For all brain expression and replication cohort analyses, statistical tests were performed using open source R software packages. R-scripts are available upon request. For evaluations of the effect of disease status, SNP genotype, and gender on TMEM106B expression, linear regressions were used to compute p-values in univariate models. We evaluated assumptions of linearity by checking QQ plots (observed vs. predicted under normal distribution). For pairwise comparisons within the linear models, risk allele homozygotes and GRN mutants, respectively, were designated the reference group for marginal t-tests evaluating genotype effects and the effects of GRN mutations on expression. Normalized gene expression sample genotype and gender data are provided in Supplementary Data 1 and 2. For evaluations of the independent contributory effects of SNP genotype and TMEM106B expression on disease state, logistic regressions were used to compute AIC values in multivariate vs. univariate models (Supplementary Table 4a). For evaluations of the independent contributory effects of SNP genotype and disease state on TMEM106B expression, linear regressions were used in multivariate vs. univariate models (Supplementary Table 4b). For analyses of association of SNP genotypes with disease in our TaqMan replication cohort, Cochran-Armitage trend tests were used to compute P-values under a codominant model.
