Bidirectional Relationships between Testosterone and Aggression: A Critical Analysis of Four Predictions

Abstract

Experimentally elevated testosterone (T) often leads to enhanced aggression, with examples across many different species, including both males and females. Indeed, the relationship between T and aggression is among the most well-studied and fruitful areas of research at the intersection of behavioral ecology and endocrinology. This relationship is also hypothesized to be bidirectional (i.e., T influences aggression, and aggression influences T), leading to four key predictions: (1) Individuals with higher T levels are more aggressive than individuals with lower T. (2) Seasonal changes in aggression mirror seasonal changes in T secretion. (3) Aggressive territorial interactions stimulate increased T secretion. (4) Temporary elevations in T temporarily increase aggressiveness. These predictions cover a range of timescales, from a single snapshot in time, to rapid fluctuations, and to changes over seasonal timescales. Adding further complexity, most predictions can also be addressed by comparing among individuals or with repeated sampling within individuals. In our review, we explore how the spectrum of results across predictions shapes our understanding of the relationship between T and aggression. In all cases, we can find examples of results that do not support the initial predictions. In particular, we find that Predictions 1–3 have been tested frequently, especially using an among-individual approach. We find qualitative support for all three predictions, though there are also many studies that do not support Predictions 1 and 3 in particular. Prediction 4, on the other hand, is something that we identify as a core underlying assumption of past work on the topic, but one that has rarely been directly tested. We propose that when relationships between T and aggression are individual-specific or condition-dependent, then positive correlations between the two variables may be obscured or reversed. In essence, even though T can influence aggression, many assumed or predicted relationships between the two variables may not manifest. Moving forward, we urge greater attention to understanding how and why it is that these bidirectional relationships between T and aggression may vary among timescales and among individuals. In doing so, we will move toward a deeper understanding on the role of hormones in behavioral adaptation.

Introduction

From its inception, the discipline of endocrinology has been closely linked with the study of behavior. In the classic experiment credited for launching the field, Berthold removed the testes of male chickens, noting that the capons (castrated males) exhibited lower levels of aggression. Critically, the capons regained their aggressive tendencies after surgically replaced testes became vascularized, suggesting that something produced by the gonads and secreted into the bloodstream was responsible for the behavioral change (Berthold 1849; Berthold and Quiring 1944). That “something” was later isolated and characterized as the steroid hormone testosterone (T).

Since Berthold, many experiments have further implicated T as a key mediator of aggressive behavior. In the lab, castration can lead to a reduction in aggression, while treatment with exogenous T can restore the behavior (e.g., Beeman 1947; Barkley and Goldman 1977; Moore 1988). Manipulations in free-living vertebrates provide further evidence of a causal link between T and aggression: animals whose T levels are experimentally increased are more aggressive than animals given sham treatments (reviewed in Lynn 2008; Rosvall et al. 2020). Pharmacological treatments with drugs that prevent the binding or metabolism of androgens can also diminish aggression (e.g., Schlinger and Callard 1990; Sperry et al. 2010; but see Apfelbeck et al. 2013). Though aggression is not always affected by T treatment (Lynn et al. 2005; Lynn 2008), the many cases in which the hormone does appear to promote aggression include both male and female subjects in many vertebrate taxa.

With the refinement of methods for measuring circulating hormone levels (Wingfield and Farner 1975), researchers began examining more relationships between T and aggression in free-living animals. These studies often use similar techniques (e.g., trapping or mist-netting, collecting a blood sample, and radioimmunoassays or enzyme immunoassays for quantifying hormone titers) and behavioral measures (e.g., simulated territorial intrusions and scan or focal sampling for measurements of aggression). Much of this work began in wild birds (Wingfield et al. 1990), leading to an over-representation of birds in this line of research (Vitousek et al. 2018a; Wingfield et al. 2019). Nevertheless, relevant research has also used reptiles, fish, and mammals, including humans (Archer 2006; Moore et al. 2019). Across this line of research , studies vary in how T and aggression are examined in relation to one another.

We identified four common types of questions that attempt to relate circulating T levels to aggressive behavior (Fig. 1). For each question, we present a prediction that is based on the positive link between exogenous T and aggression seen in experimental manipulations. Prediction 1: Individuals with higher T levels are more aggressive than individuals with lower T. Prediction 2: Seasonal changes in aggression mirror seasonal changes in T secretion. Prediction 3: Aggressive territorial interactions stimulate increased T secretion. Prediction 4: Temporary elevations in T temporarily increase aggressiveness. These predictions vary in their directionality: Do T fluctuations influence aggression, or do aggressive interactions change T secretion? And some predictions explore co-variation between T and aggression without addressing causality. These questions and predictions can be framed at the level of the individual, or they may apply to comparisons among individuals. Some are tested at different timescales, ranging from minute-to-minute changes in hormones or behavior to those changes that are visible over the course of weeks, reproductive cycles, or seasons. Others have recently explored these questions at even longer (evolutionary) timescales (i.e., comparisons among species; Husak et al. 2021), and so we have elected to focus on questions pertaining to within-species variation. Careful examination of the nature of the questions being asked is critical, as growing evidence suggests that relationships may not be conserved across different levels of analysis (Goymann et al. 2007; Nussey et al. 2007; Breuner and Berk 2019).

Fig. 1.

Four central questions and predictions linking T and aggression in nature. Although many of these questions are explored using among-individual analyses, they also can be explored withinindividuals, except for question 1, which explores a snapshot in time. Shapes represent different individuals and lines represent reaction norms as individuals vary in either T or aggression across different timescales and in response to different stimuli or hormonal treatments.

In our review, we seek to explore how the spectrum of results for each prediction can and should be interpreted. It was our a priori impression that Predictions 1–3 have been tested most frequently, largely via among-individual comparisons. As we discuss below, all three predictions have some support, but there are also cases in which results appear contradictory. Prediction 4, on the other hand, is something that we identify as a core assumption of past work; however, it has rarely been tested directly. Our goal here is to integrate results across predictions, extending prior reviews that have primarily focused on one or another prediction (see corresponding reviews cited for each prediction below). As a case study, we also present findings from a series of studies, by ourselves and collaborators, in female tree swallows (Tachycineta bicolor), a songbird that competes intensely for nesting sites. Collectively, this exploration of past and ongoing work sheds light on the complex, bidirectional relationship between T and aggression across levels of analysis.

Prediction 1: differences in circulating T levels explain among-individual variation in aggression

If T experimentally promotes aggression, then a natural first question to ask is: Are circulating T levels and aggression positively correlated among individuals? Several studies have tested this prediction by measuring baseline T levels (i.e., when the animal is generally undisturbed) and by assaying aggression levels from those same individuals at another time. These assays typically use a simulated conspecific competitor, such as a taxidermic mount, a live caged conspecific, or playback of conspecific vocalizations. Alternatively, they may measure the outcomes of staged, paired contests between residents and live intruders. Some of these studies have found significant positive correlations between baseline T and aggression among individuals (Harding 1983; Geniole et al. 2020). Of course, a positive correlation is not proof in itself that T causes aggression; because T and aggression may co-vary with some third unknown variable, further experimentation is needed (i.e., via blockers or other pharmaceutical supplements). Critically, however, the inverse is also true: A lack of positive correlations (e.g.,Silverin et al. 2004; Ball and Balthazart 2008;Baird et al. 2014) is not necessarily proof against T facilitating aggression.

We can think of several reasons as to why experimental manipulations of T could increase aggression, yet natural variation in T among individuals would fail to predict among-individual difference in aggression. For one, hormones influence many different components of the phenotype, including behavior, via step-wise or other non-linear effects (Hews and Moore 1997; Adkins-Regan 2005). Another possibility is that the labile nature of circulating T levels, especially in response to environmental stimuli or changing internal states (reviewed in Kempenaers et al. 2008), may make direct relationships between T and aggression difficult to detect. In particular, basal T levels in songbirds exhibit circadian patterns, with peak levels seen at night (Greives et al. 2021). If the timing of sampling is not highly controlled, then “baseline” samples may not be directly comparable to one another.

More controlled measures of an individual's hormonal reactivity are possible. For instance, T levels measured following a standardized injection of gonadotropin-releasing hormone (GnRH; i.e., a “GnRH challenge”) are highly repeatable within individuals, at least within the same reproductive stage (Jawor et al. 2006). This approach essentially maximizes gonadotropin output, stimulating the gonad to secrete T, with a peak about 30–60 min after the GnRH injection (Jawor et al. 2006; Rosvall et al. 2016). There is substantial among-individual variation in T production, and some of this hormonal variation stems from standing variation in steroidogenic machinery in the gonad (Rosvall et al. 2016). In birds, prior research using GnRH challenges has linked T production with individual differences in aggressive and parental behavior (McGlothlin et al. 2007), ornamentation (McGlothlin et al. 2008), and reproductive success (Cain and Pryke 2016). Hormonal secretion also may have a heritable component (reviewed in Greives et al. 2017), suggesting that, even if there is noise in T secretion, some of that variation may reflect true individual differences.

A lack of correlation between circulating T and aggression may also occur if individuals vary in other mechanistic steps on the pathway between T and behavior. Within the androgen signaling system, many factors can contribute to behavioral differences—for example, binding globulins (Breuner and Orchinik 2002), tissue sensitivity (Ball and Balthazart 2008), or metabolism (Soma et al. 2015). For example, in a now classic study, Grunt and Young (1952) measured mating behavior in male guinea pigs and found substantial variation among individuals. Each male was then gonadectomized and treated with the same dose of T via subcutaneous implant. Not only did the T treatment restore mating behavior, but it also restored the same among-individual differences in behavior—males with the highest behavioral expression before any manipulation resumed high mating behaviors, whereas those with previously low behavioral expression resumed low mating behavior, despite all males having the same T inputs. This result has been taken to indicate that important behavioral variation lies in other aspects of the androgen signaling system, beyond T levels in circulation. Indeed, expression of steroid receptors and steroid-binding enzymes in socially relevant brain regions has been found to correlate with aggression ( Delville et al. 1984a;Trainor et al. 2006; Rosvall et al. 2012;Horton et al. 2014). Similarly, androgen receptor expression in muscle can also explain individual variation in androgen-mediated signaling behavior (Fuxjager et al. 2015). Though one or another of these endocrine parameters may be correlated with T production, that is not necessarily the case (Lipshutz et al. 2019), suggesting that some individuals may have lower T levels in circulation but they get “more bang for their buck” via high sensitivity to that T (sensu Canoine et al. 2007). Therefore, upon finding that circulating T and aggression are not correlated among individuals—despite experimental evidence that T does increase aggression—one must also consider other parts of the signaling pathway.

Even if individual differences in aggression do stem from differences in T to some degree, correlations between hormones and behavior may not occur when quantified using among individual analyses. This seemingly counterintuitive idea relates to observations that individuals may differ in quality or condition in ways that mask or reverse the relationship between two traits, with correlations only emerging when evaluating within-individual changes in both traits over time (Van Noordwijk and de Jong 1986; Careau and Wilson 2017; Breuner and Berk 2019; Laskowski et al. 2021 ). Addressing this issue requires that we delve further into temporal variability in T, aggression, and the relationship between them, not only among individuals but within an individual over time. We expand upon these points in the remaining predictions below.

Prediction 2: seasonal changes in aggression mirror seasonal changes in T secretion

Seasonal changes in T and aggression are one central component of the influential Challenge Hypothesis, introduced by Wingfield et al. (1990). The Challenge Hypothesis provides a series of explanations for why circulating T levels show so much variation within and among species, and one of these solutions lies in seasonally varying environments. In seasonally breeding animals, the need for aggression often changes with time of the year including, for example, winter vs. summer or different phases of a reproductive cycle. Often, competition is most intense during earlier stages of reproduction, such as territory establishment, courtship, or mating. In later breeding stages, conspecific aggression may become less necessary or beneficial, particularly if aggression trades off with time or energy that could be invested in parental care (Stiver and Alonzo 2009).

Over the past three decades, there has been much support for this “seasonal” aspect of the Challenge Hypothesis, especially in free-living songbirds (reviewed in Ketterson et al. 2005; Goymann et al. 2007; Rosvall et al. 2020). T levels tend to be higher during early stages of the breeding season, when competition for territories and mates is most intense. T levels often decline during periods of parental care, and further diminish during the non-breeding season. Patterns of aggression mirror these same temporal patterns, at least within the breeding season (Gowaty 1981; Sandell and Smith 1997; Ward and McLennan 2006). This broad-scale temporal alignment of T and aggression at the population level is consistent with observations that experimental T treatment often increases aggression.

So what can we learn from situations in which changes in aggression do not match seasonal changes in T? The most obvious examples occur when population-level expression of aggression remains relatively high, despite little to no gonadal T production. For example, male song sparrows (Melospiza melodia) and spotted antbirds (Hylophylax naevioides) still respond aggressively to simulated territorial intruders during the non-breeding months, even though T levels in circulation are essentially undetectable (Wingfield 1994; Hau et al. 2004). In both species, experimental and correlational evidence suggests that dehydroepiandrosterone (DHEA), an androgen precursor secreted largely by the adrenals, promotes aggression in the non-breeding months, in lieu of gonadal T (Hau et al. 2004; Wacker et al. 2008). A similar story can be found in both male and female Siberian hamsters, which are more aggressive during non-breeding seasons than during the breeding season (Jasnow et al. 2000; Munley et al. 2022). Though gonadal T production is much lower among non-breeding hamsters, DHEA levels are elevated (Scotti et al. 2008; Rendon and Demas 2016). The non-reproductive brain also may have greater sensitivity to sex steroids (Canoine et al. 2007; Wacker et al. 2010), which may be derived via neurosteroidogenesis or metabolism of adrenal precursors into more active forms (Soma et al. 2015). These studies highlight that relationships between T and aggression can change over the annual cycle, with stronger coupling at some times of the year and decoupling at other times of the year. These patterns also reveal how nature solved the apparent mis-alignment of selective pressures shaping T and aggression: when territorial aggression is adaptive but gonadal T is potentially costly (e.g., in the non-reproductive season; Wingfield et al. 2001), then aggression is facilitated without gonadal T (reviewed in Munley et al. 2018).

Most of the research evaluating Prediction 2 focuses on the population level, with limited consideration of individual differences. We know that, behaviorally, not all individuals track population-typical patterns—that is, variation can exist among individuals as to how their aggression changes from one breeding stage to the next (Araya-Ajoy and Dingemanse 2017). It is an unresolved empirical question whether some of this behavioral variation stems from individual differences in the seasonal regulation of T. As in the song sparrows, antbirds, and Siberian hamsters, any such deviations have the potential to tell us about the diverse mechanisms that may interact with or override variation in T to influence the adaptive expression of aggression in one or another context.

Prediction 3: T levels rapidly increase in response to a social challenge

Another key part of the challenge hypothesis is that social challenges, such as aggressive interactions with competitors, promote T elevation (i.e., the Challenge Hypothesis; Wingfield et al. 1990). In systems in which T promotes competitive phenotypes, a socially induced elevation of T may be an adaptive response to adjust behavior to match the current demands of social competition (i.e., the “winner effect”; reviewed in Hsu et al. 2006; Gleason et al. 2009). Initial evidence in support of this prediction was found in male song sparrows: males sampled after experiencing a simulated territorial intrusion had higher circulating T levels than passively caught controls (Wingfield 1985; Wingfield and Wada 1989). Over the past few decades, similar tests have been conducted in a variety of vertebrate taxa, including other songbird species (reviewed in Goymann et al. 2019), fish and reptiles (reviewed in Moore et al. 2019), and even humans (reviewed in Geniole et al. 2020). Many of these experiments suggest that T levels elevate in response to a social challenge, but there are also many examples in which T levels do not elevate, most notably in other songbird species (Goymann et al. 2019). In fact, several studies have found lower T levels in animals exposed to a social challenge, compared to controls, suggesting that T levels decreased in response to a social challenge (Harding and Follett 1979; Elekonich and Wingfield 2000; Pinxten et al. 2004; Landys et al. 2007; Adreani et al. 2018; Muck and Goymann 2018; George et al. 2022).

How do we reconcile situations in which T demonstrably promotes aggression, yet circulating T levels do not rise after a social challenges? One solution relates to the potential costs of elevated T, which are thought to stem from trade-offs among mating effort, parental effort, and somatic maintenance (Wingfield et al. 2001). Individuals may face selection to be “judicious” in their T responsiveness, only responding to a certain frequency, intensity, or type of challenge in which the presumed benefits of elevated T outweigh any ensuing costs. Subsequent results from male song sparrows are consistent with this idea, as T responses vary with the stimuli used (e.g., audio playback and live decoys), and it is the combination of these stimuli that elicits the greatest T elevation (Wingfield and Wada 1989). In addition, male spotted antbirds only appeared to elevate T after ≥2 h of continuous audio playback, suggesting that shorter playback durations did not sufficiently induce a hormonal response (Wikelski et al. 1999).

Some studies have found significant among-individual correlations between aggressive responses during a simulated challenge and post-challenge T levels (e.g., Buck and Barnes 2003; George et al. 2022), which is consistent with the idea that the intensity of the challenge or extent of behavioral response affects subsequent T secretion. Notably, though, much of the process by which a social challenge alters T secretion remains a black box (Ball and Balthazart 2020). This cascade should include sensory systems (i.e., visual, auditory, and olfactory cues) as well as ensuing effects on motivational and motor processes, but it is difficult to disentangle whether T levels are affected by the perception of a competitor vs. the subject's own aggressive intent or performance.

It is also reasonable, if not parsimonious, to expect animals of different quality or condition to have different degrees of social responsiveness, such that some individuals’ T secretion changes more quickly or with greater magnitude (i.e., animals differ in their reactive scope). Studies using GnRH challenges have shown that an individual's ability to elevate T is correlated with proxies of condition or quality, such as body mass or ornamentation (Millesi et al. 2002; McGlothlin et al. 2008; George et al. 2021). Furthermore, there is at least one study, in male zebrafish (Danio renio), showing that individuals vary in the timing of their androgen response to a social challenge, ranging from 2 to 15 min, with some individuals exhibiting a second peak 90 min later (Félix et al. 2020). Recent work on glucocorticoid responsiveness and flexibility highlights that endocrine reaction norms can vary among individuals (Guindre-Parker 2020), in part based on individual condition or resource acquisition (Breuner and Berk 2019). If individuals vary in the degree to which the social environment affects T levels, such a variation among individuals could obscure potential changes occurring within individuals. Collectively, these processes may affect our ability as researchers to see socially induced changes in T secretion when analyzed among individuals.

T levels may also appear unchanged or lower after a social challenge if there are other processes—beyond production—that affect T levels in circulation. For instance, T levels can rapidly decrease if T is used up or converted into other active metabolites, such as estradiol (E₂) or dihydrotestosterone, in the process of an individual responding aggressively. Activity levels of aromatase, the enzyme that converts T to E₂, can change rapidly in the brain in response to changes in the physical or social environment (reviewed in Cornil 2018; Trainor et al. 2006), and elevations in neural E₂ levels have been shown to increase aggression within minutes via non-genomic pathways (Heimovics et al. 2015). Similarly, if T is metabolized into inactive forms after hormone-receptor binding, we might expect to see lower T among animals in which T has recently activated behavioral changes. Although we are not aware of a direct empirical example with aggression, there is comparable evidence for courtship and copulatory behavior in Japanese quail: T and E₂ experimentally enhance or restore these behaviors (Balthazart et al. 1990; Cornil et al. 2006), but males exposed to a prospective female have lower T levels than controls (Delville et al. 1984b). Applying this phenomenon to the context of social challenges, the observation that T often decreases after a simulated territorial challenge could in fact be evidence for, rather than against, T's involvement in mediating the behavioral response.

Prediction 4: temporary elevations in T with an individual's reactive scope temporarily increase aggressiveness

We already know that males given T implants for days or weeks display increased mating effort, including heightened aggression, oftentimes alongside lower investment in parental effort (reviewed in Stiver and Alonzo 2009; Lynn 2016). Less clear is whether comparable behavioral shifts occur in response to short-lived, intra-individual fluctuations in T levels that occur as a part of an animal's natural hormonal reactive scope (i.e., within-individual changes in hormone secretion).

One direct approach to this question is to use within-subjects sampling of T levels, say, before and after an aggressive encounter. However, such repeated sampling also introduces handling-induced stress effects that may influence subsequent reactivity along the hypothalamo-pituitary-gonadal (HPG) axis (Deviche et al. 2012; Abolins-Abols et al. 2018), or influence an individual's aggressive response (Summers and Winberg 2006). Another direct approach is to use a more fleeting treatment, such as injections with T (Rutkowska et al. 2005), or less invasive methods like fast-acting hormone gels (Vitousek et al. 2018b) and hormone-laced food (Remage-Healey and Bass 2006; Heimovics et al. 2015). A third approach is to temporarily block either the binding of T to receptors or conversion of T to estradiol via pharmaceutical injections (e.g., Soma et al. 2000).

Finally, the GnRH challenge is another way to induce brief fluctuations in T, though this approach also affects other hormones too, both via the pleotropic effects of GnRH (Adkins-Regan 2005) as well as any stress responses induced by handling and restraint before the final blood draw (George et al. 2021). GnRH challenges are primarily used as bioassays of an individual's T production capabilities. However, they can also be used as an experimental treatment to temporarily maximize T production within an individual's current capabilities or reactive scope. In European ground squirrels (Spermophilus citellus), for example, males exhibit more aggression on the day after treatment with GnRH (Millesi et al. 2002). Presumably, this behavioral effect outlasts the initial (<1 h) hormonal effect because T's genomic effects take time, that is, to bind receptors and recruit various co-factors before eventually altering transcription and translation (Etgen and Pfaff 2010). On the other hand, a similar experiment with black redstarts (Phoenicurus ochruros) did not affect aggression (Goymann et al. 2015). Similarly mixed results are seen in response to GnRH injections and their effects on paternal care. GnRH injection of male redstarts leads to reduced paternal care in the 2 h post-treatment (Goymann et al. 2015; Goymann and Dávila 2017 ), whereas in male tree swallows (Tachycineta bicolor), it leads to increased parental care over the subsequent day (George et al. 2021). These case studies make it clear that behavior does not necessarily respond to endogenous T elevation in the same way that behavior responds to exogenous treatment.

We believe this apparent contradiction lies in the fact that GnRH injections limit T production to an individual's own physiological capabilities—which are likely to be both condition-dependent and highly variable among individuals. Treatment with exogenous T, on the other hand, may override such self-regulation to elicit a behavioral change that would not otherwise be induced. Along these lines, a recent meta-analysis found some indication that fitness effects of hormone implants, including T, depend upon starting hormone levels (Bonier and Cox 2020). In cases in which aggression and the ability to elevate T are both correlated with condition, individuals with greater hormonal reactive scopes may already be more aggressive, and therefore may not exhibit much of a behavioral change in response to endogenous T elevations. At the same time, individuals with the greatest potential to increase aggression may be the ones with the smallest potential to elevate T, resulting in little visible effect of endogenous T elevations (say, elicited with GnRH injections) on behavior. In this way, patterns seen within individuals could be masked or even reversed to generate positive correlations among individuals, if those individuals vary in condition or quality (Van Noordwijk and de Jong 1986; Laskowski et al. 2021 ). Presumably individual condition also may change over time, and further investigation of these within-individual perspectives will advance our understanding of the causal relationship between T and aggression, and how these relationships affect performance in nature.

Testing predictions in an aggressive female songbird

Though the vast majority of studies involving T and aggression have been conducted in male vertebrates, it is well-established that vertebrate ovaries are also capable of producing and secreting T (Staub and De Beer 1997; Ketterson et al. 2005; Goymann and Wingfield 2014; Geniole et al. 2017 ), and that T in circulation can influence female behavior, including aggression (Staub and De Beer 1997; Duque-Wilckens and Trainor 2017; Rosvall et al. 2020). Thus, female vertebrates that exhibit T-mediated aggression can be well-suited for testing the predictions outlined above.

We have been systematically examining the relationship between T and aggression in female tree swallows (Tachycineta bicolor), an obligate secondary cavity-nesting songbird species that readily breeds in artificial cavities (i.e., nestboxes) across North America. Females of this species often face intense competition for limited nesting sites (Leffelaar and Robertson 1985), and more aggressive females are more likely to obtain a nesting territory when territories are limited (Rosvall 2008). Throughout the breeding season, territorial females defend their nest sites against intruding, territory-less females (i.e., floaters) that threaten to evict or even kill them (Stutchbury and Robertson 1985). Past experimental work has shown aggression to be at least partially mediated by T in females of this species. Compared to controls, females given T implants behave more aggressively toward a same-sex decoy (Rosvall 2013). However, experimentally elevated T also disrupts incubation, an essential maternal behavior (Rosvall 2013). Based on these experimental results, and what we know about the birds’ life history, we would expect predictions in Fig. 1 to apply to this system. As discussed above, results that deviate from initial predictions may provide evidence of more complex processes at play.

In line with Prediction 1, there is some evidence that baseline T levels and aggression are positively correlated among individuals, at least during territorial establishment. Territorial females with higher circulating T levels spent more time attacking a taxidermic decoy during a simulated territorial intrusion at their nest boxes, whereas lower T females spent less time attacking (Lipshutz and Rosvall 2021).

In line with Prediction 2, baseline and GnRH-induced T levels decrease over the course of a breeding season, corresponding with observed changes in the competitive environment: higher baseline T levels were seen during earlier stages with more naturally occurring intrusions, with lower T levels during later stages with parental care (George and Rosvall 2018). Stage-to-stage variation in aggression directed at a staged territorial intruder was also higher during territorial establishment when T levels are higher, and lower during the chick period when T levels are low (Bentz et al. 2019b). In response to GnRH injection, females robustly elevated T, but only during those earlier breeding stages with more competition (George and Rosvall 2018). Changes in ovarian steroidogenic gene expression mirror these stage-to-stage patterns of variation (Bentz et al. 2019a), providing further indication that Prediction 2 generally holds in this system.

To evaluate Prediction 3, we tested for socially induced changes in T levels in circulation during territorial establishment. During this breeding stage, females are physiologically capable of elevating T to activation of the HPG-axis (George and Rosvall 2018). However, we found that females had lower T levels after real or simulated social challenges, relative to controls (George et al. 2022), echoing several other studies in songbirds (summarized above). We also found that the amount of time females spent behaving aggressively negatively correlated with T levels observed 30–60 min after the start of the simulated challenge, suggesting to us that the act of behaving aggressively caused T to go down. In fact, we recently found that while females engaged in intense competition for nestboxes, ovarian gene networks related to steroidogenesis were significantly downregulated, relative to stage-matched controls (Bentz et al. 2022). This aggression-sensitive gene network included steroidogenic acute regulatory protein, CYP11A1, P450scc, and adrenodoxin, all of which are key players in producing steroids from cholesterol. Notably, we found no immediate or lasting effects of competition on ovarian aromatase gene expression, suggesting that aggressive interactions do not lower T levels via conversion to E₂, at least in the ovary at the level of mRNA. We continue to explore potential effects of aggressive behavior on steroid metabolism in the brain, but our results thus far point to competition-induced changes in neural gene activity within non-steroid-mediated pathways (e.g., dopamine synthesis; Bentz et al. 2021); those same pathways have been shown to interact with steroid hormones in other systems, thus suggesting an additional layer of complexity to the interplay between T and aggression in this system.

Throughout these multiple ways of exploring Prediction 3, we have not yet evaluated the degree to which different individuals respond to challenges with varying effects on T production. However, we suspect this to be true, considering our observations that T is positively correlated with aggression at baseline (Lipshutz and Rosvall 2021) but negatively correlated after STI (George et al. 2022). Although much of the work on T and aggression assumes that individuals do not differ the slope of these reaction norms (note roughly parallel reaction norms represented as dotted lines in Fig. 1), we imagine that individuals vary in both their baseline hormone levels and their hormonal response to a challenge. If both of those traits are constrained by or proportional to other components of the phenotype, such as condition, then we would expect them to co-vary. We propose this scenario as a possible explanation for our initially perplexing findings that pre-challenge T levels positively correlated with aggression (Lipshutz and Rosvall 2021; Fig. 2A) but post-challenge T levels negatively correlated with aggression (George et al. 2022; Fig. 2B). We observed that birds exposed to a simulated territorial intruder had lower T than controls (George et al. 2022; bar graph within Fig. 2C). If hormonal responses to challenges are related to an individual's aggressive phenotype, then reaction norms of hormonal reactivity will not be parallel and an individual's slope of hormonal reactivity will be correlated with its intercept (Fig. 2C). Under this scenario, a “reversal” in the sign of the co-variation can readily occur (Fig. 2C). We do not yet know if this is the case, and empirically testing this idea has logistical hurdles, but greater attention to these sorts of within- vs. among-individual patterns will be crucial in the future.

Fig. 2.

In our tree swallow case study, we hypothesize that the most aggressive individuals experience the greatest change in T post-challenge. (A) Baseline T can be positively correlated with aggression among individuals and (B) post-challenge T can be negatively correlated with aggression among individuals if (C) within-individual hormonal reaction norms (dotted lines) co-vary with aggressiveness. Unique shapes represent unique individuals, each with a different level of aggression, represented by the intensity of shading. Bars in (C) represent average T levels across individuals at each time point.

We have not tested Prediction 4, whether brief T elevation within an individual's reactive scope leads to a temporary increase in aggressive behavior. However, we have shown that temporary GnRH-induced T elevations did not negatively impact parental care in male tree swallows (George et al. 2021), even though T treatment itself negatively impacts parental care in females (Rosvall 2013). This result suggests that once again, endogenous hormonal and behavioral responses may not mirror effects of exogenous hormone treatments on behavior.

Conclusions and next steps

Bidirectional relationships between T and territorial aggression are woven into the fabric of evolutionary behavioral endocrinology. There is good evidence that elevated T promotes aggression. In our qualitative review of the most common types of studies that measure T and aggression in relation to one another, we found mixed results. As we highlight above, changes in aggression across different breeding stages mirror similar temporal changes in T secretion, though it is clear from studies of territorial aggression in the non-breeding season that these two traits can vary quite independently over the course of weeks and months. We found mixed support for predictions about among-individual variation and within-individual variation at shorter, more immediate timescales. For instance, at baseline, individual differences in T production and aggression are not necessarily correlated. Endogenous T elevation with an individual's reactive scope does not always lead to enhanced aggression, and social challenges do not necessarily lead to further T elevation, even in systems where exogenous T enhances aggression. Our own research on the tree swallow echoes these patterns. Thus, it is clear that T and aggression may not co-vary at every timescale or every level of analysis, but T and aggression may still be mechanistically linked.

Our synthesis therefore underscores two main takeaways, which we see as priorities for future research. First, we reiterate past work showing that the mechanisms linking T and aggression vary across timescales, such that longer-term feedbacks do not necessarily occur at shorter timescales (Goymann et al. 2007; Goymann 2009). Second, the mechanisms linking T and aggression may vary among individuals, such that among-individual differences in either trait do not directly translate to within-individual changes in hormones or behavior.

In both cases, we believe that conflating levels of analysis or timescales introduces the potential for errors in causal inference. We view apparent inconsistencies as opportunities to embrace and empirically explore greater complexity in behavioral endocrinology. Substantial work has already highlighted the one such layer of complexity, related to functional variation in target tissue sensitivity to or metabolism of T (Hau 2007; Ketterson et al. 2009; Goymann et al. 2019; Rosvall et al. 2020; Schuppe et al. 2020). By broadening the scope of inquiry beyond circulating hormone levels, these analyses have already opened new doors into understanding how diverse endocrine molecular mechanisms shape and are shaped by behavior, and thus have the potential to influence behavioral divergence in nature. Moving forward, we advocate for greater attention to two additional layers of complexity, regarding (1) short- vs. long-timescales, and (2) within- vs. among-individual levels of analysis.

Relationships between T and aggression vary among timescales

The idea that hormonal mechanisms operating on different timescales may evolve independently from one another is not new. In particular, past meta-analyses in male songbirds have shown that seasonal variation in T production can be completely unrelated to short-term T responses to a social challenge (Goymann et al. 2007; Goymann 2009). Therefore, individual differences in T levels (Prediction 1) and seasonal patterns in T secretion (Prediction 2) are unlikely to stem from simple additive effects of variation in recent aggressive experiences and ensuing feedback on T production. Similarly, our research shows that T levels before a fight can predict aggressiveness during a fight, even if that aggressive interaction does not further elevate T. Other, shorter timescales (e.g., circadian changes) are also an exciting area of growth (Greives et al. 2021), though more work is needed to understand whether circadian patterns of T correspond to changes in behavior throughout the day (Elderbrock et al. 2021). We encourage greater attention to integrating these timescales to better understand what environmental factors do regulate variation in T at one timescale and not another, and why.

Relationships between T and aggression vary among individuals

It is clear that the observed relationship between T and aggression can differ based on whether we sample multiple individuals in a population at a single point in time vs. if we sample one individual repeatedly over time. It follows then that a null relationship between T and aggression among individuals does not preclude the existence of a positive relationship within an individual over time, and vice versa. Over the years, most research on the relationship between T and territorial aggression has focused on among-individual comparisons (e.g., control vs. STI groups); while this perspective has key logistical advantages, it does not allow us to evaluate the likely condition-dependent, rank-specific, or other individually variable connections between T and aggression. We know that T production varies with these other phenotypic qualities, and we know that high T can have some costs, but as of yet, we have insufficiently explored the next logical step in this line of thinking, that the specific effects of T or behavior may also be individually variable. As described above, we imagine that a specific degree of T elevation may be beneficial to some individuals and costly to others. Likewise, the adaptive value of a given behavioral response (e.g., responding with more or less aggression in the future) is likely to vary among individuals.

We recognize that testing these hypotheses will have methodological challenges due to the need for combined within- and among-subjects experimental designs. However, it is increasingly clear that the characterization of variation in reaction norms is essential in understanding the evolution of plastic traits (Nussey et al. 2007), including hormones (Taff and Vitousek 2016; Guindre-Parker 2020; Grindstaff et al. 2022). By coupling experimental results with longitudinal data on T, condition, current and future reproductive prospects, and other fitness-related traits, we can more deeply integrate the experimental approaches started by Bertold, alongside the more evolutionary approaches championed in recent behavioral endocrinology (Zera et al. 2007; Ketterson et al. 2009; Hau and Goymann 2015). In doing so, we will move toward a deeper understanding on the role of hormones in behavioral evolutionary trajectories.

Funding

This work was supported by the National Science Foundation [grant number IOS-1656109 to K.A.R.] and the Graduate Research Fellowship to E.M.G., and by the National Institutes of Health [grant number T32HD049336 to K.A.R. and E.M.G.].

Conflict of interest

The authors declare that there is no conflict of interest.