Genetic variants affecting bone mineral density and bone mineral content at multiple skeletal sites in Hispanic children

Abstract

Context:

Osteoporosis is a major public health burden with significant economic costs. However, the correlates of bone health in Hispanic children are understudied.

Objective:

We aimed to identify genetic variants associated with bone mineral density (BMD) and bone mineral content (BMC) at multiple skeletal sites in Hispanic children.

Methods:

We conducted a cross-sectional genome-wide linkage analysis, genome-wide and exome-wide association analysis of BMD and BMC. The Viva La Familia Study is a family-based cohort with a total of 1,030 Hispanic children (4–19 years old at baseline) conducted in Houston, TX. BMD and BMC were measured by Dual-energy X-ray absorptiometry.

Results:

Significant heritability were observed for BMC and BMD at multiple skeletal sites ranging between 44 and 68% (P<2.8×10⁻⁹). Significant evidence for linkage was found for BMD of pelvis and left leg on chromosome 7p14, lumbar spine on 20q13 and left rib on 6p21, and BMC of pelvis on chromosome 20q12 and total body on 14q22–23 (logarithm of odds score > 3). We found genome-wide significant association between BMC of right arm and rs762920 at PVALB (P = 4.6×10⁻⁸), and between pelvis BMD and rs7000615 at PTK2B (P = 7.4 ×10⁻⁸). Exome-wide association analysis revealed novel association of variants at MEGF10 and ABRAXAS2 with left arm and lumber spine BMC, respectively (P<9×10⁻⁷).

Conclusions:

We identified novel loci associated with BMC and BMD in Hispanic children, with strongest evidence for PTK2B. The findings provide better understanding of bone genetics and shed light on biological mechanisms underlying BMD and BMC variation.

Introduction

Osteoporosis and low bone mass are currently estimated to be a major public health threat for almost 54 million people aged 50 and older in the U.S [1]. Mexican Americans have a higher overall age-adjusted prevalence of osteoporosis and the most rapid projected increase in osteoporosis burden, compared to other ethnicities [1,2]. However, the correlates of bone health in Hispanic individuals, especially Hispanic children, are understudied.

Genetic factors account for up to 80% of the variance in bone mass [3]. Multiple family-based linkage studies have identified chromosomal regions linked to bone mineral density (BMD) and bone mineral content (BMC) [3,4]. Common variants identified through genome-wide association studies (GWAS) explain only about 5% of the genetic variance in BMC and BMD, and the missing heritability may be explained by low-frequency and rare variants, whose contribution to BMC and BMD are still largely unknown [3,5]. Most genetic studies have been conducted in adults, while only a few studies have investigated the genetic effects on bone in children and adolescents [6–9]. Although genetic variants affecting metabolic diseases have been discovered in both adults and children, some novel variants have been found only in children [9–11]. Compared to adults, bone mass in children is largely affected by bone acquisition and less affected by environmental factors [11]. Pediatric genetic studies provide opportunities to identify novel loci affecting bone health. As genetic studies related to bone health have been conducted in European or Asian populations and mainly in older adults, it is not certain whether those variants are also associated with bone health in Hispanic children. Therefore, we elected to identify genetic variants associated with gold standard Dual-Energy X-Ray Absorptiometry (DXA)-derived BMD and BMC at multiple skeletal sites in a family-based study, Viva La Familia Study (VFS) which includes children and their siblings of ages 4 −19 years [12,13]. By studying a mixed-age pediatric population, the results might be more generalizable to the broad pediatric population, instead of a smaller age range or puberty stage.

Materials and Methods

Study Population

The Viva La Familia Study (VFS).

The VFS is a family study designed to examine the genetic and environmental factors influencing obesity and its comorbidities in 1030 Hispanic children (4–19y) from 319 families, enrolled in Houston, TX between November 2000 and August 2004, with details described previously [14]. Pedigrees were validated using genomic data. Phenotypic assessments were conducted over the course of 3 study visits. During the first visit, interviews were conducted to obtain family pedigree, sociodemographic, lifestyle information, medical histories, and puberty stages. Anthropometric traits were assessed in both children and their parents.

BMC and BMD of skeletal sites were determined by DXA with a Delphi-A whole-body scanner (Hologic Inc., Waltham, MA, USA). We examined total body and lumbar spine bone sites as primary outcomes because these measures are the most reproducible skeletal sites for BMC and BMD in children [15]. As hip BMD is recommended to measure to diagnose osteoporosis in adults [15], we also considered pelvis BMC and BMD in our children cohort as primary outcomes. Other skeletal sites (e.g. leg, arm) were examined as secondary outcomes. Left and right bone sites were considered separately due to limb bone bilateral asymmetry [16].

Written informed consent or assent was given by all enrolled children and their parents. The protocol was approved by the Institutional Review Boards for Human Subject Research at Baylor College of Medicine and Affiliated Hospitals, Texas Biomedical Research Institute and the analysis was approved by University of North Carolina at Chapel Hill.

Genotyping

Genotyping of Microsatellites for Genome-wide Linkage Analysis.

Genotyping of microsatellites has been described in detail previously [17]. In short, DNA was prepared from whole blood with the Wizard Genomic DNA Purification Kit (Promega, Madison, WI, USA). Short tandem repeats (STR) that were spaced at an average interval of 10cM (range, 2.4–24.1), were genotyped for all participants. The autosomal markers used were from the ABI PRISM Linkage Mapping Set-MD10 Version 2.5 (Applied Biosystems, Foster City, CA, USA). Polymerase chains reaction (PCR) was used to amplify each marker and the PCR products were pooled with others with a Robbins Hydra-384 microdispenser. The STRs were quantified using fluorescent emissions by comparing with a standard, and the genotype scoring was performed using the genotype software package (Applied Biosystems). Pedigree errors were detected by PREST and Mendelian errors were detected by SIMWALK2 [18]. LOKI was used to compute the identify-by-descent matrix for linkage analysis [19]. The chromosomal map was developed based on marker locations reported by deCODE Genetics [20].

Genotyping of Common Variants.

Genotyping for 1.1 million SNPs was conducted using marker assays included on the Illumina HumanOmni1-Quad v1.0 BeadChip [21]. Genotype calls were obtained after scanning on the Illumina BeadStation 500GX and were analyzed using the GenomeStudio software. The genotyping error rate (based on duplicates) was 2 per 100,000 genotypes. The average call rate per individual sample was 97%. Specific markers were removed from analysis if they had call rates < 95% (~4000 SNPs) or deviated from Hardy-Weinberg equilibrium at a 5% false discovery rate (12 SNPs). Single nucleotide polymorphisms (SNP) genotypes were checked for Mendelian consistency using the program SimWalk2 [18]. The estimates of allele frequencies and their standard errors were obtained using the software program Sequential Oligogenic Linkage Analysis Routines (SOLAR version 7.6.2) [22].

Whole Exome Sequencing and Genome Variant Identification.

The custom NimbleGen VCRome 2.1 capture reagent was used to capture the entire exome for each DNA sample [23]. The reagent targets coding exons from the consensus coding sequence, Vega human genome annotations, and NCBI RNA reference sequences. Sequencing on the Illumina platform using standard protocols followed capture enrichment and Illumina sequence analysis was conducted based on the Human Genome Sequencing Center’s integrated Mercury pipeline [24,25]. Burrows-Wheeler Aligner was used to map the sequencing reads to the GRCh37 human reference sequence and the Atlas suite was used to call single-nucleotide variants. Variant annotation was accomplished using the Cassandra annotation suite. Quality control was performed using a custom pipeline, which followed the guidelines for the CHARGE-S project [26].

Covariates Measurements

Body weight was measured with a digital balance to the nearest 0.1 kg, and height was measured to the nearest 1 mm with a stadiometer. BMI was calculated as kg/m². BMI Z-score, a relative weight measure adjusted for age and sex, was determined using reference data from the Centers for Disease Control and Prevention (CDC) [27].Tanner stages of sexual maturation were self-reported based on pubic hair and breast and penis development illustrated with drawings. Puberty stages were categorized based on Tanner stages. Children in Tanner stage 1 were considered as “pre-pubertal” and children in other Tanner stages were considered as “pubertal”.

Statistical analysis

Linkage analysis.

The variance components decomposition method was used to estimate heritability and identify the chromosomal location(s) affecting variation in bone outcomes. To estimate the genetic component in the variation of bone outcomes, heritability was estimated first. A multipoint linkage analysis was conducted to find a putative quantitative trait locus (QTL) or loci (QTLs) that affect bone outcomes [22]. The method has been described in detail elsewhere [22]. In brief, this is an extension of the variance components method where a QTL component is added to the basic model. The correlations of phenotype within families were modeled as the cumulative effects of identity by descent (IBD) for the family at a specific QTL associated with a marker, with residual genetic and environmental effects. The model was adjusted for age, sex, age², age × sex, age² × sex and BMI Z- score and binary puberty stages for the genome-wide linkage analysis. A logarithm of odds score (LOD) >3 was considered as significant and a LOD between 2 and 3 was considered as suggestive evidence for linkage. Further association analysis was conducted with SNPs in the one-LOD support interval and relevant bone phenotypes. Bonferroni correction was used for each one-LOD support interval to identify significance for association.

Genome-Wide Association Study (GWAS).

A measured genotype analysis (MGA) approach was used to identify genetic loci associated with bone phenotypes. This approach accounts for both the random effects of kinship and the main effects of SNP genotypes [28]. Inverse normalization was used for all bone outcomes to ensure normality assumptions. Because BMI and puberty stages could affect bone density and bone content among children, the covariates age, sex, age*sex, age², age²*sex, BMI Z-score, and binary puberty stage were included in the model [29]. In SOLAR, each SNP genotype was converted to 0, 1, or 2 based on the number of minor alleles. These were included in variance-components mixed models for MGA compared to null models which included the random effects of kinship and fixed effects of age and sex. The thresholds for genome-wide significant and suggestively significant association were based on the distribution of P values obtained from 10,000 simulation GWASs [21]. Each SNP was tested independently as a 1 degree of freedom likelihood ratio test with a P value threshold for significant evidence of association set at 1.01×10⁻⁷ and suggestive evidence at 1.0 ×10⁻⁶ [21]. The genome-wide P value threshold was computed accounting for kinships. A quantitative transmission disequilibrium test was used to test for population stratification [30]. Conditional analysis was then performed using an additive model for the remaining SNPs, conditioned on the previously identified genome-wide significant SNPs in the GWAS analysis. The software SOLAR (version 7.6.2) was used for the statistical analysis [22].

Exome-Sequencing Association analysis.

Single-variant association analysis was conducted using RVtest software [31]. All analyses used additive genetic models, and the empirical pedigree information was transformed into a kinship matrix and included in the association analysis. Covariates used in this analysis were the same as those included in the GWAS. Principal component analysis (PCA) was conducted by the PC-AiR method in the GENESIS package [32]. Population structure was accounted for by adding the top 10 PCs in the model. Single variant association score summary statistics and variant-specific parameters were generated. Single-variant analysis included variants with minor allele frequency (MAF) greater than or equal to 0.01. The significance level was calculated for each phenotype using a Bonferroni correction.

SNP Lookups for Novel Loci Identified in VFS.

To investigate if SNPs identified in our cohort associated with bone outcomes in multiple skeletal sites in childhood cohorts of other ethnicities and Hispanic adults, we performed a look-up of published pediatric cohort results from the Bone Mineral Density in Childhood Study (BMDCS) [8], and Hispanic adults cohort of the San Antonio Family Osteoporosis Study(SAFOS) [33].

Potential Functional Roles for Identified Loci.

To examine the potential functional roles of identified loci, we looked up the SNPs in GTEx (https://gtexportal.org/home/) and eQTL Catalog (https://eqtl.onderzoek.io/) [34,35].

Results

A total of 1,030 children participated in the study (boys = 510, girls = 520). The mean age and BMI were 11.0 ± 4.1 y and 25.1 ± 7.6 kg/m², respectively. The phenotype and genotype data were available for about 750 children. The heritabilities of BMC and BMD were significant for all skeletal sites ranging from 0.44 to 0.68 (Table 1).

Table 1.

Heritability for BMC and BMD

	Mean	SD	h2	SE	p value
Bone Mineral Content (g)
Left arm	94.3	49.2	0.61	0.083	8.5×10−17
Left leg	291.5	140.4	0.59	0.082	2.3×10−16
Left rib	68.3	28.3	0.44	0.083	2.8×10−9
Lumbar spine	29.6	15.7	0.58	0.081	7.2×10−16
Pelvis	158.1	92.9	0.56	0.085	1.6×10−13
Right arm	100.7	52.6	0.59	0.081	2.4×10−16
Right leg	292.4	139.9	0.57	0.082	1.7×10−15
Right rib	65.1	27.6	0.48	0.083	8.5×10−11
Thoracic spine	68.3	33.6	0.44	0.081	1.1×10−9
Total body	1510.5	635.3	0.6	0.082	8.9×10−17
Bone Mineral Density (g/cm2)
Left arm	0.62	0.10	0.68	0.08	3.4×10−20
Left leg	0.97	0.19	0.51	0.081	1.3×10−12
Left rib	0.57	0.09	0.56	0.081	5.8×10−15
Lumbar spine	0.80	0.18	0.58	0.083	3.7×10−15
Pelvis	1.00	0.24	0.51	0.083	3.4×10−12
Right arm	0.62	0.10	0.66	0.081	5.3×10−20
Right leg	0.97	0.19	0.44	0.079	3.1×10−10
Right rib	0.56	0.09	0.67	0.082	7.2×10−20
Thoracic spine	0.66	0.15	0.44	0.081	4.8×10−10
Total body	0.92	0.14	0.6	0.082	8.8×10−17

The models were adjusted for adjusted for age, age², sex, age*sex, age²*sex, kinship, puberty stage and BMI Z- score.

Linkage analyses are used to identify broad genomic regions that contain the putative disease loci. The genome-wide linkage analysis results are shown in Table 2 and Figure 1. The most significant result was the linkage of pelvis BMD (LOD = 4.8) with genetic loci within the chromosomal region of 7p14 between markers D7S484 and D7S510, which was also linked with left leg BMD (LOD = 3.4). Other regions that were significantly linked with BMD were 20q13 for lumbar spine (LOD = 3.2) and 6p21 for left rib BMD (LOD = 3.0). For BMC, the best LOD score was obtained for pelvis BMC (LOD = 3.4) on chromosomal region of 20q12 followed by total body BMC (LOD = 3.0) on chromosomal region of 14q22. We also identified 10 loci with suggestive evidence of linkage at various chromosomal regions (2 < LOD ≤ 3) for different sites, with details shown in Table 2. To narrow down to smaller regions, association analyses were conducted in the identified statistically significant linkage regions (LOD > 3) for BMC or BMD. The rs6018245 at EYA Transcriptional Coactivator and Phosphatase 2 (EYA2) (β (SE) = −1.41 (0.32), P = 1×10⁻⁵) on 20q13.12 was found to be statistically significantly associated with pelvis BMC based on the extent of multiple testing correction applied. After accounting for rs6018245 on pelvis BMC in conditional linkage, the LOD score on pelvis BMC decreased from 3.4 to 2.4 on chromosome 20, indicating this SNP could explain about 29% of the variation of pelvis BMC.

Table 2.

Genome-wide Linkage Results on BMC and BMD

Trait	Chr	cM	Location	Nearby Markers	LOD
Bone Mineral Content
Left arm	19	65	19q12, 19q13.13	D19S414, D19S220	2.0
Left leg	20	31	20p12.3, 20p12.2	D20S115, D20S186	2.5
Left rib	7	61	7p14.2, 7p14.1	D7S484, D7S510	2.2
Pelvis	20	75	20q12, 20q13.12	D20S107, D20S119	3.4
Right arm	9	35	9p24.1, 9p22.3	D9S286, D9S285	2.4
Right leg	19	65	19q12, 19q13.13	D19S414, D19S220	2.8
Thoracic spine	9	41	9p22.2, 9p21.3	D9S157, D9S171	2.6
Total body	14	65	14q22.3, 14q23.2	D14S276, D14S63	3.0
Bone Mineral Density
Left arm	3	123	3p12.3, 3q12.2	D3S3681, D3S1271	2.6
Left leg	7	61	7p14.2, 7p14.1	D7S484, D7S510	3.4
Left rib	6	64	6p21.2, 6p12.1	D6S1610, D6S257	3.0
Lumbar spine	20	79	20q13.12, 20q13.13	D20S119, D20S178	3.2
Pelvis	7	62	7p14.2, 7p14.1	D7S484, D7S510	4.8
Right leg	7	61	7p14.2, 7p14.1	D7S484, D7S510	2.6
Thoracic spine	1	275	1q42.2, 1q43	D1S2800, D1S2785	2.1
Total body	7	59	7p15.1, 7p14.2	D7S516, D7S484	2.9

The models were adjusted for age, age², sex, age*sex, age²*sex, and kinship. Statistically significant results (Logarithm of Odds score (LOD) >3) are depicted in bold) and suggestive results (LOD>2) are shown.

Figure 1.

Multipoint LOD scores for linkage of BMC and BMD at different skeletal sites (only significant results are shown). BMDLRI, left rib BMD; BMCLRI, left rib BMC; BMDLLE, left leg BMD; BMDPEL, pelvis BMD; BMDRLE, right leg BMD; DXBMD, total body BMD; DXBMC, total body BMC; BMCLLE, left leg BMC; BMCPEL, pelvis BMC; BMDLSP, lumbar spine BMD.

To complement linkage analysis with broad region identification, we also performed genome-wide association study (GWAS) to identify common variants, usually with smaller effects than linkage [36]. In GWAS, two significant loci (rs7000615 and rs762920) were identified, with results for primary and secondary skeletal sites shown in Table 3 and Supplemental Table 1, respectively. The Manhattan plots and regional association plots for the two signals are shown in Figures 2 and 3 respectively. The SNP rs7000615 near the gene encoding protein tyrosine kinase 2 beta (PTK2B) was statistically significantly associated with pelvis BMD (P = 7.4×10⁻⁸). In addition, multiple other SNPs at PTK2B also nominally associated with both pelvis BMC and BMD (Table 3). The SNP rs762920 was found to be genome-wide statistically significantly associated with right arm BMC (P = 4.6×10⁻⁸) and it was also nominally significantly associated with left arm BMC (P = 1.8×10⁻⁷) and right leg BMC (P = 4.1×10⁻⁷) (Supplemental Table 1). In further conditional analysis to identify secondary signals, no statistically significant or suggestive associations were observed.

Table 3.

Genome-wide association results for BMD and BMC

SNP	Effect Allele (Freq)	Chr	Position	Nearest Gene	Total body		Lumbar Spine		Pelvis
					Beta(SE)	P value	Beta(SE)	P value	Beta(SE)	P value
Bone Mineral Density
rs7000615	C(0.36)	8	27405916	PTK2B	−0.22(0.06)	2.1×10−4	−0.24(0.06)	2.2×10−5	−0.28(0.05)	7.4×10−8
rs11775958	C(0.35)	8	27412367	PTK2B	−0.17(0.06)	2.7×10−3	−0.23(0.06)	7.3×10−5	−0.27(0.05)	2.5×10−7
rs891391	A(0.34)	8	27417992	PTK2B	−0.17(0.06)	3.2×10−3	−0.25(0.06)	2.5×10−5	−0.27(0.05)	2.9×10−7
rs1541845	A(0.36)	8	27421883	PTK2B	−0.18(0.06)	2.5×10−3	−0.25(0.06)	1.4×10−5	−0.28(0.05)	1.8×10−7
rs2241652	G(0.36)	8	27422188	PTK2B	−0.19(0.06)	1.6×10−3	−0.25(0.06)	1.8×10−5	−0.28(0.05)	1.6×10−7
rs3739214	A(0.34)	8	27243762	STMN4	−0.19(0.06)	1.3×10−3	−0.29(0.06)	2.4×10−7	−0.22(0.05)	2.5×10−5
Bone Mineral Content
rs12567355	A(0.03)	1	110921002	LRIF1	−0.65(0.15)	1.3×10−5	−0.87(0.17)	2.2×10−7	−0.68(0.16)	3.4×10−5
rs1712674	C(0.03)	1	110923431	LRIF1	−0.61(0.14)	1.8×10−5	−0.81(0.17)	3.9×10−7	−0.63(0.15)	5.5×10−5
rs1780569	C(0.03)	1	110927141	LRIF1	−0.61(0.14)	1.8×10−5	−0.81(0.16)	3.9×10−7	−0.63(0.15)	5.5×10−5
rs3739214	A(0.34)	8	27243762	STMN4	−0.22(0.05)	3.1×10−5	−0.27(0.06)	6.1×10−6	−0.28(0.06)	7.1×10−7
rs3735759	G(0.34)	8	27345522	PTK2B	−0.20(0.05)	1.8×10−4	−0.20(0.06)	5.8×10−4	−0.29(0.06)	4.8×10−7
rs7834529	C(0.35)	8	27353816	PTK2B	−0.20(0.05)	5.7×10−5	−0.18(0.06)	1.4×10−3	−0.28(0.05)	2.4×10−7
rs7000336	A(0.33)	8	27388820	PTK2B	−0.20(0.05)	1.6×10−4	−0.20(0.06)	6.5×10−4	−0.30(0.06)	2.6×10−7
rs11987089	A(0.33)	8	27390918	PTK2B	−0.19(0.05)	3.5×10−4	−0.19(0.06)	1.1×10−3	−0.28(0.06)	8.4×10−7
rs2059970	A(0.34)	8	27401909	PTK2B	−0.19(0.05)	3.1×10−4	−0.20(0.06)	5.3×10−4	−0.29(0.06)	4.3×10−7
rs7000615	C(0.36)	8	27405916	PTK2B	−0.24(0.05)	5.0×10−6	−0.24(0.06)	4.3×10−5	−0.30(0.06)	1.7×10−7
rs11775958	C(0.35)	8	27412367	PTK2B	−0.22(0.05)	4.5×10−5	−0.24(0.06)	7.4×10−5	−0.29(0.06)	7.4×10−7
rs891391	A(0.34)	8	27417992	PTK2B	−0.22(0.05)	2.5×10−5	−0.25(0.06)	3.1×10−5	−0.29(0.06)	4.0×10−7
rs1541845	A(0.36)	8	27421883	PTK2B	−0.23(0.05)	1.9×10−5	−0.25(0.06)	2.3×10−5	−0.30(0.06)	3.2×10−7
rs2241652	G(0.36)	8	27422188	PTK2B	−0.23(0.05)	1.7×10−5	−0.25(0.06)	3.6×10−5	−0.29(0.06)	4.4×10−7
rs1055256	A(0.39)	10	124758023	METTL10	0.18(0.05)	5.1×10−4	0.28(0.06)	6.5×10−7	0.20(0.06)	3.8×10−4
rs10901818	G(0.39)	10	124760859	METTL10	0.18(0.05)	5.1×10−4	0.28(0.06)	6.1×10−7	0.20(0.06)	3.7×10−4
rs11245366	G(0.38)	10	124794280		0.20(0.05)	1.6×10−4	0.29(0.06)	3.4×10−7	0.22(0.06)	5.8×10−5
rs10901823	G(0.38)	10	124800091	ABRAXAS2	0.18(0.05)	3.2×10−4	0.28(0.06)	9.2×10−7	0.22(0.06)	7.2×10−5
rs2303611	G(0.38)	10	124829420	ABRAXAS2	0.18(0.05)	3.2×10−4	0.28(0.06)	9.2×10−7	0.22(0.06)	7.2×10−5
rs17112849	A(0.01)	14	106513959		−1.33(0.30)	7.8×10−6	−1.73(0.33)	2.1×10−7	−1.60(0.32)	6.3×10−7

SNPs with statistically significant (p<1×10⁻⁷) or suggestive results (p<1×10⁻⁶) for any of the three primary sites are shown. The models were adjusted for age, age2, sex, age*sex, age2*sex, kinship, BMI Z-score and puberty stages, with statistically significant results depicted in bold.

Abbreviations: PTK2B, protein tyrosine kinase 2 beta; STMN4, stathmin 4; LRIF1, Ligand Dependent Nuclear Receptor Interacting Factor 1; MEGF10, multiple EGF-like-domains 10; ABRAXAS2, abraxas 2, BRISC complex subunit.

Figure 2.

Genome-wide association of BMC and BMD on multiple skeletal sites (only significant results are shown). A. Manhattan Plot of the genome wide association analysis of right arm BMC and B. Manhattan Plot of the genome wide association analysis of pelvis BMD. The analyses were adjusted for age, sex, age*sex, age²*sex, kinship, puberty stage and BMI Z-score.

Figure 3.

Association Plots. A. SNP association plot for right arm BMC-associated region at chromosome 22q12.3. B. SNP association plot for pelvis BMD-associated region at chromosome 8p21.2. Genetic coordinates are according to hg19/1000 Nov 2014 Genomes AMR.

As GWASs are usually used to identify common variants, exome-wide association analysis was also used to examine exonic variants, which include low-frequency variants. After Bonferroni correction, the P value significance level was 9.2 × 10⁻⁷. The exome-wide association results for primary and secondary skeletal sites are shown in Table 4 and Supplemental Table 2, respectively. We observed three novel statistically significant associations including one synonymous variant rs2303611 in the gene encoding abraxas 2 BRISC complex subunit (ABRAXAS2) with lumbar spine BMC (Table 4) and two missense variants rs17164935 (P = 5.8×10⁻⁷) and rs3812054 (P = 8.9×10⁻⁷) near gene encoding ‘multiple EGF like domains 10’ (MEGF10) associated with left arm BMC (Supplemental Table 2).

Table 4.

Exome sequencing association results for BMD and BMC

SNP	Effect Allele (Freq)	Chr	Position	Gene	SNP type	Total body		Lumbar Spine		Pelvis
						Beta (SE)	P value	Beta (SE)	P value	Beta (SE)	P value
Bone Mineral Density
rs370055571	A(0.01)	2	39485723	MAP4K3	missense	−1.19 (0.26)	5.3×10−6	−7.29 (0.26)	4.8×10−2	−0.89 (0.24)	1.7×10−4
rs532648	C(0.35)	17	26864302	FOXN1	missense	−0.22 (0.05)	8.0×10−5	−0.18 (0.05)	7.5×10−4	−0.24 (0.05)	9.7×10−7
Bone Mineral Content
rs56287545	A(0.02)	3	45989461	CXCR6	3‘ UTR	0.45 (0.15)	2.0×10−3	0.78 (0.17)	4.5×10−6	0.51 (0.16)	1.4×10−3
rs41289620	T(0.02)	3	46003735	FYCO1	missense	0.46 (0.15)	2.2×10−3	0.80 (0.18)	4.4×10−6	0.52 (0.17)	1.6×10−3
rs3812054	A(0.06)	5	126732427	MEGF10	missense	0.46 (0.10)	4.2×10−6	0.32 (0.11)	4.8×10−3	0.44 (0.11)	6.4×10−5
rs17164935	A(0.07)	5	126791282	MEGF10	missense	0.43 (0.09)	2.1×10−6	0.32 (0.10)	2.6×10−3	0.43 (0.10)	1.6×10−5
rs2303611	G(0.38)	10	126517989	ABRAXAS2	synonymous	−0.17 (0.05)	6.9×10−4	0.28 (0.06)	7.7×10−7	−0.23 (0.05)	2.5×10−5
rs151179905	T(0.01)	15	28517394	HERC2	synonymous	0.90 (0.20)	7.3×10−6	0.66 (0.23)	4.6×10−3	0.79 (0.22)	3.2×10−4

SNP with statistically significant (p<9.2×10⁻⁷⁾ or suggestive results (p<9.2×10⁻⁶) on any of the three primary sites were shown. The model adjusted for age, age², sex, age*sex, age²*sex, kinship, BMI Z-score and puberty stages, with statistically significant results depicted in bold.

Abbreviations: MAP4K3, Mitogen-Activated Protein Kinase Kinase Kinase Kinase 3; FOXN1, Forkhead Box N1; CXCR6, C-X-C Motif Chemokine Receptor 6; MEGF10, multiple EGF-like-domains 10; ABRAXAS2, abraxas 2, BRISC complex subunit; HERC2, HECT and RLD Domain Containing E3 Ubiquitin Protein Ligase 2.

To investigate whether SNPs identified in VFS were associated with any skeletal sites BMD or BMC in other studies, we performed SNP look-ups in one pediatric cohort BMDCS and a Hispanic adults cohort SAFOS (Table 5). Two SNPs (rs17164935 and rs3812054) were not available in BMDCS and rs762920 was not available in SAFOS. Results of SNPs that were available in both cohorts (rs7000615 and rs2303611) are shown in Table 7. In SAFOS, rs7000615 was nominally associated with Ward’s triangle BMD (P = 0.049) and Ward’s triangle BMC (P = 0.045) and the directions were consistent with the initial observations. No evidence of suggestive or significant evidence of association was found in BMDCS.

Table 5.

BMDCS and SAFOS lookups for novel loci identified in VFS

Chr	SNP	Position	Nearest Gene	Trait	MAF	SAFOS	P value	MAF	BMDCS	P value
						Beta(SE)			Beta(SE)
8	rs7000615	27405916	PTK2B	Total Hip BMC	0.33	−0.45(0.45)	0.31	-	-	-
8	rs7000615	27405916	PTK2B	Total Hip BMD	0.33	−0.01(0.01)	0.18	0.16	0.05(0.05)	0.33
8	rs7000615	27405916	PTK2B	Interchrochanter BMC	0.33	−0.19(0.33)	0.57	-	-	-
8	rs7000615	27405916	PTK2B	Interchrochanter BMD	0.33	−0.02(0.01)	0.11	-	-	-
8	rs7000615	27405916	PTK2B	Femoral Neck BMC	0.33	−0.03(0.05)	0.46	-	-	-
8	rs7000615	27405916	PTK2B	Femoral Neck BMD	0.33	−0.01(0.01)	0.31	0.16	0.08(0.05)	0.11
8	rs7000615	27405916	PTK2B	Spine BMC	0.33	−0.31(0.79)	0.69	-	-	-
8	rs7000615	27405916	PTK2B	Spine BMD	0.33	−0.01(0.01)	0.26	0.16	0.02(0.05)	0.60
8	rs7000615	27405916	PTK2B	Trochanter BMC	0.33	−0.2(0.1)	0.05	-	-	-
8	rs7000615	27405916	PTK2B	Trochanter BMD	0.33	−0.01(0.01)	0.22	-	-	-
8	rs7000615	27405916	PTK2B	Ward’s Triangle BMC	0.33	−0.02(0.01)	0.045	-	-	-
8	rs7000615	27405916	PTK2B	Ward’s Triangle BMD	0.33	−0.02(0.01)	0.049	-	-	-
8	rs7000615	27405916	PTK2B	Radius BMD	-	-	-	0.16	0.06(0.05)	0.19
8	rs7000615	27405916	PTK2B	Total body less head BMC	-	-	-	0.16	0.05(0.04)	0.15
10	rs2303611	126517989	ABRAXAS2	Total Hip BMC	0.44	−0.5(0.44)	0.25	-	-	-
10	rs2303611	126517989	ABRAXAS2	Total Hip BMD	0.44	−0.01(0.01)	0.33	0.43	0.05(0.04)	0.22
10	rs2303611	126517989	ABRAXAS2	Interchrochanter BMC	0.44	−0.41(0.32)	0.21	-	-	-
10	rs2303611	126517989	ABRAXAS2	Interchrochanter BMD	0.44	−0.01(0.01)	0.42	-	-	-
10	rs2303611	126517989	ABRAXAS2	Femoral Neck BMC	0.44	−0.04(0.05)	0.43	-	-	-
10	rs2303611	126517989	ABRAXAS2	Femoral Neck BMD	0.44	−0.01(0.01)	0.31	0.43	0.04(0.04)	0.29
10	rs2303611	126517989	ABRAXAS2	Spine BMC	0.44	−0.68(0.77)	0.38	-	-	-
10	rs2303611	126517989	ABRAXAS2	Spine BMD	0.44	−0.003(0.01)	0.74	0.43	0.03(0.04)	0.45
10	rs2303611	126517989	ABRAXAS2	Trochanter BMC	0.44	−0.06(0.1)	0.53	-	-	-
10	rs2303611	126517989	ABRAXAS2	Trochanter BMD	0.44	−0.01(0.01)	0.37	-	-	-
10	rs2303611	126517989	ABRAXAS2	Ward’s Triangle BMC	0.44	−0.001(0.01)	0.92	-	-	-
10	rs2303611	126517989	ABRAXAS2	Ward’s Triangle BMD	0.44	−0.001(0.01)	0.94	-	-	-
10	rs2303611	126517989	ABRAXAS2	Radius BMD	-	-	-	0.43	−0.03(0.04)	0.47
10	rs2303611	126517989	ABRAXAS2	Total body less head BMC	-	-	-	0.43	0.05(0.03)	0.062

BMDCS, Bone Mineral Density in Childhood Study; SAFOS, San Antonio Family Osteoporosis Study; VFS, the Viva La Familia Study; BMC, Bone mineral content; BMD, Bone mineral density; PTK2B, protein tyrosine kinase 2 beta; ABRAXAS2, abraxas 2 BRISC complex subunit

With further investigations in the GTEx and eQTL catalog, SNP rs7000615 at PTK2B was found to be cis-regulated in testis tissue in GTEx and eQTL Catalog, while no eQTL information was available for rs762920 at PVALB. In GTEx, SNP rs2303611 at ABRAXAS2, rs17164935 and rs3812054 was found to be cis-regulated in multiple tissues including muscle-skeletal tissue, but no significant evidence was found for cis or trans-regulation for these three SNPs in eQTL Catalog.

Discussion

Our genome-wide linkage and association studies identified several novel loci for BMD and BMC, with stronger evidence on rs7000615 at PTK2B. Our findings support the site-specific genetic effects on BMD and BMC. Significant linkages for BMD and BMC were found at 7p, 20q, 6p and 14q for different skeletal sites. Further examination of SNP association under the linkage peak yielded a significant variant, rs6018245 at EYA2, after Bonferroni correction. In genome-wide and exome-wide association analysis, we found five novel loci: rs762920 (PVALB) with right arm BMC and rs7000615 (PTK2B) with pelvis BMD, rs17164935 (MEGF10) and rs3812054 (MEGF10) with left arm BMC, and rs2303611 (ABRAXAS2) with lumbar spine BMC. Moreover, rs7000615 (PTK2B) was found to be suggestively associated with BMD in SAFOS with directional consistency, and to be cis-regulated in both GTEx and eQTL Catalog, indicating the potential functional roles of this locus. The exonic SNPs rs17164935, rs3812054 and rs2303611 were also cis-regulated in multiple tissues based on GTEx.

The significant linkage regions identified in our study were previously reported in literature, which increases confidence in our findings. The highest linkage peak observed on 7p14.2–7p14.1 has been previously linked to lumbar spine and femoral neck BMD in Caucasians [37,38], and total hip BMD in Chinese populations [39]. The chromosome 20q12-q13 region was also linked to BMD and bone geometry in Europeans [38,40,41]. The chromosome 14q22-q23 and 6p21–6p12, were linked to femoral neck BMD in younger Amish populations [42,43] and lumbar spine BMD in Caucasians [10,44]. The linkage regions were identified in various populations, suggesting these broad regions harbor important genes related to bone health across races. However, as many bone-related genes could be located in the same linkage region, the specific genes or genetic variants on identified linkage regions could still be race-specific, which could be identified in smaller regions by association analysis. In further association analysis under the linkage peaks, SNP rs6018245 at the EYA2 locus on chromosome 20 was significantly associated with pelvic BMC and the linkage signal was not significant after adjusting for rs6018245 in conditional linkage, suggesting EYA2 is the functional gene. The EYA2 locus was associated with heel BMD in Europeans of UK Biobank but the reported SNPs were not available in our data [45,46]. It is possible that there is a different tag or locus heterogeneity as we investigated in Hispanic population, rather than Europeans.

In genome-wide and exome-wide association analysis, four novel genes, PTK2B, PVALB, MEGF10 and ABRAXAS2, were identified in association with BMD or BMC. The SNP rs7000615 (PTK2B) was genome-wide significantly associated with pelvis BMD in Hispanic children of VFS and suggestively associated with BMC and BMD in Hispanic adults of SAFOS, but not in Caucasian/African Americans of BMDCS. It is possible that this SNP is ethnic-specific since VFS and SAFOS are both primarily Hispanic/Mexican American population but more studies are needed to confirm this possibility. There was also evidence in GTEx and eQTL catalog shown that rs7000615 was cis-regulated, suggesting the functional role of this locus. Biologically, PTK2B has been found to have a role in osteoclast maturation and bone resorption in vitro, and PTK2B/PYK2 inhibitors might be used to treat osteoporosis based on findings from mice studies [47]. PVALB and MEFG10 are also biologically relevant to bone metabolism. PVALB is a high affinity calcium ion-binding protein and one study in Pvalb knockout mice found that the lack of parvalbumin is associated with stronger bones and abnormal calcium handling [48] and MEGF10 is a Runx2-responsive genes that is related to osteoblast proliferation [49]. ABRAXAS2 is involved in processes related to protein metabolism, deubiquitination pathways, and myocardial infarction [50], but the mechanism underlying its relationship with bone health is still unknown. The identified genes in GWAS and exome-wide association analysis were not overlapped mainly due to the fact of variants sequenced are different. The majority of GWAS identified SNPs are intronic, while exome-wide analysis could complement GWAS results with exonic variants [51].

Our study has a few limitations. First, the sample size is small and may suffer from insufficient power and a high false discovery rate, which are potential problems for genome-wide scans [52]. We also had limited ability to detect rare variants (MAF < 0.01) due to small sample size. Second, the identified SNPs may not be causal and could be in LD with the underlying causal variants and we also do not provide definitive biological mechanisms of the association found in our study. Future studies on potential functional SNPs with gene expression and translational studies are needed to elucidate the findings. Third, DXA-derived measurement may poorly capture the effects of genetic variants on bone structure and geometry, which are also important aspects for bone health [53]. Moreover, there could be residual confounding that is not accounted for, therefore future studies on mechanisms and function are needed to confirm the findings. Finally, obesity can impact BMD and BMC, especially as our cohort was made up of about two-thirds of children who were overweight/obese, which may limit our generalizability [54]. However, about one-third of children were normal-weight and the analysis was adjusted for BMI Z-score to account for the effects of obesity on bone outcomes.

Our study also has several strengths. This study examining genetic variation underlying BMD and BMC is novel and fills the knowledge gaps in three prospective. First, the study was conducted in Hispanic population, which is understudied compared to Europeans and provides opportunity to identify novel variants as well as support for early identified variants[55]. Second, the study was conducted in children with bone accrual playing a main role and less environmental influence accumulated compared to adults, and a few novel loci related to BMD were only identified in pediatric GWAS studies. The discovery of variants associated with peak bone mass might be used to predict osteoporosis risk as adult. Third, as site specificity of genetics on BMD were found [56–58], multiple skeletal sites were examined to identify novel loci and better understand the genetic heterogeneity of different effects on BMD or BMC. Besides novelty, another strength is that we used a combination of genome-wide linkage and association (genome-wide and exome-wide) analyses approaches. Linkage studies with family data can find larger genomic regions while association analysis could narrow down to smaller regions. Also, linkage in families may be able to detect rare variants, while genome-wide and exome-wide association could identify common and low frequency variants. Finally, we use both BMD and BMC to assess pediatric bone health, where areal BMD values may better reflect bone matrix mineralization and BMC may better capture volumetric bone mass [59].

In summary, our study is the first genome-wide linkage scan, with GWAS and exome-wide analysis to examine underlying genetic loci associated with BMD and BMC in Hispanic children. Our findings provide insights into the biological mechanism underlying BMD and BMC in Hispanic children. The genetic loci identified may be used to predict osteoporosis risk in adults and contribute to identification of new drug targets for osteoporosis prevention and treatment. Further functional studies and replication studies are warranted to elucidate the precise biological mechanisms and to validate our findings.

HIGHLIGHTS.

1. Osteoporosis and low bone mass are a major public health threat for US

2. The correlates of bone health in children are understudied.

3. We aimed to identify genetic variants associated with bone health in Hispanic children utilizing genome-wide linkage, genome-wide association and exome-based analysis approaches.

4. We identified novel loci associated with BMC and BMD in Hispanic children, with strongest evidence for PTK2B