Phenotype-resolved cis-eQTL Mendelian randomization and colocalization at two osteoarthritis-associated regions, with partial replication in FinnGen

Background Osteoarthritis (OA) has no approved disease-modifying therapy, and genome-wide association studies often identify OA loci without resolving the causal gene. We evaluated two OA-associated regions on chromosomes 16 and 20, selected from an unpublished exploratory transcriptome-wide screen by the same authors. Methods We analyzed 14 genes using whole-blood cis-eQTL data from eQTLGen ( N = 31,684) and four Genetics of Osteoarthritis (GO) phenotypes (AllOA, KneeHipOA, KneeOA, HipOA; N = 353,388–826,690). Evidence from cis-eQTL Mendelian randomization (MR), coloc.abf colocalization, and summary-data-based MR (SMR) with HEIDI was graded with a three-level framework (FDR q < 0.05). The two lead signals were tested in FinnGen release 12 (up to 101,454 cases), an independent cohort, with the same instruments and pre-specified criteria (MR: same direction, p < 0.05; colocalization: PP.H4 ≥ 0.8). Results In the re-run (15,697 genes), the chromosome 20 region ranked first; MAPK3 was not the strongest gene in its region. No cell reached Level 1. The UQCC1/GDF5 locus, instrumented by UQCC1 expression because GDF5 has no whole-blood cis-eQTL, showed a risk-increasing locus-level association with Level 2 evidence for AllOA (single-instrument) and KneeHipOA. MR replicated in FinnGen in all three phenotypes (AllOA OR 1.116), and colocalization replicated for AllOA (PP.H4 = 0.929). MAPK3 met Level 2 in the four GO phenotypes only as single-instrument (AllOA, KneeHipOA, KneeOA; rs4548895) or SMR-only (HipOA) evidence. In FinnGen, the multi-instrument MAPK3 MR estimate did not replicate (AllOA OR 0.964, p = 0.12), colocalization favoured distinct variants (PP.H3 = 0.770), and only the rs4548895 association was reproduced. At a more conservative colocalization prior (p12 = 2 × 10⁻⁶), PP.H4 fell below 0.8 in four of the six Level 2 cells (MAPK3 × AllOA and HipOA; UQCC1/GDF5 locus × AllOA and KneeHipOA). Conclusions The UQCC1/GDF5 locus is the more robust signal, supported in discovery and independent FinnGen data, but at the locus rather than the gene level. MAPK3 evidence is single-instrument and did not replicate, so MAPK3 remains a tentative candidate. Colocalization support for most Level 2 cells depends on the prior, and discovery effect sizes, particularly for MAPK3, may be inflated by selection. These results are hypothesis-generating. Clinical trial number Not applicable.