S. Lykkjen, Department of Basic Sciences and Aquatic Medicine, Norwegian School of Veterinary Science, Post-box 8146 Dep., N-0033 Oslo, Norway. E-mail: email@example.com
Osteochondrosis (OC), a disturbance in the process of endochondral ossification, is by far the most important equine developmental orthopaedic disease and is also common in other domestic animals and humans. The purpose of this study was to identify quantitative trait loci (QTL) associated with osteochondrosis dissecans (OCD) at the intermediate ridge of the distal tibia in Norwegian Standardbred (SB) using the Illumina Equine SNP50 BeadChip whole-genome single-nucleotide polymorphism (SNP) assay. Radiographic data and blood samples were obtained from 464 SB yearlings. Based on the radiographic examination, 162 horses were selected for genotyping; 80 of these were cases with an OCD at the intermediate ridge of the distal tibia, and 82 were controls without any developmental lesions in the joints examined. Genotyped horses descended from 22 sires, and the number of horses in each half-sib group ranged from 3 to 14. The population structure necessitated statistical correction for stratification. When conducting a case–control genome-wide association study (GWAS), mixed-model analyses displayed regions on chromosomes (Equus callabus chromosome – ECA) 5, 10, 27 and 28 that showed moderate evidence of association (P ≤5 × 10−5; this P-value is uncorrected i.e. not adjusted for multiple comparisons) with OCD in the tibiotarsal joint. Two SNPs on ECA10 represent the most significant hits (uncorrected P =1.19 × 10−5 in the mixed-model). In the basic association (chi-square) test, these SNPs achieved statistical significance with the Bonferroni correction (P =0.038) and were close in the permuted logistic regression test (P =0.054). Putative QTL on ECA 5, 10, 27 and 28 represent interesting areas for future research, validation studies and fine mapping of candidate regions. Results presented here represent the first GWAS of OC in horses using the recently released Illumina Equine SNP50 BeadChip.
Osteochondrosis (OC) is a common and clinically important disease that affects developing joints in horses as well as cattle, pigs, dogs, poultry and humans (Olsson & Reiland 1978). OC is defined as a disturbance in the process of endochondral ossification (Rejnö & Strömberg 1978). At the epiphyseal growth cartilage, the disturbance can lead to the formation of partially or completely detached fragments (osteochondrosis dissecans (OCD), fissures, or subchondral bone cysts (Rejnö & Strömberg 1978; Ytrehus et al. 2007). Lesions occur at predilection sites that are specific to the species and joint in question (Ytrehus et al. 2007). By causing joint inflammation and osteoarthritis, OC is a frequent cause of pain, lameness and reduced performance in young athletic horses (McIlwraith 1993), dogs (Harari 1998) and humans (Schenck & Goodnight 1996). In swine, OC is regarded as the most important cause of leg weakness (Jørgensen et al. 1995).
Microsatellite-based whole-genome scans in Hanoverian WB and SGC have identified a moderate number of quantitative trait loci (QTL) on different chromosomes linked to OC and OCD in tibiotarsal, metacarpo- and metatarsophalangeal joints (Dierks et al. 2007; Wittwer et al. 2007b). A QTL on Equus callabus chromosome (ECA) five for fetlock OCD in Hanoverian WB has been refined (Lampe et al. 2009), and candidate genes associated with OCD in fetlock joints and OC in fetlock and hock joints in SGC are reported on ECA 4 and ECA 18, respectively (Wittwer et al. 2008, 2009). In this study, we performed a whole-genome scan, using the newly developed Illumina Equine SNP50 BeadChip® (San Diego, CA, USA), to identify loci associated with osteochondrosis dissecans at the intermediate ridge of the distal tibia in Norwegian SB trotters. We hypothesized that the high-density BeadChip offered a good opportunity to identify QTL associated with OCD status.
Materials and methods
Study population and phenotypes
Radiographic data and blood samples were obtained from 363 Norwegian SB trotter yearlings in 2007 and from 101 yearlings in 2008, representing 50% and 14% of the total cohort, respectively. Offspring of sires that had 10 or more progeny in 2006 were chosen as the sample population. The combined 464 horses were offspring of 22 different sires and 424 dams, and the number of animals per sire varied from 9 to 55. At examination, the animals had a mean (±SD) age of 12.1 (±1.8) months, a minimum age of 8.0 months and a maximum age of 17.9 months.
The horses were reared by their breeders, and feeding, housing and exercise levels varied among the animals. Because of extensive use of artificial insemination, the examined progeny of most sires were geographically well distributed in Norway and thus representative of the Norwegian SB population. All veterinary interventions were in accordance with the Norwegian Animal Welfare Regulations.
Horses were examined radiographically at the Norwegian School of Veterinary Science (NSVS) (n = 110) and by staff from NSVS at nine regional equine clinics (n = 354). Digital radiography was the predominant examination technique, but conventional radiography was performed at two clinics (n = 107). Before examination, the horses were given detomidine (Domosedan vet®, Orion; 1 mg/100 kg I.V.) and butorphanol (Torbugesic®, Fort Dodge; 1 mg/100 kg I.V.).
The radiologic examination comprised ten views to reveal OC, OCD and other bony fragments in the metacarpo- and metatarsophalangeal joints and the tibiotarsal joint. Lateromedial (90°) views were used for the metacarpophalangeal joint. The metatarsophalangeal joints were examined by dorso (45°) proximal (35°) lateral-plantarodistomedial and dorso (45°) proximal (35°) medial-plantarodistolateral views. The tibiotarsal joints were examined by dorso (45°) medial-plantarolateral oblique and dorso (30°) lateral-plantaromedial views.
Phenotypic traits – Predilection sites and interpretation
All radiographs were scrutinized by two of the authors, a professor in equine surgery and an experienced equine veterinary surgeon, for the presence of all types of bony fragments and OC. In the tibiotarsal joints, three predilection sites were evaluated bilaterally for OC and OCD; the intermediate ridge of the distal tibia, the lateral trochlear ridge of the talus and the medial malleolus of the tibia. Criteria for OC (Butler et al. 2008) included (i) the presence of irregular texture of the bone with variable radiopacity (e.g. radiolucency of the subchondral bone) and (ii) changes of the regular bone contour such as an irregularly flattened, smaller or larger concavity at the predilection site. OCD was diagnosed when isolated radiodense areas (osteochondral fragments) in the joint space were visible at the predilection sites. A corresponding defect in the underlying bone often accompanied the OCD lesion.
The metacarpophalangeal and metatarsophalangeal joints in particular were evaluated for OC/OCD at the dorsal sagittal ridge of the third metacarpal bone, and osteochondral lesions at the dorsoproximal rim of the proximal phalanx, the palmar and plantar processes of the proximal phalanx, the attachment sites of the short distal sesamoidean ligaments, and also at the proximal sesamoid bones.
The genome-wide association analysis (GWAS) encompassed 162 of the examined yearlings (Table S1), including 80 quite uniform cases (39 females, 41 males), preferably with OCD just at the intermediate ridge of the distal tibia, and 82 controls (40 females, 42 males) without any developmental lesions in the joints examined. One hundred and twenty-eight of these yearlings were born in 2006 and 34 in 2007. They were offspring of 22 sires and 158 dams, and the number of animals per sire varied from 3 to 14.
For genotyping, 5–10 μg genomic DNA (100 μl, 50–100 ng/μl) was isolated from 400 μl of proteinase-K-treated EDTA blood using an animal blood and tissue kit (QIAGEN). The DNA was measured using a Nanodrop® spectrophotometer ND-1000 (Thermo scientific, Delaware, CO, USA). The Equine SNP50 BeadChip® (Illumina) was used for genotyping. The chip includes 54 602 evenly distributed SNPs (average probe spacing of 43.2 kb across all autosomes) derived from the EquCab2.0 SNP collection (http://www.broad.mit.edu/mammals/horse/), which were discovered in light coverage sequencing of a single horse of each of the following breeds: Arabian, Andalusian, Akhal-teke, Icelandic, SB, TB and Quarter horse. In SB, the BeadChip has an average call rate of 99.55%, 45.715 polymorphic loci (MAF > 0.05) and average MAF of 0.20.
The SNP array analysis of the 162 Norwegian SB trotter yearlings was performed by the Mayo Clinic’s Shared Genotyping Resource (Rochester, MN, USA). The laboratory protocols were all according to the manufacturers’ instructions and used Illumina’s software to call the genotype data.
Genotype quality assurance
All data were subject to quality control procedures, which were as follows. First, only samples with a minimum call rate of 95% were included. With an average genotype call rate of 99.18%, all samples met this criterion. Second, we discharged the SNPs with: (i) a call rate < 95% in the total sample (n =470); (ii) those deviating from Hardy–Weinberg equilibrium in cases and controls (P <0.001), or having a differential case/control missing of P < 0.01 (n =88); and (iii) MAF < 0.05 in the total sample (n =13 265). A total of 41 170 SNPs passed our quality control criteria.
Osteochondrosis dissecans cases and healthy controls were compared by a basic association (chi-square) test in the whole-genome association analysis toolset PLINK; the Bonferroni correction was initially applied to address the problem of multiple comparisons. When deviation from expected distribution of P-values was demonstrated with quantile–quantile plots, suggesting inflation of the chi-square statistic because of population stratification, association with the disease phenotype was evaluated using logistic regression with sire treated as a covariate, using Cochran–Mantel–Haenszel (CMH) test 2 × 2 × K (K = 16) and finally by using mixed-model analyses (Yu et al. 2006).
Logistic regression analysis in PLINK was followed by 10 000 t-max permutations. The CMH test in PLINK was based upon 16 clusters generated by complete linkage agglomerative pairwise identity-by-state (IBS) clustering. The clustering process was modified by restrictions based on significance distance (P =0.01) and a phenotype criterion (i.e. all clusters must contain at least one case and one control). Mixed-model analyses were performed in R using the pedigreemm package. The model includes a correction for population structure based on individual inbreeding coefficients as well as pairwise values of kinship. The relationship matrix was constructed by using the option ‘know pedigree relationships’ as implemented in pedigreemm.
Quantile–quantile (QQ) plots showing the difference between expected and observed test statistics, and whole-genome association plots of significance, were generated in R. The association plots display the results for each chromosome as a negative log10 of the P-value. For genome-wide association studies, uncorrected P-values <5 × 10−7 provide strong evidence of association, whereas uncorrected P-values between 5 × 10−5 and 5 × 10−7 are considered to provide moderate evidence (Wellcome Trust Case Control Consortium 2007). When applying the Bonferroni correction or permutation, P-values ≤0.05 were considered to provide strong evidence of association (Hirschhorn & Daly 2005). Genes and homologue regions in equine and human genomes were identified by the NCBI map viewer and Basic Local Alignment Search Tool (BLAST) (http://www.ncbi.nlm.nih.gov/mapview/), the Ensemble genome browser (http://www.ensembl.org/index.html) and the UCSC genome browser (http://genome.ucsc.edu/).
The mixed-model statistics of the GWAS analysis revealed seven SNPs that showed moderate evidence of association with OCD status (P* ≤5 × 10−5; −log P*≥ 4.30; the asterisk (*) denotes an uncorrected P-value i.e. not adjusted for multiple companisons); these SNPs were located on ECA 5, 10, 27 and 28 (Fig. 1). Table 1 gives significant test results (P ≤ 5 × 10−5) with different test statistics, SNP positions on the chromosome, minor allele frequencies (MAF), odds ratio (OR) and genes within1 Mb of actual SNPs.
Table 1. Genomic regions associated with osteochondrosis dissecans at the distal intermediate ridge of the tibia in Norwegian Standardbred trotters.
On ECA 5, a single SNP (BIEC2-920265) showed moderate significant association in the mixed-model, the chi-square test and in the logistic regression statistics (Table 1).
Two SNPs on ECA 10 (BIEC2-132748 and BIEC2-132753) represent the most significantly associated hits in both the mixed-model (P* = 1.19 × 10−5), the chi-square statistics (P* = 9.31 × 10−7; P =0.038 after Bonferroni correction) and also in the logistic regression statistics (P* = 1.28 × 10−5; P =0.054 after t-max permutation) (Table 1). For both SNPs, the minor allele had an odds ratio (OR) of 0.23 (95% confidence interval (CI); 0.13–0.43) and allele frequencies of 0.10 and 0.32 in cases and controls, respectively (Table 1). The SNPs on ECA 10 were neighbours and flanked by two less significantly associated SNPs (Table 1).
Two SNPs on ECA 27 (BIEC2-721410 and BIEC2-722382) showed moderate evidence of association with OCD status in the mixed-model, but not in the chi-square or the logistic regression statistics (Table 1). The associated SNPs on ECA 27 were located approximately 1 Mb apart, and BIEC2-721410 was flanked by two less significantly associated SNPs (Table 1).
The neighbouring SNPs on ECA 28 (BIEC2-744792 and BIEC2-744794) displayed moderate evidence of association with OCD status in the mixed-model (P* =4.66 × 10−5), but not in the chi-square or the logistic regression statistics (Table 1). Downstream, these SNPs were flanked by two less significantly associated SNPs (Table 1).
In the chi-square statistics, an additional seven SNPs on ECA 1, 3, 5, 10, 18 and 28 were found to be moderately associated with OCD status (Table 1).
The chi-square and logistic regression models showed evidence that the results were influenced by population substructure (identified by QQ-plots showing differences between expected and observed results, Fig. 2). This was not unexpected because of uneven distribution of cases and controls in the different half-sib family groups. The mixed-model and CMH analyses showed partial and almost full accounts for population substructure (Fig. 2).
After conducting a case–control genome-wide association study to identify loci associated with OCD status in Norwegian SB trotters, we identified seven SNPs located on ECA 5, 10, 27 and 28 (Table 1) that in the mixed-model showed moderate evidence of association (P* < 5 × 10-5; the astrerisks (*) denotes an uncorrected P-value i.e. not adjusted for multiple comparisons). The moderate number of QTL on different chromosomes found for OCD in the tibiotarsal joint, suggests that several genes are possibly involved in the development of the condition (Table 1).
SNPs on ECA 10 (BIEC2-132748 and BIEC2-132753) represent the most significant hits in the mixed-model, as well as in the other statistical models (Table 1). The SNPs achieved statistical significance in the basic association (chi-square) test with Bonferroni correction, and they also were close in the permuted logistic regression test (Table 1). The fact that the SNPs on ECA 10 are neighbours and also flanked by two less significantly associated SNPs (Table 1) increases the likelihood that they represent a true QTL associated with OCD in the tibiotarsal joint.
Disease-causing variants are seldom directly typed in GWAS studies, and the pattern of linkage disequilibrium (LD) is of importance when looking for candidate genes. LD in the horse is moderate, with long-range haplotype sharing among breeds (Wade et al. 2009). Table 1 lists candidate genes within 1 Mb of the significantly associated SNPs, although it is possible that genes further away can be involved, as the length of haplotypes in the population (or in different areas of the genome) is as yet unknown. LOC100073151 (protein coding gene similar to serum/glucocorticoid regulated kinase), located 0.15 Mb upstream of SNP BIEC2-132748 (Tables 1 & 2), represents a predicted gene closest to the associated SNPs. In a homologous region on human chromosome 6, the serum/glucocorticoid regulated kinase 1 gene (SGK1) can be identified (Table 2). SGK1 encodes a serine/threonine protein kinase that covers a wide variety of physiological functions and plays an important role in cellular stress response (Table 2). In rat brain, an increased level of SGK1 gene expression has been demonstrated after both focal and global brain ischaemia (Lu et al. 2003; Nishidaa et al. 2004). Ischaemic necrosis seems to play an important role in the aetiopathogenesis of OC (Carlson et al. 1995; Ytrehus et al. 2004a,b; Olstad et al. 2008), implying that SGK1 is potentially relevant to the pathogenesis of the disease.
Table 2. Candidate genes and their function.
LOC100073151 (protein-coding gene similar to serum/glucocorticoid regulated kinase)
In the syntenic region on human chromosome 6, the homologue Serum/glucocorticoid regulated kinase 1(SGK1) can be identified (position; 134,490,387-134,639,196bp). SGK1 encodes a serine/threonine protein kinase that plays an important role in cellular stress response. This kinase activates certain potassium, sodium and chloride channels, suggesting an involvement in the regulation of processes such as cell survival, neuronal excitability and renal sodium excretion
Chloride channel, calcium activated, family member 4 (CLCA4)
CLCA4 encodes a calcium-sensitive chloride conductance protein (chloride channel), reported to be found in epithelial cells, neurons, cardiac and smooth muscle cells, as well as blood cells. The channels modulate excitability in neurons and muscle cells, they are involved in signal transduction in olfactory receptor cells. In epithelial cells, they play an important role in transepithelial transport.
Collagen, type XXIV, alpha1 (COL24A1)
COL24A1 encodes collagen type XXIV. COL24A1 is found to be expressed in the forming skeleton of the mouse embryo and is also transcribed in the trabecular bone and periosteum of the newborn mouse. (Matsuo et al. 2008).
F-box protein 25 (FBXO25)
FBXO25 encodes a member of the F-box protein family. F-box proteins are determinant in ubiquitin-mediated proteolysis and are positioned as key regulators in many pathways of cell signalling, transcription and cell cycle
TBC1 domain family, member 22A (TBC1D22A)
The gene encodes a GTPase-activating protein that regulates small GTPases (guanine nucleotide–binding proteins). GTPases are proteins involved in cellular signal transduction. The small GTPases regulate a wide variety of processes in the cell, including growth, cellular differentiation, cell movement and lipid vesicle transport
On ECA 5, a single SNP (BIEC2-920265) located within the chloride channel, calcium activated, family member 4 (CLCA4) gene yielded moderate significant associations in all test statistics (Tables 1 & 2). Lampe et al. (2009) have identified collagen type XXIV alpha1 (COL24A1) as a potential candidate gene responsible for fetlock OC in WB horses (Table 2). COL24A1 is located 0.36 Mb downstream of BIEC2-920265, which might also reflect correspondence between this gene and OCD in tibiotarsal joints of Norwegian SB.
The two associated SNPs on ECA 27 (BIEC2-721410 and BIEC2-722382) are located upstream of the gene F-box protein 25 (FBXO25) (Table 2). The SNPs displayed moderate evidence of association with OCD status in the mixed-model, but are not supported with corresponding results in the remaining test statistics (Table 1). BIEC2-721410, being flanked by two less significantly associated SNPs, contributes supporting evidence that the SNPs on ECA 27 represent a true QTL associated with the disease (Table 1).
Moderately associated SNPs on ECA 28 (BIEC2-744792 and BIEC2-744794) are located within the gene TBC1 domain family, member 22A (TBC1D22A) (Table 2). Their statistics are quite similar to those encountered for SNPs on ECA 27; moderate evidence of association is only displayed in the mixed-model, but the fact that the SNPs are neighbours and flanked downstream by two less significantly associated SNPs suggests that they represent a true QTL for OCD.
Previous studies have reported QTL, potential candidate genes and differentially expressed genes related to OC in different horse breeds, pigs and humans. Dierks et al. (2007) performed a genome-wide search for microsatellite markers associated with OC/OCD in Hanoverian WB horses. Although the study identified a moderate number of QTL on different chromosomes, only two QTL on ECA5 represent a putative correspondence between the previous and the present studies. In a follow-up study, Lampe et al. (2009) reported refinement of a QTL on ECA5 and identification of COL24A1 as a potential candidate gene for OCD in fetlock joints. As previously mentioned, COL24A1 might also be involved in the pathogenesis of OCD at the intermediate ridge of the distal tibia in Norwegian SB. Wittwer et al. (2007b) mapped QTL for OC, OCD and palmar/plantar osseous fragments in fetlock joints of SGC horses. Numerous QTL were identified by the use of 250 microsatellite markers, but they were all different from the ones observed in the present study. An expression study in SB foals identified two upregulated genes, the equine tousled-like kinase 2 (TLK2) and an unknown gene (EST: CD465746.1) (Austbøet al. 2010). The genes on ECA 11 and 13 do not correspond to QTL identified in the present study.
In pigs, Andersson-Eklund et al. (2000) identified QTL for OC on Sus scrofa chromosomes (SSC) 5 and 13. The QTL on SSC 13 did not correspond to any QTL identified in horses here. The QTL on SSC 5 is homologous to human chromosome 12q14-q24, which harbours the ALX homeobox 1 gene (ALX1). A BLAST of the human ALX1-region identifies ALX1 on ECA 28 at position 12.03–12.05 Mb. The large interval (approximately 30 Mb) between ALX1 and the QTL on ECA 28 observed in the present study gives no support to a putative correspondence. Laenoi et al. (2010) found downregulation of the matrix Gla protein gene (MGP) in OC compared to healthy cartilage and suggested that MGP might play an important role in the pathogenesis of OC in pigs. MGP on SSC 5 did not correspond to any QTL identified in horses here.
In humans, a missense mutation in the aggrecan C-type lectin domain gene disrupts extracellular matrix interactions in cartilage and causes dominant familial osteochondritis dissecans (Stattin et al. 2010). The region on human chromosome 15 harbouring the mutated gene aggrecan (ACAN) shares homology with ECA 1, but does not correspond to QTL identified in the present study.
A possible explanation for the differences in the QTL identified between studies could be that the trait analysed is associated with different loci in different species and breeds. However, the fact that a genomic region is associated only in one horse breed does not necessarily mean the region is not a general risk factor, rather the genomic markers may have different levels of penetrance in different breeds. In addition, a lack of statistical power caused by too few animals and/or markers could result in important markers falling below the significance thresholds, or smaller risk factors being missed (Hirschhorn & Daly 2005). A huge difference in marker coverage in microsatellite-based versus SNP chip-based genome scans represents a third explanation for the unequal study results. The equine SNP chip provides an evenly distributed and approximately 250-fold increase in genomic coverage and should therefore be well suited to detect QTL in GWAS (Wade et al. 2009). On the other hand, thousands of markers represent a multi-testing problem, possibly resulting in more false positive results (Balding 2006).
Various phenotypic criteria applied when selecting cases can be another explanation for the different results, as lesions at different predilection sites are possibly influenced by different genes. An apparent advantage of the present study is that the GWAS sample came from a homogenous population containing cases with a stringent phenotype (McCarthy et al. 2008). Also, as radiographic identification of OC and OCD lesions can be challenging, including only horses with OCD at the intermediate ridge of the distal tibia might facilitate correct identification of cases.
The results of gene expression studies in OC are especially difficult to interpret. As long as the aetiopathogenesis, timeline and sequence of events in OC have not been clarified, it is difficult to know whether the differentially expressed genes reflect events prior to the disturbance in enchondral ossification, the disease process itself, or secondary repair processes within the bone. Subsequently, a lack of correlation between the present study and past expression studies might simply reflect different stages in the disease process.
The goal of GWAS studies is to identify patterns of polymorphisms that vary systematically between individuals with different disease states and could therefore represent the effects of risk-enhancing or protective alleles (Balding 2006). However, the genome is so large that patterns that are suggestive of causal polymorphism could well arise by chance. To help distinguish causal from spurious signals, tight standards for study design, quality control and statistical approach need to be established. A potential limitation of our study is that the GWAS results may have been affected by discrete population structure, as most QQ-plots (Fig. 2) showed some difference between observed and expected results (Hirschhorn & Daly 2005; McCarthy et al. 2008). Dealing with association within but not between individual strata, the IBS-clustered CMH test successfully accounted for population substructure, but resulted in less significant SNP associations. As mixed-model analyses are considered to be better corrections of population stratification in GWAS (Balding 2006; Yu et al. 2006), the interpretation of our results is primarily based on these statistics.
In conclusion, this is the first report of applying the Equine SNP50 BeadChip® (Illumina) for genome-wide association analysis of OCD in horses. The study identified several SNPs that showed evidence of association with OCD in the tibiotarsal joint in Norwegian SB trotters. Putative QTL on ECA 5, 10, 27 and 28 represent the most significant hits and would be interesting areas for future research. Further studies must include replication and validation of these study results by including a larger number of animals or joint (meta-) analysis of data from comparable GWAS. When confirmed, refinement of candidate regions will eventually follow.
The current study was funded by the University of Minnesota, the Norwegian School of Veterinary Science and the Norwegian Horse Center. Funding for the equine SNP chips was provided by the US Department of Agriculture (National Research Initiative Competitive Grants Program) and the Morris Animal Foundation. The authors are indebted to Prof. Theo Meuwissen and Morten Mattingsdal for assistance in selecting samples for genotyping and interpretation of analyses. We gratefully acknowledge the generous assistance of the staff at the Section for Equine Medicine and Surgery, Radiology and Genetics of the Norwegian School of Veterinary Science, as well as staff at the Department of Veterinary and Biomedical Sciences of the University of Minnesota.