Introduction

In the last decade, numerous genome wide association studies (GWAS) have been conducted as a strategy to identify new genes and variants associated with complex diseases. In the field of osteoporosis, a number of GWAS have been performed to determine the association with bone mineral density (BMD) and osteoporotic fracture. BMD is a complex quantitative trait that is mostly genetically determined, therefore it is a good biological marker for the determination of bone status. Initial GWAS studies1,2,3,4,5,6 have identified more than 50 loci associated with BMD, among which the Wnt/βcatenin signaling pathway was especially enriched. Specifically, in the study by Estrada et al.6, 14 BMD-associated loci were also found to be associated with fracture risk, and 6 of them belonged to the Wnt pathway: LRP5, SOST, DKK1, WNT4, WNT16 and CTNNB1. More recently a low-frequency variant with large effects on BMD has been described in the WNT16 gene in a whole genome sequencing study7. In fact, several monogenic bone phenotypes are caused by rare mutations in genes of the Wnt pathway. For example, sclerosteosis8,9,10 and van Buchem11,12 – two monogenic bone overgrowth diseases- are caused by rare loss of function mutations or the deletion of an enhancer element (ECR5) of SOST, and the osteoporosis-pseudoglioma syndrome is caused by rare loss of function mutations of LRP513.

The Wnt pathway regulates a wide variety of cellular processes during embryogenesis and in adult regenerative tissues, as is the case for bone. Wnt proteins can activate up to 3 signaling pathways: 2 non-canonical and one canonical pathway or β-catenin-dependent pathway. When this signaling pathway is activated in osteoblasts, a series of important genes for the differentiation and formation of bone are induced14. The canonical pathway is finely regulated by extracellular inhibitors, including DKK1 and sclerostin, encoded by DKK1 and SOST, respectively. Sclerostin is produced mainly in osteocytes, although it is also expressed in other cell types differentiated from the mineralized matrix, such as cementocytes or chondrocytes15.

Despite the large number of loci associated with BMD, they account for only a small part of the genetic impact on the femoral neck BMD (<6%)6. Thus, the variability of BMD cannot be explained solely by tag SNPs, which is why a search for new polymorphisms and rare variants is necessary. Our objective has been to identify variants within 3 genes of this pathway: WNT16, DKK1 and SOST. We have approached the characterization following a two-step process. First, we have resequenced the coding regions of these genes in two extreme BMD groups of postmenopausal women of the BARCOS cohort to look for rare variants responsible for the phenotype, or common variants associated with BMD. Secondly, we analyzed the association of the common and low frequency variants with BMD, in the full cohort, and we explored the distribution of the rare variants.

Results

Search for Variants in a Truncated Selection of Samples from the BARCOS Cohort and Analysis of their Frequencies in the Extreme Groups

We resequenced WNT16, DKK1 and SOST coding exons, intronic flanks, UTRs and regulatory regions, in 108 postmenopausal women in the extremes of the BMD distribution of the BARCOS cohort (55 with HBM and 53 with LBM, see Methods). A total of 19, 14 and 18 single nucleotide variants (SNVs) were identified in WNT16, DKK1 and SOST, respectively (Tables 1–3 and Fig. 1), of which only the DKK1 variant c.406 + 195 G > A was novel. Fifteen of them have frequencies above 5% (common variants), 12 have frequencies between 1–5% (low frequency variants, LFV), and the 24 remaining have frequencies below 1% (rare variants), according to EUR population in 1000 genomes (sequencing chromatograms of the 24 rare variants are shown in Supplementary Figures S1–3). Only the DKK1 LFV rs74711339 showed significant differences between the genotype frequencies of the two extreme groups (Fisher’s exact test, p = 0.0224). This SNP is located in the 3′UTR and the minor allele (G) was found overrepresented in the LBM group (Table 2).

Table 1 Variants found in the resequencing of WNT16.
Full size table
Table 2 Variants found in the resequencing of DKK1.
Full size table
Table 3 Variants found in the resequencing of SOST.
Full size table
Figure 1
figure 1

Variants found in WNT16 (a), DKK1 (b) and SOST (c). Horizontal lines below the genes indicate amplicons used to resequence the exons and flanking intronic regions (not to scale). #Rare variants (MAF < 0.01). ■Novel variant.

Full size image

In silico Analysis of Functionality of the Identified Variants

To better assess the importance of the identified variants we have explored their possible functionality using in silico tools and databases.

Four of the variants were missense, of which 2 common SNPs in WNT16 (rs2908004; p.Gly72Arg/p.Gly82Arg and rs2707466; p.Thr253Ile/p.Thr263Ile) and one LFV in SOST (rs17882143; p.Val10Ile) were predicted to be tolerated and benign by SIFT and PolyPhen-2, respectively. The other missense change, the DKK1 rare variant rs149268042 (p.Arg120Leu), was predicted to be harmful (deleterious by SIFT, probably damaging by PolyPhen-2, and disease causing by Mutation Taster).

Fifteen other variants might play some regulatory role. One of them, the rs55710688 SNP, located in the 5′UTR of WNT16a, is an insertion of 4 nt (CCCA) that affects the Kozak consensus sequence, and might alter translation initiation rate. The DKK1 intron 3 SNP rs1569198 was predicted to affect splicing, where the G allele might generate a new splice acceptor site 42 nt upstream of the original one, although with a lower score. One rare variant in SOST, rs570754792, located 23 bp upstream the transcription start site, lies within the extended TATA box motif. The change is a C (most probable in the consensus TATA box sequence) for a T (least probable). Five 3′UTR variants were predicted to have an effect on the binding of 13 miRNAs (Supplementary Table S1): 3 SNPs in SOST (rs17883310, rs17886183, rs75901553), 1 SNP in WNT16 (rs17143305) and 1 rare variant in WNT16b (rs190011371), not present in the other mature transcript of this gene. Several variants were shown to be eQTLs of WNT16, DKK1 or SOST, according to GTEx. The WNT16 SNP rs55710688 that affects the Kozak sequence mentioned above, was also described as an eQTL of WNT16 in adipose tissue. Regarding DKK1, two SNPs (rs74711339 and rs41281546) are eQTLs in transformed fibroblasts. For the SOST gene, we have found 3 SNPs (rs1237278, rs851058 and rs10534024) that act as eQTLs in more than one tissue. This information together with additional eQTL information relative to other genes, is shown in Supplementary Table S2. Finally, the WNT16 SNP rs142005327 (a 2-bp intronic insertion) and the rare variant rs113001389 lie in a potential regulatory site in osteoblasts, according to ENCODE data (and rs142005327 is described as an eQTL of CPED1 and FAM3C).

Association of Common and Low Frequency Variants with BMD in the Complete Cohort

In total, six common variants and four low frequency variants were genotyped in the complete BARCOS cohort (n = 1490) to test their association with BMD (Tables 1–3, in bold). For all of them, we obtained MAF values that are similar to those found in 1000 genomes for the European population (Supplementary Table S3). Nominal significant differences were obtained for lumbar spine BMD (LS-BMD) with the WNT16 SNPs rs55710688, rs142005327, rs2908004 and rs2707466, under an additive model, and for the SOST SNP rs17882143, under a dominant model (Table 4, italics). In all cases, the minor allele had a protective effect on LS-BMD (Supplementary Figure S4). Moreover, nominal significant differences were obtained in the comparison of the femoral neck BMD (FN-BMD) with the WNT16 SNPs rs55710688, rs142005327 and rs2707466 and for the DKK1 SNP rs1569198, under an additive model. Only the WNT16 SNPs rs55710688 and rs142005327 remained significant after Bonferroni’s correction (p < 0.0038) (Table 4, bold). No significant association was found between the genotyped variants and fractures. The SOST SNP rs17882143 presented the largest effect size (β = 0.041) of all associated SNPs described (Supplementary Table S4). In our cohort, the associated SNPs in WNT16 and SOST were found to be in linkage disequilibrium with the respective GWA hits (rs3801387, WNT16; rs4792909, SOST) of Estrada et al.6 (Fig. 2a,c), while the DKK1 SNP rs1569198 was not (Fig. 2b).

Table 4 Association between LS-BMD and FN-BMD and genotypes of common and LFV WNT16, DKK1 and SOST variants.
Full size table
Figure 2
figure 2

Haploview linkage disequilibrium plots of variants of WNT16 (a), DKK1 (b) and SOST (c) genotyped in the BARCOS cohort. *GWAS hits from Estrada et al.6. The numbers within the squares and the color scale both refer to D’/LOD values (with bright red: D’ = 1 and LOD ≥ 2; white: D’ < 1 and LOD < 2; blue: D’ = 1 and LOD < 2 and shades of pink/red: D’ < 1 and LOD ≥ 2).

Full size image

Haplotype analyses of WNT16 associated SNPs

We next tested whether haplotypes might display a stronger association with BMD than any given WNT16 SNP alone. We built haplotypes including the 4 associated SNPs and the WNT16 tag-SNP from Estrada et al.6 in the complete BARCOS cohort and could detect 4 different haplotypes with frequencies above 2% (Table 5). The two most frequent ones consisted on the combination of all the major alleles or all the minor alleles, as expected, given the strong LD present in the region. We tested association between each of the 4 major haplotypes and LS-BMD and found that only the two most frequent (DGDAC and IAIGT) were nominally significantly associated (Table 5), although they did not reach significance after Bonferroni’s correction. DGDAC had a BMD-lowering effect, opposite to IAIGT, which was protective. These effects are in agreement with those of the individual SNPs. We could not find an association of any haplotype with FN-BMD (data not shown).

Table 5 Haplotype association with LS-BMD.
Full size table

We also tested for significant frequency differences of the 4 major haplotypes between the HBM and LBM groups of the truncate selection, and found none. We only observed a slight overrepresentation of DGDAC in LBM and a slight overrepresentation of IAIGT in the HBM group, as expected (Table 6). Finally, DGDAC was found to be significantly overrepresented in the lower half of the LS-BMD distribution in the complete BARCOS cohort (Table 6).

Table 6 Haplotype distribution in extreme BMD groups, and in the two halves of the complete BARCOS cohort according to LS-BMD mean (upper half > 0.8509; lower half < 0.8509).
Full size table

Analysis of Rare and Low Frequency Variants

In the resequencing of WNT16, DKK1 and SOST in the extreme BMD groups, we found 24 rare and 12 low frequency variants. In WNT16, the minor alleles of these variants were distributed between 23 HBM and 22 LBM women. In DKK1, they were found in 8 HBM and 6 LBM women. Finally, in SOST, they were distributed in 16 HBM and 9 LBM women. However, this difference did not reach significance.

Three rare variants found in two LBM and one HBM women and located in putative regulatory elements (two in WNT16 and one in SOST), were genotyped in the complete BARCOS cohort (in italics in Tables 1 and 3, respectively). For the LBM variants (rs190011371 and rs570754792), the rare allele was found in heterozygosis in 3 women whose BMD was below the mean of the BARCOS cohort (0.737 g/cm2 and 0.7610 g/cm2 vs. 0.8509 g/cm2). For the HBM variant rs113001389 (WNT16), no other instance of it was found in the complete BARCOS cohort.

Discussion

WNT16, DKK1 and SOST have been previously associated with osteoporosis6. In this study, we have investigated variants within these genes as possible causes of the association. Specifically, we resequenced them in a truncate selection of women of the BARCOS cohort based on their Z-score. We found 19, 14 and 18 variants, respectively, 15 of which were common, 12 low frequency and 24 rare variants. Ten variants, with in-silico predicted functionality, were tested for association in the complete BARCOS cohort, and 6 of them (4 in WNT16, one in SOST and one in DKK1) gave nominal significant results with LS-BMD, FN-BMD or both. After correction for multiple testing, only two WNT16 variants (one affecting the Kozak sequence and the other a putative intronic enhancer) remained significant. Finally, 2 rare variants present in putative regulatory sites (TATA box of SOST and 3′UTR of WNT16b) were each found in only 3 women of the BARCOS cohort, all 6 with BMD values below the cohort average.

Association with BMD

Of the 4 WNT16 variants nominally associated with BMD, rs2707466 and rs2908004 were missense. These results agree with those published by others in different cohorts using several bone parameters16,17,18,19,20,21. Although these missense variants were predicted to be tolerated and benign by SIFT and PolyPhen-2, respectively, rs2707466 is a change of a threonine to an isoleucine that directly abolishes a phosphorylation site, which causes a deleterious effect on the WNT16 protein18, while rs2908004 is a change from a glycine to an arginine, two amino acids with very different biochemical properties. These functional changes might somehow contribute to the association.

The other two WNT16 variants nominally associated with BMD remained so after multiple-testing correction and thus, deserve further attention. One, rs55710688, modifies the Kozak sequence of WNT16a, and has been shown to produce changes in the translation of this transcript21. The transcription of WNT16 generates two RNA isoforms, WNT16a and WNT16b, which differ in the first exon and in the 3′UTR region and are independently controlled by two alternative promoters22. The expression of the WNT16a is restricted to the pancreas, whereas WNT16b is expressed in multiple organs22, although bone expression data are not available. It would therefore be interesting to know which isoform is being expressed in bone. In the work of Hendrickx et al.21, they carried out in vitro translation studies and showed a higher translation efficiency for the minor allele (insCCCA). Also, this allele was found over-represented in the HBM group, similar to our results. Interestingly, the rs55710688 variant is listed as an eQTL for WNT16 (subcutaneous adipose tissue) and also for FAM3C (skin). The other variant, rs142005327, is an insertion of 2 nucleotides in intron 2, which lies in a potential regulatory site in osteoblasts, and is an eQTL of FAM3C (skin and tibial nerve) and of CPED1 (tibial artery). Consistent with our results, Hendrickx et al.21 found differences between genotype frequencies of this variant in two extreme BMD groups, where the minor allele (insCT) was over-represented in the HBM group. Overall, our results for these two variants are in the same line of those of Hendrickx et al.21, and we describe, for the first time, a significant association in a full cohort, reinforcing their importance as determinants of bone mass.

The 4 variants of WNT16 described above are in high linkage disequilibrium with the GWAS hit from Estrada et al.6, rs3801387, which lies in the last intron of the gene. We note that this variant is also an eQTL of FAM3C (skin). Given its location next to WNT16, there has been a debate on the role of FAM3C in BMD determination (for example see ref.16). This gene encodes a citokine-like factor involved in epithelial to mesenchymal transition and recently has been shown to modulate osteogenic differentiation through Runx2 downregulation23. It is tempting to speculate that the biological effect of these 3 significantly associated variants of WNT16 happens in part through their effects on FAM3C expression.

The DKK1 rs1569198, found nominally associated with FN-BMD, is predicted to affect splicing. This same variant was found associated with hip axis length (HAL) by Piters et al.24, in young Caucasian men. However, Koromila et al.25 failed to find association with BMD or bone turnover markers in Greek postmenopausal women. This SNP is not in LD with the GWA hit (rs1373004, Estrada et al.6), which lies 3.5 kb downstream of the gene, and therefore may constitute an independent signal. On the other hand, the DKK1 LFV rs74711339, that showed significant differences in the two extreme groups and is a DKK1 eQTL, showed a trend for association with FN-BMD, but did not reach significance in the full cohort, probably due to lack of power.

The only SOST variant found nominally associated with BMD was rs17882143 (p.Val10Ile), a missense change affecting the signal peptide of the sclerostin protein. This change may alter the cellular localization of the protein as shown for another signal peptide missense variant (p.Val21Leu) by Kim et al.26 To our knowledge, this is the first time that the rs17882143 SNP is found associated with BMD. It is in linkage disequilibrium with the SOST GWAS hit of Estrada et al.6, rs4792909, which, interestingly, did not show association in our cohort (see Table 4). We did not obtain association with BMD for the rest of the SOST SNPs tested, including rs2023794, which Zhang et al.27 had found associated with LS-BMD in postmenopausal Chinese women.

Interesting rare variants

The WNT16b 3′UTR rs190011371, found in one LBM woman, was genotyped in the full cohort and two other cases presented with the minor allele, both with LS-BMD values below the mean of the cohort. According to in silico predictors, only the major allele (G) would bind the miRNA hsa-miR-383. Interestingly, this miRNA was found significantly under-expressed in osteoporotic bone (our unpublished results; see Supplementary Table S1). It is tempting to speculate that the inability of this miRNA to bind to the mutated 3′UTR from WNT16b would have a similar biological outcome, and this would be in agreement with the presence of the variant in 3 women with LS-BMD below the mean. Similarly, the SOST variant rs570754792, located within the extended TATA box motif, was found in one LBM woman, and was also present in two other women of the full cohort with a BMD below the mean. It should be noted that when rare and low frequency variants of each gene were taken into account, only those of SOST presented an unbalanced distribution, with an over-representation in the HBM extreme group. If we assume that most of them might be damaging, this would agree with the loss of the inhibitory activity of sclerostin in HBM women.

Additional variants

Beside the 13 variants chosen to be genotyped in the complete BARCOS cohort due to the evidence of functionally, there are other variants which deserve discussion. Among them, we highlight the SNPs rs17143305 and rs75901553 in the 3′UTR region of the WNT16 and SOST genes, respectively. According to various functional predictors, rs17143305 may affect the binding of miRNAs hsa-miR-541 and hsa-miR-4263 and has been reported as an eQTL for genes RP11-3L10.1 and FAM3C in different tissues. Interestingly, hsa-miR-541 has been described as a negative regulator of osteoblastic differentiation28, in the only study in which it is studied in relation to bone tissues, to the best of our knowledge. On the other hand, as explained above, FAM3C was reported to have a negative effect on osteogenic differentiation23. Regarding rs75901553, different functional predictors suggest that the binding of up to 8 miRNAs might be affected, two of them being hsa-miR-98-5p and hsa-miR-let-7f. The former is known to regulate the expression of BMP2, and consequently the osteogenic differentiation of hBMSC29, while the latter was shown to be crucial for the regulation of β-catenin activity and osteogenic differentiation by TIMP-130. For these reasons, it may be interesting to further study the association of these variants with BMD, as well as their functionality.

Strengths and weaknesses

The resequencing approach in a truncate selection carried out in this study, together with deep in silico functional assessment, allowed us to describe potentially relevant common and rare variants present in extreme BMD groups in our cohort. We were able to identify variants that are influencing BMD in the general population, as well as rare variants that may explain extreme phenotypes. This has enabled us to find the rare variants variants rs190011371, rs113001389 and rs570754792, which would have gone unnoticed in association studies. We have performed a comprehensive analysis of the whole coding region, intronic flanking regions, 5′ and 3′UTR and relevant regulatory regions of three important Wnt pathway genes. On the other hand, our approach has two main limitations: the sample size of the study and the lack of analysis of most of the introns and far regulatory regions.

Final remarks

This work provides variants with indications of functionality in 3 important genes of the Wnt pathway. Specifically, 4 missense variants have been found, three of them nominally associated with BMD. However, the only variants significantly associated were two putative regulatory variants in WNT16: rs55710688 (Kozak sequence) and rs142005327 (putative enhancer). In addition, two rare variants in functional regions (WNT16 in 3′UTR and SOST in TATA box) may help determine a low BMD phenotype. Therefore, this work reinforces the importance of regulatory variants in these genes in bone function and opens new ways of research. To confirm our hypotheses, future research may include functional studies to validate the in silico predictions of the associated variants and replication of the association studies in different (and preferably larger) cohorts.

Methods

Study Cohort

The BARCOS cohort consisted of 1490 postmenopausal women from the Barcelona area, monitored at the Menopausal Unit of the Hospital del Mar (Barcelona, Spain), all of Spanish descent. BMD of all participants was measured at the lumbar spine (LS) and the femoral neck (FN) by dual energy X-ray absorptiometry (DXA). The following data were also recorded: age, age of menarche and menopause, number of fractures and anthropometric measures such as weight and height. DNA is available from all samples of the cohort. Details of the cohort and DNA extraction have been described previously31,32.

For the resequencing of the candidate genes, two extreme BMD groups were selected from the BARCOS cohort. For the election of the extreme groups the statistical Z-score was used. The 55 women with the highest Z-score scores (from 2.98 to 0.695) were included in the HBM (High Bone Mass) group and the 53 women with the lowest Z-scores (−2.40 to −4.26) were included in the LBM (Low Bone Mass) group.

Written informed consent was obtained from all patients in accordance with the regulations of the Clinical Research Ethics Committee of Parc de Salut Mar, which approved the study. Samples from patients were obtained in accordance with the Helsinki Declaration, as revised in 2000. All experiments and protocols were approved by the Bioethics Committee of Universitat de Barcelona (IRB00003099).

Re-sequencing

Resequencing of the WNT16, DKK1 and SOST genes was performed in two extreme subgroups of the BARCOS cohort according to the Z-score statistic score. The selection of fragments to be amplified from the WNT16, DKK1 and SOST genes and the design of the primers were based on the consensus sequences ENSG00000002745, ENSG00000107984 and ENSG00000167941, respectively (GRCh37.p13, ENSEMBL). WNT16 was amplified into 6 fragments; DKK1 in 5 fragments and SOST in 7 fragments. The resulting amplicons included all of the coding exons, the UTR regions, the corresponding flanking intronic regions and the two different protein coding transcripts of WNT16. Finally, a pair of primers were used for the amplification of the 252 bp of the ECR5 enhancer of SOST. Amplification of each of these fragments was done by PCR using Go Taq Flexi DNA polymerase (Promega). The temperature and magnesium conditions of each PCR, and the sequence of the primers are presented in Supplementary Table S5. The PCR fragments were analyzed by agarose gel electrophoresis and purification was done in MultiScreen TM Vacuum Manifold 96-well plates (Merck Millipore). The purified PCR products were sequenced by the Sanger method by the CCiTUB genomic service (Genómica, Parc Cientific, Barcelona, Spain). The tagging kit used is BigDye™ Terminator v3.1 Cycle Sequencing Kit, detection and electrophoresis are performed on automated capillary sequencer models 3730 Genetic Analyzer and 3730xl Genetic Analyzer.

SNV Genotyping

Genotyping of 13 selected variants in the complete BARCOS cohort was carried out at LGC Genomics (Hoddesdon, UK). In addition, as BARCOS was included in the replication phase of the study by Estrada et al.6, the genotyping results of SNPs rs3801387 (WNT16), rs1373004 (DKK1) and rs4792909 (SOST) were also available. The genotyping of 12% of the samples was performed in duplicate, as a genotyping quality control, and showed a concordance above 99%.

Linkage Disequilibrium Calculation

The Haploview software from HapMap (http://hapmap.ncbi.nlm.nih.gov/) was used to calculate and represent the degree of linkage disequilibrium between genotyped polymorphic variants33 using the default parameters.

Haplotype construction and association

Haplotypes of WNT16 associated SNPs were built using PLINK (http://pngu.mgh.harvard.edu/purcell/plink) with default parameters34. Association of haplotypes with LS- or FN-BMD was assessed with the same software. Correction for multiple testing was performed using the Bonferroni’s method. Fisher’s exact test was used to assess the WNT16 haplotype frequency differences between the two extreme groups of BMD, and between the two halves of the LS-BMD distribution of the complete BARCOS cohort.

Statistical Analysis

Statistical analyses were performed using Rstudio version 3.3.2. Hardy-Weinberg equilibrium (HWE) was calculated using Chi-square test (X2). p-values less than 0.01 were considered significant. Fisher’s exact test was used to determine the differences between the two extreme groups of BMD on the genotype frequencies of all detected variants. Linear regression analysis adjusted by years since menopause (YSM), using SNPassoc package, was used to determine the association between BMD and the genotype of each genotype variant. All analyses were performed by testing the additive, the recessive and the dominant model. p- values less than 0.05 were considered nominally significant. Correction for multiple testing was performed using the Bonferroni’s method for the number of SNPs tested (n = 13).

Bioinformatic Analysis and in silico Predictions of the Effect of the Variants

To identify and characterize all variants, we used information from the Ensembl database GRCh37.p13 (http://www.ensembl.org). Information was sought on the MAF of each of the variants found in the European population (EUR) and in the Iberian Peninsula (IBS). We used SIFT (http://sift.bii.a-star.edu.sg/), PolyPhen (http://genetics.bwh.harvard.edu/pph2/) and MutationTaster (http: //www.mutationtaster. org /) to test the effect of the missense variants. MiRNSP (bioinfo.bjmu.edu.cn); miRNA-SNP (www. bioguo.org/miRNASNP); SNP Function Prediction (https://snpinfo.niehs.nih.gov/snpinfo/snpfunc.html); MicroSNiPer (http: // epicenter.ie-freiburg.mpg.de/services/microsniper/); miRdSNP (http://mirdsnp.ccr.buffalo.edu); miRTarBase (http://mirtarbase.mbc.nctu.edu.tw/) were used to analyze the binding of miRNA to the variants in the 3′UTR region. The GTEx database (www.gtexportal.org/home/) was used to identify variants that act as eQTLs. Human Splicing Finder (umd.be) and SplicePort (http://spliceport.cbcb.umb.edu/) were used to evaluate the predicted splicing sites. To characterize regions that include the variants found, the UCSC Genome Browser GRCh37/hg19 was enriched with ENCODE data from osteoblasts.