It is now widely accepted that classification of tumors based on their genetic markers is an important prerequisite for designing and choosing an appropriate disease-management strategy. Classification of lung cancer, however, still relies on the visual evaluation of the morphology of the tissue. What makes it even more difficult is that each of the major histological types of lung cancer is a heterogeneous collection of tumor subtypes.
The World Health Organization has attempted to describe these subtypes based on morphological appearance, but this classification is of limited usefulness in clinical practice due to high subjectivity of such characterization.
“Other cancers, such as breast cancer, have clearly identifiable genetic characteristics that profoundly influence the choice of therapies,” says D. Neil Hayes, M.D., Lineberger Comprehensive Cancer Center, University of North Carolina. “We were able to lay the background work for clear and reproducible identification of lung cancer subtypes based on DNA microarrays.”
Dr. Hayes’ group supplemented its own studies with the meta-analysis of previously published independent datasets, resulting in a sizable cohort of over 1,000 patients. “Historically, reconciliation of the results of individual gene-expression studies proved to be very difficult. Although tumor subtypes seemed to exist, there was no consensus of their number on nature.
“We realized that if we correct for technical imperfections, we should be able to uncover biological bases characteristic to each subtype.”
The researchers analyzed the studies generated on several different gene-expression platforms and utilized integration correlation statistics to select for the probes that measured the expression reproducibly across the chosen platforms. Out of 3,000–5,000 genes selected by this method, about one-third demonstrated reproducible differential expression of varying degrees.
The analysis placed all adenocarcinoma samples into three subtypes, and squamous cell carcinomas into four subtypes. The subtypes have statistically significant survival differences and patient demographics, independent of disease stage. They are comprised of tumors with differing underlying rates of mutations in key lung cancer genes, including KRAS and EGFR.
Predicting Radiation Resistance
Sixty percent of cancer patients undergo radiation therapy (RT). However, the majority of epithelial cancers is only marginally sensitive to RT, requiring very large doses to produce a measurable effect. Some types of cancer (such as renal cell and melanoma) are notoriously resistant to RT.
“We started with the idea that tumor regulatory pathways are organized in complex networks where a significant redundancy could be expected,” comments Javier F. Torres-Roca, M.D., division of experimental therapeutics, H. Lee Moffitt Cancer Center and Research Institute and CMO at Cvergenx. “Therefore, to understand tumor radiosensitivity, one needs to understand the structural molecular components of radiation sensitivity networks.”
The team began by analyzing individual gene expression in cancer cell lines and correlating it with cellular response to radiation. “However, just gene expression was not enough to develop the mathematical algorithm. The model had to include other biological data, such as mutation status of RAS and p53 genes. Once we added the biological data, we were able to use linear regression to fit a predictive model to an observed dataset.”
The studies supported linear regression analysis as a valid approach to correlate gene expression with intrinsic radiosensitivity of the cells lines. The resulting map of 500 interconnecting genes was further distilled to the network of 10 gene hubs. The predictive value of the model was validated using knockouts of hub genes, which increased tumor radiosensitivity.
“Not surprisingly, the analysis of known radiosensitizer drugs showed that they interfered with just a few of the hubs, suggesting that our clinical approach needs to include combination strategies, overcoming the redundancy of signals,” says Dr. Torres-Roca.
The current model predicts a radiosensitivity index (RSI) for tumors of epithelial origin. A positive predictive RSI value was 86% when correlated with the pathological response to RT in esophageal, rectal, and head-and-neck patients. And while the model is still being validated, it may be the first step to understanding how RT may be tailored for the maximum patient benefit.