Page 307 - Personalised medicine of fluoropyrimidines using DPYD pharmacogenetics Carin Lunenburg

for integrative QC assessment. In order to evaluate the possibility of population stratification or outliers, multidimensional scaling (MDS) analysis was performed in PLINK. In addition, pairwise identity by state (IBS)/identity by descent (IBD) statistics was calculated to assess duplicates. MDS, IBS and IBD were computed using PLINK. Patients who were identified as outliers based on IBS clustering were excluded from the analysis. MDS coordinates were extracted and used as covariates in the association analysis. SNP imputation was performed using the programs shapeit and impute2 with default parameters in which the reference panel 1000Genomes build version 3 was used with total, ‘cosmopolitan’, set of individuals.31 An MDS plot was created to compare self-reported ethnicity of patients.
Statistical analyses
Genetic variants were tested for an association with the onset of severe fluoropyrimidine- induced toxicity. The primary outcome was severe (grade ≥3) fluoropyrimidine-induced toxicity, compared to grade 0 or 1 fluoropyrimidine-induced toxicity. Grade 2 toxicity was excluded from this analysis to maximize the contrast between toxicities. Gender, age, baseline BSA and treatment type (grouped as previously published)5 were used as pre-specified covariates. Statistical analyses were performed in R statistics version 2.3.2. Base packages stats, survival and MASS were used to evaluate logistic, Cox, and ordinal regressions, respectively. Associations with a p-value ≤5x10-8 were considered statistically genome-wide significant. Associations with a p-value between 5x10-8 and ≤5x10-6 were considered suggestive. Post association QC was performed by visual inspection of Quantile- Quantile (QQ) plots of p-values of association tests and computation of the inflation factor given as: λ=(median(T1,...,Tn)/0.675)2, where T1,...,Tn are square roots of χ2 quantiles.
A Polygenic Risk Score (PRS) was constructed by extracting all SNPs with a p-value <0.01 in the association test. To avoid problems due to collinearity, in the list sorted according to p-values, SNPs in a window of 100 kb were excluded after inclusion of a SNP. A penalized regression model was fitted using R-package glmnet. Included clinical parameters were gender, age, baseline BSA and treatment schedule.
Results
Patients
Sufficient DNA was available for 1,146 out of 1,181 recruited patients (97%). These patients entered the QC procedure prior to association analyses. The flowchart on patient inclusion is shown in Figure 1. The observed individual genotype call rates varied between 97% and 100% and therefore meet the quality criteria. Based on subsequent QC steps, 45 patients (3.9%) were excluded from the analyses. Of these 45 patients, 30 patients (2.6%) were excluded due to missing genotypes, four (0.3%) patients were excluded due to a gender mismatch with the clinical data, six patients (0.5%) were excluded based on outlier removal of IBS plots. The inbreeding coefficient was 0.01 (-0.03─0.004), therefore, five (0.4%) patients were excluded. Of the 1,101 remaining patients, screen failures (N=55), patients with missing BSA at baseline (N=24) and DPYD variant allele carriers who received initially reduced dosages (N=80) were excluded (Figure 1). In addition, we chose to exclude patients who experienced grade 2 toxicity (N=343) from the primary analysis to maximize contrast between severe and non-severe toxicity.
12
Genome-wide association study
305