The objective of this study was to identify QTL associated with milk production traits and SCS in Holstein-Friesian cattle from a low-input production system. Both a frequentist and a Bayesian statistical approach were employed to test for association between genotypes and phenotypes. The QTL identified using both the sire and cow populations were spread across all 29 autosomes; the location and frequency of these QTL were in general agreement with those previously reported [3
A large number (1,529) of significant associations were detected across all traits. The majority of these significant associations were located within known QTL for the trait of interest. This shows that our methodology is effective in detecting associated regions of the genome. Also, our findings will help to further refine QTL regions previously detected with microsatellites [47
]. The detection of a large number of known QTL regions in our study would suggest that a large number of QTL that are important in high-input, confinement, concentrate-based systems are also important in low-input, pasture-based systems such as ours. In spite of this, 276 novel SNPs were detected in the sires using the single SNP regression approach. Of these novel SNPs, a number of promising clusters of SNPs were identified for each trait which may indicate potential new QTL regions. These regions include an area of chromosome 13 significantly associated with milk yield and fat percentage. Also, significant novel associations were detected on chromosome 20 for fat yield and somatic cell score close to the GHR and PRLR genes reported to be associated with milk production traits and SCS [38
]. In addition, particular areas of interest were separately detected for protein yield and percentage on chromosomes 8 and 15, respectively. These genomic regions may consist of QTLs that are unique to or advantageous in a low-input system such as ours.
Several significant associations, both within and outside known QTL regions, were detected for SCS. However, associations were considerably less numerous and weaker compared with those for the milk production traits. This may have been due to several inherent problems with the SCS phenotype resulting in reduced power to detect associations. Firstly, the reliability of the SCS proofs, which is an indicator of the amount of information available for an animal, was lower than that of the milk production traits in both the sires and cows. Decreased reliability of SCS means greater uncertainty as to the true breeding value of the animal for that trait. Furthermore, in the sire population, there were 138 fewer animals used to test for associations with SCS which would also decrease the power to detect significant associations. In addition, the lower heritability of SCS when compared to that of the milk production traits may also contribute to the weaker associations identified for SCS (i.e. a greater number of animals may be required for SCS
The Bayesian analysis used provided a number of advantages/alternatives to the standard single SNP regression approach. This Bayesian approach fits all markers in the analysis simultaneously and it was noticeable that this approach detected only 1-2 significantly-associated SNPs where the single SNP regression detected a cluster of numerous significantly-associated SNPs (i.e. on chromosome 14 for fat percentage). Additionally, the ability to allow a priori
information to be factored into the statistical model appears to have merit where different traits mat be controlled by varying numbers of genetic variants [49
]. The prior appears to be robust, with similar genomic regions detected as significant even when using different priors.
Genome-wide association studies are susceptible to detection of false positives due to the large number of statistical approaches being performed. One method to confirm or validate a SNP association/QTL is via replication of the association in a separate population as the probability of detecting the same associated variant in two separate populations is small [50
]. Our use of a separate population of Holstein-Friesian cows allowed validation of a number of associations from the sires, however, the size of this population was probably insufficient to validate SNPs of smaller effect (i.e. power was lower). Also the reliabilities of all traits in the cows were much lower than those of the sires resulting in potentially less accurate phenotypes to quantify the associations.
A number of SNPs in this analysis were significantly associated with more than one trait suggesting that genes with pleiotropic action may have been detected. Typically, in this study, a SNP affected multiple production traits with no association with SCS. Examples of this are three SNPs which were significantly associated with all five production traits (Additional file 3
). This indicates that certain regions of the genome may affect various different production-related traits and this should be taken into consideration when selecting animals for a particular breeding goal. In addition, four SNPs were significantly associated with a production trait and SCS, in particular three SNPs on chromosome 20 were associated with a concurrent decrease in milk yield and SCS. This observation agrees with the well-known positive correlation that exists between milk yield and SCS [51
]. Of these three SNPs, two lie in close proximity to the PRLR gene which has been reported to be associated with milk production [48
] and changes in SCC [52
]. QTL regions such as this may help elucidate how to select for increased milk yield without the associated detrimental effect on resistance to mastitis.