Evaluation of hair cortisol as an indicator of long-term stress responses in dogs in an animal shelter and after subsequent adoption

Shelter dogs are exposed to a variety of stressors. Among non-invasive techniques, hair cortisol concentration (HCC) is suggested an easy to collect biomarker for giving insight into long-term stress responses. We evaluated HCC as an indicator of long-term cortisol responses in dogs in an animal shelter over different chronological time points during sheltering and after adoption. Hair samples were collected from the neck region following a shave/re-shave protocol of shelter dogs (total n = 52) at four different time periods: T1 intake at shelter (pre-shelter period, n = 51); T2 after 6 weeks in the shelter (n = 23); T3 6 weeks after adoption (n = 24); T4 6 months after adoption (n = 22). HCC at T2 was significantly higher than HCC at T1, T3 and T4 (effect of sample collection moment: F3,41 = 12.78, p < 0.0001). The dog’s weight class, age class, sex, reason for admission, kennel history and melanin type also explained HCC variability. No significant difference in HCC was found between shelter dogs T1 and control pet dogs in their own homes (n = 20, one sample, t = − 1.24, p = 0.219). A significant but moderate positive correlation between HCC and urinary cortisol:creatinine ratios was found (т = 0.3, p < 0.001). As HCC increased in the shelter, the use of this non-invasive parameter appears a useful additional tool in dog welfare research.

HCC as a measure of long-term HPA-axis reactivity in shelter dogs. This study aimed to further validate HCC to measure chronic stress in dogs. For research in shelters, HCC can be an interesting measure with multiple practical applications, e.g., evaluating retrospective HCC levels at intake 6 and subsequent HCC monitoring with shave/re-shave methods 10,29 during the shelter period to provide more specific insight in longer-term stress levels over the different chronological phases of sheltering and after adoption into a new home. Therefore, in this study we aimed more specifically to (1) compare HCC in shelter dogs at the moments of relinquishment (pre-shelter), during the shelter period (in-shelter) and after adoption (post-shelter) and compare pre-shelter levels with those of control pet dogs; and (2) explore the relationship between HCC and urinary cortisol levels (UCCR) of the same dogs at the shelter and after adoption.
In an earlier study, urinary cortisol responses in dogs in the shelter were higher than after adoption 30 . Likewise, for this study in-shelter HCC levels were also expected to be higher than post-shelter HCC levels, and higher than HCC of pet dogs. Pre-shelter HCC levels were expected to be more variable, as dogs could come from all kinds of situations before entering the shelter, including high and low stressful situations. HCC levels 6 weeks in-shelter and 6 weeks post-adoption were expected to correlate with urinary cortisol levels of urine samples collected during in-shelter and post-adoption, similar to the correlations of cortisol levels with other matrices in the literature as described above.

Materials and methods
Subjects, demographics, and housing. Fifty-two shelter dogs admitted to the largest animal shelter in the Netherlands (Animal Shelter DOA) between October 2018 and August 2019 were included (for demographics, see Supplementary Table 1). Exclusion criteria for the dogs to participate in the study are described earlier in more detail 30 , and comprised being affected by a physical health condition, high levels of anxiety-or aggressionrelated behaviour, being younger than 1 or older than 13 years of age, and being housed in pairs. The final shelter dog group had a mean age of 3.8 years (range 1-13 years), with 18 females (6 neutered, 9 entire, 3 unknown) and 34 males (12 neutered, 22 entire). The duration of stay in the shelter for these dogs ranged from 5 days up to 8 months, but most stayed in the shelter for at least two weeks. Based on the appearance, an experienced shelter employee assigned the dogs to a breed group 31 . Breed labelling of shelter dogs is highly unreliable 32,33 , but we labelled the dogs in order to match a control group. Dogs were housed as described earlier 30 : individually, in kennels with a glass-fronted indoor and bar-fronted outdoor enclosure, and kennels were only accessible to staff, volunteers, and the researchers of this study.
A control group of twenty pet dogs, balanced for characteristics of the shelter dog group on breed group 31 , size (body weight class), age class 30  where the same spot was resampled, at several time periods in-and post-shelter with equal regrowth periods in between to assure coherent time periods per sample (see Fig. 1): after intake (T1), after 6 weeks in the shelter (T2), 6 weeks after adoption (T3) and 6 months after adoption (T4, cut by new owners). Participating new owners of the shelter dogs received written instructions and tools for urine collection and hair sampling at T4 sample collection period. Hair samples of control pet dogs were collected once.
Collection hair samples. Hair samples of dogs were preferably shaved to the level of the skin to remove all hair from the sample area, except for being cut with scissors when the dog was scared for the noise of the razor or when the sample was taken by the owners 6 months after adoption. Also, when dogs entered as a stray, a smaller intake sample was cut for aesthetic reasons, as stray dogs can be recollected by their owners within two weeks. Our pilot study with pet dogs (n = 5) revealed no differences in HCC in shaved versus cut hair samples (see Supplementary Fig. 1, two-tailed paired-samples t-test, t = − 0.57, p = 0.597), therefore both shaved and cut samples were included in the analysis. In total, exactly half of the samples were shaved, and half were cut. Not all hair samples for all shelter dogs were collected, due to recollection of the dogs from the shelter within 6 weeks or no participation after adoption. Where possible, a shave/re-shave protocol was used 16,18 . A 3 × 3 cm area was shaved or cut bald on T1 and pre-adoption to allow a re-shave of newly grown hair at 6 weeks in-shelter and 6 weeks post-adoption. Hairs were collected following a standard protocol in the dorsal neck region as this region is often used in other studies and is easy to access [23][24][25]28 . The researcher or owner wore nitrile gloves and shaved (Wahl® professional Cordless Trimmer-Super Trim, Type 1592) or cut (blunted scissors) the dog's hair as close to the skin as possible. Razor and scissors were cleaned between every sample collection. Hairs were kept in aluminium foil in an envelope in a dark closet until analysis, which occurred 6 -18 months after sampling.
Hair selection, processing, and analysis. Both wool and guard hair were used for analysis, as these are strongly correlated 23 . From samples T2 and T3, longer and therefore older hairs were removed to prevent contamination of hair outside of the shave/re-shave area. In the case of small amounts of differently coloured hairs being present in the samples, these hairs were removed to make all the samples from the same dog unicoloured to control for colour. The predominant hair melanin type was visually registered 17 . Shelter dog samples per dog were predominantly eumelanin (n = 24) or pheomelanin (n = 13), with fewer white (n = 5), agouti (n = 1) or mixed (n = 9) samples. Pet dog hair samples were eumelanin (n = 11), pheomelanin (n = 2), white (n = 6) and mixed (n = 4).
The hair processing protocol was adapted from previous studies 12 . Samples were washed twice with isopropanol and dried in a stove at 37 °C. Hair samples were cut into small pieces and powdered using metal beads (Lab Services BV Biospec beads, 3.2 mm) in a TissueLyser II (QIAGEN). 1.5 mL methanol was added to each powdered hair sample (30 ± 5 mg, samples that weighed less were marked) for steroid extraction. Samples were incubated overnight at room temperature on an end-over-end roller. 1 mL of the methanol supernatants were dried in a Speed Vac Concentrator (CentriVap Concentrator Labconco, 42 °C). Supernatants were dissolved in phosphate buffer provided by the commercial ELISA essay kit (Salimetrics, LLC), with diluent amount differing per hair sample weight: < 10 mg sample in 55 µL diluent, 10-14 mg in 65 µL, 15-24 mg in 135 µL, > 25 mg in 200 µL.
Cortisol determination with the ELISA kit was performed according to the protocol provided by the manufacturer. Four microtitre plates were used for the ELISA, with samples of one dog assigned to one plate to prevent inter-assay differences within one dog. Plates were read at optical density 450 and 490 nm. Results were expressed as cortisol concentration in pg of cortisol per mg of hair sample. Samples were run in duplicate and inter-assay coefficient of variability ranged from 3.1% to 16.5%; intra-assay variability ranged from 1.3 to 3.1 µL/dL.  Figure 1. Timeline of shelter dog hair and urine sample collection moments: after intake (T1), after 6 weeks in the shelter (T2), 6 weeks after adoption (T3) and 6 months after adoption (T4). Stray dogs were cut instead of shaved at intake to prevent that a large bald spot was visible if dogs would be recollected by their owners.
Scientific Reports | (2022) 12:5117 | https://doi.org/10.1038/s41598-022-09140-w www.nature.com/scientificreports/ Urinary cortisol:creatinine ratio. For shelter dogs, urine samples were collected on day 1, 2, 3, 7 and 12 both during their time in the shelter and after adoption. Samples were collected after intake in the shelter (by the researchers) and after adoption (by the new owners). Urine was collected between 6:00-12:00 AM (median 8:36 AM). Naturally voided morning urine was captured mid-stream with a ladle and transferred with a pipette to a polyproetheen tube. If the dogs in the shelter were not naturally urinating outside of their kennel, urine on the kennel floor was used. As no significant difference was found in urinary cortisol:creatinine ratio (UCCR) between urine samples of the same dog on the same morning captured directly versus collected from the kennel floor (n = 19, Paired Wilcoxon test: V = 117.5, p = 0.376), samples from both methods were included for statistical analysis.
Samples were stored and analysed by the University Veterinary Diagnostic Laboratory of the Faculty of Veterinary Medicine at the Utrecht University, the Netherlands, as described earlier 30 .
Data handling and statistics. Data were stored and cleaned in Microsoft Excel® files (Microsoft Corporation). Statistical software program RStudio (version 1.0.136 -©RStudio, Inc.) was used for linear mixed model analysis with the package 'Nlme' 34 and Spearman's rank correlation with the package 'Hmisc' . Graphs were created in Graphpad Prism (version 8.3.0 ©GraphPad Software, LLC).
Three HCC outliers (mean ± 3s.d. per time point T) were removed: one on T1, T2 and T4. HCC data was rightskewed and therefore (natural) log transformed before statistical testing and back transformed for interpretation. Back transformed (exp) log model values resulted in ratios, with a ratio < 1 meaning a lower value and > 1 a higher value than the reference mean. Alpha level was set at p < 0.05. For mixed models, 95% confidence intervals (CI) ranges < 1 or > 1 were considered significant.
UCCR mean of day 1 + 2 + 3 + 7 + 12 per dog was calculated to determine one comprehensive outcome for both in-shelter and post-adoption. The relation between UCCR means and corresponding original HCC outcome on T2 and T3 was evaluated with a Spearman's rank correlation test for non-parametric variables, as both UCCR and original HCC were not normally distributed (Shapiro-Wilk tests and visual inspection of boxplots and quantile-quantile plots).
HCC data of I) T1 of shelter and pet dogs and II) T1 of relinquished versus stray shelter dogs were compared using a two-tailed t-test on the log transformed HCC data.
A linear mixed effects model was fit on the outcome variable HCC data of the shelter dogs with a fixed effect for 'sample collection moment' (T1-4) and a random effect for 'dog ID' (individual identity). A full model with variables described below was fitted and variables were dropped with the 'drop1' function based on a backward selection approach, using the Akaike information criterion (AIC) to determine the best model fit with Maximum Likelihood estimation. The best fitting model included the factors as seen in Table 1 and described in the results. Variables dropped from the model based on AIC were: 'sample collection moment*body weight class' , 'season of collection' (Dec-Feb/March-May/June-Aug/Sept-Nov), 'sample collection moment*kennel history' (no/unknown/yes) and 'neuter status' (no/unknown/yes). To test for the best fitting model using the AIC, various correlational and variance structures were added to the model. The best fit was a model with a correlation structure (continuous autoregressive model of the order 1, CAR1) and variance transformation (weights = varPower for 'sex'). Restricted Maximum Likelihood estimation was used for the final model. Models were evaluated by visual inspection of the residuals (normality and constant variance).
Ethical note. Methods in this study were performed in accordance with relevant guidelines and regulations.
All owners agreed and volunteered to participate in this study and signed informed consent. All experimental protocols were approved by the institutional committee Utrecht Animal Welfare Body of Utrecht University, The Netherlands. The study was carried out in compliance with the ARRIVE guidelines.

HCC in shelter dogs.
In the shelter dog group, the factors 'sample collection moment' , 'weight class' , 'sex' , 'reason for admission to the shelter' and 'kennel history' and an interaction effect between 'sample collection moment' and 'age class' and between 'sample collection moment' and 'melanin type' , significantly explained HCC variability. See Table 1 for full mixed model results with estimates in ratio, 95% confidence interval and F-test statistics of sample collection moment effect, other main effects, and interaction effects as described below).
Sample collection moment. Regarding the moment of sampling, the 6 weeks in-shelter HCCs (T2, mean ± s.d.: 21.8 ± 9.4 pg/mg, n = 23) were significantly higher than the intake sample of the shelter dogs (T1, 16.0 ± 6.8 pg/ mg, n = 51), where post-adoption samples were not (T3, 17.7 ± 8.5 pg/mg, n = 24 and T4, 18.4 ± 9.5 pg/mg, n = 22, see Fig. 2a for the data of all dogs, 2b for the subset of dogs that were not adopted yet at T2 and therefore had an available sample at T2).
Other main effects. Overall, low weight dogs (i.e., smaller dogs, < 10 kg, 20.2 ± 7.9 pg/mg, n = 18 or 35% of all dogs) had higher HCCs than larger weight dogs (20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) 2 Estimated ratio of mean of specified sample collection moment and mean at T1. 3 Estimated ratio of mean of specified weight class and mean in reference weight class. 4 Estimated ratio of mean of specified sex and mean in reference sex. 5 Estimated ratio of mean of specified relinquishment reason and mean in reference relinquishment reason. 6 Estimated ratio of mean of specified kennel history information and mean in reference kennel history information. 7 Estimated ratio of mean of specified melatonin type and mean in reference melatonin type at same sample collection moment. 8 Estimated ratio of mean of specified age class and mean in reference age class at same sample collection moment. www.nature.com/scientificreports/ Interaction effects. The sample size per sample collection moment for the interaction effects was often very low, therefore the following results need to be interpreted with caution and only sample size is mentioned. A significant interaction effect between 'sample collection moment' and 'age class' was found, where dogs from age class 5-7 years had lower HCC on T3 (n = 2) and T4 (n = 1) but higher HCC on T2 (n = 4) compared to dogs from age class 1-4 years (1-4 years: T2 n = 15; T3 n = 18; T4 n = 18). Another significant interaction effect was found between 'sample collection moment' and 'melanin type' , with eumelanin samples having higher HCC than pheomelanin samples on T2 (n = 14 versus n = 2), lower HCC than agouti on T2 (n = 14 versus n = 1), lower HCC than white samples on T3 (n = 13 versus n = 3), and higher HCC than mixed samples on T4 (n = 8 versus n = 6).

Discussion
In this study we compared HCC in shelter dogs over different chronological time points during sheltering and after adoption with a shave/re-shave method, aiming to further validate HCC to measure chronic stress in dogs.
We also compared pre-shelter levels with those of control pet dogs and explored the relation between HCC and urine cortisol levels in shelter dogs. In short, we found HCC levels of shelter dogs to be higher in samples of hair that grew during the first 6 weeks in the shelter than in samples collected at intake in the shelter, and to be higher than in two samples taken 6 weeks and 6 months after adoption into a new home. This implies that cortisol levels in the shelter were higher and thus dogs appear more stressed during the first shelter period. HCC levels from www.nature.com/scientificreports/ the intake samples of the shelter dogs did not differ from the control group of pet dogs, reflecting no overall difference in cortisol levels between the two groups when living in familiar homes. HCC was moderately but significantly correlated with mean urine cortisol creatinine ratio, showing a relation between the two measures.
Half of the dogs (27/52) were already adopted out or returned to the former owner after 6 weeks in the shelter and were therefore not available for hair sampling 6 weeks in the shelter. Dogs could still be in the shelter after 6 weeks for many reasons; breed type, body size and in-kennel behaviour are known to influence adoption likelihood 35 . Therefore, a bias can be present in the type of dogs that provided more samples at T2. For example, at T2 more large dogs provided samples than smaller dogs (> 30 kg 35% versus < 10 kg 17% of total dogs at T2, compared to 21% versus 35% in total). Another possibility is that dogs that cope worse with the shelter environment were still in the shelter at T2. However, as seen in Fig. 2B, these were not the dogs with higher HCC levels in general at T1, making this option less plausible.
HCC levels of dogs did not appear to be elevated during the period before the dogs were admitted to the shelter compared to after adoption, which implies no profound stressful pre-shelter period. However, it is possible that part of the hair shaft re-shaved at T2 did develop prior to the first shave at intake, as hair in the follicle and up to ~ 1 mm out of the follicle could not be shaved due to proximity to the skin. During the late anagen phase of hair follicles, hair follicles in dogs reach a length of ~ 2.5 mm 36 . Hair grows at a mean of 1 cm per month in humans 37 and ranged from 5.3-12.0 mm per month in pigs and 3.5-17.0 mm per month in cattle 38 . In dogs, hair regrowth was a mean of 82% at 6-12 weeks after shaving in German Shepherds and Labrador Retrievers 17 and fully regrown to pre-shave length in an average of 14.7 weeks in Labrador Retrievers 39 . Therefore, our re-shaved hair shaft 6 weeks in the shelter could contain ~ 3.5 mm of hair that had already developed up to weeks before intake at the shelter. To control for this 'old' hair growth, future studies can remove the first few millimetres of the hair shaft that are not of interest for analysis.
We found several dog-related factors to influence HCC. Interestingly, we found an effect of weight class, with smaller dogs having higher HCC than larger dogs. Urinary cortisol:creatinine levels were also higher in small shelter dogs than larger dogs in earlier studies 30,[40][41][42] , which could be explained by relatively little creatinine production in smaller dogs as creatinine production is proportional to muscle mass 43 , however not relevant to HCC levels. Also, salivary cortisol levels were lower in giant/large dogs 44 and were negatively correlated with body weight 41 , therefore cortisol levels in general seem to be higher in smaller dogs. Bowland et al. 27 also found HCC to be negatively associated with size (based on morphometric measurements) in hunting village dogs in a reserve in Nicaragua, although this effect disappeared in a model with all factors incorporated. To our best knowledge, no other studies have investigated the effect of body size on HCC in dogs. However, it has been suggested that plasma cortisol varies with mass-specific metabolic rate in mammals, where larger mammals have lower cortisol levels 45 . Furthermore, an effect of sex on HCC was also found, with females having higher HCC than males. Higher HCC in the hair of female dogs has been found in recent studies in herding dogs 25 , solitary hunting breeds 26 , and hunting village dogs 27 , although reproductive status of the female (pregnancy or lactation) could have affected HCC. Other studies, however, did not find an effect of sex on HCC in dogs 21,24 . In general, across species, an inconsistent influence of sex on HCC has been found between studies 9 . However, female dogs showed a higher behavioural and HPA-axis response to stressors 1,46,47 and it is known that HPA-axis responses are higher in females in rodents and humans as a result of circulating estradiol levels 48,49 , which can be an explanation for the higher HCC. We also found a puzzling interaction effect of age, where dogs of 5-7 year had higher HCC levels at T2 but lower at T3 and T4 than younger dogs of 1-4 year. However, the number of samples of dogs of 5-7 years was very low, and therefore results need to be interpreted with caution. No effect of age on HCC has been found before in dogs 21,25 but for several species an age-dependent decline in HCC from young to adult age groups was found 9 . And last, stray dogs overall had lower HCC than relinquished dogs, but this difference was not visible in the intake samples in the shelter. Interestingly, Willen et al. 50 also found no difference in HCC between relinquished and stray dogs at intake in a shelter. Therefore, it seems that stray dogs respond different to sheltering than relinquished dogs, which can therefore depend on previous life experiences. This supports the hypothesis that prior life history of dogs can mediate coping to a shelter environment 51 , although it must be mentioned that stray dogs in the Netherlands are mostly on the streets only for a few hours to a few days before they are captured and admitted to a shelter. To summarize, these dog-related factors need to be considered in future studies as their effect on HCC is present or not yet clear.
We found no overall effect of melanin type on HCC, but an effect that differed per sample collection moment. However, sample sizes per sample collection moment were often very low and therefore these results need to be interpreted with caution. Compared to eumelanin (brown-grey-black) hair samples, pheomelanin (blondeyellow-red) contained lower HCC at T2, white contained higher HCC at T3, and mixed samples of hair contained lower HCC at T4. Therefore, we did not find a consistent effect of colour on HCC. So far, in the literature, the effect of colour on HCC is also inconsistent across and within species 9,52 . In general, due to the variety in hair colours between and within dogs, visual evaluation of the dominant pigment type in hair samples can be difficult. Therefore, chemical analysis of eumelanin and pheomelanin in dog hair sample or analysis of pigment genes in fundamental studies can help to objectively evaluate differences in HCC in dog hair samples. For example, in humans, the effect of hair pigmentation on HCC was studied by evaluating hair pigmentation genes 53 .
Overall, HCC in shelter dogs is highly promising to provide information on cortisol responses retrospectively and in the long-term. HCC can be a 'stable' measure for evaluating these responses of dogs during their lives, for example to evaluate lifestyles and personality traits in dogs 23,25,26 . For adaptation to a novel environment however, an acute changing situation may not be reflected as the HCC in the hair shaft gets 'averaged' or 'diluted' during analysis 54,55 and therefore other matrices such as urine are a better choice for evaluating acute cortisol levels. In our study, HCC correlated moderately but significantly with UCCR levels, showing a form of construct validity. Correlations between cortisol matrices with different time spans are complex, as some matrices such as blood or urine reflect a 'snap shot' in time while others integrate all these moments in time, such as hair and nails 54 .

Scientific Reports
| (2022) 12:5117 | https://doi.org/10.1038/s41598-022-09140-w www.nature.com/scientificreports/ Therefore, even when averaging urinary cortisol over a prolonged period, it can be difficult to find a relationship between HCC and urinary cortisol levels (such as in humans 56,57 ). To relate HCC with cortisol levels in another biological substrate, nail tissue is a more logical option 28,58 , however cortisol levels in the nails of dogs have not yet been validated. Furthermore, behavioural responses to stress provide important information to draw conclusions on physiological responses such as HCC 1 , for example on the valence of the stress response, i.e., positive or negative arousal. The average levels of HCC in our study are comparable to other studies. However, comparisons of HCC levels between studies are difficult, as cortisol extraction from the hair might depend on the way the hair is processed. For example, powdering the hair can result in a 3.5 fold increase in cortisol measurement compared to when the hair is only chopped 52 . More standardized protocols for both the analysis of HCC and the collection of the hairs would improve comparability of studies in this field. Shave/re-shave methods are preferred when aiming to evaluate cortisol levels over a certain period, as this assures a known timeline of cortisol incorporation into the new hair and avoids including follicles in the sample 10 . Also, better insight of cortisol deposition in different types of dog fur, for example related to hair growth (breed dependent), can improve reliability of this measure to evaluate cortisol levels in dogs.
To conclude, hair cortisol concentration in shelter dogs increased after 6 weeks spent in the shelter and correlated moderately with UCCR levels. Therefore, to improve canine welfare, HCC analysis can be a reliable and feasible non-invasive method to evaluate cortisol levels in shelter dogs, especially when comparing HCC levels over longer periods within dogs. A shave/re-shave method is preferred when aiming to evaluate HCC levels during a specific period. However, we and others 9,23,24 raise concerns on methodological issues and missing information on how cortisol is deposited in the hair shaft in a broad range of dog breeds with different hair types. Therefore, further studies are needed to shed light on these concerns.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.