Testing hormonal responses to real and simulated social challenges in a competitive female bird

Abstract

Competitive interactions often occur in series; therefore animals may respond to social challenges in ways that prepare them for success in future conflict. Changes in the production of the steroid hormone testosterone (T) are thought to mediate phenotypic responses to competition, but research over the past few decades has yielded mixed results, leading to several potential explanations as to why T does not always elevate following a social challenge. Here, we measured T levels in tree swallows (Tachycineta bicolor), a system in which females compete for limited nesting cavities and female aggression is at least partially mediated by T. We experimentally induced social challenges in two ways: (1) using decoys to simulate territorial intrusions and (2) removing subsets of nesting cavities to increase competition among displaced and territory-holding females. Critically, these experiments occurred pre-laying, when females are physiologically capable of rapidly increasing circulating T levels. However, despite marked aggression in both experiments, T did not elevate following real or simulated social challenges, and in some cases, socially challenged females had lower T levels than controls. Likewise, the degree of aggression was negatively correlated with T levels following a simulated territorial intrusion. Though not in line with the idea that social challenges prompt T elevation in preparation for future challenges, these patterns nevertheless connect T to territorial aggression in females. Coupled with past work showing that T promotes aggression, these results suggest that T may act rapidly to allow animals to adaptively respond to the urgent demands of a competitive event.

When female animals fight, does their testosterone elevate? We used two approaches to explore this question in tree swallows, a cavity-nesting territorial songbird. First, we simulated intrusions with taxidermic models, and second, we reduced territory availability to increase actual competition. T levels were not higher in experimental birds compared to controls, and in fact were sometimes lower and negatively correlated with the degree of aggression. Testosterone may therefore be depleted to promote adaptive aggression.

INTRODUCTION

Competition can lead to contests among individuals over access to limited resources. Often, competitive interactions are not isolated events; rather, they can be predictive of similar challenges in the near future. Animals may therefore respond to social challenges in ways that enable them to prepare for likely future conflict (i.e., social priming; Rosvall and Peterson 2014). In fact, there is evidence from a variety of taxa that changes in the competitive environment alter expression of competitive phenotypes—including overall aggressiveness (Apfelbeck et al. 2011; Bentz et al. 2013), display intensity (Wagner Jr. 1989; Langmore and Davies 1997), blood glucose levels (Meehan et al. 1987), and gamete production (Awata et al. 2006; Ramm and Stockley 2009).

Changes in the production of the steroid hormone testosterone (T) are often thought to mediate phenotypic responses to social challenges, especially in male vertebrates. There is strong correlational and experimental evidence that elevated T promotes competitive traits, including aggression (Goymann et al. 2019; Moore et al. 2019; Rosvall et al. 2020). As part of the Challenge Hypothesis, Wingfield and colleagues posited that males in unstable social environments temporarily increase circulating T levels after social challenges in order to prepare for future challenges (Wingfield et al. 1990). Experimental support for this hypothesis first came from male song sparrows (Melospiza melodia), which were captured and bled shortly after facing a taxidermic mount and song playback on their breeding territory (i.e., a simulated territorial intrusion, or STI). Males that experienced a simulated social challenge had higher circulating T levels than passively caught controls (Wingfield 1984, 1985; Wingfield and Wada 1989), suggesting that males rapidly increased T levels after engaging with the perceived competitor. Separate experiments showed that treatment with exogenous T increased aggression (Wingfield 1984). Together, these results supported the hypothesis that competition increases T in order to upregulate competitive phenotypes, and thus prepare for future challenges.

Over the past few decades, these ideas have been applied to a variety of species. While there has been evidence of socially mediated T elevations in both male and female vertebrates (Moore et al. 2019; Rosvall et al. 2020), there have been many examples of STIs not increasing T levels, most notably in other songbird species (Goymann et al. 2019). In some of those cases, individuals appeared to decrease T in circulation after experiencing an STI (e.g., Pinxten et al. 2004; Landys et al. 2007). We still lack a clear understanding as to why such a variety of hormonal responses have been observed following experimental social challenges, though several potential explanations have been proposed.

Often, methodological constraints are evoked as reasons why socially-induced T elevations were not observed. For one, the stimulus used to imitate a competitor may not elicit the same hormonal response as true competition. For example, male song sparrows respond aggressively toward (1) caged live decoys coupled with playback, (2) devocalized live decoys without audio, and (3) song playback alone, but T levels were significantly higher than controls only for the first type of STI (Wingfield and Wada 1989). Likewise, male zebrafish (Danio rerio) elevate androgen levels (11-ketotestosterone) following agonistic interactions with actual competitors, but not following aggression directed at their own mirror image, arguably because the mirror assay does not have a clear resolution with a winner and loser of the challenge (Teles and Oliveira 2016). Frequency and/or duration of a challenge, as well as timing of sampling relative to the challenge, can also affect whether hormonal responses are observed. Again in zebrafish, the timing of peak androgen responses to STI varied among males, ranging from 2 to 15 min with some individuals exhibiting a second peak 90 min later (Félix et al. 2020). While male song sparrows exhibited elevated T levels within 10 min of an STI (Wingfield and Wada 1989), male spotted antbirds (Hylophylax naevioides) only showed elevated T levels after 2 h of continuous audio playback (Wikelski et al. 1999). To tease apart potential methodological confounds, one would need a system in which both real and simulated competition could be induced.

Even after animals experience a social challenge (real or simulated), they still may not rapidly increase T if selection has already acted against that potential response mechanism. Elevated T may confer costs, such as reduced immune function, interference with parental care, and increased risk of injury (reviewed in Wingfield et al. 2001). These costs could explain the lack of observed T responses in some systems (Goymann 2009), particularly if high T is costly relative to its benefits, such that selection may have already limited an animal’s ability to elevate T. Researchers often use a “GnRH challenge” as a standardized method for measuring an individual’s current T production potential, or reactive scope (Jawor et al. 2006; Apfelbeck and Goymann 2011; George and Rosvall 2018). That an animal can elevate T in response to exogenous GnRH is by no means a guarantee that it will elevate T after a social challenge (e.g., Apfelbeck and Goymann 2011; Rosvall et al. 2014). However, if an animal cannot elevate T following a GnRH challenge, then it is physiologically incapable of doing so after a social challenge as well. Therefore, an ideal system in which one would expect to find socially modulated T levels would be one in which individuals are physiologically capable of rapidly increasing circulating T, elevated T is known to cause increased aggression, and the benefits of T-induced phenotypic changes for competition should outweigh the costs.

Female tree swallows are excellent candidates in which to test for socially-induced T elevations. As obligate secondary cavity-nesters, females face intense competition for limited nesting sites that are necessary for reproduction (Winkler et al. 2020). When nesting sites are limited, female aggressiveness predicts success in intraspecific contests, with more aggressive females more likely to obtain a nesting site (Rosvall 2008). T appears to facilitate aggression, as exogenous T (delivered via silastic implants) increases female aggressiveness (Rosvall 2013). During the territory establishment and nest building stages, females also exhibit the ability to roughly double circulating T levels within 30 min of GnRH injection, suggesting that selection has not acted to completely limit females’ T production capabilities at the time when initial competition and defense of nesting territories are most intense (George and Rosvall 2018). That tree swallows readily adopt artificial cavities, or nestboxes, as nest sites also makes this system ideal for addressing questions about competition. Researchers can increase or decrease the amount of competition within a population by changing the location and number of available nestboxes (e.g., Stutchbury and Robertson 1985; Rosvall 2008; Bentz et al. 2013). Once a female successfully claims a nestbox, she reliably defends the box and its immediate surrounding area (radius of at least 10–15 m, Winkler et al. 2020), and will respond quite aggressively to a simulated “competitor” (i.e., decoy; as in Bentz et al. 2019b). Finally, tree swallows can easily be captured upon entering their nestboxes (Stutchbury and Robertson 1986), thus facilitating the collection of blood samples for circulating hormone measurements.

Here we used separate experiments to manipulate the competitive environment in two different ways: (1) we presented taxidermic decoys at nestboxes to simulate a challenge from a competitor, and (2) we temporarily removed some nestboxes shortly after initial territory establishment to reduce nest site availability and induce competition among free-living birds. In both experiments, we captured females shortly after they were involved in these aggressive interactions and measured circulating T, which we compared to date- and breeding stage-matched controls. If the challenge hypothesis applies to females, then T should be higher among experimental birds compared to controls in both experiments. Deviations from this overarching prediction will shed light on when and why aggression alters the secretion of T.

METHODS

Study system, populations

All subjects were female tree swallows captured in artificial nesting cavities (nestboxes) located around Bloomington, IN, USA (39°9 N, 86°31 W) in springs 2017–2019. Experiment 1 occurred from April 22 to May 11, 2018 and variations of Experiment 2 took place in 2 years: April 21 – May 7, 2017 (hereafter Y1) and April 16 – April 18, 2019 (Y2). Years varied in their exact breeding phenology, but all experiments occurred after breeding pairs had established territories at boxes but prior to yolk production and egg laying, which we confirmed by regularly checking nestbox contents. We selected this timing because we have previously shown that females are physiologically capable of elevating T in response to exogenous GnRH in this breeding stage (George and Rosvall 2018).

Experiment 1: Simulating social challenges

Using decoys to simulate territorial intrusions

In order to simulate a conspecific intruder, we placed a tree swallow decoy on a nestbox for 30 min. Decoys were made from female skins/feathers, affixed to a 3D-printed swallow body, and positioned in a defensive posture at the entrance to a nestbox (described in Bentz et al. 2019b); skins were sourced from birds collected for our genomics work (Bentz et al. 2019c). We only used decoys made from 2+ year old females with iridescent blue/green plumage because some studies have found that younger (brown) females evoke less aggression (Coady and Dawson 2013; but see Stutchbury and Robertson 1987b). We rotated among seven separate decoys between trials to minimize pseudoreplication. During each STI, we also played an audio file containing aggressive tree swallow calls. Audio files were made by repeating a single vocalization clip several times within a file, so that one playback file consisted of a (10 s call, 10 s silence) sequence repeated three times, then 30 s silence, then back to (10 s call, 10 s silence) × 3, and so on. We rotated among four audio files made from four separate clips, which were each recorded from unique individuals in our population in years prior to the experiment (2016–2017). Audio was played using a Sandisk Sportclip mp3 player and a portable speaker (Altec Lansing iM227, iMW257, or iMW270), which were attached behind or directly below (within 1ft of) a nestbox. Playbacks were normalized to naturalistic volume. Post hoc, we tested for effects of specific decoy or playback IDs on behavioral and T responses (Supplementary Table 1); upon finding none, we excluded playback and decoy IDs from further analyses. All STIs were conducted between 8:17 AM and 1:10 PM.

Measuring aggressive responses and capturing experimental females

Upon placing the decoy and playback equipment at a nestbox, we retreated at least 30 m to observe behavioral responses to the STI. We recorded all observations into a digital voice recorder and later transcribed behaviors using the event-recording software JWatcher (Blumstein et al. 2006). Observed aggressive behaviors included actions without direct physical contact: approaches (flying within 1 m of the decoy), dives (steep, rapid flights within 1 m of the decoy, often accompanied by vocalizations or clicks), hovers (stationary flight within 1m of the decoy), as well as direct physical contact: hits, pecks, and perches directly on decoy. As a measure of how much females engaged in aggressive behaviors throughout a 30 min trial, we calculated the proportion of minutes in which she performed at least one aggressive behavior (“proportion of time aggressive”). After a 30 min STI, we set a trap inside the nestbox (Stutchbury and Robertson 1986) to capture females. On average, females were caught 59.9 ± 4.6 min after the start of the STI (range = 31–80 min). Of the 33 STIs performed, 16 resulted in successful capture and sampling of females afterward. The proportion of time a female spent being aggressive during the 30 min trial did not predict likelihood of capture (Χ ² = 0.81, P = 0.37).

Control females

Stage-matched control females were trapped passively in nestboxes, with no exposure to STI decoys or audio (n = 23). Five of these 23 controls had their partially built nests removed 1–2 days prior to capture, as a part of another experiment. However, T levels were not significantly different between these two sets of females (t(29) = 0.64, P = 0.53), and so we did not differentiate them in downstream analyses.

Upon capture of females in both treatment groups, we immediately took blood samples and collected morphometric data (see “Both experiments” section below). All females were captured and sampled between 8:38 AM and 1:14 PM.

Experiment 2: Inducing real social competition

Manipulating competition for nestboxes

We experimentally induced genuine social instability by temporarily reducing the number of available nestboxes, as shown in Figure 1. Iterations of this experiment in Y1 and Y2 were conceptually similar but contained some methodological differences, detailed below. Additionally, a subset of females collected in Y2 were included in a project on neurogenomic responses to competition (Bentz et al. 2021).

Figure 1.

Conceptual overview of Experiment 2, wherein a subset of boxes were made newly accessible, while even more boxes were simultaneously made inaccessible. In Y1 (depicted), we accomplished this by closing boxes that were previously open, and opening “NEW” boxes that were previously closed. In Y2, boxes were completely removed or installed anew. In both years, “NEIGHBOR” boxes were untouched, occupied territories in close proximity to those boxes with altered accessibility.

In both years, as birds initially arrived at the breeding sites, we regularly monitored nestbox contents and tree swallow activity to identify territories. The placement of a few pieces of nesting material (grass, goose feathers) is often used by tree swallows to claim a box even before nest-building occurs in earnest (Winkler et al. 2020). In Y1, leg bands from previous capture allowed for visual approximations of some birds consistently seen at particular nestboxes. In Y2, the consistent presence of banded individuals was also confirmed using RFID readers at nestboxes (see Bentz et al 2021 for additional details). Tree swallows are known to occasionally defend more than one adjacent nestbox (e.g., Harris 1979; Robertson and Gibbs 1982). Within such “superterritories,” we determined which nestbox was likely to be the primary or preferred site based on nesting materials in and time spent at one box versus another in the days leading up to the experiment.

In both years, we designated neighborhoods of nestbox territories into experimental (Y1 n = 5; Y2 n = 2) or control (Y1 n = 4; Y2 n = 6) replicate treatment groups. Each replicate consisted of 3–16 occupied nestboxes, and most replicates were >300 m apart. This distance is sufficient because tree swallows reliably defend the immediate area around their box (<30 m, Robertson and Gibbs 1982; Rendell and Robertson 1989), and the behavioral effects of temporary nestbox manipulations are quite localized (as detected by in-person observations and RFID-equipped nestboxes; Rosvall 2008; Bentz et al 2021). In Y1, three replicates were <300 m apart; however, we temporally separated these groups over several days, such that spillover effects from geographically adjacent birds were unlikely.

For each replicate group, we waited until sunset, when birds leave for communal roosting sites (Stutchbury and Robertson 1990) to enact either the experimental or control treatment. Our goal at the experimental sites was to increase the frequency of aggressive conspecific interactions, among females seeking to acquire new territories and those defending pre-existing claims. To do this, we rendered approximately two-thirds of then-occupied boxes inaccessible at each experimental replicate. In Y1, we accomplished this by stapling cardboard over the nest entrances. In Y2, to increase the salience of our manipulation, we completely removed poles and boxes from the sites. The remaining one-third of occupied boxes within an experimental group were designated as “neighbors”; those boxes and their nest contents were left untouched. By changing box availability in the vicinity of these neighbors, we sought to measure the effects of surrounding competition on a bird who had not gained or lost its own nestbox. For every two occupied boxes that we closed (Y1) or removed (Y2), we also made one new nestbox available. In Y1, this involved a combination of opening some preemptively closed boxes (which had been stapled shut prior to the arrival of any birds at the sites), as well as erecting altogether new boxes. In Y2, we only added boxes via the second method of installing new boxes on the eve of the experiment start. All newly available boxes (hereafter “new”) were positioned roughly equidistant among inaccessible boxes and untouched neighbor boxes. This process served to concentrate competition among the displaced birds at the new boxes and near the aforementioned neighbors (Figure 1). To further increase the likelihood of conflict among birds, we also closed (Y1) or removed (Y2) every box that had not yet been claimed by a tree swallow, as well as surplus boxes that were part of apparent superterritories.

At control groups, we did not open/close or remove/install any nestboxes. Instead, in Y1 we removed the partially built nests of roughly half the breeding pairs. This generated two types of controls, allowing us to differentiate between potential effects of competition versus nest-building. Control females whose nest contents were removed faced a setback in nest-building status similar to experimental females whose boxes were closed/removed, but without requiring them to compete for new boxes prior to rebuilding. As control comparisons for neighbor experimental boxes, we also left some control boxes completely unmanipulated; these control and experimental females differed only in whether surrounding birds were competing or not. In Y2, informed by the lack of hormonal differences seen among control categories in Y1 (rebuilding or unmanipulated; see Results), we left all control boxes completely unmanipulated.

Quantifying aggressive behaviors

We used personal observations to measure behavioral responses to the experimental or control treatments. On the morning following each treatment start, prior to sunrise, we arrived at each replicate site. Over the following 2 to 3 h, as birds arrived back from their overnight roosts, we conducted sequential 20 min observations, usually with 10 min pauses between sessions (Y1 n = 92 experimental and 42 control observation sessions between 6:53 AM and 12:05 PM; Y2 n = 18 experimental and n = 15 control sessions between 7:16 AM and 1:24 PM). At larger sites, where multiple observers were monitoring concurrently, we ensured that each box was only covered by one observer at a time. During each session, the observer used digital voice recorders to note the presence and behavior of all tree swallows in their line of sight (on average 4.3 ± 0.2 occupied boxes per observer). Continuous behavioral observations were used in order to maximize the likelihood that aggressive events were recorded. We defined “physical aggression” as attempted or successful physical contact (e.g., hits or grappling in the air or on the ground), because these are the same behaviors that naturally occur as birds establish territories. We later transcribed audio recordings and calculated rates of aggression for each session as the proportion of observation minutes that contained any acts of physical aggression, divided by the estimated number of tree swallow pairs within the observed area. For experimental groups, this number was the combined total of neighbor and closed/removed boxes; for controls, this was simply the number of boxes occupied by tree swallows.

Using RFID to measure territorial intrusions

In Y1, a subset of nestboxes were equipped with RFID systems that we used to detect territorial intrusions (n = 7 experimental, 10 control boxes). We placed RFID readers beneath each nestbox (Gen2 model, Bridge and Bonter 2011) and 2.5 in-diameter antennas around nestbox entrances, which detected and recorded time and ID whenever birds with PIT tag-embedded leg bands were perched at or went through the box entrances. RFID readers were programmed to scan every 500 ms to detect any PIT tags at the box entrance and record date, time, and tag identity. To account for individuals perching or hovering at the antenna for an extended period of time, we filtered out sequential reads by the same individual. We considered any reads not within 3 s of one another to be separate RFID “events.”

Visual inspection of RFID events over time, overlayed with PIT tag ID, made apparent the identity of owners at a given box, or revealed that the box owners were not already banded. We thus considered any events by individuals who were not the apparent owners to be “intrusions.” We calculated the total number of female intrusion events per box on the first and second mornings post-treatment (prior to 1 PM, to correspond with capture efforts—see below). This method probably underestimates the true rate of intrusions because not all birds were equipped with PIT tags before the experiment began. However, the proportion of tagged to untagged females captured during the experiment did not differ between treatment groups (X² = 0.072, P = 0.79). Therefore, while these data potentially underestimate intrusions from untagged birds, they do so equally well between treatments.

Capturing after treatment

Approximately 2 to 3 h after birds arrived at nest sites on the morning post-treatment, we set nestbox traps in newly available and neighbor boxes. In Y1, captures occurred on the first two mornings after treatment (n = 19 experimental, 10 control). In Y2, all captures occurred on the first morning birds experienced the treatment (n = 11 experimental, 6 control). All captures occurred between 9:00 AM and ~1:30 PM (Y1 range= 9:18 AM to 12:43 PM; Y2 range = 9:07 AM to 1:32 PM). Within experimental groups, we primarily categorized females based on the type of box in which they were captured: “neighbor,” “floater,” or “new.” Neighbors were females known to be the owners of neighbor boxes. The vast majority of neighbor females were captured within their own boxes, but a few were trapped while inspecting or defending a new box, presumably in an attempt to defend it as a superterritory. Floaters were females trapped in neighbor boxes that they clearly did not own at the time of the experiment, based on observations and/or RFID. Given that cavities are a limiting resource necessary for reproduction, female floaters are common in tree swallow populations. While intruding floaters are usually chased off quickly by territorial females, they can succeed in entering nestboxes (Stutchbury and Robertson 1987a). Any female captured in a new box that was not known to currently own a neighbor territory was considered “new.” This category included a mixture of females who were territory owners of boxes that were closed or removed for the experiment as well as females who were floaters (i.e., territory-less) prior to the experiment. Post hoc, we separately tested whether experimental category (Y1: n = 11 new, 6 neighbor, 2 floater; Y2: n = 1 new, 6 neighbor, 4 floater) or control category (Y1: n = 7 emptied, 3 neighbor) had any effect on T levels within each treatment group and year (Supplementary Table 2). We found no evidence of T differences between categories in any year or treatment group, so we did not include them in further analyses. For illustrative purposes, we still differentiate them by shape in Figure 4.

Figure 4.

Females caught after a reduction in available nestboxes had significantly lower circulating T levels compared to stage-matched controls in Y2 but not Y1 of Experiment 2. There was no effect of box-type (neighbor, new, etc.; denoted by shape). All T concentrations were log-transformed. Lines and shading represent mean ± SE.

Both experiments

Blood collection and morphometrics

Upon capture in a nestbox, we immediately (within 5 min) took blood samples for the purpose of measuring circulating T levels. Using 27-gauge needles and heparinized microcapillary tubes, we collected 80–100 µL of blood from the alar vein. In Y2 of Experiment 2, females were euthanized and tissues collected for another project (Bentz et al. 2021), so we instead collected trunk blood immediately post-decapitation using heparinized collection tubes (BD Microtainer no. 365965). Blood samples were stored on blue ice in the field. We centrifuged blood that same day and used a 100 µL Hamilton syringe to isolate plasma, which was stored at −20 °C until hormone assays.

If previously unbanded at the time of capture, a bird was given a unique numbered USGS aluminum band on one leg and a color band with an embedded PIT (Passive Integrated Transponder) tag on the other leg (2.3 mm EM4102, IB Technology, U.K.- modified to fit tree swallow tarsi). Approximately two-thirds of captured females were already banded prior to sampling (Experiment 1 Y1: 36 out of 49, Experiment 2 Y1: 20 out of 29, Experiment 2 Y2: 9 out of 17).

Prior to release, we also collected data on body mass, wing length, sex, and age. Because we aimed to sample females shortly after territory establishment and prior to egg laying, we could not use the presence of brood patches or cloacal protuberances to determine sex. Instead we used a combination of plumage color (Hussell 1983; Dakin et al. 2016), wing length (Stutchbury and Robertson 1987c), behavioral observations, and capture information during prior years and subsequent breeding stages. Furthermore, we later confirmed that all females in both experiments had T levels comparable to what we have previously observed in females (George and Rosvall 2018) and not at all comparable to what we have seen in males at this breeding stage (Lipshutz and Rosvall 2021). For all euthanized birds, sex was confirmed post-mortem. Tissue collection also allowed us to confirm a pre-fertile breeding stage for females in Experiment 2 Y2, as they all had recrudesced ovaries but no yolky follicles. For females in Experiment 1 and Y1 of Experiment 2, we continued to monitor their nests post-release to confirm that they did not begin laying within 6 days—the approximate duration of egg production in this species (Ardia et al. 2006). Any females who did so were excluded from further analyses to avoid any confounds with fertility status.

Due to the unique delayed plumage maturation seen in female tree swallows, we were able to determine a female’s age category, which has been associated with a number of behavioral and life history traits (Winkler et al. 2020). Females with <50% blue/green dorsal plumage were classified as 1 year old (aka “yearlings”) and females that were >90% blue/green as 2+ years old. A few females in Experiment 2 could not be aged exactly because they exhibited intermediate levels of ornamentation (i.e., their dorsal plumage was 50–90% blue/green; n = 5). Since they were likely to be 1–2 years old (Hussell 1983), we included them in the yearling age category. In all, there were 14 yearlings and 24 older females in Experiment 1, 7 yearlings and 22 older females in Experiment 2 Y1 and 7 yearlings and 15 older females in Experiment 2 Y2.

Measuring T concentrations

T concentrations were measured using High Sensitivity Testosterone ELISA kits (Enzo #ADI-900-176, Farmingdale, NY, USA), previously validated in this species (George and Rosvall 2018). For the vast majority of samples, we extracted hormones from 40 μL of plasma (3× extractions with diethyl ether) and reconstituted in 250 μL assay buffer. In a few cases with <40 μL of plasma, we supplemented plasma with ultrapure H₂O to reach a total volume of 40 μL, and mathematically corrected for the change in starting plasma volume. All samples were run in duplicate, and treatments were randomized and balanced across plates, but each experiment/iteration was run separately. A standardized pool of previously extracted hormones was run in multiple times on each plate in order to measure intra- and inter-plate variation. For Experiment 1, inter-plate variation was 6.1% and average intra-plate variation was 3.4 ± 1.8% (across three plates). For Experiment 2, inter-plate variation was 1.2% (Y1), and average intra-plate variation was 5.5± 0.4% (Y1, two plates) and 3.8% (Y2, one plate).

Statistical analyses

All analyses used R (v 3.6.1, R Core Team 2019). We used either linear models (LMs), linear mixed models (LMMs), or generalized linear mixed models (GLMMS), as dictated by the data and questions. We created all LMMs using the “lmer” function in package “lme4” (Bates et al. 2015). We used type III analysis of variance with Satterthwaite’s method to test for significance of fixed effects within each LMM, via the package “lmerTest” (Kuznetsova et al. 2016). We confirmed that all linear models satisfied assumptions of normality and homoscedasticity by visually inspecting normal quantile plots of final model residuals as well as plots of predicted versus residual values (Zuur et al. 2010). We generated all GLMMs with the package “glmmTMB” (Magnusson et al. 2019). We used the “simulateResiduals” function in the package “DHARMa” (Hartig 2019) to visually inspect residuals from each GLMM.

For all analyses involving T, we first log-transformed T concentration values. Each T-related model included fixed effects of body mass, because mass often covaries with T levels in songbirds (e.g., Jawor et al. 2007; George et al. 2021), and age category, because of apparent age-related differences in GnRH-induced T levels (George and Rosvall 2018), in addition to experiment-specific effects described below. We also checked for inadvertent effects of time of day and latency from capture to blood collection. Upon finding these latter two variables to be uncorrelated with T in each experiment-year (P > 0.10 for all Spearman rank tests), we omitted them from downstream analyses to avoid overparameterizing models.

Experiment 1

We first used an LM to test for an effect of treatment (STI or control) on T levels. Within only experimental females, we then tested for effects of aggressiveness and timing of the STI relative to sampling using an LM that included proportion of time aggressive and time from STI start to capture (min) as fixed effects.

Experiment 2

Given differences in experimental protocols, we analyzed Y1 and Y2 separately. To test whether there was an effect of treatment on observed rates of aggression, we generated GLMMs with zero-inflated, log-linked Gamma distributions. Each model included time of day and treatment as fixed effects, with replicate group ID as a random effect. Y1’s model also included experiment day (first or second morning after treatment) and a treatment by day interaction as additional fixed effects. To test for an effect of treatment on RFID intrusions in Y1, we created a GLMM with a zero-inflated Poisson distribution, in which the dependent variable was the cumulative number of female intrusion events per box. Treatment was a fixed effect, and replicate ID was a random effect. To assess the effects of increased competition on circulating T levels, we generated LMMs for each year that included treatment as a fixed effect and replicate ID as a random effect. Y1’s model also included a fixed effect of experiment day as well as a treatment by day interaction (not included in Y2’s model because all samples in that year were collected on the first day that birds experienced the treatment).

RESULTS

Experiment 1

There was a significant effect of treatment, such that experimental females had lower T levels compared to controls (Table 1, Model 1; Figure 2a). T concentrations also positively covaried with body mass, with heavier females having higher T levels. In general, experimental females responded quite aggressively toward the STIs (mean proportion minutes with aggression = 0.77 ± 0.06). There was a significant effect of time spent being aggressive on log- T, such that females who were more aggressive had lower T levels after the STI (Table 1, Model 2; Figure 2b). However, there was no effect of elapsed time from STI to capture.

Table 1.

Results of linear models testing for effects of STI and aggression on circulating T levels

Fixed effect	Estimate ± SE	F-statistic	P-value
Model 1 (both treatment groups)
Treatment	−0.51 ± 0.25	F(1, 37) = 8.11	0.0074
Mass (g)	0.18 ± 0.08	F(1, 37) = 4.84	0.035
Age category	0.09 ± 0.24	F(1, 37) = 0.14	0.72
Model 2 (STI-only)
Proportion of time with aggression	−0.87 ± 0.64	F(1, 14) = 7.79	0.019
Latency from STI to capture (min)	−0.002 ± 0.005	F(1, 14) = 0.27	0.62
Mass (g)	0.29 ± 0.11	F(1, 14) = 5.53	0.04
Age category	−0.25 ± 0.29	F(1, 14) = 0.73	0.41

Model 1 includes n = 15 experimental females captured after 30 min STIs and n = 23 control females, while Model 2 includes only the experimental females. For both models, T concentrations were log-transformed. Treatment effects are relative to controls and age effects are relative to older (2+ years old) females. Bold P-values indicate a significant (P > 0.05) result.

Figure 2.

(a) Females caught after a 30 min STI had significantly lower circulating T levels compared to controls. Lines and shading represent mean ± SE. (b) Females that spent a greater proportion of time behaving aggressively during an STI had significantly lower circulating T levels. Shading represents 95% CI. T concentrations are log-transformed for normality.

Experiment 2

In both iterations, rates of physical aggression were higher at experimental sites compared to controls (Table 2; Figure 3).

Table 2.

Effects of treatment on rates of physical aggression (proportion of minutes per observation session that contained physical aggression, divided by the number of estimated tree swallow pairs)

Fixed effect	Estimate ± se	Z-statistic	P-value
Y1
Treatment	0.59 ± 0.22	2.72	0.0066
Experiment day	−0.23 ± 0.35	0.66	0.51
Treatment*Exp day	−0.09 ± 0.39	0.23	0.82
Time of day	0.10 ± 0.07	1.45	0.15
Y2
Treatment	2.21 ± 0.34	6.55	<0.0001
Time of day	0.12 ± 0.07	1.69	0.09

Models for both years were zero-inflated GLMMs with Gamma distributions. Effects of treatment are relative to control sites and the effect of experiment day in Y1 is relative to the first morning. Bold P-values indicate a significant (P > 0.05) result.

Figure 3.

Physical aggression occurred more frequently at experimental replicates than controls during both iterations of Experiment 2. Rates were calculated as the proportion of minutes within a single observation session in which aggression was observed, divided by the estimated number of tree swallow pairs in the vicinity. Each dot represents one 20 min observation period. Additional fixed effects are described in Table 2.

Intrusions by non-territory holding females occurred more frequently at experimental versus control boxes in Y1, as detected via RFID reads (average of 4.7 ± 2.4 cumulative intrusions per experimental box, 0.2 ± 0.2 intrusions per control box; β = 3.15 ± 0.89, z = 3.52, P = 0.00043). There was no effect of treatment on T levels in Y1 (Table 3, Y1; Figure 4a), but T levels were significantly lower among experimental females than controls in Y2 (Table 3, Y1; Figure 4b).

Table 3.

Effects of treatment on log-transformed testosterone levels (Y1 n = 19 experimental, 10 control females; Y2 n = 11 experimental, 11 control)

Fixed effect	Estimate ± se	F-statistic	P-value
Y1
Treatment	−0.21 ± 1.31	F(1, 1.5) = 0.10	0.79
Experiment Day	−0.04 ± 0.31	F(1, 16.3) = 0.24	0.63
Treatment*Exp Day	0.28 ± 0.40	F(1, 17.3) = 0.49	0.49
Mass	0.04 ± 0.06	F(1, 22.5) = 0.32	0.57
Age category	0.04 ± 0.24	F(1, 20.8) = 0.02	0.89
Y2
Treatment	−0.12 ± 0.47	F(1,18) = 6.88	0.017
Mass	−0.31 ± 0.20	F(1,18) = 2.42	0.14
Age category	−0.19 ± 0.35	F(1,18) = 0.30	0.59

Effects of treatment are relative to the control group, effect of Day is relative to the first treatment morning, and effects of age are relative to older females (2+ years old). Bold P-value indicates a significant (P > 0.05) result.

DISCUSSION

We used two separate experimental approaches to explore how defense of a nesting territory influences female aggression and testosterone secretion. In one experiment, we used the classic method of a simulated territorial intrusion via taxidermic decoy on a female’s nestbox. In another experiment with two iterations across years, we exacerbated rates of natural intrusions by temporarily reducing local nestbox availability. In all cases, we observed marked aggressive behaviors by females, directed toward their live or taxidermic challengers. Following the challenge hypothesis originally applied to male vertebrates, we predicted that females exposed to either real or simulated social challenges would have elevated T in circulation. Instead, we found that experimental females did not have higher T levels, and in some cases, circulating T concentrations were lower in socially challenged females than in controls. Likewise, T levels after STIs were negatively correlated with time spent engaging in aggressive behaviors. Although these patterns are not in line with the idea that social challenges prompt T elevation, they nevertheless connect T to territorial aggression in females. Below, we explore why T levels do not elevate and why they may sometimes decrease, particularly if high T is deployed to facilitate the urgent demands of a discrete competitive event.

Comparing simulated social challenges and genuine social competition

Females exhibited strong behavioral responses to manipulations of their competitive environment, with repeated dives, hits, and pecks at taxidermic decoys during simulated territorial intrusions, and frequent chases, hits, and grappling after a reduction in available nest sites. However, we found no evidence that females exhibited elevated T levels following either type of experimental manipulation. Our two experimental approaches complement one another in their strengths and limitations, allowing us to reject a few potential explanations for the lack of T elevations that are more akin to methodological artifacts than insights into how birds truly respond to changes in their social environment.

A major strength of the STIs used in Experiment 1 is that they are standardized, so each subject receives a comparable challenge. Additionally, direct observations of the territory-holder during each challenge allowed us to measure how long and to what extent each female responded behaviorally to the challenge. A potential drawback of STIs is that they are in fact simulated. If subjects do not perceive taxidermic decoys as competitors, then STIs cannot address how hormone levels respond to social challenges. Given the robust and prolonged aggressive responses we observed during STIs, however, this seems unlikely to be the case. Another possible explanation is that one 30 min intrusion by a single intruder is insufficient to elicit a hormonal response in this system. While male song sparrows have been shown to elevate T levels after only 10 min of STI exposure (Wingfield and Wada 1989), male spotted antbirds only showed an increase in T levels after 2 hours of conspecific song playback (Wikelski et al. 1999). Multiple temporally-separated intrusions, challenges from multiple individuals, or a single intrusion lasting longer than 30 min may be more predictive of significant changes in the competitive social environment, and thus more effective stimuli for eliciting hormonal responses that, at least in principle, should prepare animals for further competition (e.g., Wikelski et al. 1999; Oyegbile and Marler 2005).

Conversely, subjects may view decoys as especially formidable competitors, since they are permanently in a defensive posture and refuse to retreat. The perceived ability of a competitor has been shown to affect the magnitude and direction of behavioral and hormonal responses in some systems. For example, male Japanese quail (Coturnix japonica) and cichlid fish (Oreochromis mossambicus) increase androgen production in response to real fights but not to simulated challenges with their own mirror image, presumably because the mirror “competitors” equally match the subjects in fighting ability (Oliveira et al. 2005; Hirschenhauser et al. 2008). Male European robins (Erithacus rubecula) show stronger aggressive responses to stuffed versus live decoys during simulated intrusions (Scriba and Goymann 2008), suggesting that stationary decoys are perceived as greater threats, similar to the mirrored fish. Likewise, systemic T levels in male California mice only increase after a male wins, and not if he loses a contest (Oyegbile and Marler 2005).

However, results from Experiment 2 also provide support to the idea that female T levels do not increase after social challenges, so the nature of STIs cannot fully explain the lack of T elevation. Behavioral observations and RFID activity show that we did successfully increase competition for nest sites among females in the experimental treatment groups. After we reduced the number of available nestboxes at a site, we observed more instances of aggression at the experimental sites and we detected more PIT-tagged intruders at claimed boxes. Among experimental birds, T levels did not differ between winners at new boxes and neighbor females who maintained box ownership throughout, suggesting we did not overlook a winner effect that was dependent on the duration of box ownership. That T levels were not higher in experimental females overall compared to date- and stage-matched controls suggests that the increased rate of real territorial intrusions did not cause T levels to go up, despite the “realness” of the social challenges they experienced.

One potential drawback of the real challenges in Experiment 2 is that they are inherently more variable than the standardized challenges of Experiment 1. Even among females in Experiment 2, individuals may not have experienced the exact same intensity or frequency of challenges. If T is dependent on the recency and/or duration of a social challenge, then we might expect greater variation in T responses among Experiment 2 females. Though we did not directly compare years, the degree of experimentally induced competition seemed to vary between Y1 and Y2, likely driven by changes in our methods: birds seemed more inclined to investigate and compete over new boxes once their old box was completely gone (as in Y2), rather than merely inaccessible (as in Y1). These methodological differences may have contributed to different patterns of T responses seen in Y1 and Y2. However, in neither year did we find any evidence of socially-induced T elevation, consistent with our results from Experiment 1.

Explaining the lack of challenge-induced T elevations

This lack of competition-induced T elevation is noteworthy in light of previous experimental work on T and aggression in this system. Past work has shown that, when nestboxes are limited, females who respond more aggressively to simulated intruders are more successful at obtaining a nesting site (Rosvall 2008). Additionally, experimental increases in circulating T levels (via subcutaneous implants) cause females to behave more aggressively toward simulated intruders within a few days of treatment (Rosvall 2013). Together, these studies suggest that increasing T production after a social challenge, for the purposes of enhancing aggression, should be an adaptive response to increased competition. Given that this mechanism of response occurs in a wide variety of vertebrate taxa (Hirschenhauser and Oliveira 2006; Moore et al. 2019), the question remains: why does it not occur in female tree swallows? One possibility is phylogenetic inertia, if songbirds generally do not elevate T following social challenges (Goymann et al. 2019). However, given other cases of socially-induced T elevations in males and females of this taxonomic order (e.g., Wikelski et al. 1999; Gill et al. 2007) this is not a wholly satisfactory solution. Alternatively, females may already be secreting T at maximal levels within their own reactive scope. In a prior experiment, though, we showed that 30 min after an exogenous influx of GnRH, females in this same breeding stage can roughly double T levels in circulation (George and Rosvall 2018). Therefore, females could elevate T, but they do not do so following social challenges.

If there is not a physiological limitation on production, might there be additional costs of elevated T levels that outweigh putative benefits to females’ competitive abilities? Decades of comparative and experimental work have linked higher T to immunosuppression, risk of injury, and energy expenditure (reviewed in Wingfield et al. 2001). Exogenous T also may interfere with aspects of female reproduction including mate choice (McGlothlin et al. 2004), nest-building (Searcy 1988), and egg laying (Clotfelter et al. 2004; Veiga and Polo 2008; Berzins et al. 2018). Whether costs of T are visible after transient elevations, within an individual’s own reactive scope, is an important empirical question that is rarely tested, with limited and mixed results (Millesi et al. 2002; Peluc et al. 2012; Goymann et al. 2015; Goymann and Dávila 2017; George et al. 2021).

Moving beyond these changes in T in circulation, there may also be mechanisms to promote aggression in other ways. Social challenges can alter the steroid hormone microclimate within the brain, via local synthesis or conversion of adrenal steroid hormones (Soma et al. 2008; Pradhan et al. 2010; Rendon and Demas 2016) or changes in androgen receptor density (Hattori and Wilczynski 2014). Likewise, a recent study using a subset of individuals involved in Y2 of Experiment 2 showed that social challenges can upregulate brain gene regulatory networks related to catecholamines and dopamine (Bentz et al. 2021), all of which have connections to aggression (Duque-Wilckens and Trainor 2017; Rosvall et al. 2020). Indeed, there are many diverse mechanisms that may shape how females respond to increased rates of competition.

Interpreting the paradox that T enhances aggression, but aggression can suppress T

Beyond finding that T levels clearly did not elevate following a social challenge, we also found evidence of T levels decreasing in response to heightened competition and territorial intrusions. T levels were lower in experimental females in both Experiment 1 and the Y2 iteration of Experiment 2. Among females exposed to an STI in Experiment 1, T levels negatively correlated with time spent being aggressive, suggesting that the act of engaging with intruders is what caused T levels to decrease and resulted in the overall differences between treatment groups. T responses in this direction are not unique to female tree swallows: lower T levels following a social challenge have been documented in female song sparrows (Elekonich and Wingfield 2000), male great tits (Pinxten et al. 2004), male blue tits (Landys et al. 2007), female eastern bluebirds (Navara et al. 2006), male rufous horneros (Adreani et al. 2018), and female barred buttonquails (Muck and Goymann 2018). This suggests that female tree swallows’ T responses are not an exception, but rather part of a broader pattern of one type of physiological response to competition.

It is therefore worth considering how and why circulating T levels may decrease following competition. Changes in the social environment are known to alter several components of the hypothalamo-pituitary-gonadal axis, which regulates T production. For example, male Japanese quail, which rapidly decrease T levels upon viewing females (Cornil et al. 2009), also respond to that social stimulus by increasing gonadotropin-inhibiting hormone (GnIH) expression in the hypothalamus, and by reducing circulating levels of luteinizing hormone (Tobari et al. 2014). Acute stress can also lead to rapid decreases in circulating T levels, apparently driven by changes in gonadal activity (Deviche et al. 2010) that may or may not be regulated by concurrent changes in corticosterone levels (McGuire et al. 2013; Deviche et al. 2017). Given that corticosterone levels often increase following STIs (Van Duyse et al. 2004; Deviche et al. 2014), the lower T levels among our experimental birds could be at least partially due to interactions between sex steroids and corticosteroids. Changes in the social environment also may affect metabolic pathways that inactivate and clear T from the body—a process that occurs predominantly in the liver (Mueller et al. 2015). However, there is little evidence that changes in hepatic activity drive changes in circulating T levels (e.g., Wilson and LeBlanc 2000; Bentz et al. 2019a). Regardless of the exact mechanism, a reduction in T levels after real or simulated competition could be part of an adaptive strategy to temporarily reduce allocation toward sex steroid-mediated sexual or reproductive behavior, to focus on the challenge at hand.

T levels could also decrease if T is “used up” or converted into other active metabolites, such as estradiol (E₂) or dihydrotestosterone. Indeed, there is evidence that changes in the activity of aromatase (the enzyme that converts T to E₂) can occur rapidly in response changes in the physical or social environment (reviewed in Trainor et al. 2006; Cornil 2018) and that elevations in neural E₂ levels can increase aggression on the timescale of minutes via non-genomic pathways (Heimovics et al. 2015). Our finding that T levels negatively correlate with time spent being aggressive during the preceding 30 min STI supports this interpretation. Notably, baseline or pre-challenge T levels positively correlate with aggressiveness in female tree swallows (Lipshutz and Rosvall 2021). This positive correlation with aggression at baseline, coupled with a negative correlation with aggression when challenged, is consistent with the idea that T levels may decline following social challenges because T or its metabolites bring about higher aggression (sensu Rosvall 2013). After all, individuals who began a challenge with the highest T may have more T on hand to facilitate a sustained aggressive response, even if T is depleted after the encounter. As behavioral ecology becomes increasingly intertwined with endocrinology, we encourage greater attention to the complex and bidirectional mechanisms linking hormones and behavior (Lipshutz et al. 2019), particularly since different mechanisms may have distinct implications for present and future behavior.

FUNDING

This work was supported by National Science Foundation grants to KAR (IOS-1656109), EMG (NSF graduate research fellowship). EMG, SEW, ABB, and KAR were all also supported by the National Institute of Health (T32HD049336). EMG was also supported by a research grant from IU’s Research and Teaching Preserve.

Conflict of Interest statement: The authors declare no conflict of interest.

Data availability: Analyses reported in this article can be reproduced using the data provided by George et al. 2021.