Effects of breed, management and personality on cortisol reactivity in sport horses

Sport horses need to fulfill high physical and psychological requirements during training and competition. These as well as certain conditions of modern husbandry may affect their wellbeing. Here we aimed to (1) evaluate effects of demographic and management factors as well as personality traits on stress reactivity of sport horses, (2) investigate if elite sport horses have elevated stress levels compared to amateur sport horses, and (3) assess whether different equestrian disciplines differentially influence equine adrenal cortex responsiveness. For this purpose, we visited 149 healthy elite (n = 94) and amateur (n = 54) sport horses in Switzerland and performed an adrenocorticotropic hormone (ACTH) stimulation test. Additionally, a person who was familiar with the horse completed a questionnaire about demographic and management factors and their horses’ personality traits. Linear models were calculated to assess associations between the questionnaire data and salivary cortisol 60 (T60) and 90 (T90) minutes after ACTH stimulation. While the model at T90 was not significant, post-stimulatory cortisol after 60 minutes appears most informative in line with a previous study and was significantly affected by the breed and by three management factors: “number of riders”, “hours spent outside” and “group housing” (adjusted r2 = 15%, p<0.001). Thoroughbred and Warmblood horses displayed an increased adrenal response compared to Franches-Montagnes horses. Horses with several riders had a less pronounced reaction than horses with one rider, and horses that spent more time outside had a decreased response compared to horses that were stabled most of the time. Horses living in groups showed higher post-stimulatory cortisol values than horses that were housed singly. However, no significant associations of cortisol responsiveness with personality traits were found, and neither the use as elite or as amateur sport horses nor the discipline had an effect on the cortisol response. This suggests that optimizing husbandry conditions may be more important for improving equine welfare than changing their use.


Introduction
Using ACTH stimulation tests and questionnaire surveys, the aims of this study were to explore (1) effects of breed, demographic and management factors and personality traits on the adrenal reactivity of sport horses, (2) whether elite sport horses have elevated stress-levels in comparison to amateur horses, and (3) if sport discipline is an influencing factor.

Sample population
The study was approved by the cantonal authority for animal experimentation, the Veterinary Office of the Canton of Bern, Switzerland, and subsequently by all other included cantons (27608/BE26/16). Before participating in the study, all owners gave their informed written consent. This prospective cross-sectional study was carried out between January and August 2017. Swiss private practitioners and official team veterinarians of the different disciplines were contacted to establish contacts with riders and owners in order to recruit horses of different equestrian disciplines (show jumping, dressage, eventing, endurance, driving, vaulting, para-equestrian) for both an Elite Sport Horse Group (ESHG) and an Amateur Sport Horse Group (ASHG), respectively. In order to participate in the study, horses had to have competed at predefined levels of competition (Table 1). Overall, 149 horses from 17 different cantons in Switzerland were included. Ninety-five of them were part of the ESHG and 54 of the ASHG (Table 1).

Experimental setup
The first author visited all horses in their 'home stable' environment. Each horse underwent an ACTH stimulation test, and a person who was familiar with the horse (owner, rider or caretaker) completed a questionnaire. Horses did not participate in any physical activity two hours before the visit, and they were not transported and did not attend competitions during the preceding 48 hours. They received no medication or additives except for over-the-counter  [41]. The general examination included an assessment of mucous membrane color, lymph nodes, heart rate, respiratory rate, and rectal temperature at rest. Only horses in a good general condition with normal general examination findings were included in the study. The body condition score (BCS) was recorded, according to the 9-point Henneke BCS system ranging from 1 (poor) to 9 (extremely fat) [42]. During the time of the test, horses remained in their normal environment such as their stall or open stable. They were prevented from drinking during the time of the test to avoid dilution of saliva before cortisol measurements. For this reason, self-watering troughs were covered with plastic bags. ACTH stimulation test. For the ACTH stimulation test, based on the estimated weight, a dose of 1 μg/kg BW synthetic ACTH1-24 (Synacthen tetracosactidum 0.25 mg/ml equivalent to 25 IU/mL; Novartis, Vilvoorde, Belgium) was injected intravenously into the jugular vein with a 20 gauge needle. The dosage was chosen based on previous publications [23,26,43,44]. Saliva samples were taken with salivettes (Sarstedt, Nuembrecht, Germany) before (0 minutebaseline; T0) and 60 (T60) and 90 (T90) minutes after ACTH administration. The salivette swab was placed into the horse's mouth for at least 40 seconds until it was completely soaked with saliva. Afterwards, the swab was replaced into the salivette and stored in a box with cool packs until return to the clinic. Salivettes were then directly centrifuged at room temperature for 10 minutes at 185 x g and stored at -20˚C until the analysis. For the determination of salivary cortisol concentrations, a competitive enzyme immunoassay (cELISA, Salimetrics, Newmarket, United Kingdom) was used, which has been previously validated for use in horses [23,26,44].
Questionnaire. The first part of the questionnaire was designed to record information about demographic and management factors such as housing, use and feeding, stereotypies, and clinical signs such as teeth grinding or yawning ( Table 2, S1 File). After termination of the study, some questions were re-coded because there were too few observations in one of the categories ( Table 2).
The aim of the second part of the questionnaire was to evaluate personality traits of the participating horses. For this purpose, a previously validated questionnaire [45], relying on the opinion of caretakers regarding the personality of horses, was adapted for the current study (Table 3, S1 File). The original version by Momozawa et al. [45] consisted of 20 items (nervousness, concentration, self-reliance, trainability, exitability, friendliness towards people, curiosity, memory, panic, cooperation (during grooming), inconsistent emotionality, stubbornness, docility, vigilance, perseverance, friendliness towards horses, competitiveness, skittishness, and timidity), leading to four factors after a principal component analysis. These factors were labelled 'Anxiety', 'Trainability', 'Affability' and 'Gate Entrance'. For the current study, the question about the horses' behavior at the starting gate ("gate entrance") was excluded, since it is only relevant for racehorses. Three additional questions regarding cooperation during riding, trailer loading and transportation and four questions about the horses' resilience during or after putatively stressful events such as competitions or transports including (1) recovery time; (2) susceptibility to infection (e.g. respiratory infections); (3) loss of appetite; and (4) signs of discomfort (e.g. flehmen, colic, prolonged resting times) were added instead. Out of the 26 questions of the second part of the questionnaire, those on 'competitiveness' and 'docility' were removed for statistical analysis. Many horses did not have direct contact with other horses; therefore, it was not possible to evaluate competitiveness. German and French translations of docility turned out to be unsatisfactory and respondents were confused about the meaning. Questions of this second part of the questionnaire were answered using a scale ranging from (1) poor to (9) excellent. To identify possible pitfalls or misunderstandings, four independent volunteer horse owners completed both parts of the questionnaire. The questionnaire was adapted according to their comments, and German and French translations were prepared for the participants. A person who was familiar with the horse was asked to complete the questionnaires during the time the ACTH stimulation test was performed.

Statistical analysis
All statistical analyses were performed in R [46]. Skittishness The left column shows the questions from the second part of the questionnaire. Variables written in italics were not part of their original survey by Momozawa et al. [45]. The remaining columns display the principal components (PC) 1 to 7 and the respective factor loadings for each variable. Values written in bold were grouped under the corresponding factor. Underlined values were grouped together under the respective factors by Momozawa et al. [45].

Principle component analysis on personality traits.
A Principle Component Analysis PCA (function prcomp) was performed on the second part of the questionnaire (personality traits) in order to reduce the number of variables and obtain principle components for further analysis. A Varimax rotation was carried out, and components with Eigenvalues >1 were retained.
Evaluation of the effect of competition level and discipline on stimulated cortisol levels. For the comparison of adrenal reactivity in the ESHG versus the ASHG, the disciplines para-equestrian, driving and vaulting were excluded, since no comparable groups of amateur sport horses were available. Thus, 71 horses of the ESHG and all 54 horses of the ASHG were included in the analysis. Separate linear regression models were computed to evaluate the effect of category (ESHG versus ASHG) and the influence of the different disciplines on cortisol levels after exogenous stimulation. The residuals of the models were checked visually via scatterplots for independence and via quantile-quantile-plots, histograms and Shapiro-Wilk tests for normal distribution. The models were tested for homoscedasticity of variance via visual inspection and Breusch-Pagan test (package car, function ncvTest). If necessary, squareroot transformation of the respective response variable was performed to meet linear model assumptions. Level of significance was set at p < 0.05.
Evaluation of the effects of demographic and management factors as well as personality traits on stimulated cortisol levels. As neither the competition level nor the discipline showed a significant effect on stimulated cortisol levels (also see results section), data of the ESHG and the ASHG were pooled for further analysis. Linear regression models (function lm) were calculated separately for the dependent variables T60 and T90, with separate models for (1) demographic and management factors and (2) personality traits, respectively. In the first model, 24 demographic and management factors from the first part of the questionnaire (Table 2) in addition to the exact time of day of the start of the ACTH stimulation test and the horse's BCS were included as predictors. In the second model, the seven principal components derived from the personality questionnaire were used as explanatory variables. Initially, all variables were entered in the models, which were then reduced by stepwise backward selection (function step). The fit of the models was assessed using Akaike's Information Criterion (AIC), and the model with the lowest AIC was retained as the final model. Results were corrected for multiple testing via Bonferroni correction. In the Results section, we report original p-values and subsequently indicate whether they can still be considered significant after Bonferroni correction. As above, model assumptions were checked and were adequate for all models. Variables of the models were checked for multicollinearity via use of the variance inflation factor (package car, function vif). There was no multicollinearity in the variables included in the final models. The 95% confidence intervals and likelihood-ratio tests (LRT) were calculated for the final reduced models.

Principal component analysis on personality traits
The PCA on the remaining 24 items yielded seven cumulative factors with an Eigenvalue > 1 accounting for 63.4% of the common variance. The factors partially confirmed the factor structure determined for the original version of the questionnaire [45]. Factor 1 included five out of six items (nervousness, excitability, panic, skittishness, timidity) that were grouped as "anxiety" by Momozawa et al. [45]. In the present study, it contained two additional items (concentration, perseverance). Factor 2, 'trainability', consisted of two (trainability, memory) out of four items grouped by Momozawa et al. [45] and of two additional items (stubbornness, cooperation during riding). Factor 3 consisted of four items that were not part of the original questionnaire and was named 'recovery' (recovery time after efforts, infections after efforts (e.g. respiratory infections), inappetence after efforts and discomfort after efforts). Factor 4, 'social interaction' and factor 5, 'alertness', did not form part of the factors identified by Momozawa et al. [45]. Factor 4 consisted of 'self-reliance', 'friendliness towards horses' and 'discomfort after efforts', and factor 5 of 'concentration', 'curiosity', and 'vigilance'. In agreement with Momozawa et al. [45], factor 6 was labeled 'affability'. It consisted of two (friendliness towards people, cooperation (during grooming)) out of the four items that were grouped in the original questionnaire and of one further item (inconsistent emotionality). The newly defined factor 7, 'transportability', consisted of the newly added questions 'cooperation during trailer loading' and 'cooperation during transport' (Table 3 and Fig 1).

Evaluation of the effects of demographic and management factors as well as personality traits on stimulated cortisol levels
The final reduced linear regression model evaluating the effect of demographic and management factors on cortisol levels 60 minutes after ACTH stimulation was significant after Bonferroni correction (T60: p < 0.001; adjusted R 2 = 15%; F 1, 122 = 3.305). Likelihood ratio tests revealed four variables with a significant effect: 'breed' (LRT: p = 0.02), 'number of riders' (LRT: p = 0.009), 'hours spent outside' (LRT: p = 0.002) and 'group housing' (LRT: p = 0.045) ( Table 4). Regarding 'breed', positive parameter estimates for Warmblood and Thoroughbred horses indicate higher post-stimulatory cortisol values in these breeds compared to Franches-Montages horses that served as the reference in the model. Regarding the 'number of riders', horses with one rider displayed higher post-stimulation cortisol values than horses with two or more riders. Furthermore, horses that spent more time outside of their stalls on pasture or for training showed lower values than horses being less outside. Predicted values for cortisol 60 minutes after ACTH stimulation are plotted for the predictors 'breed', 'number of riders', 'hours spent outside', and 'group housing' in Fig 2. The linear regression model evaluating demographic and management factors influencing cortisol levels 90 minutes after ACTH stimulation (Table 4) was not significant (T90: p = 0.085; Adjusted R 2 = 4%; F 1, 122 = 1.897).

Discussion
This study found that 'breed', 'number of riders', 'hours spent outside', and 'group housing' are significantly related to salivary cortisol concentrations in horses 60 minutes after ACTH stimulation, which is considered as a measure of long-term effects of stress [11]. There was, however, no difference in post-stimulatory cortisol levels between elite sport horses and amateur sport horses, nor between horses of different disciplines. Furthermore, personality traits were unrelated to the horses' cortisol responsiveness.
Notably, 60 minutes post-stimulation appeared to be the most informative time point, as linear regression models evaluating cortisol concentrations after 90 minutes showed no significant outcome. This is in agreement with our previous study [26], where sampling after 60 minutes also revealed the best association with the occurrence of glandular gastric disease. The present study supports our previous hypothesis [26] that predominantly the initial increase of cortisol after ACTH stimulation is important for the evaluation of the cortisol responsiveness in the context of stress-related disorders.
The current study was the first known study to compare stress of high-performance athletes with amateur sport horses and found no difference in post-stimulatory cortisol levels between the two groups. However, studies investigating acute stress in relation to training [47][48][49] and competition [33,34] showed that experienced horses had lower basal plasma cortisol values than inexperienced horses [47]. Furthermore, horses with a superior physical training level had decreased basal cortisol levels, indicating that trained horses experience less stress during exercise than untrained horses [47][48][49]. Although the level of experience of riders had no Effects of breed, management and personality on cortisol reactivity in sport horses impact on physiological stress parameters of horses during exercise in one study [50], the effect of the rider on equine welfare during exercise is not to be underestimated. Indeed, the interaction between horse and rider was shown to influence the horses' level of stress in previous studies [35,51]. Potentially, the increased requirements and the elevated stress levels that elite sport horses encounter may be compensated by their greater experience, their better fitness, and possibly by their more professional riders. This may be the reason why measurable indicators of long-term stress, such as cortisol responsiveness to external stimulation do not differ between elite and amateur sport horses. Overtraining, however, defined as a depletion of performance capacity, was revealed to lead to deleterious consequences including weight loss, behavioral changes, and reluctance to exercise [52]. It was also reported to be associated with lower baseline cortisol values [40] and-quite contradictory-with an increased [53], decreased [54], or unchanged [40] adrenal response to exogenous ACTH. Interestingly, Bruin et al. [53] reported an increased response in horses with early stages of overtraining, whereas Persson et al. [54] found a decreased response in horses that were progressively and severely affected. In comparison to these previous investigations, the present study tried to evaluate and compare the overall adrenal capacity of the sport horse in a normal environment without external stimuli such as training or transportation. Our results indicate that elite performance levels do not lead to increased adrenal cortex responsiveness to ACTH. We cannot specify, though, if this effect reflects targeted selection of horses that are able to withstand the enhanced requirements in high-performance competitions. Also, with respect to the effect of discipline, we found no difference in adrenal cortex reactivity between horses used for dressage, show jumping, eventing, or endurance. While two previous studies reported increased cortisol levels in dressage horses [47,55] compared to horses of other disciplines in a competition environment, our results indicate no long-term effect of discipline on HPA axis responsiveness.   Effects of breed, management and personality on cortisol reactivity in sport horses Thoroughbreds and Warmbloods displayed an increased adrenal response in comparison to Franches-Montagnes horses. This is in line with a previous study of our group, finding a lower response to ACTH in Franches-Montagnes horses compared to Warmblood horses [44]. In this previous study, however, all Franches-Montagnes horses were stallions and so it was not possible to differentiate whether the effect was due to the sex or the breed of the horses. The present study investigated the potential effects of sex and breed as separate explanatory variables, indicating that the breed (but not sex) might be influencing equine adrenal reactivity. This suggests a possible underlying genetic effect of adrenal reactivity to ACTH stimulation-but note that the sample size for the Franches-Montagnes horses was small at n = 12, and 10 of these were driving horses trained by only three different trainers. Thus, results need to be interpreted with caution, and further confirmatory studies are needed.
Concerning the 'number of riders', interestingly, horses with two, three, or more riders had a less pronounced adrenal response than horses with only one rider. Conceivably, horses that are accustomed to higher levels of stimulation in everyday life may be better equipped to deal with challenging situations [56][57][58]. In accordance, horses that spent more time outside of their stalls on pasture or for training and competition had a decreased reaction to exogenous ACTH. This is in line with previous studies indicating better welfare in pastured horses compared to stalled horses [6,32,59], possibly because the possibility to graze outside more closely reflects the natural living conditions of the horse [6,60].
'Group housing' was another variable that significantly influenced post-stimulatory cortisol concentrations. Horses living full-time in groups and horses living in groups on pasture showed higher post-stimulatory cortisol values than single housed horses. Contradictory results have been reported in previous studies investigating stress levels of domestic horses in different types of housing. One study found that horses living in single housing without physical contact to other horses had higher fecal glucocorticoids compared to horses with visual, auditory, and tactile contact to other horses and horses living in groups of two or more individuals [18]. The authors concluded that integrating social contact into the housing of the horse improves equine welfare [18]. Other studies found no effect of the type of housing, neither in baseline salivary cortisol concentrations in adult horses [61] nor in fecal glucocorticoid metabolites and baseline salivary cortisol in young horses stabled for the first time [62,63]. Similar to the results of the present study, Visser et al. [27] found lower cortisol concentrations in response to a CRH stimulation test in individually housed horses. In contrast to our interpretation, however, they proposed a decreased, not an increased response as a potential sign for chronic stress. In line with the analysis of the other variables in the present study, we conclude that group housing might represent a further stressor activating the HPA axis. Although group housing appears overall to have beneficial effects on horse welfare, it is also associated with social challenges [64], which might be reflected by the increased cortisol reactivity of group housed horses in the present study. In this context, social stability of groups might be important to consider in future studies, which itself may be influenced by many factors such as group size, age, and sex composition, frequencies of changes or feeding management, available space and infrastructure (e.g. feeding places in comparison to the number of horses).
Regarding personality, the present study could confirm the overall factor structure of the original questionnaire by Momozawa et al. [45] to some extent, although it yielded some additional factors and some differences in loadings of individual items. Some differences in factor structure are expectable given that firstly the study population was more varied than in Momozawa et al. [45], where only racehorses were investigated, secondly not all questions were identical (two questions from the original version were removed and four questions were added), and thirdly German and French translations of the questionnaire were used. Still, the similarity with the corresponding factors in Momozawa et al. [45] indicates validity of the questionnaire.
As in previous studies using the questionnaire by Momozawa et al. [45,65], the individuals rating the horses' personality were not behavior specialists. In their capacity as either the horses' owners, riders or caretakers, however, they knew their horses best. Nevertheless, the analysis revealed no significant associations between the cortisol response to ACTH and personality factors.
Previous studies investigating associations between HPA axis function and personality traits in horses yielded diverse results. In line with our results, no correlations were found between temperament and basal serum cortisol values of horses used in therapeutic riding programs [66]. Another study investigated acute stress in stallions during transportation and found that calmer and more experienced stallions had a more pronounced increase in plasma cortisol following road transport than inexperienced, nervous stallions [67]. The authors concluded that lower cortisol concentrations in the latter group might represent a failure of the HPA axis to adapt to stress perceived during transportation [67]. As these examples demonstrate for horses, the evidence for associations between HPA axis function and personality has been inconclusive also in our own species [68] where both increased [37] and decreased cortisol concentrations [67] have been interpreted as signs of disturbed HPA function. Further research is needed in order to gain more specific insights into possible relations between HPA axis function and equine behavior.
A central limitation of this and similar studies remains the uncertain interpretation of the results of ACTH stimulation tests. While a decreased response has usually been interpreted as a sign of chronic stress in horses through desensitization of the HPA axis [27,39,54] (in its extreme form an Addison crisis), various studies demonstrate an increased response in potentially stressed horses [23,25,26,53]. Irrespective of these fundamental questions, a blunted response, typically described in the context of severe distress [17], was not expected in horses of the present study. At the time the study was conducted, all participating animals successfully took part in competitions and were in good health. Therefore, we hypothesized that increased responses might indicate higher stress responsiveness.
A further difficulty for the interpretation of the results is the use of an immunoassay for the measurement of cortisol concentrations in equine saliva. As shown for humans [69] and only recently for use in horses [70], cortisol results derived from immunoassays need to be interpreted with caution. While baseline values lack specificity, stimulated cortisol concentrations might be overestimated [70]. However, an evaluation of overall effects, at a group level rather than at an individual level, should still be reliable. Consequently, only stimulated cortisol concentrations were considered for the analysis of the present study.
Previous studies described seasonal variability and a diurnal rhythm within cortisol concentrations [61,71,72]. Due to the need to adjust to horse owners' and riders' schedules, it was not possible to collect samples at the same time of the day or within only one season. However, baseline cortisol values, which are the most likely to be affected by natural occurring release patterns, were not analyzed here. Cortisol releases induced by exogenous ACTH administration are expected to override diurnal or seasonal changes [73], and we statistically controlled for this effect by including time of day as a predictor in the models. This confirmed that time of day had no significant effect on the post-stimulatory salivary cortisol levels reported here.
In conclusion, from the current data, there is no evidence that the superior performance and the high demands that elite sport horses have to fulfill in comparison to amateur sport horses have long-term adverse consequences on their welfare as determined by their cortisol reactivity. However, breed and management parameters, such as number of riders, time spent outside, and housing, influence the amount of stress hormones released. Therefore, our results suggest that optimizing husbandry conditions, rather than changing performance level, may be more important for improving the welfare of horses.
Supporting information S1 File. Original questionnaire about demographic, management and personality factors. The questionnaire consists of two parts. The first part contains questions about demographic and management factors. The second part about personality traits was adapted from a previously validated questionnaire by Momozawa et al. [45]. In addition to the English version, the German and French translations of the questionnaire are included, as they were used in the study. (PDF)