Integrating genome-wide association study with regulatory SNP annotations identified novel candidate genes for osteoporosis

Abstract

Aims

Osteoporosis (OP) is a metabolic bone disease, characterized by a decrease in bone mineral density (BMD). However, the research of regulatory variants has been limited for BMD. In this study, we aimed to explore novel regulatory genetic variants associated with BMD.

Methods

We conducted an integrative analysis of BMD genome-wide association study (GWAS) and regulatory single nucleotide polymorphism (rSNP) annotation information. Firstly, the discovery GWAS dataset and replication GWAS dataset were integrated with rSNP annotation database to obtain BMD associated SNP regulatory elements and SNP regulatory element-target gene (E-G) pairs, respectively. Then, the common genes were further subjected to HumanNet v2 to explore the biological effects.

Results

Through discovery and replication integrative analysis for BMD GWAS and rSNP annotation database, we identified 36 common BMD-associated genes for BMD irrespective of regulatory elements, such as FAM3C (p_{discovery GWAS} = 1.21 × 10^-25, p_{replication GWAS} = 1.80 × 10^-12), CCDC170 (p_{discovery GWAS} = 1.23 × 10^-11, p_{replication GWAS} = 3.22 × 10^-9), and SOX6 (p_{discovery GWAS} = 4.41 × 10^-15, p_{replication GWAS} = 6.57 × 10^-14). Then, for the 36 common target genes, multiple gene ontology (GO) terms were detected for BMD such as positive regulation of cartilage development (p = 9.27 × 10^-3) and positive regulation of chondrocyte differentiation (p = 9.27 × 10^-3).

Conclusion

We explored the potential roles of rSNP in the genetic mechanisms of BMD and identified multiple candidate genes. Our study results support the implication of regulatory genetic variants in the development of OP.

Cite this article: Bone Joint Res 2023;12(2):147–154.

Article focus

• To access the genetic mechanism of osteoporosis (OP) using the integrative analysis of genome-wide association study (GWAS) and regulatory single nucleotide polymorphism (rSNP) annotation information of bone mineral density (BMD).

• We performed functional enrichment analysis of the common candidate genes associated with BMD.

Key messages

• A total of 36 target regulatory genes were detected for BMD such as FAM3C, CCDC170, SOX6, and PLEKHM1.

• We detected 12 BMD-associated gene ontology terms such as muscle cell differentiation, positive regulation of cartilage development, and positive regulation of chondrocyte differentiation.

Strengths and limitations

• Integrating analysis of GWAS and rSNP has a boosting power for regulatory genetic variant detection.

• One of the main limitations is that rSNP analysis could not take tissue or cell types into account.

Introduction

Osteoporosis (OP) is a metabolic bone disease, characterized by a decrease in the density of bone.¹ Multiple risk factors contribute to the development of OP such as hormone changes, gut microbiome, the use of certain drugs, and cigarette smoking.^2-4 In the USA, millions of people either already have OP or are at high risk due to low bone mass.⁵ OP is more common in postmenopausal women,⁶ which leads to heavy burden on the healthcare system and society.

Bone mineral density (BMD) is one of the major diagnostic indexes of OP. A recent study suggested that bone density had a strong genetic determination, with an estimated heritability ranging from 61% to 83%.⁷ More than 50 susceptibility loci have been identified for BMD or OP, such as LRP5, MEPE, SPTBN1, and DKK1.^8-10 For instance, genome-wide association studies (GWASs) have led to the identification of 100 loci associated with BMD and other bone traits related to risk of fracture.¹¹ However, only a small percentage of the heritability of BMD can be explained by the susceptibility genetic variants discovered so far,¹² suggesting the existence of undiscovered causal genetic factors for BMD.

Although GWASs have achieved great success in identifying the susceptible loci of complex diseases, there are several issues and limitations associated with these genetic association studies.^13,14 GWAS signals often lie in non-coding regions, which may harbour regulatory elements affecting gene expression, such as expression quantitative trait loci (eQTLs) and methylation quantitative trait loci (meQTLs).^15,16 However, it is difficult to identify the precise causal SNP and causal gene for these noncoding variants.¹⁷ These causal genetic variants within non-coding regulatory regions are indistinguishable from the neighbouring markers.

Regulatory single nucleotide polymorphisms (rSNPs) refer to the SNPs that are associated with the regulation of gene expression levels through transcription factor binding, chromatin interaction, circular RNA (circRNA)-mediated post-transcriptional regulation, and so on.^18,19 Hindorff et al¹⁴ have illustrated that a lot of risk SNPs could affect phenotypes in a non-coding manner, for instance impacting gene regulation. Integrating GWAS datasets with rSNPs has the potential to reveal novel susceptibility genetic variants for human complex diseases.²⁰ A previous study has shown that a rSNP in the Cx43 promoter region plays a critical role in Tetralogy of Fallot (TOF), by impacting the transcriptional activity.²¹ Furthermore, Yeo et al²² reported that low-effect chronic obstructive pulmonary disease (COPD) risk SNPs were identified through enrichment of cis-regulatory SNPs in genes.

In this study, we conducted an integrative analysis of GWAS and large-scale rSNP annotation data for BMD, containing seven regulatory genetic elements. The significant loci identified by the GWAS of BMD were firstly annotated with the rSNP annotation database to obtain BMD-associated regulatory genetic elements and their target genes. To explore the functional relevance of identified candidate genes, the common candidate genes identified by both discovery and replication studies were further subjected to gene ontology (GO) functional enrichment analysis.

Methods

Discovery GWAS summary dataset of BMD

The GWAS summary dataset of BMD was drawn from the GEnetic Factors for OSteoporosis (GEFOS) Consortium,²³ containing a total of 52,236 individuals of European ancestry. Briefly, the GEFOS project used meta-analysis of whole genome sequencing, whole exome sequencing, and deep imputation of genotype data to identify candidate variants associated with the risk of BMD. Single variants with a minor allele frequency (MAF) > 0.5% were tested for an additive effect on femoral neck BMD (FN-BMD), lumbar spine BMD (LS-BMD), and forearm BMD (FA-BMD), adjusting for sex, age, age_squared, and weight as covariates. Detailed information of the subjects, genotyping, imputation, and quality control can be found in a previously published study.⁸

Replication GWAS summary dataset of BMD

The replication GWAS summary dataset of BMD was drawn from the UK Biobank database.^24-26 In summary, the UK Biobank BMD GWAS dataset contains 452,264 participants. The BMD values of bones and joints were measured using dual-energy X-ray absorptiometry in this study. DNA was extracted from stored blood samples and shipped to Affymetrix Research Services Laboratory for genotyping. SNP genotyping was performed using the UK Biobank Axiom array.²⁷ Genotypes were imputed by IMPUTE4.²⁸ Principal component analysis (PCA) was used to account for potential population structure in both marker- and sample-based quality control procedures. Detailed information of the subjects, genotyping, imputation, and quality control can be found in two previously published studies.^24,25

Annotation datasets of rSNPs

The rSNP annotation data were collected from the rSNPBase 3.1 database.^29,30 There are seven types of regulatory elements, including 7,562,592 annotation terms for transcription factor binding regions (TFBRs) from the Encyclopaedia of DNA Elements (ENCODE) project,³¹ 212,837 for chromatin interactive regions (CIRs) from the ENCODE project,³¹ 2,794 for mature microRNA (miRNA) regions from miRbase,³² 384,284 for predicted miRNA target sites from TargetScan³³ and miRNAda,^34-37 211,749 for long non-coding RNA (lncRNA) regions from LNCipedia,³⁸ 38,916 for topologically associated domains (TADs) from ENCODE-processed data, and 312,673 for circRNAs from CircNet.³⁹ Except for circRNAs, target gene analyses were performed on the other six types of regulatory elements to get corresponding elements-gene (E-G) pairs.

For regulatory E-G pairs of transcriptional regulation, chromosome location was acquired from ENCODE Consortium, among which TFBR information was acquired from proceeded ChIP-seq peak data, CIR information was acquired from Chromatin Interactions by 5 C and ChIA-PET, and TAD information was acquired from proceeded Hi-C data. Ensembl recorded genes on hg19 coordinate⁴⁰ was used to analyze the genome locations of these three types of regulatory elements through the potential promoter region (from 2 k upstream to 1 k downstream of transcription). For E-G pairs of non-coding RNAs, miR2Disease, miRTarBase, and lncRNA2Target databases were integrated into rSNPBase for miRNA-gene and lncRNA-gene pairs with experimentally supported target relations through RT-PCR, microarray, or RNA-seq.^41-43 Moreover, TargetScan and miRnada databases were used to obtain predicted miRNA target sites,^34-36 which were mapped to human genome sequence (on hg 19 coordinate). The functional association between the included regulatory elements and target genes was analyzed with genomic proximity or by using widely used reference databases as mentioned above. ^{31,32,34-36–38-43}

Integrating analysis of GWAS and rSNP annotation information

Firstly, significant SNPs were selected from the discovery GWAS dataset with p-value < 5 × 10^-8, and then were annotated in rSNPBase 3.1 to obtain BMD-associated SNP regulatory elements and E-G pairs. Then, the other GWAS replicate summary dataset of BMD was used to validate the integration results with p-value < 5 × 10^-8. Finally, we selected the common genes between the discovery and replication analysis results. A flowchart for this study is depicted in Figure 1.

Fig. 1.

Flowchart for integrating analysis of genome-wide association study (GWAS) and regulatory single nucleotide polymorphism (rSNP) annotation data for bone mineral density (BMD).

Functional enrichment and annotation analysis

For the BMD-associated genes overlapping in both discovery and replicate studies, GO enrichment analysis was conducted by HumanNet v2,⁴⁴ a database of human gene networks.⁴⁵ The network of HumanNet-XC (Functional gene network extended by Co-citation) and the network-based disease gene prediction of HumanNet were used in our study.

Statistical analysis

The significant SNPs with GWAS p-value < 5 × 10^-8 were selected from both the discovery and replication GWAS summary dataset of BMD. To explore the functional relevance of identified common genes shared by both the discovery and replicate studies, GO and pathway enrichment analyses were performed by HumanNet-XC and the network-based disease gene prediction of HumanNet. The significance threshold was set as p < 0.05.

Results

BMD-associated rSNPs and their target genes

In discovery GWAS dataset study, for FA-BMD, we detected six rSNPs for TFBRs, 14 rSNPs for TADs, and 12 rSNPs for CIRs, corresponding to four, three, and six target regulatory genes, respectively. For circRNA region, 108 rSNPs were identified for FA-BMD, 16 of which have been demonstrated as eQTLs in a previous study by Guo et al²⁹ (Supplementary Table i). For FN-BMD, we detected 20 rSNPs for TFBRs, 140 rSNPs for TADs, 53 rSNPs for CIRs, and 15 rSNPs for lncRNAs, corresponding to 19, 33, 49, and four target regulatory genes, respectively. For circRNA region, 135 rSNPs were identified for BMD, 100 of which have been demonstrated as eQTLs by Guo et al²⁹ (Supplementary Table ii). For LS BMD, we detected 32 rSNPs for TFBRs, 268 rSNPs for TADs, 69 rSNPs for CIRs, and six rSNPs for lncRNAs, corresponding to 18, 31, 44, and four target regulatory genes, respectively. For circRNA region, 400 rSNPs were observed for BMD, 204 of which have been identified as eQTLs by Guo et al²⁹ (Supplementary Table iii).

For the replication results, we detected 533 rSNPs for TFBRs, 4298 rSNPs for TADs, 1336 rSNPs for CIRs, one rSNP for miRNA, six rSNPs for predicted miRNA target site, and 365 rSNPs for lncRNAs, corresponding to 1,136, 2,645, 2,275, four, 12, and 380 target regulatory genes, respectively. For circRNA region, 6,446 rSNPs were identified for BMD, 90 of which have been demonstrated as eQTLs by Guo et al²⁹ (Supplementary Table iv).

After comparing the discovery and replication study results, we identified four overlapped target regulatory genes for TFBRs, 30 overlapped target regulatory genes for TADs, and 14 overlapped target regulatory genes for CIRs, respectively (Table I). Irrespective of different regulatory genetic elements, 36 target regulatory genes were detected for BMD, such as FAM3C for TADs and CIRs (p_{discovery GWAS} = 1.21 × 10^-25, beta = 0.186, p_{replication GWAS} = 1.80 × 10^-12), CCDC170 for TADs (p_{discovery GWAS} = 1.23 × 10^-11, beta = -0.053, p_{replication GWAS} = 3.22 × 10^-9), SOX6 for TADs (p_{discovery GWAS} = 4.41 × 10^-15, beta = 0.080, p_{replication GWAS} = 6.57 × 10^-14), and PLEKHM1 for TADs (p_{discovery GWAS} = 1.75 × 10^-8, beta = 0.085, p_{replication GWAS} = 1.56 × 10^-15).

Table I.

List of common regulatory single nucleotide polymorphisms and their target regulatory genes shared by both discovery and replication studies.

Gene	Discovery				Replication		Regulatory elements
	GWAS-rSNPs	EAF	Beta	pdiscovery	GWAS-rSNPs	preplication
FAM3C	rs7776725	0.259	0.186	1.21 × 10-25	rs7776725	1.80 × 10-12	CIRs TADs
PTPRZ1	rs7776725	0.259	0.186	1.21 × 10-25	rs7776725	1.80 × 10-12	CIRs TADs
CCDC122	rs9533094	0.404	-0.083	2.80 × 10-20	rs150081494	1.05 × 10-15	TADs
DCLK1	rs9533094	0.404	-0.083	2.80 × 10-20	rs191115702	1.52 × 10-16	TADs
TPTE2P5	rs9594738	0.403	-0.082	3.63 × 10-20	rs150081494	1.05 × 10-15	TADs
ENSG00000253603	rs7006553	0.385	0.078	9.99 × 10-18	rs140103265	1.80 × 10-12	CIRs
C11orf58	rs16931831	0.195	0.080	4.41 × 10-15	rs138547759	6.57 × 10-14	TADs
SOX6	rs16931831	0.195	0.080	4.41 × 10-15	rs138547759	6.57 × 10-14	TADs
ULK4	rs369145	0.451	-0.057	1.57 × 10-13	rs117982632	2.26 × 10-9	CIRs
	rs10896328	0.238	-0.081	2.09 × 10-13	rs117982632	2.26 × 10-9	TADs
PPP6R3	rs12272917	0.237	-0.074	4.58 × 10-13	rs17149179	3.14 × 10-10	CIRs
	rs12274114	0.255	-0.072	1.17 × 10-12	rs67446678	2.32 × 10-9	TFBRs
	rs10896328	0.238	-0.081	2.09 × 10-13	rs72936524	5.15 × 10-10	TADs
CHKA	rs12272917	0.237	-0.074	4.58 × 10-13	rs2186937	6.70 × 10-10	CIRs
CTTN	rs4084149	0.237	-0.075	3.43 × 10-13	rs374538301	1.16 × 10-11	TADs
ELMO1	rs1357651	0.355	-0.068	3.75 × 10-13	rs147344245	2.23 × 10-14	TADs
AMPH	rs1357651	0.355	-0.068	3.75 × 10-13	rs147344245	2.23 × 10-14	TADs
NDUFA3P2	rs12272917	0.237	-0.074	4.58 × 10-13	rs2186937	6.70 × 10-10	CIRs TADs
ENSG00000234985	rs7776725	0.259	0.072	6.85 × 10-13	rs7776725	1.80 × 10-12	CIRs TADs
GEMIN6	rs73215359	0.080	0.183	4.77 × 10-12	rs114805039	5.80 × 10-11	CIRs TFBRs
ADGB	rs7740522	0.505	-0.053	1.23 × 10-11	rs115947573	9.21 × 10-24	TADs
SYNE1	rs7740522	0.505	-0.053	1.23 × 10-11	rs115947573	9.21 × 10-24	TADs
PLEKHG1	rs7740522	0.505	-0.053	1.23 × 10-11	rs145135839	4.74 × 10-17	TADs
CCDC170	rs7740522	0.505	-0.053	1.23 × 10-11	rs182402542	3.22 × 10-9	TADs
CBX5	rs3803042	0.417	-0.061	3.23 × 10-11	rs57292110	7.11 × 10-16	CIRs TFBRs
COG6	rs9533110	0.496	0.055	9.83 × 10-10	rs116613004	1.38 × 10-11	TADs
EPSTI1	rs9533110	0.496	0.055	9.83 × 10-10	rs554647317	4.39 × 10-8	TADs
ENSG00000254789	rs12797380	0.477	0.047	2.05 × 10-9	rs539468022	1.63 × 10-10	TADs
PSMA1	rs12797380	0.477	0.047	2.05 × 10-9	rs539468022	1.63 × 10-10	TADs
SUV420H1	rs7935347	0.153	-0.075	3.08 × 10-9	rs183720308	1.03 × 10-10	CIRs
RNU2-63P	rs4528686	0.362	0.071	6.58 × 10-9	rs116151937	5.28 × 10-14	CIRs TFBRs
EPDR1	rs1376264	0.271	0.055	9.30 × 10-9	rs147344245	2.23 × 10-14	TADs
ANLN	rs1376264	0.271	0.055	9.30 × 10-9	rs138651438	6.42 × 10-10	TADs
OTOG	rs11024028	0.147	0.066	1.44 × 10-8	rs372278011	1.02 × 10-14	TADs
ENSG00000246225	rs11024028	0.147	0.066	1.44 × 10-8	rs77490678	3.00 × 10-16	TADs
USAH1C	rs11024028	0.147	0.066	1.44 × 10-8	rs138547759	6.57 × 10-14	TADs
BDNF-AS	rs11024028	0.147	0.066	1.44 × 10-8	rs187573374	3.33 × 10-17	TADs
PLEKHM1	rs34814687	0.053	0.085	1.75 × 10-8	rs144518135	1.56 × 10-15	TADs
CSAD	rs4759021	0.248	0.053	4.04 × 10-8	rs556893559	8.42 × 10-12	CIRs

p_discovery denotes the p-value of discovery GWAS-rSNPs; p_replication denotes the p-value of replication GWAS-rSNPs.

CIRs, chromatin interactive regions; EAF, effect allele frequency; GWAS, genome-wide association study; rSNP, regulatory single nucleotide polymorphism; TADs, topologically associated domains; TFBRs, transcription factor binding regions.

Gene ontology and pathway enrichment analysis

For the 36 target genes shared by both discovery and replication study, GO enrichment analysis detected 12 GO terms with p < 0.01 for BMD, such as positive regulation of cartilage development (p = 9.27 × 10^-3), positive regulation of chondrocyte differentiation (p = 9.27 × 10^-3), muscle cell differentiation (p = 5.31 × 10^-3), and regulation of p38 mitogen-activated protein kinase (p38MAPK) cascade (p = 5.31 × 10^-3) (Table II).

Table II.

Gene ontology enrichment analysis results of the common genes identified by both discovery and replication studies.

Term ID	Term description	p-value
GO:0007417	central nervous system development	2.86 × 10-3
GO:2000741	positive regulation of mesenchymal stem cell differentiation	5.31 × 10-3
GO:1900744	regulation of p38MAPK cascade	5.31 × 10-3
GO:0042692	muscle cell differentiation	5.31 × 10-3
GO:0090286	cytoskeletal anchoring at nuclear membrane	6.63 × 10-3
GO:2000726	negative regulation of cardiac muscle cell differentiation	6.63 × 10-3
GO:0007096	regulation of exit from mitosis	6.63 × 10-3
GO:0045987	positive regulation of smooth muscle contraction	6.63 × 10-3
GO:0043666	regulation of phosphoprotein phosphatase activity	7.95 × 10-3
GO:0050957	equilibrioception	7.95 × 10-3
GO:0061036	positive regulation of cartilage development	9.27 × 10-3
GO:0032332	positive regulation of chondrocyte differentiation	9.27 × 10-3

GO, gene ontology; MAPK, mitogen-activated protein kinase.

Discussion

Recent studies have demonstrated the important roles of regulatory genetic variants in the genetic mechanism of human complex diseases.^21,22 To explore the functional relevance of regulatory genetic variants with BMD and identified novel candidate genes for OP, we conducted an integrative analysis of GWAS with rSNP annotation information. We identified multiple rSNPs and their target genes for BMD. Further functional analysis of the identified target genes supports the important roles of rSNPs in the development of OP.

One important finding of this study is FAM3C, which was detected between discovery and replication studies. FAM3C, a member of the family with sequence similarity 3 (FAM3), encodes a secreted protein with a GG domain. In a three-stage genome-wide association (GWA) meta-analysis study, FAM3C was confirmed to be associated with BMD.⁴⁶ Additionally, a follow-up study in Caucasians observed that rs7776725 influenced BMD at multiple skeletal sites.⁴⁷ Interestingly, the variant rs7776725 was located in the first intron of the FAM3C gene. Notably, several other SNPs identified in this gene were also associated with BMD at multiple sites,⁴⁷ indicating that FAM3C may play a notable role in bone metabolism. Furthermore, the SNP rs7776725 in FAM3C was also proved to be associated with a genome-wide substantially increased risk of forearm fracture.⁴⁸FAM3C was demonstrated to play a functional role in the regulation of osteoblast differentiation. In differentiating osteoblasts, knockdown of FAM3C increased alkaline phosphatase expression and activity whereas overexpression of FAM3C reduced it. Furthermore, FAM3C and TGF-β1 were found to regulate each other reciprocally.⁴⁹ Furthermore, our analysis found that rs7776725 targeted FAM3C and PTPRZ1 through regulating CIRs and TADs, which belong to transcriptional regulation. Previous studies have shown that FAM3C indeed has an effect on bone metabolism-related disease through transcriptional regulation, which was consistent with our analysis.^46-49

SOX6 is another important gene, which encodes transcription factor SOX-6. It has been reported that the SOX6 gene is an essential transcription factor in chondrogenesis and cartilage formation.^50,51 For example, SOX6 was discovered to have differential expression during osteoblast development.⁵² Homozygous Sox6 mutant (Sox6-/-) mice were born with relatively mild skeletal anomalies and mesenchymal condensations, but there was no overt chondrocyte differentiation.⁵⁰ Moreover, expression of chondrocyte marker genes was severely reduced in Sox6 mutant mice. Sox5, Sox6, and Sox9 are necessary for chondrogenic differentiation at different steps, the combination of which provides a new in vitro chondrogenic differentiation model.⁵¹ In addition, several studies have identified that hip BMD was associated with rs7117858, which is located downstream of the SOX6 gene.^52-54 Furthermore, multiple SNPs located in the SOX6 gene were identified to be associated with hip BMD in both Caucasian and Chinese populations,⁵⁵ which further highlights the importance of the role of SOX6 in influencing BMD variation. Besides, Sox9/Sox6 and Sp1 are involved in the insulin-like growth factor-I-mediated upregulation of human type II collagen gene expression in articular chondrocytes,⁵⁶ which is consistent with our findings that rs16931831 and rs138547759, as regulatory SNPs, target SOX6 through TADs.

Additionally, we detected several novel candidate genes for BMD, such as PLEKHM1, PTPRZ1, CCDC122, and DCLK1. PLEKHM1 encodes pleckstrin homology domain-containing family M member 1, which is suggested to be involved in vesicular transport in osteoclasts.⁵⁷PLEKHM1 gene mutation results in osteopetrosis both in humans and rats, which is characterized by diminished bone resorption by osteoclasts. Mechanistically, loss of PLEKHM1 abrogates the peripheral distribution of lysosomes and bone resorption in osteoclasts through regulating lysosome positioning and secretion through RAB7.⁵⁸ Rptpzeta, which is encoded by PTPRZ1, has been proved to play a physiological role in bone remodelling, and was thus identified as the first protein tyrosine phosphatase (PTP) regulating bone formation in vivo.⁵⁹CCDC122 was identified as a susceptibility gene for leprosy in 3,614 individuals, involving two family-based and three independent case-control samples.⁶⁰ Moreover, a study for exploring the relationship of testicular atrophy to bone metabolism in 31 male leprosy patients and 31 healthy control men was conducted and identified that BMD of the forearm significantly correlated with free testosterone levels (r = 0.689, p < 0.001), which indicates that low BMD may be due to testicular atrophy in leprosy patients.⁶¹ In addition, DCLK1 is upregulated in osteoblast-prone clones compared with non-differentiating clones through transcriptome sequencing and confirmed by real-time quantitative PCR.⁶² rs7776725 targeting PTPRZ1, rs34814687 and rs144518135 targeting PLEKHM1, rs9533094 targeting CCDC122 and DCLK1, rs150081494 targeting CCDC122, and rs191115702 targeting DCLK1 are all regulated through TADs. Further biological studies are warranted to explore the potential roles of PLEKHM1, PTPRZ1, CCDC122, and DCLK1 in the development of BMD.

We identified 12 GO terms enriched in the overlapped candidate genes identified by discovery and replication studies, such as positive regulation of mesenchymal stem cell (MSC) differentiation GO term (GO:2000741) and positive regulation of cartilage development (GO:0061036). Bone-forming osteoblasts are derived from MSCs. In a study of transcriptional profiling of human femoral MSCs in OP, some genetic changes in MSCs were found to be involved in the pathophysiology of OP.⁶³ Research of mouse pathological models and patients with postmenopausal osteoporosis (PMOP) indicated that MEG3 regulates the expression of miR-133a-3p, and inhibits the osteogenic differentiation of bone marrow mesenchymal stem cell (BMSC)-induced PMOP.⁶⁴ It has also been reported that cell-based arthroplasty therapy via the use of MSCs may become one of the strategies for OP treatment.⁶⁵

Another interesting GO term is positive regulation of cartilage development. There is a growing amount of research focus on the relationship between systemic BMD and cartilage properties. For example, Zhu et al⁶⁶ have demonstrated that low BMD, particularly at the hip, was positively associated with knee cartilage defects. In addition, it has been reported that longitudinal BMD loss is related to progressive cartilage loss in knees with osteoarthritis.⁶⁷ Furthermore, a cross-sectional study demonstrated that systemic BMD is positively associated with knee cartilage volume in healthy, asymptomatic adult females.⁶⁸

To the best of our knowledge, this study represents one of the larger studies to explore the roles of regulatory genetic variants in the genetic mechanism of BMD. The main advantage of this study is that rSNPBase 3.1 extends the scope of SNP-related annotations, the regulatory elements to SNP-related regulatory element target gene pairs, therefore it supports SNP-based gene regulatory network analysis. In addition, we further used the replication analysis to validate the discovery results, enhancing the reliability and persuasiveness of our study. However, there are several limitations in our study. First, although data used in our study were from a large sample and we performed a replication analysis, further studies need to be conducted using larger sample sizes, different genetic populations, and different gene variations. Additionally, the biological effect of target genes in different tissue or cell types was not considered in this functional annotation information of rSNP. Our rSNP analysis could not take tissue or cell types into account; therefore, the roles of different tissue and cell types should be warranted in further studies. Moreover, in this analysis, integrative analysis and functional analysis were performed, however some other regulatory elements are not included in the rSNPBase 3.1 database, for instance m⁶A.⁶⁹ Finally, this study focused on rSNP, which occupies a small part of the whole genome.

In conclusion, we identified SNP-related regulatory elements and regulatory element-target gene pairs associated with BMD by integrating rSNPBase 3.1 with both discovery and replication GWAS summary datasets of BMD. We anticipate that the findings of this investigation will bring new insights into the aetiology and treatment of OP. Further research is required to substantiate and elucidate the underlying mechanisms of the identified genes implicated in the development of OP.