, altered body proportions, increased reproductive output and older age14,60,61. Status, health, life history variables and clan composition are known and routinely monitored, as are pregnancies, detectable at 3–4 weeks of gestation. We used palpation, weight gain and shape change to determine gestational stage in real time, and later confirmed estimated conceptions by backdating 70 days (the duration of meerkat gestation) from the known dates of birth22,28.
Focal subjects and datasets
Five of our 57 unique focal dams changed status during the study, producing 31 dominant dams (a subset of which we treated with antiandrogens, see below) and 31 subordinate dams that carried 89 pregnancies to term, all with live births within the clan (i.e., excluding abortions and evictions22). Given field complexities, dams contributed differentially to endocrine and behavioural datasets across pregnancy stages. Endocrine data derived from 35 individuals contributing a total of 49 serum samples (~1 blood sample/pregnancy) and 36 individuals contributing a total of 156 faecal samples, with 18 individuals contributing both types of biological samples (Table 1). Adult behavioural data derived from 37 unique individuals during 48 pregnancies, observed for 1598 focal sessions, totalling 422 h (Table 2). The focal offspring were 103 unique individuals from 32 litters, representing 14 dominant control (DC), 13 subordinate control (SC) and 5 dominant treated (DT) litters (Table 3). Offspring behavioural data from 5661 focal sessions totalled 1362 h of observation.
Biological sample collection and processing
For blood sampling, we individually captured a target DC, SC or DT female upon emergence from the underground den, gently picked her up by the base of her tail, placed her into a cotton sack and anaesthetised her with 4% isoflurane (Isofor; Safe Line Pharmaceuticals, Johannesburg, South Africa) in oxygen, administered via a mask attached to a vehicle-mounted vaporiser. Once fully sedated, we reduced isoflurane dosage to 1–2% and used a 25G needle and syringe to draw 0.2–2 ml of blood from the jugular vein. We immediately transferred the blood to a serum separator tube (Vacutainer®, Becton Dickinson, Franklin Lakes, NJ, USA), allowing it to clot at ambient temperature. We then performed our additional procedures, such as health monitoring, as well as our antiandrogen treatment of DT dams, as described below (all capture, anaesthetic, monitoring and sampling methods across subjects were identical). Once recovered from anaesthesia (~20–30 min post-capture), the female was returned to her clan and closely monitored, including by veterinarians, throughout that and the following day (during this time, behavioural data collection on treated females was suspended, see below). We later centrifuged the blood (500 × g for 10 min at 24 °C) and stored the decanted serum, on site, at minimally −20 °C.
We also opportunistically collected, into clean plastic bags, roughly half the amount of any freshly voided faeces (to avoid disrupting faecal scent marking), immediately placed samples on ice and later stored them at minimally −20 °C. We transported all serum and faecal samples, on ice, to Duke University (Durham, North Carolina) and stored them at −80 °C. Within a year of collection, we lyophilised, pulverised and sifted faecal samples into fine powder and extracted steroid metabolites15,22. We weighed 0.2 g of dry powder, mixed it with 2 ml of 90% methanol, placed the mixture on a rotating shaker (30 min), centrifuged it twice, discarding the sediment each time, and stored the methanol-extracts at −80 °C until analysis.
Endocrine assays
To measure serum androgen concentrations, we used competitive enzyme immunoassay (EIA) kits (ALPCO diagnostics, Salem, NH, USA), previously validated in meerkats (Supplementary Methods)15,43. The serum androstenedione assay has a sensitivity of 0.04 ng/ml using a 25 µl dose, with intra-assay and inter-assay coefficients of variation (CVs) of 5.23% and 8.7%, respectively. The serum testosterone assay has a sensitivity of 0.02 ng/ml using a 50 µl dose, with intra-assay and inter-assay CVs of 7.9% and 7.3%, respectively. We also assayed fAm via EIA (Supplementary Methods). The fAm assay has a sensitivity of 0.2–12.5 ng/ml per plate, with intra-assay and inter-assay CVs of 7.7% and 6.2%, respectively.
To measure fGCm concentrations, we used the ImmuChem double-antibody 125I radioimmunoassay kit for corticosterone (MP Biomedicals, Irvine, CA), also validated in various vertebrates, including meerkats (Supplementary Methods)22. The inter-assay CVs were 10.93% and 9.05% for low and high faecal pools, respectively. Intra-assay CV was 5.55%. All individual sample CVs were below 15%.
Behavioural observations
We tracked and observed clans minimally every three days; all observation protocols limit human presence to several days/week. For candidate dams, deemed healthy and living in stable clans, we began behavioural observations after confirming pregnancy, continuing throughout gestation and PP. For offspring, we began behavioural observations when pups were ~1 month old (when they leave the natal den and reliably join the foraging clan). We used focal-observation protocols62, as these are less vulnerable to detection bias than are critical-incident protocols and allowed us to capture detailed or cryptic behaviour, including across seasons when vegetation can differentially obscure the animals.
We observed focal subjects, in randomised order, during two daily periods, one beginning in early morning, after emergence, until midday (AM), the other resuming late afternoon until the animals retired underground (PM). We recorded behaviour in real time, using CyberTracker software (version 3.263, CyberTracker Conservation) uploaded to handheld palm pilots (Palm T|X, Palm, Inc.), and paused recording whenever the subject was out of view. We tailored our protocols to the meerkats’ major activity patterns at different times and locales: 5 min den focals occurred at either end of the day, during the meerkats’ brief periods spent above ground at the den, whilst mostly sedentary and prosocial; 30 min ranging focals occurred in the interim, whilst meerkats primarily foraged throughout their territory. The shorter focals at the den allowed us to rotate through our subjects minimally once prior to the clan’s departure from or descent into the den.
For adults, we present data on 1) aggression (including competition over an acquired food item and other high-intensity aggression or ‘HIA’; the latter comprises the former), 2) scent marking, 3) prosociality, 4) submission and 5) nearest-neighbour (N-N) associations (as a proxy of an ego network; Supplementary Methods). For categories 1, 3 and 4, we recorded the partner and directionality (i.e., initiator versus recipient). For category 5, every 2–3 min or whenever partners changed, we recorded the identities of meerkats within 10 cm or 2 m of the focal dam (for den and ranging focals, respectively). We also report on the frequency and duration of non-target or neutral behaviour performed by all meerkats whilst ranging, including 6) vigilance, a form of cooperative behaviour44, and 7) the duration of digging for food. For offspring, we present on category 1 (with the addition of targeted begging). We omitted from analysis any focal in which a disrupting interclan interaction occurred. We assessed intra- and inter-observer reliability, with scores exceeding 87% (see Supplementary Methods for additional details on reliability). For the much longer span of offspring observations, we also confirmed that including observer identity as a random variable in the analyses had no significant impact on the findings.
Selecting controls
We planned our experimental manipulation, using steadily dissolving antiandrogen pellets that require no removal, in dominant dams only, as their reproductive success is most reliable. Local permitting required that we avoid implanting sham pellets into matriarchs—the key individual per clan—to minimise the number of clans potentially affected by our protocols (i.e., to facilitate uninterrupted, long-term collection of normative life-history data by other researchers). Therefore, we ran multiple alternative controls, the first of which involved a substitute manipulation (Feb 2011–Jan 2012) in three groups of adult, subordinate, male meerkats: a flutamide-treated (~15 mg/kg/day) group that received 1 pellet (300 mg, Innovative Research of America, Sarasota, FL), a sham-treated group that received 1 sham pellet of the same matrix, and untreated controls that received no pellet43 (see below). Males are ideal for delineating androgen-mediated traits as a basis for assessing flutamide effects in females19,39,41,42,46.
We then supplemented this sham control in males with several other controls in females (see below), that allowed us to test multiple predictions. Notably, we compared the following: androgen-mediated vs. nonandrogen-mediated behaviour in DT dams, health measures (via abortions, stress hormones and weight) in DC vs. DT dams; indirect behavioural effects in SC dams cohabiting with DC vs. DT dams; match-paired effects on social centrality (via nearest-neighbour analyses) in DC vs. DT dams and; the behaviour of offspring from DC vs. DT and DT vs. SC dams.
Antiandrogen treatment
We brought in full-time veterinarians, licensed in South Africa, to perform the minimally invasive implants and monitor the animals before, during and after treatment. Previously, in males43, we found (a) no negative health effects in any of the three groups, (b) no behavioural differences between sham and control groups, (c) the same significant differences between the flutamide-treated group and each of the two control groups, (d) the specifically predicted changes in behaviour owing to androgen receptor blockade and (e) consistency of androgen-specific inhibition, as obtained in other flutamide studies, regardless of administration mode (i.e., oral, injection, topical, subcutaneous implant)34,39,40,41,42. Neither the procedures associated with minor incision nor the pellet matrix significantly impacted male meerkats43.
Here, following this male validation, we treated 11 pregnancies from 10 dominant dams with flutamide, at the same dose used in males (~15 mg/kg/day), targeting the estimated last 21 days (LP) of their 70-day gestation (Supplementary Fig. 3). During routine capture, the veterinarian provided each experimental animal with a subcutaneous injection of a non-steroidal, anti-inflammatory painkiller (either 0.2–0.3 mg/kg meloxicam, Metacam, Boehringer or 0.1 ml of metacam, plus 0.1 ml of lentrax or other long-lasting penicillin, depending on availability). Under sterile procedures, the veterinarian used a scalpel to make a 1–2 cm dorsal skin incision between the shoulder blades, blunt dissection to create a small subcutaneous pocket, forceps to insert two 21-day release flutamide pellets (because dominant females are twice the weight of adult, subordinate males43), dissolvable material (Vicryl) to suture the skin incision and surgical glue to cover the suture for added protection.
Recovery and release of all dams was as described above (see biological sample collection); however, treated females were monitored more extensively and for longer periods than is routine. Again, no adverse effects of treatment were detected; dominant females maintained their status and continued to fill their matriarchal role (as expected for this short treatment window). Whereas 21% of the contemporaneous dominant controls aborted22, only one treated female (9%) aborted and that pregnancy was thus excluded from the study. Given costs of natural androgens to dominant female meerkats57,58, flutamide-treated dams had potentially less health risk than their control counterparts. A few subjects received antibiotic ointment to treat minor infections at the suture site, but these typically occurred after parturition (i.e., after the cessation of focal observations) and had no noticeable effects on behaviour. Owing to difficulties assessing pregnancy stage in real time, particularly under drought conditions22, treatment periods ultimately extended into PP, so we matched the LP + PP time span in controls (Supplementary Fig. 3). Overshooting treatment was preferable to undershooting, given that the critical period of offspring behavioural masculinisation could extend perinatally, through a combination of androgen transfer via the placenta and maternal milk.
Analysis of endocrine data
To examine the relationship between androgens and status in control females across pregnancy, we ran three models in R (version 3.4.363). We analysed serum concentrations using GLMMs in the MASS package (version 7.3-4764). We used a Gamma error distribution and log link function, accounting for females sampled across multiple pregnancies by including individual identity as a random factor. We analysed faecal concentrations using LMMs in the lme4 package (version 1.1-2165). After log transformation, the response variables conformed to the normal distribution, so we used a Gaussian error distribution with an identity link function. Because we obtained multiple faecal samples per pregnancy, we included litter identity nested within individual as a random factor.
To test if androgen concentrations changed naturally with pregnancy stage, we included the interaction between status (dominant, subordinate) and stage (EP, MP, LP, PP) as fixed factors in the full models, as well as female age (continuous in years; meerkat age does not covary with status for endocrine data) and total monthly rainfall (continuous in mm2; rainfall highly correlates with year and territory, but was the most sensitive measure of temporal quality). Owing to drought conditions, we excluded clan identity from most analyses, because membership and location changed too often between pregnancies to be a reliable measure. Because we obtained all serum samples in the morning, we included collection period (AM or PM) only as a covariate in the fAm model (Supplementary Table 1).
Each full model included all probable independent terms and biologically relevant interactions; we then obtained minimal models by sequentially removing the least significant factors (P < 0.05), starting with two-way interactions. We confirmed validity of the final models (full and minimal) using a forward stepwise procedure66 and verified our assumptions by checking residuals for normality and homogeneity of variance. We assessed collinearity between main effects via variance inflation factors (VIFs), all of which were <2, in the R package “car” (version 2.1-667). We determined significance of fixed factors through maximum likelihood estimation and likelihood ratio tests following a χ2 distribution. Significant interactions and pairs from three-level factors (e.g., EP, MP, LP or MP, LP, PP) were compared to each other using post hoc pairwise comparisons (LSD) in the lsmeans package (version 2.30-068). Statistical tests were two-tailed throughout and, unless otherwise stated, we present means and standard errors.
Lastly, we analysed the effect of antiandrogen treatment on the concentrations of androgens and stress hormones in dominant dams. Because treated females did not contribute serum samples and because faecal sampling is noninvasive and better captures natural changes in the stress response, we addressed this question using faecal samples. We ran new models (comparable to the prior fAm model, above) to test if flutamide influenced the fAm or fGCm concentrations of treated versus control dominant dams. The full models included experimental condition (control, treatment), age, total monthly rainfall and collection period. Additionally, we included pregnancy stage (LP, PP) as a random factor to control for the timing of treatment.
Analysis of adult life-history data
We tested if dam age, weight or clan size related to treatment condition, using LMs in the stats package (version 3.6.1) in R (version 3.6.163), for which we included the dominant female’s treatment condition as the fixed factor. A first analysis of values at conception served to verify that we had randomly assigned treatments. Indeed, there were no differences by age (t1,26 = 0.45, P = 0.655), weight (t1,27 = 1.34, P = 0.191) or clan size (t1,29 = 0.52, P = 0.609) between dominant control and treated females. A second analysis of postpartum weight served to confirm the lack of negative health effects of flutamide treatment, as would be revealed by weight loss (see “Results” section). To control for pre-treatment condition, this analysis also included weight at conception as a covariate.
Analysis of observed adult behaviour
To test our behavioural predictions for dams by 1) status, 2) treatment condition (for dominant females), or 3) matriarch’s treatment condition (for subordinate females), we analysed rates of aggression, scent marking, prosociality and submission using zero-inflated GLMMs in the glmmADMB package (version 0.8.3.369) in R (version 3.4.363). We focused on the MP-PP increase in behaviour, rather than the potential PP decrease, because behavioural changes often lag behind endocrine changes. We used either a Poisson or negative binomial error distribution, and used the distribution with the lowest Akaike’s Information Criterion (AIC) value for subsequent model selection, with focal duration as an offset. To examine the effects of antiandrogen treatment on non-targeted behaviour, including rates of digging and vigilance, we used models for which focal duration was included as an offset. For duration measures, we used an LMM with a gaussian (and a log link) error distribution for digging, and a GLMM with zero-inflated Gamma (and a log link) distribution for vigilance, using the glmmTMB package (version 1.0.2.170) in R (version 3.6.163). To account for repeated sampling (of individuals and pregnancies), we included litter nested within individual as a random factor, unless convergence failed, in which case we included only litter as a random factor, as this was the more sensitive measure.
First, we tested if normative behaviour (save submission, the direction of which depends on status) differed by status as the unmanipulated pregnancies progressed. We included the interaction between status and pregnancy stage (MP, LP, PP) in each full model, as well as clan size (continuous in number of members present during observation), total monthly rainfall and collection period. For behaviour that occurred irrespective of location, such as aggression (initiated or received) and scent marking, we included the type of focal (at den, whilst ranging) as a covariate. We did not do so for food competition (which occurred primarily whilst ranging) or prosociality and submission (which occurred primarily at the den). Owing to collinearity between age and status when comparing behaviour across classes, as assessed via VIFs (all other variables >2), we did not include age in these models (Supplementary Table 2).
Second, we tested if flutamide directly affected the behaviour of treated dominant dams, relative to control dominant dams (Supplementary Table 3), and third, if flutamide indirectly altered the behaviour of cohabiting subordinate dams, for which we compared the contemporaneous behaviour of subordinate dams whose matriarch was treated to that of subordinate dams whose matriarch was left untreated (Supplementary Table 6). In both cases (involving within-class comparisons), we included the dominant female’s treatment condition as a fixed factor in each full model, as well as the respective dominant or subordinate female’s age, clan size, total monthly rainfall and collection period. The type of focal was included as a covariate whenever relevant. Additionally, pregnancy stage for focal dominant (LP, PP) or subordinate (MP, LP, PP) dams was included as a random factor to control for potential gestational differences. Because the behaviour of subordinate dams remained consistent across pregnancy (Fig. 1d–f), we increased their sample size by extending this analysis to include subordinates at MP.
As in our endocrine analyses, we initially included all probable independent terms and biologically relevant interactions in the full models. A minimal model was obtained by sequentially removing terms based on AIC, with validity of the final models being confirmed using a forward stepwise procedure66. Significance of fixed factors was determined through maximum likelihood estimation and likelihood ratio tests following a χ2 distribution, and significant interactions and three-level factors were compared using post hoc pairwise comparisons (LSD) in the lsmeans package (version 2.30-068).
Analysis of inferred evictions
We tried several ways to analyse the number of evictions by dominant control versus treated females during LP (for n = 31 pregnancies) using changes in clan composition. First, we excluded the pregnancies of dominant females when there were no other adult females present. Then, we accounted for differences in clan size, the number of adult females present (i.e., the number of potential eviction victims), the number of times the same female was repeatedly evicted, and the timing for the onset of treatment (i.e., if victims had already been evicted when dominant females started treatment late). We selected the mean percent of adult females evicted (out of the total number of adult females residing in the clan); nevertheless, accounting for these important variables for so few occurrences of evictions by treated matriarchs reduced the power necessary to properly model this question.
Analysis of nearest-neighbour (N-N) associations
To determine if antiandrogen treatment altered the matriarch’s social relationships (or centrality), we calculated dyadic rates of N-N associations (using scan sampling) for dominant dams and their adult clan members, during both treatment and control periods. We restricted this analysis to the four dominant females observed for more than an hour during both a treated and control pregnancy (totalling eight pregnancies); however, the sample size, based on Satterthwaite’s approximation method, reflects the number of adults associating with these dominant dams. The resultant dataset contained 110 dyads, including 43 adult female and 67 adult male partners (27 and 40 of which, respectively, were unique), representing 77 total hours of focal observation. Dominant females averaged 1.83 N-N scans per adult clan member per hour across periods (S.D. = 1.20, range = 0–6.35). We ran an LMM in R (version 3.6.163), using the lme4 package (version 1.1-2165), with dyadic rates of N-N association as the outcome variable and partner sex, condition (treatment versus control), adult clan size, and indicator variables for dominant males and subordinate dams as main effects, and dominant female identity and partner identity as random effects. We calculated VIFs using the R package “car” (version 3.0-567); all VIFs were <1.2, indicating adequate independence of predictor terms (Supplementary Table 5).
Analysis of observed offspring behaviour
We used a combined model to examine the offspring’s rates of initiated aggression by maternal condition, including status (dominant versus subordinate control) and treatment (dominant control versus treated; Supplementary Table 6). We used GLMMs with a negative binomial error distribution in the glmmTMB package (version 1.0.170) in R (version 3.6.363) to account for overdispersion and zero-inflation, with focal duration as an offset. To account for repeated sampling, we included individual nested within litter nested within dam as a random factor, unless convergence failed, in which case we included only individual nested within litter, as this was a more sensitive measure. As in our analyses of adults, we included all probable independent terms (i.e., maternal condition, offspring sex, offspring age in months, clan size, rainfall, focal location and collection period) and biologically relevant interactions (i.e., between maternal condition and offspring age, and between clan size and rainfall) in the full model. We obtained a minimal model by sequentially removing terms based on AIC and confirmed final-model validity using a forward stepwise procedure65. Fixed factor significance was determined through maximum likelihood estimation and likelihood ratio tests following a χ2 distribution, and significant interactions and three-level factors were compared using post hoc pairwise comparisons (LSD) in the lsmeans package (version 2.30-067).
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com
