The Musculoskeletal Knowledge Portal: Making Omics Data Useful to the Broader Scientific Community

Abstract

The development of high-throughput genotyping technologies and large biobank collections, complemented with rapid methodological advances in statistical genetics, has enabled hypothesis-free genome-wide association studies (GWAS), which have identified hundreds of genetic variants across many loci associated with musculoskeletal conditions. Similarly, basic scientists have valuable molecular cellular and animal data based on musculoskeletal disease that would be enhanced by being able to determine the human translation of their findings. By further synthesizing these large scale human genomic musculoskeletal datasets with complementary evidence from model organisms, new and existing genetic loci can be statistically fine-mapped to plausibly causal variants, candidate genes and biological pathways. Genes and pathways identified using this approach can be further prioritized as drug targets including side-effect profiling and the potential for new indications. To bring together these big data, and to realize the vision of creating a knowledge portal, the International Federation of Musculoskeletal Research Societies (IFMRS) established a working group to collaborate with scientists from the Broad Institute to create the Musculoskeletal Knowledge Portal (MSK-KP) that would consolidate -omics datasets from humans, cellular experiments, and model organisms into a central repository that can be accessed by researchers. The vision of the MSK-KP is to enable better understanding of the biological mechanisms underlying musculoskeletal disease and apply this knowledge to identify and develop new disease interventions.

Introduction

Musculoskeletal diseases are extremely common and contribute to pain, disability, and death. In the United States, musculoskeletal disease represents a significant public health burden that affects an estimated one in two individuals, equivalent to the prevalence of cardiovascular and pulmonary conditions combined.¹ Aggregate incremental costs (direct and indirect) attributed to musculoskeletal disease are estimated at $321.6 billion per year, ² and will increase with the aging of the world’s population unless more effective interventions are developed.³ Musculoskeletal diseases are the leading cause of morbidity in the workforce and the most common cause of early retirement of patients and burden for their families.⁴ Musculoskeletal conditions represent a global threat to healthy aging.⁵.

Twin and family studies have yielded robust evidence about the genetic contribution underlying musculoskeletal diseases.⁶ Therefore, genetic epidemiological approaches are a logical strategy to better understand the biological underpinnings of musculoskeletal disease onset, and pathogenesis. The development of high-throughput genotyping technologies and large biobank collections, complemented with rapid methodological advances in statistical genetics, has enabled hypothesis-free genome-wide association studies (GWAS), which have identified hundreds of genetic variants across many loci associated with musculoskeletal conditions such as fracture,^7,8 osteoarthritis,⁹ and sarcopenia, and their related endophenotypes including bone mineral density,^10-13 lean mass,¹⁴ grip strength^15,16 and even back pain.¹⁷ As large- scale whole genome (WGS) and exome sequencing data become available, even more variants will be identified.¹⁸ These studies provide a fertile ground for understanding the biologic pathways underlying the genetic architecture of musculoskeletal traits. Furthermore, summary association results produced by GWAS offer a powerful and proven means of identifying biological targets relevant to human disease pathophysiology that may be amenable to pharmacotherapeutic intervention.¹⁹ Indeed, identification of disease-associated variation in the gene coding for a drug target protein, is currently the single best-known predictor of whether a compound acting on that target will survive phase III testing and subsequently be approved for clinical use.²⁰ Moreover, when genome-wide significant variants are followed up in large-scale phenome-wide association studies (PheWAS) that encompass thousands of traits and diseases, important information can be garnered regarding the likelihood that a pharmacological intervention affecting the desired pathway/target will produce unintended side effects²¹ and/or pinpoint new therapeutic indications (drug repurposing).²² Consequently, the results of GWAS are increasingly being used by the pharmaceutical industry as an effective means of prioritizing compounds for development as well as for repurposing existing medications for new indications.²⁰ Although investigators are able to download and use these data for new research discoveries, it can be a daunting task to curate and integrate such data with bioinformatics resources that can move musculoskeletal genetics into the “post GWAS” era.

To date the central challenge facing GWAS (and consequently a better understanding of disease pathogenesis and identification of new drug targets) is how to reliably map genome-wide significant variants to their causal genes and/or biological pathways. Because of the correlated structure of the genome, a phenomenon referred to as linkage disequilibrium (LD),²³ each GWAS associated variant will often have hundreds of other variants in LD with it. All of these variants that span large regions of the genome and contain many genes that also may be significantly associated with the trait. In addition to the challenges in disentangling LD and functional variants, data mapping the three dimensional chromosomal structure in the form of topological associated domains (TAD’s) has made it possible to identify self-interacting TADs that physically interact with each other more frequently than with those outside the TAD.²⁴ Variants in these regions harbor important elements of the gene regulation landscape, including long range interactions that cannot be characterized using physical distance alone.²⁵ Thus, in summary there are growing collections of important GWAS data and supporting information about variant function that are available for application to the field of musculoskeletal genomics, yet the integration of these data is challenging. Similarly, basic scientists have valuable molecular cellular and animal data based on musculoskeletal disease that would be enhanced by being able to determine the human translation of their findings.

One way to address this challenge is to establish a dedicated “knowledge portal” that consolidates the summary association results from GWAS and integrates this information with functional data from gene expression, proteomic, metabolomic and epigenetic studies. Data coming from projects such as the ENCODE (https://www.encodeproject.org/),²⁶ Roadmap Epigenomics project (http://www.roadmapepigenomics.org/)²⁷ and GTeX (https://www.gtexportal.org/home/) projects²⁸ all provide essential data on regulatory function of non-coding elements in the genome across multiple human tissues. Nevertheless, across ENCODE, the ROADMAP Epigenomics project and GTEx, there is limited musculoskeletal representation, confined to one osteoblast-derived cell line with epigenetic data. For basic and translational scientists to make better use of the wealth of genomic data emerging from consortia such as the Genetic Factors for Osteoporosis (“GEFOS”), Cohorts for Heart and Aging Research in Genomic Epidemiology (“CHARGE”), Trans-omics for Precision Medicine (“TOPMed”), Genetics of Osteoarthritis (“GO”), and the UK Biobank, a one-stop, integrated knowledge base that creates a functional interface between the GWAS results and the supporting musculoskeletal datasets is needed. The datasets are available because the musculoskeletal community has been generating carefully analyzed and validated results over the past decade.

By further synthesizing these integrated musculoskeletal datasets with complementary evidence from model organisms, new and existing loci can be statistically fine-mapped to plausibly causal variants, candidate genes and biological pathways. Genes and pathways identified using this approach can be further prioritized as drug targets including side-effect profiling and the potential for new indications.

Establishing the Musculoskeletal Knowledge Portal

As a first step towards achieving the goal of creating a knowledge portal, the International Federation of Musculoskeletal Research Societies (IFMRS) established a “Big Data” working group, with representation of all its member societies. The IFMRS mission is to advance musculoskeletal research globally in order to prevent and treat musculoskeletal diseases by collaboration of international societies to share resources, raise public awareness and provide education. The Big Data working group has been meeting for the past five years and have created an inventory of data that may be useful for musculoskeletal research (http://www.ifmrs.org/publications/big-data/). However, this effort served to catalogue existing data, but did not create an integrated resource to translate association results to biologic mechanisms.

Recognizing these limitations, the Big Data working group set out to develop a “Musculoskeletal Knowledge Portal” (MSK-KP) modeled on a successful “Type-2 Diabetes Knowledge Portal” (http://www.type2diabetesgenetics.org/) developed to facilitate functional studies of genetic factors underpinning Type-2 diabetes (T2D) associated loci. Similar to the T2D knowledge portal, the MSK-KP would be designed to support experimentalists, rather than to serve simply as a repository for musculoskeletal GWAS datasets generated by individual laboratories and large consortia.

To begin the process of establishing the MSK-KP, an additional working group dedicated to the knowledge portal was formed to represent the international musculoskeletal research community within the IFMRS. The working group agreed that the ultimate vision of the MSK-KP was to consolidate -omics datasets from humans, cellular experiments, and model organisms into a central repository that can be accessed by researchers. The vision of the MSK-KP is to enable better understanding of the biological mechanisms underlying musculoskeletal disease and apply this knowledge to identify and develop new disease interventions. To realize this vision, the working group collaborated with the team from the Broad Institute of Harvard & the Massachusetts Institute of Technology, who designed the software and data platform that powers the T2D Knowledge Portal. Together they identified several objectives necessary to realize the vision of MSK-KP, which included: (i) identifying, obtaining, curating and integrating various -omic datasets from the international musculoskeletal research community; (ii) hosting a unified omics data resource on a web server that is publicly available to the scientific community; (iii) develop an intuitive and flexible web interface that enables non-specialist users to mine omics data easily; (iv) promote the use of the knowledge portal, encourage data sharing and necessitate a greater community collaborative spirit. The MSK-KP went live in March, 2020 with the beginnings of these ambitious goals.

Plans to achieve the objectives of the Musculoskeletal Knowledge Portal

The setup of the MSK-KP, builds upon multidisciplinary international collaboration, comprising the collection and integration of Big Data from different expertise fields of MSK research. Figure 2 displays the envisioned flow of information, analysis and synthesis of data within the MSK-KP.

Figure 2:

A roadmap underlying the MSK-KP flow of data (Image provided by the GEMSTONE COST ACTION CA18139)

The envisioned approach encompasses several steps. The first challenge is to take the thousands of summary association results from GWAS, as well as whole-genome and exome sequencing,¹⁸ and apply bioinformatics and statistical processing to narrow the number of genes and variants in these genes that may be causally related to pertinent musculoskeletal disease and their endophenotypes. The next step is to capitalize on cellular, tissue, and organism knock-outs or models of editing genes/variants to identify and understand how newly identified genes of significance contribute to disease pathogenesis. Ultimately to translate this to the clinic, the long-term goal is to identify and prioritize causal genes and/or pathways as potential drug targets for new therapeutic development, risk stratification, and molecular disease redefinition, also using artificial intelligence.²⁹ The need for new therapeutics and/or tailored therapies for musculoskeletal disease is underscored by several hard facts: 1) the osteoporosis drug pipeline has largely dried up; 2) there are currently no disease modifying drugs to treat osteoarthritis; 3) similarly, there are no pharmacological treatments for age-related muscle loss, or “sarcopenia”

Osteoporosis, the first target area for the MSK-KP

Osteoporosis represents one of the promising musculoskeletal diseases in which the MSK-KP can realize its vision and leverage information various human and non-human omic datasets to better understand disease pathophysiology and identify and prioritize putative drug targets. This is because:

1. Bone mineral density (BMD) is a highly heritable polygenic trait (h² >0.4)⁶ that is the strongest known risk factor for osteoporosis and predictor of low trauma fracture.³⁰ Therefore, inter-individual differences in BMD are largely genetically determined, suggesting that GWAS holds much promise for identifying individual genetic determinants that contribute to disease pathogenesis. In light of this, the MSK-KP has access to GWAS summary results statistics from several efforts of worldwide consortia such as GEFOS (http://www.gefos.org/), and CHARGE (http://depts.washington.edu/chargeco/wiki/Main_Page), and large scale GWAS like the UK Biobank Study that involve BMD derived from dual-energy x-ray absorptiometry (DXA), peripheral quantitative computed tomography (pQCT), and quantitative ultrasound (QUS) respectively, and that encompass tens to hundreds of thousands of individuals.^11,12,31 Access to these resources provided the MSK-KP with unprecedented opportunity to explore genetic loci including low frequency loss of function variants which are often particularly good targets for pharmacotherapy.¹⁹ As a proof of concept, a recent GWAS of ultrasound-derived BMD identified genome-wide significant variants in genes coding for existing OP drug targets: [e.g. romosozumab (SOST), denosumab (RANKL), raloxifene (ESR1)].^11,12 Importantly, this resource is likely to expand with the recent announcement that the UK-Biobank Study will be sequencing whole-genomes and exomes in all participants from their study, and that GEFOS is committed to making the summary association result statistics available via the MSK-KP.

2. Publicly-available gene expression and methylation data from human and/or murine osteoblasts, osteoclasts and osteocytes are increasingly becoming available, which can be leveraged to implicate the functional genes underlying significant GWAS associations (Table 1).³² Conversely, GWAS data can be leveraged by cell biologists to establish whether differentially expressed genes that define a cell type or cell state in a model organism, are enriched for human homologues involved in musculoskeletal disease.^33,34 In light of this, the MSK-KP working group has recently identified several independent laboratories committed to providing MSK-KP with unique access to specialized -omics datasets originating from skeletal tissue and specific bone cell types.

3. Measurement of BMD in knock-out mouse models is highly relevant to osteoporosis, and through collaborators within the musculoskeletal research community, access can be obtained to large-scale knockout mouse phenotyping programs such as the International Mouse Phenotyping Consortium (IMPC)³⁵ and the Origins of Bone and Cartilage Disease (OBCD)³⁶ that measure up to 19 different skeletal parameters reflecting a combination of bone structure, function and strength. This resource enables the skeletal phenotypes of corresponding candidate genes to be followed up in a rapidly expanding dataset of more than a thousand deeply phenotyped mouse knockout lines.

Table 1.

Omics datasets related to osteoporosis that can potentially be hosted by the MSK-KP

Type	Source	Primary Cell Line	Repository	Identifier
RNA expression	Mouse	Osteoblasts	NCBI-GEO	GSE54461
RNA expression	Mouse	Osteoclasts	NCBI-GEO	GSM1873361
RNA expression	Mouse	Osteocytes*	N/A	N/A
RNA expression	Human	Osteoblasts	NCBI-GEO	GSE15678
DNA methylation	Human	Osteoblasts	NCBI-GEO	GSM683928
RNA expression	Human	Osteoclasts	MSK-KP	†

^*Kindly made available by AI Croucher (Garvan Institute of Medical Research)

^†https://doi.org/10.6084/m9.figshare.11953674.v2

Identifying and prioritizing osteoporosis drug targets using the MSK-KP:

A hypothetical MSK-KP workflow for identifying drug targets for osteoporosis could include several stepwise approximations as follows:

• Candidate loci: Identifying independent regions of the genome that contain genetic variants (i.e. loci) robustly associated with BMD or other osteoporosis-related traits by stepwise conditional association analysis of GWAS, whole exome sequencing and/or whole genome sequencing summary statistics.

• Gene prioritization: Prioritizing genes harboring low frequency, loss of function variants in protein coding regions of the genome for downstream analysis, as they often represent excellent potential drug targets.¹⁹ Also based on population genetic theory and recent empirical studies, rare genetic variants are likely to be enriched for functional and deleterious effects, thus being disproportionately represented among trait associated variants.³⁷ Candidate genes can be further analyzed using gene set enrichment analysis³⁸ to identify genes that cluster with defined anabolic signaling pathways that may be of therapeutic relevance.

• Causal variants: Annotating and fine-mapping new and existing loci to identify candidate causal variants and genes using advanced software tools.³⁹ Using Mendelian randomization⁴⁰ approaches in conjunction with publicly available expression quantitative trait studies (including in osteoblasts)⁴¹ and osteoclasts^42,43 to identify non-coding variants (and their target genes) that may causally influence osteoporosis by altering gene and protein expression levels. This is particularly crucial for gene annotation considering that most of the BMD variants identified by GWAS/GWS will be located in non-coding regions of the genome.

• Further refinement of genes and variants: The resulting list of candidate genes and variants can be further refined by mining publicly available gene expression and epigenetic resources including data from the RoadMap Epigenomics Project,²⁷ and the Encyclopedia of DNA Elements Consortium²⁶ both of which can be used to help identify candidate genes that are located in transcriptionally active genomic regions specific to human chondrocyte and/or osteoblast cells.

• Leveraging animal cell models: Publicly available RNA-seq datasets derived from murine osteoblasts, osteoclasts and osteocytes can be mined to identify whether candidate genes are transcriptionally active in skeletal cells.

• Leveraging mouse knock-out models: Importantly, skeletal phenotypes of corresponding candidate genes can be followed up in deeply phenotyped mouse knockout lines generated by the IMPC³⁵ and the OBCD.³⁶

• Leveraging other functional databases: Databases from the Zebrafish Information Network (ZFIN),⁴⁴ Online Mendelian Inheritance in Man (OMIM),⁴⁵ and International Skeletal Dysplasia Society Nosology of Skeletal Disorders⁴⁶ can also be mined to identify candidate genes, that when knocked out, result in musculoskeletal disease.⁴⁷ Candidate genes can be further analyzed using gene set enrichment analysis³⁸ to identify genes that cluster with defined anabolic signaling pathways that may be of therapeutic relevance.

Pleiotropic effects of genetic variants associated with musculoskeletal traits

The resulting list of plausibly causal variants, genes and pathways, can be followed up by mining several publicly available PheWAS datasets, which encompass > 4000 traits and diseases to identify pleiotropic associations and to predict unintended side effects⁴⁸ Also, since the MSK-KP is part of a larger initiative with several other knowledge portals (e.g., T2D, cardiometabolic), there are opportunities to identify potential pleiotropy by investigating variants and genes with other traits in the portal collection. Studying genetic pleiotropy may provide novel insights into disease pathophysiology,⁴⁹ with benefits including potential clinical implications of incorporating molecular signatures into the etiology and nosology of musculoskeletal disease, and the possibilities of estimating overlap with rare bone diseases⁵⁰. Thereafter, prioritized genes and biological pathways could be interrogated in large drug-gene interaction databases (e.g. DGIdb 3.0, consisting of 56,000 compounds that target 6000 genes⁵¹ and Open Targets⁵²) to inform drug screening projects.

The future of the MSK-KP

The MSK-KP is already a wonderful resource for skeletal biology and in the longer term, for other musculoskeletal traits. Skeletal phenotypes were a logical starting point given their availability at the time the portal was launched in March 2020). Populating the portal with bone mineral density measured by DXA and heel ultrasound, hip geometry and fracture placed these large datasets in a convenient and useable format. More refined phenotypes will soon be available such as high resolution peripheral quantitative computed tomography and even biochemical phenotypes such as serum sclerostin and osteocalcin concentrations. Also the portal can support the addition of monogenic skeletal dysplasia disorders. Importantly, the musculoskeletal system encompasses many other tissues (e.g., muscle, joints and tendons) and diseases (e.g., osteoarthritis, scoliosis) with their own genomic data. Osteoarthritis data were added in May 2020. The MSK-KP is in its nascent phase and the above workflows and innovative uses of the portal will require years to develop. Further data can be added on many other musculoskeletal phenotypes or diseases including lean mass, muscle strength, the multitude of arthropathies, scoliosis, cartilage and other soft tissue disorders, and others. Equally important will be the addition of the basic laboratory data to enable the interpretation of the human omic data. Thus there are many opportunities to expand the MSK-KP to be the largest collection of integrated data for molecular into musculoskeletal disorders. At the same time the portal serves the needs of the research community to make omic data accessible to a greater portion of the scientific community to inform experimental prioritization. In the long term, the portal will bridge gaps in knowledge by bringing together multidisciplinary expertise so that these –omic resources ultimately improve the health of the millions of patients world-wide suffering from musculoskeletal disease

Figure 1:

The Musculoskeletal Knowledge Portal (http://mskkp.org) integrates relevant genetic and genomic annotations and represents the results in distilled tables and interactive visualizations.