Identification of a Linkage Disequilibrium Block in Chromosome 1q Associated With BMD in Premenopausal White Women

Abstract

Osteoporosis is a complex disease with both genetic and environmental risk factors. A major determinant of osteoporotic fractures is peak BMD obtained during young adulthood. We previously reported linkage of chromosome 1q (LOD = 4.3) with variation in spinal areal BMD in healthy premenopausal white women. In this study, we used a two-stage genotyping approach to identify genes in the linked region that contributed to the variation of femoral neck and lumbar spine areal BMD. In the first stage, 654 SNPs across the linked region were genotyped in a sample of 1309 premenopausal white women. The most significant evidence of association for lumbar spine (p = 1.3 × 10⁻⁶) was found with rs1127091 in the GATAD2B gene. In the second stage, 52 SNPs around this candidate gene were genotyped in an expanded sample of 1692 white women. Significant evidence of association with spinal BMD (p < 10⁻⁵), and to a lesser extent with femoral neck BMD, was observed with eight SNPs within a single 230-kb linkage disequilibrium (LD) block. The most significant SNP (p = 3.4 × 10⁻⁷) accounted for >2.5% of the variation in spinal BMD in these women. The 230-kb LD block contains 11 genes, but because of the extensive LD, the specific gene(s) contributing to the variation in BMD could not be determined. In conclusion, the significant association between spinal BMD and SNPs in the 230-kb LD block in chromosome 1q indicates that genetic factor(s) in this block plays an important role in peak spinal BMD in healthy premenopausal white women.

INTRODUCTION

Age-related osteoporosis is a common skeletal disease, affecting up to 30% (6–9 million) of postmenopausal white women in the United States.(1–4) The disease is characterized by a reduction in bone strength caused, in large part, by a decrease in BMD, leading to an increased risk of fractures from minor trauma. A major predictor of bone strength and osteoporotic fracture risk in later life is peak BMD achieved during early adulthood.(5) BMD is a complex quantitative trait, which involves the interaction of multiple genetic and environmental components.(5) Environmental factors, including nutrition and physical activity, influence the attainment of BMD; however, genetic factors account for as much as 80% of the variability in BMD attained in early adulthood.(5–7)

To identify genes contributing to variation in BMD observed among healthy individuals, several ongoing studies have performed linkage analysis using a variety of bone phenotypes. Results suggest that multiple genes, rather than a single major gene, contribute to the observed variation in BMD.(8) Furthermore, there is growing evidence that BMD is influenced by sex-specific genes(8–10) and skeletal site-specific genes for hip and spine.(8,10–12) In a large sample of premenopausal American white sisters, we previously found significant evidence of linkage of spinal BMD to chromosome 1q (LOD = 4.3).(11,12) This linkage was corroborated in a recent meta-analysis of linkage data from multiple independent studies.(8) A variety of mouse studies also support the presence of genes influencing bone mass in this chromosomal region.(13–17) Similar to the studies in humans, this chromosomal region seems to affect BMD only in female mice.

To identify the gene(s) underlying these linkage findings, we genotyped 654 SNPs across the chromosome 1q region in healthy premenopausal white women. Because of their high density in the human genome, SNPs are particularly useful in association analysis that exploits linkage disequilibrium (LD) to map genes contributing to a phenotype. The most significant association found in the 654 SNPs was followed up with additional SNP genotyping in an enlarged sample of women. Here we report a 230-kb LD block in chromosome 1q, which is likely to harbor sequence variations that contribute to the normal variation in spinal BMD in healthy premenopausal white women.

MATERIALS AND METHODS

Sample

Families made up of healthy premenopausal white sisters were recruited from the state of Indiana to identify genes contributing to bone mass.(11,12) Recruitment focused on families with two or more healthy sisters (N = 766). In addition, 633 parents of the sisters donated a blood sample for DNA extraction. A detailed medical history of the sisters was obtained through administration of health and lifestyle questionnaires. Studies were performed at the General Clinical Research Center of Indiana University School of Medicine. The study was approved by the Institutional Review Board of Indiana University-Purdue University Indianapolis. Informed written consent was obtained from all subjects before their participation in the study.

The sample of sisters ranged in age from 20 to 51 yr. Sisters were required to be within 10 yr of each other in age. Women who had irregular menses or a history of pregnancy or lactation within 3 mo before enrollment were excluded. Women taking oral contraceptives were not excluded. Additional exclusion criteria included a history of chronic disease, use of medications known to affect bone mass or metabolism, and an inability to have BMD measured because of obesity.

BMD, height, and weight measurements

Areal BMD (g/cm²) at the lumbar spine (vertebrae L₂–L₄) and femoral neck were measured by DXA, using two DPX-L and one Prodigy instrument (GE Lunar Corp., Madison, WI, USA). All three instruments were cross-calibrated weekly using a step-wedge phantom. There was no detectable systematic difference between the three machines over the course of the study. The CV on duplicate measurements after repositioning the subject was 1.0% for femoral neck and 0.52% for lumbar spine. Sisters were measured on the same instrument usually at the same visit. BMD was not measured in the parents. Height and weight were measured using a Harpenden stadiometer and a Scale-Tronix weighing scale, respectively.

SNP genotyping

SNP genotyping was performed in two stages.

Stage I:

The Illumina Fine Map SNP Panel 4, which contains SNPs limited to chromosome 1q, was genotyped in a sample of 1309 premenopausal white women and 273 parents from 578 families (Table 1). Genotyping was performed at the Center for Inherited Disease Research (CIDR) on a BeadLab system using the Illumina Golden Gate assay.(18) CIDR released genotypic data for 654 SNPs to Indiana University for analyses. The average SNP spacing across the region was 61 kb. The SNP minor allele frequencies ranged from 0.0 to 0.50, with a mean of 0.26. The Mendelian inconsistencies (0.008%) were removed using the similar method described below (see Stage II). The overall missing genotypic data were 0.13%. As part of internal quality control, CIDR genotyped a parent-child trio plus one duplicate child, which yielded a duplicate error rate of 0.014% and an overall parent-child discordance rate of 0.017%.

Table 1.

Sample Characteristics of Premenopausal White Women

	Stage I		Stage II
	Mean ± SD	Range	Mean ± SD	Range
Number of families	578	—	766	—
Number of sisters	1309	—	1692	—
Number of genotyped parents	273	—	633	—
Age (yr)	33.3 ± 7.3	20.0–50.7	33.1 ± 7.2	20.0–50.7
Height (cm)	165.3 ± 6.1	146.5–192.3	165.5 ± 6.0	146.5–192.3
Weight (kg)	70.1 ± 16.8	41.2–166.0	69.9 ± 16.5	41.2–166.0
Lumbar spine BMD (g/cm2)	1.27 ± 0.14	0.80–1.78	1.27 ± 0.14	0.80–1.78
Femoral neck BMD (g/cm2)	1.02 ± 0.14	0.63–1.46	1.02 ± 0.13	0.63–1.60

Genotyping quality was evaluated by computing Hardy-Weinberg equilibrium using the program Haploview.(19) Markers with significant (p < 0.001) deviation from Hardy-Weinberg equilibrium were removed from further analysis (N = 1). Markers with a minor allele frequency <0.05 were removed from further analysis because of their low informativeness (N = 42). SNPs outside of our chromosome 1q linkage region were also excluded (N = 2). Thus, a net of 609 SNPs, which spanned 143–183 Mb on chromosome 1q, were analyzed to test for an association with areal BMD (see Statistical analyses).

Stage II:

The most significant SNP in stage I, along with 51 additional SNPs surrounding it, were evaluated in an enlarged sample, which was made up of all individuals in stage I augmented with sister families that had been recruited subsequently. The enlarged sample was made up of 1692 premenopausal white women and 633 parents from 766 families (Table 1). The 51 SNPs were selected to ensure that the entire LD block containing the most significant SNP was covered and that some SNPs were in the flanking LD blocks. The LD blocks were defined by a method described by Gabriel et al.(20) Most of the 51 SNPs were selected using SNP Tagger(21) by setting the minor allele frequency in the CEPH Utah population to be at least 0.2 in the HapMap database (http://www.hapmap.org). Based on their potential functional importance, nonsynonymous SNPs were also included when possible. Genotyping was performed using the Sequenom iPLEX genotyping assays on the MassARRAY platform, which is based on allele-specific primer extension with mass-modified terminators (Sequenom, San Diego, CA, USA). The average missing rate for the iPLEX genotyping assays was 2.6%, with a range from 0.5% to 8.0%. Using one randomly selected sister in each family, each SNP was tested for significant (p < 0.001) deviation from Hardy-Weinberg equilibrium. None of the SNPs significantly deviated from Hardy-Weinberg equilibrium. In addition, the SNPs were genotyped in the available parents to identify Mendelian inconsistencies in the SNP genotypes, using the program PedCheck.(22) Inconsistent genotypes were individually reviewed, and the minimum number of genotypes was removed to resolve the Mendelian inconsistencies.

Statistical analyses

Stepwise regression analysis was performed using height, weight, oral contraceptive use, pack-years of smoking, and age. Only age and weight were significant covariates and accounted for 17.2% and 12.1% of the phenotypic variation in spine and hip BMD, respectively. Therefore, regression residuals, representing age- and weight-adjusted BMD, were computed and used in all subsequent analyses.

Previous analyses using genotypic data from a 10-cM microsatellite genome screen(10) did not provide evidence for population stratification in our sample of premenopausal white women.(23) The program Haploview(19) was used to examine the extent of LD between the SNPs to ensure that the SNP density was sufficient to evaluate evidence of association. LD (D′ statistic) was evaluated using one randomly selected individual in each family. The program SNP Tagger(21) (http://www.broad.mit.edu/mpg/tagger/) was used to estimate how well the selected SNPs represented the genetic information contained in nongenotyped SNPs in the region containing the most significant SNP. The extent to which the genotyped SNPs correlated with all HapMap SNPs in this region was evaluated for different levels of LD, with the r ² statistic.

In stage I, a population-based association test was performed using a linear mixed model framework. SNP genotype was modeled as a fixed effect (taking on three levels corresponding to the observed genotypes) and family (sibship) as a random effect in the mixed model. Correlation between subjects in the same family was modeled by assuming an error covariance matrix exhibiting compound symmetry. The model was fitted using the MIXED procedure in the SAS statistical software (version 9.1; SAS Institute, Cary, NC, USA). The mixed model association test uses data from all siblings with both genotype and phenotype. Significant SNPs were defined as those with an association with areal spine BMD that exceeded the p < 0.01 threshold.

In stage II, association analyses were performed using the same mixed model approach. Given the large size of our samples, we used a threshold of p < 0.01 to ensure that we detected robust evidence of association and a clinically significant portion of the BMD variability. Mean BMD comparisons between genotypes for specific SNPs were also made with the mixed model. The proportion of BMD variation explained by each SNP was estimated by the r ² measure for the mixed model framework as proposed by Xu.(24) This measure is analogous to the common r ² measure in traditional ANOVA analysis.

Gene expression analysis

To prioritize genes in the region associated with BMD, expression of the genes in bone-related tissues/cells was tested by RT-PCR. First-strand cDNAs were synthesized from 1 μg each of total RNAs, using the Advantage RT-for-PCR Kit (Clontech, Mountain View, CA, USA). Total RNAs used in the cDNA synthesis were from bone (Stratagene, La Jolla, CA, USA); bone marrow (BD Biosciences Clontech); osteogenic sarcoma (SaOS-2) and osteosarcoma (MG-63) (Ambion, Austin, TX, USA); and osteoclasts.(25) The synthesized cDNAs were used for PCR amplification using the Multiplex Kit (Qiagen, Valencia, CA, USA). PCR primers were designed using Primer3 v.0.4.0 (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi). The RT-PCR products were resolved on 2% agarose gels and visualized under the UV light.

RESULTS

Sample characteristics

A sample of 1692 premenopausal white women completed detailed phenotypic assessment (Table 1). BMD at lumbar spine and femoral neck, age, height, and weight were essentially the same in the women studied in stage I and stage II.

Stage I

Three SNPs provided significant evidence of association (p < 0.01) with lumbar spine BMD, and five SNPs yielded significant association with femoral neck BMD (Table 2; Fig. 1). The most significant SNP with spinal BMD (p = 1.3 × 10⁻⁶) was rs1127091 located in exon 11 (3′ untranslated region) of the GATAD2B gene. This SNP was also associated with femoral neck BMD (p = 6.9 × 10⁻⁴). The nearest SNPs on either side were located >120 kb away and did not provide significant evidence of association at the p < 0.01 level.

Table 2.

SNPs Significantly Associated With BMD in Stage I

SNP	Position (bp)*	Gene	Phenotype	p
rs1127091	152,046,036	GASTAD2B	Lumbar spine	1.3 × 10−6
rs736022	168,954,629	PRRXI	Lumbar spine	3.1 × 10−3
rs1569475	167,890,304		Lumbar spine	5.7 × 10−3
rs1127091	152,046,036	GATAD2B	Femoral neck	6.9 × 10−4
rs990707	174,045,596		Femoral neck	1.4 × 10−3
rs1750293	174,073,672		Femoral neck	1.8 × 10−3
rs1046332	148,084,132	HIST2H2AA3/4†	Femoral neck	1.8 × 10−3
rs764024	164,800,141		Femoral neck	8.5 × 10−3

* Chromosome positions are based on the NCBI Human Genome Assembly v.36.2.

^† Located ~3.2 kb upstream of rs1046332.

FIG. 1.

Stage I association tests using the 609 SNPs in the chromosome 1q region. The chromosome location is shown in the baseline (dbSNP Build 128); −log of p values are shown on the y-axis. The circles indicate rs1127091 in GATAD2B. Blue diamond, lumbar spine BMD; red square, femoral neck BMD.

Stage II

To confirm the evidence of association in the region around GATAD2B, 51 additional SNPs, as well as rs1127091, were genotyped in an enlarged sample of 1692 white women. The LD block harboring rs1127091, which is defined by the method described by Gabriel et al.,(20) spans ∼230 kb in our sample (block 3 in Fig. 2). The LD pattern in our sample was consistent with that observed in the HapMap sample of Utah residents with European ancestry. To further evaluate how well the genotyped SNPs represented the known sequence variation in this chromosomal region, the genotypic information from these 52 SNPs were compared with the genotypic data available from all SNPs genotyped in this region by the International HapMap Project.(26) From this comparison, it was estimated that the 52 SNPs captured the genotypic information (with an r ² ≥ 0.5) contained in 97% of the SNPs in the region that have a minor allele frequency ≥ 0.05.

FIG. 2.

LD (D′) structure around the GATAD2B gene. LD was computed between the genotyped SNPs in this region (see Table 3). The pairwise D′ statistic is shown within each box (range, 0–100; 0 indicating no LD; 100 indicating complete LD). The degree of shading represents the extent of LD. Darker shades indicate greater LD, lighter shades indicate less LD. When no number is shown within a box and the box is black, the corresponding D′ is 100 (complete linkage disequilibrium). The two SNPs that define LD block 3 are marked with thick black lines.

Among the 52 SNPs analyzed, 11 SNPs (including rs1127091) within the LD block 3 were associated with spinal BMD (p = 3.4 × 10⁻⁷–8.5 × 10⁻⁴), whereas those in the flanking LD blocks showed no evidence of association (all p > 0.01) except rs4845605 (SNP 41), which has high LD with SNPs in block 3 (Fig. 3; Table 3). Many of these same SNPs were also significantly associated with femoral neck BMD, although the strength of the association was weaker (p = 4.3 × 10⁻⁴–5.6 × 10⁻³). The eight most significant SNPs (p < 10⁻⁵) were highly polymorphic, with a minor allele frequency between 0.46 and 0.48.

FIG. 3.

Stage II association tests using the 52 SNPs around GATAD2B. The arrows denote genes located in this region (the orientation in the genome). The chromosome location is shown in the baseline (dbSNP Build 128); −log of p values are shown on the y-axis. The circles indicate rs1127091 in GATAD2B. Blue diamond, lumbar spine BMD; red square, femoral neck BMD.

Table 3.

Stage II SNPs and Association Results

	SNP	Gene	Position within gene	Position (bp)*	MAF†	Lumbar spine p value	Femoral neck p value
1	rs11264342	INTS3	Intron 1	151,977,892	0.48	0.37	0.57
2	rs12041194	INTS3	Intron 3	151,980,876	0.41	0.67	0.33
3	rs9426902	INTS3	Intron 3	151,985,370	0.44	0.32	0.48
4	rs6424268	INTS3	Intron 5	151,987,314	0.46	0.098	0.86
5	rs4345834	INTS3	Intron 25	152,009,941	0.43	0.19	0.74
6	rs6663011	INST3	Exon 28 (T945T)	152,011,752	0.07	0.55	0.59
7	rs10158450	SLC27A3	Intron 9	152,018,763	0.48	0.010	0.93
8	rs12030242			152,019,703	0.17	0.26	0.33
9	rs10908474			152,020,349	0.35	0.75	0.73
10	rs6665637			152,022,707	0.28	0.046	0.45
11	rs10796953	LOC343052	Exon 2	152,034,803	0.12	0.076	0.13
12	rs12738361	LOC343052	Exon 3	152,035,364	0.47	9.5 × 10−7	6.5 × 10−4
13	rs11583896	LOC343052	Exon 4	152,035,606	0.29	0.037	0.60
14	rs9426935	LOC343052	Intron 4	152,036,024	0.46	2.6 × 10−6	3.9 × 10−3
15	rs12133923	LOC343052	Intron 8	152,039,864	0.45	4.3 × 10−4	0.022
16	rs4434872	LOC343052	Exon 12	152,040,900	0.17	0.010	0.013
17	rs1139620			152,044,472	0.48	9.1 × 10−7	1.5 × 10−3
18	rs1127092	GATAD2B	Exon 11 (3′ UTR)	152,045,860	0.12	0.027	0.038
19	rs1127091	GATAD2B	Exon 11 (3′ UTR))	152,046,036	0.46	2.4 × 10−6	5.1 × 10−4
20	rs9426938	GATAD2B	Exon 11 (3′ UTR)	152,046,922	0.47	7.8 × 10−7	4.5 × 10−4
21	rs4363451	GATAD2B	Intron 5	152,057,125	0.30	0.046	0.59
22	rs9427232	GATAD2B	Intron 2	152,059,582	0.47	3.4 × 10−7	4.3 × 10−4
23	rs6427298	GATAD2B	Intron 2	152,061,620	0.12	0.12	0.34
24	rs10796956	GATAD2B	Intron 2	152,062,193	0.30	0.073	0.80
25	rs7512766			152,077,188	0.29	0.052	0.59
26	rs12129468			152,088,568	0.37	8.5 × 10−4	56 × 10−3
27	rs7552274			152,100,215	0.29	0.026	0.48
28	rs7414227			152,115,620	0.47	7.4 × 10−7	1.4 × 10−3
29	rs12043354			152,121,048	0.13	0.013	0.048
30	rs12033532			152,155,444	0.13	0.023	0.051
31	rs17364880			152,155,666	0.11	0.82	0.037
32	rs10494303			152,159,647	0.41	3.6 × 10−4	0.068
33	rs17364880	SLC39AI	Exon 5 (3′ UTR)	152,198,894	0.12	0.71	0.35
34	rs11264743	CREB3L4	Exon 3 (P95S)	152,208,138	0.28	0.063	0.35
35	rs4845360	NUP210L	Intron 34	152,233,284	0.46	3.9 × 10−6	4.6 × 10−3
36	rs11264824	NUP210L	Intron 33	152,239,795	0.12	0.022	0.28
37	rs11264875	NUP210L	Exon 28 (V1215I)	152,261,271	0.28	0.25	0.80
38	rs11264886	NUP210L	Intron 26	152,263,970	0.28	0.29	0.75
39	rs4845597	NUP210L	Intron 8	152,356,062	0.33	0.68	0.86
40	rs6667928			152,381,329	0.33	0.61	0.87
41	rs4845605	TPM3	Intron 2	152,425,011	0.49	4.2 × 10−3	0.69
42	rs1194604	UBAP2L	Intron 15	152,493,863	0.46	0.018	0.73
43	rs1194600	UBAP2L	Intron 24	152,502,633	0.46	0.013	0.66
44	rs1212352	UBAP2L	Intron 25	152,505,907	0.39	0.99	0.78
45	rs4103781			152,547,246	0.50	0.22	0.68
46	rs11265530	LOC653690	Intron 1	152,551,676	0.27	0.59	0.82
47	rs6685323	AQP10	Exon 3 (H123Y)	152,562,216	0.31	0.86	0.69
48	rs6702754	ATP8B2	Exon 6 (N120N)	152,570,600	0.31	0.85	0.57
49	rs11586379	ATP8B2	Intron 11	152,574,254	0.29	0.50	0.68
50	rs1194587	ATP8B2	Exon 21 (S759S)	152,583,777	0.47	0.42	0.66
51	rs1194585	ATP8B2	Intron 24	152,585,188	0.47	0.43	0.64
52	rs10908831			152,595,167	0.42	0.88	0.70

* Chromosome positions are based on the NCBI Human Genome Assembly v.36.2.

^† Minor allele frequencies calculated using one sibling from each family.

To further examine the association with BMD, we focused on the most significant SNP, rs9427232, which is located in intron 2 of GATAD2B. As shown in Table 4, women homozygous for the G allele had significantly higher mean lumbar spine BMD compared with women who were homozygous for the A allele or heterozygous at this marker. Femoral neck BMD was also significantly greater in women with the GG genotype compared with the heterozygous women, with a trend toward significance in women with the AA genotype (p = 0.10). Using the r ² measure proposed by Xu,(24) we found that the genotypes at the eight significantly associated SNPs accounted for 2.0–2.6% of the variation in spinal BMD and up to 1.1% of femoral neck BMD.

Table 4.

Mean BMD ± SE (G/CM²) as a Function of Rs9427232 Genotype

Genotype	N	Lumbar spine	Femoral neck
AA	428	1.270 ± 0.006*	1.023 ± 0.005
AG	807	1.257 ± 0.005*	1.010 ± 0.004*
GG	397	1.298 ± 0.006	1.040 ± 0.006

* p < 0.001 for comparison with mean BMD of GG genotype.

Candidate genes

The 230-kb LD block between rs10796953 and rs11264886 (SNPs 11 and 38) harbors nine annotated genes and two uncharacterized genes (Fig. 3; Table 3). However, high LD extends to the region between rs12030242 and rs1212352 (SNPs 8 and 44), which contains four additional genes. Because of the high LD spanning the region (∼500 kb), it is impossible to determine which gene (and functional sequence variations) affects BMD on the basis of association. We speculated that genes expressed in bone would have a higher likelihood of affecting peak BMD. Therefore, we tested for expression of these genes in bone-related tissues or cells by RT-PCR. Because it is unclear where LD is completely lost, two genes immediately outside of the extended LD region (between rs12030242 and rs1212352) were also included in this analysis. Thirteen of these genes were expressed in all bone-related tissues or cell lines (Table 5). Thus, this approach did not substantially limit the number of genes, and most remain viable candidates.

Table 5.

Gene Expression in Bone-Related Tissues and Cell Lines

Gene symbol	Gene name	Bone	Osteoclast	Bone marrow	Osteogenic sarcoma (SaOS-2)	Osteosarcoma (MG-63)
SLC27A3	Solute carrier family 27 (fatty acid transporter), member 3	+	+	+	+	+
LOC343052	Similar to putative neuronal cell adhesion molecule	−	−	−	−	−
GATAD2B	GATA zinc finger domain containing 2B	+	+	+	+	+
RP11–216N14.7 (LOC645965)	Adipose differentiation-related protein pseudogene	+	+	+	+	+
DENND4B	DENN/MADD domain containing 4B	+	+	+	+	+
CRTC2	CREB regulated transcription coactivator 2	+	+	+	+	+
SLC39A1	solute carrier family 39 (zinc transporter), member 1	(+)	+	(+)	+	+
CREB3L4	cAMP responsive element binding protein 3-like 4	(+)	+	+	+	+
JTB	Jumping translocation breakpoint	+	+	+	+	+
RAB13	RAB13, member RAS oncogene family	+	+	+	+	+
RPS27	Ribosomal protein S27 (metallopanstimulin 1)	ND	ND	ND	ND	ND
NUP210L	Nucleoporin 210 kDa-like	−	−	−	−	(+)
TPM3	Tropomyosin 3	+	+	+	+	+
Clorf189	Chromosome 1 open reading frame 189	(+)	+	−	(+)	(+)
Clorf43	Chromosome 1 open reading frame 43	+	+	+	+	+
UBAP2L	Ubiquitin associated protein 2-like	(+)	+	(+)	+	+
HAX1	HCLS1-associated protein X-1	+	+	+	+	+

The presence and absence of gene expression determined by RT-PCR are denoted by + and −, respectively. Weak, but detectable, amplification is denoted by (+).

Expression of RPS27 could not be determined reliably because of the presence of highly homologous genes in the human genome.

ND, not determined.

DISCUSSION

We previously detected linkage of spinal BMD to chromosome 1q in a sample of premenopausal white women(11,12); however, the linked region was broad, spanning 40 cM. To explore this linkage, we performed moderate-density SNP genotyping of the linked region (143–183 Mb) in healthy white women and found that rs1127091 in GATAD2B was highly associated with spinal BMD. Subsequent analysis of 52 SNPs in the enlarged sample of white women confirmed that eight highly polymorphic SNPs in or around GATAD2B were highly associated with spinal BMD. The most significant SNP (rs9427232) accounted for >2.5% of the spinal BMD variation, strongly suggesting that genetic variations around GATAD2B influence spinal BMD in premenopausal white women. The same SNPs were also associated, but to a lesser extent, with femoral neck BMD, suggesting that the gene(s) have a generalized effect on BMD. Because the region of high LD spans >500 kb, it is statistically impossible to identify which of the 17 genes located in this region is responsible for the observed association. The expression data of the 17 genes allowed us to gain some insight into which genes might be important in bone function; however, expression in these skeletal-related tissues does not guarantee a role in the attainment of BMD. Thus, additional functional studies will be necessary to identify which gene(s) or regulatory element(s) influence spinal and femoral BMD.

Among the 11 genes located in the 230-kb LD block, GATAD2B and SLC39A1 are of particular interest because of their reported functions. The GATAD2B gene encodes P66beta (GATA zinc finger domain containing 2B), which is a subunit of the methyl-CpG-binding protein, MeCP1 complex.(27,28) The complex represses transcription through preferential binding, remodeling, and deacetylation of methylated nucleosomes,(29) which might regulate genes involved in bone development.

The SLC39A1 gene encodes zinc transporter 1 (ZIP1). Zinc deficiency is associated with retardation of bone growth and development of osteoporosis.(30,31) Zinc is also known to increase mineralization of osteoblast-like cells.(32) A recent study shows that SLC39A1 is expressed in osteoclast and may inhibit osteoclast function through zinc uptake.(33) In addition, overexpression of ZIP1 induces expression of osteoblast-specific markers, including alkaline phosphatase, osteopontin, and Cbfa1/Runx2 and increases differentiation and mineralization of pluripotent mesenchymal stem cells.(34) These lines of evidence suggest that expression of ZIP1 may play an important role in bone development.

Our study had several strengths. First, we had a relatively large sample of 1692 healthy premenopausal white women, yielding 90% power to detect an association accounting for only 1% of the variation in BMD with an α = 0.001. Second, previous analyses indicated that there was no evidence of stratification in our sample; therefore, population stratification was unlikely to account for the significant associations found in our population sample.(23) Third, rather than testing only a few SNPs, we tested a range of SNPs that spanned from strong LD to no LD with the most significant SNP (rs9427232). This also ensured that all genes in this region were appropriately tested for association with BMD. In addition, LD structure allowed us to compare the pattern of association and reduce the likelihood of a false-positive association. Finally, we selected SNPs that had high heterozygosity to maximize our statistical power to detect association.

In summary, we detected consistent evidence of significant association between spinal BMD and common SNPs in the 230-kb LD block on chromosome 1q. The sequence variations in this block explain a clinically meaningful portion of the spinal BMD variation in premenopausal white women. Positive association results in this study have significant clinical implications for osteoporosis because identification of multiple genes with small effects could lead to development of a potential screening panel for osteoporosis risk in healthy individuals and also allow for early institution of preventative measures and/or treatment against future bone loss. However, because of the large LD block spanning >200 kb, it remains challenging to identify specific gene(s) or regulatory element(s) responsible for the observed association.