|Home | About | Journals | Submit | Contact Us | Français|
Coat color and type are essential characteristics of domestic dog breeds. Although the genetic basis of coat color has been well characterized, relatively little is known about the genes influencing coat growth pattern, length, and curl. We performed genome-wide association studies of more than 1000 dogs from 80 domestic breeds to identify genes associated with canine fur phenotypes. Taking advantage of both inter- and intrabreed variability, we identified distinct mutations in three genes, RSPO2, FGF5, and KRT71 (encoding R-spondin–2, fibroblast growth factor–5, and keratin-71, respectively), that together account for most coat phenotypes in purebred dogs in the United States. Thus, an array of varied and seemingly complex phenotypes can be reduced to the combinatorial effects of only a few genes.
The tremendous phenotypic diversity of modern dog breeds represents the end point of a >15,000-year experiment in artificial and natural selection (1, 2). As has been demonstrated for traits such as body size (3) and coat color (4), marker-based associations with phenotypic traits can be explored within single breeds to initially identify regions of genetic association, and then expanded to multiple breeds for fine-mapping and mutation scanning (5, 6). Coat (pelage) phenotypes are particularly amenable to this strategy as they show a huge amount of variation across breeds but still allow for simple variation within single breeds (7). This offers a unique strategy for advancing the genetic understanding of a complex phenotype.
We used the structured pattern of fur variation in dogs to localize the genetic basis of three characteristics of the canine coat: (i) the presence or absence of “furnishings,” the growth pattern marked by a moustache and eyebrows typically observed in wire-haired dogs; (ii) hair length; and (iii) the presence or absence of curl. To accomplish this, we generated three genome-wide single-nucleotide polymorphism (SNP) data sets using the Affymetrix version 2.0 canine SNP chip (8, 9). The first data set consisted of 96 dachshunds segregating three coat varieties: wire-haired with furnishings, smooth, and long-haired without furnishings. The second data set comprised 76 Portuguese water dogs (PWDs), segregating the curl phenotype. The final data set, termed CanMap, included 903 dogs from 80 breeds representing a wide variety of phenotypes. An additional data set used to map furnishings included a panel of microsatellite markers (10), genotyped on a 96-dachshund pedigree segregating all three coat varieties.
The same strategy was used to map all three traits. First, a genome-wide association study (GWAS) within a breed segregating the phenotype was conducted to determine the most strongly associated locus. To rule out false-positives caused by population structure within the breeds (11), we did a second GWAS that used the CanMap data set divided into cases and controls based on the presence or absence of the phenotype in question. Fine-mapping of significant, concordant peaks was used to define the smallest shared haplotype, followed by sequencing to identify the putative causative mutations. Each mutation was validated in a large panel of at least 661 dogs from 108 breeds, including cases and controls for all phenotypes (table S1).
We initially mapped furnishings in the dachshund using smooth-coated and long-haired dogs as controls and wire-haired dogs as cases (Fig. 1A). Single-marker analysis of the dachshund GWAS data set and concurrent linkage analysis of the dachshund pedigree identified the same locus on canine chromosome 13 (CFA13) surrounding nucleotide 11,095,120 [P = 3.4 × 10−27, lod score (logarithm of the odds ratio for linkage) = 5.6; Fig. 1B]. We confirmed the association on CFA13 in the CanMap data set at nucleotide 11,659,792 (P = 10−241; Fig. 1C and table S2). A 718-kb homozygous haplotype in all dogs fixed with furnishings was located within both the original 3.4-Mb haplotype observed in the dachshund-only GWAS, and a 2.8-Mb haplotype identified in crossover analysis within the dachshund pedigree (Fig. 1D).
Fine-mapping allowed us to reduce the homozygous region to 238 kb spanning only the R-spondin–2 (RSPO2) gene, excluding the 5′ untranslated region (5′UTR) and the first exon (Fig. 1D, fig. S1, and table S3). RSPO2 is an excellent candidate for a hair-growth phenotype as it synergizes with Wnt to activate β-catenin (12), and Wnt signaling is required for the establishment of the hair follicles (13, 14). Moreover, the Wnt/β-catenin pathway is involved in the development of hair-follicle tumors, or pilo-matricomas (15), which occur most frequently in breeds that have furnishings (16). Recent studies have shown that a mutation in the EDAR gene, also involved in the Wnt pathway, is responsible for a coarse East-Asian hair type found in humans (17), with some similarity to canine wirehair.
All exons and conserved regions of RSPO2 were sequenced in dogs from seven breeds (table S4). Only an insertion of 167 base pairs (bp) within the 3′UTR at position 11,634,766 was perfectly associated with the furnishings trait in dogs from both the case/control study and the extended pedigree (table S5). The result was further confirmed in a set of 704 dogs of varying phenotypes. In total, 297 of 298 dogs with furnishings were either homozygous (268) or heterozygous (29) for the insertion, and all 406 dogs lacking the trait were homozygous for the ancestral state, as is consistent with a dominant mode of inheritance (table S1).
This mutation does not affect the protein-coding region of the RSPO2 gene. However, because the 3′UTR frequently encodes elements that influence mRNA stability [reviewed in (18)], we examined whether the insertion was associated with a change in the expression level of the RSPO2 gene. We found a threefold increase in RSPO2 transcripts in muzzle skin biopsies of dogs with furnishings, consistent with a transcript effect (fig. S2).
We applied the same mapping strategy to hair length. Previously, mutations in the FGF5 gene were identified in Welsh corgis segregating an atypical “fluffy” or long-haired phenotype (19) and associated with excess hair growth in mice and cats (20–22). Our study replicates these findings in an extended breed set. Indeed, association analyses in both the dachshund and CanMap data sets highlight the region on CFA32 containing FGF5 with P values of 3 × 10−27 and 9 × 10−44, respectively. After fine-mapping, a 67-kb homozygous region highlighted the FGF5 gene (Fig. 2A, fig. S3, and table S6). The strongest association was observed at position 7,473,337 (P = 1 × 10−157), in which a highly conserved Cys is changed to Phe (Cys95→Phe) in exon 1 of FGF5, consistent with the previous study (19). Sequencing within the homozygous haplotype revealed no SNPs with stronger association (table S7).
This diagnostic SNP was typed in several hundred additional dogs of varying hair length. Within the dachshunds, all long-haired dogs had the TT genotype, whereas all short or wire-haired dogs had either the GT or GG genotypes, suggesting a recessive mode of inheritance, as predicted previously (23). Across all breeds, the T allele was found in 91% of the long-haired dogs, in only 3.9% of the short-haired dogs, and accounts for ~30% of genotypes found in medium-haired dogs. Three breeds with very long hair, including the Afghan hound, neither carry the Cys95→Phe variant nor show an association with CFA32, suggesting that additional loci exist that contribute to hair length in dogs (table S1).
To identify the gene that causes curly coat, we conducted a GWAS using PWDs (Fig. 2B) and identified a single associated SNP at position 5,444,030 on CFA27 (P = 4.5 × 10−7). A SNP in close proximity (5,466,995; P = 6.9 × 10−28) was associated with curly coat in the CanMap data set. Fine-mapping revealed a shared homozygous haplotype that included two keratin genes (Fig. 2C, fig. S4, and table S8). Sequence data covering 87% of the homozygous region identified one SNP at position 5,542,806 that segregated with the trait. Non–curly haired dogs carried the CC genotype; curly coated dogs had the TT genotype. In breeds where the trait segregates, such as PWDs, all three genotypes were observed. The relevant SNP is located in the KRT71 gene (previously called K6irs1, Kb34, and K71) and causes a nonsynonymous Arg151→Trp alteration (table S9). Genotyping an additional 661 samples at this SNP validated the association (P = 3 × 10−92) (table S1).
Keratins are obvious candidates for hair growth [reviewed in (24)], and mutations in KRT71 have been described in curly coated mice (25). The mutation described in our study is within the second exon of the gene and may affect either or both of two protein domains: a coiled-coil and a prefoldin domain (www.ensembl.org/Canis_familiaris/). Conceivably, sequence alterations in these domains could affect cellular targeting, receptor binding, or proper folding of the protein after translation [reviewed in (26)].
Notably, these three mutations in various combinations explain the observed pelage phenotype of 95% of dogs sampled, which include 108 of the ~160 American Kennel Club (AKC)–recognized breeds. A total of 622 dogs representing all identifiable coat phenotypes were genotyped at all three loci (table S10). By analyzing each of the three major traits both within and across multiple breeds, we show that combinations of these genotypes give rise to at least seven different coat types, encompassing most coat variation in modern domestic dogs (Fig. 3). Specifically, short-haired breeds display the ancestral state in all three genes. Wire-haired breeds, all of which have furnishings, carry the RSPO2 insertion. Dogs that carry both the RSPO2 and KRT71 mutations display “curly-wire” hair that is similar in texture to wire-hair but longer and curled or kinked rather than straight. Long-haired breeds carry the variant form of FGF5. Dogs carrying the FGF5 mutation, along with the RSPO2 insertion, have furnishings and long soft coats, rather than wiry ones. When dogs carry variants in both FGF5 and KRT71, the pelage is long and curly. Not surprisingly, coats must be of sufficient length to curl, and all curly haired dogs in our study were homozygous for the FGF5 mutation. Finally, if all three mutations are present, the phenotype is long and curly with furnishings.
None of the mutations we observed were found in three gray wolves or the short-haired dogs, indicating that short-haired dogs carry the ancestral alleles (table S1). Our finding of identical haplotypes surrounding the variants in all dogs displaying the same coat type suggests that a single mutation occurred for each trait and was transferred multiple times to different breeds through hybridization. Because most breeds likely originated within the past 200 years (27), our results demonstrate how a remarkable diversity of phenotypes can quickly be generated from simple genetic underpinnings. Consequently, in domesticated species, the appearance of phenotypic complexity can be created through combinations of genes of major effect, providing a pathway for rapid evolution that is unparalleled in natural systems. We propose that in the wake of artificial selection, other complex phenotypes in the domestic dog will have similar tractable architectures that will provide a window through which we can view the evolution of mammalian form and function.
We gratefully acknowledge support from NSF grants 0733033 (R.K.W.) and 516310 (C.D.B.), NIH grants 1RO1GM83606 (C.D.B.) and GM063056 (K.G.L. and K.C.), the Nestlé Purina company, the AKC Canine Health Foundation, the University of California–Davis Veterinary Genetics Laboratory, and the Intramural Program of the National Human Genome Research Institute. We thank L. Warren and S. Stafford for providing pictures. Finally, we thank the many dog owners who generously provided us with samples from their pets.