Genetic parameters for longevity and informative value of early indicator traits in Danish show jumping horses

Livestock Science 184 (2016) 126–133

Contents lists available at ScienceDirect

Livestock Science journal homepage: www.elsevier.com/locate/livsci

Genetic parameters for longevity and informative value of early indicator traits in Danish show jumping horses T. Seierø a, T. Mark a, L. Jönsson a,b,n a b

Department of Veterinary Clinical and Animal Sciences, Faculty of Health and Medical Sciences, University of Copenhagen, 1870 Frederiksberg, Denmark Danish Warmblood Association, Vilhelmsborg Alle 1, 8320 Mårslet, Denmark

art ic l e i nf o

a b s t r a c t

Article history: Received 16 October 2015 Received in revised form 18 December 2015 Accepted 22 December 2015

The study aim was to investigate the usefulness of different longevity measures in Danish show jumping horses as well as potential early indicator traits for longevity in jumping. The analyses comprised jumping competition data from 9592 horses born during 1981 to 1994, 30,435 young horse records, and their pedigree. Genetic parameters and breeding values were estimated using AI-REML and mixed models including fixed effects of birth year, age at first placing, sex and number of offspring. Four longevity traits were investigated: no. years in competition from first to last entry (NYC), no. active years in competition i.e. only years with a registered start/placing (NAY), NAY plus no. foals carried to term by mares (NAYF), and accumulated lifetime points (LDP) combining longevity and competition success. Longevity defined as NAY was found most useful for the Danish Warmblood. The heritability of NAY was 0.11. Young horse jumping traits had moderate to high genetic correlation with longevity (rg: 0.51–0.74) and highest value as indicator trait among young horse traits (rg
rIA: 0.23–0.44). Conformation had lower informative value for longevity (highest |rg
rIA| was 0.10). Including information of young horse capacity and rideability during jumping in a multivariate analysis increased the accuracy of NAY breeding values of younger horses from 0.32 to 0.49 and increased model predictive ability compared to a univariate longevity evaluation. & 2016 Elsevier B.V. All rights reserved.

Keywords: Longevity Show jumping Genetic correlation Indicator trait

1. Introduction Longevity of domestic animals refers to the length of the animal’s service or productive life. For example, longevity in dairy cows can refer to the duration of which the cow produces milk (e.g. Pritchard et al., 2013), and longevity in sport horses to the time in competition (e.g. Jönsson et al., 2014b). Longevity in horses is important because it increases the time they can be used for their primary purpose, competitions, relative to the investments associated with rearing and training them. Thus, longevity is considered one of the most important traits of a horse according to potential horse buyers on the market (Hennessy et al., 2008). Furthermore, longevity may be favourably associated with animal welfare traits such as orthopaedic health. It is difficult to measure longevity in horses, due to the risk of indirectly measuring talent and/or other factors. Furthermore, it is typically impossible to distinguish between voluntary and involuntary termination of the sport career in practice as the cause of it is not routinely recorded. n Corresponding author at: Grønnegårdsvej 3, Department of Veterinary Clinical and Animal Sciences, Faculty of Health and Medical Sciences, University of Copenhagen, 1870 Frederiksberg C, Denmark. E-mail address: [email protected] (L. Jönsson).

http://dx.doi.org/10.1016/j.livsci.2015.12.010 1871-1413/& 2016 Elsevier B.V. All rights reserved.

Ricard and Blouin (2011) proposed the use of number of years from the first entry to the last entry in competition as a measure of longevity, defined as NYC in the present study. The heritability of NYC was estimated at 0.10 in a French riding horse population by Ricard and Blouin (2011), and at 0.17 for Hungarian jumping horses (Posta et al., 2014). In the study by Braam et al. (2011) only years in which the horse had placings, were accounted for, i.e. number of active years in competition (presently defined as NAY). Braam et al. (2011) found that NAY in show jumping was on average 3.3 years, the heritability was found to range between 0.07 and 0.17 depending on model. The lowest heritability was found when age at first placing was taken into consideration. Jönsson et al. (2014b) likewise used NAY and estimated the heritability to be 0.20. The differences in heritability from Braam et al. (2011) could be due to differences between the populations considered as Braam et al. (2011) studied competing Swedish Warmblood (SWB) males born 1967 to 1991, while Jönsson et al. (2014b) studied all SWB horses participating in young horse performance tests (both gaits and jumping) from 1983–2005 (except 1985–87). In addition to NAY, Jönsson et al. (2014b) considered another trait where points were given upon lifetime results, i.e. number of accumulated lifetime points (defined as LDP in the present study). The highest points were given upon better placing’s in the higher level

T. Seierø et al. / Livestock Science 184 (2016) 126–133

Number of horses per birth year 1200

Number of horses

of competition. The top 25% horses in a competition are given placings. Lifetime performance is therefore a mixture of ability to perform well in a single competition (preferably at high level) and ability to have a long competition career. The heritability for lifetime performance in competition was found to be 0.24 (Jönsson et al., 2014b). Furthermore, number of starts has been used as a longevity measurement (Arnason et al., 1982) for trotters, with an estimated heritability of 0.10. For race longevity and race persistence heritabilities have been estimated at 0.10 and 0.12, respectively, in similar records of Thoroughbred race horses (Velie et al., 2015). Only few of the European breeding associations have breeding goals directly focusing on longevity or health of horses. Most of the breeding associations instead put emphasis on soundness and/or conformation (Koenen et al., 2004), which in some cases may be indirect measures of longevity (Jönsson et al., 2014b). It is, however, beneficial to consider the breeding goal trait directly in genetic evaluations while also including correlated information from early indicator traits via a multivariate analysis rather than only evaluating the indicator traits. Early selection reduces the generation interval and improves genetic progress provided use of accurate selection criteria. But uncensored longevity records are not available until after the horse’s sports career, which makes it challenging to improve longevity of sport horses through breeding unless genomics or informative early indicator traits are used. In addition to being available earlier in the life of a horse, potential early indicator traits such as young horse performance and conformation records are also expected to be less influenced by training, rider and random events compared to competition results later in life which is expected to result in higher heritabilities for the former traits. The aim of this study was therefore to investigate the usefulness of different longevity measures in Danish show jumping horses as well as potential early indicator traits that can be used to improve the accuracy and predictive ability of genetic evaluations for longevity.

127

1000 800 600 400 200 0

Birth year

Longevity was studied, based on available information supplied by SEGES, Danish Warmblood (DWB) and the Danish Riding Association (DRF), as no. years in competition from first to last entry (NYC), no. active years in competition i.e. only years with a registered start/placing (NAY), NAY plus no. foals carried to term by mares (NAYF), and accumulated lifetime points (LDP) combining longevity and competition success. 2.1.1. Competition data The competition data was recorded by DRF and registered by the Danish national centre for animal databases and breeding evaluations (SEGES). The competition data included 728,311 repeated competition observations of 22,034 DWB horses with 9739 riders from show jumping competitions held in Denmark in the years from 1986 to 2013. From 1986 to 1997 only placings were registered, since 1998 all entries independent of success in the competition were included. Records of 184 horses (0.8%) were omitted due to unrealistic age ( o5 or 428 years) and of 11 horses born before 1970 due to few horses (o 10) per age group. Years in competition were described up until a maximum of 14 years as few horses had competed longer (0.2% for NAY and 0.6% for NYC). To avoid right censuring regarding years in competition, only horses expected to have finished their sports career were included in the dataset. Hence, horses born after 1994 were excluded from the dataset. In total, observations on 10,631 horses were removed,

Unused

Fig. 1. Number of horses per birth year in the original competition data set; data from years marked with dark colour were discarded to avoid left and right censoring.

Table 1 Data description including phenotypic standard deviation (sP) of untransformed edited longevity data. Trait

No. records Mean

Median 75% quantile

spf

Kurtosis Skewness

NAYa NYCb LDPc NAYFd AJRe

9592 9592 7836 9592 9592

2.00 2.00 31.0 3.00 8.06

2.58 3.05 952 2.96 1.31

5.56 4.33 236 4.71 3.77

3.12 3.56 250 3.61 7.79

4.00 5.00 131 5.00 8.81

1.61 1.37 12.3 1.42  0.87

a

NAY – Number of active years in competition; NYC – Number of years in competition; LDP – Accumulated lifetime points; d NAYF – Number of active years in competition plus number of foals; e AJR-Average jumping results; f sp – Phenotypic standard deviation. b c

Table 2 Estimated heritabilities (h2) and additive genetic standard deviation (sA) of longevity traits, as well as their genetic correlations with average jumping results (rg(AJR)) for different traits and models (SE is the asymptotic standard error from AIREML of the given parameter). Traita

h2

SE(h2)

sA

rg(AJRf)

SE(rg(AJR))

NYCb NAYc NAYFd LDPe AJRf

0.099 0.114 0.233 0.311 0.190

0.020 0.021 0.027 0.034 0.032

0.246 0.241 0.370 1.054 0.441

0.068 0.086 0.526 0.432 –

0.138 0.130 0.094 0.091 –

2. Materials and methods 2.1. Material

Used

a All traits are transformed by natural logarithm, to ensure best possible normal distribution. b NYC – Number of years in competition; c NAY – Number of active years in competition; d NAYF – Number of active years in competition plus number of foals; e LDP – Accumulated lifetime points; f AJR¼Average jumping results.

leaving 11,208 horses born from 1970 to 1994. To avoid left censuring horses born before 1981 were omitted as competition records included data from 1986 and later, and horses can start competing in classes registered in DRF from the age of five. Thus, 9592 horses remained in the final competition dataset (3643 mares and 5949 males). Number of horses per birth year ranged from 409 (1981) to 1044 (1994) horses (Fig. 1). Geldings and stallions were categorised together as males, because information was missing about whether or when stallions had been castrated, and 30 geldings had registered offspring. NAYF was studied as the sum of NAY and number of foals carried to term as a mare (i.e. only mares were credited for foals) irrespective of whether the mare competed before or/and after parturition. The data did not include information of twin births. In total, 2252 mares had registered offspring. Horses had on average 3.56 NYC, 3.12 NAY, 250 LDP and 3.61 NAYF (Table 1). Mares had on average slightly lower longevity

128

T. Seierø et al. / Livestock Science 184 (2016) 126–133

Table 3 Number of records used in analyses (N), estimated heritability (h2) with standard error (SE h2), and estimated additive genetic standard deviation (sA) for potential indicator traits. Furthermore, estimated genetic correlations (rg) with standard errors (SE rg) between the indicator trait and longevity (for longevity both NAY and LDP were considered) from bivariate analysis. Indicator trait

h2 (SE h2)

N

sA

rg (SE rg) NAYd

Conformation Type Frontparta Toplineb Forelimbsc Hindlimbsc Correctness of movement Elasticity, gaits Overall conformation Young horse jumping Rideability, cooperativeness Capacity Canter

LDPe

24,808 24,809 24,809 13,512 13,513 13,502 13,508 24,809

0.288 0.277 0.211 0.168 0.160 0.195 0.347 0.294

(0.018) (0.017) (0.017) (0.020) (0.020) (0.021) (0.024) (0.017)

0.423 0.383 0.342 0.335 0.325 0.339 0.554 0.455

0.018 0.013 0.101 0.052 0.227  0.150  0.133 0.001

(0.096) (0.096) (0.108) (0.120) (0.118) (0.115) (0.098) (0.093)

 0.027  0.033 0.114  0.034 0.055  0.207  0.088  0.040

(0.076) (0.076) (0.084) (0.099) (0.100) (0.095) (0.081) (0.074)

6732 12,637 4771

0.253 0.401 0.319

(0.033) (0.026) (0.042)

0.484 0.664 0.513

0.660 0.740 0.561

(0.106) (0.083) (0.119)

0.510 0.712 0.545

(0.089) (0.058) (0.090)

a

Saddle area, shoulder, withers. Topline and hindquarter. c Not including correctness of movement. d NAY – Number of active years in competition. e LDP – Accumulated lifetime points. b

Table 4 Mean accuracy (rIA) of EBVs for different indicator traits and standard deviation (SD) of the rIA, predicted in bivariate analyses with longevity (for longevity both NAY and LDP were considered). Furthermore potential value (|rg
rIA|) as indicator trait for old horses with both longevity records, tested younger horses, and stallions with more than ten offspring with at least one young horse performance record. Indicator traits

|rg
rIA|d

Mean rIA (SDrIA)

NAYe

Conformation Type Frontparta Toplineb Forelimbsc Hindlimbsc Correctness of movement Elasticity, gaits Overall conformation Young horse jumping Rideability, cooperativeness Capacity Canter

LDPf

Old horsesg

Younger horsesh

Stallionsi

Old horsesg Younger horsesh Stallionsi Old horsesg Younger horsesh Stallionsi

0.476 0.470 0.445 0.406 0.401 0.419 0.475 0.475

0.591 0.583 0.531 0.460 0.453 0.481 0.576 0.596

(0.075) (0.076) (0.085) (0.118) (0.117) (0.119) (0.131) (0.074)

0.772 0.766 0.723 0.502 0.495 0.525 0.618 0.776

0.009 0.006 0.045 0.021 0.091 0.063 0.063 0.001

0.011 0.007 0.054 0.024 0.103 0.072 0.077 0.001

0.014 0.010 0.073 0.026 0.112 0.079 0.082 0.001

0.013 0.016 0.051 0.014 0.022 0.087 0.042 0.019

0.016 0.019 0.060 0.016 0.025 0.100 0.051 0.024

0.021 0.025 0.082 0.017 0.027 0.109 0.055 0.031

0.398 (0.196) 0.445 0.466 (0.218) 0.596 0.379 (0.379) 0.440

(0.144) (0.138) (0.154)

0.433 (0.228) 0.263 0.694 (0.179) 0.345 0.459 (0.235) 0.212

0.293 0.441 0.247

0.286 0.514 0.257

0.203 0.332 0.206

0.227 0.424 0.239

0.221 0.494 0.250

(0.180) (0.183) (0.175) (0.173) (0.171) (0.177) (0.193) (0.185)

(0.094) (0.096) (0.105) (0.195) (0.194) (0.199) (0.215) (0.093)

a

Saddle area, shoulder, withers. Topline and hindquarter. Not including correctness of movement. d Potential value as indicator trait (rg
rIA) calculated as: genetic correlation (rg) between young horse trait and longevity measure multiplied with mean accuracy of EBV of young horse trait (rIA). e NAY – Number of active years in competition. f LDP – Accumulated lifetime points. g Old horses: Horses with own longevity (NAY and LDP) records (n¼ 7836) h Younger horses: Horses with at least one young horse result, and with no longevity (NAY or LDP) record (n ¼27,795) i Stallions: Stallions with more than 10 offspring with at least one young horse result (no longevity (NAY or LDP) record) (n¼ 146) b c

(NYC 3.45, NAY 3.01, LDP 234 and NAYF 4.30) than males (3.63, 3.18, 260 and 3.18) except for NAYF. The study was focused on longevity in show jumping competitions only, due to the pronounced specialisation of DWB horses, where only 3–6% of studied horses were actively competing in both dressage and show jumping the same competition year. Average jumping results (AJR), calculated as the average of repeated observations of ranking points was used to measure competition success where:

ranking point = 11 −

original ranking + (6 − level of competition) × 5

where the part within the square root was the adjustment of ranking points based on level of the competition class (level 1¼ lowest level to level 6 ¼highest level) so that a 1st placing on level 4 was equal to the ranking points for a 6th placing on level 5, or a 11th placing on level 6. For best normalisation (Boelling, 2011), 11-the square root of the adjusted ranking points was used. Furthermore the competition data included records of accumulated lifetime DRF points (LDP), which were assigned to the horses

0.278 0.550 0.550 0.556 (0.117) (0.110) (0.110) (0.112) 0.323 0.484 0.485 0.492 (0.086) (0.126) (0.127) (0.128) 0.432 0.457 0.457 0.462

Accuracies were from bivariate analysis rEBVfull,EBVred is the Pearson correlations between estimated EBV in full and reduced data for horses with observations set missing. In the reduced data all results of horses born between 1990 and 1994 were set missing. The intercept was set to zero. bEBVfull,EBVred is the regression coefficient between estimated EBVs in full and reduced data for horses with observations set missing. In the reduced data all horses born between 1990 and 1994 were set missing. The intercept was set to zero. c Univariate: Longevity traits in univariate analysis. d Bivariate: Longevity traits and capacity in bivariate analysis. e 3 traits: NAY – NAY, capacity and rideability/cooperativeness in multi-variate analysis; LDP-LDP, capacity and canter in multi-variate analysis. f 4 traits: NAY – NAY, capacity, rideability/cooperativeness and canter in multi-variate analysis; LDP-LDP, capacity, canter and rideability/cooperativeness in multi-variate analysis. g NAY – Number of active years in competition. h LDP – Accumulated lifetime points. i Old horses: Horses with own longevity (NAY and LDP) records (n¼ 7836). j Younger horses: Horses with at least one young horse result, and with no longevity (NAY or LDP) record (n¼ 27,795). k Stallions: Stallions with more than 10 offspring with at least one young horse result (no longevity (NAY or LDP) record) (n¼ 146). b

0.343 0.564 0.575 0.575 (0.126) (0.106) (0.109) (0.109) 0.385 0.507 0.515 0.516 (0.053) (0.071) (0.072) (0.072) 0.608 0.625 0.628 0.630 1.742 1.169 1.151 1.184 (0.191) (0.144) (0.144) (0.150)

0.666 0.777 0.780 0.777

Stallions Younger horses Old horses bEBVfull,EBVred rEBVfull,EBVred Old horses

i

Mean rIA (SErIA)

Younger horses

j

Stallions

k

Univariatec Bivariated 3 traitse 4 traitsf

a

0.702 0.727 0.728 0.724 (0.218) (0.153) (0.156) (0.156)

0.617 0.673 0.674 0.671

bEBVfull,EBVred rEBVfull,EBVred

Model predictive ability

k j i

Mean rIA (SErIA) Model predictive ability

LDPh NAYg Analysis

Table 5 Mean accuracies (rIA)a, standard deviation (SD) of the rIA, and model predictive ability (rEBVfull,EBVred; bEBVfull,EBVred)b for longevity traits (NAY and LDP) with univariate and three different multi-trait analysis. Results are presented for three.

T. Seierø et al. / Livestock Science 184 (2016) 126–133

129

upon placings in competitions from regional competition and upwards. The numbers of points awarded increase with increased level of show, class and number of placing. 2.1.2. Young horse performance scores The data was recorded in years 1984 to 2013, predominantly including 3–4 years olds at time of recording. Results were accompanied with information of age, gender, and date and place of recording (event), where information from events with at least 10 participating horses was studied. Studied data included results for 30,435 young horses, 24,809 horses with conformation scores and 12,928 horses with jumping scores. Some young horse traits were not recorded during the entire period, hence scores obtained from free jumping and jumping under rider were pooled for each trait, using the mean in cases of information on both, and only traits with at least 100 horses having records for both the young horse trait and competition were further considered. 2.1.3. Pedigree data The pedigree data included known ancestors seven generations back from each horse with competition and/or young horse conformation and performance data. For NAY and NYC it consisted of 36,707 horses with information of dam and sire, for LDP 32,019 horses, and when including all pedigree data for both horses with longevity records and young horse performance scores it consisted of 104,801 horses. Unknown dam or sire was replaced with an age dependent fictive phantom group identification number, so each birth year cohort obtained an individual fictive dam and sire number, as horses born the same year could be assumed to have more similar genetic makeup compared to horses born in other years. 2.2. Statistical analyses Longevity traits were transformed using the natural logarithm to ensure better approximation to normal distribution, as found most appropriate in Box-cox transformation trials, similar to some previous studies of competition data (Viklund et al., 2010; Jönsson et al., 2014b). Transformed longevity traits showed a skewness of 0.33 to 0.48 and a kurtosis of 0.74 to 1.19. Non-genetic effects included in the model used for the analyses were chosen based on initial analyses of variance using the “Generalised Linear Models” (GLM) procedure in R (R Core Team, 2013). For estimating genetic parameters, the average information restricted maximum likelihood (AI-REML; Jensen et al., 1997) as implemented in the DMU package was applied (Madsen and Jensen, 2013). Both univariate and multivariate analysis were performed, for computing accuracies of estimated breeding values (EBV). The following BLUP animal model (1) was utilised for the longevity trait (either NAY, NYC, NAYF or LDP):

yijkn = BYi + AFEj + Sk + an + eijkn

(1)

where y was the log-transformed longevity trait, either NAY, NYC, LDP or NAYF; BY was the fixed class effect of birth year (i ¼1981, …, 1994); AFE was the fixed class effect of age at first placing (j¼5, …, 14), where ages above 14, was set to 14 (3.3% of studied horses); S was the fixed class effect of sex (k¼ 1 (male), 2 (female)); a was the random additive genetic effect of the nth horse, and e the random residual effect for the ijknth observation. The random effects were assumed to follow a multivariate normal distribution with zero means and variance equal to,

⎡ A ⎤ ⎡ Aσa2 0 ⎤ ⎥, var ⎢ ⎥ = ⎢ ⎣ E ⎦ ⎣⎢ 0 Iσ 2 ⎥⎦ e where A was the additive genetic relationship matrix and I was an

130

T. Seierø et al. / Livestock Science 184 (2016) 126–133

incidence matrix. Only the most informative indicator traits were considered in multivariate BLUP analysis. The following animal model (2) was used for the young horse traits in the multivariate analysis which also included one of the longevity traits:

yijn = AGi + LDj + an + eijn

(2)

where y was the young horse trait; AG was a combined fixed class effect of age and sex n(i) ¼20 for conformation traits (including 3– 12 year olds), and n(i)¼ 3–4 for jumping traits (including 4–5 year olds, or 3–5 year olds respectively); LD was a combined fixed class effect of location and date of the grading (n(j)¼ 357); a was the random additive genetic effect of the nth horse and e the random residual effect, where A was the relationship matrix and I was an incidence matrix of appropriate size corresponding to no. observations. The (co)variance structure for the bivariate models was (similar assumptions for 4 2 traits):

⎡ Aσ 2 Aσ 0 0 ⎤ a1,2 a1 ⎥ ⎡ a1⎤ ⎢ ⎢ a ⎥ ⎢ Aσa1,2 Aσa22 0 0 ⎥ 2 ⎥ var ⎢ ⎥ = ⎢ ⎢ e1 ⎥ ⎢ 0 0 Iσe21 Iσe1,2 ⎥ ⎥ ⎣ e2 ⎦ ⎢ ⎢⎣ 0 0 Iσe1,2 Iσe22 ⎥⎦ where sa1,2 and se1,2 are additive genetic and residual covariances, respectively. The accuracy of EBVs (rIA) were calculated as

rIA =

⎛ σ2 ⎞ ⎟ 1 − ⎜ EBV ⎝ σA2 ⎠

Heritabilities were calculated using results of single-trait analyses, while genetic correlations were calculated using results of bi-variate analyses. The mean accuracy of EBVs was calculated for (1) horses with own longevity records (“old horses”), (2) horses with at least one young horse performance record and no longevity record (“young horses”), and (3) stallions with more than 10 offspring that each had at least one young horse performance record but no longevity record (“stallions”). The three above groups were chosen to study the scenarios of selection based on waiting for own longevity results, compared to relying on only own young horse results, or stallion evaluation based on only offspring young horse results, respectively. The impact of young horse information traits was studied by comparing univariate evaluation of the longevity trait with multivariate evaluations including the most informative young horse traits (1–3 young horse traits) for mean accuracies and model predictive ability. The model predictive ability was studied according to Reverter et al. (1994) using Pearsons correlation and regression coefficients between EBVs estimated using full and reduced data to study consistency between consecutive evaluations and indications of bias, respectively. In the reduced data set phenotypic records of horses born 1990–1994 were excluded. The regression coefficient was calculated with no intercept, based on EBVs for the horses with observations set missing in the reduced data (EBVred ) regressed on EBVs for the same horses from the full data (EBVfull ) ; EBVfull = b (EBVred ).

3. Results The estimated heritabilities of the longevity traits were low to moderate (h2: 0.10–0.31; Table 2). The estimated heritability of the young horse performance and conformation traits (Table 3) ranged from 0.16 to 0.40, with canter, elasticity in gaits and capacity in jumping being most heritable (h2: 0.32, 0.35, 0.40 respectively). The estimated genetic correlation between longevity and AJR ranged from 0.07 to 0.53 (Table 2). Genetically, NYC and NAY were

almost independent of AJR (rg ¼ 0.1) whereas both NAYF and LDP had a positive genetic correlation with AJR (rg ¼ 0.4–0.5). The estimated genetic correlations between young horse conformation traits and longevity traits were generally low (rg:  0.15 to 0.23 for NAY and 0.21 to 0.11 for LDP; Table 3). However, the estimated genetic correlations between young horse jumping traits and longevity traits were moderate to high (rg: 0.56–0.74 for NAY and 0.51–0.71 for LDP; Table 3). When considering the most informative young horse performance and conformation trait, by multiplying the genetic correlation between the indicator and longevity trait with the mean accuracy of the EBV of the indicator trait, the young horse jumping traits provided most information about longevity (rg
rIA(jump): 0.25–0.44 for NAY; 0.23–0.42 for LDP; Table 4). The accuracy of EBVs for the longevity traits increased substantially when using bivariate compared to univariate models for ‘young horses’ (i.e. horses having at least one young horse performance and conformation record, but no longevity record) as well as for stallions with more than ten young horse progeny records (Table 5). The mean accuracy increased further, but with smaller margins, as one and especially two additional indicator traits were added to the multivariate analysis. That is, the mean accuracy for longevity was 0.32–0.49 for NAY and 0.39–0.52 for LDP. The consistency between consecutive evaluations for NAY was moderate to high (rEBVfull,EBVred: 0.66–0.78). Indications of bias in NAY EBVs was relatively low for the multivariate analyses (bEBVfull,EBVred: 1.15–1.18), but higher in the univariate analysis (bEBVfull,EBVred: 1.74). For LDP the model predictive ability was poorer compared to NAY except for indications of bias in the univariate evaluations. Considering both consistency and indications of bias there was a noticeable improvement in model predictive ability for multivariate evaluations compared to univariate evaluations, and especially for NAY.

4. Discussion 4.1. Study settings and limitations associated with available data A substantial (56%) amount of data was excluded in data editing of the competition data. As censored records could not be appropriately handled in the linear models considered, it was necessary to exclude horses born after 1994 in order to avoid left censuring. Thus, horses used in the statistical analysis were quite old (421 years of age). The birth year 1994 was chosen as cut off for the current data, because less than two percent of horses above 19 years old were still active in show jumping competitions (Fig. 1) in 2013, which was the last recorded year in the competition data. Due to technological developments, a change of the stored data occurred in 1998. Before this year, only entries with placings were registered, while all entries in competition regardless of placing were registered from 1998 and onwards. It may also have affected horses with few entries, due to lower chance of getting placings and thus a risk of exclusion from the competition statistics of the early period. Number of horses with only one active competition year was fairly stable for horses born from 1981 to 1985 but increased after 1985 which may reflect registration changes. Further, competition statistics may not be comparable between time periods, due to genetic and veterinary progress, training changes and easier access to participate in competitions in recent years due to more shows and easier transportation of horses. Thus, studied traits for longevity may have changed during the study period. These changes may however, be partly adjusted for through the inclusion of birth year as an effect in statistical analyses. Studied data included only show jumping competition results, where horses could have competed in dressage without any credit in the estimate of longevity for show jumping. However, only 3–6% of

T. Seierø et al. / Livestock Science 184 (2016) 126–133

horses with jumping results were also active in dressage competitions during the same competition year, suggesting that active careers in both disciplines at the same time are very rare for DWB horses. Proportion of studied horses with results in both disciplines in their lifetime career, on the other hand, was somewhat larger (25%), raising the question of reason for shifting discipline. The average lifetime result in dressage for these horses was markedly lower (68 points), compared to the average of all dressage competing horses (136 points), suggesting a rather short and/ or unsuccessful dressage career. The change may in some cases be due to change of owner interest for different disciplines, but could also be due to altered fitness/health of the horse where coping with physical strains of competing in jumping was no longer possible. In such case, inclusion of dressage competition results, would give an overestimate of longevity for jumping. Due to the risk of favoring horses that no longer could compete in jumping, in combination with a distinct specialisation structure where most DWB horses only compete in one discipline at the same time and where most young horses only are evaluated for talent in one of the disciplines, longevity was chosen to be evaluated for jumping results separately. In a less specialised population the optimal manner of considering the competition data may have been different. Studied data was collected from competitions from regional to international level in Denmark, but data from local competitions were not included, similar to competition recordings in other studies (Viklund et al., 2010; Jönsson et al., 2014b). Thus, the competition data was most likely pre-selected for talent. Horses and close relatives to horses that primarily tend to participate in local competitions may therefore have underestimated longevity. It may also have caused underestimated genetic correlations between longevity and jumping talent traits as the most talented horses are expected to participate in higher level competitions to a larger extent. Further, Danish horses that compete in international competitions abroad would not have been awarded the correct amounts of entries and points due to placings, as this information was not available. However, at least one entry or placing in a Danish competition will likely occur each year for these horses, thus, minor or no influence on the longevity traits NYC, NAY and NAYF may be expected. For further studies efforts should, however, be made to include data from all levels of competition. Exported horses are expected to be more talented than the average horse, but with a potential risk of right-censored competition results, which may affect study results to some extent. However, only few jumping horses are exported from Denmark. Number of imported jumping horses is expected to be higher, but should not affect the results as age at first entry was corrected for. Survival analyses were not considered as they were considered too demanding in practical implementations, especially in a multiple-trait genomic scenario, for the Danish horse industry which has few human resources for developing and running genetic evaluations. Theoretically survival traits are advantageous, as they allow appropriate use of censored records (Ducrocq and Casella, 1996). However, in practice appropriate separation of time dependent environmental and genetic effects can be challenging (e.g. Nielsen et al., 2003). 4.2. Choice of longevity measurements The longevity measures that could be considered in this study were restricted because number of starts and the cause of career termination were not available. In the present as well as in earlier studies where NYC (Ricard and Blouin, 2011) and NAY (Braam et al., 2011; Jönsson et al., 2014b) have been considered, the analyses were based on whole years, whereas a more precise estimate could be expected if months, quarters or number of starts were

131

studied. A further problem with measuring longevity using NYC is horses that may have periods in-between years of the competition career start and end where it is not competing due to health problems. In the case of Denmark where health traits is not being evaluated it is desirable that longevity takes into account and penalise horses with injuries. For this reason NAY would be preferred over NYC. Additionally, the heritability of NAY was highest for the two. LDP was a more indirect way of measuring longevity as horses with a high LDP would need many years of training and competition experience to reach the highest level where the highest points are earned. However, this trait included also horse capacity, talent and environmental factors, such as rider and trainer to a larger extent. This was reflected by the estimated genetic correlation of 0.43 between LDP and average competition result (Table 2). LDP could be advantageous when a combined evaluation of both talent and longevity is desired, but presents a disadvantage if breeders desire information on both traits e.g. if different economic weights are desired. NYC and NAY were also influenced by talent, as less talented horses would have fewer entries in competitions and certainly less placings, but to a smaller extent. Further, horses competing in high level of competition, may participate in less competitions due to a predetermined training and competing schedule, which would influence LDP but not NYC and NAY as they compete in more than one show per year. Mares with foals had, on average, higher NAY, NYC and LDP compared to mares without foals, which likely indicate that mares used for breeding had more talent than the average horse. This could also explain the substantially higher genetic correlation between mean jumping performance and NAYF (rg(AJR): 0.53) compared to NAY (rg(AJR): 0.09). NAYF was considered as a longevity trait because the number of foals a mare had produced potentially influenced NYC, NAY and LDP negatively. On that ground the trait NAYF was included to compensate mares having foals with one year per foal. This was a rough estimate as it is possible to compete in show jumping until five months before term and again four months after. Therefore some mares could, although unlikely, compete the same year they produced a foal, resulting in an over-compensation. The high genetic correlation between NAYF and AJR of 0.53 is, however, problematic if a longevity measure being independent of AJR is strived for, as well as the fact that injuries are sometimes used as an opportunity for mares to have foals. Instead of modifying the dependent variable based on number of produced foals of mares, the impact of accounting for foals as an explanatory effect was also tested for NYC, NAY and LDF. This was conducted by creating a more elaborate effect of sex in which the females were grouped according to number of foals. However, this only had minor impact on the results (slightly smaller genetic variance and heritabilities; results not shown) compared to presented results. Therefore, and because number of foals is also expected to be favourably associated with talent and performance, the more elaborate sex effect was disregarded. 4.3. Results compared to literature The heritability of the longevity traits was low to moderate in following order NYC oNAY oNAYF oLDP (Table 2). As NAYF and LDP included information related to capacity and talent in addition to longevity, the higher heritability for these traits was expected. The mean longevity and genetic parameters for the longevity traits in the present study were quite similar compared to earlier published studies. Braam et al. (2011) found a mean NAY for males to be 3.3, while it was 3.2 in present study (Table 1). The heritabilities of NYC were identical with the earlier study by Ricard and Blouin (2011) being 0.10 in both. For NAY earlier studies have found heritabilities of 0.07–0.17 (Braam et al., 2011) and 0.20 (Jönsson

132

T. Seierø et al. / Livestock Science 184 (2016) 126–133

et al., 2014b) while it was found to be 0.11 in the current study. The changes can be due to differences among populations as Braam et al. (2011) studied competing geldings, Jönsson et al. (2014b) studied horses with young horse results, while the current study was based on competing horses regardless of gender independent of whether the horse participated in young horse conformation and performance tests. Jönsson et al. (2014b) found the heritability of lifetime performance to be 0.24, compared to 0.31 in the present study. Early prediction of accurate genetic merit for longevity is crucial for genetic progress. As the trait itself is not recorded until late in life, the information to predict accurate genetic merit could come from genomics, correlated traits recorded early in life or both. The genetic correlations between the studied conformation traits and longevity in show jumping were generally low. Hence, the conformation traits were found less valuable as longevity indicators. Only topline (which includes hindquarter) and hindlimbs had some, genetic association with NAY (rg was 0.10 and 0.23, respectively; Table 3). This suggests that young horses with a balanced conformation and a good hindquarter tends to be able to last longer as active show jumping horses, similar to results of the SWB (Jönsson et al., 2014b), but also that genetic variation for longevity, as measured here, is influenced by many other factors. 4.4. Implications and future challenges The young horse jumping traits had high value as indicator traits for longevity regardless of whether NAY or LDP was considered as the longevity trait of interest. This is similar to earlier study results of strong correlations between young horse talent traits and success in dressage competitions later in life within the same population (Jönsson et al., 2014a). However, their value as indicator trait for LDP was slightly lower. Capacity contributed most information to longevity EBVs, showing that horses with high capacity in young horse tests had better prerequisites to stay in competition for a long time. Capacity breeding values alone could yield an accuracy of NAY breeding values of 0.44 for younger horses (rg, capacity, NAY
rIA, capacity, Table 4), while the NAY accuracy in the same horses increased to 0.48 in a bivariate analysis (Table 5), where the parent average for NAY also contributed with information, and to 0.49 when information from additional indicator traits was also incorporated. This should be compared to an accuracy of 0.32 which can be achieved for younger horses in a univariate longevity evaluation through parent information. Also among older horses, with own NAY records, the mean accuracy improved from 0.43 to 0.46 when including indicator traits, as did the model predictive ability. The improvements in predictive ability were likely due to a reduced impact of pre-selection when young horse results on horses without longevity records were included in multi-trait genetic evaluations. The optimal model for NAY was found to be a 3-variate analysis with NAY, capacity and rideability/cooperativeness, while it for LDP was a 3-variate analysis with LDP, capacity and canter (Table 5). The genetic correlations with the young horse jumping traits were quite similar for the two longevity traits, where only rideability/cooperativeness showed a difference with 0.66 for NAY and 0.51 for LDP (Table 3). A reason for this difference may be that easy rideable horses were preferred by the riders, maybe to a larger degree the amateurs, and was therefore kept longer in the sport on low and less demanding levels of competition, whereas LDP also includes talent which is less affected by the rideability. Another explanation could be that a highly rideable horse may be less exposed to physical strains during training, in terms of number, length and intensity of sessions, to reach a competable state, which may improve the length of the active career (NAY). The mean accuracies of EBVs for longevity of younger horses without own longevity records were

about half of theoretical maximum and the EBVs were slightly biased despite the efforts to exclude censored data from the genetic evaluations. Nevertheless young horse traits are the only available information to base selection decisions upon this early in life (3–4 years of age). Delaying selection until the horse has an own longevity record is not an efficient alternative as it would roughly triple the generation interval and therefore result in much less genetic progress for longevity. Further, signs of bias were substantially reduced by including the young horse jumping traits in multi-trait genetic evaluations, due to inclusion of earlier phenotypic information with less environmental influence, and with less expose to pre-selection. Further, the use of a multi-trait animal model allows the full use of all family information to estimate the goal trait. This study only focused on participation in show jumping shows. Termination of the jumping career does not necessarily mean that it is no longer fit for competing. It might be competing in another discipline, used as leisure horse or for breeding. The latter is supported by earlier study results in other horse populations, that mares have a slightly shorter competitive life (Ducro et al., 2009; Jönsson et al., 2014b) but a longer lifetime span from birth to death (Wallin et al., 2000), compared to males. In the present study, account is taken for breeding mares, but a more complex measure of longevity which also takes into account the cause of termination of the sports career as well as the duration and intensity of other activities may provide a trait definition that is more in line with breeder expectations of longevity. However, it would be challenging in practice to record the necessary information as well as combining all information appropriately during evaluation. Further, longevity defined as number of starts may be considered in the future, when sufficient amount of data has been recorded (began in 1998). In such a scenario, competition level should also be accounted for. Number of starts has, to the authors knowledge, not been applied to riding horses, possibly due to lack of records, but has been used for trotters (Arnason et al., 1982).

5. Conclusion NAY was found to be the best of studied measurements for longevity in show jumping to use when a trait with least dependency of talent is desired as for Danish Warmbloods. LDP is useful if a combined longevity and performance trait is desired. The estimated heritability for NAY was 0.11 and for LDP it was 0.31. The estimated genetic correlations between young horse jumping traits and longevity traits for show jumping were moderate to high (rg: 0.56–0.74 for NAY and 0.51–0.71 for LDP). The multivariate analysis for NAY improved the accuracy for younger horses from 0.32 to 0.49 when NAY, capacity and rideability/cooperativeness were analysed simultaneously in a 3-variate analysis. A 3-variate analysis including LDP, capacity and canter was best for LDP and increased the accuracy of LDP breeding values from 0.39 to 0.52 while reducing bias.

Conflicts of interest Authors have no conflicts of interest.

Acknowledgements This research is part of the GenHors project which is partly funded by the Innovation fund in Denmark and Asta og Jul. P.

T. Seierø et al. / Livestock Science 184 (2016) 126–133

Justesens Fund. Data was provided by the Danish Warmblood Association and the horse section at SEGES.

References Arnason, T., Darenius, A., Philipsson, J., 1982. Genetic selection incides for Swedish trotter broodmares. Livest. Prod. Sci. 8, 557–565. Boelling, D., 2011. Description of competition index (in Danish). Word-document from the Danish warmblood association, 6pp. (received 27.09.13). Braam, Å., Näsholm, A., Roepstorff, L., Philipsson, J., 2011. Genetic variation in durability of Swedish warmblood horses using competition results. Livest. Sci. 142, 181–187. Ducro, B., Gorissen, B., van Eldik, P., Back, W., 2009. Influence of foot conformation on length of competitive life in a Dutch warmblood population. Equine Vet. J. 41, 144–148. Ducrocq, V., Casella, G., 1996. A Bayesian analysis of mixed survival models. Genet. Sel. Evol. 28, 505–529. Hennessy, K.D., Quinn, K.M., Murphy, J., 2008. Producer or purchaser: different expectations may lead to equine wastage and welfare concerns. J. Appl. Anim. Welf. Sci. 11, 232–235. Jensen, J., Mäntysaari, E.A., Madsen, P., Thompson, R., 1997. Residual maximum likelihood estimation of (co)variance components in multivariate mixed linear models using average information. J. Indian Soc. Agr. Stat. 49, 215–236. Jönsson, L., Christiansen, K., Holm, M., Mark, T., 2014a. Genetic correlations between young horse and dressage competition results in Danish warmblood horses. In: Proceedings of the 10th World Congress of Genetics Applied to Livestock Production (WCGALP), Vancouver, BC, Canada, 17–22 August 2014, No. 791. Jönsson, L., Egenvall, A., Roepstorff, L., Näsholm, A., Dalin, G., Philipsson, J., 2014b. Health status and conformation of young Swedish warmblood riding horses –

133

associations to longevity and lifetime competition performance. J. Am. Vet. Med. Assoc. 244, 1449–1461. Koenen, E.P.C., Aldridge, L.I., Philipsson, J., 2004. An overview of breeding objectives for warmblood sport horses. Livest. Prod. Sci. 88, 77–84. Madsen, P., Jensen, J., 2013. An user’s guide to DMU. A package for analysing multivariate mixed models. Version 6, release 5.2. Danish Institute of Agricultural Science, Department of Animal Breeding and Genetics Research center Foulum, Tjele, Denmark, pp. 1–32. Nielsen, U.S., Pedersen, O.M., Toivonen, M., 2003. Time dependent effects as source of bias in estimating breeding values for longevity and fertility traits. Interbull Bull. 30, 29–34. Posta, J., Rudiné Mezei, A., Mihók, S., Mészéros, G., 2014. Evaluation of the length of competitive life in Hungarian sport horses. J. Anim. Breed. Genet. 131, 529–535. Pritchard, T., Coffey, M., Mrode, R., Wall, E., 2013. Understanding the genetics of survival in dairy cows. J. Dairy Sci. 96, 3296–3309. R Core Team, 2013. R: A language and environment for statistical computing. R. Found. Stat. Comput. Reverter, R., Golden, B., Bourdon, R., Brinks, J., 1994. Technical note-detection of bias in genetic predictions. J. Anim. Sci. 72, 34–37. Ricard, A., Blouin, C., 2011. Genetic analysis of the longevity of French sport horses in jumping competition. J. Anim. Sci. 89, 2988–2994. Velie, B.D., Hamilton, N.A., Wade, C.M., 2015. Heritability of racing durability traits in the Australian and Hong Kong Thoroughbred racing populations. Equine Vet. J. http://dx.doi.org/10.1111/evj.12436. Viklund, Å., Braam, Å., Näsholm, A., Strandberg, E., Philipsson, J., 2010. Genetic variation in competition traits at different ages and time periods and correlations with traits at field tests of 4-year-old Swedish Warmblood horses. Animal 4, 682–691. Wallin, L., Strandberg, E., Philipsson, J., Dalin, G., 2000. Estimates of longevity and causes of culling and death in Swedish warmblood and coldblood horses. Livest. Prod. Sci. 63, 275–289.