Association of Bone Morphogenetic Proteins With Otosclerosis

Abstract

We studied the role of polymorphisms in 13 candidate genes on the risk of otosclerosis in two large independent case-control sets. We found significant association in both populations with BMP2 and BMP4, implicating these two genes in the pathogenesis of this disease.

Introduction

Otosclerosis is a progressive disorder of the human temporal bone that leads to conductive hearing loss and in some cases sensorineural or mixed hearing loss. In a few families, it segregates as a monogenic disease with reduced penetrance, but in most patients, otosclerosis is more appropriately considered a complex disorder influenced by genetic and environmental factors.

Materials and Methods

To identify major genetic factors in otosclerosis, we used a candidate gene approach to study two large independent case-control sets of Belgian-Dutch and French origin. Tag single nucleotide polymorphisms (SNPs) in 13 candidate susceptibility genes were studied in a stepwise strategy.

Results

Two SNPs were identified that showed the same significant effect in both populations. The first SNP, rs3178250, is located in the 3′ untranslated region of BMP2. Individuals homozygote for the C allele are protected against otosclerosis (combined populations: p = 2.2 × 10⁻⁴; OR = 2.027; 95% CI = 1.380–2.979). The second SNP, rs17563, is an amino acid changing (p.Ala152Val) SNP located in BMP4. The G allele, coding for the amino acid alanine, confers susceptibility in both populations (combined populations: p = 0.002; OR = 1.209; 95% CI: 1.070–1.370).

Conclusions

These results indicate that polymorphisms in the BMP2 and BMP4 genes, both members of the TGF-β superfamily, contribute to the susceptibility to otosclerosis and further strengthen the results from the recently reported association of TGFB1 with this disease.

INTRODUCTION

Otosclerosis is a primary disorder of bone metabolism unique to the human temporal bone and is characterized by disordered resorption and deposition of bone. The disease shows distinct stages of bone resorption, new bone formation, and final eburnation of the affected bone.(1) The first active and vascular stage is referred to as otospongiosis.(2) A predilection site for otosclerotic foci is a region anterior to the oval window.(3) Otosclerotic foci in this region may cause fixation of the stapedial footplate, leading to conductive hearing loss. Otosclerosis can also be associated with sensorineural hearing loss. In this case, it is called cochlear otosclerosis.(4)

Otosclerosis can be divided into histological and clinical types.(3) Clinical otosclerosis is one of the most frequent causes of acquired hearing loss in whites, with a prevalence of 0.3–0.4%.(5) It leads to conductive and/or sensorineural hearing loss and affects more females than males.(3) However, otosclerosis occurs more frequently as a histological disease in which clinical symptoms are absent. In this case, it is only discovered in histological temporal bone sections and through CT. This form has a prevalence of 2.5% in the white population.(5) In African blacks, Asians, and American Indians, otosclerosis is rare.(6) Despite extensive research to identify factors involved in the etiology of otosclerosis, the true cause and pathogenesis remain to be elucidated. Various hypotheses about the etiology have been proposed, including genetic factors, immunologic factors, and viral infection. Among viral factors, mainly persistent measles virus infection has been proposed as the causative agent.(7)

Strong familial aggregation of individuals affected by otosclerosis typically shows an autosomal dominant pattern with reduced penetrance and has facilitated the identification of six loci for otosclerosis on chromosome 15q25–26 (OTSC1), chromosome 7q34–36 (OTSC2), chromosome 6p21.2–22.3 (OTSC3), chromosome 16q21–23.2 (OTSC4), chromosome 3q22–24 (OTSC5), and chromosome 6q13–16.1 (OTSC7).(8–13) Two additional loci (OTSC6 and OTSC8) have been reserved with the Human Genome Organization nomenclature committee, although these localizations have not been published yet. None of the corresponding genes has been cloned yet.

In most patients with otosclerosis, however, mendelian segregation of the disease is absent, suggesting that otosclerosis is more appropriately considered a complex disease caused by environmental and genetic factors. Associations with COL1A1 have been reported for the complex form,(14–16) although these associations fail to replicate across populations.(17) Recently, a first susceptibility gene (TGFB1) was successfully identified to be associated with the complex form of otosclerosis in two independent populations.(18)

TGFB1 is an important factor in bone remodeling(19) and in the induction of embryogenesis in the otic capsule. It can either stimulate chondrogenesis to promote otic capsule growth or selectively inhibit this process to permit perilymphatic space formation and capsular modeling.(20) Functional analysis of a variant of TGF-β1 (rs1800472; p.Thr263Ile) showed that the protective variant (Ile) is more active, suggesting that this allele decreases otosclerosis susceptibility by inhibiting osteoclast differentiation and activation and thereby preventing the otospongiotic phase of the disease.(18)

This study aims at the identification of genetic variants that influence the risk of otosclerosis by means of a genetic association study. We have selected 13 candidate genes that (1) are in the TGFB1 interacting network; (2) have a putative function in the metabolism and chondrogenesis of the otic capsule; (3) are known to be involved with syndromic or nonsyndromic forms of stapes fixation; and (4) are candidates based on specific hypotheses about the etiology of otosclerosis (Table 1). Tag single nucleotide polymorphisms (SNPs) were selected across these candidate genes and were genotyped in two large independent populations in a stepwise strategy.

Table 1.

Selected Candidate Genes

Gene	Abbreviation	Function/potential role in disease
Bone morphogenetic protein 2	BMP2	Role in otic capsule chondrogenesis and bone formation(35,57)
Bone morphogenetic protein 4	BMP4	Role in otic capsule chondrogenesis and bone formation(58,59)
		Involved in the generation of inner ear sensory epithelia(60)
Noggin	NOG	Antagonist BMPs, with high affinity for BMP2 and 4(61)
		Mutations in this gene cause syndromic stapes fixation(55,56)
Fibroblast growth factor 2	FGF2	Role in otic capsule chondrogenesis(62)
		Synergistic interaction with TGF-β1 is required for induction otic capsule formation(62)
		Localized in higher concentration lining the otic capsule in the Palmerston North autoimmune mouse, which develops otic capsule sclerotic lesions(63)
Osteoprotegerin	TNFRSF11B	Primary regulator of bone metabolism(64)
		Abnormal bone remodeling within the otic capsule and progressive hearing loss in knockout mice(65)
Retinoic acid receptor α	RARA	Crucial role in initial differentiation of otic placode derivatives(66)
		Blocking of RARA in cultured periotic mesenchyme from otocysts reduces levels of TGF-β1 and suppresses chondrogenesis(67)
Otoraplin	OTOR	Role in initiation of periotic mesenchym chondrogenesis(68)
Parathyroid hormone	PTH	Key regulator in bone turnover(69)
Parathyroid hormone receptor 1	PTHR1	Receptor that binds PTH and PTHLH(70)
		Lower expression of mRNA in otosclerotic stapes(71)
Diastrophic sulfate transporter	SLC26A2	Acts as a sulphate exchanger in chondrocytes(72)
		Mutations in this gene cause diastrophic dysplasia, a systematic skeletal disorder(73)
		Has a higher activity in cells from stapes and the external auditory canal of otosclerosis patients(74)
POU domain class 3, transcription factor 4	POU3F4	Mutations in this gene cause DFN3, characterized by a conductive hearing loss due to fixation of the stapes and progressive sensorineural hearing loss(75)
		Expressed in regions of the otic capsule that immediately surround the stapes during otic development(76)
Membrane co-factor protein	CD46	Receptor of measles virus(77)
Signaling lymphocyte activation molecule	SLAMF1	Receptor of measles virus(78)

MATERIALS AND METHODS

Clinical diagnosis

Individuals in whom otosclerosis was surgically confirmed were considered affected. In nonoperated persons, the diagnosis of otosclerosis was based on audiologic and clinical data. Pure-tone audiometry was performed in all persons with air conduction at 125, 250, 500, 1000, 2000, 4000, and 8000 Hz and bone conduction at 250, 500, 1000, 2000, and 4000 Hz. Otoscopy and tympanostomy were performed to exclude outer or middle ear pathology, respectively, and stapedial reflexes were measured to asses the mobility of the stapes.

Populations

This project took place according to the guidelines of the ethical committee of the University of Antwerp. Informed consent was obtained from each patient. A total of 632 unrelated Belgian or Dutch otosclerosis patients were collected by the Department of Otolaryngology of different hospitals: University Hospital of Antwerp (Belgium), St-Augustinus Hospital, Antwerp (Belgium), University Hospital of Ghent (Belgium), and the University Medical Center St-Radboud, Nijmegen (The Netherlands). The same number of unrelated controls (632) were selected from a DNA repository of the Department of Medical Genetics, Antwerp, and matched according to sex and ethnicity. A total of 455 unrelated French otosclerosis patients were collected through the Jean Causse Ear Clinic in Colombiers, France. A total of 485 unrelated control samples, matched based on ethnicity were collected by the Laboratoire de Genetique Moleculaire et Chromosomique CHU, IURC Montpellier in France. Controls for both populations were not screened for otosclerosis because the frequency of the disease is only 0.3–0.4%. Calculations have shown that the power loss under these circumstances is negligible.(21) Genomic DNA was isolated from either fresh or frozen blood, using standard techniques.

SNP selection and genotyping

A total of 92 SNPs were selected for genotyping from the international HapMap project phase II (release 20), including one SNP (rs4988235) to assess population stratification. TagSNPs were selected using the aggressive (multimarker) method with the program Tagger,(22) an implement of haploview (version 3.2).(23) The minimum allele frequency was set at 0.05, and the r ² threshold was set at 0.8. In addition to the tagSNPs, the Tagger program provides specific combinations of two or three single SNP alleles (haplotypes) that serve as proxies for a hidden SNP. We refer to these haplotypes as multimarker predictions. SNPs located in coding regions or SNPs that were previously associated with related diseases were preferentially included as tagSNPs using the force include option in the Tagger program. For BMP2, one extra SNP (rs1049007) from the dbSNP database (www.ncbi.nlm.nih.gov/projects/SNP/) was included, because this variant was previously shown to be associated with BMD.(24) To include regulatory or possible promotor regions, a margin of ∼3000 bp was taken around the transcribed region of each gene. An exception was made for POU3F4, where a margin of 1 MB was taken upstream of the gene, because microdeletions and more complex chromosomal arrangements associated with DFN3 have been located up to 900 kb upstream to the gene.(25,26) Genotyping was performed by Kbiosciences (Herts, UK), using a competitive allele-specific PCR system (KASPar) (www.kbioscience.co.uk).

Quality control methods

Samples with a poor call rate (>10% missing calls) were excluded from the analysis in the Belgian-Dutch population. In the French population, samples were excluded if more than three calls were missing, because the number of SNPs was limited. After removal of samples with a poor call rate, SNPs with a bad call rate (>3% missing calls) were excluded in both populations. Hidden intraplate duplicate controls were used as an additional quality control of the assay. Hardy-Weinberg equilibrium (HWE) was calculated on the controls using the program Genotype Transposer (version 1.0).(27) SNPs with p < 0.001 were excluded.(28) For SNPs located on the X-chromosome, HWE was only calculated for females. In the Belgian-Dutch population, unknown family relationships and individuals with an outlying genetic background were traced using the program Graphical Representation of Relationships (GRR) and the program CHECKHET, respectively.(29,30) Traced individuals were excluded. This procedure was not done in the French population because the number of SNPs genotyped in this population was small.

Association testing

Analyses were performed using the statistical program R (2.3.1) (www.r-project.org) and SPSS 12.0 for Windows (SPSS, Chicago, IL, USA). Association testing was performed using the likelihood ratio test (LRT), in a logistic regression framework. The genotype was coded linearly, thereby assuming an additive model. To test the association between a SNP and otosclerosis susceptibility, allowing for effect modification (interaction) by sex, a logistic regression using backward selection was performed. First, a saturated model was fitted including main effects for sex and genotypes and the interaction term sex × genotype. If the interaction term was significant, meaning that sex has an influence on the effect of the SNP, females and males were tested separately. If the interaction term was not significant, the term was omitted, and a new model was fitted including only the main effect for genotype. SNPs on the X-chromosome were tested separately in males and females. Genotype frequencies were inspected visually to detect possible dominant/recessive effects. If present, genotypes of the dominant allele were combined. Correction for multiple testing was performed on all the calculated single SNP p values using a false discovery rate (FDR) analysis. The method of Storey and Tibshirani was used,(31) as implemented in the R package Qvalue. Multimarker predictions were calculated using the program WHAP (SNP haplotype analysis package).(32) For each multimarker prediction, we tested the specific haplotype that predicts the hidden SNPs versus all other haplotypes combined.

The subset of SNPs genotyped in the French population was chosen based on their significance in the Belgian-Dutch population. When interaction with sex was found in the Belgian-Dutch population, this SNP was also tested separately for males and females in the French population. Similarly, when a dominant/recessive effect was tested in the Belgian-Dutch population, this was also done in the French population.

To determine whether a significant association in the French population represented a true replication of a significant association in the Belgian-Dutch population, we tested for homogeneity of the genetic effect across the two populations. In a logistic regression model that included genotype, population, and the interaction population × genotype, an LRT was performed to test the significance of the interaction term. Nonsignificance of the interaction term implies that the effect of the SNP on disease susceptibility is not significantly different between the two populations. In this case, the common genetic effect across the two populations was estimated by fitting a reduced logistic regression model that only included the main effects for genotype and population. The common OR and its 95% Wald CI were estimated by the regression coefficient for genotype and its large-sample CI, respectively. The p values for the common genetic effect were obtained through the LRT comparing a model including population and genotype to a model only including population as a predictor. For moderately large sample sizes within strata, this test usually gives results similar to the Cochran-Mantel-Haenszel test. An advantage of using a model-based approach is that it provides estimates of effect size.(33)

RESULTS

Quality control

After quality control, the final Belgian-Dutch population consisted of 1224 samples (608 controls and 616 patients). SNP rs7054313 and SNP rs11573897 were excluded because of, respectively, a bad call rate and deviation from HWE (data not shown). All samples and SNPs passed quality control in the French population. SNP assays had a 0% error rate in both populations for the hidden duplicate controls, with the exception of rs7061408 in the Belgian-Dutch population, showing only one error, which was considered to be acceptable.

Association testing in the Belgian-Dutch population

A total of 92 SNPs across 13 candidate genes (including SNP rs4988235 selected to asses population stratification) were genotyped in the Belgian-Dutch population. Association analysis was performed on the 90 SNPs that passed quality control. Because otosclerosis is a disease that affects more females then males, we studied sex × genotype interaction, testing whether the effect of a SNP differs between males and females. Assuming an additive model, seven SNPs showed a significant interaction term, and for these SNPs, the effect of genotype was tested separately. A total of 11 SNPs showed a significant difference between patients and controls in either females, males, or both sexes together under an additive model. Systematic inspection of genotype frequencies revealed a potential dominant/recessive effect in SNPs rs3178250 (BMP2) and rs167428 (FGF2). Both SNPs were significant acting through a dominant/recessive effect (Table 2); no interaction for sex was detected for these two SNPs under a dominant/recessive model (data not shown). SNP rs4988235, located near the lactase gene (LCT), showed no significant difference in allelic frequency between patients and controls (Table 2). Analysis of multimarker predictions did not reveal a significant result (data not shown).

Table 2.

SNPs With Significant p Values for Association With Clinical Otosclerosis in the Belgian-Dutch Population

Gene	SNP	Region	Nucleotide change*	Amino acid change†	Interaction‡	Additive§	Dom/Rec¶
BMP2	rs3178250	3′UTR	g.6700201T>C		0.303	0.215	0.007
BMP4	rs17563	Exon 4	g.35417272A>G	p.Ala152Val	0.404	0.028
CD46	rs2796267	5′ gene flanking region	g.1442685G>A		0.022	F:0.392
						M:0.024
	rs2796270	Intronic	g.1455391G>A		0.003	F:0.062
						M:0.021
FGF2	rs167428	Intronic	g.48321586T>C		0.560	0.052	0.027
	rs17407577	Intronic	g.48327488T>C		0.275	0.044
	rs1960669	Intronic	g.48330715C>A		0.982	0.006
OTOR	rs6044284	3′ gene flanking region	g.16674689G>A		0.349	0.016
PTH	rs6254	Intronic	g.12301504C>T		0.003	F:0.008
						M:0.087
	rs751610	Intronic	g.12303620C>T		0.009	F:0.116
						M:0.039
RARA	rs4890109	Intronic	g.2232400G>T		0.867	0.012
SLAMF1	rs2295612	Exon 1	g.11107058G>T	p.Phe11Leu	0.637	0.019
LCT	rs4988235**	5′ gene flanking region	g.25316568G>A		0.472	0.242

* The following nucleotide reference sequences were used:for BMP2, NT_011387.8; for BMP4, NT_026437.11; for CD46, NT_021877.18; for FGF2, NT_016354.18; for OTOR, NT_011387.8; for PTH, NT_009237.17; for RARA, NT_010755.15; for SLAMF1, NT_004487.18; for LCT, NT_022135.15.

^† The following protein reference sequences were used:for BMP4, NP_001193.1; for SLAMF1, NP_003028.1.

^‡ Interaction for sex assuming an additive model.

^§ p value assuming an additive model; F, females; M, males.

^¶ p value assuming a dominant/recessive model.

** Included to assess population stratification.

To account for multiple testing, we performed an FDR analysis on all the single SNP p values. Results are shown for all the significant SNPs in Table 3, with the exception of rs1485286. The q value is the expected proportion of false positives among all features as or more extreme than the observed one.(31) If, for example, the p value of rs17407577 is declared significant, all SNPs with a lower p value are declared significant as well. The q value of 0.305 indicates that, among these 12 SNPs, 30.5% are expected to be false. Conversely, this means that 3 or 4 of the 12 SNPs that are nominally significant are expected to be false.

Table 3.

FDR Analysis of All the Significant SNPs in the Belgian-Dutch Population

SNP	p value	q value
rs1960669	0.006	0.224
rs3178250	0.007*	0.224
rs6254	0.008†	0.224
rs4890109	0.012	0.235
rs6044284	0.016	0.235
rs2295612	0.019	0.235
rs2796270	0.021‡	0.235
rs2796267	0.024‡	0.235
rs167428	0.027*	0.235
rs17563	0.028	0.235
rs751610	0.039‡	0.298
rs17407577	0.044	0.305

* p value assuming a dominant/ recessive model.

^† p value in females.

^‡ p value in males.

Association testing in the French population

To confirm the positive associations detected in the Belgian-Dutch population, a replication study in a French population was completed. All SNPs that were significant in the Belgian-Dutch population, under either an additive or a dominant/recessive model, were analyzed in the French population, with the exception of rs1485286 that showed borderline significance only in males. A total of 12 SNPs were selected (Table 2). For none of the SNPs was the sex × genotype interaction term significant under an additive model (data not shown). Nevertheless, SNPs showing a significant sex × genotype interaction in the Belgian-Dutch population (rs2796267, rs2796270, rs6254, rs751610) were also analyzed separately in males and females in the French population. rs3178250 and rs167428 were also tested for a dominant/recessive effect, because this effect was detected in the Belgian-Dutch population (Table 2). The sex × genotype interaction was not significant assuming a dominant/recessive model for these two SNPs (data not shown). Results from the French population are shown in Table 4. Three SNPs showed a significant p value in this population: rs3178250 located in the untranslated region (UTR) of BMP2, rs17563 located in exon 4 of BMP4, and rs2295612 located in exon 1 of SLAMF1. SNP rs4988235 (LCT), showed no significant difference between patients and controls (Table 4).

Table 4.

Analysis of SNPs in the French Population

Gene	SNP	Additive*	Dom/Rec†
BMP2	rs3178250	0.088	0.011
BMP4	rs17563	0.037
CD46	rs2796267	0.559
		F: 0.355
		M: 0.694
	rs2796270	0.682
		F: 0.836
		M: 0.910
FGF2	rs167428	0.713	0.786
	rs17407577	0.909
	rs1960669	0.736
OTOR	rs6044284	0.110
PTH	rs6254	0.833
		F: 0.538
		M: 0.985
	rs751610	0.589
		F: 0.711
		M: 0.608
RARA	rs4890109	0.359
SLAMF1	rs2295612	0.013
LCT	rs4988235‡	0.541

Significant p values are marked in bold.

* p value assuming an additive model; F, females; M, males.

^† p value assuming a dominant/recessive model.

^‡ Included to assess population stratification.

Combined analyses of the Belgian-Dutch and the French population

Although three of the SNPs showed a significant association in both populations, true replication most likely occurs if the disease-associated allele is the same in both populations. Therefore, we tested the homogeneity of the SNP effect on the phenotype by logistic regression. If there is no significant interaction between genotype and population, the effect of the SNP on the disease susceptibility does not differ significantly between the two populations. For SNP rs2295612 in the SLAMF1 gene, interaction was very significant, and opposite allelic effects were found. In this case, the association was most likely a false positive. In contrast, SNP rs3178250 (BMP2) and rs17563 (BMP4) showed the same effect in both populations (Table 5; Fig. 1). For each of the two SNPs, a common OR was estimated, and a p value was calculated for both populations combined (Table 5). For SNP rs3178250 (BMP2), the C allele is recessive and protective against disease. The common OR, comparing homozygotes for the C allele versus carriers of the T allele, was estimated to be 2.027, which means that homozygotes for the C allele are two times more protected against disease than carriers of the T allele. For SNP rs17563 (BMP4), subjects possessing the G allele have a higher risk in developing otosclerosis, compatible with an additive model. The common allelic OR was estimated to be 1.209, which means that the odds of disease increase multiplicatively with a factor 1.2 for every extra copy of the G allele.

Table 5.

Analysis Combining Both Populations

Gene	SNP	Homogeneity test	p value common genotype effect	OR [95% CI]
BMP2	rs3178250	0.935*	2.2 × 10−4*	2.027 [1.380–2.979]*
BMP4	rs17563	0.902†	0.002†	1.209 [1.070-1.370]†
SLAMF1	rs2295612	6.6 × 10−4†	NA	NA

* Assuming a dominant/recessive model.

^† Assuming an additive model.

NA, not applicable.

FIG. 1.

Bar charts of SNP rs3178250 (BMP2) and rs17563 (BMP4) showing the distribution of genotypes in the Belgian-Dutch and French population. (A) Bar chart of rs3178250 in the Belgian-Dutch population. Number of patients: CC, 23; CT, 207; TT, 378. Number of controls: CC, 44; CT, 189; TT, 371. (B) Bar chart of rs3178250 in the French population. Number of patients: CC, 18; CT, 165; TT, 271. Number of controls: CC, 38; CT, 171; TT, 275. (C) Bar chart of rs17563 in the Belgian-Dutch population. Number of patients: GG, 203; GA, 319; AA, 94. Number of controls: GG, 176; GA, 307; AA, 121. (D) Bar chart of rs17563 in the French population. Number of patients: GG, 114; GA, 243; AA, 94. Number of controls: GG, 105; GA, 247; AA, 129.

DISCUSSION

Although several monogenic loci for otosclerosis have been mapped, attempts to identify the disease-causing genes at these loci have been unsuccessful. In addition, a candidate gene association study of COL1A1 and otosclerosis has generated contradictory results. Recently, however, TGFB1 has been associated with otosclerosis in two large independent populations.

This study indirectly supports this finding by showing that polymorphisms in BMP2 and BMP4, two members of the TGF-β superfamily, also contribute to otosclerosis susceptibility.

Members of the TGF-β superfamily are involved in many biological contexts.(34) However, it is noteworthy that both BMPs and TGF-β1 are critical regulators of bone formation within this superfamily.(19,35,36) In addition, interplay has been shown between the BMP and TGF-β pathways. For example, TGF-β1 inhibits the expression of BMP2 mRNA in cultures of fetal rat calvarial osteoblasts, whereas BMP2 enhances gene expression of TFGB1 in normal human osteoblast-like cells.(37,38) Given the important functions in both bone regulation and otic capsule chondrogenesis and given the previous association with TGFB1, both BMP2 and BMP4 were considered to be strong candidate genes for this study. BMP2 and BMP4 are not located in one of the previous reported loci in monogenic families with otosclerosis.

Six SNPs in BMP2 were genotyped in the Belgian-Dutch population, one of which (rs3178250) was significant and showed the same effect in both the Belgian-Dutch and French populations (Table 5; Fig. 1). rs3178250 is located in the 3′ UTR of BMP2. Individuals possessing the T allele have a higher risk of developing otosclerosis, whereas individuals homozygous for the C allele seem to be protected against otosclerosis. This could be a direct effect of the SNP as part of a regulatory region or of another unidentified causative variation in linkage disequilibrium (LD) with this SNP. In support of the first possibility is evidence implicating the 3′ UTR of mRNA in the regulation of gene expression by controlling nuclear export, polyadenylation status, subcellular targeting, rates of translation, and degradation of mRNA.(39) Alterations in the 3′ UTR-mediated functions are involved in the pathogenesis of different diseases.(39) In this way, replacement of the C allele by a T allele could lead to aberrant protein regulation. Genetic variants in BMP2 have been associated with low BMD in Icelandic, Danish, and Korean populations,(24,40) and BMP2 is also associated with osteoarthritis.(41)

Two tagSNPs in BMP4 were genotyped in the Belgian-Dutch population. rs17563 showed a significant association with otosclerosis, which was replicated in the French population (Table 5; Fig. 1). BMP4 mRNA is transcribed from four exons, of which the first two exons are untranslated. There is evidence that alternate first exons are used.(42) rs17563 is an amino acid changing (p.Ala152Val) SNP located in exon 4 of BMP4. The G allele, which codes for alanine at position 152 of the protein, is more prevalent in otosclerosis patients (Fig. 1). Although we cannot exclude LD of this SNP with a yet unidentified causative variation elsewhere in BMP4, it is tempting to speculate that carrying the G allele leads to altered protein production or function. The same variant has also been found to be associated with BMD in postmenopausal women.(43) This polymorphism reduces intertrochanter and total hip BMD in subjects homozygous for the alanine coding allele compared with subjects either homozygous or heterozygous for the valine coding allele.(43) Choi et al.(24) have also reported an association between BMP4 and BMD in young Korean men and women but not with the same variant.

It has been reported that patients with osteoporosis have significantly more hearing loss compared with the general population.(44) In addition, Clayton et al.(45) found that osteoporosis is significantly more frequent in female otosclerosis patients compared with presbyacusis patients. In aggregate, these findings and the association of both BMP2 and BMP4 with otosclerosis and osteoporosis, suggest a possible shared genetic etiology. Interestingly, osteoporosis has also been associated with polymorphisms in both TGFB1 and COL1A1.(46,47)

In case-control study designs, undetected population stratification can lead to spurious associations.(48) To prevent this, we used genetically homogeneous populations by matching our controls to our patients by ethnicity and by tracing and excluding individuals with a different genetic background using the program CHECKHET in the Belgian-Dutch population.(30) Homogeneity across the population also enhances the power to detect disease-influencing polymorphisms and markers in LD with them.(30) In addition, an SNP (rs4988235) near the LCT gene was genotyped on both populations especially to look for stratification. The frequency of this marker varies greatly even among closely related populations, making it informative in detecting subtle levels of population stratification in an apparently homogeneous study populations.(49) This SNP did not show any differences between cases and controls in either populations, not indicating the presence of population stratification.

Correction for multiple testing remains a controversial issue in association studies. Because SNPs are not independent from each other because of LD, these comparisons do not represent independent tests. The Bonferroni correction is too conservative in these circumstances, because each new test does not provide a completely independent opportunity for a false-positive result.(50) An alternative approach is the FDR correction. We used the method of Storey and Tibshirani(31) to estimate how many of the association tests reaching a given significance level are expected to be false (Table 3). The results show that less than one third of all nominally significant SNPs are expected to be false, but there is no indication which of the significant SNP p values represent true findings. A replication study is suitable to find SNPs that represent genuine results. This stepwise strategy to replicate associations through a large independent second population, looking for SNPs where the same allele confers susceptibility (harmful or protective), should greatly minimize false association. Because replications in independent populations represent a stronger confirmation of an association than a very low p value,(51) we feel that both BMP2 and BMP4 represent real associations. In contrast, we found several SNPs in other genes, especially FGF2, showing nominally significant associations or showing a trend toward association in the Belgian-Dutch population. Although these associated SNPs failed to replicate in the French population, we cannot conclusively exclude FGF2 or these other genes as important factors involved in the pathogenesis of otosclerosis. Because the French population is slightly smaller in sample size, its power to detect subtle differences is lower and small effects could be overlooked, leading to falsely negative replication. Differences in environmental or genetic backgrounds between both populations can also lead to false-negative results.(52) The latter could be because of the fact that a complicated set of interactions between environmental and genetic factors or between genes could be important in the development of otosclerosis. Furthermore, subtle differences of LD structure between populations could lead to nonreplication.(51,53) It is also possible that different disease-causing alleles could predominate in different study populations.(53) On the other hand, an SNP in SLAMF1 (rs2295612) showed association in both sample sets, but the risk alleles were opposite. This might indicate a false-positive result, although a true replication cannot be ruled out. Theoretical modeling showed that this phenomenon could occur when the studied variant is correlated through interactive effects or LD with the causal variant at another locus.(54)

BMP2 and BMP4 play an important role in otic capsule chondrogenesis, are associated with BMD, and based on this study, are also associated with otosclerosis. Consistent with our results are reports describing mutations in NOG, a BMP2 and BMP4 antagonist, with syndromic stapes fixation and our previous association implicating TGFB1 with otosclerosis.(18,55,56) Functional studies on both proteins focusing on their interplay and interaction with TGF-β1 can provide insight into the pathophysiology of otosclerosis. This insight may facilitate the development of therapeutic approaches to stabilize or prevent the disease. At present, there is only a possibility to cure the conductive component of the hearing loss, but there is no effective treatment for the sensorineural hearing loss that results from cochlear otosclerosis.