The results from this study demonstrate the potential power of EBV to improve the predicted accuracy of selection against hip and elbow dysplasia in many dog breeds in the UK, including 3 of the 10 most popular breeds. The mean accuracies of EBV are always higher than would be obtained via selection on available phenotypes (using either individual or both parental phenotypes). Furthermore, a far greater proportion of juvenile animals have EBV with a higher accuracy than can at present be obtained by selection using both parental phenotypes. Thus, reliable information is available on much more of the population than currently exists, which will allow breeders to make more accurate selections earlier in the life of the dog. The accuracy of selection is directly linked to genetic progress, meaning more accurate selection will lead to greater progress in breeding for health. We have demonstrated this to be the case in a wide range of breeds. The broader impact can be realised by noting that the average annual number of registrations of the 15 breeds included in this study, and so that will each have an EBV, is in excess of 80,000, approximately 1/3 of all annual registrations with the UK Kennel Club.
Substantially faster genetic improvement is expected to come via both increased accuracy and greater selection intensity, as the provision of EBV could have a major impact on the ways in which dogs are selected by breeders and pet owners (accompanied by appropriate user information). Currently mate selection is based on ancestral phenotypes and two dogs’ own phenotypes (if known). Using EBV owners of breeding bitches would be able to more accurately assess the genetic merit of potential sires resulting in an improved response to selection, whether phenotypes are available or not. In addition, EBV will be available for all registered animals of the breed, increasing selection intensity opportunities. For example: the projected time to achieve an improvement of 5 points in the median hip score via phenotypic selection, under the guidelines which were in place for the majority of the period covered by the data, range from 30 to over 300 years (NEWF and BEARD respectively) mainly due to weak selection intensity . Although these guidelines have now been amended to promote selection from below the median rather than the mean phenotype, the opportunity to increase selection intensity is more readily presented by EBV (their universality within a breed removing the random sampling of genetic risk from the use of un-scored animals). Selecting breeding stock with EBV below the breed mean is projected to achieve such an improvement in between 9 years for NEWF and 18 years for BEARD. The increases in the proportion of animals with breeding value accuracies greater than that provided by parental phenotypes illustrate that EBV provide, per phenotype, more information on more animals, enabling wider comparison by breeders. An additional benefit from publishing EBV could be the indirect introduction of selection pressure through potential pet owners more accurately differentiating the genetic risk of hip (and elbow) dysplasia among available litters.
It is crucial however that participation in the BVA/KC screening schemes continues – the availability of EBV does not mean scoring is no longer necessary. Phenotypes are the basis of accurate breeding values, and accuracies will rapidly decline if phenotypic information were to become sparse. Theory predicts that EBV accuracy would be expected to increase with participation, and a plot and regression of mean EBV accuracy at birth on the mean annual proportion of sires with hip scores provides empirical support (Additional file 5: Figure S1). Experience in livestock sectors reinforces the theory, where widespread and routine data collection and very large family sizes (i.e. thousands of progeny) can yield EBV accuracies of >0.9, although, it must be noted, accuracies are rarely so high at the time of selection. The resulting message to breeders is simple: continued scoring will maintain and further enhance the accuracy of selection of breeding stock for healthy joints, as well as increasing the pool of animals with reliable information. Moreover, the phenotypic score is of value to breeders and pet owners alike in providing an indicator of not only the genetics but the environmental influence on an individual animal’s hip/elbow joints. While the EBV should guide breeding decisions, the phenotype is useful to inform the appropriate care of the dog that may ameliorate the severity of hip and elbow dysplasia where it occurs.
The accumulation of phenotypes will be particularly critical for future analyses of elbow dysplasia, where the extent of recording is much less than for hip dysplasia, and since elbow score is less heritable than hip score (possibly due in part to the collection of traits described by the elbow score). This study only managed to detect a genetic correlation between hip and elbow scores with enough precision to be statistically significantly different to zero in the LAB. Previously, we demonstrated that bivariate analysis of hip and elbow data can confer significant benefits to the accuracy of EBV for elbow scores, where a favourable genetic correlation exists . Additional elbow score data will be essential to determine more precise genetic correlations between hip and elbow score in BMD, GR, GSD and ROTT, although reported estimates from other studies indicate there may be wide variation across breeds [9, 14, 16].
While genetic parameters are often (correctly) viewed as specific to each breed, questions can arise as to whether the genetic parameters (h2 and rA) from one breed may be useful in BLUP analyses (EBV calculation) of another. This is particularly relevant where small population size means that breed-by-breed parameter calculation is not feasible. The analysis of 15 breeds in this study using the same model provided a good opportunity to explore this matter. Results from the meta-analysis indicate that there is more between breed variation in estimates of genetic correlation between hip and elbow score than for heritability of elbow score, across the five breeds for which both traits were analysed. While additional elbow scoring data will therefore be expected to result in more consistent estimates of heritability across breeds as sampling variance is reduced, the estimates of genetic correlation between hip and elbow score are expected to reflect the greater between breed variation in the true parameter. The slightly higher estimate of between breed variance of heritability for hip score compared to that for elbow score may reflect the greater number of breeds included in the analysis for that trait, and inclusion of additional breeds not currently in the sample may prove an outlier to this current collection. Nevertheless, results from the meta-analysis suggest that the heritability of both hip and elbow score are remarkably consistent across breeds, and that most of the observable variation in estimates is due to sampling variation. The across breed estimate of the residual correlation between hip and elbow score is small (with a small s.e., 0.024 ± 0.035), and the meta-analysis revealed only small between breed variation in such estimates (Additional file 6: Table S3). This implies that across breeds there is a large degree of independence in non-genetic environmental risk factors on dysplasia of the hip and elbow joint. This finding across multiple breeds supports an earlier observation on the small environmental correlation between hip and elbow score in LAB  and is somewhat surprising given that both dysplasias are developmental orthopaedic diseases.
All breeds included in this study showed an improving genetic trend with respect to hip and elbow score, except the SHUSK, suggesting that phenotypic selection to date has had a small but beneficial impact. The increasing genetic propensity towards hip dysplasia in the SHUSK was matched by the phenotypic trend (regression of total hip score on date of birth showing a yearly rise of 0.075 score points), which has been observed previously . However, the SHUSK had the best hip scores of all the 15 breeds analysed here. It may be that the historical role of the SHUSK as a sled dog has entailed de facto selection against lameness, but that increasing popularity as pets or show dogs has weakened this tacit selection pressure. The popularity of the breed in the UK has risen quickly recently, from 829 registered in 2000 to 2,209 in 2010. While the general hip condition of the SHUSK remains better than for many other breeds, breeders should be aware of the detrimental trend. It serves as an example that the transition to a popular pet breed be accompanied by tools, such as EBV, that protect the qualities of the breed for which it is valued.
The results presented here indicate that the GDN has been subject to the greatest selection intensity for reduction in hip score, equivalent to excluding the 18% of animals with the worst hip scores from breeding. This is in line with former breeding guidelines based on the mean hip score and has been accompanied by a phenotypic decline in hip score of over 0.6 points per year (from regression of total hip score on date of birth) and a fall in the mean hip score from 24.35 in 1990 to 14.77 in 2010. The GDN is not a numerous breed, with a mean of 324 dogs registered per year from 2000–2010, but appears to have a large proportion of breeders committed to including health traits in selection objectives; for example over 80% of sires and dams undergo hip scoring. While slightly greater genetic progress was observed in the NEWF, a larger estimate of heritability and shorter generation interval meant that the derived selection intensity was smaller than for the GDN. However across all breeds and traits, regression of genetic gain on the proportion of breeding animals scored did not show significant association (P > 0.05). This demonstrates that quantity of data alone does not guarantee genetic improvement, but that it must be accompanied by the appropriate breeding advice and the motivation by breeders to act upon it. Across comparable breeds, the rates of genetic progress calculated in this study were broadly typical of those that have been previously reported .
Substantial improvements in the predicted accuracy of selection, and therefore genetic progress, based on estimating breeding values have been quantifiably demonstrated here for a wide range of breeds, including a number of the more uncommon breeds. For the more uncommon breeds, selection against diseases such as hip dysplasia is more problematic when based on phenotypes alone as there may be only a small number of the candidates with a record, and so making a small breed smaller. Therefore an approach to increasing numbers of candidates with usable information, as demonstrated here, should be welcome. Rarer breeds are more likely to suffer the effects of genetic over-contribution of some animals to future generations, usually through the widespread use of popular sires. Where selection does take place in small populations (which it must do to improve welfare where hip dysplasia is prevalent, as argued in the introduction) a balance must be struck between genetic progress in reducing the burden of disease on the one hand, and minimising the risk of the emergence of a novel genetic disease on the other, which can be measured by the rate of inbreeding. The inbreeding coefficient per se was found to be largely unrelated to, and have only a small effect on, hip and elbow score in this study. However, one drawback with the use of EBV based on pedigrees and phenotypes is that they too can promote greater rates of inbreeding in the course of generating more progress . This need not be inevitable, but instead places an emphasis on increasing awareness of inbreeding among breeders, and making more tools available to help them manage rates of inbreeding as EBV are introduced.
In this study we elected to conduct a deterministic prediction of the superiority of EBV accuracy over that of selection using phenotype. An alternative method would be to use simulation. However, simulations are stochastic and can be prone to error in some situations. A further disadvantage of simulation is a lack of insight into the underlying causes, which when encountered through deterministic use of empirical data can then be used in induction. The reported superiority of mean EBV accuracies over the accuracy of selection on individual hip score phenotype reported here were smaller than reported by Malm et al. using simulation , however there tended to be fewer animals with phenotypes in our data, implying less information. Comparison of EBV accuracy with selection on parental phenotypes shows the improvement was of similar magnitude.
EBV for hip and elbow dysplasia are routinely computed and published in Norway, Finland and Denmark for up to 38 breeds and in Sweden for 5 breeds (K Maki, personal communication), in Germany for GSD, and in the USA for LAB. The public release of EBV described in this study is anticipated in the UK in 2013. The abundance of EBV for hip and elbow dysplasia in so many countries raises the prospect of the globalisation of scoring and evaluation schemes. Analyses determining the genetic correlations between individual scoring protocols would enable dogs to be evaluated under any (participating) scheme (UK registered dogs evaluated under the FCI scheme and Scandinavian dogs participating in the BVA/KC scheme for example) while still having an EBV in the country of registration . It should be noted, however, that not all scoring protocols may be equal in terms of predicting the lameness associated with hip and elbow dysplasia and consequential OA . To address this further research focussing on identifying OA and lameness later in the life of scored dogs would be welcome. Fortunately, the manner in which EBV for canine health are presented offers an ‘outward continuity’, allowing improvements to be made to the computational model or to the evaluation protocol, as well as the utilisation of international data, without noticeable disruption to the end user .