Figure 4.

Central nervous system (CNS) and transcription factor (TF) associations by subtype

(A) CNS associations by subtype. Heatmaps showing intersample correlations (purple, positive; cyan, negative) between mRNA profiles of tumors in PDX and patient metastasis compendium datasets (columns) and mRNA profiles of Fantom²⁰ cell types or tissues related to the CNS (rows).

(B) Top TF associations by subtype. Encode²¹ data on TF binding was extracted for 158 TFs, with gene associations defined by binding within 2 kb upstream of the gene start. For the set of 35 TFs represented, there was both significant overlap (p < 1E−6, one-sided Fisher’s exact test or chi-square test) between the TF-bound genes and the genes overexpressed in the expression subtype and significantly higher or lower levels of the TF gene in that same subtype (p < 0.05 by t test), for both PDX and patient metastasis compendium datasets.

(C) Of the 35 TFs represented in (B), 10 were represented in an expression profiling dataset of siRNA knockdown of specific genes.²² For three TFs—E2F1, HDAC2, and MYC—the genes were highly expressed in the associated s1 subtype, with the corresponding siRNA knockdown signature scoring significantly negative (p < 1E−6, t test) for both PDX and patient tumor metastases datasets. For these three TFs, the associated patterns are represented across PDX and patient tumor metastasis compendium datasets: average differential expression of the TF-bound genes (red-blue heat maps), differential expression of the TF genes (yellow-blue heat maps), and differential levels of the TF gene siRNA signature (purple-cyan heat maps). A negative siRNA signature association indicates that knocking down the TF gene would result in a global pattern opposite TF gene overexpression.