Genetic and Gene Expression Resources for Osteoporosis and Bone Biology Research

Abstract

Purpose of Review

The integration of data from multiple genomic assays from humans and non-human model organisms is an effective approach to identify genes involved in skeletal fragility and fracture risk due to osteoporosis and other conditions. This review summarizes genome-wide genetic variation and gene expression data resources relevant to the discovery of genes contributing to skeletal fragility and fracture risk.

Recent Findings

Genome-wide association studies (GWAS) of osteoporosis-related traits are summarized, in addition to gene expression in bone tissues in humans and non-human organisms, with a focus on rodent models related to skeletal fragility and fracture risk. Gene discovery approaches using these genomic data resources are described. We also describe the Musculoskeletal Knowledge Portal (MSKKP) that integrates much of the available genomic data relevant to fracture risk.

Summary

The available genomic resources provide a wealth of knowledge and can be analyzed to identify genes related to fracture risk. Genomic resources that would fill particular scientific gaps are discussed.

Introduction

Osteoporosis-related fractures are common worldwide and are associated with high health care costs and increased risk of disability and mortality [1,2]. In the United States in 2005, there were more than 2 million incident fractures that were estimated to result in a cost of $17 billion [3]. Fractures, particularly hip fractures, are associated with increased risk of disability and mortality [2]. Low bone mineral density (BMD), which is commonly used to diagnose osteoporosis, is associated with fracture in men and women [4,5]. Areal BMD (aBMD) is typically measured using dual-energy x-ray absorptiometry (DXA), and osteoporosis defined by BMD T-score ≤ −2.5 is common, particularly among the elderly [2].

DXA-based BMD assesses bone mass, but whole bone strength is determined not only by bone mass but also bone microarchitecture and several other factors that are collectively referred to as bone quality. While BMD is a strong risk factor for fracture, several factors other than bone mass also contribute to fractures, as the majority of postmenopausal women experiencing fractures do not have osteoporosis defined by BMD [6,7]. Bone microarchitecture and volumetric BMD (vBMD), which can be assessed with high-resolution peripheral quantitative computed tomography (HR-pQCT), contributes to fracture risk independently of aBMD. A meta-analysis that combined evidence from eight studies, resulting in a total of 7254 individuals that experienced 765 incident fractures over a mean follow-up of 4.63 years, found that failure load, trabecular volumetric vBMD, total vBMD, trabecular number, and cortical area were independently associated with incident fracture after adjustment for aBMD or FRAX score [8]. Importantly, 86% of those with incident fracture were not considered to have osteoporosis based on aBMD T-score alone [8]. Combinations of HR-pQCT parameters to create bone phenotypes were also found to be associated with incident major osteoporotic fracture independently of aBMD [9]. Diagnostics and therapies to substantially reduce fracture risk need to address whole bone strength, which includes bone mass, bone morphology, bone microarchitecture, and bone material quality.

Gene identification with genetics and gene expression

Genomics provide an opportunity to identify biological factors that could lead to effective diagnostics and therapies. Developing new therapies is an expensive and lengthy process, so the choice of which drug target to prioritize is critical [10]. Promising drug targets are typically genes whose manipulated activity can impact the trait or disease of interest while having minimal or acceptable off-target effects [11]. To identify such drug targets, there has been an increased reliance on human genome-wide association studies (GWAS) [12–16]. This approach is supported by the finding that drugs with genetic support can have an approximately two times higher chance of FDA approval [17]. In this context, a drug with genetic support means that the target of a drug harbors genetic associations with the same trait as the drug’s lead indication. For example, widely used low-density lipoprotein (LDL)-lowering statin drugs target HMG-CoA reductase (HMGCR), a gene that contains genetic variants associated with LDL levels [11].

The GWAS approach has identified significant genetic associations with thousands of traits, providing target identification information for a wide range of disease areas. The GWAS catalog, a repository of GWAS results, contains approximately 400,000 curated genetic variant trait associations for approximately 3,500 traits as of July 2022 [18]. Osteoporosis-related traits, such as BMD and fracture, have been the subject of large GWAS efforts, resulting in the identification of hundreds of associated genetic variants [19]. While the GWAS approach can discover chromosomal regions that harbor variants causally associated with a trait, translating this knowledge to drug targets requires identifying the gene(s) contributing to the variant association [20]. There are multiple challenges to identifying the genes underlying GWAS association regions. First, trait-associated genetic variants are rarely missense mutations and are often located in non-coding DNA, so the associated variant does not directly alter the protein sequence coded by a gene [21]. Second, the extensive linkage disequilibrium (LD) in the human genome means that neighboring genetic variants are often not independent of each other and that trait-associated genetic variants are often not the causal variant themselves but rather are in LD with causal variant(s) [22,23]. Third, it is often unknown whether a trait-associated genetic variant is also associated with a molecular phenotype (gene expression, chromatin accessibility, DNA methylation, protein levels, etc.) in a tissue type relevant to the trait that would link the associated genetic variant with a gene. Given these challenges, it is not surprising that functional studies have lagged behind the discovery of genetic variant trait associations from GWAS [24].

An expression quantitative trait locus (eQTL) study is a particularly effective approach to identify genes underlying GWAS findings. As gene expression can be considered a quantitative trait, an eQTL study is a GWAS of gene expression. An eQTL study requires gene expression to be measured in a tissue or cell type relevant to the trait or disease of interest and genome-wide genetic variation to be measured in the same individuals. Gene expression from a tissue unrelated to the trait of interest can lead to spurious results due to cross-tissue differences in levels of expression and strengths of eQTL associations [25]. If a genetic variant is associated with a trait and gene expression in a trait-relevant tissue, that gene is a strong candidate to be a causal gene in the GWAS association region [26].

Recognizing that gene expression and eQTL variant associations differ between tissues, the Genotype-Tissue Expression (GTEx) project was established as a large eQTL study performed in 838 individuals with gene expression measured in 49 tissues [27]. The GTEx project is a valuable resource to identify genes affected by trait-associated variants discovered by GWAS, but gene expression in bone tissue was not measured. For skeletal fragility research, it is particularly important to rely on molecular traits measured in bone. However, it is well understood that bone tissue is heterogeneous in terms of cellular content and spatial organization, and osteocytes, osteoblasts, and osteoclasts play distinct roles in bone metabolism to maintain bone mass and structure. With the advent of single cell sequencing technologies, the ability to probe cell-specific and spatial gene expression is becoming possible, which should provide a deeper understanding of the molecular interactions between cells and structures within bone.

Gene expression is useful not only for linking trait-associated genetic variants with genes in an eQTL study, but also for identifying differentially expressed genes (DEGs) related to skeletal fragility in humans and non-human vertebrates. Osteoporosis-related DEGs and gene expression networks discovered in human bone have been used to prioritize genes within GWAS association regions [28,29]. In addition to human bone gene expression, mouse bone gene expression can also be used to prioritize candidate genes related to skeletal fragility in human GWAS [30]. Going in the opposite direction, results from GWAS of osteoporosis-related traits (BMD and fracture) in humans can add value to mouse bone transcriptomic studies by helping to prioritize mouse DEGs for further mechanistic studies [31••].

Recognizing the value of GWAS and gene expression for discovery of fracture-associated genes, this review summarizes publicly-available genomic data resources for GWAS of skeletal fragility-related traits, genome-wide gene expression in bone from humans and rodents, and eQTL studies in human bone. We also describe web-based resources that make these genome-wide results easily searchable and browsable.

GWAS of skeletal fragility traits

Of the bone traits that contribute to whole bone strength, including bone mass, morphology, microarchitecture, and material quality, the vast majority of GWAS to-date have focused on bone mass (DXA-based BMD) and fracture due to the availability of these measures in clinical research studies [19]. The first GWAS of BMD was performed in 2007 using data from the Framingham Study [32]. Since then, GWAS results from multiple studies with a common set of imputed genotypes have been combined using meta-analysis, resulting in large sample sizes and significant genetic associations with consistent genetic associations with BMD and fracture across studies, as reviewed in Zhu et al. [19]. For example, the meta-analysis approach reached a sample size sufficient to achieve reasonable statistical power in a GWAS meta-analysis of fracture, resulting in the identification of 15 genome-wide significant loci (P-value ≤ 5×10⁻⁸) [33].

Over the past few years, GWAS have been performed in the UK Biobank, which has enabled the sample size to reach nearly 500,000 individuals. DXA-based BMD was not available for most of the UK Biobank participants, so to leverage the large sample size, heel quantitative ultrasound was used as an estimate of BMD, i.e., eBMD. GWAS of eBMD was conducted in 426,824 UK Biobank participants, resulting in the identification of 518 genome-wide significant loci and the ability to explain 20% of eBMD variance with genetics [34•].

The genetic architecture of bone morphology and bone microarchitecture is not as well characterized as BMD or fracture, owing to the smaller number of research participants with these traits measured and genome-wide genotypes available. Despite these limitations, genome-wide significant associations have been identified for hip shape derived from statistical shape modeling of DXA scans [35], bone size using bone area from DXA scans [36], hip bone geometry [37], and cortical BMD from pQCT [38].

GWAS publications typically highlight genome-wide significant (P-value ≤ 5×10⁻⁸) variant associations in the main text, tables, or supplemental material. These reported associations are only a small fraction of the approximately 10 million genetic variant associations that are available in a GWAS imputed to modern haplotype reference panels such as 1000 genomes [39] or Haplotype Reference Consortium (HRC) [40]. The full GWAS results for all of the millions of genetic variants tested for trait associations are needed to take advantage of powerful post-GWAS analysis methods, such as two-sample Mendelian Randomization [41], cross-trait genetic correlation [42], GWAS-by-subtraction [43], or an imputed transcript-wide association study [44]. Recently, the GWAS-by-subtraction approach was used with the full GWAS results of fracture, eBMD, and BMD at the femoral neck and lumbar spine to identify genetic variants potentially associated with fracture independently of BMD [45•]. The GWAS-by-subtraction approach did not confirm any risk loci harboring non-BMD effects on fracture risk, suggesting that either DXA BMD genetics explains most fracture associations or there is a limitation to the computational approach. Moreover, full genome-wide GWAS results allow researchers to look-up genetic associations for particular candidate genes. It has been recognized that there are likely true positive associations that do not reach the strict genome-wide significance threshold but that could provide useful information in a candidate gene look-up motivated by independent prior knowledge [46]. We summarize large GWAS meta-analyses of osteoporosis-related traits and bone size with full genome-wide results publicly available (Table 1).

Table 1.

Osteoporosis-related GWAS, eQTL, and bone gene expression open data resources.

First author, date	PMID	Trait	Study type	Access	Ref
Estrada, 2012	22504420	BMD (FN, LS)	GWAS	gefos.org	[102]
Zheng, 2015	26367794	BMD (FN, LS, FA)	GWAS	gefos.org	[103]
Medina-Gomez, 2017	28743860	Pediatric TB BMD, lean mass	Bivariate GWAS	gefos.org	[104]
Kemp, 2017	28869591	eBMD	GWAS	gefos.org	[105]
Medina-Gomez, 2018	29304378	age-specific TB BMD	GWAS	gefos.org	[106]
Trajanoska, 2018	30158200	fracture	GWAS	gefos.org	[33]
Morris, 2019	30598549	eBMD and fracture	GWAS	gefos.org	[34•]
Styrkarsdottir, 2019	31053729	Bone size	GWAS	decode.com/summarydata	[36]
Trajanoska, 2020	32999390	Fall history	GWAS	gefos.org	[107]
Grgic, 2021	34754074	Skeletal age	GWAS	gefos.org	[108]
Mullin, 2020	32216834	Osteoclast GEX	eQTL	gefos.org	[50]
Reppe, 2010	19922823	BMD	GEX	ArrayExpress: E-MEXP-1618	[52]
Farr, 2015	26402159	Age, Estrogen	GEX	GEO: GSE72815	[54]
Weivoda, 2020	31911667	DMAb treatment	GEX	GEO: GSE141614, GSE72815	[55]

BMD = bone mineral density, FN = femoral neck, LS = lumbar spine, FA = forearm, TB = total body, eBMD = estimated BMD by quantitative heel ultrasound, GEX = gene expression.

Gene expression and eQTL studies in bone in humans

Genes underlying GWAS association regions can be identified with eQTL studies that link trait-associated genetic variants with gene expression. This is a key step in bridging genetic variants from GWAS results to genes. At least partly due to the technical difficulty of isolating RNA from mineralized tissue, there are only two eQTL studies analyzed gene expression in bone tissue. The earliest bone eQTL study measured gene expression using microarrays from cultured human osteoblasts obtained through explant outgrowth from 95 donors undergoing hip or knee replacement surgery [47]. While the cultured cells exhibited osteoblastic features, the cell source was not able to be unambiguously defined from the donors [47]. Unfortunately, the full eQTL results are not available from this study, limiting its utility. The second eQTL study’s genome-wide results are publicly available, but the gene expression was not measured in primary bone cells [48–50]. Instead, this study was conducted using gene expression measured in osteoclast-like cells differentiated from peripheral blood mononuclear cells (PBMCs) obtained from 158 women. The subjects were aged 30–70 years and all had self-reported European ancestry. Quantification of gene expression in the osteoclast-like cells was performed using RNA sequencing (RNAseq) and genotypes were imputed to the HRC reference panel. Full genome-wide results are available from this eQTL study, enabling the results to be fully leveraged (Table 1). One caveat to this study is whether gene expression in osteoclasts differentiated in culture from PBMCs is representative of gene expression in osteoclasts in vivo.

In addition to eQTL studies, bone gene expression without DNA genotyping can still be very effective in the identification of genes whose expression is associated with skeletal fragility. Since skeletal fragility DEGs have a molecular function (gene expression) associated with a bone trait, these genes are excellent candidates to be causal genes within a GWAS association region. Progress in gene expression analysis in human bone tissue has been reviewed recently [51]. Highlighted here are studies with larger sample sizes, as well as publicly available raw data and processed results from microarray and RNAseq.

Two publications reported results from the analysis of microarray gene expression from iliac crest bone biopsies. In a study of 84 postmenopausal females, eight genes, which included well-established skeletal fragility related genes SOST and DKK1, were significantly associated with total hip BMD [52]. A reanalysis of this same microarray expression data categorized the 84 female participants into an osteoporotic group (BMD T-score ≤ −2.5 and ≥ 1 fracture) and controls (BMD T-score > −1 and no fractures), and identified 256 genes associated with osteoporotic fracture independently of age and BMI [53].

As opposed to previous studies that used RNA microarrays, subsequent studies used RNAseq to measure gene expression from iliac crest bone biopsies. An RNAseq study of bone biopsies obtained from 58 women identified DEGs associated with age and estrogen therapy [54]. This RNAseq data from 58 women were combined with RNAseq data from iliac crest bone biopsies collected in an intervention study using denosumab (DMAb) to remove osteoclasts, which identified osteoclast-secreted factors suppressed by DMAb [55]. DEGs identified from these human bone gene expression studies can potentially be used to prioritize candidate genes identified from other gene discovery approaches.

Given the importance of obtaining gene expression from tissues relevant to the disease of interest in eQTL studies [25], there is a critical need for robust eQTL studies of primary bone tissue from diverse populations.

Gene expression studies of skeletal fragility in rodent models

The prevailing method for transcriptomic studies that yield gene expression information is bulk RNA sequencing (RNAseq), which has become increasingly cost-effective, leading to an abundance of data in this field. In an effort to characterize heterogeneous cell populations and their transcriptional profiles, single-cell RNA sequencing (scRNAseq) has gained traction in musculoskeletal (MSK) and bone research. Lastly, spatial transcriptomics has emerged as a promising tool in recent years which provides both cell-based gene expression information and spatial localization of gene expression. Using these transcriptomic methods in rodent models, specifically in mouse models, has yielded important discoveries and identified novel genes and pathways in bone development, remodeling, and diseases [31••,56••, 57–61]. While small fish models, notably zebrafish, have also contributed to advancing musculoskeletal research [62–65], this review concentrates on resources provided by rodent models.

Recent publications have provided comprehensive summaries for each sequencing technique; however, it is important to acknowledge that these techniques still come with their individual challenges [66–70]. Bulk RNA sequencing provides valuable insights into gene expression profiles but lacks the ability to distinguish among specific cell types or to spatially localize gene expression within a tissue. In contrast, single-cell RNA sequencing offers cell-based information and more precise gene expression data, but it currently lacks spatial information regarding where genes are expressed within the tissue. Furthermore, it is still expensive to implement. Spatial transcriptomics addresses certain limitations of these technologies; however, it is not yet able to achieve single-cell resolution and has lower read depths compared to scRNAseq.

Transcriptomics in MSK research face additional technical challenges when investigating bone tissue due to its complex structure and extracellular matrix composition. The abundant extracellular matrix tends to favor the profiling of superficial bone cells rather than deeply-embedded osteocytes. Within bone cells, extraction and isolation of osteocytes from cortical bone are complicated and require specific expertise [71]. As a result, many single cell and spatial omics studies in MSK research have primarily focused on bone marrow cells [67,72]. Profiling osteocytes remains challenging despite their crucial role in regulating osteoblasts and osteoclasts, bone remodeling, mechanical load sensing, mineral homeostasis, and perilacunar/canalicular remodeling [73–79]. Table 2 highlights recent transcriptomic studies aimed at identifying gene candidates associated with specific interventions in rodent models such as osteogenesis imperfecta, the anabolic response of bone to mechanical load, the effect of age on bone, and others that target bone cells, while excluding the influence of the bone marrow component.

Table 2.

Rodent bone gene expression data resources.

First author, date	Method	Tissue	Intervention	Raw Data access	Result access	Ref
Mantila Roosa, 2011	Exon array	Ulnae	Mechanical loading	No	DEGs	[82]
Ayturk, 2013	RNAseq	Tibiae	Altering Lrp5 Mutations	No	Full	[109]
Galea, 2017	RNAseq	Tibiae	Aging and mechanical loading	No	Full	[84]
Zimmerman, 2019	RNAseq	Femurs, tibiae	Osteogenesis Imperfecta CrtapKO and oim/oim	No	DEGs	[61]
Chermside-Scabbo, 2020	RNAseq	Tibiae	Aging and mechanical loading	No	Full	[85]
Ayturk, 2020	scRNAseq	Calvaria, femurs, tibiae	Sclerostin-neutralizing antibody injection	No	Full	[88]
Moffatt, 2021	RNAseq	Calvaria	Osteogenesis Imperfecta Jrt and oim	No	Count	[80]
Vrahnas, 2019	RNAseq	Femurs	Brittle bone - EphrinB2 deletion in osteocytes	GSE110795	Top DEGs	[81]
Wang, 2020	RNAseq	Calvaria, tibiae	None	GSE151971	No	[87]
Spatz, 2021	RNAseq	Hindlimb	Unloading	GSE169292	DEGs	[86]
Wang, 2021	scRNAseq	Calvaria, femurs, tibiae	Absence of Sp7 in osteoblasts and osteocytes	GSE154719	Full	[57]
Youlten, 2021	RNAseq	Calvaria, humeri, femurs, tibiae	Aging and sex	E-MTAB-5532, 7447, 5533	Full	[56••]
McDonald, 2021	scRNAseq	Long bones	Osteomorphs and osteoclasts distinction	PRJNA507938	DEGs	[91]
Kaya, 2022	RNAseq	Humeri	Aging	PRJNA695408	Full	[31••]
Chlebek, 2022	RNAseq	Tibiae	Mechanical loading	GSE210827	No	[83]
Hanai, 2023	scRNAseq	Femurs	Calcitriol (1,25-dihydroxyvitamin D3) injection	GSE220836	No	[89]
Agoro, 2023	RNAseq scRNAseq,	Femurs, tibiae, humeri	Chronic kidney disease Sost-Cre and sex	GSE205792 GSE208152	Full	[58]

Raw Data Access as FASTQ files:

1. For GSE #: visit https://www.ncbi.nlm.nih.gov/geo/

2. For E-MTAB #: visit https://www.ebi.ac.uk/biostudies/

3. For PRJ #: visit https://www.ncbi.nlm.nih.gov/bioproject/

Result Access in the supplemental materials section:

4. No = No results available

5. Full = Full genome-wide gene expression results available.

6. Count = Raw counts table available.

7. DEGs = Differentially expressed gene results available.

8. Top DEGs = Top differentially expressed gene results available.

Several bulk RNA sequencing studies have made significant contributions to our understanding of bone cells and related interventions. One area of focus has been brittle bone conditions, specifically osteogenesis imperfecta (OI), which is characterized by bone fragility due to disruptions in collagen type I production, processing, or function, including in the control of signaling. Comparative studies investigating osteocyte transcriptomes across different OI mouse models, including those utilizing oim (Col1a2 mutation) and jrt (Col1a1 mutation) mice [80] or examining Crtap knockout and oim/oim mice, both harboring distinct collagen I mutations [61], have revealed dysregulated genes related to Wnt and TGFβ signaling pathways, providing new insights into potential therapeutic targets. Additionally, another brittle bone model, osteocyte-targeted EphrinB2 deletion has shed light on the impact of Ephrin B2 deficiency on mineral and matrix composition, identifying dysregulated autophagy genes [81].

Transcriptomic studies have provided key insights into mechanisms responsible the effects of physical activity on bone adaptation, remodeling, and mass, using in vivo loading studies of mice. Microarray analysis in mechanically loaded bones has revealed distinct gene expression patterns, implicating the involvement of Wnt/β-catenin and TGFβ signaling pathways [82]. Axial loading experiments in different regions of cortical bone have identified spatially unique transcriptomic responses, with the mid-diaphysis region showing the greatest number of differentially expressed genes [83]. Furthermore, RNA sequencing of osteocyte-enriched cells from femur, tibia, and humerus has demonstrated similarities in the osteocyte-enriched transcriptome among bones, with a small number of differentially expressed genes (27 DEGs) identified between skeletal sites [56••].

Age-related changes in bone mass, architecture, osteocyte density, and mechanical properties have been investigated using mouse models. Transcriptomic analysis targeting osteocytes in young and aged female mice subjected to axial tibial loading has revealed rapid and sustained transcriptomic responses, with DEGs related to proliferation and bioenergetics between age groups [84]. Similarly, age-related differences in response to more prolonged post-axial loading have been observed, with diminished Wnt signaling, extracellular matrix, and neural responses in aged mice compared to young mice [85]. Bulk RNA sequencing of wild-type osteocyte-enriched humerus samples at multiple time points has highlighted changes in pathways associated with axon guidance, focal adhesions, and ECM-receptors during aging [31••]. Additionally, sex differences in the osteocyte transcriptome have been identified starting at 16 weeks, as well as the specific expression of 1239 osteocyte signature genes, including upregulation of Sost and Mepe with age [56••].

Unbiased transcriptomic analysis of hindlimb unloaded mice has uncovered novel genes, metabolic processes and Wnt signaling involved in bone’s response to altered mechanical loading [86]. A cross-species RNAseq study comparing cranial and tibial osteocytes of mice, rats, and rhesus macaques have provided valuable insights into site-specific differences and conserved gene expression patterns, highlighting upregulated genes associated with bone growth and remodeling and the involvement of the Wnt signaling pathway [87].

Several scRNAseq studies have elucidated the heterogeneity of bone cells and their roles in skeletal development and disease. Ayturk et al. [88] examined osteoblasts in neonatal mouse calvarial cells, revealing differences in cell type abundance and transcriptomes between freshly isolated and cultured cells. Wang et al. [57] identified transcription factor Sp7 as essential for osteocyte dendrite development and demonstrated defective osteocyte maturation in Sp7 deletion. Hanai et al. [89] studied FGF23-expressing osteocytes and their response to 1,25-dihydroxyvitamin D3, identifying subpopulations of osteocytes expressing Fgf23 and their sensitivity to 1,25-dihydroxyvitamin D3 for phosphate metabolism. In a mouse model of chronic kidney disease, scRNAseq was performed on fluorescently-labeled osteoblasts/osteocytes isolated from long bones, revealing distinct populations of osteoblasts/osteocytes. The cluster associated with osteocytes exhibited the highest expressions of Phex and Dmp1 [58]. Moreover, beyond the exploration of osteocytes, investigation of the development of the coronal suture in mice identified distinct pre-osteoblast signatures, ligament-like populations, and chondrogenic-like populations [90]. Also, an intriguing scRNAseq analysis revealed the existence of a unique cell type called osteomorphs, which actively involved in bone resorption [91]. Osteomorphs exhibited distinct transcriptional profiles that distinguished them from osteoclasts and macrophages. Importantly, several osteomorph genes are implicated in monogenic skeletal disorders, and some of these genes are also associated with eBMD, suggesting their potential involvement in the development of osteoporosis [91]. Therefore, scRNAseq approach will certainly provide valuable insights as it is more widely adapted and standardized for its application in MSK and bone research.

Despite its limitations, MSK gene expression studies have extraordinary potential to advance our knowledge of mechanisms involved in development, homeostasis, degeneration, and disease of bone and other musculoskeletal tissues. To fully leverage the power of MSK genomics and facilitate further research, there is a pressing need for consensus approaches to consolidate these studies and to present the results in a user-friendly and easily navigable manner. Although challenging, the consolidation of multi-modal genomic data (single cell and bulk gene expression, RNA splicing, chromatin accessibility, DNA methylation, protein abundance, and genetic associations with these molecular traits) in humans and mice in a user-friendly manner presents a great opportunity for MSK research.

Integrative and browsable resources

Translational research that bridges clinical and laboratory science requires information sharing in a form that is understandable to scientists across both fields. Until recently, there have been significant barriers preventing lab researchers from asking questions as simple as, “is my favorite mouse gene also associated with osteoporosis and fracture in humans?” Likewise, it has also been challenging for human geneticists to narrow down the potentially long list of genes within a GWAS association region or genetic linkage locus to a shorter list of genes functionally related to bone traits by gene expression or laboratory-based mouse experiments. We highlight two resources, MSKKP and mouse2human, that address these needs in the scientific community.

Accessing and interpreting human GWAS results has been challenging until recent resources have been established. Imputed GWAS results with approximately 10 million rows of data cannot be easily opened, browsed, and searched without computer programming skills. Interpretation of GWAS results by laboratory scientists is challenging as well, due to the fact that genetic variant identifiers from dbSNP are accession numbers that require linkage to other NIH databases to retrieve chromosomal physical position. Even after a genetic variant’s physical location is determined, the gene to which it corresponds is not always obvious. Finally, in a single human gene region, there can be thousands of genetic variants with trait associations from a human GWAS. For laboratory researchers, it is not obvious what the P-value significance threshold should be for a single gene with a large set of genetic variants, many of which are correlated due to linkage disequilibrium (LD). The number of genetic variants, the extent of LD, and gene size are all gene-specific and require gene-specific summaries of thousands of genetic variant associations to provide information regarding whether a gene is associated with osteoporosis. While genetic variants associated at the genome-wide significance level (P-value ≤ 5×10⁻⁸) clearly pass multiple testing in a single gene region, this significance threshold is too stringent when examining a single gene.

Until recently, accessing GWAS results has required consortia to host their own GWAS results on custom websites. For skeletal traits, gefos.org and decode.com have hosted GWAS results (Table 1). This burden has largely been removed now that the GWAS catalog has expanded their service to host full genome-wide GWAS results [18]. Hosting GWAS result files provides access to the public, but the issue of interpretation and searching large GWAS results for candidate genes remains a barrier.

The Musculoskeletal Knowledge Portal (MSKKP, http://msk.hugeamp.org/) addresses the issue of browsing and interpreting MSK GWAS results by providing an easy-to-use display similar to a genome browser [92••,93]. This allows users to easily search for a gene of interest to observe genetic trait associations in their physical positions along a chromosomal region. For some traits, there have been multiple GWAS publications in which some of the same research participants enrolled in clinical research studies were used across multiple publications. To combine and summarize genetic trait associations across multiple studies of the same trait while accounting for sample overlap, the MSKKP displays a “bottom-line” genetic trait association for all genome-wide genetic variants. The “bottom-line” results for genetic variants in a gene region are summarized to a single gene score using the default parameters and a 50 kb window size using MAGMA software [94]. MAGMA’s default gene-based score is the SNP-wise mean model, which calculates the mean test-statistic of variant associations in a gene region, and uses numerical integration to estimate the gene-based P-value. This method resorts to an empirical permutation method only when numerical integration fails. The statistical power of the SNP-wise mean model is higher for genes with significant variant associations within a region of high LD. However, statistical power is lower for genes with only a small proportion of variants that are significant, as is the case with rare variants that are typically not in LD with a large number of variants.

One of the advantages of the MSKKP is that it is part of the larger Knowledge Portal Network (KPN) of the Human Genetics Amplifier (HuGeAMP) software platform. The KPN provides genome browsers for GWAS traits for several different disease areas, such as common metabolic disease, cardiovascular disease, and lung disease. As such, the MSKKP can leverage this information to display how an MSK-associated genetic variant is associated with other traits and diseases curated by the KPN, thus providing a wider context of association for a genetic variant.

The MSKKP also provides additional genomic layers of data viewable alongside GWAS results. On the gene-based page of MSKKP, in addition to common and rare variant GWAS results, users can also find gene expression levels from different human tissues, even bone. Gene expression values from one of the human bone gene expression datasets highlighted in Table 1 [54] are shown graphically, but values from the study’s experimental contrasts (young, old, and estrogen therapy) are grouped together. With this information, users can determine whether a gene of interest is expressed in bone, but not whether the gene is differentially expressed relative to experimental groups, which could provide important biological evidence for a gene’s role in skeletal fragility. The MSKKP also provides lists of pre-computed predicted effector genes for different GWAS. Predicted effector traits combine multiple genomic assays to prioritize candidate genes within GWAS association regions [95].

While the MSKKP hosts more genomic resources, www.mouse2human.org (mouse2human) [31••] is a web-based tool that illustrates a strategy to integrate human and mouse genome-wide data. Mouse2human displays gene-based scores calculated using MAGMA from GWAS of eBMD and fracture, with a few key differences from MSKPP. First, rather than using bottom-line GWAS results that are combined across multiple studies of the same trait, mouse2human.org has chosen results from specific GWAS, namely, eBMD and fracture from the UK Biobank [34•]. Second, a different gene-based score calculation from MAGMA was used. While the MSKKP used the default SNP-wise mean model, mouse2human chose the SNP-wise top model, which tests whether the most significant SNP trait association in a gene region is significant using an adaptive permutation approach. The SNP-wise top model can detect genetic associations without a high degree of LD at the most significant SNP, which is consistent with what could occur with rare SNPs and even sometimes common SNPs. The permutation approach for the SNP-wise top model generates a null distribution that is specific to the size, number of SNPs, and extent of LD of each gene region using 1000 Genomes reference genotypes in the region along with a randomly generated phenotype variable [94]. In this way, these gene-based scores correct for the number of variants in a gene region, not the entire genome, which avoids the stringent genome-wide significance threshold. Moreover, there is a batch query tab on the mouse2human website where multiple gene-based scores can be queried with multiple test correction applied. In addition to the different type of gene-based scores, mouse2human provides the mouse homolog for each human gene, which includes paralogs, along with links to the corresponding mouse or human NCBI gene entry. Recent updates to mouse2human include gene-based associations for human osteoarthritis at four different sites (hip, knee, hip and knee, and any site) using UK Biobank and arcOGEN GWAS [96].

Conclusion and Future Directions

A wealth of genomic data can be used to drive discovery of novel genes related to skeletal fragility, and current integrative and navigable web resources like the MSKKP and mouse2human make genomic findings accessible to the wider MSK scientific community of clinical researchers and laboratory scientists. As additional genomic studies related to bone and skeletal fragility emerge, integrating results from different genomic assays will provide multiple lines of evidence to support a gene’s role in bone biology and fracture risk. Of the many future directions available in the expanding field of MSK genomics, we identify four future directions that would impact translational science and integrative MSK genomic resources.

1. Facilitate discovery of gene expression resources.

While there are excellent repositories of raw and processed gene expression data (Sequence Read Archive [SRA]: https://www.ncbi.nlm.nih.gov/sra/ and the Gene Expression Omnibus [GEO]: https://www.ncbi.nlm.nih.gov/geo/), the entries are not always easy to discover, and the results are not easy to browse and navigate. Full genome-wide results of fold changes representing differential gene expression between experimental groups can sometimes be found at GEO, and sometimes as publication supplemental files. The Gene Expression Database (GXD, https://www.informatics.jax.org/expression.shtml) provides an easy-to-use interface to help the scientific community discover mouse gene expression datasets that match a question of interest, allowing users to search for mouse SRA and GEO listings from RNAseq or microarray studies [97•]. Search filters based on anatomical structure, developmental stage, and sex are available. Using the GXD, we were able to find 11 studies of gene expression collected from bone tissue without bone marrow with GEO/SRA entries among the 1514 postnatal RNAseq studies. Data deposition into GXD is voluntary, so some mouse bone RNAseq studies are not listed in the GXD. We recommend that future mouse gene expression studies of bone should deposit data into the GXD to facilitate discovery by the scientific community. GXD facilitates discovery by using gene expression data within the Mouse Genome Informatics (MGI, https://www.informatics.jax.org/) platform developed by the Jackson Laboratory. The MGI website integrates gene expression data with a repository of genomic and other biological data that related to clinically relevant outcomes. Moreover, while considerable resources are readily available for mice, a diverse array of genomic resources is available for other animal models. For instance, the Zebrafish Information Network (ZFIN, https://zfin.org/) serves as a primary hub for the zebrafish model [98]. The Drosophila RNAi Screening Center and Transgenic RNAi Project provides resources for evolutionary comparisons across vertebrate and invertebrate models. Its original focus was on tools for Drosophila research, but this repository has expanded to include RNAseq/scRNAseq resources for various tissues in mice, humans, zebrafish, and flies (https://fgr.hms.harvard.edu/tools) [99]. Currently, MSK tissues at this resource are limited to bone marrow in humans.

2. Integration of multiple genomic results in MSKKP.

Raw gene expression data can be found at the GXD, GEO, and SRA, but differential gene expression results are not typically provided at these resources. The deep MSKKP resources include gene expression to the extent that users can visualize whether a particular gene is expressed in human bone and other tissues, but the visualizations do not leverage the study designs from which the data is drawn. For instance, the MSKKP displays bone gene expression from Farr et al., 2015 (Table 1), but does not display the gene expression fold changes reported for age or estrogen from the study. Displaying not only expression values but also fold changes and P-values for experimental contrasts would inform users that a gene might be expressed in bone and also functionally related to skeletal fragility traits. The MSKKP currently visualizes multiple types of omics, and combining differential gene expression results with GWAS results can help to provide multiple lines of independent evidence supporting a gene’s role in musculoskeletal traits and conditions. Along these lines, there is a need to measure other molecular phenotypes, such as chromatin accessibility, DNA methylation, and protein expression, in MSK tissues to facilitate discovery. There are few such datasets for MSK tissues, but it is notable that the MSKKP hosts DNA methylation QTLs (mQTLs) in cartilage and synovium [100], along with eQTLs in cartilage and synovium, protein QTLs (pQTLs) in cartilage, and differential gene expression and protein abundance in cartilage [101]. Moreover, the availability of deep coverage of germline genetic variation with whole-exome and whole genome DNA sequencing is continuing to expand in clinical research studies of MSK traits, which will certainly contribute to the identification of novel genetic associations with molecular and clinical traits. We expect that integrating new genomic results into the MSKKP will enhance the utility of these valuable datasets.

3. Conduct GWAS for additional skeletal fragility traits.

BMD does not tell the whole story of skeletal fragility, especially in the context of aging, diabetes, and glucocorticoid use. Collaborative GWAS meta-analyses have led to a well-developed characterization of the genetic architecture of osteoporosis, i.e., BMD and fracture. However, there have been relatively fewer GWAS of bone microarchitecture from HR-pQCT. While the sample size of GWAS with HR-pQCT will be smaller than those with more commonly-measured osteoporosis-related traits, GWAS results for bone microarchitecture can be used to distinguish genetic variants and genes associated with microarchitecture, BMD, and fracture risk.

4. Produce robust and publicly available eQTL resources for bone.

Finally, there is a need for eQTL studies of primary bone tissue from diverse populations. This should be a top funding priority that is needed to provide the critical missing link between GWAS and cellular and molecular mechanisms. Also, to facilitate reliable and reproducible gene expression measurements in MSK tissues, there is a need for robust and publicly available protocols that overcome unique challenges presented by MSK tissues for spatial and single cell analysis.

Data availability.

Not applicable.