Estimation of the genetic parameters for temperament in Haflinger horses based on the results of a character test using a pedigree-based relationship matrix

Abstract Literature provides indications that the personality and temperament of a horse are sufficiently genetically determined. This background was the impetus to continue collecting phenotypic information on the temperament in the South Tyrolean Haflinger horses, not only to have marketing tools but also to be able to estimate genetic parameters for a future genetic evaluation. The latter should open the possibility to include temperament traits in the breeding goal. Therefore, the aim of the present study was to consider for the very first-time phenotypic information from character testing for estimating genetic parameters for temperament in Haflinger horses. For the present study, data from routine character testing of 210 South Tyrolean Haflinger mares between 2019 and 2021 were considered. A multivariate estimation of heritability (h2) and genetic correlations (r) was carried out for the criteria ‘Interest’, ‘Activity’ and ‘Excitement’. In addition, univariate and bivariate estimation runs were also carried out. The pedigree-based heritability for the behavioural trait Interest was lowest with 0.07 and a standard error of ±0.04. For the other two criteria, Activity and Excitement, h2 was 0.19 ± 0.05 and 0.21 ± 0.04, respectively. The genetic correlation between Interest and Activity and between Interest and Excitement was in both cases negative with −0.52 and −0.61, respectively while the genetic correlation between Excitement and Activity was positive (0.99). Results of the present study highlight the potential of considering temperament traits in breeding of Haflinger horses. BULLETED HIGHLIGHTS Character testing in Haflinger horses provided phenotypic data for estimating heritability for selected temperament traits. Heritability for the temperament traits describing interest, activity, and excitement in Haflinger horses were low to moderate. In future, a larger number of tested animals is required for obtaining more robust data for genetic parameters and implementation of a routine genetic evaluation.


Introduction
The Haflinger horse breed has its origin in the Alpine region of South Tyrol, very northern part of Italy (Kofler et al. 1995).The name 'Haflinger' is of South Tyrolean origin: it was named after the small village of Hafling (Feddersen 1999).Originally the name Haflinger was used to describe hard-working, easyfooted, robust, compact, and versatile horses.On May 2, 1894, the name 'Haflinger' was officially approved as a breed name (Druml et al. 2018).The Haflinger breed has developed mainly in a geographically closed area.Thereby natural selection conditions were given (e.g.sparse feed supply, mountainous topography), which required an optimal adaptation to the Alpine environment.Only those animals, which were able to cope with such adverse circumstances, were used as working and breeding animals (Schwark and Petzold 1986).Today the Haflinger horse breed is a worldwide known and bred horse breed (Schweisgut 1995;Samor e et al. 1997;Druml et al. 2018).The breed fascinates by its special appearance and its versatile abilities.It has evolved from a carrying and pulling animal into an excellent riding horse as both the conformation and the interior have been changed over the last decades through changing breeding goals, targeted mating and partly through crossbreeding with horses with higher bloodlines, for example the Arabians (Schweisgut 1995, Samor e et al. 1997, Wulf 2007;Druml et al. 2018).Consequently, nowadays Haflinger horses are represented in almost all equestrian sports.At the same time, the demand for sociable and easy to handle horses for leisure riding has increased (Gille and Spiller 2010;K€ uhl et al. 2019).In order to meet the wishes and demands of the different riders and breeders, and the multipurpose use of the Haflinger, character testing for the South Tyrolean Haflinger population was implemented in 2019 to assess temperament and behavioural traits (Zanon et al. 2021).The latter proved to be beneficial, as a higher willingness to pay for a horse with tested temperament traits has been described in previous studies (Gille and Spiller 2010;Graf et al. 2013a;Hornauer et al. 2022).Moreover, the results of a character test enable to better interpret behavioural problems (Olsson 2010) and to make specific adjustments in the husbandry management and training program accordingly (Olsson 2010;Sackman and Houpt 2019).Furthermore, in the long term, the information generated from the character testing could be used for breeding for the personality of Haflinger horses.From the literature, there are indications that the personality and temperament of a horse are sufficiently genetically determined (Samor e et al. 1997;Lloyd et al. 2008;Graf et al. 2013a;Kim et al. 2018;Olsen and Klemetsdal 2019;Rankins and Wickens 2020).For instance, several studies have shown low to moderate heritabilities for temperament and personality traits (Samor e et al. 1997;Mantovani et al. 2010;Rothmann et al. 2014a;Kim et al. 2018).However, this requires enough data, both on the maternal and paternal side, and an appropriately large breeding population to make accurate estimates (Rothmann et al. 2014b;Willam and Simianer 2017).
This background was the impetus to continue collecting phenotypic information, not only to have marketing tools but also to be able to estimate genetic parameters for a future genetic evaluation.The latter should open the possibility to consider temperament traits for breeding purposes.In Samor e et al. (1997), heritabilities for some temperament traits in Haflinger horses were already estimated based on linear scoring.However, the assessment of parameters for analysing the horses' personality was not based on an objective, reliable and repeatable test design at that time, which, however, is essential during performance testing to generate viable phenotypic data for describing temperament traits (Graf et al. 2014).Therefore, the aim of the present study was to consider for the very-first time phenotypic information from character testing for estimating genetic parameters for temperament in Haflinger horses.

Data collection
For the present study, data from character testing of 210 South Tyrolean Haflinger mares between 2019 and 2021 were considered.The character testing was performed considering the test design originally designed by Graf et al. (2014) and further developed by Zanon et al. (2021).The test design includes several parameters for the evaluation of the temperament and behaviour of a horse in a well-defined test course with specific obstacles (¼stimuli).The obstacles (¼stimuli) were selected by Graf et al. (2014) considering specific guidelines which were easy realisation, low injury risk, and objective and reliable measurements which were previously tested under standardised conditions in form of two pilot studies by K€ onig von Borstel et al. (2012) and Schmidt (2009).The high interobserver reliability observed in K€ onig von Borstel et al. (2012) [0.89-0.96],Graf et al. (2014) as well as Zanon et al. (2021) [0.80-0.86]highlight the robustness of the test design.The final test course consists of 5 obstacles (¼stimuli), which have to be mastered in a given order (Figure 1).For the obstacles 'Lying balls', 'Rolling balls', 'Carpet' and 'Tunnel' three criteria were assessed, respectively, namely Interest in the obstacle (Interest), Activity at the obstacle (Activity), and Excitement during the exercise (Excitement).For the obstacle 'Standing' only the criterion Excitement was considered.For mastering a respective obstacle, the horse had 60 s test time available.The grading scale used for judging the criteria at respective obstacle is published in Zanon et al. (2021).Briefly, if an animal was unable to master the obstacle within specified test time, a score of 1 was given for 'Failure'.The maximum score for evaluating a criterion at respective obstacle was 10.On the day of official performance testing the test horses were assessed independently by two a priori trained judges at the same time.For each respective year, different judges were used for assessing the temperament.The character testing was performed at the test location Alps Coliseum in Egna,South Tyrol (46 35'57.742''N,11 31'4.627''E)370 m asl.All participating horses had to be brought to the test location.All tested horses were not previously exposed to the test course prior to the official performance test.
Moreover, a total of 5 ancestral generation for respective horse were available and retrieved form the database of the South Tyrolean Haflinger Breeding Association.

Statistical analysis
The software package SAS v. 9.4 (SAS Institute Inc., Cary, NC) was used for data editing and pre-analyses of the model (data not shown).Three animals were excluded from the analysis, two animals due to missing pedigree, one animal that appeared in two years and the correct ID was no longer clearly traceable.After editing, 207 mares remained in the final data set.The Haflinger mares were classified into 5 age groups (Table 1).The criteria Interest, Activity and Excitement were scored from 1 to 10 separated by obstacle and judge (Zanon et al. 2021).Thus, up to 10 repeated measurements were available per horse.Missing values were defined as 'missing'.
Animal models were used for the estimation of (co)variance components applying the restricted maximum likelihood approach implemented in the VCE6 programme (Neumaier and Groeneveld 1998;Groeneveld et al. 2010).Apart from a multivariate estimation of heritabilities (h 2 ) and genetic correlations for the criteria 'Interest', 'Activity' and 'Excitement', univariate and bivariate estimation runs were also performed for comparison and control purposes.The multi-trait mixed model used for the analysis was as follows: where, Yijklm is the score of the animal received for passing test course, obstacle i is the fixed effect of obstacle (5 classes for interest, 4 classes for activity and excitement as described in Zanon et al. ( 2021)), judge-Ã year j is the fixed interaction effect of judge and year (6 classes), age class k is the fixed effect of age class (5 classes as described in Table 1), pe l is the random permanent environmental effect of the animal (207 animals), a m is the random genetic animal effect and e ijklm is the random residual.A total of 5 ancestral generations were available.Thus, pedigree data set comprises 2,131 animals.Finally, the fixed effect of horse leader was not significant in previous analyses on the phenotypic level (p > 0.05, data not shown); therefore, it was excluded from the final model.Most horse owners only attended the test with only one mare (135  from 163 owners, 82.82%), hence the horse owner was also not considered in the model.Heritabilities were estimated as r 2 a /(r 2 a þ r 2 pe þ r 2 e ) and repeatabilities as ( e ) with r 2 a , r 2 pe and r 2 e denoting the additive genetic, permanent environmental and residual variance, respectively.

Results
Overall descriptive statistics of investigated temperament traits are reported in Table 2.The mean score for the behaviour criteria Interest was 6.5 with a standard deviation of 1.8.The mean score for Activity and Excitement was in both cases 7.5 with a standard deviation of 2.3 and 2.2, respectively (Table 2).The coefficient of variation (CV) for Interest, Activity and Excitement was 28%, 31% and 29%, respectively (Table 2).
Descriptive statistics for both judges are reported in Table 3.The mean score for Interest was 6.4 and 6.5, while for Activity and Excitement mean scores were 7.4 vs. 7.5 and 7.5 vs. 7.5, respectively (Table 3).Coefficients of variation varied between 27% and 32% between judges and criteria (Table 3).
Table 4 depicts the variance components and heritabilities for the investigated behavioural traits Interest, Activity and Excitement.All variance components were significantly different from zero except the genetic variance (0.11 ± 0.10 ns ) for the criterion Interest (Table 4).Heritability for Interest was lowest with 0.05 and a standard error of ±0.04.For the other two criteria Activity and Excitement, h 2 was 0.16 ± 0.05 and 0.18 ± 0.06, respectively (Table 4).Heritabilities from the univariate analyses were similar with somewhat higher standard errors (h 2 values of 0.03 ± 0.05, 0.20 ± 0.11 and 0.18 ± 0.10, respectively).In the bivariate analyses, heritabilities were also within the range of uni-and multivariate analyses.The repeatability levels of Interest, Activity and Excitement were 0.17, 0.28 and 0.37, respectively (Table 4).
Phenotypic and genetic correlation coefficients of the data regarding investigated behavioural traits are shown in Table 5.The phenotypic correlation between Interest and Activity was À0.31 and between Interest and Excitement À0.21 while the phenotypic correlation between Activity and Excitement was 0.82 (Table 5).The genetic correlation between Interest and Activity and between Interest and Excitement was in both cases negative with À0.60 and À0.78, respectively (Table 5).In contrast, the genetic correlation between Excitement and Activity was positive and very high (0.97) (Table 5).Similar results were achieved with the bivariate analyses (-0.49,À0.64 and 1.00, respectively).

Discussion
South Tyrol, as the region of origin of the Haflinger breed, plays an important role in the breeding, spreading and preservation of this horse breed.At present, about 3,000 Haflinger horses are present in South Tyrol, of which approx.1,800 are broodmares, 60 are stallions, 150 are geldings as well as 720 young horses, and are farmed by about 1,200 horse breeders and owners (Provinz 2021).Despite its history, Haflinger horse breeding is of secondary importance in this region, as the focus of the South Tyrolean livestock sector is on dairy cattle.Furthermore, due to machinery the original purpose of Haflinger to be work horses for mountain farms is no longer relevant.
Previous studies reported the inclusion of temperament traits in horse marketing and breeding to become more and more relevant (Graf et al. 2013a;Zanon et al. 2021;Hornauer et al. 2022).Consequently, several studies have already been performed in the past for developing an objective cost-efficient test design, which enables assessing temperament traits in a fast, objective, and repeatable way (Olsson 2010;K€ onig von Borstel et al. 2012;Graf et al. 2013bGraf et al. , 2014;;Kozak et al. 2018;Olsen and    Interest 0.30 ± 0.10 0.11 ± 0.10 2.04 ± 0.07 0.05 ± 0.04 0.17 ± 0.06 Activity 0.52 ± 0.19 0.75 ± 0.26 3.27 ± 0.12 0.16 ± 0.05 0.28 ± 0.07 Excitement 0.83 ± 0.22 0.79 ± 0.28 2.73 ± 0.09 0.18 ± 0.06 0.37 ± 0.06 r pe 2 permanent environmental variance; r a 2 genetic variance; r e 2 residual variance; h 2 heritability; r repeatability; SE standard error. Klemetsdal 2019) and on the other hand, for estimating genetic parameters for some temperament and behavioural traits in horses (Mantovani et al. 2010;Rothmann et al. 2014b;Kim et al. 2018).In the present study, temperament and behavioural traits of 207 Haflinger mares were assessed using the test design first described in Graf et al. (2014) and further developed by Zanon et al. (2021).For the assessment only young-aged horses were considered (Table 1), as the temperament test was performed during the admission procedure of young mares to the stud book.However, Hausberger et al. (2004) and Abdel-Azeem and Emeash (2021) observed no significant relationship between horse temperament or personality traits and age.Furthermore, Mantovani et al. ( 2010) concluded that temperament, among other traits, is heritable if assessed at an early stage in horse's life, which agrees with our sampling strategy.Moreover, all animals were tested in the same equestrian hall to guarantee similar environmental and handling conditions and thus avoid potentially arising confounding effects as far as possible (Rankins and Wickens 2020).The CV ranged between 28% for Interest and 31% for Activity (Table 2) and were similar to CVs reported in Kim et al. (2018) for the temperament traits assessed in Jeju crossbred.In addition, there was a high degree of agreement in the judges' scoring (Table 3) which highlights the good repeatability and objectivity of the test design as already reported in Graf et al. (2014) and Zanon et al. (2021).The assessed temperament traits in the present study showed low to moderate heritabilities (Table 4), which is in line with previously published findings on personality and temperament assessment in horses.For instance, Rothmann et al. (2014b) reported a heritability for reactivity to be 0.17 while Samor e et al. (1997) estimated heritabilities for temperament and docility in Haflingers to be 0.06 and 0.02.Moreover, Kim et al. (2018) reported heritability estimates of 0.32, 0.40, 0.40, 0.53 and 0.08 for gentleness, patience, aggressiveness, sensitivity, and friendliness and Molina et al. (1999) and Mantovani et al. (2010) estimated heritabilities for temperament of 0.08 and 0.26, respectively.The repeatability levels for Interest, Activity, Excitement reported in Table 4 were lower compared to values reported for different temperament traits in Kim et al. (2018) which might indicate a moderate inter-evaluation reliability.The latter might be explained by the limited number of animals considered in the present study (Rothmann et al. 2014b).Another possible explanation for the moderate inter-evaluation reliability might be related to the multi-test approach.However, the briefing of the judges, the test design as well as the location for testing the temperament was for all test runs the same and further, Schmidt (2009), K€ onig von Borstel et al. (2012) and Zanon et al. (2021) observed high interobserver reliability as well as highlighted the robustness of the test design.Therefore, for obtaining more robust data, a larger number of animals as well as repetitive collection of phenotypic performance data over a longer period are required (Rothmann et al. 2014b;Willam and Simianer 2017).Moreover, the strong correlation between Activity and Excitement (Table 5) highlights the fact that these two criteria are very similar and therefore could be considered as one criterion for future investigation.It should be noted, however, that both the multivariate and the bivariate analyses between Activity and Excitement did not finish with the highest status, while all other runs did.This suggests that the estimates for the relationship between Activity and Excitement should be interpreted with caution and a re-estimation of variance components is advised as soon as more data are available.
Finally, the present heritability estimates provide a first indication on which traits could be worth studying further on a molecular genetic level (Berglund 2021), as some studies already identified candidate genes involved in equine temperament as well as genetic markers for temperament traits (Momozawa et al. 2007;Ren et al. 2017;Rankins and Wickens 2020;Wickens and Brook 2020).The latter are highlighting the potential for genomics-assisted breeding.Therefore, in future, it might be possible to consider genotype information for the genetic evaluation of temperament traits in Haflinger horses in addition to phenotypic performance testing, which was reported in the present study.However, as far as variance components are concerned, differences between estimates based on the pedigree or on genomic information were previously found to be rather small (e.g.Aldridge et al. 2020), particularly in case of random genotyping (Wang et al. 2020).Besides, according to Kennedy et al. (1988), variance components estimated by the REML method and an animal model using the pedigree-based relationship matrix are unbiased if selection is based on phenotype and pedigree information and an infinitesimal model is justifiable.

Conclusion
In the present study, low and moderate heritabilities were observed for the temperament traits Interest, Activity and Excitement, which highlight the potential of considering temperament traits in Haflinger horses' breeding.Although the quality of the phenotypic performance data collected during the character testing was good, the restricted number of animals limited the accuracy of the estimated genetic parameters.As next step, a continuation of the collection of phenotypic temperament performance on mares and possibly also on stallions is crucial for increasing the robustness of the data and therefore, improve the accuracy in heritability estimation.The latter is fundamental for estimating breeding values with appropriate reliabilities.Meanwhile the test results can individually be used for the marketing of single horses as many customers are interested in horses which have either been proven to be 'easy to handle' or which are showing are certain level of 'temperament'.

Figure 1 .
Figure1.Test course design for assessing the temperament and behaviour in Haflinger horses(Zanon et al. 2021, modified).

Table 1 .
Age classes for Haflinger mares used for defining the fixed effect.

Table 2 .
Descriptive statistics of investigated behaviour traits.

Table 3 .
Descriptive statistics of investigated behaviour traits for both judges.

Table 4 .
Variance components, heritabilities and repeatabilities from the multivariate analysis for Interest, Activity and Excitement in Haflinger horses.

Table 5 .
Genetic and phenotypic correlation coefficients from the multivariate analysis for Interest, Activity and Excitement in Haflinger horses.Phenotypic correlations are above, and genetic correlations are below the diagonal; ±standard error.