Identification of IDUA and WNT16 Phosphorylation-Related Non-Synonymous Polymorphisms for Bone Mineral Density in Meta-Analyses of Genome-Wide Association Studies

Abstract

Protein phosphorylation regulates a wide variety of cellular processes. Thus, we hypothesize that single nucleotide polymorphisms (SNPs) that may modulate protein phosphorylation could affect osteoporosis risk. Based on a previous conventional genome-wide association (GWA) study, we conducted a three-stage meta-analysis targeting phosphorylation-related SNPs (phosSNPs) for femoral neck (FN)-, total hip (HIP)-, and Lumbar Spine (LS)-BMD phenotypes. In stage 1, 9,593 phosSNPs were meta-analyzed in 11,140 individuals of various ancestries. Genome-wide significance (GWS) and suggestive significance were defined by α = 5.21×10⁻⁶ (0.05/9,593) and 1.00×10⁻⁴, respectively. In stage 2, 9 stage 1-discovered phosSNPs (based on α = 1.00×10⁻⁴) were in silico meta-analyzed in Dutch, Korean, and Australian cohorts. In stage 3, four phosSNPs that replicated in stage 2 (based on α = 5.56×10⁻³, 0.05/9) were de novo genotyped in two independent cohorts. IDUA rs3755955 and rs6831280, and WNT16 rs2707466 were associated with BMD phenotypes in each respective stage, and in 3 stages combined, achieving GWS for both FN-BMD (P-value = 8.36×10⁻¹⁰, 5.26×10⁻¹⁰, and 3.01×10⁻¹⁰, respectively) and HIP-BMD (P-value = 3.26×10⁻⁶, 1.97×10⁻⁶, and 1.63×10⁻¹², respectively). Although in vitro studies demonstrated no differences in expressions of wild-type and mutant forms of IDUA and WNT16B proteins, in silico analysis predicts that WNT16 rs2707466 directly abolishes a phosphorylation site, which could cause a deleterious effect on WNT16 protein, and that IDUA phosSNPs rs3755955 and rs6831280 could exert indirect effects on nearby phosphorylation sites. Further studies will be required to determine the detailed and specific molecular effects of these BMD-associated non-synonymous variants.

INTRODUCTION

Osteoporosis, a complex disease characterized by reduced bone mass, results in microarchitectural deterioration of bone tissue, and increased bone fragility and susceptibility to fracture.⁽¹⁾ It has been estimated that the prevalence of osteoporosis in the United States will increase to >14 million people in 2020,⁽²⁾ and by 2025 it is projected that there will be >3 million fractures/year in the U.S., costing $25.3 billion annually.⁽³⁾ A diagnosis of osteoporosis for both males and females is attained when bone mineral density (BMD) is 2.5 standard deviations or more below the young adult mean.⁽⁴⁾ BMD, a highly heritable polygenic trait, is the best predictor for skeletal fragility.⁽⁵⁾

Protein phosphorylation represents the most widespread post-translational modification (PTM), which plays a critical role in essential cellular processes, e.g., metabolism, cell signaling, differentiation, and membrane transportation.⁽⁶⁾ Large-scale phosphoproteomics studies suggest that more than half of all eukaryotic proteins are phosphorylated.⁽⁷⁾ The most common phosphorylation sites in eukaryotes are serine (S), threonine (T), and tyrosine (Y) residues,⁽⁸⁾ which are catalyzed by S/T-specific, Y-specific, and dual-specificity protein kinases.⁽⁹⁾ Single nucleotide polymorphisms (SNPs) constitute almost 90% of genetic variations in the human genome.⁽¹⁰⁾ Non-synonymous SNPs (nsSNPs), defined as SNPs resulting in amino acid changes that include either missense or nonsense mutations,⁽¹¹⁾ represent 60% of known disease-causing mutations.⁽¹²⁾ Of them, those that create/alter/abolish phosphorylation sites, called phosphorylation-related SNPs (phosSNPs), have been recognized as functional variants for a spectrum of human diseases, e.g., lung cancer (CSF1R rs10079250),⁽¹³⁾ prostate cancer (TP53 rs1042522),^(13,14) long QT syndrome (KCNH2 rs1805123),^(15,16) systemic lupus erythematosus (VEGR2 rs2305948),⁽¹⁷⁾ and tuberculosis (TLR2 rs5743708).^(18,19) Each phosphorylation site consists of an acceptor residue surrounded by an evolutionarily conserved motif consisting of 7–12 amino acid residues on either flanking region. Based on the hypothesis that a sequence motif surrounding an acceptor residue represents a main determinant of protein kinase specificity, phosphorylation sites can be predicted in silico, and nsSNPs affecting such sites can be identified. From 91,797 nsSNPs of NCBI dbSNP Build 130, by applying the Group-based Phosphorylation Scoring (GPS) 2.0 program (a kinase-specific phosphorylation site predictor)⁽²⁰⁾, Ren et al. (2010)⁽²¹⁾ identified 64,035 phosSNPs residing in 17,614 human proteins, which were categorized into five distinct types: I, II, III, IV and V based on the different effects they exert on phosphorylation sites.

Among at least 60 loci identified by > 40 previous genome-wide association (GWA) studies and meta-analyses of these studies for osteoporosis, WNT16 locus has been found to be an important genetic determinant of osteoporosis risk.⁽²²⁾ The human WNT16 gene spans ~16 kb from initiation to termination codons, encoding two protein isoforms — WNT16A (40.5 kD) and WNT16B (40.7 kD).⁽²³⁾ As depicted in Supplementary Figure S1, these two WNT16 isoforms have different first exons (i.e., 1a and 1b, respectively), independently controlled by two alternative promoters P1 and P2, respectively.⁽²⁴⁾ Expression of the WNT16A isoform has been shown to be restricted to the pancreas in humans, whereas WNT16B is expressed in multiple organs.⁽²⁴⁾ Compared to WNT16A, the role of WNT16B as a key regulator of osteoclastogenesis has been more extensively characterized.⁽²⁵⁾

Meta-analyses of GWA studies have significant potentials for detecting subtle genetic effects.⁽²⁶⁾ However, because conventional GWA studies often include a large number of variants of unknown functional effects, the significance threshold attained by Bonferroni correction becomes overly conservative, producing a high rate of type II error (i.e., β). PhosSNPs are more likely disruptive to protein function than other protein-coding missense mutations.⁽²⁷⁾ However, such potentially causal missense mutations could be missed by conventional GWA approaches, due to very strict control for type I error (i.e., α). Power to detect disease-causing variants can thus be increased by focusing exclusively on SNPs with higher prior probabilities of functional effects, either as in a whole-exome sequencing⁽²⁸⁾ approach targeting solely exonic SNPs, or as we apply here, targeting exclusively potentially functional phosSNPs. However, such a functional candidate genomic region approach⁽²⁹⁾ could be susceptible to a higher rate of false positive results.⁽³⁰⁾ Therefore, to guard against an inflated α, we employed a three-stage approach, such that those phosSNPs attaining genome-wide significance (GWS) in stage 1 (i.e., GWA discovery) are required to be replicated in independent cohorts of stages 2 and 3, respectively, based on their corresponding Bonferroni-corrected α thresholds.

MATERIALS AND METHODS

A detailed description of study participants, phenotype measurement and modeling, DNA genotyping, quality control (QC), and genotype imputation, association tests, meta-analysis methods, and regional association plots of the three-stage GWA meta-analysis is given in Supplementary Materials and Methods. At stage 1, seven GWA cohorts were included, and a suggestive significance threshold of α = 1.00×10⁻⁴ was applied for phosSNP selection. At stage 2 (in silico replication), three GWA cohorts were included, and at stage 3 (de novo genotyping replication), two independent cohorts were included, and at each stage, a Bonferroni-corrected significance threshold was applied.

PhosSNPs in Potential Phosphorylation Sites

The phosSNP-centric GWA meta-analysis focuses exclusively on 9,593 phosSNPs in stage 1 of the conventional GWA meta-analysis.⁽³¹⁾ Details about phosSNP selection are given in the Supplementary Materials and Methods.

In Silico Bioinformatics Analyses

(1) Computational Predictions of Phosphorylation Sites Affected by phosSNPs

Phosphorylation sites that could be affected by the three significant phosSNPs — IDUA rs3755955 (R105Q) and rs6831280 (A361T), and WNT16 rs2707466 (WNT16B T263I), were predicted by two commonly used online software programs: NetPhos2.0⁽³²⁾ and NetPhosK1.0.⁽³³⁾ Details about these programs are given in the Supplementary Materials and Methods.

(2) Computational Predictions of Functional Impacts of phosSNPs

Functional effects of the three significant phosSNPs — IDUA rs3755955 (R105Q) and rs6831280 (A361T), and WNT16 rs2707466 (WNT16B T263I) were computed using four online software tools: (i) Mutation Assessor,⁽³⁴⁾ (ii) BLOSUM62,⁽³⁵⁾ (iii) PMut,⁽³⁶⁾ and (iv) PANTHER.⁽³⁷⁾ Details about these tools are given in the Supplementary Materials and Methods.

(3) Computational Prediction of Protein Secondary and Tertiary Structures

Protein secondary and tertiary structures were predicted by Protein Homology/analogy Recognition Engine Version 2.0 (Phyre²).^(38,39) The Phyre² server predicts a protein’s secondary structure based on the amino acid sequence. In brief, this program converts a protein sequence into a hidden Markov model (HMM) based on sequence homologs retrieved from experimentally determined known protein structures using PSI-Blast.⁽⁴⁰⁾ The HMM of the query sequence is then scanned against a non-redundant library of HMMs of proteins with experimentally determined structures. The 3D model of the query sequence is then constructed on the basis of alignments between the HMM of query sequence and the HMMs of known structures. Phyre² program can generate highly accurate models at low sequence identities (e.g., 15–25%).⁽³⁹⁾

In Vitro Protein Expression Studies

To assess whether mutant (MUT) alleles of respective phosSNPs, i.e., IDUA rs3755955, rs6831280, and WNT16 rs2707466, could affect protein expression levels in vitro, we designed and constructed plasmid pcDNA3.1-Myc/His vectors harboring either wild-type (WT) or MUT allele of each phosSNP and transfected each of them into Chinese hamster ovary (CHO) cells. Details about cloning and transfection and Western blot analysis are given in the Supplementary Materials and Methods.

RESULTS

Cohort characteristics at three stages were presented in Supplementary Table S1. A detailed comparison of study designs of current study with those of two previous conventional GWA meta-analysis studies^(31,41) is shown in Figure 1. In stage 1, current study restricted association tests to exclusively phosSNPs (~10K), as opposed to the entire set of genotyped and imputed SNPs (~5800K) of previous conventional study.⁽³¹⁾ As a result, different SNP sets were selected from stage 1 for stage 2 in silico replication [9 phosSNPs for current study, and none overlapped with 33 SNPs of previous conventional study⁽³¹⁾]. In stage 2, different SNP selection criteria were employed between current study and previous conventional study.⁽³¹⁾. Four stage 2-selected phosSNPs (i.e., IDUA rs3755955 and rs6831280, WNT16 rs2707466, and ESPL1 rs56358776) of current study were entirely different from those three stage 2-selected SNPs of previous conventional study⁽³¹⁾ (i.e., SMOC1 rs227425, CLDN14 rs170183, and intergenic SNP rs6827815).

FIGURE 1.

Diagrammatic representations of study designs of three-stage GWA meta-analysis of current study (top panel), Zhang et al. (2014)⁽³¹⁾ (middle panel), and Estrada et al. (2012)⁽⁴¹⁾ (bottom panel)

Stage 1 (GWA Discovery)

Table 1 presents a comparison of 33 SNPs selected in stage 1 of previous conventional study,⁽³¹⁾ with those 9 phosSNPs selected in stage 1 of current study, which include four phosSNPs (located in three gene regions) attaining phosSNP-centric GWS (i.e., α = 0.05/9,593 = 5.21×10⁻⁶) — IBSP rs1054627 for FN-BMD in female-specific sample, IDUA rs6831280 and rs3755955 for FN-BMD in gender-combined (i.e., male and female) sample, and WNT16 rs2707466 for HIP-BMD in gender-combined sample, and another five phosSNPs (located in four gene regions) attaining only suggestive significance (i.e., α = 1.00×10⁻⁴) — SRMS rs310655 for FN-BMD in gender-combined sample, DNAH8 rs61748601 for HIP-BMD in gender-combined sample, ESPL1 rs56358776 and rs1318648 for LS-BMD in gender-combined and female-specific samples, respectively, and GPATCH1 rs2287679 for FN-BMD in female-specific sample.

Table 1.

Comparison of 33 SNPs (previous conventional study⁽³¹⁾) and 9 phosSNPs (current study) selected in stage 1^*

Stage 1 of Zhang et al. (2014)(31)						Stage 1 of Current Study
SNP ID	Locus	Gene	Phenotype	Analysis	P-value	SNP ID	Locus	Gene	Phenotype	Analysis	P-value
rs34920465	1p36.12	ZBTB40	HIP-BMD	Combined	3.06×10−9	---	---	---	---	---	---
rs1430740	1p31.3	MIR1262	LS-BMD	Combined	8.46×10−9	---	---	---	---	---	---
rs11582394	1q32.1	PLEKHA6	FN-BMD	Combined	7.14×10−7	---	---	---	---	---	---
rs11696050	2q34	PTH2R	HIP-BMD	Combined	3.72×10−8	---	---	---	---	---	---
rs11130082	3p21.31	FYCO1	LS-BMD	Combined	9.73×10−7	---	---	---	---	---	---
rs6827815	4p16.3	FGFRL1	FN-BMD	Combined	6.01×10−7	rs6831280	4p16.3	IDUA	FN-BMD	Combined	1.21×10−6
						rs3755955	4p16.3	IDUA	FN-BMD	Combined	1.80×10−6
rs17813558	4p16.1	PSAPL1	FN-BMD	Combined	9.19×10−7	---	---	---	---	---	---
rs4974930	4p14	WDR19	HIP-BMD	Combined	6.20×10−7	---	---	---	---	---	---
rs1463104	4q22.1	HSP90AB3P	LS-BMD	Combined	7.08×10−7	rs1054627	4q22.1	IBSP	FN-BMD	Female	3.54×10−7
rs4703541	5q13.2	ZNF366	FN-BMD	Combined	6.52×10−7	---	---	---	---	---	---
rs6894139	5q14.3	MEF2C	FN-BMD	Combined	2.02×10−9	---	---	---	---	---	---
---	---	---	---	---	---	rs61748601	6p21.2	DNAH8	HIP-BMD	Combined	4.83×10−5
rs1871859	6q25.1	C6orf97	LS-BMD	Female	5.04×10−	---	---	---	---	---	---
rs28529426	7p22.1	FOXK1	LS-BMD	Combined	8.12×10−7	---	---	---	---	---	---
rs2529750	7p21.1	PRPS1L1	LS-BMD	Female	7.04×10−7	---	---	---	---	---	---
rs10429035	7q21.3	FLJ42280	HIP-BMD	Combined	4.24×10−9	---	---	---	---	---	---
rs13242054	7q22.3	PIK3CG	LS-BMD	Combined	5.83×10−7	---	---	---	---	---	---
rs10242100	7q31.31	WNT16	HIP-BMD	Combined	4.63×10−8	rs2707466	7q31.31	WNT16	HIP-BMD	Combined	1.77×10−6
chr8:78734104	8q21.12	PKIA	LS-BMD	Combined	1.98×10−7	---	---	---	---	---	---
rs4424296	8q24.12	TNFRSF11B	LS-BMD	Combined	5.94×10−10	---	---	---	---	---	---
rs10868819	9q21.12	KLF9	FN-BMD	Combined	6.31×10−7	---	---	---	---	---	---
chr9:87907046	9q21.33	AGTPBP1	HIP-BMD	Male	1.88×10−7	---	---	---	---	---	---
rs7025969	9q31.3	ACTL7B	FN-BMD	Female	7.77×10−8	---	---	---	---	---	---
rs7108738	11p15.1	SOX6	FN-BMD	Combined	6.73×10−9	---	---	---	---	---	---
rs4267051	11p15.1	HPS5	HIP-BMD	Female	7.97×10−7
rs525592	11q13.2	LRP5	LS-BMD	Combined	9.04×10−7	---	---	---	---	---	---
rs471753	11q14.2	TMEM135	FN-BMD	Combined	2.20×10−7	---	---	---	---	---	---
---	---	---	---	---	---	rs1318648	12q13.13	ESPL1	FN-BMD	Female	9.87×10−5
						rs56358776	12q13.13	ESPL1	FN-BMD	Combined	5.36×10−5
rs9533095	13q14.11	AKAP11	LS-BMD	Female	3.96×10−7	---	---	---	---	---	---
rs227425	14q24.2	SMOC1	LS-BMD	Combined	2.69×10−7	---	---	---	---	---	---
rs11848357	14q32.11	RPS6KA5	LS-BMD	Combined	3.29×10−7	---	---	---	---	---	---
chr16:86714715	16q24.1	FOXL1	HIP-BMD	Female	1.86×10−9	---	---	---	---	---	---
rs10756	19q12	C19orf2	FN-BMD	Combined	7.60×10−7	---	---	---	---	---	---
---	---	---	---	---	---	rs2287679	19q13.11	GPATCH1	FN-BMD	Female	6.59×10−5
rs12481249	20q13.33	OSBPL2	LS-BMD	Combined	1.20×10−7	rs310655	20q13.33	SRMS	FN-BMD	Combined	3.57×10−5
rs170183	21q22.13	CLDN14	HIP-BMD	Female	3.71×10−7	---	---	---	---	---	---

^*Abbreviations: BMD, bone mineral density; FN, femoral neck; HIP, total hip; LS, lumbar spine; phosSNP, phosphorylation-related SNP; SNP, single nucleotide polymorphism. Combined refers to Male and Female. Those SNPs that have attained the conventional genome-wide significance (GWS) (i.e., α = 5.00×10⁻⁸) were highlighted in bold font. Shaded regions were matched regions between these two studies, although different SNPs were identified (Note: a region could contain more than one SNP).

Stage 2 (in silico Replication)

In stage 2, the above 9 stage 1-discovered phosSNPs were subject to replication in three in silico independent cohorts. A meta-analysis within stage 2 revealed 6 phosSNPs at Bonferroni corrected α = 5.56×10⁻³ (i.e., 0.05/9): WNT16 rs2707466 for FN-BMD in gender-combined sample, IBSP rs1054627 for FN-BMD in gender-combined sample, ESPL1 rs1318648 and rs56358776 for LS-BMD in gender-combined sample, and IDUA rs3755955 and rs6831280 for FN-BMD in gender-combined sample. Of these, IBSP encodes a well-known bone matrix protein that is important for bone mineralization^(42–44) which, consequently, was not further tested in stage 3. For ESPL1 phosSNPs rs1318648 and rs56358776, neither reached GWS (i.e., 5.21×10⁻⁶ ≤ P-values < 1.00×10⁻⁴) in stage 1 phosSNP-centric GWA meta-analysis. ESPL1 rs1318648 is a previously known nsSNP suggestively associated with FN- and LS-BMD phenotypes⁽⁴¹⁾, whereas ESPL1 rs56358776 is a novel nsSNP that was not reported in either of two previous conventional studies^(41,45), which is in high linkage disequilibrium (LD) with ESPL1 rs1318648 [r² = 0.798 in 1000 Genomes (1KG) Pilot 1 CEU Population by applying the SNP Annotation and Proxy search (SNAP) tool⁽⁴⁶⁾ of Broad Institute]. Therefore, we selected 4 stage 2-replicated phosSNPs — IDUA rs6831280 and rs3755955, WNT16 rs2707466, and potentially novel phosSNP ESPL1 rs56358776, for stage 3 de novo genotyping replication.

Stage 3 (de novo Genotyping Replication)

In stage 3, the above four stage 2-selected phosSNPs identified were subject to further replication by de novo genotyping. Three of these phosSNPs were replicated by stage 3-specific meta-analysis at Bonferroni corrected α = 0.0125 (i.e., 0.05/4). WNT16 rs2707466 was consistently replicated for HIP-, FN-, and LS-BMD phenotypes in gender-combined sample. IDUA rs3755955 and rs6831280 were significantly associated with FN- and HIP-BMD phenotypes in gender-combined sample. ESPL1 rs56358776 was not replicated at this stage (P-value = 0.79, 0.78, and 0.32 in gender-combined sample for FN-, HIP-, and LS-BMD, respectively).

Stage 1+2+3 Meta-analysis

Table 2 presents ethnicity-specific and combined meta-analysis results aggregating these 3 stages for stage 1-discovered (α = 1.00×10⁻⁴), and stage 2- and 3-replicated (Bonferroni-corrected α = 5.56×10⁻³ and 0.0125, respectively) phosSNPs — IDUA rs6831280 (A361T), IDUA rs3755955 (R105Q), and WNT16 rs2707466 (WNT16B T263I), respectively. In ethnicity-specific meta-analyses, in Caucasians, all three attained phosSNP-centric GWS (i.e., α = 5.21×10⁻⁶) for FN-BMD and only WNT16 rs2707466 attained this threshold for HIP-BMD and in Asians, only WNT16 rs2707466 attained phosSNP-centric GWS for HIP-BMD. The effects of these phosSNPs were consistent between Caucasian and Asian ethnicities. In combined meta-analysis across 3 stages, IDUA rs3755955 was significantly associated with FN- and HIP-BMD phenotypes (P-value = 8.36×10⁻¹⁰ and 3.26×10⁻⁶, respectively). Likewise, IDUA rs6831280 was significantly associated with FN- and HIP-BMD phenotypes (P-value = 5.26×10⁻¹⁰ and 1.97×10⁻⁶, respectively). Similarly, WNT16 rs2707466 was significantly associated with FN- and HIP-BMD phenotypes (P-value = 3.01×10⁻¹⁰ and 1.63×10⁻¹², respectively). Regional association plots were generated for these three significant phosSNPs — IDUA rs3755955 and rs6831280 (Figure 2), and WNT16 rs2707466 (Figure 3).

Table 2.

GWA Meta-analysis Results of Stage 1+2+3 for BMD-Associated PhosSNPs in Ethnicity-Specific and Combined Cohorts^*

Gene Symbol	SNP ID (WT/MUT Alleles; AA Change)	PhosSNP Type	Phenotype
			FN-BMD	HIP-BMD	LS-BMD
Caucasian (7 Cohorts)
IDUA	rs3755955 (G/A; R105Q)	Type II(+)	2.38×10−8	2.34×10−4	0.0188
IDUA	rs6831280 (G/A; A361T)	Type III	1.98×10−8	1.83×10−6	0.0366
WNT16	rs2707466 (C/T; T263I**)	Type I(−)	1.13×10−7	4.54×10−7	0.0109
Asian (3 Cohorts)
IDUA	rs3755955 (G/A; R105Q)	Type II(+)	0.03	0.0106	0.263
IDUA	rs6831280 (G/A; A361T)	Type III	0.033	0.0117	0.289
WNT16	rs2707466 (C/T; T263I**)	Type I(−)	1.20×10−3	1.04×10−6	6.67×10−3
Total (12 Cohorts)
IDUA	rs3755955 (G/A; R105Q)	Type II(+)	8.36×10−10	3.26×10−6	8.50×10−3
IDUA	rs6831280 (G/A; A361T)	Type III	5.26×10−10	1.97×10−6	0.0147
WNT16	rs2707466 (C/T; T263I**)	Type I(−)	3.01×10−10	1.63×10−12	1.17×10−4

^*Abbreviations: AA, amino acid; BMD, bone mineral density; FN, femoral neck; HIP, total hip; GWS, genome-wide significance, i.e., P-value < 5.21×10⁻⁶; LS, lumbar spine; phosSNP, phosphorylation-related SNP; SNP, single nucleotide polymorphism. PhosSNPs attaining GWS level are highlighted in bold font. Combined refers to Male and Female. PhosSNP types are defined as follows: Type I (−), a non-synonymous (nsSNP) that removes the phosphorylation site; Type II (+), an nsSNP that creates one or multiple adjacent phosphorylation sites; Type III, an nsSNP that induces changes of protein kinase type(s) at adjacent phosphorylation site(s) as defined in Ren et al. (2010),⁽²¹⁾ which were predicted by GPS2.0 software.⁽²⁰⁾

^**The amino acid position at WNT16B protein isoform (SWISS-PROT ID: Q9UBV4-1) is indicated.

FIGURE 2.

Regional association plots for chromosome 4p16.3 loci IDUA Exon 3 phosSNP rs3755955 (R105Q), Exon 8 phosSNP rs6831280 (A361T), intergenic SNP rs6827815, and FGFRL1 3’-untranslated region SNP rs4647940 based on RefSeq accession number NG_008103.1 for FN-BMD (most significant phenotype). (A) IDUA rs6831280 with flanking ± 100-kb (B) a zoomed-in view of the center region [indicated by the dashed box of (A)] — IDUA rs6831280 with flanking ± 40-kb. PhosSNPs were highlighted in bond font

FIGURE 3.

Regional association plot for WNT16 rs2707466 with flanking ± 100-kb for HIP-BMD (most significant phenotype), with chromosome 7q31.31 WNT16 Exon 2 nsSNP rs2908004 (WNT16B, G82R), Intron 3 SNP rs3801387, Exon 4 phosSNP rs2707466 (WNT16B T263I), and intergenic SNP rs10242100 based on RefSeq accession number NG_029242.1 indicated. The phosSNP was highlighted in bond font

Phosphorylation Sites Predicted to be Affected by IDUA and WNT16 PhosSNPs

Based on predictions by NetPhos2.0 and NetPhosK1.0, four phosphorylation sites (have either a NetPhos2.0 score > 0.5 or a NetPhosK1.0 score > 0.5) were predicted by these two in silico bioinformatics tools that could be affected by these three BMD-associated phosSNPs (Table 3). Detailed information on 96 and 54 predicted phosphorylation sites for IDUA and WNT16B were presented in Supplementary Tables S2 and S3, respectively. IDUA phosphorylation sites T98 and S102 were potentially affected by their neighboring phoSNP IDUA rs3755955 (R105Q), whereas IDUA phosphorylation site T366 was potentially affected by a neighboring phoSNP IDUA rs6831280 (A361T). WNT16B phosphorylation site T263 was potentially directly abolished by phoSNP WNT16 rs2707466 (WNT16B T263I). Of them, WNT16B T263 (affected by WNT16 rs2707466) has been experimentally validated to be phosphorylated in vivo,⁽⁴⁷⁾ whereas IDUA T98 and S102 (potentially affected by IDUA rs3755955) and T366 (potentially affected by IDUA rs6831280) have yet not been experimentally confirmed.

Table 3.

In Silico Predicted Phosphorylation Sites of Three PhosSNPs Associated with BMD Phenotypes^*

Gene Symbol	Predicted Phosphorylation Site [represented by PSP(7,7)]	PhosSNP ID (WT/MUT Alleles; AA Change)	PhosSNP Type	NetPhos2.0 Score (prediction)	NetPhosK1.0 Score (prediction)
IDUA	HWLLELVTTRGSTGQ AA Pos: 98	rs3755955 (G/A; R105Q)	Type II(+)	0.417 (Probable)	0.68 (Yes)
IDUA	ELVTTRGSTGQGLSY AA Pos: 102	rs3755955 (G/A; R105Q)	Type II(+)	0.987 (Yes)	0.51 (Yes)
IDUA	PFTQRTLTARFQVNN AA Pos: 366	rs6831280 (G/A; A361T)	Type III	0.600 (Yes)	0.51 (Yes)
WNT16	SIQISDKIKRKMRRR AA Pos: 263**	rs2707466 (C/T; T263I**)	Type I(−)	0.833 (Yes)	0.86 (Yes)

^*Abbreviations: AA, amino acid; BMD, bone mineral density; NetPhos, Neural Network Phosphorylation Predictor; NetPhosK, Neural Network Phosphorylation Kinase-Specific Predictor; phosSNP, phosphorylation-related SNP. Pos, position; PSP, phosphorylation site peptide; SNP, single nucleotide polymorphism; WT, wild-type. PhosSNP types were defined as in footnote of Table 1. The phosphorylation acceptor residue is underlined, and the phosSNP site (mutant allele shown) is highlighted in bold and italic font. Predicted phosphorylation sites for respective phosSNPs have either a NetPhos2.0 score > 0.5 or a NetPhosK1.0 score > 0.5. GPS2.0 scores were directly extracted from PhosSNP1.0 database,⁽²¹⁾ where these scores were greater than their respective thresholds ― 3.26, 2.85, and 2.43 for IDUA protein (SWISS-PROT ID: P35475) positions 98, 102 and 366, respectively, and 4.48 for WNT16B protein (SWISS-PROT ID: Q9UBV4-1) position 263. For NetPhos2.0 and NetPhosK1.0 scores, “Yes”, “Probable” and “No” refer to a score > 0.5, 0.1―0.5, and < 0.1, respectively.

^**The amino acid position at WNT16B protein isoform (SWISS-PROT ID: Q9UBV4-1) is indicated.

Predicted Functional Impacts of IDUA and WNT16 PhosSNPs

As shown in Supplementary Table S4, although IDUA rs6831280 (A361T) and rs3755955 (R105Q) were predicted to have no (Mutation Assessor and BLOSUM62 scores) or low (PMut and PANTHER scores) functional impacts, WNT16 rs2707466 (WNT16B T263I) showed highest Mutation Assessor score (0.705; nearly reaching a “low impact” threshold 0.80), lowest BLOSUM62 score of (−1.00; indicative of “evolutionarily less acceptable”), highest PMut pathogenicity score (0.3099; indicative of a “moderate pathogenicity”), and lowest PANTHER subSPEC score (−1.92476; indicative of a deleterious effect corresponding to a highest deleteriousness probability P_deleterious = 0.25441). Further, evolutionary analysis by multiple sequence alignment method revealed that a 27-amino-acid peptide (−14 — +12) surrounding the T263 phosphorylation site is conserved across three mammalian species — human, mouse and rat (Supplementary Figure S1), supporting a likely functional significance of this phosSNP. Based on these bioinformatics prediction results, we further assessed the potential impact of WNT16 rs2707466 (WNT16B T263I) on WNT16B secondary and tertiary structures.

Predicted Secondary and Tertiary Structures of WT and MUT alleles for WNT16 PhosSNP

The secondary and tertiary structures of protein isoforms encoded by WT and MUT alleles for WNT16 rs2707466 (WNT16B T263I) predicted by Phyre² server are presented in Supplementary Figures S2 and S3, respectively. With respect to secondary structures, this phosSNP (i.e., T263 residue) is located in a disordered region (indicated by a tract of “?” symbols) typical for a phosphorylation site,⁽⁴⁸⁾ downstream of a predicted β-strand (SIQISDK) for either isoform (Supplementary Figure S2) with potential functional effects (Supplementary Table S4). A comparison of the local 3D structures between WT and MUT isoforms near the T263 residue clearly shows different spatial patterns (Supplementary Figure S3, dashed boxes).

Effects of IDUA and WNT16 PhosSNPs on Protein Stability

In CHO cells, Western blot results showed that, at the protein level, IDUA rs6831280 (A361T) and rs3755955 (R105Q) MUT alleles were expressed at equivalent levels compared with the IDUA WT allele (Panel As for Supplementary Figures S4 and S5, respectively). The WNT rs2707466 (WNT16B T263I) MUT allele was also expressed at equivalent levels compared with the WNT16 WT allele (Panel B for Supplementary Figures S4 and S5, respectively). Overall, the protein expression of the MUT allele is equivalent to that of the WT allele for each of these three phosSNPs, suggesting that their influences of protein phosphorylations could be important, rather than on expression levels per se.

DISCUSSION

In human genome, nsSNPs account for 60% of mutations that cause diseases.⁽¹²⁾ However, not all nsSNPs lead to a functional impact. Therefore, it is essential to select only those nsSNPs that are most plausible causal variants. Our study is unique in associating phosSNPs affecting the most common type of PTM with BMD phenotypes by taking a three-stage approach to protect against an inflated false positive rate. Beyond detecting genetic association, we also performed in silico and in vitro functional characterizations of identified significant nsSNPs. At stage 1, four chromosomal loci i.e., 4p16.3 (IBSP), 4q22.1 (IDUA), 7q31.31 (WNT16), and 20q13.33 (GPATCH1), were detected by both the current and conventional studies,⁽³¹⁾ but were represented by totally different SNPs, and for 4q22.1 and 20q13.33, represented by different genes. At 7q31.31, the previous study detected association with intergenic SNP rs10242100 (with no apparent functional significance) near WNT16 gene, which is in moderate LD with the WNT16 SNP rs2707466 detected by our current study [r² = 0.462 in 1KG Pilot 1 CEU Population by applying SNAP tool⁽⁴⁶⁾]. Overall, three phosSNPs (IDUA rs6831280 and rs3755955, and WNT16 rs2707466), were discovered in stage 1 and were independently replicated in stages 2 and 3, respectively. In ethnicity-specific meta-analyses, their effects were consistent in subgroups of Caucasian and Asian ancestries, and statistical significances were greater in Caucasian than in Asian samples in part because of a larger Caucasian sample size (Table 2). In combined stage 1+2+3 meta-analysis, all three phosSNPs reached conventional GWS for FN-BMD, and WNT16 rs2707466 attained conventional GWS for HIP-BMD also. By applying NetPhos2.0 and NetPhosK1.0, 96 and 54 predicted phosphorylation sites in IDUA and WNT16B proteins, respectively (Supplementary Tables S3 and S4). IDUA encodes a glycosyl hydrolase that hydrolyzes the terminal alpha-L-iduronic acid residues of two glycosaminoglycans, dermatan sulfate and heparan sulfate⁽⁴⁹⁾. Wang et al. (2010)⁽⁵⁰⁾ created Idua-W392X mouse model, and found that 35-week-old homozygous Idua-W392X mice showed a 24% increase in femur BMD, and bone abnormalities such as thickening of the zygomatic arch and aberrations in the length and width of the femur were also observed⁽⁵⁰⁾.

For IDUA protein, a predicted phosphorylation site T366 could be indirectly affected by IDUA rs6831280 (A361T), a Type III phosSNP, and two predicted phosphorylation sites T98 and S102, could be indirectly affected by IDUA rs3755955 (R105Q), a Type II(+) phosSNP (Table 3). For WNT16B protein, phosphorylation site T263 could be directly abolished by WNT16 rs2707466 (WNT16B T263I), a Type I(−) phosSNP. Of them, only WNT16B T263 has been experimentally validated to be a genuine phosphorylation site by mass spectrometry technology in a phosphoproteomic analysis of human embryonic stem cells in vivo.⁽⁴⁷⁾ Whether IDUA T98 and S102, and T366 are actual phosphorylation sites influenced by nearby IDUA phosSNPs rs3755955 (R105Q) and rs6831280 (A361T) remain to be experimentally determined.

WNT16 encodes a member of the wingless-type MMTV integration site family, which has been reported to mediate signaling via both canonical and non-canonical Wnt pathways. Wnt proteins are known to play important roles in vertebrate skeletal development.^(51–53) Wnt16 is expressed in osteoid tissue of craniofacial bones during embryonic development in mice, and suppresses osteoblast differentiation through the canonical β-catenin pathway in MC3T3-E1 preosteoblasts.⁽⁵⁴⁾ Several GWA meta-analysis studies have demonstrated that WNT16 intron 3 SNP rs3801387,⁽⁴¹⁾ exon 2 rs2908004 (WNT16B G82R) and exon 4 rs2707466 (WNT16B T263I)^(55,56) as well as intergenic SNP rs10242100⁽³¹⁾ are associated with BMD phenotypes (Figure 3). However, functional roles of non-coding SNPs rs10242100 and rs3801387, which are in almost perfect LD [r² = 0.915 in 1KG Pilot 1 CEU Population by applying SNAP tool⁽⁴⁶⁾], remain unclear. WNT16 exon 2 rs2908004 (WNT16B G82R) and exon 4 rs2707466 (WNT16B T263I) are shown to be in nearly complete LD [r² = 0.933 in 1KG Pilot 1 CEU Population by applying SNAP tool⁽⁴⁶⁾], which could represent the same phosphorylation association signal (i.e., WNT16 rs2707466). Consistent with our results, exon 2 nsSNP rs2908004 was significantly associated with upper limbs BMD, lower limbs BMD, as well as skull BMD phenotypes, and is the top signal in the chromosome 7q31.31 region in a GWA meta-analysis of the Avon Longitudinal Study of Parents and their Children and Generation R Study.⁽⁵⁷⁾ The phosSNP WNT16 rs2707466 results in a substitution of threonine by isoleucine in both WNT16A (amino acid position 253) and WNT16B (amino acid position 263) isoforms. This phosSNP is predicted to exert a modest impact on protein function (by Mutation Assessor), and to be evolutionarily less acceptable (by BLOSUM62) and moderately deleterious (by PMut and PANTHER) (Supplementary Table S4). Because WNT16B T263 has been experimentally confirmed to be a phosphorylation site in vivo,⁽⁴⁷⁾ in silico secondary structure prediction shows that T263 is located in a disordered region (Supplementary Figure S2). This is in agreement with findings of Dephoure et al.,⁽⁵⁸⁾ which demonstrated that phosphorylation sites mostly occur in disordered regions, and the addition of a phosphate group to acceptor residue upon phosphorylation can lead to a disorder-to-order transition.⁽⁵⁹⁾ Predicted local 3D structures also indicate notable differences between WT and MUT isoforms around T263 phosphorylation site (Supplementary Figure S3). Taken together, it is highly probable that T263, a phosphorylatable residue located in a disordered region of WNT16B protein, acts as a switch for regulating protein-protein interactions,⁽⁵⁹⁾ and WNT16 rs2707466, a type I(−) phosSNP that abolishes this phosphorylation site, constitutes a causal variant for BMD phenotype. This is supported by observations of wnt16 null mice, which had significantly reduced total body BMD, thinner cortical bones at the femur midshaft, and reduced bone strength of both the femur and tibia.^(55,56). Further, local injection of WNT16B (WT form) could increase BMD, providing direct experimental evidence that WNT16 gene is critical for skeletal development.⁽²⁵⁾

There are several limitations of our study. First, 9,593 phosSNPs included in stage 1 (GWA discovery) of current study represent 14.98% of the entire 64,035 phosSNP set. Nevertheless, the original 5,842,825 autosomal SNPs either directly-typed or imputed in stage 1 of conventional GWA study⁽³¹⁾ only covered 15.92% of entire 36.7 million human autosomal SNP set⁽⁶⁰⁾. Therefore, although these included phosSNPs appear limiting, they constitute a similar proportion of total phosSNPs as the original stage 1 SNP set of previous conventional study⁽³¹⁾. Second, our in vitro protein expression experiments of WT and MUT alleles of IDUA rs3755955 (R105Q), rs6831280 (A361T), and WNT16 rs2707466 (WNT16B T263I) only demonstrated relatively equivalent protein expression levels between WT and MUT alleles (Panel Bs of Supplementary Figures S4 and S5, respectively). Additional experiments by applying phospho-specific antibodies could be insightful to reveal whether these phosSNPs truly affect protein phosphorylations either directly (for WNT16 phosSNP rs2707466) or indirectly (for IDUA phosSNP rs6831280 and rs3755955). However, such experiments are time-consuming and the extents of such differences may be challenging to detect, because both IDUA and WNT16B proteins can have multiple phosphorylation sites, and these phosSNPs may only affect 1–2 among them. It also remains to be shown whether a fraction of BMD variation is attributed to impacts of IDUA rs6831280 (A361T) and rs3755955 (R105Q) on their neighboring IDUA putative phosphorylation sites T98, S102 and T366, and to abolishment of WNT16B T263 phosphorylation site by WNT16 rs2707466 (WNT16B T263I). Nevertheless, the study of Moverare-Skrtic et al.⁽²⁵⁾ clearly demonstrated a pivotal role of WNT16B WT isoform in skeletal development, and phosSNP rs2707466 could indeed play a major functional role in regulating bone metabolism.

The collective findings from our multi-stage phosSNP-centric GWA meta-analysis identified and robustly validated three phosSNPs, IDUA rs6831280 and rs3755955 and WNT16 rs2707466, to be significantly associated with FN- and HIP-BMD. These results could offer new mechanistic insights of causal variants for osteoporosis. Because currently there is a lack of bone-specific phosphorylation maps, for those phosphorylation sites that are impacted by these BMD-associated phosSNPs, more studies are necessary to elucidate whether phosphorylations affected by them are present in these various types of bone cells.