72 Genetic analyses Genotyping and imputation Genome-wide genotyping data was generated as described in.22 Genotype data processing is described in.36 Briefly, microarray data were generated on CytoSNP and ImmunoSNP platforms and processed on the Michigan Imputation Server.37 Haplotype phasing was performed using SHAPEIT and imputation using the HRC version R1 as reference (Consortium and the Haplotype Reference Consortium, 2016). Genetic preprocessing We used GenotypeHarmonizer38 for imputation (minimum posterior probability of 0.4), call rate (minimal call rate of 95% of samples), Hardy-Weinberg equilibrium (minimal P-value allowed of 1x10-6) and SNP ambiguity filtering. We then computed identity by descent among samples using PLINK v1.939 on linkage disequilibrium (LD)– pruned genotypes (window size 50 Kb, variance inflation threshold 5 and maximum R2 between variants 0.2). We estimated identity by descent between all samples using PLINK and randomly selected a sample from the pairs with a PI_hat value > 0.2, which resulted in the removal of 14 samples from subsequent analysis (total of 1,255 available samples). Heritability and genetic correlation GCTA40 was used to compute a genomic relationship matrix (GRM) using genotyped SNPs with a minor allele frequency (MAF) of at least 0.05. The GRM was used to estimate antibody-bound peptide heritability using a linear mixed model between unrelated individuals (GREML approach) 41 while controlling for age, sex and PhIP-Seq plate. Similarly, genetic correlations between peptides were estimated using GCTA.42 Genome-wide association For each of the available antibody-bound peptides, we conducted an association analysis between genotypes (MAF > 0.05) and presence/absence profile. PLINK v1.939 logistic mode was run while controlling for age and sex and using the genotype in an additive model. This analysis was reproduced in a recessive model between 49.1 and 49.3 Mb in chromosome 19. Genetic meta-analysis A second study using the same PhIP-Seq library panel and protocol has been conducted in an IBD cohort from the Netherlands.28,29 Genotyping information is available for this cohort and was previously described in.43 The same quality control steps and analysis methods have been used as described above, while the disease subtype (Crohn’s disease or ulcerative colitis) was also added as an extra covariate in the logistic regression. Summary statistics from both the LLD and 1000IBD cohorts were meta-analyzed using METAL.44 We performed a P-value–based fixed-effects meta-analysis. A study-wide significance threshold was estimated by dividing the genome-wide significance threshold of 5x10-8 by the number of independent peptides included in the GWAS. The number of PCs needed to reach 90% Chapter 3
RkJQdWJsaXNoZXIy MjY0ODMw