Altered Transmission of HOX and Apoptotic SNPs Identify a Potential Common Pathway for Clubfoot

Abstract

Clubfoot is a common birth defect that affects 135,000 newborns each year worldwide. It is characterized by equinus deformity of one or both feet and hypoplastic calf muscles. Despite numerous study approaches, the cause(s) remains poorly understood although a multifactorial etiology is generally accepted. We considered the HOXA and HOXD gene clusters and insulin-like growth factor binding protein 3 (IGFBP3) as candidate genes because of their important roles in limb and muscle morphogenesis. Twenty SNPs from the HOXA and HOXD gene clusters and 12 SNPs in IGFBP3 were genotyped in a sample composed of nonHispanic white and Hispanic multiplex and simplex families (discovery samples) and a second sample of nonHispanic white simplex trios (validation sample). Four SNPs (rs6668, rs2428431, rs3801776 and rs3779456) in the HOXA cluster demonstrated altered transmission in the discovery sample, but only rs3801776, located in the HOXA basal promoter region, showed altered transmission in both the discovery and validation samples (p=0.004 and p=0.028). Interestingly, HOXA9 is expressed in muscle during development. A SNP in IGFBP3, rs13223993, also showed altered transmission (p=0.003) in the discovery sample. Gene-gene interactions were identified between variants in HOXA, HOXD and IGFBP3 and with previously associated SNPs in mitochondrial-mediated apoptotic genes. The most significant interactions were found between CASP3 SNPS and variants in HOXA, HOXD and IGFBP3. These results suggest a biologic model for clubfoot in which perturbation of HOX and apoptotic genes together affect muscle and limb development, which may cause the downstream failure of limb rotation into a plantar grade position.

INTRODUCTION

Clubfoot, or talipes equinovarus (TEV), is a common birth defect which is easily recognized and diagnosed because of the characteristic forefoot adduction, hindfoot varus and equinus, midfoot cavus, and hypoplastic lower limb musculature [Dietz, 1985]. Isolated, or idiopathic, cases occur in the absence of any other anomaly and/or known cause(s) [Moorthi et al., 2005]. The birth prevalence varies among ethnic groups from 1/150-2500, with a worldwide rate of approximately 1/1000 [Moorthi et al., 2005]. A variety of approaches including genome scans and candidate gene testing have been used to identify the underlying causes of complex genetic disorders. Interrogation of minimal overlapping chromosomal deletions in syndromic clubfoot led to the identification of a group of candidate genes, the HOXD gene cluster (UCSC Genome Bioinformatics) [Brewer et al., 1998].

The HOX genes encode transcription factors that direct patterning of the axial skeleton and muscle and limb development [Mark et al., 1997; McGinnis and Krumlauf, 1992]. This family consists of 39 genes distributed in four paralogous clusters (A, B, C and D). Each of the HOX clusters share common genomic structure and are temporally and spatially expressed during embryogenesis. The 5’ HOXA genes are expressed in the muscles of both the fore and hind limbs, and are involved in patterning and differentiation of muscles in embryonic limbs and during muscle repair of adult limbs [Houghton and Rosenthal, 1999]. Mutations in HOX genes perturb normal limb development; for example, the HOXD10 956T>A/319M>K mutation causes autosomal dominant congenital vertical talus (CVT) [Dobbs et al., 2006; Shrimpton et al., 2004].

Conditional HOXA13 knockout mice show decreased expression of insulin-like growth factor binding protein 3 (IGFBP3) in the limbs (Dr. S. Stadler, personal communication), which has been shown to have a role in apoptosis [Baxter and Martin, 1989]. This observation is important because we previously reported an association between genetic variants in Caspase genes and clubfoot [Ester et al., 2007; Heck et al., 2005].

Given the role of HOXA, HOXD and IGFBP3 in limb development, this study was undertaken to determine whether variation in the HOXA, HOXD or IGFBP3 genes is associated with isolated clubfoot and to further evaluate whether the 956T>A/319M>K HOXD10 mutation causes clubfoot.

MATERIALS AND METHODS

Clubfoot study samples and sample preparation

SNPs from the HOXA and HOXD gene clusters and IGFBP3 were genotyped in the discovery population composed of 1783 genotyped individuals from 598 singly ascertained families as described previously [Ester et al., 2007; Heck et al., 2005]. This included a total of 179 extended families (122 nonHispanic white (NHW) and 57 Hispanic) and 331 simplex families (130 NHW and 201 Hispanic) and 88 trios with a positive family history (42 NHW and 46 Hispanic). The validation population consisted of 144 NHW simplex trios.

Families in the discovery dataset were identified through a proband with clubfoot from one of six orthopedic centers: Shriners Hospital for Children of Houston, Los Angeles and Shreveport, Texas Scottish Rite Hospital for Children of Dallas, University of Iowa, Department of Orthopaedics and Rehabilitation, and other collection sites, as previously described [Ester et al., 2007; Heck et al., 2005]. Individuals from the validation population were ascertained in the Department of Orthopedics at Washington University in St. Louis (144 probands). All affected individuals were diagnosed with clubfoot using standardized clinical criteria and radiographic examination as previously described [Lochmiller et al., 1998]. Age of ascertainment for probands generally ranged from birth to 18 years of age. Cases were excluded if they were caused by a chromosomal abnormality or were syndromic. Ethnicity was self-reported and informed consent was obtained.

Blood or saliva samples were collected from the affected individuals and family members. DNA was extracted using either the Roche DNA Isolation Kit for Mammalian Blood (Roche, Switzerland) or the Oragene Purifier for saliva (DNA Genotek, Inc., Ontario, Canada) following the manufacturer’s protocol.

956T>A/319M>K HOXD10 mutation detection

To determine whether the 956T>A/319M>K HOXD10 mutation played a causal role in isolated clubfoot, 494 clubfoot probands (267 NHW and 227 Hispanic) and 595 unrelated, unaffected controls (380 NHW and 215 Hispanic) were genotyped using a custom TaqMan assay designed using FileBuilder 3.0 (ABI, Foster City, CA). One hundred forty-four probands ascertained later were not genotyped for this mutation. The positive control in all assays was DNA from two individuals with CVT and the HOXD10 956T>A/319M>K mutation.

Genotyping

SNPs were identified using NCBI and Ensembl websites and the software program Haploview [Barrett et al., 2005]. Final SNP selection was based on heterozygosity (>0.3), inter- and intragenic positions, coverage of the gene, and tagging ability. Using these criteria, nine HOXA, 11 HOXD and 12 IGFBP3 SNPs were identified (Table I). SNPs were genotyped using TaqMan^® Genotyping Assays (Applied Biosystems, Foster City, CA) and detected on a 7900 HT Sequence Detection System (Applied Biosystems, Foster City, CA).

Table I.

Allele frequencies for HoxA, HoxD and IGFBP3 SNPs

Gene	dbSNP	Pos.	Alleles	MAFa	HMAFb,c	Location
HoxD	rs6749771	SNP1	A/G	0.391	0.685	3' of HoxD cluster
	rs1446575	SNP2	C/A	0.385	0.697	HoxD3-HoxD1
	rs1318778	SNP3	G/C	0.289	0.145	3' of HoxD3
	rs1542180	SNP4	G/A	0.378	0.692	5' of HoxD3
	rs1867863	SNP5	C/A	0.341	0.606	5' of HoxD4
	rs2113563	SNP6	T/C	0.306	0.257	HoxD8-HoxD4
	rs2592394	SNP7	C/T	0.305	0.182	HoxD9-HoxD8
	rs847146	SNP8	G/T	0.354	0.361	HoxD11 (intronic)
	rs741610	SNP9	G/A	0.316	0.443	HoxD12-HoxD11
	rs711812	SNP10	T/G	0.325	0.438	5' of HoxD12
	rs6758117	SNP11	T/C	0.302	0.594	5' of HoxD cluster
HoxA	rs2462907	SNP12	C/T	0.491	0.461	3' of HoxA cluster
	rs6668	SNP13	C/T	0.375	0.413	3' of HoxA2
	rs2428431	SNP14	C/G	0.367	0.409	HoxA2 (intronic)
	rs3757640	SNP15	C/T	0.267	0.415	HoxA5 (intronic)
	rs3801776	SNP16	G/A	0.226	0.315	5' of HoxA9
	rs3779456	SNP17	T/C	0.442	0.540	HoxA10 (intronic)
	rs1859164	SNP18	T/C	0.479	0.386	HoxA10 (intronic)
	rs6968828	SNP19	G/T	0.462	0.445	HoxA11 (intronic)
	rs3807598	SNP20	C/G	0.451	0.305	5' of HoxA cluster
IGFBP3	rs2132571	SNP21	A/G	0.306	0.218	upstream
	rs2132572	SNP22	A/G	0.227	0.478	upstream
	rs2854744	SNP23	A/C	0.472	0.316	upstream
	rs2854746	SNP24	C/G	0.415	0.264	exon 1 (missense)
	rs2854747	SNP25	C/T	0.415	0.631	intron 1–2
	rs3793345	SNP26	C/T	0.199	0.150	intron 1–2
	rs2471551	SNP27	C/G	0.192	0.154	intron 1–2
	rs3110697	SNP28	A/G	0.422	0.629	intron 3–4
	rs10255707	SNP29	C/T	0.244	0.545	intron 3–4
	rs2453839	SNP30	C/T	0.197	0.130	intron 4–5
	rs6670	SNP31	A/T	0.216	0.129	3' UTR
	rs13223993	SNP32	A/G	0.248	0.289	downstream

^aMAF = Minor allele frequency in nonHispanic white sample

^bHMAF = Frequency of NHW minor allele in Hispanics

^cHMAF signficiantly different from MAF (p<0.001) in bold

Statistical analysis

Allele frequencies and Hardy-Weinberg Equilibrium (HWE) were calculated using SAS (v9.1). The SNPs not in HWE (p<0.001) were identified and appropriate methods for correction were used in the analyses. Pair-wise linkage disequilibrium values (D′ and r²) were calculated using GOLD [Abecasis and Cookson, 2000]. Multiple methods for assessing linkage and/or association were performed to extract the greatest amount of information from each dataset. Parametric and non-parametric linkage analyses were conducted using Merlin [Abecasis et al., 2002]. Two parametric models based on segregation analysis were tested on the 179 extended families [de Andrade et al., 1998; Rebbeck et al., 1993; Wang et al., 1988]. Model I used penetrances of 0.000, 0.020, 0.494 and 0.0, 0.008, 0.358 while model II used 0.0, 0.002, 0.067 and of 0.0, 0.008, 0.358 for males and females, respectively. The rsq option (rsq=0.16) of Merlin was used to remove the effects of LD.

Association in the Presence of Linkage (APL) was used to analyze association because this program incorporates data even when a parental genotype is missing [Chung et al., 2006]. The APL software was also used to assess transmission of 2-SNP haplotypes within a gene or gene clusters. Generalized Estimating Equations (GEE) were used to detect gene-gene interactions for all SNPs [Hancock et al., 2007].

Identification of transcription factor binding sites

In silico analyses of all the over-transmitted SNP sequences in regulatory regions were performed using three online algorithms: Alibaba2, Patch and Transcription Element Search Software (TESS) [Grabe 2002; Matys et al., 2006; Schug 2003]. Ancestral and alternate (www.ncbi.nlm.nih.gov/) allele sequences were inputted into the program and the output was compared.

This study was reviewed and approved by the Committee for the Protection of Human Subjects at the University of Texas Heath Science Center (HSC-MS-04-239).

RESULTS

The 956T>A/319M>K HOXD10 mutation was not detected in any of the 494 probands (267 NHW and 227 Hispanic) or the 595 unrelated controls (380 NHW and 215 Hispanic). This excluded a maximum carrier frequency 0.005 with over 80% power in both ethnic groups.

We next interrogated the HOXA and HOXD gene clusters and IGFBP3 in the discovery population. Only rs10255707 was not in HWE (p<0.0001) in either ethnic group (data not shown). Comparisons of SNP allele frequencies between NHW and Hispanics detected significant differences in the majority of SNPs in all genes even after Bonferroni correction (p<0.001) (Table I). Therefore, the data were stratified by ethnicity and examined separately. The data was combined across ethnicities for SNPs showing no differences in frequency.

Affected individuals had slightly higher levels of linkage disequilibrium (LD) than unaffected individuals in both ethnic groups (Supplemental Tables I–IV). Parametric and nonparametric multipoint linkage analysis detected suggestive evidence (LOD>1.5) for linkage to IGFBP3 in the Hispanic subset (Table II). No evidence of linkage was found for HOXA or HOXD with clubfoot (data not shown).

Table II.

IGFBP3 multipoint linkage results for discovery Hispanic sample

			Nonparametric		Parametric
Gene	dbSNP	Position	LOD	p-value	HLODa
IGFBP3	rs2132571	SNP21	2.06	0.001	1.93
	rs2854746	SNP24	2.13	0.0009	2.01
	rs2453839	SNP30	2.22	0.0007	2.12
	rs6670	SNP31	2.29	0.0006	2.19
	rs13223993	SNP32	2.29	0.0006	2.19

^aAll families linked under model II

The APL analysis in the discovery population found significant evidence for association with SNPs in the HOXA and HOXD gene clusters and IGFBP3 (Table III). The strongest associations were seen for rs3801776 in the combined discovery NHW multiplex and simplex group (p=0.004) and when stratified by family history, and for rs13223993 in IGFBP3 in the simplex families (p=0.003) (Table III). Additionally, a strong association for rs6749771 in HOXD was found in the NHW simplex families (p=0.009). In contrast, in the Hispanic sample, the strongest association was with rs2428431 (p=0.008) in the HOXA gene cluster, with no evidence for association found with SNPs in HOXD or IGFBP3. Interestingly, when the ethnicities were combined, rs2428431 in HOXA was still significant (Table III).

Table III.

Results of APL analysis of HoxA, HoxD and IGFBP3 in discovery sample ^a,^b

			All			Multiplex		Simplex
	dbSNP	Pos	All Ethn.	NHW	Hisp	NHW	Hisp	NHW	Hisp
HoxD	rs6749771	SNP1	—	0.015	0.899	0.505	0.699	0.009	0.754
	rs1446575	SNP2	—	0.021	0.334	0.026	0.440	0.224	0.134
HoxA	rs6668	SNP13	0.101	0.802	0.025	0.968	0.119	0.872	0.073
	rs2428431	SNP14	0.006	0.345	0.008	0.158	0.081	0.876	0.043
	rs3801776	SNP16	—	0.004	0.143	0.032	0.114	0.046	0.449
	r3779456	SP17	—	0.179	0.845	0.042	0.459	0.980	0.702
IGFBP3	rs3793345	SNP26	—	0.015	0.307	0.066	0.497	0.101	0.393
	rs2471551	SNP27	0.045	0.032	0.556	0.041	0.489	0.311	0.769
	rs13223993	SNP32	—	0.047	0.320	0.852	0.398	0.003	0.552

^aSNPs with p<0.05 shown in bold,

^bp-values uncorrected for multiple testing,

NHW=nonHispanic white, Hisp.=Hispanic

Significantly over-transmitted haplotypes for HOXA and HOXD were seen in NHW and only for HOXA in Hispanics (Table IV and Supplemental Tables 5 – 7). These haplotypes primarily involve the SNPs identified in the single SNP association analyses (Table III). Among the NHW, the most significant haplotype for HOXA involved rs3801776 (HOXA9) and rs1859164 (HOXA10), which were located in the 5’ region of the HOXA cluster; both HOXA9 and HOXA10 are expressed during limb development (Table IV). In Hispanics, haplotypes involving rs2428431 or rs2428431 (HOXA) were also over-transmitted (0.006≤p≤0.009, Table IV and Supplemental Table V).

Table IV.

Overtransmitted HoxA and HoxD haplotypes in discovery sample^a,^b

A. nonHispanic White
	Haplotype	p-value
HoxD	SNP1/SNP2	0.007
	SNP1/SNP6	0.007
	SNP1/SNP7	0.006
HoxA	SNP16/SNP18	0.004

B. Hispanic
	Haplotype	p-value
HoxA	SNP14/SNP15	0.009
	SNP14/SNP19	0.006
	SNP16/SNP12	0.006

^aHaplotypes with p<0.01 shown,

^bp-values uncorrected for multiple testing

There was no evidence for interactions between HOXA and HOXD in the discovery NHW group. However for the discovery Hispanic dataset, there was evidence for interactions between HOXA and HOXD for rs6758117/ rs2428431 and rs1867863/ rs3801776 (p=0.006 and p=0.007, respectively, data not shown). Interactions were found in the discovery sample for SNPs in HOXA/HOXD and IGFBP3 (Table V). The strongest interaction was between rs3801776 (HOXA) and rs13223993 (IGFBP3) in the NHW subset (p=0.0001, Table V and Supplemental Table IX and X). Interestingly, significant HOXA interactions were found for rs13223993 and 5’ HOX SNPs expressed in limb development in both NHW and Hispanic groups.

Table V.

Gene-gene interactions for Hox and IGFBP3 SNPs^a,^b

A. Discovery
Gene1	SNP1	Gene2	SNP2	p-value	Pop
HOXD	rs2113563 (SNP6)	IGFBP3	rs3793345 (SNP26)	0.008	NHW
	rs2113563 (SNP6)		rs2471551 (SNP27)	0.007	NHW
	rs2592394 (SNP7)		rs3793345 (SNP26)	0.003	NHW
	rs2592394 (SNP7)		rs2471551 (SNP27)	0.001	NHW
	rs741610 (SNP9)		rs13223993 (SNP32)	0.002	H
	rs711812 (SNP10)		rs13223993 (SNP32)	0.003	H
HOXA	rs3801776 (SNP16)		rs13223993 (SNP32)	0.0001	NHW
	rs6968828 (SNP19)		rs2854747 (SNP25)	0.003	NHW
	rs6968828 (SNP19)		rs3110697 (SNP28)	0.001	NHW

B. Validation
Gene1	SNP1	Gene2	SNP2	p-value
HOXD	rs1542180 (SNP4)	IGFBP3	rs2854747 (SNP25)	0.002
	rs1542180 (SNP4)		rs3110697 (SNP28)	0.0003
	rs1542180 (SNP4)		rs2453839 (SNP30)	0.002
	rs2592394 (SNP7)		rs3110697 (SNP28)	<0.0001
	rs2592394 (SNP7)		rs2453839 (SNP30)	0.004
	rs2592394 (SNP7)		rs2854747 (SNP25)	0.0006
	rs847146 (SNP8)		rs2854747 (SNP25)	0.004
	rs847146 (SNP8)		rs3110697 (SNP28)	0.001
HOXA	rs2462907 (SNP12)		rs2471551 (SNP27)	0.007

^aOnly p<0.01 shown, p<0.001 in bold,

NHW=nonHispanic white H=Hispanic,

^bp-values uncorrected for multiple testing

All SNPs were then tested in the validation dataset. There was evidence for association with rs3801776 in HOXA (p=0.028). Analysis of 2-SNP haplotypes in this sample identified seven HOXA haplotypes with altered transmission, none of which were identified in the discovery population (Table IV and Supplemental Table VIII). The most significant haplotypes in the validation population were rs3779456/ rs3807598 (p=0.007) and rs1859164/ rs3807598 (p=0.0003, Supplemental Table VIII).

There was no evidence for an interaction between SNPs in HOXA and HOXD in the validation dataset (data not shown). In contrast, multiple interactions between SNPs in HOXA and IGFBP3 and HOXD and IGFBP3 were identified in the validation population; the majority of these interactions involved SNPs in HOXD and SNPs in IGFBP3 (Table VI and Supplemental Table XI) but are different from those found in the discovery population (Supplemental Tables IX and XI).

Table VI.

Gene-gene interactions for Hox IGFBP3 and mitochondrial-mediated apoptotic variants in discovery dataset^a,^b

A. nonHispanic White
Gene 1	SNP 1	Gene 2	SNP 2	p-value
Casp3	rs1049216	HoxD	rs6749771 (SNP1)	0.002
Casp3	rs1405937		rs6749771 (SNP1)	0.005
Casp3	rs1049253		rs2592394 (SNP7)	0.001
Casp3	rs1049216		rs2592394 (SNP7)	0.008
Bid	rs8919		rs6758117 (SNP11)	0.007
Casp3	rs2720378	HoxA	rs6668 (SNP13)	0.004
Casp3	rs1049216		rs1859164 (SNP18)	0.004
Casp3	rs2720378		rs1859164 (SNP18)	0.004
Casp10	rs3900115		rs3779456 (SNP17)	0.008
Bid	rs181405		rs3779456 (SNP17)	0.001
Bid	rs181405		rs1859164 (SNP18)	0.005
Apaf1	rs2132572	IGFBP3	rs2132572 (SNP22)	0.008
Casp3	rs1049210		rs2132572 (SNP22)	<.0001
Casp3	rs1049210		rs2854744 (SNP23)	0.006
Casp3	rs1049210		rs10255707 (SNP29)	0.002
Casp3	rs1049210		rs2854746 (SNP24)	0.005
Casp8	rs3769825		rs10255707 (SNP29)	0.003
Casp10	rs3900115		rs2854747 (SNP25)	0.008

B. Hispanic
Gene 1	SNP 1	Gene 2	SNP 2	p-value
Casp3	rs4647602	HoxD	rs1318778 (SNP3)	0.0005
Casp3	rs4647602		rs2113563 (SNP6)	0.009
Casp3	rs2696057		rs741610 (SNP9)	0.008
Bcl2	rs1564483	HoxA	rs3801776 (SNP16)	0.002
Casp9	rs4233533		rs3779456 (SNP17)	0.001
Apaf1	rs1866477		rs6968828 (SNP19)	0.009
Apaf1	rs2278361	IGFBP3	rs2132571 (SNP21)	0.007
Apaf1	rs2278361		rs2453839 (SNP30)	0.002
Apaf1	rs7968661		rs2453839 (SNP30)	0.006

^aonly p<0.01 shown,

^bp-values uncorrected for multiple testing

We have previously reported an association of variants in mitochondrial-mediated apoptotic genes with clubfoot in our discovery dataset [Ester et al., 2007]. Since the genes in the current study were involved in both apoptosis and limb formation, we combined the data from the previous study with that of the current study to evaluate potential gene-gene interactions in the discovery sample. The results are presented in Table VI and Supplemental Tables XII and XIII. While multiple interactions between HOX and IGFBP3 genes with mitochondrial-mediated genes were found in both ethnicities, none of these interactions having a p<0.01 were the same in both ethnicities. There was, however, strong evidence for interaction between SNPs in HOXA and HOXD and SNPs in both CASP3 and Bid in NHW (Table VIA). The strongest interaction was between a SNP in CASP3 and one in IGFBP3 in NHW (Table VIA). In Hispanics, the strongest interactions were between SNPs in CASP3 and HOXD; two SNPs in IGFBP3 interacted with two SNPs in Apaf1 (Table VIB).

Because the HOX clusters were controlled by many different regulatory factors, SNPs in the intergenic regions of the clusters may be in DNA enhancer or suppressor binding sites. Therefore, each of the over-transmitted SNPs in these regions was assessed using three transcription factor binding site prediction algorithms and the results are shown in Table VII. rs2428431 was predicted to change ancestral binding sites by all three programs, but predicted binding factors differed (Table VII). In contrast, the ancestral rs3801776 allele was not identified as a potential binding site by any of the programs, while the alternative allele was identified as a potential binding site by two of the programs (Table VII). However, rs6749771 was not predicted to change transcription factor binding sites (data not shown).

Table VII.

Predicted transcription factor binding sites for HoxA, HoxD and IGFBP3 SNPs in S1 sample ^a,^b

		Alibaba2		Patch		TESS
		Alleles		Alleles		Alleles
		Ancestral	Alternate	Ancestral	Alternate	Ancestral	Alternate
HoxA	rs2428431 (SNP14)	MIG1, Sp1	None	Sp1, AML1	NF-E	TCF2α, MIG1, AML1	NF-E, RAF
	rs3801776 (SNP16)	None	1-Oct	None	None	None	RAF

^aOnly SNPs with p<0.01 in regulatory regions shown,

^bAbbreviations: AML1: acute myeloid leukemia 1, MIG1: MIG1, NF-E: nuclear factor E, Oct-1: octomer-binding transcription factor 1, RAF: RAF, Sp1: simian virus 40 protein 1, TCF2α: transcription factor 2 α

DISCUSSION

This study was undertaken to assess the roles of limb and muscle morphogenesis genes in clubfoot. The 956T>A/319M>K HOXD10 mutation that causes syndromic vertical talus was not found in any cases or controls. This result confirms previous findings and provides further evidence that this mutation does not cause isolated clubfoot [Gurnett et al., 2007]. Next, we evaluated whether variants in the HOXA and HOXD gene clusters and IGFBP3 are associated with clubfoot in our family-based dataset. This was accomplished by genotyping SNPs spanning the HOXA, HOXD and IGFBP3 genes first in our discovery sample with follow-up in our validation sample. Located 3’ of the HOXD cluster, SNP1 showed altered transmission and may regulate expression of genes within the cluster. In the HOXA gene cluster, rs2428431 and rs3801776 were significantly over-transmitted. In intron 1 of HOXA2, rs2428431 is changes predicted DNA binding sites (Table VII). The HOXA2 gene is not expressed in the limb, but we cannot exclude the possibility that rs2428431 may regulate expression levels of other upstream 5’ HOXA genes or is in linkage disequilibrium with SNPs that do [Sasaki et al., 2007; Tarchini and Duboule, 2006]. Located in the HOXA9 basal promoter region, rs3801776 showed altered transmission in NHW in the discovery sample and was confirmed in the validation group. Furthermore, pair-wise haplotypes involving rs3801776 were significantly over-transmitted and gene-gene interactions involving this SNP were found. The HOXA9 gene is expressed in both the proximal and distal limb and the variant allele is predicted to introduce a novel DNA binding site (Table VII). This change could alter expression levels of HOXA9 and affect normal limb or muscle development. Future studies are aimed at determining whether rs3801776 is functional or if it is in linkage disequilibrium with a functional change.

Three SNPs (rs3793345, rs2471551 and rs13223993) in IGFBP3 were over-transmitted in the discovery sample. Interestingly, rs3793345 and rs2471551 are located in a potential enhancer/regulatory region in intron 1 and the SNP variants are predicted to change DNA binding sites (data not shown). Located in the 3’ UTR of IGFBP3, rs13223993 showed the most significant results. Previously, SNPs in 3’ UTRs are have been shown to regulate protein expression through microRNAs or proteins that can stabilize or destabilize the half-life of mRNA [Wickens et al., 2002]. Additionally, rs2132572 and rs2854747 in IGFBP3 were only marginally significant in the validation group but not the discovery dataset and do not change DNA binding sites (data not shown). Functional studies are necessary to further assess these SNP variants that could potentially change DNA binding and protein expression and thus affect limb development [Wolters and Wijmenga, 2008].

Because clubfoot is a complex disorder, multiple genes are likely involved although a major gene effect cannot be excluded. This allows us to postulate a model in which multiple genes interact and together are necessary but individually are not sufficient to cause disease. Moreover, we suggest that multiple variants within a gene may cause perturbation of gene function, rather than a single common variant, and thus any one SNP may not demonstrate significant evidence (very small p-values) for association. Therefore association within a gene is considered replication, even if different SNPs are over-transmitted in the two datasets. This standard of replication also was applied to gene-gene interactions. Here, we find that rs3801776 is associated with clubfoot, and in addition, we find that variation in apoptotic genes interact with several variants in HOXA and HOXD. These results are consistent with a recent study demonstrating that apoptosis is important for muscle morphogenesis [Rodriguez-Guzman et al., 2007].

Apoptosis is associated with interdigital removal of excess skin; however, it has recently been shown to play a far more subtle role in later limb development by shaping muscle and tendons [Rodriguez-Guzman et al., 2007]. Specifically, CASP3 has been shown to modulate apoptosis in later muscle and tendon development. While we did not find a strong association with CASP3 and clubfoot alone [Ester et al., 2007], we found evidence for interactions between CASP3 and variants in HOX genes and IGFBP3. This implies that individual variation in CASP3 may not be sufficient to cause clubfoot but that the interaction with genes in other limb and muscle developmental pathways, such as HOXA and HOXD, are necessary in order to disrupt the developmental process. The interactions observed between the many mitochondrial-mediated apoptotic genes and HOXA and HOXD genes suggest that there may be a common mechanism controlling these two pathways. Perturbations of this regulatory pathway may also contribute to clubfoot.

The goal of this study was to identify high-risk haplotypes that would contribute to clubfoot. The first step in this process is to identify genetic variants that are associated with clubfoot and to determine whether they interact. The results of our studies suggest a model for clubfoot that is shown in Figure 1 and includes the interactions between different genes. It is possible that environmental exposures alone can cause clubfoot, but it is more likely that an individual’s genotype confers an increased susceptibility to these exposures. This model further suggests that severity of phenotype may relate to the number of genetic variants and/or genetic pathways affected by the variants (Fig. 1). In this study, we found evidence for an association between a basal promoter SNP in HOXA9 and clubfoot, and provide further evidence that interactions between genes involved in limb development play a role in clubfoot.

Figure 1.