Discovery of new myositis genetic associations through leveraging other immune-mediated diseases

Abstract

Genome-wide association studies (GWASs) have been successful at finding associations between genetic variants and human traits, including the immune-mediated diseases (IMDs). However, the requirement of large sample sizes for discovery poses a challenge for learning about less common diseases, where increasing volunteer numbers might not be feasible. An example of this is myositis (or idiopathic inflammatory myopathies [IIM]s), a group of rare, heterogeneous autoimmune diseases affecting skeletal muscle and other organs, severely impairing life quality. Here, we applied a feature engineering method to borrow information from larger IMD GWASs to find new genetic associations with IIM and its subgroups. Combining this approach with two clustering methods, we found 17 IMDs genetically close to IIM, including some common comorbid conditions, such as systemic sclerosis and Sjögren's syndrome, as well as hypo- and hyperthyroidism. All IIM subtypes were genetically similar within this framework. Next, we colocalized IIM signals that overlapped IMD signals, and found seven potentially novel myositis associations mapped to immune-related genes, including BLK, IRF5/TNPO3, and ITK/HAVCR2, implicating a role for both B and T cells in IIM. This work proposes a new paradigm of genetic discovery in rarer diseases by leveraging information from more common IMD, and can be expanded to other conditions and traits beyond IMD.

Figure 1

Heatmap showing overall significant (FDR <1%) IIM datasets from Miller and Rothwell studies across 13 features Colors represent projection values; similar colors mean the projections are closer in a given feature. Full dots represent the dataset is significant for the feature at a FDR of less than1%, and hollow dots represent significance at a FDR of less than 5%. For 7 of 13 features, at least 1 myositis dataset was significant at a FDR of less than 1%. IIM (M)/(R), meta-analyses; M, Miller; R, Rothwell.

Figure 2

Relationships among DPMUnc and D_B clusterings DPMunc called seven clusters, while we called nine using D_B. IIM datasets are allocated to cluster 2 in DPMUnc and cluster 5 in D_B, along with other IMDs (highlighted in light blue). Other IMDs in DPMUnc cluster 2 are located in D_B clusters 7 and 9. CR(E)ST, CR(E)ST syndrome; Felty, Felty syndrome; IgG⁺ NMO, IgG⁺ neuromyelitis optica; JIA, juvenile idiopathic arthritis; LOMG, late-onset myasthenia gravis; PR, palindromic dermatitis.

Figure 3

PPs for shared causal variants (H4) between IIM and selected IMDs at eight top candidate SNPs A colored square indicates the colocalization analysis was performed, and the shade of color represents the PP H4 for the test. Medium confidence associations (0.5 < PP ≤ 0.8) are marked in yellow rectangles and high confidence (PP > 0.8) associations in green rectangles. Top candidate SNPs are labeled with their nearest gene or with a nearby strong IMD candidate gene when we find one. JIA, juvenile idiopathic arthritis; LOMG, late-onset myasthenia gravis. (M) and (R) represent that the dataset comes from the Miller or Rothwell study, respectively. IIM labels represent meta-analysis datasets.

Figure 4

Comparison of p-values of SNPs from FinnGen R5 and FinnGen R10 used in our validation process (A and C) SNPs identified using our pairwise FDR and colocalization approach (pwFDR + coloc) at two PP H4 levels (H4 > 0.5 and H4 > 0.8). (E) SNPs identified using a conventional genome-wide significant threshold (P < 5e−8). (B, D, and F) Zoomed-in view of the data showed in (A, C, and E). The diagonal red line represents the validation threshold (−log10(P) R5 = −log10(P) R10). The vertical blue line represents the genome-wide significant threshold. Numbers in green represent the number of SNPs that validate (above red line) and those that do not (below red line).