Supplementary Data from Transcriptome-Wide Association Study Identifies New Candidate Susceptibility Genes for Glioma

Supplementary Figure 1: Quantile-Quantile Plots of -log10 (P-value) associations. (a) TWAS for GBM glioma; (b) TWAS for GBM glioma (lower 90% of associations); (c) TWAS for non-GBM glioma; (d) TWAS for non-GBM glioma (lower 90% of associations); (e) GWAS meta-analysis for GBM glioma; (f) GWAS meta-analysis for GBM glioma (lower 90% of associations); (g) GWAS meta-analysis for non-GBM glioma; (h) GWAS meta-analysis for non-GBM glioma (lower 90% of associations). Supplementary Figure 2: Manhattan Plots of SNP genomic co-ordinates against GWAS meta-analysis -log10 (P-value). (a) GBM glioma; (b) Non-GBM glioma. The red line represents the Bonferroni-corrected threshold of P{less than or equal to}5Ã-10-8. Supplementary Figure 3: Dendograms to display gene module clustering in brain tissues. Clustering was performed using the WGCNA (11) package for each tissue. (a) Amygdala; (b) Anterior cingulate cortex (BA24); (c) Caudate (basal ganglia); (d) Cerebellar Hemisphere; (e) Cerebellum; (f) Cortex; (g) Frontal Cortex (BA9); (h) Hippocampus; (i) Hypothalamus; (j) Nucleus accumbens (basal ganglia); (k) Putamen (basal ganglia); (l) Spinal cord (cervical c-1); (m) Substantia nigra. Supplementary Figure 4: Heatmaps illustrating gene module correlations in brain tissues. Darker colours indicate stronger associations. (a) Amygdala; (b) Anterior cingulate cortex (BA24); (c) Caudate (basal ganglia); (d) Cerebellar Hemisphere; (e) Cerebellum; (f) Cortex; (g) Frontal Cortex (BA9); (h) Hippocampus; (i) Hypothalamus; (j) Nucleus accumbens (basal ganglia); (k) Putamen (basal ganglia); (l) Spinal cord (cervical c-1); (m) Substantia nigra. Supplementary Figure 5: Module clustering of brain tissues with amygdala. Modules were created by clustering according to the amygdala tissue using WGCNA (11). These modules were applied to each tissue in turn and the Z statistic is a quantitative measure of module preservation. (a) Amygdala (reference); (b) Anterior cingulate cortex (BA24); (c) Caudate (basal ganglia); (d) Cerebellar Hemisphere; (e) Cerebellum; (f) Cortex; (g) Frontal Cortex (BA9); (h) Hippocampus; (i) Hypothalamus; (j) Nucleus accumbens (basal ganglia); (k) Putamen (basal ganglia); (l) Spinal cord (cervical c-1); (m) Substantia nigra. Supplementary Figure 6: TWAS power plots. Simulation analysis based on 12,488 glioma cases (6,183 GBM, 5,820 non-GBM) and 18,169 controls. Gene expression was generated from the distribution of gene expression levels from the respective brain tissue. Statistical power was calculated at P 0.4, case and control MAF > 0.01, PHWE in controls > 1x10-8. For the meta-analysis we considered all SNPs meeting the above criteria in all eight studies as well as having heterogeneity I2 < 75 in GBM/non-GBM glioma. a For the UCSF/Mayo study an imputation threshold of > 0.8 was applied. Supplementary Table 2: Prediction models trained on GTEx v7 data and covariance files obtained from http://predictdb.org/. Genes were retained if nested cross-validated correlation between predicted and actual levels > 0.10 (R2>1%) and P-value of the correlation test < 0.05. Supplementary Table 3: Previously reported glioma risk SNPs. Supplementary Table 4: Brain tissue, soft power threshold and number of independent genes. Soft power thresholds were chosen at the 90% threshold unless otherwise indicated. + For Cerebellar Hemisphere and Cerebellum, the 90% threshold was not achieved and 80% was imposed. For this reason such tissues were excluded from median calculations for soft power and number of independent genes. Supplementary Table 8: S-PrediXcan results for 31 genes in 922 whole-blood samples from the Depression Genes and Networks (DGN) study. Supplementary Table 9: Annotation of identified genes by presence within COSMIC cancer gene census and overlap with tumour copy number changes. *Cosmic cancer gene census annotation for CDKN2A was used.