Applied Animal Behaviour Science 254 (2022) 105707 Available online 14 July 2022 0168-1591/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Acute changes in oxytocin predict behavioral responses to foundation training in horses Taru Niittynen a, Veera Riihonen b, Liza R. Moscovice c,*,1, Sonja E. Koski a,b,**,1 a University of Helsinki, Organismal and evolutionary biology research programme Biocentre 3, Viikinkaari 1, 00014 University of Helsinki, Finland b University of Turku, Department of Biology, 20014 University of Turku, Finland c Research Institute for Farm Animal Biology, Institute of Behavioral Physiology, Wilhelm-Stahl-Allee 2, 18196 Dummerstorf, Germany A R T I C L E I N F O Keywords: Horse Training Oxytocin Cortisol Behavior Stress A B S T R A C T Ensuring horse welfare is a central aim in equestrian activities. Training is an important context for welfare, as horses form long-lasting representations of people and actions at a young age. However, only a few studies have addressed horses’ emotional responses during early training with humans. In this study, we followed N ˆ 19 young horses, including naïve yearlings and more experienced two- to three-year-olds, through five foundation training sessions over nine months. Our goal was to combine physiological and behavioral measures to assess emotional responses to early foundation training. Specifically, we measured salivary oxytocin (sOXT) in N ˆ 100 samples and salivary cortisol (sCORT) in N ˆ 96 samples before and after training sessions. We also recoded behavioral responses during training. Changes in sOXT during training predicted individual variation in behavioral responses: Horses who showed more affiliative human-directed behaviors during training had in- creases in sOXT, while horses who showed more behavioral indicators of discomfort during training had de- creases in sOXT. Salivary cortisol was not related to individual behavioral responses, but experienced horses had lower sCORT concentrations both before and after training, and all horses showed decreases in sCORT and in behaviors indicative of fear or discomfort as training progressed. In addition, sCORT increased during longer training sessions, consistent with the established role of cortisol in responding to physical stressors. We conclude that individual variation in positive or negative behavioral responses to foundation training corresponds with acute changes in sOXT concentrations in young horses, suggesting that sOXT may be useful as a non-invasive indicator of emotional responses in young horses. 1. Introduction Horses used in sports or recreational activities require systematic training to be safe to handle and to learn the various tasks required of them. Horses are typically subjected to their task specific training from three years of age onwards, preceded by foundation training in which horses learn basic responses before being ridden (McGreevy et al., 2018). Young horses may experience considerable stress during training, as measured by changes in behavior and physiology (e.g., Schmidt et al., 2010a; Kędzierski et al., 2012). There is also concern that some training and restraint practices may cause physical injuries and discomfort (McLean and McGreevy, 2010). Therefore, increasing attention is being given to training practices and their effects on horses’ emotional well-being (McGreevy and McLean, 2007; King et al., 2018; ISES, 2018). While emotions cannot be directly assessed in animals, there have been major advances in recent years in methods for indirect assessment of emotions, based on identifying short-term changes in behavior, physi- ology and/or cognition in response to biologically-relevant stimuli that help individuals to avoid harmful situations and to seek out positive and rewarding experiences (reviewed in: Mendl et al., 2010; Paul and Mendl, 2018). As training lays the base for the horses’ experiential knowledge of interactions with people, promoting positive emotions and reducing negative emotions during training is highly relevant (Sankey et al., * Corresponding author. ** Corresponding author at: University of Helsinki, Organismal and evolutionary biology research programme Biocentre 3, Viikinkaari 1, 00014 University of Helsinki, Finland. E-mail addresses: moscovice@fbn-dummerstorf.de (L.R. Moscovice), sonja.koski@helsinki.fi (S.E. Koski). 1 Shared last authorship due to equal contributions. Contents lists available at ScienceDirect Applied Animal Behaviour Science journal homepage: www.elsevier.com/locate/applanim https://doi.org/10.1016/j.applanim.2022.105707 Received 26 April 2022; Received in revised form 9 July 2022; Accepted 12 July 2022 Applied Animal Behaviour Science 254 (2022) 105707 2 2010a; 2010b). However, it is still relatively unclear how various training practices influence horses’ emotions and welfare, and how this is related to horses’ age and previous experience with humans. Handling at a very young age has been proposed to affect horses’ later emotional expression and trainability. Neonatal handling may decrease emotion- ality and facilitate learning (Heird et al., 1986; Henry et al., 2009), although the effects may be only temporary (Lansade et al., 2005), and may result in too-familiar attitudes towards humans, which may in turn cause a lack of motivation later in life (Pereira-Figueiredo et al., 2017). Some studies show that “sympathetic” (or: “freestyle”) training, which proceeds stepwise and is adjusted to individual progress, is less stressful than standard training in 2- to 3-year-olds (Visser et al., 2009; Kędzierski et al., 2012) and results in improved relationships with people (Fureix et al., 2009). Yet, the classical European training program has been re- ported to cause only mild elevation stress in 3-year-olds, as measured by cortisol and heart rate, with the exception of first mounting, which is a significant stressor (Schmidt et al., 2010a). There is also debate over the application of “natural horsemanship” practices, which aim to adopt horses’ natural social behavior to the context of training with humans (McGreevy et al., 2009). Some studies suggest that application of natural horsemanship methods results in accelerated learning and lower heart rates than conventional training (Fowler et al., 2012), but others have argued that these methods increase stress and fear and involve erro- neous assumptions about dominance (Fenner et al., 2018; McGreevy et al., 2009). It has also been suggested that ultimately, horses’ emotional responses to the applied training depend more on the trainer’s skills than on the method as such (Rozempolska-Rucińska et al., 2013). In addition to the aforementioned forms of handling and training, horses typically go through foundation training that takes place after weaning, prior to the initiation of formal training to learn more specific skills. The timing and methods of foundation training are less formalized and less often studied, but typically they involve a mixture of sensitizing and desensitizing practices (McGreevy and McLean, 2007). The skills being trained include basic handling, standing, forward walking, being led and allowing hoof care and grooming, and possibly also more discipline-specific aspects. However, only a few studies have assessed emotional responses of horses under 3 years of age in foundation training. Thus far, studies have shown that housing and the use of bit affect emotional responses to training in young horses. Pasture-housed yearling Arabians showed calmer responses to foundation training compared to stalled yearlings (Rivera et al., 2002); group-housed foals learned faster than single-housed foals in foundation training from 6 months to 2 years of age (Søndergaard and Ladewig, 2004); and bitted two-year-olds exhibited more distress behaviors than those wearing an unbitted bridle (Quick and Warren-Smith, 2009). However, we know much less about the factors that influence positive emotional responses during early training practices, a pattern which is also evident in applied animal science research more generally (Kremer et al., 2020). Identifying behavioral and physiological correlates that are associ- ated with positive emotional responses is critical for promoting positive welfare in horses during training, and in domesticated animals more generally (Boissy et al., 2007). Behaviorally, positive emotions in a horse are recognized by relaxed eyes, extended upper lip, and backward pointing, unpinned ears in the nose line (Lansade et al., 2018). Positive behavior towards humans includes approaching, nudging, and nibbling (Sankey et al., 2010b). Regarding physiological correlates, there is increasing interest in the potential for increases in the neuropeptide hormone oxytocin to serve as an indicator of positive valence (Rault et al., 2017). Oxytocin is synthesized in the hypothalamus and can be released centrally and peripherally, in a coordinated or independent manner, in response to species-specific salient social stimuli (reviewed in Valstad et al., 2017). In domesticated animals, affiliative interactions with humans have been linked to increases in oxytocin concentrations either centrally (pigs: Rault, 2016) or peripherally (dogs: Romero et al., 2014; MacLean et al., 2017). The few prior studies assessing oxytocin in horses in relation to behavior have relied on blood sampling. Plasma oxytocin levels were positively associated with docility and friendliness and tended towards a negative association with fearfulness in adult horses (Lee and Yoon, 2021; Kim et al., 2021). Another study reported, counter-intuitively, that gentle grooming was related to decreases in plasma oxytocin concentrations in horses, while standard grooming did not influence oxytocin (Lansade et al., 2018). Finally, being involved in equine assisted therapy did not cause changes in horse plasma oxytocin concentrations (Malinowski et al., 2018). To obtain a more complete picture of emotional responses of horses during training, one should address possible negative emotions and ef- fects of stressors as well. Several behavioral indicators of acute stress have been identified in horses, such as head tossing and shaking, head upping, neck tension, tail swishing, stepping, and bucking (Konig von Borstel et al., 2017; Visser et al., 2009; Quick and Warren-Smith, 2009; Rietmann et al., 2004). Stressful situations can also provoke aggressive behaviour such as kicking and biting (Wulf et al., 2013). Physiological correlates of stress may include changes in cardiac function and/or hormone concentrations (reviewed in Konig von Borstel et al., 2017; Hall et al., 2018). Previous research has established that salivary and plasma cortisol concentrations reflect changes in heart rate in young (Schmidt et al., 2010a) and adult horses (Rietmann et al., 2004; Beck- er-Birck et al., 2013, Schmidt et al., 2010b). However, other studies comparing cortisol and behavioral indicators of stress have found somewhat contradictory results (Rivera et al., 2002; Ellis et al., 2014; Perez-Manrique et al., 2021), which may in part be due to the potential for elevations in cortisol in response to a range of arousing physical as well as psychological stimuli, regardless of valence (reviewed in Hall et al., 2018, Konig von Borstel et al., 2017). In this study, we assessed physiological and behavioral responses to foundation training in young horses. The subjects were yearlings that were completely naïve to handling, and 2- and 3-year-olds that had some, albeit still limited, previous training experience. We measured salivary concentrations of oxytocin and cortisol in horses both before and after training. Methods to measure salivary oxytocin (sOXT) are relatively new compared to the more established methods to measure salivary cortisol (sCORT), and valid concerns still exist, especially regarding the mode of transfer of oxytocin from blood to saliva (Groschl, 2009). However biological and laboratory validations of sOXT have been conducted in several species (e.g., dogs: MacLean et al., 2018; pigs: López-Arjona et al., 2020; Moscovice et al., 2022; humans: de Jong et al., 2015), and sOXT is increasingly favored as a non-invasive alternative to more established methods that rely on brain CSF or blood sampling (MacLean et al., 2019). We also measured a range of behaviors that have been linked to positive and negative emotions in other studies, as described above (Table 1). If training was rewarding, we expected horses to show increases in affiliative behavior towards their trainers, potentially linked to increases in sOXT, over the course of the training sessions. We also predicted that within a training session, sOXT would increase in horses who showed more positive behaviors towards trainers. We further predicted that as training progressed, horses would show decreases in negatively- valenced behaviors, and corresponding decreases in sCORT in later training sessions. We also expected that horses who showed more negatively-valenced behavioral responses within a training session would exhibit increases in sCORT within that session (e.g. Aurich et al., 2015). Since training sessions varied in duration and longer training sessions also required more physical activity, we controlled for any ef- fects of duration of training on sCORT in all models, as a way to account for influences of physical activity more generally on cortisol responses. Since previous studies have reported contradictory results on sex dif- ferences in responses to training (Aurich et al., 2015; Schmidt et al., 2010a), we included sex as a predictor in the models but made no directional predictions. T. Niittynen et al. Applied Animal Behaviour Science 254 (2022) 105707 3 2. Methods 2.1. Subjects and housing Subjects were one-year-old naïve (n ˆ 11) and two- to three-year old more experienced (n ˆ 8) horses (n ˆ 9 fillies, n ˆ 9 colts, n ˆ 1 gelding; Finnhorses and warmbloods, Table S1), group housed in an open shed stable at a breeding farm in Southern Finland. Two horses (FR and FU) died after the second or third training session due to an injury and illness, respectively, and one horse (FM) was dropped out of the study by the owner after the second training session. One horse (MA) failed to give the saliva samples on the first sampling round. As a result, we ob- tained only partial data from these horses. The rest of the horses participated in the full study. Most subjects had been born at the farm and all were maintained in identical conditions. Prior to weaning, the foals had lived in groups consisting of mares with dependent offspring. After weaning all subjects lived continuously in social groups of 6–14 individuals throughout the year. In the colder half of the year, from October to April, groups were housed in an open shed stable with semi- heated covered indoor spaces, peat and straw bedding and free access to the fields. Water and hay were provided ad libitum and oat and mineral supplements were given daily. From early May until late September the horses lived on large natural outdoor paddocks with streams, woodland and grassy growth, and had no visual access to neighboring groups. They did not receive any training during this time. 2.2. Training The naïve horses (yearlings) were not accustomed to handling, as it had been applied scarcely during their first year of life. They had been haltered a few times prior to the study but were not yet desensitized to a halter and had never been led. The experienced horses (two- to three- year-olds) were familiar with the halter and bridle and had practiced long reigning and lunging. They had received the year before the same training that was given to the yearlings during our study. We did not enforce a controlled training regime, as we aimed to study early training as it typically proceeds on this farm. The benefit of this approach is that we could assess the effects of the most typical practice, rather than enforcing a training system that is not typically applied. The trainers were n ˆ 12 experienced farm staff and other in- dividuals who regularly trained young and adult horses. All horses were trained by several (2 4) trainers, often within the same training session. There was no systematic rotation of the trainers or horses. Although causing variability, this also minimized possible influences of differ- ences among the trainers on the horses’ behavior or hormone responses. Trainers were not aware of the study hypotheses or the ethogram that was used. The training progressed gradually and included a mixture of stan- dard European foundation training and more individually paced, sym- pathetic/freestyle training (sensu Visser et al., 2009 and Kędzierski et al., 2012), which was consistent across all subjects. It relied on standard negative reinforcement of pressure-release but included also frequent positive reinforcement with verbal praise and gentle scratching as rewards. In negative reinforcement, the trainer removes an outcome that the horse wants to avoid to increase the occurrence of a desired behavior. In the pressure-release technique the negative outcome is pressure applied by the trainer and, upon the correct response, released. Positive reinforcement, in turn, means addition of an outcome the horse values to increase the occurrence of a desired behavior (McGreevy et al., 2018). The general training regime was the same across all horses within each age group: the naïve horses (yearlings) received training on accepting a lead rope, standing, walking with a trainer and, on the latter three training sessions, trotting, long-reining, bit in mouth, and carriage training. The experienced horses (two- and three-year-olds) received lunging, riding and carriage training, which often lasted for longer and required more physical exertion. The pace of training followed the idea of sympathetic training in that the trainers accounted for the horses’ individual skill and confidence at each phase when determining the progression of the training. There were therefore differences among the subjects in the number of training sessions received and in the particular tasks trained within a session, which unfortunately could not be controlled for. The subjects received 11–17 training sessions applied in two periods in 2020: Period 1 from March– April and Period 2 from October– December. Between these periods the horses had a five-month-long summer pasture season during which no handling or training was applied. Two authors (TN and VR) video-recorded five training sessions of each subject: two in Period 1 and three in Period 2 (Fig. 1). The fifth recorded training session was timed at the point when the subject was able to complete the required tasks. Between these five sessions, the horses received some additional training sessions (1 3) that were not recorded. Therefore, while we cannot assess the precise learning prog- ress with recorded data from each training session, we can address the general behavioral and physiological patterns over the course of training, as well as changes in individual behavior and hormones within each of the five recorded training sessions. The sessions were recorded in full length and varied in duration from ca. 5–30 min (median ˆ 11 min). When the horses were trained in a manege or arena, the researcher recorded behavior from a fixed spot; when training occurred outside, footage was taken while walking beside or behind the trainer and the horse. Table 1 Ethogram used for behavioral analyses during horse training sessions. Grouping for composite measures Behaviour Coded as Description Explosive Jump event Jumping in any direction3 Buck event Hind legs tossed up at the same time3,4 Lunge event Sudden acceleration forward or sideways3 Overrun event Pushing past the trainer with physical contact3 Fearful Rearing event Front legs are lifted from the ground, head up3 Fearful snort event Snorting with tense position5 Freeze state Afraid to move forward, tense position5 Startle event Sudden tension of muscles, possible flight5 Affiliative contact Affection event Touching trainer with muzzle2 Exploration event Touching trainer or object with lips and/or teeth2 Nibble event Fumbling the trainer with lips and pushing with muzzle6 Not grouped Head toss event Tossing head sideways or upwards2,4 Head up state Head tense and higher than withers1,3,4 1Visser et al., 2009; 2Wulf et al., 2013; 3Rietmann et al., 2004; 4Quick & Warren- Smith, 2009; 5Lansade et al., 2008; 6Sankey et al., 2010a Fig. 1. The schedule of the video-recorded training sessions and associated saliva sampling. Between each recorded training session, the subjects received 1–3 unrecorded training sessions. T. Niittynen et al. Applied Animal Behaviour Science 254 (2022) 105707 4 2.3. Hormone sampling and measurement Saliva samples were collected from subjects immediately before and after the first, third and fifth recorded training session (Fig. 1). The goals were to assess short-term changes in sCORT and sOXT concentrations during the training, and to gain insights into general patterns of hor- mone reactivity across training sessions in more and less experienced horses. By using a within-subject sampling paradigm, we minimized effects of variation in timing of sampling, individual differences, and any uncontrollable prior events on hormone concentrations (Bohak et al., 2013). In addition, all samples were collected by one researcher (TN), who also spent time habituating the horses to her presence prior to collecting samples for the study. For the pre-training sampling, the horses were walked into a tradi- tional box stable from the open shed stable in pairs or small groups where they were allowed time to adjust. When calm, horses were placed in single-horse stalls with auditory and visual, but not physical contact with other horses. Horses did not receive food during the adjustment time or training, and no food was available during sampling, to mini- mize any potential interference with sampling. Prior to collecting the pre-training saliva sample, horses were kept in the stalls for 15 min, to ensure that no food had been consumed for > 15 min before sampling. Sampling was identical for all subjects in all sampling rounds with one exception: the first pre-training sample of yearlings was taken while they were still in the shed stable, as walking them to the unfamiliar box-stable might have been stressful. After the pre-training sample was taken, the horse was tacked and taken out for the training session. Training ended when the horse was returned to the box stable, at which point the post-training sample was taken. The saliva sampling was conducted without restraining the horses and was well accepted by all horses, who did not show any signs of distress during sampling (see example video in Supplementary Mate- rials). We recorded the time of the sampling and the duration of the training. If the training lasted for less than 10 min, we waited until 15 min has passed from the onset of the training to ensure sufficient time for excretion of hormones into saliva. Previous research indicates that the release rate of cortisol into saliva depends on the duration of exercise and is detectable within 15 min from the onset of a stressor, with the peak occurring at 15–30 min following the stressor (Kędzierski et al., 2013; Peeters et al., 2011). In dogs (MacLean et al., 2017) and pigs (Lürzel et al., 2020), increases in sOXT are observed within 10–15 min following affiliative interactions with humans. Saliva was collected using synthetic SalivaBio childrens swabs (Salimetrics©, PA, USA). The swabs are designed to collect passive drool from humans and have been used in validations of salivary oxytocin in a range of animal species (e.g. dogs: MacLean et al., 2018, pigs, cattle and goats: Lürzel et al., 2020). The swab was taped onto a plastic rod, inserted into the mouth of the horse and placed gently beside or under the tongue for 1–2 min until saturated. Saturated swabs were immedi- ately placed into a polypropylene tube and stored in dry ice until transfer to a 80 C freezer. Between one to six months after collection, samples were shipped on dry ice to the Research Institute for Farm Animal Biology (FBN, Dummerstorf, DE), where they were analyzed in the Institute of Behavioral Physiology. Samples were thawed and centri- fuged, and the eluent was aliquoted into eppendorf tubes and stored at 80 C until analysis. Cortisol was run in a commercially available enzyme-immunoassay (Demeditec Diagnostics®, GmbH, Kiel, Ger- many). The standard curve ranges from 0.1 to 30 ng mL-1 and assay sensitivity is 0.019 ng mL-1. After thawing, samples were spun at 2500 g for 5 min, diluted 1:1 with assay buffer and 50 µl of supernatant were run in duplicate. We performed analytical validations of parallelism, accuracy/recovery and reproducibility. To test for parallelism, we esti- mated and compared the slopes of the fitted lines between the cortisol standard curve and a serial dilution of pooled horse saliva, using the ‘lstrends’ and ‘pairs’ functions in the package ‘lsmeans’ (Lenth, 2016). The test revealed no differences in the slopes (t-test, n ˆ 5–6, df ˆ 7, t ratio ˆ 1.159, p ˆ 0.285, see Supplementary Fig. S1a). Recovery of sCORT in three different pooled samples of known concentrations, each spiked with cortisol standards at two different concentrations (1.7 ng mL-1 and 30 ng mL-1), was 106.4 (  10.5)% (n ˆ 6 spiked samples). To determine reproducibility, we measured intra- and inter-assay correlations of variation (cvs) in low (2 ng mL-1) and high (5 ng mL-1) pooled horse saliva samples that were stored in multiple aliquots and run on each plate in replicates. The intra-assay cvs for low- and high-concentration controls were 4.2% and 3.8% respectively (n ˆ 10 samples). The inter-assay cvs for low- and high-concentration controls were 13.5% and 11.1% (n ˆ 6 samples). Prior to sOXT measurement, we performed a solid phase extraction (SPE) following a protocol modified from Lürzel and colleagues (2020). Samples were thawed on ice and centrifuged at 4 C (5 min, 2,000g). We mixed 250 µl of saliva with equal parts 0.1% trifluoroacetic acid (TFA)/ water, vortexed, and then centrifuged at 4 C (15 min, 17,000g). We loaded the supernatant on 1 mL, 30 mg sorbent, HLB cartridges (Oasis Prime®, cat no. 186008055, Waters Corporation, MA, USA), primed with 1 mL 99% acetonitrile (ACN), followed by 3 mL 0.1% TFA/water. Samples were washed with 3 mL 0.1% TFA/water and eluted with 3 mL 36% ACN/ 0.1% TFA. Eluents were evaporated using a Vacuum concentrator (SpeedVac® SC210, Thermo Fisher Scientific, MA, USA), and frozen at 80 C until measurement. We measured sOXT using a commercially available enzyme immuno-assay (Cayman Chemical®, kit no. Cay500440, MI, USA) that has been validated for sOXT in several species (dogs: MacLean et al., 2018, farm animals: Lürzel et al., 2020). The standard curve ranges from 5.9 to 750 pg mL-1 and assay sensitivity is 20 pg mL-1. Prior to measurement, we reconstituted samples in 250 µl assay buffer, and then followed the kit instructions. We performed analytical validations to confirm that the assay measures the majority of oxytocin present and is suitable for measuring oxytocin in horse saliva. Tests of parallelism revealed no differences in the slopes between the standard curve and a serial dilution of pooled horse saliva (t-test, n ˆ 4–5, df ˆ 5, t ratio ˆ 0.649, p ˆ 0.545, see Supplementary Fig. S1b). Recovery of sOXT in three different pooled samples of known concen- trations, each spiked with oxytocin standard at two different concen- trations (375 pg mL-1, 187.5 pg mL-1) was 99.9  6.9% (n ˆ 6 spiked samples). To determine reproducibility, we measured intra- and inter-assay cvs of pooled low (30 pg mL-1) and high (150 pg mL-1) horse saliva samples that were stored in multiple aliquots, extracted along with test samples and run in replicates on plates. The intra-assay cvs for low- and high-concentration controls were 6.2% and 5.4% (n ˆ 11 samples) and the inter-assay cvs were 14.6% and 12.8% (n ˆ 6 samples). 2.4. Behavior coding The coding of videos was done with Boris (version 7.10.2). We coded several behaviors from the training session videos as indicators of pos- itive (affiliative contact) and negative (explosive, fearful, head toss, head up) affective responses based on earlier research (Table 1). When the horse was not completely visible, no video was scored and we sub- tracted the not-visible time from the total duration. Behaviors that were functionally similar based on earlier behavioral studies (references in Table 1) were combined into composite scores by summing the event behaviors and, for duration behaviors, first scaling and then summing them. 2.5. Statistical analyses Statistical analyses were performed in R (v 4.0.5 R Core Team, 2021). We tested for repeated measures correlations between each horses’ cortisol and oxytocin concentrations either before or after training using the package ‘rmcorr’ (Bakdash and Marusich, 2020). We built linear mixed models (LMMs) and generalized linear mixed models (GLMMs) using the package ‘lme4’ (Bates et al., 2015). We first tested T. Niittynen et al. Applied Animal Behaviour Science 254 (2022) 105707 5 whether changes in sOXT or sCORT were related to horses’ experience level or their progression with training. Since we were particularly interested in the effects of training on changes in hormones, we included the sample context (pre- or post- training) as an additional predictor, and tested for a three-way interaction of sample context, experience and training session, since we expected that more experienced horses would acclimate to repeated training more rapidly, as measured by decreased sCORT responses and/or increases in sOXT responses to later stages of training. We also tested whether changes in hormones from pre- to post-training were influenced by the duration of training or subject sex. We included the time of each sample collection as an additional control predictor. We then tested whether changes in behavior during training were related to each subject’s characteristics, particularly their level of experience and their training progression across the five training ses- sions, as well as an interaction between experience and training session. We tested whether the two uncategorized behaviors, head up and head toss, were correlated. We treated rare behavioral categories (explosive and fear behaviors) as binomial data and modeled their likelihood of occurrence using GLMMs with a binomial error distribution. For the other behaviors (affiliative contact and head-toss), we used LMMs with gaussian error distributions to model their frequencies of occurrence. For each behavioral model, we included subject sex as a control pre- dictor, and we included the duration of training as an offset term. We also included random intercepts of subject. Finally, we tested whether changes in hormone concentrations dur- ing each training session were predicted by the occurrence (for ex- ploding and fear behaviors) or the frequency (for affiliative contact and head-toss) of distinct behavioral indicators, to provide further insights into the usefulness of using sOXT and sCORT as indicators of emotional responses in horses. To test for a relationship between specific behaviors and changes in hormone concentrations, we tested whether hormone concentrations were influenced by an interaction between each behav- ioral measure and the hormone sampling context (pre- or post-training). We included the time of sample collection as a control predictor, and the duration of training as an offset-term. For all models predicting hormone concentrations, we included random intercepts for subject and for the paired hormone samples from each subject. To avoid over-fitting the data, and to keep the type I error rate at a nominal 5% (Schielzeth and Forstmeier, 2009), we also included all theoretically identifiable random slopes (see Supplementary Materials). We log-transformed hormone concentrations to approximate normal distributions, and we z-transformed continuous predictors, to facilitate interpretability of the results. We confirmed that models were stable by comparing the model estimates derived from the full data set with those obtained from a model with each subject excluded one by one. We tested for variance inflation using the ‘vif’ function in the package ‘car’ (Fox and Weisberg, 2019) and found no evidence for collinearity of the predictors (VIFs < 1.44). For the two binomial models, dispersion parameters were be- tween 0.56 and 0.88, indicating no evidence of over-dispersion. We used likelihood ratio tests (Dobson, 2002) to compare each model to a reduced model excluding the test predictors, and present results of models that differed significantly from the reduced model. We tested for significance levels of the fixed effect predictors using t-tests with the Satterthwaite approximation (Luke, 2017) in the function lmer of the package lmerTest (v 3.1–3, Kuznetsova et al., 2017) and a model fitted with restricted maximum likelihood. To determine overall effects of experience level and training session, we further compared the full model with a reduced model excluding the specific factor of interest. Unless otherwise noted, results are presented as mean (  SD) and a P-value < 0.05 was considered significant. 2.6. Ethics The study was evaluated and granted permission by the Board of Ethical Evaluation of the University of Helsinki (decision 5/2020) and complied with the ethical guidelines of the International Society for Applied Ethology (Sherwin et al., 2003). All horse owners gave informed consent and were free to withdraw from the study at any time. All data was held according to the EU General Data Protection Regulation Act 12–14 (2016/679). The individual behavioral and hormonal means are provided in the supplementary materials; full raw data is available upon request. 3. Results 3.1. Changes in hormones related to experience and training progression Mean sOXT concentrations were similar before and after training sessions (117.33  57.25 pg mL-1 vs. 111.06  59.93 pg mL-1; see Sup- plementary Materials Table S2). The full model predicting changes in sOXT during training differed significantly from a reduced model with horse experience and training session excluded (likelihood ratio test, chi-square ˆ 19.85, df ˆ 10, p ˆ 0.03). The interaction terms were not significant and were removed from the model. In the final model, there was a significant effect of training session on overall sOXT before and after training (F2,28 ˆ 7.52, p ˆ 0.002), due to peaks in sOXT in expe- rienced and naïve horses mid-way through training (see Table 2 and Fig. 2). In addition, males had lower sOXT concentrations than females (F1,64 ˆ 12.45, p < 0.001, see Table 2). Horse experience level had no effect on sOXT concentrations (see Table 2). Mean sCORT concentrations were 2.54 (  1.40) ng mL-1 before and 3.30 (  2.62) ng mL-1 after training (see Supplementary Materials, Table S2). There was no correlation between each subject’s sCORT and sOXT concentrations in pre- (rrm ˆ 0.06, df ˆ 27, p ˆ 0.75) or post- training (rrm ˆ 0.07, df ˆ 28, p ˆ 0.73) samples. The full sCORT model differed from a reduced model excluding the effects of experience and training session (full-null model comparisons with likelihood ratio tests, chi-square ˆ 18.93, df ˆ 10, p ˆ 0.041). The three-way and several two-way interactions were not significant and were removed from the model. In the final model, the only significant predictor of changes in sCORT during training was training duration. Horses who had longer training sessions also had greater increases in sCORT from pre- to post-training samples (F1,52 ˆ 13.46, p < 0.001, see Table 3 and Fig. 3). In addition, experienced horses had lower sCORT concentrations overall (both before and after training, F1,22 ˆ 8.39, p ˆ 0.008, see Table 3 and Fig. 4). There was also a significant main effect of training session on overall sCORT concentrations (F2,23 ˆ 4.27, p ˆ 0.03), mainly due to lower concentrations on the last training session, for both experienced and naïve horses (see Table 3 and Fig. 4). 3.2. Behavioral responses related to experience and training progression Over the course of the five training sessions, we analyzed the expression of three functional behavioral categories referring to Affili- ative contact, Fear and Explosive behavior (see Table 1). Two additional uncategorized behaviors, head toss and head up, were positively Table 2 Model estimates of predictors influencing horse salivary OXT concentrations (log transformed) during training. Significant predictors are indicated in bold. Predictor Estimate Std. Error t value p value (Intercept) 4.71 0.09 experiencedy -0.00 0.10 -0.04 0.970 training (3)z 0.31 0.11 2.75 0.010 training (5)z 0.07 0.10 0.66 0.520 sex (male)x -0.26 0.07 -3.53 < 0.001 sample context (post-training) -0.07 0.06 -1.13 0.270 log training duration 0.02 0.06 0.37 0.720 time of sample collection -0.01 0.03 -0.26 0.800 y,z,x In comparison with reference level: ynaïve horses, ztraining session 1, xfemales T. Niittynen et al. Applied Animal Behaviour Science 254 (2022) 105707 6 correlated (Pearson, r ˆ 0.47, t ˆ 4.92, df ˆ 85, p < 0.001) so we excluded head up from further analyses and analyzed head toss as an indicator of mild discomfort or stress (Wulf et al., 2013; Quick and Warren-Smith, 2009). The full models containing experience level and test session explained more of the variation in each behavioral measure than reduced models that did not include the test predictors (full-null model comparisons with likelihood ratio tests, affiliative contact: chi-sq ˆ 39.15, df ˆ 9, p < 0.001, head toss: chi-sq ˆ 46.42, df ˆ 9, p < 0.001, fear: chi-sq ˆ 32.61, df ˆ 9, p < 0.001, explosive behavior: chi-sq ˆ 23.61, df ˆ 9, p < 0.001). However, only head toss was influenced by an interaction between experience and session (F4,69 ˆ 6.58, p < 0.001, see Table 4a), such that experienced horses showed consistent decreases in head tossing over sessions, while inexperienced horses did not (see Fig. 5). Regardless of experience, horses showed decreases in fearful behavior (F9 ˆ 29.94, p < 0.001, see Table 4b) and explosive behavior (F4 ˆ 22.60, p < 0.001, see Table 4c) across sessions. However, affili- ative contact also decreased over training sessions for all horses (F4,69 ˆ 8.45, p < 0.001, see Table 4d). There were no sex differences in any of the behavioral responses (see Table 4a-d), although males tended to show more explosive behavior than females (est  SEˆ 1.46  0.80, p ˆ 0.07, see Table 4c). Fig. 2. Salivary OXT concentrations in naïve and experienced horses as training progressed. Regardless of experience, horses had higher sOXT concentrations mid-way through training, during Training session 3. For full model results see Table 2. Table 3 Model estimates of predictors influencing sCORT concentrations (log trans- formed) in horses during training sessions. Significant main effects are indicated in bold. Predictor Estimate Std. Error t value p value (Intercept) 1.17 0.14 experiencedy -0.37 0.13 -2.90 0.008 training (3)z 0.04 0.13 0.29 0.776 training (5)z -0.27 0.14 -1.97 0.068 sex (male)x -0.19 0.11 -1.63 0.111 sample context (post-training)¶ 0.19 0.09 log training duration¶ -0.05 0.07 time of sample collection -0.08 0.06 -1.50 0.151 context(post-training):log training duration 0.29 0.08 3.67 0.001 y,z,xIn comparison with reference level: ynaïve horses, ztraining session 1, xfemales ¶Significance tests of main effects not indicated due to limited interpretability Fig. 3. Changes in sCORT concentrations during training in horses who received shorter or longer training sessions than on average. Note that training duration was analyzed as actual duration in the models; the categorization is included here for illustration purposes only. Fig. 4. Salivary cortisol concentrations in naïve and experienced horses as training progressed. Experience and training had independent effects on sCORT concentrations. For full model results see Table 3. Table 4a Model estimates of predictors influencing head tossing during training. Signifi- cant main effects are indicated in bold. Predictor Estimate Std. Error t value p value (Intercept) 4.37 0.44 experiencedy 1.22 0.59 training (2)y -1.69 0.46 training (3)y -0.34 0.49 training (4)y -1.51 0.51 training (5)y -0.13 0.51 sex (male)z 0.50 0.40 1.24 0.23 experienced:training (2) 0.04 0.71 0.05 0.96 experienced:training (3) -2.06 0.73 -2.82 0.01 experienced:training (4) -1.32 0.74 -1.77 0.08 experienced:training (5) -3.11 0.74 -4.18 < 0.001 ySignificance tests of main effects not indicated due to limited interpretability zIn comparison with females Fig. 5. Mean (  SE) number of head tosses per minute by experienced and naïve horses across five training sessions. For full model results refer to Table 4a. T. Niittynen et al. Applied Animal Behaviour Science 254 (2022) 105707 7 3.3. Influence of behavioral responses to training on changes in hormones The full model relating changes in sOXT concentrations during training to behavioral measures differed from the reduced model with the behav- iors excluded (likelihood ratio test, chi-square ˆ 28.02, df ˆ 9, p < 0.001). Several behavioral measures contributed significantly to explaining acute changes in sOXT during training. As predicted, horses who had more affiliative contact with trainers during a training session had increases in sOXT concentrations during training (est  SE ˆ 0.10  0.05, p ˆ 0.039, see Table 5). Interestingly, acute changes in sOXT were also strongly inversely related to behaviors associated with negative affect. Specifically, horses who showed explosive behavior during training had decreases in sOXT from pre- to post-training samples (est  SE ˆ 0.30  0.10, p ˆ 0.003, see Table 5 and Fig. 6). In addition, horses who showed fear behaviors during training tended to have decreases in sOXT concentrations after training (est  SE ˆ 0.19  0.10, p ˆ 0.06, see Table 5). The full model relating changes in sCORT concentrations during training sessions to behavioral measures differed from a reduced model with the behaviors excluded (full-null model comparison, likelihood ratio test, chi-square ˆ 17.39, df ˆ 9, p ˆ 0.043). Unlike the sOXT re- sults, there were no significant effects of any behavioral measures on changes in sCORT during training, although there was a tendency for horses who exhibited more head tossing during a training session to have increased post-training sCORT (est  SE ˆ 0.18  0.09, p ˆ 0.057, see Supplementary Table S3). 4. Discussion We assessed behavioral and non-invasive physiological indicators of emotional responses in young horses during opportunistically recorded foundation training. We specifically focused on sOXT, which to our knowledge has not been measured previously in horses, as a potential indicator of positive emotions. In addition, we measured sCORT and assessed the relationships between these hormones, horse experience and training progression, and behavioral expression during training. We found that acute changes in sOXT during training were predicted by individual behavioral responses, although effects were strongest for behavioral indicators of negative emotions. Specifically, horses that showed explosive behavior during training had decreases in sOXT. Explosive behaviors indicate horse’s discomfort and attempts to avoid a situation (Rietmann et al., 2004; Quick and Warren-Smith, 2009). Fear behaviors tended to show the same pattern. These results suggest that sOXT decreases when horses experience mental or physical discomfort. In addition, horses who had more positive contact with the trainer exhibited increases in sOXT during the training. These results are consistent with evidence that horse friendliness and trainability posi- tively predict plasma oxytocin concentrations (Lee and Yoon, 2021; Kim et al., 2021). In addition, results support evidence for increases in sOXT associated with affiliative human interactions in domesticated animals more generally (pigs: Lürzel et al., 2020; dogs: MacLean et al., 2017). The finding that positive contact behaviors did not increase over the course of training is not necessarily inconsistent with this, given that the training became progressively more demanding over the course of training, which may have reduced possibilities for affiliative behaviors. Although this is the first study to assess sOXT in horses, results are Table 4b Model estimates of predictors influencing fear behaviors during training. Sig- nificant main effects are indicated in bold. Predictor Estimate Std. Error z value p value (Intercept) 2.26 1.19 experiencedy 1.91 1.17 1.63 0.1 training (2)z -2.91 1.13 -2.56 0.01 training (3)z -4.65 1.15 -4.05 < 0.001 training (4)z -4.26 1.14 -3.73 < 0.001 training (5)z -5.15 1.17 -4.41 < 0.001 sex (male)x 1.23 1.16 1.06 0.29 y,z,x In comparison with reference level: ynaïve horses, ztraining session 1, xfemales Table 4c Model estimates of predictors influencing explosive behaviors during training. Significant main effects are indicated in bold. Predictor Estimate Std. Error z value p value (Intercept) 0.50 0.80 experiencedy 1.06 0.79 1.33 0.18 training (2)z -1.87 0.86 -2.18 0.03 training (3)z -2.35 0.82 -2.87 0.004 training (4)z -3.90 0.98 -4.00 < 0.001 training (5)z -3.07 0.87 -3.51 < 0.001 sex (male)x 1.46 0.80 1.83 0.068 y,z,x In comparison with reference level: ynaïve horses, ztraining session 1, xfemales Table 4d Model estimates of predictors influencing affiliative contact with humans during training. Significant main effects are indicated in bold. Predictor Estimate Std. Error t value p value (Intercept) 2.15 0.29 experiencedy -0.43 0.30 -1.41 0.17 training (2)z -0.25 0.21 -1.18 0.24 training (3)z -0.72 0.22 -3.22 < 0.001 training (4)z -0.96 0.23 -4.22 < 0.001 training (5)z -1.11 0.23 -4.89 < 0.001 sex (male)x 0.23 0.30 0.76 0.46 y,z,x In comparison with reference level: ynaïve horses, ztraining session 1, xfemales Table 5 Model estimates of predictors influencing changes in salivary oxytocin concen- trations (log transformed) from pre- to post-training samples. Significant main effects are indicated in bold. Predictor Estimate Std. Error t value p value (Intercept) 4.67 0.14 sample context (post-training)y 0.11 0.09 positive contacty 0.12 0.08 explode behavior (yes)y 0.50 0.17 fear behavior (yes)y -0.11 0.17 head tossy -0.35 0.09 time of sample collection 0.18 0.09 1.95 0.066 context (post-training):positive contact 0.10 0.05 2.12 0.039 context (post-training):explosive behavior (yes) -0.30 0.10 -3.09 0.003 context (post-training):fear behavior (yes) -0.19 0.10 -1.94 0.058 context (post-training):head toss 0.02 0.05 0.38 0.704 ySignificance tests of main effects not indicated due to limited interpretability. Fig. 6. Changes in sOXT concentrations in horses who exhibited explosive behavior during training, and those who did not. For full model results see Table 5. T. Niittynen et al. Applied Animal Behaviour Science 254 (2022) 105707 8 supportive of further research into the use of sOXT as a non-invasive indicator of short-term changes in emotions in horses. Both pre- and post-training levels of sOXT were higher on the third training session in comparison with the other sessions. However, we have no obvious, biologically meaningful explanation for this statisti- cally significant finding. The third recorded training session took place within days of bringing the horses back to the open-shed stable after the long summer pasture season. It is possible that some aspect of returning to the stable and being housed in different social groups after the sum- mer season may have influenced sOXT concentrations. It would be interesting to collect a larger number of sOXT samples from these horses under varied housing conditions and social contexts, to examine whether specific differences in the husbandry conditions between sea- sons may be reflected in sOXT concentrations. In line with our general predictions that young horses would adjust to training, all horses showed decreases in behavioral indicators of fear, explosive behaviors and head-tossing from their first to last training sessions during the study period. Our interpretation is supported by the decrease in overall sCORT across training sessions in both naïve and experienced horses. Further, in support of our predictions, sCORT was higher in samples taken both before and after training in naïve yearlings compared with the more experienced two- and three-year-olds. This suggests that yearlings may have had increased arousal already when being brought for tacking to the box-stable, which was an unfamiliar place to them, and that their sCORT may have remained high during training, in comparison with more experienced horses, due to greater psychological stress. We also found that sCORT increased, as predicted, during longer training sessions, which usually included more strenuous exercise. This suggests that short-term changes in peripheral cortisol during training may have been triggered primarily by physical arousal and exercise (Ferlazzo et al., 2020), while heightened sCORT both before and after training may reflect greater psychological stress. Indeed, the behavioral indicators of negative emotions within each training session did not directly predict changes in sCORT during the session, which lends further support for short-term increases in cortisol reflecting energetic, rather than psychological, stress. Our results emphasize the importance of controlling for differences in energetic activity when attempting to investigate the link between peripheral cortisol and psychological stress. An important caveat of the study is that the training regime was not standardized to follow a predetermined pace and order, and that mul- tiple trainers were often involved in a training session. A more strictly standardized system would have allowed us to specify the exact effects of each training task as well as trainer identity on hormones and behavior. However, as the training was paced individually according to each horses’ level of development, it closely resembles the ‘sympathetic training’, which is shown to be supportive of positive emotional expe- riences in young horses (Visser et al., 2009; Kędzierski et al., 2012; Fureix et al., 2009). The training method may have contributed to the horses’ positive emotional responses and, conversely, minimal stress during the training sessions. Another caveat is that we were not able to include measures of facial expressions as indicators of emotions due to the face being out of sight during much of the video recordings. Horses have highly expressive faces (Wathan et al., 2015), and many studies on horse emotions assess facial expressions (Lansade et al., 2018; Trosch et al., 2020). Integrating measures of facial expression with other behavioral and physiological responses may provide further insights into emotional responses during training. In addition, personality is likely to have influenced the horses’ responses. Adult and young horses have stable inter-individual differ- ences in responses to novelty and threat, isolation, and humans (Per- ez-Manrique et al., 2021; Rankins and Wickens, 2020). Personality is known to influences stress responses (Hausberger et al., 2019) and learning (LeScolan et al., 1997; Lansade et al., 2017; but see Christensen et al., 2012) in horses, so individual differences probably also contrib- uted to the variability observed in our study. Future research should incorporate personality trait scores in targeted test paradigms with emotional responses to foundation training to gain further insights into the factors underlying stable inter-individual variation in responses (Duberstein and Gilkeson, 2010). 5. Conclusions Our results indicate that salivary oxytocin can be measured non- invasively in young horses and may be useful as an indicator of young horses’ acute emotional responses during foundation training. Salivary oxytocin did not reflect horses’ previous training experience, nor did it show clear patterns of changes across the training period. Rather, short- term decreases in oxytocin during training corresponded with behav- ioral indicators of negative emotional responses, while short-term in- creases in oxytocin corresponded with behavioral indicators of positive emotional responses. We also found that although early foundation training was somewhat stressful for young horses, the applied training methods facilitated positive experiences. During this study, all horses exhibited reductions in cortisol and in negative behaviors during later training sessions at the end of the nine-month training period, compared to earlier training sessions. Interestingly, acute cortisol reactivity during training was not related to horse experience level or to any of the behavioral indicators, but rather reflected differences in training dura- tion and likely energetic expenditure. The combined results suggest that sOXT and sCORT, together with appropriate behavioral measures, may be useful as non-invasive indicators of more immediate and long-term emotional responses to training in young horses. Author contributions SEK and LM conceived and funded the study, TN and VR collected the data, LM conducted laboratory analyses, LM and TN conducted statistical analyses, SEK and LM wrote the manuscript and the revision. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements We are grateful to the Kylamaki Farm and the horse owners for permission to conduct the study and the farm personnel for assisting at every stage, and to Tero Pohjolainen and Anders Ingman Stable and Irina Kokko at Kulta-ahon Ratsutila. We thank Oceane Liehrmann for assistance in data collection. We are grateful to Birgit Sobczak and Petra Müntzel of the FBN Institute for Behavioral Physiology, for their assis- tance with hormone analyses. Funding: This work was supported by Kone Foundation [to SEK, grant number 201806482] and by the FBN. Appendix A. Supporting information Supplementary data associated with this article can be found in the online version at doi:10.1016/j.applanim.2022.105707. References Aurich, J., Wulf, M., Ille, N., Erber, R., von Lewinski, M., Palme, R., Aurich, C., 2015. Effects of season, age, sex, and housing on salivary cortisol concentrations in horses. Domest. Anim. Endocrinol. 52 (1), 1–16. Bakdash J.Z., Marusich L.R. 2020. rmcorr: Repeated Measures Correlation. R package version 0.4.1. 2020. hhttps://CRAN.R-project.org/packageˆrmcorri. Bates, D., Machler, M., Bolker, B., Walker, S., 2015. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67. https://doi.org/10.18637/jss.v067.i01. Becker-Birck, M., Schmidt, A., Lasarzik, J., Aurich, J., Mostl, E., Aurich, C., 2013. Cortisol release and heart rate variability in sport horses participating in equestrian T. Niittynen et al. Applied Animal Behaviour Science 254 (2022) 105707 9 competitions. J. Vet. Behav. Clin. Appl. Res. 8, 87–94. https://doi.org/10.1016/j. jveb.2012.05.002. Bohak, Z., Szabo, F., Beckers, J.-F., Melo de Sousa, N., Kutasi, O., Nagy, K., Szenci, O., 2013. Short communication: monitoring the circadian rhythm of serum and salivary cortisol concentrations in the horse. Domest. Anim. Endocrinol 45, 38–42. Boissy, A., Veissier, I., Manteuffel, G., Langbein, J., Jensen, M.B., Moe, R.O., et al., 2007. Assessment of positive emotions in animals to improve their welfare. Physiol. Behav. 92, 375–397. https://doi.org/10.1016/j.physbeh.2007.02.003. Christensen, J.W., Ahrendt, L.P., Lintrup, R., Gaillard, C., Palme, R., Malmkvist, J., 2012. Does learning performance in horses relate to fearfulness, baseline stress hormone, and social rank. Appl. Anim. Behav. Sci. 140, 44–52. https://doi.org/10.1016/j. applanim.2012.05.003. Dobson, A.J., 2002. An Introduction to Generalized Linear Models. Chapman & Hall/ CRC, Boca Raton. Duberstein, K.J., Gilkeson, J.A., 2010. Determination of sex differences in personality and trainability of yearling horses utilizing a handler questionnaire. Appl. Anim. Behav. Sci. 128, 57–63. https://doi.org/10.1016/j.applanim.2010.09.012. Ellis, A.D., Stephenson, M., Preece, M., Harris, P., 2014. A novel approach to systematically compare behavioural patterns between and within groups of horses. Appl. Anim. Behav. Sci. 161, 60–74. https://doi.org/10.1016/j. applanim.2014.09.017. Fenner, K., Mclean, A., McGreevy, P., 2018. Cutting to the chase: how round-pen, lunging and high-speed liberty work may compromise horse welfare. J. Vet. Behav. 29, 88–94. Ferlazzo, A., Cravana, C., Fazio, E., Medica, P., 2020. The different hormonal system during exercise stress coping in horses. Vet. World 13, 847–859. https://doi.org/ 10.14202/vetworld.2020.847-859. Fowler, V., Kennedy, M., Marlin, D., 2012. A comparison of the monty roberts technique with a conventional UK technique for initial training of riding horses. Anthrozoos 25, 301–321. Fox, J., Weisberg, S., 2019. An R Companion to Applied Regression, Third ed..,. Sage, Thousand Oaks, CA. Fureix, C., Pages, M., Bon, R., Lassalle, J.-M., Kuntz, P., Gonzalez, G., 2009. A preliminary study of the effects of handling type on horses’ emotional reactivity and the human–horse relationship. Behav. Process 82, 202–210. https://doi.org/ 10.1016/j.beproc.2009.06.012. Groschl, M., 2009. The physiological role of hormones in saliva. BioEssays 31, 843–852. https://doi.org/10.1002/bies.200900013. Hall, C., Randle, H., Pearson, G., Preshaw, L., Waran, N., 2018. Assessing equine emotional state. Appl. Anim. Behav. Sci. 205, 183–193. Hausberger, M., Stomp, M., Sankey, C., Brajon, S., Lunel, C., Henry, S., 2019. Mutual interactions between cognition and welfare: the horse as an animal model. Neurosci. Biobehav Rev. 107, 540–559. Heird, J.C., Whitaker, D.D., Bell, R.W., Ramsey, C.B., Lokey, C.E., 1986. The effects of handling of different ages on the subsequent learning ability of 2.year-old horses. Appl. Animl Behav. Sci. 15, 15–25. Henry, S., Richard-Yris, M.-A., Tordjman, S., Hausberger, M., 2009. Neonatal handling affects durably bonding and social development. Plos One 4, e5216. https://doi.org/ 10.1371/journal.pone.0005216. ISES, 2018. Principles of learning theory in equitation. hwww.equitationscience.com/e quitation/principles-of-learning-theory-in-equitationi. de Jong, T.R., Menon, R., Bludau, A., Grund, T., Biermeier, V., Klampfl, S.M., 2015. Salivary oxytocin concentrations in response to running, sexual self-stimulation, breastfeeding and the TSST: the Regensburg oxytocin challenge (ROC) study. Psychoneuroendocrinology 62, 381–388. Kędzierski, W., Janczarek, I., Stachurska, A., 2012. Emotional response of naive purebred arabian colts and fillies to sympathetic and traditional training methods. J. Equine Vet. Sci. 32, 752–756. https://doi.org/10.1016/j.jevs.2012.03.002. Kim, J., Park, Y., Kim, E.J., Jung, H., Yoon, M., 2021. Relationship between oxytocin and serotonin and the fearfulness, dominance, and trainability of horses. J. Anim. Sci. Technol. 63, 453–460. King, S., Wills, L., Randle, H., 2018. Early training of foals using the Ises training principles. J. Vet. Behav. 29, 140–146. Konig von Borstel, U., Visser, E.K., Hall, C., 2017. Indicators of stress in equitation. Appl. Anim. Behav. Sci. 190, 43–56. Kremer, L., Holkenborg, S.E.J.K., Reimert, I.S., Bolhuis, J.E., Webb, L.E., 2020. The nuts and bolts of animal emotion. Neurosci. Biobehav Rev. 113, 273–286. https://doi. org/10.1016/j.neubiorev.2020.01.028. Kuznetsova, A., Brockhoff, P.B., Christensen, R.H.B., 2017. lmerTest package: tests in linear mixed effects models. J. Stat. Softw. 82, 1–26. https://doi.org/10.18637/JSS. V082.I13. Lansade, L., Bertrand, M., Bouissou, M.-F., 2005. Effects of neonatal handling on subsequent manageability, reactivity and learning ability of foals. Appl. Anim. Behav. Sci. 92, 143–158. https://doi.org/10.1016/j.applanim.2004.10.014. Lansade, L., Marchand, A.R., Coutureau, E., Balle, C., Polli, F., Calandreau, L., 2017. Personality and predisposition to form habit behaviours during instrumental conditioning in horses (Equus caballus). Plos One 12, e0171010. https://doi.org/ 10.1371/journal.pone.0171010. Lansade, L., Nowak, R., Laine, A.-L., Leterrier, C., Bonneau, C., Parias, C., Bertin, A., 2018. Facial expression and oxytocin as possible markers of positive emotions in horses. Sci. Rep. 8, 14680. https://doi.org/10.1038/s41598-018-32993-z. Lee, G., Yoon, M., 2021. Association of plasma concentrations of oxytocin, vasopressin, and serotonin with docility and friendliness of horses. Domest. Anim. Endocrin 74, 106482. Lenth, R.V., 2016. Least-squares means: the R package lsmeans, 796 J. Stat. Softw. 69 (1), 1–33. https://doi.org/10.18637/jss.v069.i01. LeScolan, N., Hausberger, M., Wolff, A., 1997. Stability over situations in temperamental traits of horses as revealed by experimental and scoring approaches. Behav. Process 41, 257–266. López-Arjona, M., Mateo, S.V., Manteca, X., Escribano, D., Cerón, J.J., Martińez- Subiela, S., 2020. Oxytocin in saliva of pigs: an assay for its measurement and changes after farrowing. Domest. Anim. Endocrinol. 70, 106384 https://doi.org/ 10.1016/j.domaniend.2019.106384. Luke, S.G., 2017. Evaluating significance in linear mixed-effects models in R. Behav. Res. Methods 49, 1494–1502. https://doi.org/10.3758/s13428-016-0809-y. Lürzel, S., Bückendorf, L., Waiblinger, S., Rault, J.L., 2020. Salivary oxytocin in pigs, cattle, and goats during positive human-animal interactions. Psychoneuroendocrinol 115, 104363. https://doi.org/10.1016/j.psyneuen.2020.104636. MacLean, E.L., Gesquiere, L.R., Gee, N.R., Levy, K., Martin, W.L., Carter, C.S., 2017. Effects of affiliative human–animal interaction on dog salivary and plasma oxytocin and vasopressin. Front Psychol. 08, 1606. MacLean, E.L., Gesquiere, L.R., Gee, N.R., Levy, K., Martin, W.L., Carter, C.S., 2018. Validation of salivary oxytocin and vasopressin as biomarkers in domestic dogs. J. Neurosci. Methods 293, 67–76. https://doi.org/10.1016/j.jneumeth.2017.08.033. MacLean, E.L., Wilson, S.R., Martin, W.L., Davis, J.M., Nazarloo, H.P., Carter, C.S., 2019. Challenges for measuring oxytocin: the blind men and the elephant. Psychoneuroendocrinology 107, 225–231. https://doi.org/10.1016/j. psyneuen.2019.05.018. Malinowski, K., Yee, C., Tevlin, J.M., Birks, E.K., Durando, M.M., Pournajafi- Nazarloo, H., Cavaiola, A.A., McKeever, K.H., 2018. The effects of equine assisted therapy on plasma cortisol and oxytocin concentrations and heart rate variability in horses and measures of symptoms of post-traumatic stress disorder in veterans. J. Equine Vet. Sci. 64, 17–26. McGreevy, P.D., McLean, A.N., 2007. Roles of learning theory and ethology in equitation. J. Vet. Behav. 2, 108–118. McGreevy, P.D., Oddie, C., Burton, F.L., McLean, A.N., 2009. The horse–human dyad: can we align horse training and handling activities with the equid social ethogram? Vet. J. 181, 12–18. McGreevy, P.D., Christensen, J.W., Konig von Borstel, U., McLean, A.N., 2018. Equitation Science, 2nd edition.,. Wiley Blackwell,. McLean, A.N., McGreevy, P.D., 2010. Horse training techniques which may defy the principals of learning theory and compromise welfare. J. Vet. Behav. 5, 187–195. Mendl, M., Burman, O.H.P., Paul, E.S., 2010. An integrative and functional framework for the study of animal emotion and mood. Proc. Roy. Soc. B 2895–2904. https://doi. org/10.1098/rspb.2010.0303. Moscovice, L.R., Gimsa, U., Otten, W., Eggert, A., 2022. Salivary cortisol, but not oxytocin, varies with social challenges in domestic pigs: implications for measuring emotions. Front. Behav. Neurosci. 16, 899397 https://doi.org/10.3389/ fnbeh.2022.899397. Paul, E.S., Mendl, M.T., 2018. Animal emotion: descriptive and prescriptive definitions and their implications for a comparative perspective. Appl. Anim. Behav. Sci. 205, 202–209. https://doi.org/10.1016/j.applanim.2018.01.008. Peeters, M., Sulon, J., Beckers, J.-F., Ledoux, D., Vandenheede, M., 2011. Comparison between blood serum and salivary cortisol concentrations in horses using an adrenocorticotropic hormone challenge. Equine Vet. J. 43, 487. Pereira-Figueiredo, I., Costa, H., Carro, J., Stilwell, G., Rosa, I., 2017. Behavioural changes induced by handling at different timeframes in Lusitano yearling horses. Appl. Anim. Behav. Sci. 196, 36–43. Perez-Manrique, L., Banszegi, O., Hudson, R., Szenczi, P., 2021. Repeatable individual differences in behaviour and physiology in juvenile horses from an early age. Appl. Anim. Behav. Sci. 235, 105227. Quick, J.S., Warren-Smith, A.K., 2009. Preliminary investigations of horses’ (Equus caballus) responses to different bridles during foundation training. J. Vet. Behav. Clin. Appl. Res 4, 169–176. Rankins, E.M., Wickens, C.L., 2020. A systematic review of equine personality. Appl. Anim. Behav. Sci. 231, 105076 https://doi.org/10.1016/j.applanim.2020.105076. Rault, J.L., 2016. Effects of positive and negative human contacts and intranasal oxytocin on cerebrospinal fluid oxytocin. Psychoneuroendocrinol 69, 60–66. Rault, J.-L., Munkhof, M. van den, Buisman-Pijlman, F.T.A., 2017. Oxytocin as an indicator of psychological and social well-being in domesticated animals: a critical review. Front Psychol. 8, 1521. Rietmann, T.R., Stuart, A.E.A., Bernasconi, P., Stauffacher, M., Auer, J.A., Weishaupt, M. A., 2004. Assessment of mental stress in warmblood horses: heart rate variability in comparison to heart rate and selected behavioural parameters. Appl. Anim. Behav. Sci. 88, 121–136. https://doi.org/10.1016/j.applanim.2004.02.016. Rivera, E., Benjamin, S., Nielsen, B., Shelle, J., Zanella, A.J., 2002. Behavioral and physiological responses of horses to initial training: the comparison between pastured versus stalled horses. Appl. Anim. Behavi Sci. 78, 235–252. https://doi. org/10.1016/s0168-1591(02)00091-6. Romero, T., Nagasawa, M., Mogi, K., Hasegawa, T., Kikusui, T., 2014. Oxytocin promotes social bonding in dogs. PNAS 111, 9085–9090. Rozempolska-Rucińska, I., Trojan, M., Kosi, E., Próchniak, T., Górecka-Bruzda, A., 2013. How “natural” training methods can affect equine mental state? A critical approach – a review. Anim. Sci. Pap. Rep. 3, 185–194. Sankey, C., Richard-Yris, M.A., Leroy, H., Henry, S., Hausberger, M., 2010a. Positive interactions lead to lasting positive memories in horses, Equus caballus. Anim. Behav. 79, 869–875. Sankey, C., Richard-Yaris, M.A., Henry, S., Fureix, C., Nassur, F., Hausberger, M., 2010b. Reinforcement as a mediator of the perception of humans by horses (Equus caballus). Anim. Cogn. 13, 753–764. T. Niittynen et al. Applied Animal Behaviour Science 254 (2022) 105707 10 Schielzeth, H., Forstmeier, W., 2009. Conclusions beyond support: overconfident estimates in mixed models. Behav. Ecol. 20, 416–420. https://doi.org/10.1093/ beheco/arn145. Schmidt, A., Aurich, J., Mostl, E., Müller, J., Aurich, C., 2010a. Changes in cortisol release and heart rate and heart rate variability during the initial training of 3-year- old sport horses. Horm. Behav. 58, 628–636. https://doi.org/10.1016/j. yhbeh.2010.06.011. Schmidt, A., Biau, S., Mostl, E., Becker-Birck, M., Morillon, B., Aurich, J., Faure, J.M., Aurich, C., 2010b. Changes in cortisol release and heart rate variability in sport horses during a long-distance road transport. Domest. Anim. Endocrinol. 38, 179–189. Sherwin, C.M., Christiansen, S.B., Duncan, I.J.H., Erhard, H.W., Lay, D.C., Mench, J.A., O’Connor, C.E., Petherick, C.J., 2003. Guidelines for the ethical use of animals in applied animal behaviour research. Appl. Anim. Behav. Sci. 81, 291–305. Søndergaard, E., Ladewig, J., 2004. Group housing exerts a positive effect on the behaviour of young horses during training. Appl. Anim. Behav. Sci. 87, 105–118. Trosch, M., Pellon, S., Cuzol, F., Parias, C., Nowak, R., Calandreau, L., Lansade, L., 2020. Horses feel emotions when they watch positive and negative horse–human interactions in a video and transpose what they saw to real life. Anim. Cogn. 23, 643–653. https://doi.org/10.1007/s10071-020-01369-0. Valstad, M., Alvares, G.A., Egknud, M., Matziorinis, A.M., Andreassen, O.A., Westlye, L. T., Quintana, D.S., 2017. The correlation between central and peripheral oxytocin concentrations: A systematic review and meta-analysis. Neurosci. Biobehav Rev. 78, 117–124. https://doi.org/10.1016/j.neubiorev.2017.04.017. Visser, E.K., VanDierendonck, M., Ellis, A.D., Rijksen, C., Reenen, C.G.V., 2009. A comparison of sympathetic and conventional training methods on responses to initial horse training. Vet. J. 181, 48–52. https://doi.org/10.1016/j. tvjl.2009.03.009. Wathan, J., Burrows, A.M., Waller, B.M., McComb, K., 2015. EquiFACS: the equine facial action coding system. Plos One 10, e0131738. https://doi.org/10.1371/journal. pone.0131738. Wulf, M., Aurich, J., May, A.-C., Aurich, C., 2013. Sex differences in the response of yearling horses to handling by unfamiliar humans. J. Vet. Behav. Clin. Appl. Res 8, 238–244. T. Niittynen et al.