Among-species variation in hormone concentrations is associated with urban tolerance in birds

Abstract

As cities expand across the globe, understanding factors that underlie variation in urban tolerance is vital for predicting changes in patterns of biodiversity. Endocrine traits, like circulating hormone concentrations and regulation of endocrine responses, might contribute to variation in species' ability to cope with urban challenges. For example, variation in glucocorticoid and androgen concentrations has been linked to life-history and behavioural traits that are associated with urban tolerance. However, we lack an understanding of the degree to which evolved differences in endocrine traits predict variation in urban tolerance across species. We analysed 1391 estimates of circulating baseline corticosterone, stress-induced corticosterone, and testosterone concentrations paired with citizen-science-derived urban occurrence scores in a broad comparative analysis of endocrine phenotypes across 71 bird species that differ in their occurrence in urban habitats. Our results reveal context-dependent links between baseline corticosterone and urban tolerance, as well as testosterone and urban tolerance. Stress-induced corticosterone was not related to urban tolerance. These findings suggest that some endocrine phenotypes contribute to a species’ tolerance of urban habitats, but also indicate that other aspects of the endocrine phenotype, such as the ability to appropriately attenuate responses to urban challenges, might be important for success in cities.

1. Introduction

The global environment is currently undergoing a dramatic shift. Conversion of natural environments to urban landscapes is increasing worldwide [1] and 55% of the world's human population is now living in urban areas [2]. One consequence of this rapid urban expansion is a decrease in biodiversity, with a marked decline in species richness as habitat becomes urbanized, and an associated homogenization of biota across urban areas [3]. However, despite the apparent challenges posed by urban living, some species survive and even thrive in cities, successfully living and breeding in these novel habitats. As such, a central question arises: why do some species persist in cities, while most others do not?

Previous attempts to address this question have revealed some interesting patterns: urban-tolerant species sometimes have larger relative brain sizes [4,5], exhibit broader environmental tolerance [6] and behavioural flexibility [7], lead more generalist lifestyles [8], and are sometimes competitively dominant to their non-urban counterparts [9]. This previous work has largely focused on ecological and behavioural traits, and the potential links between species-level variation in physiological traits and urban success remain largely unexplored.

Urban environments present unique challenges to wildlife in the form of novel sights and sounds, increased human presence, and altered resources and exposure to pollutants, competitors, predators, and pathogens [10–13]. For a species to persist in urban habitats, it must be able to effectively cope with these challenges. Several endocrine axes are key components of adaptive responses to environmental challenges, through their integration of external cues to coordinate physiological, behavioural, and life-history responses, often with important fitness consequences [14–16]. As such, endocrine phenotypes (e.g. circulating hormone concentrations and sensitivity of endocrine responses) could be an important factor in determining a species' ability to cope with urban challenges and persist in urban habitats, but, to our knowledge, no previous research has directly assessed associations between among-species variation in endocrine phenotypes and urban tolerance (but see [17]). To date, several studies have compared urban and non-urban populations within species, sometimes finding differences in endocrine traits such as baseline and stress-induced glucocorticoid levels, as well as androgen levels, suggesting that urban environments might favour or induce specific endocrine phenotypes [18]. However, comprehensive reviews and analyses of population-level variation reveal a lack of consistent changes in endocrine phenotypes in response to urban challenges, illustrating the complexity of individual- and population-level responses to the urban environment, many of which are probably driven by plasticity in endocrine traits [17–22]. A comparison among species could reveal the degree to which evolved variation in endocrine phenotypes contributes to success in urban environments.

Two endocrine axes that are most likely to be associated with responses to urban habitat are the hypothalamic–pituitary–adrenal (HPA) and hypothalamic–pituitary–gonadal (HPG) axes. The HPA axis is highly conserved among vertebrates and plays a central role in mediating physiological and behavioural responses to environmental challenges, primarily through the action of glucocorticoid hormones [23,24]. While baseline and stress-induced levels of glucocorticoids can vary greatly both among and within individuals [25], species-level differences in these traits are repeatable, suggesting that species have unique, evolved glucocorticoid phenotypes [26]. Baseline levels of glucocorticoids are thought to be reflective and/or anticipatory of the routine metabolic demands facing individuals in a given life-history and environmental context [26,27]. Under acute challenges, glucocorticoid levels increase rapidly, shifting resource allocation and behaviour toward functions that promote immediate survival and recovery from the challenge [28]. While the HPA response is undoubtedly adaptive in most natural contexts, chronically or inappropriately elevated glucocorticoids could prevent some individuals or species from persisting in challenging environments. In cities, where novel stimuli are abundant, but often not representative of true threats to survival, species with lower levels of baseline and stress-induced glucocorticoids might be better able to avoid fitness costs associated with inappropriate or excessive HPA responses, allowing them to better tolerate urban challenges and minimizing the costs of elevated glucocorticoids [29]. Additionally, lower stress-induced glucocorticoids have been associated with reduced neophobia and increased boldness in some species [30,31], traits that might be important at the species level for maximizing ability to colonize urban habitats and exploit novel resources [32]. Alternatively, maintaining a robust, sensitive HPA axis could be beneficial in environments where disturbances and challenges are frequent. In this case, higher concentrations of glucocorticoids and/or a more pliable HPA axis, that can be both readily activated and shut down, could be advantageous in cities [29,33,34]. Among species, several comparative studies have revealed associations between glucocorticoid phenotype and factors associated with urban success, including brain size [35] and life-history strategy [24,26,36], suggesting an as yet uncharacterized relationship between glucocorticoid levels and species’ urban tolerance. A recent comparative analysis of baseline and stress-induced corticosterone across several species of birds and reptiles did not find a consistent association between degree of urbanization of the sampling location and endocrine phenotypes [17], which is consistent with previous studies that fail to find consistent plastic responses to urban challenges. This study also included a categorical species-level estimate of urban tolerance for birds, which was not strongly associated with corticosterone [17].

The HPG axis is also linked to behaviours and aspects of physiology that might be associated with species' success in urban environments. Androgen hormones, including testosterone, regulate reproductive function in male vertebrates and can promote aggression and territorial behaviour in animals of both sexes. Higher testosterone is also sometimes associated with increased aggression and more exaggerated signals of dominance among species [37,38]. Species occurrence and success in cities can be limited by competitive interactions [9,39,40], and therefore dominance signals and aggressive behaviour are likely to be relevant to persistence in urban habitats in some species. Although biodiversity is generally lower in urban areas, the species that thrive there tend to occur at higher densities than in non-urban environments; thus, intraspecific competition for resources may be heightened [11,12,41]. If true, then species with higher testosterone might be better able to defend territories and resources in the face of increased competition [37,42], giving them an edge in urban habitats. High testosterone has also been linked to increased boldness and exploratory behaviour among individuals [43,44], though whether this relationship exists at the species level has not been determined. Thus, higher levels of testosterone might be associated with urban tolerance. Alternatively, lower levels of testosterone might be associated with urban tolerance if selective pressures associated with altered pathogens and disease transmission dynamics in cities [13,45–47] increase the costs of immune suppression that can be caused by elevated testosterone [48–50]. Finally, as with glucocorticoids, mean circulating concentrations of testosterone might be less important for urban tolerance than androgen responsiveness to social challenges [38,51]. In this case, we might not expect a strong relationship between circulating testosterone concentrations and urban tolerance.

Here, we tested whether among-species variation in endocrine phenotypes differed among bird species that vary in their propensity to breed in cities, using a comparative analysis of the association between hormone concentrations and species-level estimates of urban tolerance in birds from across the world. Because urban environments present wildlife with numerous challenges, and endocrine phenotypes influence how species cope with challenges, we predicted that hormone concentrations would differ among urban-tolerant and urban-avoidant species, independent of the environment where they were sampled. In other words, our analyses characterize relationships between species-level variation in endocrine phenotypes and urban tolerance, rather than within-species responses to urban challenges. This study provides, to our knowledge, one of the first species-level analyses of the relationship between endocrine phenotypes and success in urban environments, offering novel insight into how evolved differences in glucocorticoid and androgen concentrations might relate to a species’ ability to cope with ongoing urbanization.

2. Methods

(a) . Urban tolerance

To classify urban tolerance of each of our focal species, we used data collected as part of a previous study that estimated species-level urban occurrence using community science methods and contributions from ornithologists and expert birders in large cities around the globe [9]. These and similar community science data have been reliably used in several published studies [6,9,52] and offer an accurate means of assessing species urban occurrence across a larger array of species and cities than would be possible with alternate methods, providing a robust dataset for our analysis. The original study focused on species for which particular ecological data were available, and so the list is not exhaustive, but provides a broad sample of birds in large cities around the world, with breeding occurrence for each species ranging from 0 (absent from cities) to 3 (widespread breeder in urban habitat) (see [9] for details). We calculated an average breeding occurrence for each species across any cities where data were available, which we used to estimate species-level variation in urban tolerance. This measure does not reflect the environment where endocrine samples were collected, as all of the species in the dataset occur in non-urban environments, and most of the endocrine data were collected outside of cities.

(b) . Endocrine phenotype data

We compiled data on avian baseline corticosterone (the primary glucocorticoid hormone in birds), stress-induced corticosterone and baseline testosterone concentrations from an open-access database, HormoneBase [53], restricting the data to species for which we had urban occurrence estimates, as described above. We use the term baseline to indicate that the sample was collected within a timespan when the authors of the original study expected circulating concentrations were unlikely to be affected by capture and handling, or the method used to lure birds to capture (e.g. song playback). HormoneBase includes most published reports of these three commonly measured hormone concentrations (in plasma) up to the year 2015, making it ideal for use in broad-scale comparisons of endocrine traits across species [53]. We revisited all original publications reported in HormoneBase to ensure data quality and gather additional information on study sites. To supplement these data, we also performed a search of the literature, incorporating published endocrine data through to the end of 2020 for as many species as possible to maximize the scope of our analyses. We searched using Google Scholar and Web of Science, with combinations of key search terms, including the common and scientific names of species for which we had urban tolerance estimates, along with cort*, glucocort*, testos*, or androg*.

We included all studies that estimated one or more of the three endocrine traits in free-ranging birds, reporting data separately for males and females, with an indication of broad breeding status (breeding or non-breeding). In several cases, data were not reported this way (e.g. data pooled across the sexes), and so we contacted study authors who often provided the data we needed for our analyses. We excluded studies of captive birds or individuals exposed to experimental treatments that might have influenced their hormone concentrations. We recorded latitude of study sites as reported in original publications or in HormoneBase. Where available, we also compiled data on latency to collect blood samples (i.e. time from initial capture until collection of the blood sample, usually reported as maximum latency for baseline corticosterone and testosterone and mean latency for stress-induced corticosterone). We also reviewed descriptions of study sites in the original publications and coded them as urban or non-urban, to allow assessment of the potential for population-level responses to urban habitat (e.g. plasticity or rapid evolution) to influence our results. In total, we compiled 1391 estimates of hormone concentrations (571 baseline corticosterone, 291 stress-induced corticosterone, and 529 testosterone) and urban occurrence scores for a total of 71 species. The vast majority of studies in our dataset sampled birds from non-urban habitats, with only 47 estimates of baseline corticosterone, 35 estimates of stress-induced corticosterone, and 25 estimates of testosterone derived from birds living in habitats that could be described as urban.

(c) . Statistical analysis

To test the hypothesis that endocrine phenotypes differ among species that vary in their tolerance of urban habitats, we conducted three main analyses, followed by supplementary analyses that controlled for phylogeny and the potential effects of sampling. All analyses were conducted in R (v. 4.2.1). We analysed each endocrine trait (baseline corticosterone, stress-induced corticosterone, and baseline testosterone) as response variables in separate linear mixed-effects models (using the package nlme) with the same fixed and random effects included in initial global models. We used hormone concentrations as the response variable, rather than urban tolerance, because hormone concentrations are dynamic and depend on many other factors (e.g. sex and life-history stage) that would be irrelevant in a model predicting urban tolerance, which is a static, species-level estimate. We also included multiple measures of hormones that were not independent (e.g. multiple measures for a species); these were best modelled using random factors, with hormone concentration as the response. Any associations between the species-level variation in urban tolerance and hormone concentrations, either in main effects or in interaction with the other fixed effects, might reflect the degree to which evolved differences among species in urban tolerance are associated with species-level variation in endocrine phenotypes.

Before analysis, we natural log-transformed baseline corticosterone (+1), stress-induced corticosterone and testosterone to improve model fit. We included the absolute latitude of the study site (scaled to a mean of 0 and s.d. of 1), sex (male or female), breeding status (breeding or non-breeding) and the species' mean urban occurrence (our estimate of urban tolerance, scaled to a mean of 0 and s.d. of 1) as fixed effects, along with all two- and three-way interactions involving urban occurrence in a global model (see the electronic supplementary material and R code for full details). Scaling permits comparison of the magnitude of effect sizes among the two continuous fixed effects. We did not include any interaction terms that did not incorporate urban occurrence, to avoid overfitting. We included species and study identity as random effects in all models. We did not include other random effects (e.g. laboratory identity and assay method) to avoid overfitting and because these other factors might introduce variation in our data, but not bias with respect to our main hypothesis that urban tolerance would be associated with endocrine phenotypes. Both the baseline corticosterone and testosterone datasets showed uneven residual variance with sex of the sampled individuals. To address this violation of model assumptions, we compared the initial global models with models that also incorporated modelling of the variance structure with sex. We assessed the effect of these variants by comparing Akaike information criterion corrected for small sample size (AICc) between models and used the global model with the lowest AICc for subsequent model selection. Inclusion of a modelled variance structure across sex improved model fit for analysis of both hormones.

We evaluated the evidence that urban occurrence predicts variation in endocrine phenotypes by comparing the AICc of all recombinant versions of the global model, including a null model with only the random effects. We report results from the best-performing model (lowest AICc) in the main text and provide model selection results in the electronic supplementary material, tables S1–S6. In cases where the relationship between urban occurrence and hormone concentrations depended on a categorical factor in the model (i.e. a significant interaction effect), we followed up the main analysis by estimating the slopes of the urban occurrence–hormone relationship within each level of the interacting factor using the emtrends function in the R package emmeans. Such post hoc tests of slopes are not possible for interactions involving more than one continuous fixed effect (e.g. urban occurrence and sampling latitude), but we plot results to illustrate these relationships, using representative low and high values for sampling latitude, and used emtrends to estimate the slope of the urban occurrence–hormone relationship at those values of latitude. We assessed fit of both the global and best-performing models by evaluating distributions of normalized model residuals, relationships between model residuals and fixed effects, model plots and Cook's distance. All models fitted well and met model assumptions, with no influential outliers.

We followed the main analyses with several supplementary analyses that controlled for phylogenetic relationships among focal species (electronic supplementary material, figures S1–S3), included latency to collect blood samples as a fixed effect, and excluded data derived from urban populations, allowing us to assess the potential effects of phylogenetic non-independence among focal species, variation in latency across studies, and plastic or rapid evolutionary responses to urban habitat on our results. We conducted an additional supplementary analysis of stress-induced corticosterone using a similar approach as described above, but with baseline corticosterone (natural log transformed+1, and scaled to a mean of 0 and s.d. of 1) included in the global model as a fixed effect. Finally, we also ran an additional supplementary analysis of testosterone concentrations in males that included a more refined estimate of life-history substage. These were unavailable for many samples, beyond the coarse breeding/non-breeding categorization included in our main analyses, but we expected variation with substage would be most important for analysis of testosterone in males, as circulating concentrations often change dramatically between early and late breeding [54]. As such, we extracted information on substages of breeding where available, which allowed categorization of samples as collected during early or late breeding stages, or in non-breeding birds. See the electronic supplementary material for detailed methods and results of all supplementary analyses. We did not repeat model selection for these supplementary analyses that had smaller sample sizes, but instead used the top-ranked model identified in the main analyses of the full dataset, with the supplementary factor added (e.g. latency), or using the subsetted dataset (i.e. urban samples excluded), or with breeding substage included in the model in place of the coarser categorization (breeding/non-breeding and excluding females). In the case of stress-induced corticosterone, we ran supplemental analyses on the top-ranked model that retained the factor of interest (urban occurrence). For the other two hormones, the top-ranked models all retained this factor. We did not include phylogeny in our main analyses because this approach might be overly conservative and could obscure biologically meaningful results [55]. We did not include latency, breeding substage, or baseline corticosterone in the main analyses because these were not available for all estimates of hormone concentrations, and so our sample sizes are reduced in these supplementary analyses. We did not explicitly assess habitat type (urban or non-urban) as a predictor in our main analyses because a large majority of the samples were collected outside of cities, and because plastic responses to the environment were not the focus of our study.

3. Results

In all results below, linear mixed-effect model effect estimates are reported with 95% confidence intervals. For all endocrine phenotypes, we focus our results on relationships with urban occurrence, including interactions, as pertaining to our research question (see the electronic supplementary material, tables S1–S6 for details of model selection and results).

(a) . Baseline corticosterone

In our main analyses of baseline corticosterone, the best-performing model retained the fixed effects of urban tolerance and breeding status, and their interaction (electronic supplementary material, table S4). Three additional models were ranked within 2AICc of the best-performing model, and urban tolerance was retained in three of the four top-ranked models (electronic supplementary material, table S1). The relationship between a species' urban tolerance and baseline corticosterone depended on breeding status (figure 1; urban occurrence by breeding status: β = −0.15 [−0.29, −0.01], p = 0.031; slopes of urban tolerance–corticosterone relationships: non-breeding birds β = −0.18 [−0.38, 0.01]; breeding birds β = −0.03 [−0.21, 0.15]). Non-breeding birds with high urban tolerance tended to have lower baseline corticosterone than birds with low-urban tolerance, whereas urban tolerance was unrelated to baseline corticosterone in breeding birds (figure 1). Supplementary analyses controlling for phylogeny, including latency as a fixed effect, and excluding data derived from urban populations produced qualitatively similar results to our main analyses, suggesting that these findings were not influenced by confounding factors (electronic supplementary material, figure S4).

Figure 1.

Variation in baseline corticosterone (y-axis) with urban tolerance (urban score, x-axis) across 49 bird species. The model-predicted relationship between a species' urban occurrence score (scaled to a mean of 0 and s.d. of 1, with high values indicating species that are widespread breeders in large cities) and 571 estimates of baseline plasma corticosterone concentrations depended somewhat on the broad breeding status (breeding birds: solid lines, non-breeding birds: dashed lines) of birds at the time of sampling (urban score by breeding status interaction: β = −0.15 [−0.29, −0.01], p = 0.031). Lines depict model-predicted relationships, with shaded areas showing 95% confidence limits. See the electronic supplementary material, table S4 for full model results.

(b) . Stress-induced corticosterone

The best-performing model explaining variation in stress-induced corticosterone concentrations in our main analyses retained only the main effects of breeding status and sex. Two additional models were ranked within 2AICc of the best-performing model, and urban tolerance was included in the second-ranked model, along with an interaction of sex with urban tolerance (electronic supplementary material, table S2). Results of this second-ranked model did not provide evidence of a relationship between urban tolerance and stress-induced corticosterone (electronic supplementary material, table S5 and figure S7; urban occurrence by sex: β = −0.09 [−0.17, 0.00], p = 0.054; slopes of urban tolerance–corticosterone relationships: males β = −0.06 [−0.21, 0.09]; females β = 0.03 [−0.13, 0.18]). Supplementary analyses including baseline corticosterone as a fixed effect in the top model that retained urban occurrence revealed a positive association between baseline and stress-induced corticosterone (β = 0.21 [0.17, 0.25], p < 0.001), but produced results that were otherwise similar to our main analyses (electronic supplementary material, figure S5). Additional supplementary analyses controlling for phylogeny, including latency as a fixed effect, and excluding data derived from urban populations produced qualitatively similar results to our main analyses, suggesting that these findings were not influenced by confounding factors (electronic supplementary material, figure S5).

(c) . Testosterone

The best-performing model explaining variation in testosterone concentrations in our main analyses retained the fixed effects of urban tolerance, breeding status, sex, latitude, and all two-way interactions with urban tolerance, as well as one three-way interaction of breeding status, latitude, and urban tolerance (electronic supplementary material, table S6). Two additional models were ranked within 2AICc of the best-performing model, and urban tolerance was retained in all three of the top-ranked models (electronic supplementary material, table S3). The relationship between urban tolerance and testosterone depended on sex, sampling latitude, and breeding status (figure 2; urban occurrence by breeding status by latitude: β = −0.31 [−0.47, −0.15], p < 0.001; urban occurrence by sex: β = 0.25 [0.07, 0.43], p = 0.005; slopes of urban tolerance–corticosterone relationships: non-breeding birds at low latitudes (22.6 degrees) β = 0.85 [0.26, 1.45]; breeding birds at low latitudes β = −0.04 [−0.61, 0.52]; non-breeding birds at high latitudes (57.7 degrees) β = −0.48 [−1.01, 0.05]; breeding birds at high latitudes β = −0.17 [−0.68, 0.33]). In breeding birds, testosterone concentrations were unrelated to urban tolerance in males, and weakly negatively related to urban tolerance in females. In birds that were not breeding, males and females with high urban tolerance had higher testosterone at low latitudes, and lower testosterone at higher latitudes, relative to birds with low-urban tolerance. At low latitudes, urban-tolerant birds appeared to maintain relatively high testosterone concentrations, independent of their breeding status, whereas urban-avoidant birds had relatively lower testosterone when they were not breeding (figure 2). Supplementary analyses produced similar results to our main analyses, indicating that confounding factors did not bias our findings (electronic supplementary material, figures S6 and S8).

Figure 2.

Variation in testosterone (y-axis) with urban tolerance (urban score, x-axis) across 47 bird species. The model-predicted relationship between species’ urban occurrence score (scaled to a mean of 0 and s.d. of 1, with high values indicating species that are widespread breeders in large cities) and 529 estimates of plasma testosterone concentrations depended on the sex (females: solid lines, males: dotted lines,) and broad breeding status (breeding: (a,b), non-breeding: (c,d)) of the sampled individuals, as well as the sampling latitude (interaction of urban occurrence by breeding status by latitude: β = 0.31 [−0.47, −0.15], p < 0.001; interaction of urban occurrence by sex: β = 0.25 [0.07, 0.43], p = 0.005). Absolute latitude of the sampling location was included in the model as a continuous variable, but is illustrated here using model predictions for 22.6 degrees latitude (low latitude) and for 57.7 degrees latitude (high latitude), which represent the 3% and 95% latitudes in the dataset. Lines depict model-predicted relationships, with shaded areas showing 95% confidence limits. See the electronic supplementary material, table S6 for full model results.

4. Discussion

Here, we characterized the relationships between species-level variation in urban tolerance and 1391 estimates of plasma concentrations of baseline corticosterone, stress-induced corticosterone, and testosterone from 186 studies in 71 species of birds, providing a global analysis of endocrine phenotypes across urban-tolerant and urban-avoidant species. We found evidence that urban tolerance is associated with endocrine phenotypes in birds, but that this association varies across ecological and life-history contexts, highlighting a complex relationship between evolved physiological traits and urban success. Baseline corticosterone was lower in urban-tolerant relative to urban-avoidant birds when they were sampled during non-breeding life-history stages (figure 1), while stress-induced corticosterone was unrelated to urban tolerance (electronic supplementary material, figure S7). Testosterone was not strongly related to urban tolerance in breeding birds, but was positively related to urban tolerance at lower latitudes and negatively related to urban tolerance at higher latitudes during non-breeding stages (figure 2). These results were robust to several supplementary analyses incorporating the phylogenetic relationship among focal species, controlling for sampling latency, excluding data derived from urban populations, controlling for baseline concentrations, and incorporating more refined breeding substages (electronic supplementary material, figures S4–S8).

Baseline corticosterone is highly plastic and generally less heritable and less repeatable within individuals than stress-induced corticosterone [56–59], with concentrations changing over short time spans in response to fluctuating metabolic demands. As such, we did not expect a strong relationship between species-level variation in this endocrine phenotype and urban tolerance. Despite this, we found evidence of a negative relationship between urban tolerance and baseline corticosterone in birds that were not breeding (figure 1). The lack of a relationship in breeding birds suggests that other selective pressures contribute to species-level variation in corticosterone during breeding, independent of a species' tolerance of urban habitats. Urban-avoidant birds had similar baseline corticosterone concentrations across breeding and non-breeding stages, whereas urban-tolerant birds had lower baseline corticosterone when they were not breeding. The ability to reduce baseline corticosterone outside of breeding stages might be associated with reduced costs of living in challenging environments, including cities. The mechanism linking low-baseline corticosterone outside of breeding and success in urban environments warrants further study.

Stress-induced corticosterone is an important component of responses to novel and challenging environments, and is generally more repeatable among individuals than baseline corticosterone [58,59]. Thus, we expected to see a more consistent relationship between this endocrine phenotype and urban tolerance. Contrary to this expectation, we did not find evidence of a link between stress-induced corticosterone and urban tolerance (electronic supplementary material, figure S7). The lack of a relationship between stress-induced corticosterone and urban tolerance could suggest that the measures we used here do not represent an endocrine phenotype that is relevant to species’ responses to urban habitat. The studies included in our dataset measured stress-induced corticosterone using a standard capture-stress protocol [60,61], which quantifies maximal circulating glucocorticoid concentrations in birds exposed to the acute stress of handling and restraint. While this measure can provide an indication of the reactivity of the HPA axis in response to a perceived threat, it does not capture the ability of the organism to alter its physiological reaction in response to different types of stimuli, a trait which might be more pertinent to urban success. In cities, where harmless disturbances such as loud noises and human presence co-occur with real dangers such as predators, traffic and competitors [62], it could be variation in the ability to separate true threats from benign stimuli and attenuate responses accordingly, rather than differences in maximal concentrations of glucocorticoids, that determines which species thrive. This idea is supported by literature on cognition in urban birds, which suggests that some successful urban species exhibit greater behavioural flexibility [7] and a stronger propensity for learning [32], sometimes reflected in larger relative brain sizes [4,5,63]. These traits could confer an increased ability to discern between different levels of threats, and to habituate to common urban stimuli like human presence, thereby minimizing inappropriate HPA axis responses. Additionally, aspects of the sensitivity of the HPA axis response, including the speed of mounting or shutting down a reaction, could be more important to tolerance of environmental challenges than maximal circulating concentrations [29,33,34]. However, relatively few studies have characterized the ability of wildlife to adjust their endocrine stress responses to match different levels of threatening stimuli (but see [64,65]) or rates of activating and shutting down responses (but see [33,34]), and none have yet linked differences in these abilities across species to variation in urban tolerance. Thus, it is possible that a relationship between evolved differences in HPA axis responses to stress and variation in urban tolerance across species exists but was not detected here.

Testosterone can have different functions in male and female vertebrates, and therefore testosterone levels across the sexes might sometimes be under conflicting selective pressures [66–68]. Our results suggest that species with lower testosterone concentrations in breeding females are somewhat more successful in urban environments, while male testosterone during breeding was unrelated to urban tolerance (figure 2). In male vertebrates, testosterone is intrinsically linked to reproductive function, with high levels of circulating testosterone needed to facilitate behaviours and physiological changes associated with breeding [69]. As such, male testosterone across species is probably constrained to optimal levels that support reproduction and might be uncoupled from variation in urban tolerance. In females, testosterone appears to play less of a role in reproduction, but can act to modulate behaviours including territoriality and aggression, with increased testosterone causing increased aggression in females of some species [70–72]. Lower testosterone in females could be beneficial for urban living as it might support tolerance of close contact with conspecifics, something that is likely a common challenge in urban habitats, where successful species tend to occur at high densities [73]. Recent work supports this idea, suggesting that greater tolerance in interactions with conspecifics is characteristic of some urban populations [74] and thus could be favoured in urban environments, though these behavioural changes have yet to be linked to circulating androgen levels.

In birds that were not breeding, we found less sex dependence and more importance of sampling latitude in the relationship between testosterone and urban tolerance. Males and females from urban-tolerant species had higher testosterone than urban-avoidant species at low latitudes, but we saw the opposite pattern at higher latitudes, with higher testosterone associated with urban avoidance (figure 2). This result could suggest that aggression or signals of dominance are important for success in low-latitude cities, whereas the benefits of lower testosterone, perhaps in relation to immune function or reduced aggression, are important in high-latitude cities. Species living at low latitudes often experience more stable climates, exhibit year-round territoriality, and have slower life histories than higher latitude species [75–77]. Cities at low latitudes can also differ from higher latitude cities [78,79], in ways that might select for different endocrine adaptations. The existing endocrine literature does not include robust sampling at low latitudes, with our testosterone dataset only including studies of six species in the tropics. As such, we are cautious in interpreting the latitudinal variation in our results and await further study to clarify these findings.

As with corticosterone, other aspects of androgen phenotypes might be more important to urban tolerance than circulating concentrations. For example, the ability to appropriately modulate testosterone in response to social stimuli could be an important component of tolerance of urban habitats, where population densities can be high [73]. Several studies have characterized modulation of testosterone in response to experimental exposure to simulated social challenges (e.g. simulated territorial intrusions) [80], but insufficient data are available across species of varying degrees of urban tolerance to assess how they are related. Comparative studies among species of social modulation of testosterone and urban tolerance are needed to test this hypothesis.

By taking a broad, among-species comparative approach, we sought to assess the role of evolved differences in endocrine traits, rather than more commonly studied plastic shifts, in shaping species responses to urbanization. Studies like this one can provide valuable insights into the degree to which pre-adaptation to urban environments contributes to species distributions, potentially helping us to predict changes in patterns of biodiversity and better target conservation initiatives as urbanization expands. Our results support previous literature suggesting that evolved differences among species can be linked to their ability to persist and thrive in urban habitat [4–9]. This work, coupled with the inconsistency of results found in previous within-species comparisons [17–21], suggests that circulating hormone concentrations might not be the aspect of endocrine phenotype that is most relevant to urban tolerance. Instead, variation in the ability to appropriately modulate endocrine responses to match different urban stimuli, and thereby avoid maladaptive reactions, might be a better predictor of which species thrive in cities, and which do not.

Data accessibility

Data and R codes used in this manuscript can be accessed online at https://osf.io/p7jgz/. See the folder ‘Datasets’ for raw data and supporting materials, and folder ‘R code’ for annotated R scripts used for analyses.

The supplemental methods, table and figures are provided in the electronic supplementary material [81].

Authors' contributions

E.C.C.S.: conceptualization, data curation, formal analysis, visualization and writing—original draft; P.R.M.: conceptualization, formal analysis, writing—review and editing; F.B.: conceptualization, formal analysis, funding acquisition, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by funding from the Government of Ontario (Ontario Graduate Scholarship to E.C.C.S. and an Early Researcher Award to F.B.), and the Natural Sciences and Engineering Research Council of Canada (Discovery Grant no. 05883–2014 to F.B.).