Cindy Boer

96 | Chapter 2.2 may provide new insights into hand OA. For heterogeneous diseases such as OA, strati- fication of OA phenotypes into different dimensions of disease is one way of increasing power in GWAS[18]. There are few GWAS of hand OA and none have examined hierar- chically defined clusters of OA joint presentation in the hands or hand OA endopheno- types that may provide new insights into hand OA pathogenesis. We therefore set out to examine the occurrence of OA across hand joints and conduct a GWAS stratified by hand OA patterns [23] to identify novel genetic loci for hand OA. Methods GWAS, discovery, replication and meta‐analysis We conducted genome-wide association studies (GWASs) on radiographic structural phenotypes for OA of the hand using data from the Rotterdam Study (RS) (n~8,700) [24]. For a detailed description of the RS, sub-cohorts and GWAS methods, see online S upplementary T ext and Table 1 . Briefly, genotypes were imputed to the Haplotype Reference Consortiumreference panel (v1.0) using theMichigan Imputation Server[25]. We assessed genetic associations in each RS sub-cohort using linear regression mod- els adjusted for age, sex and the first four genetic principal components. RVtests[26] was used for the GWAS analyses and results were quality controlled using EasyQC[27]. Variants with an imputation quality<0.3, minor allele frequency<0.05 or effective allele count<5 were excluded and genomic control correction was applied to all standard er- ror and p-values. Meta-analysis between the discovery cohorts was performed using fixed-effects inverse variance weighting with METAL[28]. Manhattan plots and QQ-plots were generated using R and R package qqman[29]. Independent variants with a p-val- ue<1x10 -06 and a chi-squared statistic test of heterogeneity p-value>1x10 -06 were select- ed for replication in the Framingham Heart Study (FHS) (n=1,203)[30]. For a detailed description of the FHS, see the online Supplementary Text . Summary level data from the discovery and replication stage were combined in a joint meta-analysis (METAL)[28]. Variants met criteria for replication if the association reached a p- value<0.05, had the same direction of effect as the discovery sample, and reached a joint meta-analysis p-value<5x10 -08 . Replicated variants were also examined for association with clinical OA (i.e., hospital diagnosed OA) based on GWAS summary statistics from a large-scale OA meta-analysis of data from the UK Biobank and Icelandic deCODE populations[14, 15]. For a detailed description of the UK Biobank and deCODE, see the online Supplementary Text. Associations that reached a p- value<0.01 were considered statistically significant.