Investigating the metabolism of tumours with an in silico chemical genetics approach

Tumours are well known to adopt a physiologic state known as aerobic glycolysis---they use glycolysis rather than mitochondrial respiration even when oxygen is abundant. We hypothesize that the predominant pathway by which a cell line generates its energy is indicative of the degree to which the cell line is transformed. Specifically, we postulate that a cell line that is more dependent on oxidative phosphorylation and mitochondrial respiration is less transformed, and a cell line that is more dependent on glycolysis is more transformed. Using pre-existing, publicly available datasets and information from the literature, I examined in silico both the basal level expression of genes in a panel of cancer cell lines, including those genes relevant to glycolysis and oxidative phosphorylation/mitochondrial respiration, as well as the growth responses of these cell lines to small molecules that target elements of the two energy pathways. To facilitate my investigations, I constructed a relational database to which I deposited the data. Using this database, I performed a global, non-supervised hierarchical clustering of cell lines vs. small-molecule inhibitors; this process revealed two major subsets of the cell lines (referred to here as subsets A and B). By visual inspection of a computed heatmap, I was able to hypothesize that ATP synthase inhibitors are the most effective classifiers of the two subsets. I next computed the rank of the genes whose expression levels in the untreated state of the cell lines best correlated with the two (sensitivity-derived) subsets of the cell lines. This marker selection enabled me to identify the genes that are the most effective classifiers of the two subsets. Most strikingly, an enrichment of the genes encoding elements of the mitochondrial respiration pathway was revealed, which supports our original hypothesis. In order to enable the database to be used to address a broader set of questions, I created a user interface that links the database to software platforms available from previous research efforts at the Broad Institute. Finally, I used my database as an analysis environment to extract knowledge not previously gained from the individual studies that yield the data. This exercise yielded three primary new hypotheses, namely: (1) the loss of a pyruvate dehydrogenase complex subunit contributes to the switch from dependency on ATP synthase to glycolysis; (2) in more transformed (i.e., more resistant to ATP synthase inhibitors) tumour cells, an isozyme of fructose-1,6-bisphosphatase that is capable of carrying out gluconeogenesis even though glycolysis is occurring is up-regulated, apparently providing tumour cells a selective advantage of survival; and (3) the expression of a major histocompatibility complex Class Ib molecule, HLA-F, also correlates increased transformation (resistance to ATP synthase inhibitors). I speculate that this part of the strategy a tumour utilizes in vivo to evade the immune system. These hypotheses now await future exploration.