Mikkola et al.

BMC Genomics (2019) 20:1027

https://doi.org/10.1186/s12864-019-6422-6

RESEARCH ARTICLE Open Access

Genetic dissection of canine hip dysplasia

phenotypes and osteoarthritis reveals three
novel loci
Lea Mikkola1,2,3, Saila Holopainen1,2,3,4, Tiina Pessa-Morikawa1, Anu K. Lappalainen4, Marjo K. Hytönen1,2,3,
Hannes Lohi1,2,3† and Antti Iivanainen1*†

Abstract
Background: Hip dysplasia and osteoarthritis continue to be prevalent problems in veterinary and human
medicine. Canine hip dysplasia is particularly problematic as it massively affects several large-sized breeds and can
cause a severe impairment of the quality of life. In Finland, the complex condition is categorized to five classes
from normal to severe dysplasia, but the categorization includes several sub-traits: congruity of the joint, Norberg
angle, subluxation degree of the joint, shape and depth of the acetabulum, and osteoarthritis. Hip dysplasia and
osteoarthritis have been proposed to have separate genetic etiologies.
Results: Using Fédération Cynologique Internationale -standardized ventrodorsal radiographs, German shepherds
were rigorously phenotyped for osteoarthritis, and for joint incongruity by Norberg angle and femoral head center
position in relation to dorsal acetabular edge. The affected dogs were categorized into mild, moderate and severe
dysplastic phenotypes using official hip scores. Three different genome-wide significant loci were uncovered. The
strongest candidate genes for hip joint incongruity were noggin (NOG), a bone and joint developmental gene on
chromosome 9, and nanos C2HC-type zinc finger 1 (NANOS1), a regulator of matrix metalloproteinase 14 (MMP14)
on chromosome 28. Osteoarthritis mapped to a long intergenic region on chromosome 1, between genes
encoding for NADPH oxidase 3 (NOX3), an intriguing candidate for articular cartilage degradation, and AT-rich
interactive domain 1B (ARID1B) that has been previously linked to joint laxity.
Conclusions: Our findings highlight the complexity of canine hip dysplasia phenotypes. In particular, the results of
this study point to the potential involvement of specific and partially distinct loci and genes or pathways in the
development of incongruity, mild dysplasia, moderate-to-severe dysplasia and osteoarthritis of canine hip joints.
Further studies should unravel the unique and common mechanisms for the various sub-traits.
Keywords: Hip dysplasia, Osteoarthritis, Dog, German shepherd, Genome-wide association study

Background is normal and E is severe CHD. In Finland, the FCI score

Canine hip dysplasia (CHD) is a common multifactorial is defined separately for both hip joints, hence the for-
hereditary disorder that has perplexed dog owners, mat is given as: left hip score / right hip score. The FCI
breeders as well as veterinarians and researchers for de- score is determined from different ‘sub-traits’ of the hip:
cades. A standardized system for CHD grading has been congruency of the joint, Norberg angle (NoA), sublux-
developed in the countries that belong to the Fédération ation degree of the joint, shape and depth of the acet-
Cynologique Internationale (FCI). The FCI score is di- abulum, and whether there are any visible signs of
vided into five categories alphabetically: A to E, where A osteoarthritis (OA) in the joint or not. FCI has derived
the grading rules, from which the Finnish Kennel Club
* Correspondence: antti.iivanainen@helsinki.fi
†
(FKC) has defined their guidelines for radiographing and
Hannes Lohi and Antti Iivanainen are co-senior authors.
1
Department of Veterinary Biosciences, University of Helsinki, P.O. Box 66
scoring hip dysplasia [1]. The above-mentioned sub-
(Mustialankatu 1), FI-00014 Helsinki, Finland
Full list of author information is available at the end of the article

© The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Mikkola et al. BMC Genomics (2019) 20:1027 Page 2 of 13

traits are not recorded for later use, only the hip score is 0.94, Fig. 1). However, all the observed associations
stored in the FKC database. throughout the loci were stronger for FHCDAE than for
As the FCI or any other combinatory score does not NoA (Table 1).
accurately correlate with the various CHD sub-traits, On chromosome 9, two SNPs demonstrated associ-
these have to be studied separately. NoA and femoral ation with FHCDAE (Fig. 2). One of these SNPs passed
head center position in relation to dorsal acetabular edge the threshold for significance with independent tests
(FHCDAE) reflect the incongruity of the hip joint, which (BICF2G630834826 with a P-value of 1.57 × 10–6, Table
impacts the development of CHD [2]. Hip joint laxity is 1). BICF2G630834826 and BICF2P742007 are located ~
a major contributor to the development of OA. How- 22 kb downstream and ~ 67 kb upstream of NOG encod-
ever, OA is suggested to develop due to many simultan- ing noggin (Additional file 2), and they are in high link-
eous pathologies, which influence the central structures age disequilibrium (LD) measured as the squared value
of the joint [3]. OA may have a distinct genetic back- (r2) of Pearson’s correlation coefficient between pairs of
ground in relation to the other hip sub-traits [4–6]. SNPs (r2 = 0.84, Additional file 3). These two SNPs also
The current consensus is that CHD is polygenic, and associated with NoA but the association was stronger
genetic contribution to the phenotype can vary from small for FHCDAE. The third SNP on chromosome 9, which
to moderate [7–14]. Variation between breeds is evident was observed only for NoA (BICF2G630837307) and
from several studies [5, 7, 9, 10, 14–16]. Some breeds are was not genome-wide significant, lies ~ 64 kb upstream
more susceptible to the disorder than others. Labrador Re- of LIM homeobox 1 (LHX1) (Additional file 2).
trievers [7, 10, 17], Bernese Mountain dogs [9], Golden Other loci with at least a suggestive association with the
Retrievers [18], and German Shepherds [4, 14, 16] have incongruity traits were on chromosomes 25 and 28 (Table
been under special interest in studies of CHD, and several 1, Fig. 2). On chromosome 25, BICF2G630468961 show-
genetic associations with different hip phenotypes have ing suggestive association with NoA was intronic to solute
been reported in these breeds. Different breeding strat- carrier family 7 member 1 (SLC7A1) (Additional file 2).
egies have been proposed to improve hip health; estimated On chromosome 28, SNPs BICF2P1046032 (in high LD
breeding values are generally considered the most efficient with BICF2P895332; r2 = 0.96, Additional file 3) demon-
approach [4, 19–22]. Also, newer methods like genomic strated significant association with FHCDAE (Table 1).
selection might bring a long awaited solution in the fight These SNPs located between CDK2 associated cullin do-
against this disorder [17, 23, 24]. main 1 (CACUL1) (~ 18 and 30 kb upstream, respectively)
To better understand the genetic etiology of CHD re- and nanos C2HC-type zinc finger 1 (NANOS1) (~ 163 and
lated phenotypes, we have carried out here a successful 174 kb upstream, respectively) (Additional file 2).
genome-wide association study (GWAS) in a cohort of
over 750 well-phenotyped German Shepherds to map loci OA maps to chromosome 1
for CHD and related sub-traits. We report three loci with We studied OA as a separate disorder. Two veterinar-
genome-wide significance and two suggestive loci for dif- ians in our group evaluated the radiographs of individual
ferent traits with physiologically relevant candidate genes. dogs for evidence of OA (see methods). The dogs exhib-
ited either no radiographic evidence of OA (controls) or
Results had mild, moderate or severe signs of OA (cases). A
The joint incongruity, measured as FHCDAE and NoA, case-control association analysis, where all controls
map to chromosomes 9, 25 and 28 (N = 492) were compared with all cases regardless of the
Incongruity of the hip joint contributes to CHD. There- severity of OA (N = 163), revealed a genome-wide sig-
fore, we carried out two different association analyses on nificant locus on chromosome 1 (Fig. 3). The SNP with
incongruity related traits, FHCDAE and NoA, which the strongest association (BICF2P468585) had a P-value
were assessed by two different veterinarians in our of 2.86 × 10–7 (Table 2). The second-best SNP
group. Both traits were measured for right and left hip, (BICF2P357728) reached a P-value of 8.93 × 10–7 (Table
but we used only the worst measure in the analysis. 2). Both SNPs passed the threshold for genome-wide sig-
NoA showed significant inter-observer variation in a lin- nificance based on the estimated number of independent
ear regression model (P = 0.028, Additional file 1), which tests determined with simpleM (1.82 × 10–6).
is consistent with earlier findings [25, 26]. Therefore, the The two genome-wide significant SNPs, as well as four
evaluator was included as a covariate in the association out of the six SNPs showing suggestive association with
analysis of NoA. For FHCDAE the inter-observer vari- OA on this chromosome, located between NADPH oxi-
ation was non-significant. The association results for dase 3 (NOX3) (except BICF2S23248027, which lies
FHCDAE and NoA indicated overlapping loci, which is within the ninth intron of NOX3) and AT-rich inter-
not surprising as these measurements were highly nega- action domain 1B (ARID1B) (Table 2, Additional file 2).
tively correlated in the study cohort (Pearson’s r = − The top SNPs BICF2P468585 and BICF2P357728 were
Mikkola et al. BMC Genomics (2019) 20:1027 Page 3 of 13

Fig. 1 Correlation plot of NoA and FHCDAE. NoA is on the Y-axis and FHCDAE on the X-axis. Above the correlation plot is the distribution of
FHCDAE measurements in the cohort. A respective distribution of the NoA measurements is on the right side of the correlation plot. Pearson’s
r = − 0.94 and P-value = 1.8 × 10–297

observed to be in high LD (r2 = 0.85, Additional file 3). LD between these two loci (r2 = 0.50–0.61, Additional
Otherwise, moderate to perfect LD (r2 = 0.63–1.00) was file 3). BICF2S23216908 located within the first intron of
observed between these six SNPs, even though the re- Transmembrane protein 181 (TMEM181) and
gion they covered was over 1.1 Mb long (Additional file BICF2S2305568 within the first intron of Dynein light
3). Thus, we concluded that these SNPs probably repre- chain Tctex-type 1 (DYNLT1).
sent just one locus that associates with the disorder. We also observed suggestive associations for chromosome
SNPs BICF2S23216908 and BICF2S2305568 (Table 2) 9 and 25 for OA. On chromosome 9, BICF2G630837240
are in perfect LD (r2 = 1.00, Additional file 3). Although locates ~ 101 kb downstream from MRM1 encoding Mito-
they are ~ 1.7 Mb away from the other SNPs that associ- chondrial RRNA Methyltransferase 1 and ~ 178 kb
ated with OA on this chromosome, we observed some upstream from LHX1 (Table 2, Additional file 2).
Table 1 Top SNPs from the GWAS on FHCDAE and NoA
Trait Chr Locus Alleles N per SNP SNP(s) P-value from FASTA(f), corrected with the
inflation factor lambda (λFHCDAE = 1.000, λNoA = 1.003)
FHCDAE 9 31,477,907 G/A 642 BICF2G630834826 1.57 × 10–6
31,387,114 A/G 642 BICF2P742007 2.13 × 10–6
28 29,111,565 A/C 627 BICF2P1046032 1.62 × 10–6
29,122,985 G/A 643 BICF2P895332 2.86 × 10–6
NoA 9 31,477,907 G/A 642 BICF2G630834826 2.22 × 10–6
31,387,114 A/G 642 BICF2P742007 5.26 × 10–6
36,694,174 A/G 640 BICF2G630837307 9.78 × 10–6
25 10,301,514 A/C 640 BICF2G630468961 9.66 × 10–6
Single-nucleotide polymorphisms (SNPs) with probability values (P-values) < 1.0 × 10–5 are listed. SNPs and the corresponding P-values are in bold font, if the P-
value passed the threshold for genome-wide significance (1.82 × 10–6) determined with the estimated number of independent tests from SimpleM (see methods).
(f)
Family-based score test for association (FASTA). Chr = chromosome. N refers to the number dogs in analysis after exclusion of dogs by FASTA because they lack
either the phenotype or a covariate. Covariates for NoA: age at radiographing, genetic cluster, genotyping batch, evaluator. Covariates for FHCDAE: age at
radiographing, genetic cluster, and genotyping batch. See also Additional file 1
Mikkola et al. BMC Genomics (2019) 20:1027 Page 4 of 13

Fig. 2 Manhattan plots for the analysis of hip joint incongruity traits FHCDAE and NoA. The upper Manhattan plot represents the results from the
analysis of FHCDAE (N = 643). The blue line indicates the threshold for significance based on the number of independent tests. The lower plot
represents the GWAS results of NoA (N = 642) with the blue line indicating the threshold for significance as in the upper plot

BICF2G630468961 on chromosome 25 is located within the dogs (Fig. 4 and Table 3). The SNPs with the strongest as-
second intron of SLC7A1 (Table 2 Additional file 2). sociation (BICF2P468585 and BICF2S23248027) passed
the threshold for significance with independent tests
Different genetic etiology of mild and moderate-to-severe (Table 3). The identified locus between NOX3 and
CHD ARID1B is the same we found for OA (Additional file 2).
To identify loci for CHD according to the FCI hip For the latter two case-control analyses with smaller num-
scores, we carried out three sets of case-control associ- ber of dogs, none of the associations reached genome-
ation analyses. In the first case-control analysis, the con- wide significance. BICF2G630837405 on chromosome 9
trols had a bilateral FCI hip score A and cases B/C, C/B lies within the eighth intron of apoptosis antagonizing
or bilateral FCI score C or worse (Ncases = 339, Ncontrols = transcription factor (AATF) and TIGRP2P126345 located
354). In the second analysis, the same controls were ~ 8 kb downstream from the same gene. These two SNPs
used but cases had a bilateral FCI score of D or worse are in high LD (r2 = 0.97, Additional file 3).
(Ncases = 166). In the third analysis we compared mild A summary of the genome-wide significant loci across
CHD dogs (B/C, C/B or bilateral FCI score C) with dogs CHD-related traits described above are listed in Table 4.
that had moderate-to-severe (at least an FCI score D or The frequencies of the effect and alternative alleles of
worse for either hip) CHD (Nmild = 124, Nmoderate-to-se- the significantly associated SNPs in cases and controls
vere = 216). The summary of the results of these three (binary analyses) are in Additional file 4. Some SNPs
comparisons are shown in Table 3. were associated with more than one trait, as expected
A genome-wide significant association was found on when the phenotypes are not independent from each
chromosome 1 for the first comparison with close to 700 other. The heritability (h2) estimates from the polygenic
Mikkola et al. BMC Genomics (2019) 20:1027 Page 5 of 13

Fig. 3 Manhattan plots for the binary trait: OA status. The Manhattan plot represents the lambda-corrected (lambda = 1.007) P-values from the
FASTA analysis of osteoarthritis (N = 655), where the blue line shows the threshold for significance with independent tests

mixed model for the different traits varied from 36 to require large and well-phenotyped study cohorts in each
64% (Additional file 5). breed. We report here a remarkable progress by mapping
three new loci on different chromosomes across key CHD
Discussion traits in German Shepherds. The locus on chromosome 1
CHD is a complex skeletal disorder and one of the leading associated with OA and the FCI hip score, and the loci on
clinical concerns in veterinary medicine. CHD is categoric- chromosomes 9 and 28 associated with the trait FHCDAE,
ally scored into five classes in screening programs of the which measures hip joint incongruity (Table 4). In addition
FCI member countries but the phenotype manifests many to the three loci with genome-wide significance, two sug-
sub-traits, which may eventually result in painful OA. The gestive loci on chromosomes 9 and 25 were uncovered for
development of OA itself is a complex process, which in- OA, NoA and different FCI hip score comparisons. Besides
volves alterations in many different tissues, including bone, revealing novel loci, the study indicates that the locus on
cartilage, synovial membrane and ligaments [27]. Given the chromosome 1 associates with two binary traits: OA and
complexity of the disorder, it is not surprising that genetic the FCI hip score with relaxed case definition (B/C, C/B, or
discoveries have also remained scarce and breakthroughs C or worse in both hips). Our study partially utilizes the

Table 2 Top SNPs from the GWAS on OA

Trait Chr Locus Alleles N per SNP SNP(s) P-value from FASTA, corrected with the
inflation factor lambda (λ = 1.007)
OA status (binary; osteoarthritis present or not) 1 45,382,633 C/A 655 BICF2P468585 2.86 × 10–7
46,279,297 A/G 650 BICF2P357728 8.93 × 10–7
46,268,586 C/A 654 BICF2P1037296 2.39 × 10–6
45,405,601 G/A 652 BICF2S23120955 4.48 × 10–6
45,161,186 G/C 655 BICF2S23248027 5.62 × 10–6
45,381,667 G/A 655 BICF2P392839 7.05 × 10–6
48,007,784 A/G 655 BICF2S23216908 7.81 × 10–6
48,065,207 A/G 655 BICF2S2305568 7.81 × 10–6
9 36,579,921 A/G 655 BICF2G630837240 4.69 × 10–6
25 10,301,514 A/C 652 BICF2G630468961 6.22 × 10–6
SNPs with P-values < 1.0 × 10–5 are listed. The SNPs and their corresponding P-values are in bold font if the P-value passed the threshold for genome-wide
significance (1.82 × 10–6) determined with the estimated number of independent tests from SimpleM (see methods). N refers to the number dogs in analysis after
exclusion of dogs by FASTA because they lack either the phenotype or a covariate. Covariates for the OA status: age at radiographing, genetic cluster, genotyping
batch and birth month. See also Additional file 1
Mikkola et al. BMC Genomics (2019) 20:1027 Page 6 of 13

Table 3 Top SNPs from the GWAS on different case-control analyses of the FCI hip score
Trait Chr Locus Alleles N per SNP SNP(s) P-value from FASTA, corrected with the
inflation factor lambda (λ = 1.010–1.024)
1st case-control analysis (normal hips 1 45,382,633 C/A 693 BICF2P468585 7.03 × 10–7
or mild-to-severe CHD; FCI scores) 45,161,186 G/C 693 BICF2S23248027 1.30 × 10–6
46,279,297 A/G 689 BICF2P357728 2.52 × 10–6
46,268,586 C/A 692 BICF2P1037296 9.51 × 10–6
2nd case-control analysis (normal hips 9 36,837,067 G/A 520 BICF2G630837405 4.96 × 10–6
or moderate-to-severe CHD; FCI scores) 36,886,621 A/G 517 TIGRP2P126345 7.75 × 10–6
3rd case-control analysis (mild CHD 9 36,837,067 G/A 340 BICF2G630837405 4.12 × 10–6
or moderate-to-severe CHD; FCI scores) 36,886,621 A/G 338 TIGRP2P126345 7.48 × 10–6
SNPs with P-values < 1.0 × 10–5 are listed. The SNPs and their corresponding P-values are in bold font if the P-value passed the threshold for genome-wide
significance (1.82 × 10–6) determined with the estimated number of independent tests from SimpleM (see methods). N refers to the number dogs in analysis after
exclusion of dogs by FASTA because they lack either the phenotype or a covariate. Covariates for the 1st analysis: age at radiographing and genotyping batch.
Covariates for the 2nd analysis: age at radiographing, genetic cluster, genotyping batch and birth month. Covariates for the 3rd analysis: genetic cluster,
genotyping batch and birth month. See also Additional file 1

study from Mikkola et al. (2019) [28] and as such cannot be modulates the activity of Rho-like proteins [36]. ARID1B,
regarded as an independent replication study. on the other hand, participates in transcriptional activa-
The locus on chromosome 1 lies in a long intergenic re- tion and repression through chromatin remodeling [37].
gion between NOX3 and ARID1B (Table 5) Neither of the Interestingly, ARID1B is associated with joint laxity via a
genes nor the intergenic region is known for functions that multisystemic Coffin-Siris syndrome (CSS); CSS is caused
could explain their role in the development of CHD or OA. by ARID1B variants and 66% of the CSS patients exhibit
However, the likely significance of this locus for CHD is joint laxity [38, 39].
highlighted by the fact that our previously observed sug- Previous studies have suggested seven different loci for
gestive association [28] was strengthened by over ten times OA, none of them overlapping our loci. A multi-breed
with a larger sample size. The association of the NOX3- study by Zhou et al. (2010) [5] suggested two loci on canine
ARID1B locus to OA was 2.5 times as strong as to the FCI chromosomes 17 and 37 for OA. Another quantitative trait
hip score (as assessed by the ratio of the P-values). The locus (QTL) study in a crossbreed experiment reported
latter is an aggregate phenotype and visible signs of OA (or putative QTLs on chromosomes 5, 18, 23 and 31 [6].
the lack of them) are part of its evaluation. Therefore, it is Chromosome 3 has also been suggested to harbor a QTL
not surprising to observe overlapping results. that regulates cranial and caudal acetabular osteophyte for-
NOX3 is a member of NADPH oxidases and an interest- mation in Portuguese Water Dogs [40]. Discrepancy to our
ing candidate for articular cartilage degradation. NADPH results may result from the genetic heterogeneity in differ-
oxidase participates in the generation of hydrogen perox- ent study populations, differences in analysis methods or
ide, which is used by myeloperoxidase as a substrate to phenotyping approaches in evaluating OA.
produce a highly reactive hypochlorous acid, and in some A locus in chromosome 9 near NOG associated with
circumstances chlorine gas [29, 30]. These two reactive the incongruity trait FHCDAE (Tables 4 and 5). The as-
molecules oxidize the pyridinoline cross-links of articular sociation of the loci with NoA were weaker than with
cartilage and initiate its degradation [29, 30]. The SNP FHCDAE. This is not surprising as NoA suffers from
BICF2P468585 with the strongest association is ~ 196 kb high inter-observer variability [25, 26], which was also
upstream from NOX3, but BICF2S23248027 (also known noted in our study. Similar bias was not seen for
as rs21911799) is located in the intron between NOX3 FHCDAE (Additional file 1). We previously found pro-
exons 9 and 10 (Tables 4 and 5). Moreover, NOX3 is tective regulatory variants upstream NOG, and demon-
mainly expressed in the inner ear and fetal tissues [31], strated the inverse correlation of their in vitro enhancer
thus, the role of NOX3 in synovial tissue inflammation re- activity with healthy hips in German Shepherds [28].
mains uncertain. Yet, among other protein-protein inter- The association of this locus with FHCDAE (as assessed
actions, a STRING [32] database search (Additional file 6) by the ratio of the P-values) was ~ 24 times as strong as
suggested possible interplay between NOX3 and matrix what we observed for the FCI hip score [28]. The puta-
metalloproteinases 2 and 9 – two matrix degrading en- tive contribution of NOG to FHCDAE remains elusive
zymes implicated in CHD and OA [33–35]. We have pre- but may offer some leads to reduced joint congruity. De-
viously discussed [28] that there is some evidence of the creased noggin activity could possibly strengthen the ac-
possible interplay between NOX3 and TRIO (trio Rho etabular bone via bone morphogenic protein (BMP)
guanine nucleotide exchange factor), another candidate signaling and help the repair of microfractures and other
gene for CHD [16]. The product of T-cell lymphoma inva- damage caused by mechanical wear in growing dogs.
sion and metastasis 2 (TIAM2) further upstream (Table 5) Interestingly, delayed ossification of the femoral head
Mikkola et al. BMC Genomics (2019) 20:1027 Page 7 of 13

Fig. 4 Manhattan plots for the case-control analyses of controls and mild to severe cases. The upmost Manhattan plot represents the case-
control analysis, where controls were dogs with an FCI score A/A and cases were dogs with an FCI score B/C, C/B, or C or worse on both hips
(N = 693). The second Manhattan plot represents the case-control analysis, where cases were dogs with an FCI score D or worse on both hips
(N = 520), and the lowest Manhattan plot is the comparison between mild cases (B/C, C/B, C/C) to moderate-to-severe cases (D or worse on both
hips) (N = 340). In each plot, the blue line shows the threshold for significance with independent tests

has been associated with CHD in later life [41, 42]. murine Nog results in osteopenia, bone fractures and
NOG is a crucial gene for many developmental pro- decreased bone formation, when the function of osteo-
cesses, such as neural tube fusion, joint formation and blasts becomes defective [47]. A recent study by Gha-
skeletal development [43, 44]. In humans, dominant dakzadeh et al. (2018) [48] showed that knocking-down
NOG mutations cause some congenital disorders with Nog in rats with small interfering RNA leads to down-
abnormal joints [45], and knocking out murine Nog regulation of Nog and increases both BMP-mediated
leads to a state where the mice lack most of the joints differentiation of osteoblasts and the mineralization
in the limbs [46]. On the other hand, overexpression of process of extracellular matrix.
Mikkola et al. BMC Genomics (2019) 20:1027 Page 8 of 13

Table 4 Summary of the genome-wide significant SNPs for different CHD-related traits
Trait Chr Locus Alleles N per SNP SNP(s) P-value from FASTA, corrected
with the inflation factor lambda
FHCDAE 9 31,477,907 G/A 642 BICF2G630834826 1.57 × 10–6
28 29,111,565 A/C 627 BICF2P1046032 1.62 × 10–6
OA status (binary;osteoarthritis present or not) 1 45,382,633 C/A 655 BICF2P468585 2.86 × 10–7
46,279,297 A/G 650 BICF2P357728 8.93 × 10–7
Case-control analysis (normal hips or 1 45,382,633 C/A 693 BICF2P468585 7.03 × 10–7
mild-to-severe CHD; FCI scores) 45,161,186 G/C 693 BICF2S23248027 1.30 × 10–6
The threshold for genome-wide significance was 1.82 × 10–6 (independent tests correction). The candidate genes in the immediate vicinity (± 200 kb) of the SNPs
demonstrating genome-wide significance are in Table 5. All genes within ±1 Mb from the associated loci for all traits, for which P-value was < 1.0 × 10–5 are in
Additional file 2

The third locus with genome-wide significance in- previous study [28]. AATF is located close to LHX1 but
volved also FHCDAE and resided on chromosome 28 its role in CHD remains uncertain. Both LHX1 and
(Tables 4 and 5). This region contains CACUL1, a cell- AATF have been associated with the levels of macro-
cycle associated gene [49], and NANOS1 that upregu- phage inflammatory protein 1b (MIP-1b) [57, 58]. MIP-
lates MMP14 a.k.a. membrane type 1-matrix metallopro- 1b is a cytokine increased in the synovial fluid in OA
teinase (MT1-MMP) thus promoting epithelial tumor and may play a role in the ingression of monocytes into
cell invasion [50]. MT1-MMP is a powerful collagenoly- osteoarthritic joints [59]. The canine gene encoding
tic element [51, 52] and Miller et al. (2009) have demon- MIP-1b (C-C motif for chemokine ligand 4, CCL4) is lo-
strated the role of MT1-MMP in human rheumatoid cated on chromosome 9, ~ 795 kb away from
arthritis with synovial invasion via collagenolysis [53]. TIGRP2P126345 and ~ 803 kb from AATF (Tables 1 and
The possible role of the NANOS1 – MMP14 interplay 3). SLC7A1 is a high affinity cationic amino acid trans-
needs to be targeted in tissues relevant to CHD. porter that belongs to the solute carrier family 7 [60]. It
Intriguingly, chromosome 28 has been previously asso- participates in the transportation of cationic amino acids
ciated with NoA in two studies of which one included arginine, lysine and ornithine across the plasma mem-
also German Shepherds [13, 54]. Although chromosome brane [60]. L-arginine and its methylated forms could
28 did not associate with NoA in our study, the reported impact OA via the nitric oxide pathway [61].
NoA locus is ~ 5.2 Mb upstream from our FHCDAE Considering the clinical complexity of CHD, it is not
locus (Table 1). Because FHCDAE and NoA are strongly surprising that we have successfully mapped several loci,
related traits (Pearsons’s r = − 0.94, Fig. 1), additional which contain candidate genes that are involved in dif-
studies across breeds are warranted to find out whether ferent biological pathways. Identification of these path-
the two loci on chromosome 28 are related or independ- ways is an important step in understanding the
ent, and if they possess variants contributing to CHD. pathophysiology of CHD. Some of the genes in these
We also observed some loci showing weaker associa- networks may have no direct function on the disorder
tions with NoA and OA on chromosomes 9 and 25 (Ta- but have a circuitous effect through other genes [62]. As
bles 1 and 2), and with FCI hip score on chromosome 9 demonstrated here and previously by Sánchez-Molano
(Table 3). These loci included relevant candidate genes et al. (2014) [7], the complexity and polygenicity of traits
LHX1, AATF (both on chromosome 9) and SLC7A1 such as CHD required large sample sizes for significant
(chromosome 25) (Additional file 2). LHX1 could be a associations. Sánchez-Molano et al. (2014) [7] had a co-
candidate for OA as it has been shown to be differen- hort of 1500 Labrador Retrievers, and observed two
tially methylated in OA [55] and is one of the most sig- genome-wide and multiple chromosome-wide significant
nificantly up-regulated genes in this disorder [56]. SNPs QTLs explaining maximum 23% of the genetic variance
near LHX1 demonstrated also a suggestive association in the analyzed traits. It is possible that larger cohorts
with CHD (quantified as the FCI hip score) in our might reveal additional loci with smaller effects.

Table 5 Candidate genes near SNPs showing genome-wide significant association with CHD-related phenotypes
Chr Locus derived from the top SNP coordinates ± 200 kb Gene(s) OA FHCDAE FCI score A vs. FCI score C to E
1 44,961,186–46,479,297 TIAM2, CLDN20, TFB1M, NOX3, ARID1B x x
9 31,187,114–31,677,907 NOG, C9H17orf67, DGKE x
28 28,911,565–29,322,985 PRLHR, CACUL1, NANOS1, EIF3A x
Total number of significant loci for the trait 1 2 1
Bold font indicates the gene(s) closest to the associated SNP. The trait associations are marked with x
Mikkola et al. BMC Genomics (2019) 20:1027 Page 9 of 13

Besides sample size, accurate and reliable phenotyping where the cases had an FCI score B/C (left/right hip), C/
is another essential factor when studying complex traits. B, or C/C or worse, and the second group with a strin-
This is particularly important when the trait comprises gent case definition, where the cases had an FCI score D
of many interconnected sub-traits that explain only or worse on both hips.
small parts of the total variation. As long as the assess- Two veterinarians in our group carefully evaluated the
ment of CHD relies on the FCI scoring, it is crucial to acquired radiographs for more specific hip phenotypes.
have standardized high-quality radiographs and a min- These phenotypes were: findings suggestive on osteoarth-
imal number of people assessing them to reduce inter- ritis (in four categories from 0 = no signs to 3 = severe
observer bias [26]. More reliable indices of joint laxity signs), NoA (in degrees), and FHCDAE (in millimeters).
such as the distraction or laxity index [25], could facili- The phenotyping process was carried out as follows: One
tate the discovery of genetic findings by removing some veterinarian (evaluator 1 in the phenotype file doi: https://
confounding factors affecting NoA and FHCDAE, as doi.org/10.6084/m9.figshare.10096595) assessed all the ra-
some laxity remains undiscovered in the extended view diographs for the study cohort that was used in our previ-
radiographs. ous study [28]. However, another veterinarian (evaluator 2
in the phenotype file (doi: https://doi.org/10.6084/m9.fig-
Conclusions share.10096595) in our group evaluated the radiographs of
In conclusion, we have performed a successful association the dogs that were genotyped during the current study. A
study with a large cohort of accurately and robustly phe- small subset of randomly chosen radiographs, which the
notyped German Shepherds and describe three loci with evaluator 1 had previously assessed were re-assessed by
genome-wide significance and two suggestive loci for the evaluator 2 to check their consistency. In case there
CHD-related traits. The candidate genes include NOX3 were any inconsistencies, the re-assessed phenotype was
and ARID1B on chromosome 1, NOG on chromosome 9, used in the analysis.
and NANOS1 on chromosome 28. Future studies will NoA varied between 70 and 108 degrees in our cohort
focus on ascertaining their role in CHD by resequencing (Table 6); the smaller the value is, the worse is the incon-
the candidate region for putative risk variants. gruity of the joint. Generally dogs with an FCI hip score A
have NoA of 105 degrees or higher [64]. Significant inter-
Methods observer variation for NoA was seen in our data (P =
Dogs 0.028, Additional file 1). We handled this in our GWAS
We acquired the data for our study from the Finnish Ken- by using the evaluator as a covariate. FHCDAE was mea-
nel Club. Before quality control we had a total of 775 sam- sured as millimeters (mm) and in our data this trait
ples of German Shepherds and of these 356 were controls, ranged between − 4 and 15 mm (Table 6). The smaller the
322 were cases with both hip joints scored C or worse and value is, the deeper the femoral head sits into the acetabu-
97 were of intermediate phenotypes with at least one hip lum in relation to the dorsal acetabular edge. OA was di-
joint scored as B. Majority of the dogs had either the same vided into four categories (the quantities for each category
FCI score bilaterally or had maximum one score grade dif- here are before quality control): no signs of arthritis (0,
ference between the right and the left hip; three dogs had N = 498), some mild changes affiliated with OA (1, N =
more than one grade difference (they had been scored A/ 57, minor osteophytes on the femoral neck and/or at the
C, C/A and B/D). The average age at radiographing was craniolateral acetabular edge), moderate changes (2, N =
1.55 years ranging from 1.01 to 5.83 years with a standard 74, larger osteophytes, also at the dorsal acetabular edge),
deviation of 0.63 years. 435 of the dogs were female and or severe osteoarthritis (3, N = 33, massive osteophytes of
340 were male. We collected at least one blood sample the femoral neck and surrounding the acetabular edge).
from all the dogs with ethylenediaminetetraacetic acid However, radiographs are relatively insensitive in detect-
(EDTA) as an anticoagulant. ing early osteoarthritic changes [65]. Therefore, the
current study is unlikely to detect any associations with
Phenotypes loci affecting exclusively the early stages of OA.
The FCI-standardized ventrodorsal extended hip radio-
graphs were taken by different veterinarians, but the hip DNA preparation and genotyping
scoring was done by two specialized veterinarians at the The original EDTA preserved blood samples for this study
FKC. Therefore, inter-observer bias was reduced in this are stored at the Dog DNA bank at the University of
data set [26]. All of the hip scores for these dogs are Helsinki. The DNA was extracted from these samples with a
available in the FKC database [63]. We had at least a Chemagic Magnetic Separation Module I with a standard
CHD score for all the dogs. We used the official FCI hip protocol by Chemagen (Chemagen Biopolymer-Technologie
scores to divide the dogs into two different case-control AG, Baeswieler, Germany). Thereafter the DNA samples
groups: the first group with a relaxed case definition, were genotyped at Geneseek (Lincoln, NE, US) with a high
Mikkola et al. BMC Genomics (2019) 20:1027 Page 10 of 13

Table 6 Median, interquartile range and minimum and number of dogs per analysis varied between 338 and 693
maximum values for the analyzed traits as FASTA dropped individual dogs from analyses if they
Trait Median Interquartile 25th percentile: Min – Max missed a phenotype or a covariate. CanFam3.1 was used
range 75th percentile as the position map for our SNPs [70]. After the GWAS
NoA (degrees) 102 15 90: 105 70–108 the genotype call quality of the top SNPs was checked to
FHCDAE (mm) 0 7 −2: 5 −4 – 15 exclude associations due to calling errors.
OA (category) 0 nr nr 0–3
nr = not relevant
Genome-wide association analysis (GWAS)
density 173 K canine SNP array from Illumina (San Diego, We performed a GWAS by using polygenic mixed
CA, US). Genotyping of the samples was done in multiple models in GenABEL, with functions “polygenic” and
batches. “mmscore” (FASTA: Score test for association in related
people) [71]. The appropriate covariates were estimated
Population structure with fitting linear regression models with the R function
We used information from a genomic relationship “lm” from the stats-package [72] for all non-binary traits.
matrix built from the SNP data to divide our highly The binary traits were analyzed with fitting generalized
stratified German Shepherd population into three sub- linear models with the R function “glm” [73]. The fol-
populations (Additional file 7). For the clustering we lowing covariates were tested: sex, age at radiographing,
used an R [66] package “mclust” [67] that utilizes covari- genetic cluster of the dog, genotyping batch, birth
ance parametrization. The selection of appropriate num- month, and evaluator, in other words the veterinarian
ber of clusters was executed with Bayesian information who evaluated the radiographs (tested for traits NoA,
criterion. We then created a covariate vector from the FHCDAE and OA). The appropriate covariates which
clustering data where every individual belonged to one had a significant effect (P-value < 0.05) for each
of the clusters. This way we could use the clustering ef- dependent trait are in Table 7 (See also Additional file
fect in our model to account for any differences in dis- 1). The inflation factor lambda for the various models
ease association between the genetic clusters. are indicated in the Tables 1-3. The corresponding Q-Q
plots are in the Additional file 8.
Quality control (QC) The r2 values for the top SNPs were estimated in R
We used PLINK [68] to merge the original three geno- with “r2fast” function [74] from GenABEL-package.
type sets from different genotyping batches. A prelimin- Bonferroni correction can be seen as a too stringent
ary QC was done on all of the genotyping batches before method to correct for multiple testing as it expects
merging, with the following thresholds: call rate per independency between the tests, which is untrue in
sample 0.10, call rate per SNP 0.05, minor allele fre- many association studies because of LD between
quency 0.05, P-value cut-off for deviation from Hardy- markers [75]. This is especially important to note in
Weinberg equilibrium (HWE) 0.00001 (from controls canine studies, as the structure of the canine genome
only). After these quality controls and data merging a is unique with strong LD due to the history of inten-
total of 100,435 SNPs and 775 samples were transferred sive selection [13]. Therefore, we used the number of
from PLINK to R. The final QC was done in R with independent tests to determine the threshold for sig-
GenABEL [69], and the thresholds were: minor allele nificance. We estimated the effective number of inde-
frequency = 0.05, per sample call rate = 0.85 and per SNP pendent tests to be 27,456 using simpleM, which uses
call rate = 0.95, and again a P-value cut-off level < dimension reduction models for filtering the correla-
0.00001 to test for deviations from HWE. After the final tions between the analyzed SNPs [76]. Based on this,
QC we had 89,251 autosomal SNPs and 769 samples to the threshold for significance 1.82 × 10–6 (0.05/27456)
use in our association analysis. However, the final is applied for P-values in this study.
Table 7 Covariates for different traits
Trait Covariates chosen with lm or glm function in R
OA status (binary; osteoarthritis present or not) Age at radiographing, genotyping batch, genetic cluster and birth month
FHCDAE Age at radiographing, genetic cluster, genotyping batch
NoA Age at radiographing, genetic cluster, genotyping batch, evaluator
Cases vs. controls (FCI score categories A vs. C-E) Age at radiographing and genotyping batch
Cases vs. controls (FCI score categories A vs D-E) Age at radiographing, genetic cluster, genotyping batch and birth month
Cases vs. controls (FCI score categories C vs D-E) Genetic cluster, genotyping batch and birth month
Mikkola et al. BMC Genomics (2019) 20:1027 Page 11 of 13

Supplementary information Availability of data and materials

Supplementary information accompanies this paper at https://doi.org/10. The datasets generated and analyzed in the current study are available at
1186/s12864-019-6422-6. FIGSHARE, doi: https://doi.org/10.6084/m9.figshare.10096595. The datasets
were anonymized to protect the owners of the animals.
Additional file 1. Summaries of glm and lm models
Additional file 2. List of genes within +/− 1 Mb from the associated loci Consent for publication
Not applicable.
Additional file 3. Pairwise LD (r2) for the top SNPs by chromosome
Additional file 4. Allele frequencies for the significant SNPs from the
case-control analyses Competing interests
2
HL provides consultancy to Genoscoper Laboratories Oy, a canine DNA
Additional file 5. Heritability (h ) estimates from the polygenic model diagnostics company, he used to partially own during the study (till 12/
(GenABEL) for all of the traits 2017). There are no other competing interests.
Additional file 6 Results from a STRING database search for the
candidate gene NOX3 Author details
1
Additional file 7. Classification plot of the population structure from R- Department of Veterinary Biosciences, University of Helsinki, P.O. Box 66
package “mclust”. The population is divided into three subpopulations, (Mustialankatu 1), FI-00014 Helsinki, Finland. 2Department of Medical and
where the green triangles represent individual dogs of one subpopula- Clinical Genetics, University of Helsinki, Helsinki, Finland. 3Folkhälsan Research
tion and the red squares and the blue dots represent the individual dogs Center, Helsinki, Finland. 4Department of Equine and Small Animal Medicine,
of the two other subpopulations University of Helsinki, Helsinki, Finland.

Additional file 8. Q-Q plots of the various association analyses corre- Received: 5 July 2019 Accepted: 22 December 2019
sponding to Tables 1-4

References
Abbreviations 1. Brass W. Hüftgelenkdysplasie und Ellbogenkrankkung im Visier der
AATF: Apoptosis antagonizing transcription factor; ARID1B: AT-rich interactive Fédération Cynologique Internationale. Kleintierpraxis. 1993;38:191–266.
domain 1B; BMP: Bone morphogenetic protein; CACUL1: CDK2 associated
2. Skurková L, Hluchý M, Lacková M, Mihalová M, Ledecký V. Relation of the
cullin domain 1; CCL4: C-C motif chemokine ligand 4; CHD: Canine hip norberg angle and position of the femoral head Centre to the dorsal
dysplasia; Chr: Chromosome; EDTA: Ethylenediaminetetraacetic acid; acetabular edge in evaluation of canine hip dysplasia. Vet Comp Orthop
FASTA: Family-based score test for association; FCI: Fédération cynologique Traumatol. 2010;23:433–8.
internationale; FHCDAE: Femoral head center position in relation to dorsal
3. Allen MJ. What’s new in Orthopaedic basic science. J Bone Jt Surg. 2016;98:
acetabular edge; FKC: Finnish kennel club; GWAS: Genome-wide association 2025–9.
study; h2: Heritability (narrow sense); HWE: Hardy-Weinberg equilibrium; 4. Wilson BJ, Nicholas FW, James JW, Wade CM, Raadsma HW, Thomson PC.
LD: Linkage disequilibrium; LHX1: LIM homeobox 1; MIP-1b: Macrophage Genetic correlations among canine hip dysplasia radiographic traits in a cohort
inflammatory protein 1b; MMP14: Matrix metalloproteinase-14; MT1-
of Australian German shepherd dogs, and implications for the design of a
MMP: Membrane type 1-matrix metalloproteinase; NANOS1: Nanos C2HC- more effective genetic control program. PLoS One. 2013;8:e78929.
type zinc finger 1; NoA: Norberg angle; NOG: Noggin; NOX3: NADPH oxidase 5. Zhou Z, Sheng X, Zhang Z, Zhao K, Zhu L, Guo G, et al. Differential genetic
3; OA: Osteoarthritis; P-value: Probability value; QC: Quality control; regulation of canine hip dysplasia and osteoarthritis. PLoS One. 2010;5:
QTL: Quantitative trait locus; r2: Square of pearson’s correlation coefficient r;
e13219.
SLC7A1: Solute carrier family 7 member 1; SNP: Single-nucleotide 6. Mateescu RG, Burton-Wurster NI, Tsai K, Phavaphutanon J, Zhang Z, Murphy
polymorphism KE, et al. Identification of quantitative trait loci for osteoarthritis of hip joints
in dogs. Am J Vet Res. 2008;69:1294–300.
7. Sánchez-Molano E, Woolliams JA, Pong-Wong R, Clements DN, Blott SC,
Acknowledgements
Wiener P. Quantitative trait loci mapping for canine hip dysplasia and its
We wish to acknowledge the expert technical assistance by Santeri Suokas,
related traits in UK Labrador retrievers. BMC Genomics. 2014;15:833.
Kirsi Lahti and Sini Karjalainen. We are also grateful to all owners for
8. Janutta V, Hamann H, Distl O. Complex segregation analysis of canine hip
donating samples from their dogs for the study.
dysplasia in German shepherd dogs. J Hered. 2006;97:13–20.
9. Pfahler S, Distl O. Identification of quantitative trait loci (QTL) for canine hip
Ethical approval and consent to participate dysplasia and canine elbow dysplasia in Bernese mountain dogs. PLoS One.
We followed the guidelines for research ethics and good scientific practices. 2012;7:e49782.
We hold a license for collecting EDTA blood samples from ELLA - Animal 10. Lavrijsen ICM, Leegwater PAJ, Martin AJ, Harris SJ, Tryfonidou MA, Heuven
Experiment Board in Finland (ESAVI/7482/04.10.07/2015) under permission of HCM, et al. Genome wide analysis indicates genes for basement membrane
The Regional State Administrative agency for Southern Finland. All dog and cartilage matrix proteins as candidates for hip dysplasia in Labrador
owners signed a form of consent and were informed of the project. retrievers. PLoS One. 2014;9:e87735.
11. Todhunter RJ, Bliss SP, Casella G, Wu R, Lust G, Burton-Wurster NI, et al.
Genetic structure of susceptibility traits for hip dysplasia and microsatellite
Authors’ contributions Informativeness of an outcrossed canine pedigree. J Hered. 2003;94:39–48.
LM, HL and AI conceived the study and designed the experiments. LM, SH, 12. Mäki K, Janss LLG, Groen AF, Liinamo A-E, Ojala M. An indication of major
AL, TPM and AI performed the experiments and analyzed the data. AI, MH, genes affecting hip and elbow dysplasia in four Finnish dog populations.
TPM, AL and HL supervised the study. LM and AI wrote the article. All Hered (Edinb). 2004;92:402–8.
authors have read and approved the final version of the manuscript. 13. Hayward JJ, Castelhano MG, Oliveira KC, Corey E, Balkman C, Baxter TL, et al.
Complex disease and phenotype mapping in the domestic dog. Nat
Commun. 2016;7:10460.
Funding 14. Fels L, Distl O. Identification and validation of quantitative trait loci (QTL) for
This study was funded by The Academy of Finland (252602, www.aka.fi, to canine hip dysplasia (CHD) in German shepherd dogs. PLoS One. 2014;9:
AI), The American Kennel Club Canine Health Foundation (01828, www. e96618.
akcchf.org, to AI), The Jane and Aatos Erkko Foundation (jaes.fi/en, to HL) 15. Bartolomé N, Segarra S, Artieda M, Francino O, Sánchez E, Szczypiorska M,
and Biocentrum Helsinki (to HL). The funders had no role in study design, et al. A genetic predictive model for canine hip dysplasia: integration of
data collection and analysis, decision to publish, or preparation of the genome wide association study (GWAS) and candidate gene approaches.
manuscript. PLoS One. 2015;10:e0122558.
Mikkola et al. BMC Genomics (2019) 20:1027 Page 12 of 13

16. Fels L, Marschall Y, Philipp U, Distl O. Multiple loci associated with canine hip 39. Santen GWE, Clayton-Smith J. The ARID1B phenotype: what we have
dysplasia (CHD) in German shepherd dogs. Mamm Genome. 2014;25:262–9. learned so far. Am J Med Genet Part C Semin Med Genet. 2014;166:276–89.
17. Sánchez-Molano E, Woolliams JA, Blott SC, Wiener P. Assessing the impact 40. Chase K, Lawler DF, Carrier DR, Lark KG. Genetic regulation of osteoarthritis:
of genomic selection against hip dysplasia in the Labrador retriever dog. J a QTL regulating cranial and caudal acetabular osteophyte formation in the
Anim Breed Genet. 2014;131:134–45. hip joint of the dog (Canis familiaris). Am J Med Genet A. 2005;135:334–5.
18. Smith GK, Mayhew PD, Kapatkin AS, McKelvie PJ, Shofer FS, Gregor TP. 41. Todhunter RJ, Zachos TA, Gilbert RO, Erb HN, Williams AJ, Burton-Wurster N,
Evaluation of risk factors for degenerative joint disease associated with hip et al. Onset of epiphyseal mineralization and growth plate closure in
dysplasia in German shepherd dogs, Golden retrievers, Labrador retrievers, radiographically normal and dysplastic Labrador retrievers. J Am Vet Med
and Rottweilers. J Am Vet Med Assoc. 2001;219:1719–24. Assoc. 1997;210:1458–62.
19. Hou Y, Wang Y, Lu X, Zhang X, Zhao Q, Todhunter RJ, et al. Monitoring hip 42. Madsen JS, Reimann I, Svalastoga E. Delayed ossification of the femoral
and elbow dysplasia achieved modest genetic improvement of 74 dog head in dogs with hip dysplasia. J Small Anim Pract. 1991;32:351–4.
breeds over 40 years in USA. PLoS One. 2013;8:e76390. 43. McMahon JA, Takada S, Zimmerman LB, Fan CM, Harland RM, McMahon AP.
20. Lewis TW, Blott SC, Woolliams JA. Comparative analyses of genetic trends Noggin-mediated antagonism of BMP signaling is required for growth and
and prospects for selection against hip and elbow dysplasia in 15 UK dog patterning of the neural tube and somite. Genes Dev. 1998;12:1438–52.
breeds. BMC Genet. 2013;14:16. 44. Brunet LJ, McMahon JA, McMahon AP, Harland RM. Noggin, cartilage
21. Edwards SM, Woolliams JA, Hickey JM, Blott SC, Clements DN, Sánchez- morphogenesis, and joint formation in the mammalian skeleton. Sci. 1998;
Molano E, et al. Joint genomic prediction of canine hip dysplasia in UK and 280:1455–7.
US Labrador retrievers. Front Genet. 2018;9:101. 45. Potti TA, Petty EM, Lesperance MM. A comprehensive review of reported
22. Oberbauer AM, Keller GG, Famula TR. Long-term genetic selection reduced heritable noggin-associated syndromes and proposed clinical utility of one
prevalence of hip and elbow dysplasia in 60 dog breeds. PLoS One. 2017; broadly inclusive diagnostic term: NOG-related-symphalangism spectrum
12:e0172918. disorder (NOG-SSD). Hum Mutat. 2011;32:877–86.
23. Sánchez-Molano E, Pong-Wong R, Clements DN, Blott SC, Wiener P, 46. Tylzanowski P, Mebis L, Luyte FP. The noggin null mouse phenotype is
Woolliams JA. Genomic prediction of traits related to canine hip dysplasia. strain dependent and haploinsufficieny leads to skeletal defects. Dev Dyn.
Front Genet. 2015;6:97. 2006;235:1599–607.
24. Guo G, Zhou Z, Wang Y, Zhao K, Zhu L, Lust G, et al. Canine hip dysplasia is 47. Devlin RD, Du Z, Pereira RC, Kimble RB, Economides AN, Jorgetti V, et al.
predictable by genotyping. Osteoarthr Cartil. 2011;19:420–9. Skeletal overexpression of noggin results in osteopenia and reduced bone
25. Broeckx BJG, Vezzoni A, Bogaerts E, Bertal M, Deforce D, Peelman L, et al. formation. Endocrinol. 2003;144:1972–8.
Comparison of three methods to quantify laxity in the canine hip joint. Vet 48. Ghadakzadeh S, Hamdy RC, Tabrizian M. Efficient in vitro delivery of noggin
Comp Orthop Traumatol. 2018;31:23–9. siRNA enhances osteoblastogenesis. Heliyon. 2017;3:e00450.
26. Verhoeven GE, Coopman F, Duchateau L, Bosmans T, Van Ryssen B, Van 49. Kong Y, Nan K, Yin Y. Identification and characterization of CAC1 as a novel
Bree H. Interobserver agreement of the assessability of standard CDK2-associated cullin. Cell Cycle. 2009;8:3552–61.
ventrodorsal hip-extended radiographs and its effect on agreement in the 50. Bonnomet A, Polette M, Strumane K, Gilles C, Dalstein V, Kileztky C, et al.
diagnosis of canine hip dysplasia and routine FCI scoring. Vet Radiol The E-cadherin-repressed hNanos1 gene induces tumor cell invasion by
Ultrasound. 2009;50:259–63. upregulating MT1-MMP expression. Oncogene. 2008;27:3692–9.
27. Loeser RF, Goldring SR, Scanzello CR, Goldring MB. Osteoarthritis: a disease 51. Shi J, Son M-Y, Yamada S, Szabova L, Kahan S, Chrysovergis K, et al. Membrane-
of the joint as an organ. Arthritis Rheum. 2012;64:1697–707. type MMPs enable extracellular matrix permissiveness and mesenchymal cell
28. Mikkola LI, Holopainen S, Lappalainen AK, Pessa-Morikawa T, Augustine TJP, proliferation during embryogenesis. Dev Biol. 2008;313:196–209.
Arumilli M, et al. Novel protective and risk loci in hip dysplasia in German 52. Itoh Y, Ito N, Nagase H, Evans RD, Bird SA, Seiki M. Cell surface
shepherds. PLoS Genet. 2019;15:e1008197. collagenolysis requires homodimerization of the membrane-bound
29. Daumer KM, Khan AU, Steinbeck MJ. Chlorination of pyridinium collagenase MT1-MMP. Mol Biol Cell. 2006;17:5390–9.
compounds. Possible role of hypochlorite, N-chloramines, and chlorine in 53. Miller M-C, Manning HB, Jain A, Troeberg L, Dudhia J, Essex D, et al.
the oxidation of pyridinoline cross-links of articular cartilage collagen type II Membrane type 1 matrix metalloproteinase is a crucial promoter of synovial
during acute inflammation. J Biol Chem. 2000;275:34681–92. invasion in human rheumatoid arthritis. Arthritis Rheum. 2009;60:686–97.
30. Steinbeck MJ, Nesti LJ, Sharkey PF, Parvizi J. Myeloperoxidase and 54. Huang M, Hayward JJ, Corey E, Garrison SJ, Wagner GR, Krotscheck U, et al.
chlorinated peptides in osteoarthritis: potential biomarkers of the disease. J A novel iterative mixed model to remap three complex orthopedic traits in
Orthop Res. 2007;25:1128–35. dogs. PLoS One. 2017;12:e0176932.
31. Breitenbach M, Rinnerthaler M, Weber M, Breitenbach-Koller H, Karl T, Cullen 55. Delgado-Calle J, Fernández AF, Sainz J, Zarrabeitia MT, Sañudo C, García-
P, et al. The defense and signaling role of NADPH oxidases in eukaryotic Renedo R, et al. Genome-wide profiling of bone reveals differentially
cells. Wien Med Wochenschr. 2018;168:286–99. methylated regions in osteoporosis and osteoarthritis. Arthritis Rheum. 2013;
32. Szklarczyk D, Franceschini A, Wyder S, Forslund K, Heller D, Huerta-Cepas J, 65:197–205.
et al. STRING v10: protein–protein interaction networks, integrated over the 56. Fei Q, Lin J, Meng H, Wang B, Yang Y, Wang Q, et al. Identification of
tree of life. Nucleic Acids Res. 2015;43:D447–52. upstream regulators for synovial expression signature genes in
33. Volk SW, Kapatkin AS, Haskins ME, Walton RM, D’Angelo M. Gelatinase osteoarthritis. Jt Bone Spine. 2016;83:545–51.
activity in synovial fluid and synovium obtained from healthy and 57. Ahola-Olli AV, Würtz P, Havulinna AS, Aalto K, Pitkänen N, Lehtimäki T, et al.
osteoarthritic joints of dogs. Am J Vet Res. 2003;64:1225–33. Genome-wide association study identifies 27 loci influencing concentrations of
34. Fujiki M, Shineha J, Yamanokuchi K, Misumi K, Sakamoto H. Effects of circulating cytokines and growth factors. Am J Hum Genet. 2017;100:40–50.
treatment with polysulfated glycosaminoglycan on serum cartilage 58. GWAS Catalog. https://www.ebi.ac.uk/gwas/variants/rs1452928. Accessed 18
oligomeric matrix protein and C-reactive protein concentrations, serum Jan 2019.
matrix metalloproteinase-2 and -9 activities, and lameness in dogs with 59. Koch AE, Kunkel SL, Shah MR, Fu R, Mazarakis DD, Haines GK, et al.
osteoarthritis. Am J Vet Res. 2007;68:827–33. Macrophage inflammatory protein-Iβ: a C-C chemokine in osteoarthritis. Clin
35. Zeng GQ, Chen AB, Li W, Song JH, Gao CY. High MMP-1, MMP-2, and MMP- Immunol Immunopathol. 1995;77:307–14.
9 protein levels in osteoarthritis. Genet Mol Res. 2015;14:14811–22. 60. SLC7A1. https://www.uniprot.org/uniprot/P30825#function. Accessed 28 Jan 2019.
36. Chiu CY, Leng S, Martin KA, Kim E, Gorman S, Duhl DM. Cloning and 61. Pascale V, Pascale W, Lavanga V, Sansone V, Ferrario P, De Gennaro Colonna
characterization of T-cell lymphoma invasion and metastasis 2 (TIAM2), a V. L-arginine, asymmetric dimethylarginine, and symmetric dimethylarginine
novel guanine nucleotide exchange factor related to TIAM1. Genomics. in plasma and synovial fluid of patients with knee osteoarthritis. Med Sci
1999;61:66–73. Monit. 2013;19:1057–62.
37. ARID1B. Uniprot. https://www.uniprot.org/uniprot/Q8NFD5#function. 62. Boyle EA, Li YI, Pritchard JK. An expanded view of complex traits: from
Accessed 23 Nov 2018. polygenic to Omnigenic. Cell. 2017;169:1177–86.
38. Schrier Vergano S, Santen G, Wieczorek D, Wollnik B, Matsumoto N, 63. Hip joint statistics - German shepherd. In: Breeding database of the Finnish
Deardorff MA. Coffin-Siris Syndrome. GeneReviews®. 1993. http://www.ncbi. Kennel Club; 2017. https://jalostus.kennelliitto.fi/frmTerveystilastot.aspx?R=
nlm.nih.gov/pubmed/23556151. Accessed 27 Nov 2018. 166&Lang=en. Accessed 18 Sep 2017.
Mikkola et al. BMC Genomics (2019) 20:1027 Page 13 of 13

64. Flückiger M, Ecvdi D. Scoring radiographs for canine hip dysplasia-the big
three organisations in the world. Eur J Companion Anim Pract. 2008;17:135–40.
65. Lust G, Todhunter RJ, Erb HN, Dykes NL, Williams AJ, Burton-Wurster NI, et al.
Repeatability of dorsolateral subluxation scores in dogs and correlation with
macroscopic appearance of hip osteoarthritis. Am J Vet Res. 2001;62:1711–5.
66. R Core Team. R: The R Project for Statistical Computing. R Foundation for
statistical computing. 2017. https://www.r-project.org/.
67. Fraley C, Raftery AE, Murphy TB, Scrucca L. mclust Version 4 for R: Normal
Mixture Modeling for Model-Based Clustering, Classification, and Density
Estimation. In: Technical Report 597. Seattle: University of Washington; 2012. p.
1–50. http://www.stat.washington.edu/research/reports/2012/tr597.pdf.
68. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al.
PLINK: a tool set for whole-genome association and population-based
linkage analyses. Am J Hum Genet. 2007;81:559–75.
69. Aulchenko YS, Ripke S, Isaacs A, van Duijn CM. GenABEL: an R library for
genome-wide association analysis. Bioinformatics. 2007;23:1294–6.
70. CanFam3.1 GCA_000002285.2. https://www.ncbi.nlm.nih.gov/assembly/GCF_
000002285.3/. Accessed 26 Apr 2018.
71. Chen W-M, Abecasis GR. Family-based association tests for Genomewide
association scans. Am J Hum Genet. 2007;81:913–26.
72. lm {stats} R Documentation. https://stat.ethz.ch/R-manual/R-devel/library/
stats/html/lm.html . Accessed 3 May 2019.
73. glm {stats} R documentation. https://stat.ethz.ch/R-manual/R-devel/library/
stats/html/glm.html . Accessed 3 May 2019.
74. r2fast. https://www.rdocumentation.org/packages/GenABEL/versions/1.8-0/
topics/r2fast . Accessed 3 May 2019.
75. Johnson RC, Nelson GW, Troyer JL, Lautenberger JA, Kessing BD, Winkler CA,
et al. Accounting for multiple comparisons in a genome-wide association
study (GWAS). BMC Genomics. 2010;11:724.
76. Gao X, Becker LC, Becker DM, Starmer JD, Province MA. Avoiding the high
Bonferroni penalty in genome-wide association studies. Genet Epidemiol.
2010;34:100–5.

Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.