We generated a DLA-wide SNP genotyping tool for fine mapping of an association between HOD in Weimaraners and DRB1 and tested two additional breeds with previously published DRB1/DQA1/DQB1 associations, Pug dogs with NME (Greer et al., 2010
) and Nova Scotia duck tolling retrievers with hypoadrenocorticism (Hughes et al., 2010
). Genotyping using the DLA-wide panel confirmed a strong association between NME in Pug dogs and the DLA; however, DLA associations in Nova Scotia duck tolling retrievers with hypoadrenocorticism and Weimaraners with HOD were not replicated using this expanded set of markers.
These results are in agreement with two previous publications: (1) a negative linkage analysis between Nova Scotia duck tolling retrievers with hypoadrenocorticism and the DLA class II region (Hughes et al., 2010
); and (2) localization of NME of Pug dogs to the DLA region on cfa12 by a genome-wide association (GWA) study (Greer et al., 2010
). Since the DLA-wide SNP panel assays the entire DLA region of 2.5 Mb containing 53 genes, including the MHC class II DRB1, DQA1 and DQB1 genes, for evidence of associations, it is unlikely that a true genetic association would be missed.
The reproducible association between DLA and NME of Pug dogs demonstrates that genetic associations can be identified using SNPs across the DLA. While SNPs from the entire DLA region were significantly associated with NME of Pug dogs, a plot of χ2 by physical position () shows that the high χ2 values are not only observed around DLA class II, but remain high across the extended DLA region. This implies that the association with NME in the Pug dog might be located downstream of, or outside, the extended DLA region on CFA12.
Since population stratification (different degrees of kinship between the case and control groups) could lead to spurious associations (Hinds et al., 2004
; Wang, 2009
; Wu et al., 2011
), this should be considered as a potential cofounding factor in canine association studies. However, it is unlikely that the inconsistency in the DLA association results could be explained by population structure, because identical samples were used for the analyses. A possible explanation for the observed spurious associations could be the small number of DRB1/DQA1/DQB1 alleles in the Nova Scotia duck tolling retriever and Weimaraner breeds.
Although the DRB1/DQA1/DQB1 genes are highly polymorphic across different breeds of dogs, only limited numbers of alleles (typically 4–7 for each locus) have been identified within a range of breeds tested (Ollier et al., 2001
; Angles et al., 2005b
; Kennedy, 2007
; Kennedy et al., 2006a
; Catchpole et al., 2008
; Wilbe et al., 2009
; Hughes et al., 2010
). In Nova Scotia duck tolling retrievers, Hughes et al. (2010)
identified seven different DRB1/DQA1/DQB1 haplotypes, comprising five DRB1 alleles, four DQA1 alleles and six DQB1 alleles. In Weimaraners, we identified nine different DRB1/DQA1/DQB1 haplotypes comprising three DRB1 alleles, four DQA1 alleles and five DQB1 alleles (). The DRB1/DQA1/DQB1 haplotypes reported by Kennedy et al. (2007a)
comprised 100 DRB1 alleles, 60 DQB1 alleles and 26 DQA1 alleles identified across 80 different dog breeds. Out of 360 dogs from 25 breeds typed for the DRB1/DQA1/DQB1 alleles, Angles et al. (2005b)
found that 40% were homozygous at the DRB1 site, 52% at the DQA1 and 44% at the DQB1 site.
Considering these findings, a significant statistical association within a breed population with a limited number of haplotypes should be considered cautiously and proof of a biological effect or linkage analysis should be sought as part of the validation of the statistical result. To illustrate that a spurious association is attainable in a breed with a small number of DLA alleles (conventionally tested), an association analysis was performed for Weimaraner sex (a trait determined by the presence or absence of the Y chromosome) and the DLA class II DRB1/DQA1/DQB1 alleles. DRB1 alleles associated with a ‘high risk’ of being a male (DRB1*1501: OR 2.68; P = 0.025) as well as with a ‘protective effect’ (DRB1*1201: OR 0.34; P = 0.013) were identified in this example of a spurious association. When the same association was performed using the DLA-wide SNP panel, no significant association was identified.
It is likely that the restricted number of DRB1/DQA1/DQB1 alleles and haplotypes do not segregate equally within a breed population due to inbreeding practices, use of popular sires, and/or genetic drift. Since both linkage disequilibrium (LD) and allelic content vary for different HLA haplotypes (Horton et al., 2008
), this may be true for the DLA. The exon 2 ‘alleles’ may represent different haplotypes that appear to be identical due to convergence (Seddon et al., 2010
). When Seddon et al. (2010)
sampled SNPs from the MHC class II region to evaluate canine cases and controls for type I diabetes, it was concluded that the published class II association with type I diabetes (Kennedy et al., 2006a
), could be outside the DRB1/DQA1/DQB1 exon 2 region or within a region larger than just exon 2. It is also possible that associations are more complex than those involving a single haplotype (Seddon et al., 2010
). Differentiation of each of these possibilities should be possible when genotyping SNPs across the extended DLA region for disease association studies.
We recognized that DRB1/DQA1/DQB1 exon 2 typing may give rise to spurious associations due to restricted numbers of haplotypes within breeds. This implies that exon 2 typing may not always be an informative tool when used to assess DLA diversity and that there is risk of misleading results due to focal low polymorphism. A DLA-wide SNP panel that spans the entire 2.5 Mb of the DLA region may present a more appropriate tool for the analysis of MHC diversity in pedigree dog populations.
The MHC region harbors the highest density of genes in the genome, yet most, if not all, of the genetic associations point to the class II classic genes in humans (Feder et al., 1996
; Lie et al., 1999a
; Undlien et al., 1999
; Fugger and Svejgaard, 2000
; Roudier, 2000
; Weyand and Goronzy, 2000
; Ota et al., 2001
; Undlien et al., 2001
; Turner and Colbert, 2002
; Raymond et al., 2005
; Thorsby and Lie, 2005
; Vandiedonck et al., 2005
; Ettinger et al., 2006
; Gorodezky et al., 2006
; Falorni et al., 2008
; Eike et al., 2009
) and dogs (Ollier et al., 2001
; Kennedy et al., 2006a
; Catchpole et al., 2008
; Wilbe et al., 2009
). In humans, an obstacle to localizing disease-predisposing genetic variants within the HLA is the strong LD typical of this region (Malfroy et al., 1997
), presumably due to selection (Raymond et al., 2005
LD or non-random association of alleles at adjacent HLA loci presents a difficulty in trying to establish which HLA locus is the primary disease-predisposing gene and which only appears to be associated because of LD to the primary associated gene (Feder et al., 1996
). LD in the MHC is considered to be complex in comparison with other genomic regions demonstrated by events of deletions, duplications, and unequal crossing-over that have been documented within an inbred mouse strain (Kumánovics et al., 2002
LD in dog breeds extends across 0.4 to 3.2 megabases depending on the size of the breed population and breed-specific history (Sutter et al., 2004
; Lindblad-Toh et al., 2005
). The long regions in LD are homozygous for breed-specific phenotypes, and show a mosaic pattern of short ancestral haplotype blocks that are shared between different breeds (Karlsson et al., 2007
). It is expected that disease association studies of DLA haplotypes will be affected by breed-specific long LD blocks as well as the restricted number of breed-specific DLA haplotypes, giving associations with linked but non-causative mutations (Seddon et al., 2010
In humans, studies targeting the entire HLA region have led researchers to identify associations outside of the classic class II genes DRB1/DRA/DQA1/DQB1. For example, type 1 diabetes mellitus, an autoimmune disease with strong genetic associations to the HLA-DRB1/DQA1/DQB1 genes (Karlsson et al., 2007
) has non-HLA risk loci identified by GWA (Wallace et al., 2010
) and a meta-analysis of SNPs located across the extended HLA region enabled the identification of genetic associations to four potential causal loci (TCF19,POU5F1, CCHCR1 and PSORS1C1), all within a 56 kb region (Cheung et al., 2011
). While it is presumed that strong LD within the HLA is a confounding factor in identifying associated genes outside DRB1/DRA/DQA1/DQB1, this could be overcome, as illustrated in this study, by utilizing novel analyzing algorithms for SNPs across the extended HLA region (Cheung et al., 2011