Animal temperament is present when there is significant among-individual variation in behavior (Réale et al., 2007; also known as personality, behavioral syndromes, and other terms, MacKay and Haskell, 2015) and is ubiquitous across taxa (Bell et al., 2009). In addition to behavior, endocrine profiles often vary substantially within populations (Kempenaers et al., 2008; Williams, 2008; Cockrem, 2013). Given the integrative role of the endocrine system in diverse physiological and behavioral activities, hormones could function as mediators of variation in temperament (Koolhaas et al., 1999), though the lower individual consistency of hormone levels relative to behavior suggests they are unlikely to drive behavioral variation (Holtmann et al., 2017).

Alternatively, endocrine systems and temperament could be coevolved, with correlational selection favoring particular temperaments in individuals with certain states (i.e., physiological characteristics; Wolf and Weissing, 2010). In particular, the glucocorticoid corticosterone (CORT) has been implicated in coping styles, or syndromes of integrated temperamental and neuroendocrine variation (Koolhaas et al., 1999). Corticosterone is produced predominantly in the adrenal or interrenal glands of vertebrates as part of the hypothalamic–pituitary–adrenal/ interrenal (HPA/I) axis, and levels of CORT exhibit individual repeatability (Taff et al., 2018). At the onset of a potential stressor (i.e., a threat to survival or homeostasis), CORT levels are generally increased from baseline levels, altering physiological activities to cope with current or future stressors, and CORT is therefore often generally referred to as a stress hormone (Sapolsky et al., 2000). One function of CORT is to mobilize energy reserves (i.e., glucose), along with a wide variety of other activities (Sapolsky et al., 2000; Vera et al., 2017). Corticosterone rises over a period of minutes to hours and has slow-acting effects via changes in gene transcription (Sapolsky et al., 2000) but can stimulate certain responses quickly by binding to membrane-bound receptors (Moore and Orchinik, 1994; Evan-son et al., 2010). Typically, individuals that exhibit proactive coping styles are characterized by readily attacking or escaping threats and having lower stress-induced CORT than conspecifics who are reactive (e.g., exhibiting an immobility response; Koolhaas et al., 1999).

Additionally, melanin-based coloration has been linked to both CORT physiology and temperament, with more melanistic individuals typically exhibiting proactive coping styles (Ducrest et al., 2008; Santostefano et al., 2019). The melanin-based coloration syndrome may be the result of pleiotropic effects of the melanocortin system, and in particular the peptide hormones derived from the proopiomelanocortin gene (POMC) and their associated receptors (Ducrest et al., 2008). The POMC protein, encoded by the POMC gene, is cleaved to produce adrenocorticotropic hormone (ACTH), an upstream hormone in the HPA/I axis that stimulates the release of CORT. However, ACTH can be further spliced into α-melanocyte stimulating hormone (α-MSH), while other splicing products of POMC include β-MSH and γ-MSH. As melanizing hormones, MSH darkens melanin-based coloration (Ducrest et al., 2008). As shared cleavage products of the POMC protein, ACTH and MSH are likely to covary in their levels depending on the expression of POMC (Roulin et al., 2011). Moreover, these hormones are all capable of binding with variable affinity to the several melanocortin receptors (MC1–5R) that regulate several biological activities related to coloration, stress physiology, and behavior (e.g., ACTH can stimulate melanocyte activity by binding to MC1R in the skin in addition to stimulating the HPA/I axis in the adrenal glands; Ducrest et al., 2008). Consequently, the melanocortin hypothesis proposes that the melanocortin system could drive the repeated covariation of these diverse traits related to hormones, coping, and temperament.

The melanocortin hypothesis has not, however, received consistent support, with some studies finding that melanin-based coloration was independent of behavior and/or stress physiology (e.g., Jenkins et al., 2013; San-Jose and Roulin, 2018; Santostefano et al., 2019). Moreover, though the link between CORT and temperament has generally become well-established in mammals and birds (Cockrem, 2007; Øverli et al., 2007; Carere et al., 2010; Hau and Goymann, 2015), evidence for this pattern is mixed (Westrick et al., 2019). For instance, the relationship between CORT and temperament can vary among age classes (Seltmann et al., 2012) and may be undetectable in wild populations under strong stabilizing selection (Archard et al., 2012 and citations therein). Furthermore, sympathetic nervous system reactivity may also be a key physiological correlate of temperament differences (Qu et al., 2018), and the combination of sympathetic and HPA/I axis reactivity could yield a more complex array of behavioral phenotypes (i.e., in addition to bold and shy temperaments, individuals may be docile or panicky, for example; Koolhaas et al., 2010).

A promising avenue for better evaluating support for the melanocortin hypothesis is to expand the taxonomic breadth of study. Nonavian reptiles (e.g., snakes, lizards, turtles, and crocodilians, which together are twice as speciose as mammals; Pincheira-Donoso et al., 2013) have been neglected in studies examining the relationships among stress physiology, temperament, and melanization as per the melanocortin hypothesis (but see Seddon and Hews, 2016), though some studies have evaluated correlations between pairs of these variables (e.g., Thaker et al., 2009; Mafli et al., 2011; Ibáñez et al., 2016; Mell et al., 2016; Currylow et al., 2017; Herr et al., 2017; Brashears et al., 2020). Furthermore, reptiles are ectotherms and therefore generally exhibit lower metabolic rates than endotherms (White et al., 2006) and lifestyles characterized by energy conservation (Sandmeier and Tracy, 2014). Consequently, energy mobilization in the stress response may carry different strategic value compared with the endotherms that have been the focus of stress physiology research (Herr et al., 2017). Indeed, reptiles with higher CORT levels typical of a reactive coping style have been found to exhibit more aggression toward humans (Mell et al., 2016; Herr et al., 2017; Brashears et al., 2020) and to flee sooner from predators (Thaker et al., 2009), behaviors that are more consistent with a proactive coping style (Koolhaas et al., 1999). A link between a reactive coping style and high CORT levels may occur because proactive behavioral strategies often require additional energy expenditure, and higher CORT levels may permit such behavior in otherwise energy-conserving animals. Additionally, as poikilotherms, reptile body temperatures are determined by a thermally variable environment. Reptile temperament (e.g., Citadini and Navas, 2013) and CORT physiology (e.g., Telemeco and Addis, 2014; Jessop et al., 2016; Moleón et al., 2018; Racic et al., 2020) can be temperature-dependent, and melanization can affect thermoregulation (Clusella Trullas et al., 2007). As a result, temperature may be independently associated with coping styles and melanization in reptiles and therefore should be accounted for when evaluating correlations between traits.

We evaluated the relationships among the CORT stress response, boldness (i.e., responsiveness to risk), and melanization of the shell in Eastern Box Turtles (Terrapene carolina, specifically the Woodland Box Turtle subspecies, T. c. carolina). We hypothesized that turtles with bolder temperaments would exhibit lower stress-induced CORT levels, based on coping style theory (Koolhaas et al., 1999), and that these bolder turtles would also have darker, more melanized shells (Mafli et al., 2011; Ibáñez et al., 2016), reflecting pleotropic effects of the melanocortin system. We also expected fluctuations in temperature (attributable to ambient conditions or performance of thermoregulatory behavior) to play a critical role: stress-induced CORT levels should be lower in cooler individuals because of limitations on physiological activity (Gienger and Urdiales, 2017), boldness should remain consistent or moderately increase with higher temperatures as observed previously (Kashon and Carlson, 2018; Pich et al., 2019; Carlson and Tetzlaff, 2020), and darker turtles should be warmer because of faster heating during basking behavior (Clusella Trullas et al., 2007).

Materials and Methods

Study Species.—Eastern Box Turtles (T. carolina) are predominantly terrestrial turtles found in much of the eastern United States, primarily occupying forested habitats (Dodd, 2001). Box turtles (Terrapene spp.) are characterized by a domed carapace (upper shell) and hinged plastron (lower shell), with the latter allowing the shell to seal completely around the head and limbs (Dodd, 2001). It is difficult for most predators to successfully attack an adult turtle enclosed in the shell, yet individuals seem to vary in the extent to which they emerge from the shell and employ more active defenses (e.g., fleeing, biting, urinating, etc.; Dodd, 2001; Pich et al., 2019). Individual Eastern Box Turtles exhibit very consistent differences in response to human handling and confinement stress, with some individuals' heads emerging consistently from their shells immediately or quickly after confinement (i.e., bold) and others remaining within their shell for a prolonged period (i.e., shy; Kashon and Carlson, 2018; Pich et al., 2019; Carlson and Tetzlaff, 2020). The head emergence time was found to be highly repeatable across several years, both in captive juveniles and in wild adults (Carlson and Tetzlaff, 2020). Moreover, this variation in boldness was not correlated with other phenotypic characteristics (e.g., sex or body size) or most testing conditions (e.g., the number of trials to which they have been subjected; Kashon and Carlson, 2018), though in some analyses there were modest reductions in head emergence time in warmer individuals (Pich et al., 2019; Carlson and Tetzlaff, 2020).

Eastern Box Turtles are also highly variable in shell color patterns (Dodd, 2001), most notably in the extent of black versus brown-yellow coloration on the carapace and plastron (Fig. 1). Ontogenetic change in coloration could be one source of color pattern variability in T. c. carolina. Ontogenetic color change occurs in some other turtle species (Tucker et al., 1995; Ennen et al., 2015), including a different subspecies of Eastern Box Turtle (T. c. triunguis; Leuck and Carpenter, 1981), the coloration of which is distinct from T. c. carolina. We know of no studies documenting the development of shell coloration across the lifespan in T. c. carolina, but a general trend is apparent: hatchlings possess almost uniformly dark brown carapaces that become lighter brown with little patterning as subadults, with mature animals being distinctly patterned with contrasting dark and light areas (Fig. 1). We therefore restricted our analysis to adult turtles (>115 mm plastron length; Williams and Parker, 1987). Dramatic differences in shell color still occur among mature adults (Dodd, 2001), and we note that in our study population, several individuals are known to have been marked 58 yr prior to this study and were already adults (‡10 yr old) at the time (Williams and Parker, 1987), and these now >68-yr-old individuals display the full range of variation in shell melanization.

Fig. 1.

Example carapace (upper row) and plastron (lower row) images of turtles demonstrating a range of melanization from low (left) to high (right). Upper row and lower row images are not paired and represent different individual turtles.

Corticosterone levels in Eastern Box Turtles have been previously reported to vary across years and seasons but not to differ between sexes (West and Klukowski, 2018). A previous study evaluated correlations between CORT and behavioral responses of Eastern Box Turtles to simulated attacks and found no relationship; however, individual animals received different types of simulated attack treatments with small sample sizes within each treatment, temperature and melanization were not incorporated into the study, and stressor duration may have been insufficient to elicit elevated CORT levels (Preston et al., 2020).

Study Sites.—We conducted our study at Wabash College's Allee Memorial Woods (AMW), a 72-ha forested preserve in Parke County, Indiana, USA, and at Weiler–Leopold and Black Rock Barrens Nature Preserves (WLBRB), two adjacent, predominately forested preserves comprising 113 ha in Warren County, Indiana, USA. Allee Memorial Woods is located approximately 56 km south of WLBRB and is the site of an Eastern Box Turtle population that was studied from 1958 to 1983, largely by capturing and marking turtles (Williams and Parker, 1987). Beginning in September 2014, we began collecting and marking turtles again at AMW. Subsets of these turtles at AMW were used for disease surveillance (having blood samples and mouth and cloacal swabs collected), tracked by radiotelemetry, and repeatedly subjected to behavioral assays over a period of 4 wk. Some of the turtles included in the present study were also used in one or both of these previous studies (1958–1983 and/or beginning in 2014), though there is little evidence of effects of repeated handling on head emergence time (Kashon and Carlson, 2018) and head emergence is consistent over years (Carlson and Tetzlaff, 2020). The turtles at WLBRB have not been studied but may have had occasional experience with human visitors to the nature preserve.

Field Data Collection.—Turtles were collected from 16 May to 15 June 2016 and between 14 May and 28 June 2018 by visual search. Once located, we immediately used an infrared thermometer to measure plastron temperature of an encountered turtle, and then we collected approximately 200 µL of blood from the subcarapacial sinus using a 22-gauge needle (the baseline sample). We recorded the time of initial encounter and the time blood draws were complete, which varied from 1 to 5 min (mean = 2.7 min). Blood samples were immediately stored in heparinized tubes on ice. We then placed the turtle in a zippered, opaque, black nylon bag (18 × 32 cm) in which a small amount of movement was possible. After 30 min, we removed the turtle and immediately drew another blood sample (∼100 µL; the stress-induced sample). This process was completed in 0.5–7 min, resulting in a total time of 32–41 min from first contact with the turtle to completion of the final blood draw (the stressor exposure time; mean = 34.7 min). One person then placed the turtle on the ground at or close to its original position and then walked away while another person quietly observed the turtle from approximately 10 m away. The observer recorded the time elapsed until the head of the turtle emerged (emergence latency), defined by the eyes surpassing the anterior margin of the carapace (modified after Ibáñez et al., 2014) and waiting up to a maximum of 10 min.

We recorded plastron length and sex of the turtle as covariates for analysis, though neither has been previously found to correlate with CORT levels or boldness (Kashon and Carlson, 2018; West and Klukowski, 2018). We used calipers to measure the length of the plastron as a metric of overall body size. We sexed the turtles primarily by the presence of a pronounced indentation on the plastron in males, as well as the position of the cloaca and length of the tail (Dodd, 2001).

We photographed the carapace and plastron of each turtle. For all photographs, we used a D7100 DSLR Camera with an AF-S DX Micro-NIKKOR 40mm f/2.8G lens (Nikon) with a macro ring flash (Sigma). We used identical settings for all photographs to minimize instrumental sources of variation (aperture = f/8, ISO = 200, RAW+JPEG file encoding, and 3 levels of exposure to ensure there was at least one well-exposed photograph per individual). Next to the turtle in each photo was an 18% gray reflectance standard to account for environmental sources of variation in color.

Within 7 h of collection, blood samples were centrifuged at 4,000 revolutions per minute for 2 min to separate plasma, which was then stored separately at -80°C until processing. We excluded from later analysis all turtles for which plasma samples were contaminated with lymph during collection or that experienced observable hemolysis during centrifugation, resulting in a final sample size of n = 68 turtles (29 females and 39 males; 59 from AMW and 9 from WLBRB).

Hormone Measurement.—We measured CORT levels in plasma using a high sensitivity enzyme immunoassay (EIA) kit for corticosterone (Immunodiagnostic Systems Limited, product number AC-15F1). The EIA kit that we used has been successfully used with lizards (Trompeter and Langkilde, 2011), snakes (Herr et al., 2017; Tylan et al., 2020), amphibians (Graham et al., 2012), and birds (Pike and Petrie, 2006). We determined recovery and linearity of dilution to validate this kit for Eastern Box Turtles. Serial dilutions of pooled plasma samples were linear and parallel to serial dilutions of kit standards. We spiked pooled plasma samples with known concentrations of CORT to evaluate recovery, which averaged 110.2%, indicating high accuracy with a small positively biased error.

Assays were performed using five kits. For the first two kits, plasma had been stored for 115–151 d prior to assays; for the third kit, samples had been stored for 361–397 d; for the fourth and fifth kits, samples had been stored for 39–84 d. CORT is generally stable for long-term storage while frozen (Reimers et al., 1982; Garde and Hansen, 2005; Herring and Gawlik, 2009), and there was no variation among CORT kits in estimated CORT levels despite the large differences in storage time (likelihood ratio test of censored regression models [see below]: χ²₄ = 5.63, P = 0.23). We diluted 15 µL of plasma in 35 µL of assay buffer and followed kit instructions. Given the specified lowest sensitivity of the kit (0.17 ng/mL) and the dilution used, we were unable to detect CORT levels in plasma below 0.57 ng/ mL. Intraassay coefficients of variation ranged from 2.7 to 6.6%, and the mean interassay coefficient of variation was 15.6%.

Color Pattern Analysis.—Our approach to characterizing the extent of melanization was to determine the proportion of shell area covered by dark (black to dark brown) versus light (yellow to light brown) pigmentation. We adopted this pattern-based, binary approach (dark versus light areas) rather than quantifying the value or lightness of the shell because of uncontrollable sources of variation in value: the curvature of the shell produces uneven reflectance of light, making certain areas appear brighter, and soil encrusted between annuli on the scutes altered perceived coloration. We used the program ImageJ (Schneider et al., 2012) for all analyses. For carapace photographs, we cropped images to only include the second vertebral scute, because it most reliably was oriented approximately parallel to the camera (whereas patterning on much of the dome-shaped carapace would be distorted as a result of perspective), and because relative differences in melanization among turtles appeared to be consistent across all carapacial scutes. We used the Smooth tool to blur out small bright spots (specular highlights) associated with surface irregularities. We measured the area of the entire scute (in pixels) and then used the Color Threshold tool to select only the dark regions of the scute. Threshold values were adjusted until there was a good match between the selected area and the actual, visually assessed patterning of the shell. The area of the selected region was measured and divided by the total area to determine the percent of the scute that was melanized (dark).

Color patterns are relatively indistinct on the plastron, with darker regions usually grading into lighter regions. To describe the proportion of the plastron that exhibited dark coloration, we visually scored the percent of the plastron that appeared dark in the photographs, treating moderately darkened areas as intermediate in melanization. Only the region of the plastron posterior to the hinge was considered because the anterior portion was frequently bent away from the plane of the camera when turtles were enclosed in the shell. To validate this subjective assessment of melanization, we used the Multispectral Imaging plugin (Troscianko and Stevens, 2015) to normalize images to the 18% gray standard included in the photographs and then measured the luminance of the posterior plastron. Our image analysis approach was objective but was limited in reliability by the presence of brighter areas associated with the curvature of the plastron. Luminance values correlated well with our visual assessments (r = 0.84), and we therefore used the latter because they describe melanization as percent of area, similar to how we analyzed the carapace. Importantly, all aspects of measuring melanization were performed blind to associated morphometric, behavioral, or hormonal data and should therefore be free of unintentional bias.

Statistical Analysis.—We first tested whether there was evidence that confinement stress increased CORT levels, demonstrating that postconfinement CORT levels reflected a stressed state. Because 92.6% and 42.6% of pre- and postconfinement CORT levels, respectively, were below the detection limit (left-censored), we used censored regression to analyze these data. We used all data without imputing a value (such as zero, or the detection limit) in place of censored values, and the analysis incorporates information on whether or not data points were above or below the detection limit as well as known values for data points that were not censored (Helsel, 2005; Chambers et al., 2011). We used the censReg package in R (Henningsen, 2017) to test for the effect of blood sampling period (pre- versus postconfinement) on log-transformed CORT values, including individual identity as a factor to allow a paired comparison.

For the remainder of the analysis, our objective was to evaluate correlations among behavior, stress physiology, and melanization, both with and without accounting for the effects of covariates. We used ln(x+1)-transformed head-emergence latency as the measure of boldness, ln(x)-transformed postconfinement CORT levels as the measure of stress physiology, and the percent of darkly colored area on both the carapace and plastron as measures of melanization. We chose to restrict the analysis to postconfinement CORT levels because only 7.4% of preconfinement CORT levels were above the detection limit, and thus there was little information available about baseline levels of CORT. We used both carapace and plastron melanization in the analysis because they separately correlate with different behaviors in tortoises (Mafli et al., 2011). To draw inferences about correlations among these four traits (boldness, CORT, and carapace and plastron melanization), we used the traits as response variables in a multivariate linear model fit with Bayesian Monte Carlo Markov Chain (MCMC) in the MCMCglmm package (Hadfield, 2010) in Program R. Our approach and this package allowed us to draw inferences about the correlations among traits (using the posterior distribution of the variance–covariance matrix of the traits), to examine the residual correlations among traits after correcting for the influence of covariates, and to use the censored Gaussian distribution for boldness (which was right-censored at a maximum observation period of 600 sec) and CORT level (which was left-censored at the detection limit; see above).

We first fit intercept-only MCMCglmm models with the four traits as responses, an unstructured residual variance–covariance matrix (permitting all covariances to be estimated), and Gaussian distributions (for carapace and plastron color) or censored Gaussian distributions (for boldness and CORT). We then fit models that also included covariates that could influence or associate with trait values: plastron length as a proxy for body size and age; sex; source population (AMW or WLBRB); plastron temperature; Julian date to account for potential seasonal effects on boldness and CORT levels; year to estimate interannual differences in boldness and CORT levels; time of day to account for diel variation in boldness and CORT levels; and total duration of stressor exposure to account for how differences in handling, confinement, and blood draw times could influence boldness and CORT levels. The MCMCglmm package uses Bayesian MCMC methods and thus requires the use of prior probabilities. For covariate effects, we used the default priors. For the variance–covariance matrices, we evaluated models with four priors to limit how much our results depended on the selected priors: 1) very low variances and belief parameter m = 5 (“V = diag(4)*1e-6, nu = 5”), the flat improper prior for correlations (J. Hadfield, pers. comm.); 2) very low variances and m = 1, the flat prior for variances and covariances (J. Hadfield, pers. comm.); 3) variance = 1 and m = 3.002, a weakly informative prior (Dingemanse and Dochtermann, 2013); and 4) the calculated raw variance–covariances among the four traits as the prior variance, with m = 3.002 (Brommer et al., 2014). All priors yielded qualitatively similar conclusions, and thus we restrict the presentation of our results to prior (1) above.

After 5 × 10⁶ iterations, we excluded the first 1 × 10⁶ as burn-in and generated the posterior distributions from 1,000 values (thinning to every 4,000th) from the MCMC chain. This yielded good convergence as assessed from plots of posterior distributions, traces, and Geweke's diagnostic, with low levels of autocorrelation (r < 0.10). We report results as 95% Bayesian credibility intervals (CIs, estimated using the highest posterior density interval) and MCMC P-values (pMCMC, using the package MCMC.OTU; Green et al., 2014), and we use α = 0.05 to determine statistical significance. The data sets and code used during this study are available in the Figshare repository ( https://doi.org/10.6084/m9.figshare.19103282).

Results

Postconfinement CORT levels were significantly higher than baseline CORT levels in Eastern Box Turtles (censored regression: t₇₀ = 13.88, P < 0.0001). Therefore, our confinement and handling techniques successfully induced a stress response in the subjects (Fig. 2).

Fig. 2.

Plasma corticosterone (CORT) levels from before and after confinement stress. Values from the same individual are linked by the lines. Values below 0.57 ng/mL (indicated by the horizontal dotted line) were below the detection limited and treated as censored in the analysis. The y-axis is presented on a logarithmic scale.

There was little evidence for correlations among boldness, stress-induced CORT levels, and shell melanization that are predicted by the melanocortin hypothesis, regardless of whether trait correlations were adjusted for covariates (Table 1; Fig. 3). Carapace and plastron melanization were significantly though not strongly correlated (Fig. 3), indicating a tendency for the melanin content of both parts of the shell to vary in similar directions but with substantial independence. However, neither boldness nor stress-induced CORT levels were associated with the melanization of either part of the shell (Fig. 3), indicating that darker colored turtles did not predictably differ in behavior or stress response. When not adjusting for covariates, there was a trend toward turtles with darker carapaces having low stress-induced CORT levels (P_MCMC = 0.11; Table 1; Fig. 3). Additionally, there was a potential relationship between boldness and stress-induced CORT levels such that bolder (fast-emerging) turtles tended to have higher levels of CORT (Fig. 3), but again with 95% CIs that marginally included 0 (raw trait correlation: P_MCMC = 0.08, covariate-corrected trait correlation: P_MCMC = 0.12), indicating only moderate support for this relationship (Table 1).

Table 1.

Correlations among behavioral, corticosterone (CORT), and shell melanization traits as estimated from posterior distribution of the variance–covariance matrix in multivariate Generalized Linear Mixed Models. Values shown are correlations (95% credibility interval), and Bayesian Monte Carlo Markov Chain P-values, with significant relationships in bold text. Values above the diagonal are uncorrected for covariates, while values below the diagonal account for covariates (extrinsic factors associated with sampling conditions or intrinsic differences between individuals; see Table 2).

Fig. 3.

Relationships among behavioral, corticosterone (CORT), and shell melanization traits. Stress-induced plasma CORT and boldness (head emergence latency) are shown on a logarithmic scale. Censored values of CORT levels (<0.57 ng/mL) and boldness (head emergence latencies >600 sec) are indicated by triangular symbols.

In the model that included covariates, neither plastron length nor sex had an effect on any of the four traits (Table 2). Effects of date, year, and time of day were only assessed for boldness and stress-induced CORT levels (for which they could have biologically meaningful effects), and they did not influence either trait (Table 2). Populations differed in boldness and plastron melanization, with WLBRB turtles taking significantly longer to emerge from the shell and having significantly lighter colored plastrons (Fig. 4). Plastron temperature and total stressor exposure time were both only associated with stress-induced CORT levels: turtles that were warmer or had been handled longer had higher stress-induced CORT levels (Fig. 5).

Table 2.

Covariate effects on behavioral, corticosterone (CORT), and shell melanization traits estimated from multivariate Generalized Linear Mixed Models. Results are shown as the estimated effect β (95% credibility interval) and Bayesian Monte Carlo Markov Chain P-values, with significant effects in bold text. Unestimated effects are blank.

Fig. 4.

Differences in boldness (A) and plastron melanization (B) between the two populations, Allee Memorial Woods (AMW) and Weiler–Leopold/Black Rock Barrens (WLBRB) nature preserves. Boldness (head emergence latency) is depicted on a logarithmic scale, and each circle represents a value for an individual turtle. Boxes represent the 25th to 75th percentiles, thick lines within the boxes represent the median, and whiskers (vertical lines) extend to all observed values within a range of 1.5 times the interquartile range from the boxes; points beyond 1.5 times the interquartile range are shown unconnected by whiskers in (B).

Fig. 5.

Effects of plastron temperature (A) and total stress exposure time (B) on stress-induced plasma corticosterone (CORT) levels. Total stress exposure time is the elapsed time from initial contact to completion of the final (stress-induced) blood sampling, including confinement, handling, and bleed times. CORT levels are shown on a logarithmic scale. Censored values of CORT levels (<0.57 ng/mL) appear below the horizontal line representing the detection limit.

Discussion

Our results generally failed to support the melanocortin hypothesis in Eastern Box Turtles. We found no significant correlations among behavioral, CORT, and shell melanization traits, and correlations that approached significance were weak. Moreover, the near significant relationship between stress-induced CORT levels and boldness (P = 0.08 uncorrected for covariates, P = 0.12 corrected for covariates) is in the opposite direction from what we would have expected, with higher stress-induced CORT levels tending to occur in the bolder, more proactive turtles who emerged faster from the shell; in previous studies, higher CORT levels were generally associated with more reactive and shy individuals (Koolhaas et al., 1999; Atwell et al., 2012; Baugh et al., 2012; Mazza et al., 2019). We found a significant difference between populations in plastron melanization and boldness and a clear effect of plastron temperature on stress-induced CORT levels. Curiously, we also found that plastron melanization and carapace melanization were significantly correlated but apparently the correlation was too weak to demonstrate differences in the degree of carapace melanization between populations.

When uncorrected for covariates, we found no significant effect of melanization on stress-induced CORT levels although the value of our P-value was small (P = 0.11). Therefore, melanized turtles did not tend to have lower CORT levels. The POMC gene is expressed as a peptide chain (POMC) that is spliced into 5 melanocortin hormones, one of which (ACTH) stimulates the release of CORT from the adrenal–interrenal glands, affecting numerous aspects of behavior and physiology (Sapolsky et al., 2000; Ducrest et al., 2008). The ACTH peptide chain can be further modified to produce α-MSH, which darkens coloration by increasing melanin in the integument (Ducrest et al., 2008). Therefore, CORT and melanin are linked because of their reliance on a single hormone (ACTH; Ducrest et al., 2008). Splicing ACTH to produce α-MSH disables its stimulatory effect on CORT release and, as such, higher melanization should yield lower CORT release. For instance, lizard populations with higher circulating α-MSH levels had lower baseline CORT levels (Seddon and Hews, 2020). However, greater expression of the POMC gene could produce both higher ACTH and α-MSH levels, resulting in a positive relationship between CORT and melanin levels (Roulin et al., 2011). In addition, ACTH is also capable of directly increasing melanization by binding to the MC1R receptor, which would mean elevated ACTH could both increase melanization and CORT release. As such, higher CORT could be correlated positively or negatively with melanization depending on the mechanism behind melanization in a specific organism (Ducrest et al., 2008). In Eastern Box Turtles, this complexity may have obscured the relationship between CORT and melanization.

Moreover, the mechanisms underlying differences in shell coloration among individual turtles are unknown and may be unrelated to POMC expression. For example, melanization in one species of lizard was uncorrelated with α-MSH levels (Seddon and Hews, 2020). The MC1R gene encodes several melanocortin receptors to which POMC products bind, and thus MC1R may play a role in melanization in some species (Hosoda et al., 2005; Mundy, 2005), and agouti-signaling protein acts as an antagonist to α-MSH and other melanocortin hormones and promotes production of phaeomelanin (responsible for yellow and brown coloration) over eumelanin (responsible for blacker color; Ollmann et al., 1998). Production of predominately phaeomelanin when compared with eumelanin could explain the lack of relationship we see between coloration and behavior, warranting further research on the mechanisms underlying turtle color patterns. Further research on the development of coloration within Eastern Box Turtles would also be valuable for understanding drivers of variation in carapace coloration. There is a well-known but poorly documented pattern of ontogenetic shell color change, and the development of carapace coloration is subject to some environmental influence: a complete absence of sunlight on carapacial scutes prevents darkening of the scutes in growing T. c. carolina (Belzer and Seibert, 2010), and juveniles of many turtle species develop darker carapaces when reared on dark substrates, though this tendency was nonsignificant in T. c. carolina specifically (Rowe et al., 2014). Our observations point toward substantial coloration differences even among mature adults from the same habitat, but in-depth experimental and longitudinal studies of carapace coloration and comparisons of related individuals could yield important insights into ontogenetic, extrinsic, and heritable drivers of shell coloration in Eastern Box Turtles.

Our nearly significant findings in which bolder turtles tended to have higher stress-induced CORT is the reverse of what has been suggested as a general pattern (Koolhaas et al., 1999) in which bolder individuals are expected to have lower levels of CORT, whereas in these turtles, bolder individuals tended toward higher levels of CORT. Similar relationships have been observed in other reptiles, where bolder or more aggressive animals have higher, rather than lower, levels of CORT (Mell et al., 2016; Herr et al., 2017; Brashears et al., 2020; but see Thaker et al., 2009; Claunch et al., 2017), and in other animals the links between boldness and CORT may be variable among species (Tablado et al., 2021). In Eastern Box Turtles, the nature of the defensive strategies used by the putatively bold individuals may be important to explaining their tendency toward higher stress-induced CORT levels. Specifically, fleeing is likely more energetically demanding than remaining enclosed within the shell, and therefore high glucose mobilization following CORT elevation may be advantageous in bolder individuals. Moreover, neuroendocrine activity in the sympathetic nervous system may play a complicating or predominant role in Eastern Box Turtle behavior under stress. In particular, norepinephrine and epinephrine levels represent another physiological variable that interacts with CORT, and this physiological system may be most important in fight–flight–freeze responses to immediate threats (Wingfield, 2005). Indeed, in rats these hormone levels can positively correlate with at least some types of proactive behaviors (which typically would correlate with boldness; Ducrest et al., 2008). Direct evaluation of sympathetic nervous activity may better illuminate the physiological correlates of differences in boldness in Eastern Box Turtles.

The general lack of correlations predicted by the melanocortin hypothesis may also reflect a statistical limitation for our data set. Though boldness in Eastern Box Turtles is highly repeatable (r = 0.69–0.77; Kashon and Carlson, 2018; Pich et al., 2019; Carlson and Tetzlaff, 2020), repeatability of CORT levels is not known but is often moderate in other species (approximately r = 0.20–0.40, depending on conditions; Schoenemann and Bonier, 2018; Taff et al., 2018). Even with substantial repeatability, any given measurement of behavior or hormone levels will deviate from an individual animal's mean value. Such deviations mean that the use of single values for each animal (rather than repeated measurements) could statistically obscure the underlying correlation between traits; moreover, such variance in the data can conflate correlations that occur between individuals' mean values (e.g., between an animal's long-term boldness and its average CORT level) with correlations that occur between observation-level deviations within individuals (e.g., between individual measurements of boldness and CORT levels; Garamszegi and Herczeg, 2012; Fürtbauer et al., 2015; Niemelä and Dingemanse, 2018). Further exploration of the melanocortin hypothesis in turtles would be strengthened by repeated measurements of behavior and CORT levels within individuals.

We found that temperature had a strong positive effect on CORT levels. Many studies in reptiles, amphibians, and birds have found similar effects (Narayan et al., 2012; Telemeco and Addis, 2014; Lynn and Kern, 2017; Brischoux et al., 2018; Racic et al., 2020), though low temperature stress can also be associated with higher CORT (Dupoué et al., 2013; Telemeco and Addis, 2014). Furthermore, the scale can matter when examining stress responses because temperature may play a role in stress physiology in birds but at a climate-scale, in which long-term average conditions shape stress physiology (Jessop et al., 2016). Increased CORT in warmer animals may reflect stress responses to thermal challenge or, especially in ectotherms, enhanced physiological activity and increased metabolic rates. Notably, higher temperatures can slightly increase boldness in Eastern Box Turtles (Pich et al., 2019) and bolder turtles tend to maintain higher body temperatures (Kashon and Carlson, 2018), and therefore higher temperatures could cause a relationship between boldness and stress-induced CORT levels. Alternatively, in an observational study such as this, existing differences in CORT can affect body temperature by influencing an organism's thermoregulation. Manipulation of CORT levels can alter thermoregulatory behavior in reptiles: in one lizard species, thermoregulatory behavior increases body temperatures in CORT-elevated individuals (Preest and Cree, 2008), whereas in another lizard species the elevation of CORT either increased or decreased time spent basking, depending on whether CORT was applied to prenatal or postnatal lizards (Belliure and Clobert, 2004; Belliure et al., 2004). We know of no research on the effects of CORT levels on thermoregulation in turtles.

Our populations differed in plastron coloration and in boldness, with turtles from the AMW population being bolder and their plastrons less melanized than turtles from the WLBRB population (Table 2). The AMW turtles presumably had more contact history with humans (Williams and Parker, 1987) but this is unlikely to explain their increased boldness because previous research has found that repeated behavioral trials did not affect behavior (Kashon and Carlson, 2018). The two populations may differ in predator abundance or assemblage because exposure to predators has been found to influence boldness via plasticity (Hellström and Magnhagen, 2011) and differential mortality (Lapiedra et al., 2018). Alternatively, other behaviors or phenotypic traits that are correlated with boldness (Koolhaas et al., 1999; Sih et al., 2004) may be targets of selection. With respect to plastron melanization, substrate and habitat differences can influence carapace coloration of some turtles (Rowe et al., 2006, 2014; McGaugh, 2008) and could be driving differences in the two populations in this study, but such influences on plastron coloration have not been documented in other turtles (Reinke et al., 2018).

In conclusion, we find little support for the melanocortin hypothesis in Eastern Box Turtles, a species with highly consistent individual differences in boldness and with often dramatic variation in melanization. However, we found a clear effect of temperature on stress-induced CORT levels and population differences in melanization and boldness, pointing to the added complexity that comes with expanding this research question to more diverse taxa. Further work in this and similar systems promises to enhance our understanding of the melanocortin system and its generality.