Fig. 3: Tumour subtypes demonstrate differential NJ expression.
From: Tumour-wide RNA splicing aberrations generate actionable public neoantigens
a,b, Density (left) and box (right) plots showing the total putative NJs expressed in IDHmut (orange) and IDHwt (green) cases in TCGA GBM and LGG (IDHwt, n = 166; IDHmut, n = 263; a) and spatially mapped GBM and LGG data (IDHwt, n = 258; IDHmut, n = 277; b). c,d, Histograms and box plots depicting NJ counts in IDHwt (blue), IDHmut-A (yellow), and IDHmut-O (red) in TCGA GBM and LGG (c) and in-house GBM and LGG datasets (d). e,f, Volcano plots illustrating significantly upregulated (P < 0.05 and NES > 1, blue) and downregulated (P < 0.05 and NES < –1, red) gene sets comparing IDHmut-O versus IDHwt (left), IDHmut-A versus IDHwt (centre), and IDHmut-O versus IDHmut-A (right). GOBP (e) and Gene Ontology Cellular Component (f) gene sets were investigated. Splicing-related gene sets are denoted in yellow. NES, normalized enrichment score. g,h, Box-and-whisker plots depicting log2[RNA-seq by expectation–maximization (RSEM)] of splicing-related genes from GOBP sets with significant (P < 0.05) log2[fold expression] differences: increased (log2[fold increase] ≥ 1.5) between IDHmut-A (yellow) and IDHmut-O (red) cases when compared to IDHwt cases (blue) (g) and decreased (log2[fold decrease] ≤ 1.5) between IDHmut-O when compared to IDHmut-A and IDHwt cases (h). i,k, Pearson correlation of glioma-specific NJs against CELF2 (i), SNRPD2 (k, left) and SF3A3 (k, right) in IDHmut-O (z axis), IDHmut-A (y axis) and IDHwt (x axis) cases. NJs with correlations of ≥0.10 (purple) or ≤−0.10 (yellow) are highlighted, with NJACAP2 (i,k (left)) and NJPEA15 (k, right) analysed. j,l, Expression of splicing-related genes was assessed in LGG (SF10417; j) or GBM (GBM115; l) cell lines transduced with dCAS9–KRAB and control single guide RNAs (sgRNAs; n = 6), CELF2 sgRNAs (j) SNRPD2 sgRNAs (l, left, n = 3) or SF3A3 sgRNAs (l, right, n = 3). m,n, Box plots (left) and heat maps (right) showing NJ expression per case and Wilcoxon rank-sum test results across iCluster (C) subtypes in TCGA LIHC (iCluster 1, n = 65; iCluster 2, n = 55; iCluster 3, n = 63) (m) and LUAD (iCluster 1, n = 26; iCluster 2, n = 19; iCluster 3, n = 47; iCluster 4, n = 31; iCluster 5, n = 18; iCluster 6, n = 61) (n). Further statistical details are provided in Supplementary Table 3. NS, not significant; **P < 0.01; ***P < 0.001; ****P < 0.0001.
