Regulation of metabolism is one of the tumor development hallmarks. Here we tried to characterize such regulation across many tumor types in an unbiased manner: without specifying any of the standard metabolic pathways. For the analysis we used gene expression and somatic mutation data from 28 TCGA tumor types.

First, we for each tumor type we found “metabolic modules”: clusters of genes that both have similar gene expression pattern across samples and are localized close in a network of metabolic reactions. For that we developed an iterative algorithm that combines ideas of co-expression clustering and method GAM (Sergushichev et al, 2016, NAR) for finding metabolic modules using differential expression data .

We found that three groups of modules are largely shared between all tumor types. These groups seems to be related to immune system, however it remains to be further investigated. Other modules are more type-specific and are either unique or shared between small number of tumor types.

Next, we tried to find somatic mutations that are associated with (and potentially causal for) regulation of the found modules. We applied Kolmogorov-Smirnov test to find mutations associated with changes of module expression and filtered indirect associations using Bayesian network-like approach.

This resulted in a few dozen associations. Mainly these were associations with major oncogenes like TP53 or KRAS. We also found several known metabolic-specific regulators of tumor metabolism such as KEAP1 and NRF2 in lung adenocarcinoma or IDH in lower grade glioma. However, some of the found genes are not commonly known to be associated with metabolism and we are now investigating one of such hits.

Overall, we developed a pipeline to unbiasedly search for metabolic modules and their potential regulatory events in tumor samples. We applied this pipeline to TCGA data and were able to analyse overall landscape and find specific modules with their potential regulatory genes.