Assessing long-term stress levels is of key importance in wildlife ecology and management, as physiological stress is one of the major links between populations dynamics and anthropogenic perturbations (Chown and Gaston 2008). Given limited funds available (Wilson et al. 2006), research demands a toolkit of methods suitable to effectively and efficiently monitor stress levels within populations (Ceballos et al. 2015), possibly adopting non-invasive sampling methods (Zemanova 2020).

Stress can be measured using glucocorticoids (GCs) and, more precisely cortisol (CORT) in most mammals, as biomarkers (Sapolsky et al. 2000). In response to a stressor, CORT is secreted trough activation of the hypothalamic–pituitary–adrenal (HPA) axis. Once released, CORT stimulates the mobilization of energy through glucogenesis while downregulating other life-sustaining functions that do not immediately contribute to survival, such as immunocompetence, growth or reproduction (Möstl and Palme 2002). This allows individuals to adequately cope with the stressor (Sapolsky et al. 2000). CORT can be traced in well-established matrices such as blood, saliva and feces (Sheriff et al. 2011). These, however, reflect hormone levels over short time periods (e.g., from a few minutes or hours—blood and saliva—to a few days—feces), and their application is restricted to the assessment of acute stress. Although, in principle, they may allow researchers to investigate chronic stress, this requires repeated sampling (Meyer and Novak 2012), which may be challenging due to time and field constraints. CORT, however, can also be traced in other samples, such as hair or feathers (Sheriff et al. 2011).

Hair integrates baseline levels and short-term peaks of CORT over longer periods, up to several months (Salaberger et al. 2016). Thus, it provides the option to assess long-term stress levels easily and effectively within a single sample. Moreover, hair can be obtained easily and through minimally invasive procedures, which reduces sampling bias (Sheriff et al. 2011). Accordingly, HCCs are increasingly used in research, and recent studies have highlighted the role of HCC as valuable indicators of chronic stress and ecological patterns (e.g., Koren et al. 2002; Mastromonaco et al. 2014; Carlitz et al. 2014). Despite these advantages, an ecological informative interpretation of HCC levels is challenging.

HCC is influenced by multiple factors of different origin—internal, external, physiological, and environmental. Hair cortisol levels were found to vary with sex (Lafferty et al. 2015), age (Laudenslager et al. 2012; del Rosario et al. 2011), body condition (Macbeth et al. 2012), personality (Sauveroche et al. 2020), pregnancy (Fairbanks et al. 2011), season of the year (Roth et al. 2016), or even features of the hair itself such as hair color (Yamanashi et al. 2013). However, their effects on HCC are inconsistent across species and may be caused by different mechanisms (Heimbürge et al. 2019). Sex, for example, affects HCC in American black bears Ursus americanus (Lafferty et al. 2015), polar bears Ursus maritimus (Bechshøft et al. 2011), chacma baboons Papio hamadryas ursinus (Bergman et al. 2005) or humans Homo sapiens (Raven and Taylor 1996), but not Asiatic black bears Ursus thibetanus (Malcolm et al. 2013), chimpanzees Pan troglodytes (Yamanashi et al. 2013) or Canada lynx Lynx canadensis (Terwissen et al. 2013). In chacma baboons, higher HCCs in males were related to social stress, caused by the comparably high social instability of male hierarchies (Bergman et al. 2005). Likewise, in humans, men show generally higher HCCs, but simply for physiological reasons, namely a lower activity of glucocorticoid-metabolizing enzymes in men (Raven and Taylor 1996). Hence, if not considered, such diversely influencing factors may confound results when using HCC as biomarker of chronic stress with regard to a specific research question or when comparing individuals or populations (Heimbürge et al. 2019).

Here, we explore different ecological and physiological determinants of HCC in the Alpine marmot Marmota marmota, a large, ground-dwelling and highly social rodent widespread in the mountainous areas of central and southern Europe. Marmots are subjected to a wide array of abiotic and biotic stressors, mainly including harsh environmental conditions and natural predation, but also challenges of group living and increasing levels of anthropogenic disturbance, for example caused by tourism and hunting (Arnold 1990; Hackländer et al. 2003; Oberosler et al. 2017). Because the Alpine marmot is protected in large parts of its distribution area (Bern convention, Annex III: protected fauna species), there is an interest in establishing tools, such as the use of HCC, to monitor the status of populations.

We studied marmots in the Lombardy sector of the Stelvio National Park, Central Italian Alps (46.53867 N, 10.40938 E). The area covers about 37.6 hectares with altitude ranging from 2341 to 2671 m a.s.l. (cf. Figure 1 in Corlatti et al. 2020). The climate is alpine, with mean daily temperature ranging from − 12 °C in winter to 23 °C in summer. Vegetation within the study site is mainly composed of alpine and boreal meadows of Carex curvula and Nardus stricta. CORT is incorporated into the hair during hair growth at the follicles via passive diffusion from the bloodstream; thus, the time frame which the HCC represents depends upon hair growth species-specific characteristics (Meyer and Novak 2012). Even though we do not have knowledge of these characteristics in Alpine marmot, HCCs measured at the time of sampling should reflect stress experienced during the molting period, i.e., a few weeks around June, right after the reproductive season (Barash 1989).

Hair samples were collected during marmot captures, conducted in 2019, from 6 to 16 June, when marmots emerged from the burrows after hibernation. A total of 20 Tomahawk’s traps were located within the study area and kept open each day from 08:00 to 20:00 h. Dandelion flowers Taraxacum officinalis were used as bait to maximize capture probability. Each captured animal was marked with a Tracer Bayer transponder PIT and with a different combination of colored ear-tags, which made it possible to recognize each individual. The manipulation of the animal was handled by a veterinarian and did not exceed 30 min; animals were not sedated (cf. Corlatti et al. 2020). During manipulation, before the injection of the PIT, a sample of hair was torn out from the interscapular area. Because hair does not sit tight in marmots, a small sample can easily be pulled out by hand without causing major signs of distress in the animals. Samples were conserved in a uniquely identified plastic bag at room temperature in a dry and dark place until analysis. Biometric information on each captured animal, i.e., length (cm) and body mass (kg), was collected, alongside sex, age class (juvenile or adult), and rectal temperature (°C)—the latter used as an index of body temperature. Body length (BL) and mass (BM) were used to calculate a body condition index, the Scale Mass Index (SMI; Peig and Green 2009), in a two-step process. First, we performed a standardized major axis regression on log-transformed value pairs of length and mass. Second, we used the slope of the best fit line (bSMA) to calculate a scaled index for each individual as SMIi = BMi [BL0/BLi]^bSMA, where BL0 denotes the arithmetic mean of all measured BLs, and BLi − BMi, respectively, represent the individual measurements of BL and BM.

Before hair cortisol extraction (Acker et al. 2018), the hair was cut (segment length < 0.5 cm) and weighed in glass vials. Samples were washed (750 µl /sample) with 100% methanol and they were vortexed for 10 s to remove surface contaminants. After removing the wash liquid, each vial was left open for 5 min in a fume hood to guarantee the complete evaporation of methanol. Cortisol was extracted using 2 ml 100% methanol for each sample for 24 h on an orbital shaker. At the end of the extraction period, samples were centrifuged at 3500 rpm for 10 min. Hair was discarded, and methanol was pipetted into a new vial and dried in a fume hood under constant air flow for 48 h. Dried extracts were reconstituted by adding 150 μl of phosphate buffer, vortexed for 10 s, and kept frozen until assay. Cortisol concentrations were determined by Cortisol DetectX® (Ann Arbor, Michigan 48108–3284 USA). The intra- and inter-assay coefficients of variation were 4.6 and 8.2, respectively. Despite unvalidated for hair extract, the assay is based on monoclonal antibody, results should reflect real hair cortisol concentration without influent cross-reactivity and several studies have used this assay with hair extract in humans (Ling et al. 2019, 2020).

To investigate the relationship between hair cortisol (response variable) and the explanatory variables, we used a linear modelling approach. Hair cortisol concentration is a strictly positive variable, but its conditional distribution is not normal. We thus built a ‘beyond optimal’ generalized linear model with Gamma distribution and a log link. The model included the following explanatory variables: sex, SMI, age class and body temperature. Prior to model fitting, multicollinearity was tested with the ‘vif’ function (Variance Inflation Factor) in the ‘car’ package (Fox and Weisberg 2019). Model selection was conducted using data dredging with the function ‘dredge’ in the MuMin package (Barton 2020). This method allows testing multiple hypotheses by searching for all possible variable combinations and returns models ranked based on their AICc (Akaike Information Criterion corrected for small samples) value. If more models had delta AICc < 2, they were considered competitive; models that were more complex versions of the top-ranked one were excluded from the candidate set (Burnham and Anderson 2004). To check if body temperature was confounded by outside temperature (Clarke et al. 2010) at the timepoint of capture or mainly explained by body mass (Clarke and Rothery 2008), we did a simple linear regression of body temperature (response variable) against outside temperature and body mass (explanatory variables), respectively, for a subset of the data. The analyses were conducted using R v. 4.0.4 (R Core Team 2020) in RStudio v. 1.3.1056 (RStudio Team 2020). The model was inspected for quantile residual distribution using ‘DHARMa’ (Hartig 2020), using a QQ plot (observed vs. expected values) and a plot of residual vs. predicted values.

During summer of 2019, a total of 33 hair samples were collected from 33 individuals and used for the analyses. Outside temperature during capture (used to check for confounding effects on body temperature) was only available for a subset, 23 out of 33 individuals.

The residual diagnostics of the global model did not show major violation of assumptions, and none of the variables had VIF values > 3; therefore, multicollinearity was considered inconsequential (cf. Zuur et al. 2010). Only two models had delta AICc < 2. In particular, the first model included sex and body temperature (AICc = 346.5), while the second one included sex, the body condition index and body temperature (AICc = 347.4). Since the second-best model was a more complex version of the top-ranked model, only the latter was selected for inference (Burnham and Anderson 2004). Parameter estimates of the selected model showed a significant difference between males and females, with lower cortisol concentration in males than in females (Table 1; Fig. 1B). Furthermore, hair cortisol concentration and body temperature were positively related (Table 1; Fig. 1A). Body temperature was not related to outside temperature or body mass (Outside temperature: β = − 0.18; 95% CI = − 0.58, 0.21; Body mass: β = 0.11; 95% CI = − 0.33, 0.54). Figure 1 suggests the presence of a potentially influential datapoint, which could affect the estimate of the regression coefficients and their confidence intervals. A robust generalized linear model with the same conditional distribution and link function was thus fitted using the ‘robustbase’ package (Maechler et al. 2020). The robust model supported the absence of major differences in parameters estimates from the previous model (sex [male]: β = − 0.393; 95% CI = − 0.774, − 0.012; Body temperature: β = 0.226; 95% CI = 0.030, 0.423). Additionally, we refitted the same model by removing the outlier and we obtained the following parameters estimate: sex [male]: β  = − 0.346; 95% CI = − 0.687, − 0.010; Body temperature: β = 0.168; 95% CI = − 0.029, 0.353.

Table 1 Model selected to explain variation in hair cortisol concentration in a marmot population within the Stelvio National Park in 2019
Full size table
Fig. 1
figure 1

Marginal effects of the model selected to explain variation in hair cortisol concentration in a marmot population within the Stelvio National Park in 2019. A Hair cortisol concentration as a function of sex (red for females and blue for males). The bars represent 95% confidence interval. B Hair cortisol concentration as function of body temperature. The grey shaded area represents 95% confidence interval

Full size image

To better understand the influence of the various factors influencing HCC, a novel and easily measured biomarker of long-term stress, we tested the effect of sex, age, physical condition and body temperature on the HCC levels in alpine marmots. Sex and body temperature were major determinants of HCC, while age class and physical condition were not influential. To our knowledge, this is the first investigation of HCC and its determinants in the Alpine marmot.

The generally higher levels of HCC in females than in males provide some additional information to the somewhat inconsistent picture of sex-specific effects on HCC in various taxa (Kudielka and Kirschbaum 2005). Different mechanisms might be responsible for the observed difference. First, the sex effect may be caused by physiological differences between males and females, as sex-specific gonadal steroid hormones may affect HPA-axis activity (McCormick and Mathews 2007; Laudenslager et al. 2012). Alternatively, higher HCCs in females might be related to generally poorer body conditions in females than in males, possibly associated with nutritional stress (Heimbürge et al. 2019). However, this latter explanation seems unlikely in our study, as sex and body condition were not correlated in our study animals, nor was body condition related to HCC. On the other hand, higher HCC levels in females may be truly stress-related and reflect higher levels of chronic stress in females than in males. Social factors can be a major source of stress that, depending upon the species-specific social system, may affect sexes differently (Creel et al. 2013). In Alpine marmots, for example, female reproduction suppression was identified as a major source of short-term stress, mediated through aggressive behavior directed from dominant to subordinate members in a social group, as indicated by a more than twofold rise of plasma cortisol in females during the reproductive period (Hackländer et al. 2003). To clarify the mechanisms responsible of sex-specific differences, further research on HCCs in Alpine marmots should include observations of social behavior and possibly complements the measurement of long-term stress with measures of short-term stress based on feces cortisol metabolites, blood or saliva.

More surprising than the sex difference was the observed positive relationship between HCC and the single measurement of body temperature during one point in time, that is at the capture of subjects. To our knowledge, no studies have yet investigated the relationship between body temperature and chronic stress based on cortisol levels in hair or other matrices, neither in Alpine marmots nor other species. In contrast, numerous studies have reported an increase in body temperature in response to acute stress, often looking at stress induced by handling in a variety of species, including rodents (Briese and Cabanac 1991; Long et al. 1990) and other mammals (Moe and Bakken 1997), as well as birds (Carere and Oers 2004) and reptiles (Cabanac and Bernieri 2000). This reaction has been termed ‘emotional fever’, because it is usually accompanied by an increase in heart rate (Cabanac and Aizawa 2000; Cabanac and Guillemette 2001). However, how long-term stress levels proxied by the HCC may be related to the strength of ‘emotional fever’ response during handling still remains unclear. It is well known that the behavioral and neurophysiological components of individual stress response are not fully flexible, but rather constrained into alternative response patterns, known as coping style (Koolhaas et al. 1999, 2010) or personality (Coppens et al. 2010). The presence of coping styles was confirmed in a variety of species, including Alpine marmots (Ferrari et al. 2013). Thus, the observed higher body temperatures in some animals, if indeed an ‘emotional fever’ response towards the stressful experience of handling, might be part of an individual marmot’s coping style. Because the general and stress-related responsiveness of the hypothalamus–pituitary–adrenal axis is part of the individual coping style (Koolhaas et al. 2010), chronic stress as indicated by the HCC could be related likewise. For example, a study on great tits Parus major indicates a relationship between increase in body temperature due to handling stress and personality differences (Carere and Oers 2004) and, more recently, basal and stress-induced plasma cortisol levels were related to coping styles in Alpine marmots (Costantini et al. 2012). Yet, given the low sample size and narrow range of body temperature, this supposition remains to be tested with a larger dataset and by taking repeated measures of body temperature (shortly before and after handling) in combination with an explicit investigation of coping styles. It is worth noting that the relationship between hair cortisol and body temperature was influenced by the occurrence of an unusual value: as a matter of fact, the removal of the outlier reduced the effect size of the predictor. However, we argue that removing this datapoint is not justifiable on biological ground: the removal of a datapoint is legitimate when the datapoint is a result of an incorrect observation (Faraway 2016). In most of the cases, using a robust linear approach is preferred as it downweighs the effects of larger errors (Venables and Ripley 2002).

Despite the preliminary nature of our investigation, due to limited sample size and lack of cross-validation of HCC with other measure of stress, our findings underpin the complexity of the neuro-endocrinological modulation on the HCC and narrow the knowledge gap regarding its determinants in a protected wildlife species. Many questions remain open, with respect to the mechanisms underlying variation in HCC. These could be tackled by better investigating cortisol incorporation into the hair, for example with regard to species-specific hair growth characteristics, to clarify the retrospective timeframe represented in the samples. Furthermore, multilevel studies combining hormonal and behavioral correlates of stress would be necessary to unravel the effects of different determinants on HCC.