The increasingly widespread use of DNA and RNA sequencing technologies to identify molecular determinants of cancer has enabled the classification of cancer based on molecular criteria. This invaluable resource can then form the foundation for guiding the design of potential targeted therapies. The holy grail of human cancer sequencing studies is to discover mutated genes that definitely confer a selective advantage to cancer cells. It is often implied that these driver mutations affect common oncogenic pathways across tissue types, however, many commonly mutated cancer genes have very complex biology, and regulate a variety of different biological processes. This poses a formidable challenge to pinpointing the mechanistic basis of carcinogenesis and developing targeted therapies based on molecular aberrations.
To tackle this problem using a new approach, investigators from the Human Biology division led by Dr. Bruce Clurman and his laboratory joined forces with the Paddison and Hockenbery labs. Together with their collaborators from the Pacific Northwest and abroad, they theorized that pertinent biological insights into oncogenic mutations may be obtained via gene expression signatures that can predict the mutational status of a tumor for a specific gene in different tumor types. To acquire these gene expression signatures, the authors used a type of machine-learning technique, known as kernerlized Bayesian transfer learning (KBTL) to infer transcriptional signatures predictive of mutation status across ten tumor types within The Cancer Genome Atlas (TCGA) datasets. Transfer learning is a machine learning method whereby a model developed for one task is repurposed for use on a second task; such optimization allows enhanced performance when modeling the second task. Compared to traditional learning approaches, KBTL has the advantage in having more robust statistics due to the larger sample size from combining multiple datasets, and also has the capacity to extract signals common across datasets, thereby revealing shared biological processes across organ sites. In this study, each tumor type was considered a separate task. KBLT was applied to infer predictors of the mutational status of a given gene in each tumor type based on gene expression.
This approach identifies gene expression patterns associated with mutation status across all tumor types, and was developed with the intention to reveal the core consequences of mutations. The authors selected the F-box/WD repeat-containing protein 7 (Fbw7) ubiquitin ligase as a test case because it is one of the most commonly mutated human tumor suppressors, yet its tumor suppression mechanism remains poorly understood. Furthermore many of its substrates are oncoproteins as well as master transcription factors, implying that Fbw7 mutations may broadly impact gene expression. Fbw7 epitomizes a highly complex and understudied cancer gene, likely to drive tumorigenesis through the involvement of multiple oncoproteins rather than a single driver, and is therefore a perfect candidate for KBTL. Using informatics and engineered cancer cells, the authors showed that Fbw7 mutations cause metabolic reprogramming by increasing oxidative phosphorylation and metabolic vulnerabilities that may be targeted therapeutically, and recently published these findings in the journal PNAS.
From the TCGA analysis, five distinct cancers from different organ sites presented with sufficient mutant samples for KBTL analysis, and were modeled to derive transcriptional signatures inferred to predict Fbw7 mutational status. Gene set enrichment analysis was performed on the 500 most predictive genes to identify biological pathways enriched within these signatures. Surprisingly, genes associated with mitochondrial function, termed “mitochondrial signature genes” (MSG) were found to be the dominant biological process enriched within Fbw7 predictive gene signatures. MSGs were similarly associated with loss of Fbw7 through mRNA repression, suggesting that Fbw7 loss is associated with metabolic dysregulation in primary tumors. Of these Fbw7-associated cancers, colorectal cancer emerged to be most enriched for MSG expression. To validate a causal relationship between Fbw7 mutations and MSG expression, isogenic colorectal cancer cell lines that differed in Fbw7 status were engineered to either impair or restore Fbw7 function. Experiments using these cell lines confirmed the KBTL prediction that Fbw7 mutations directly increase MSG expression. The authors further showed that Fbw7 mutations affected cellular metabolism by measuring the rate of oxygen consumption and extracellular acidification. Global metabolite profiles of metabolic changes dependent on Fbw7 were also performed, and revealed increased oxidative metabolism and context-dependent changes in central carbon metabolism in Fbw7-mutant colorectal cancer cells.
In summary, the authors report an approach to infer the physiologic consequences of oncogenic mutations by obtaining gene expression signatures from TCGA datasets that predict the mutational status of a gene across different tumor types, and identifying the shared biological pathways enriched within these signatures. This approach was employed to further study mutations in the Fbw7 ubiquitin ligase, which revealed a new role for Fbw7 in regulating cellular metabolism. Excitingly, this approach may be applied to study other complex cancer genes, exposing previously untapped therapeutic vulnerabilities and biological knowledge.
Davis R, Gonen M, Margineantu DH, Handeli S, Swanger J, Hoellerbauer, Paddison PJ, Gu H, Raftery D, Grim JE, Hockenbery DM, Margolin AA, Clurman BE. 2018. Pan-cancer transcriptional signatures predictive of oncogenic mutations reveal that Fbw7 regulates cancer cell oxidative metabolism. PNAS. 32(7-8):512-523.
Funding was provided by the National Cancer Institute Cancer Center support grants, the Turkish Academy of Sciences and the Science Academy of Turkey.