Identifying molecular markers that stratify tumor samples into meaningful subtypes is an important goal in cancer genomics. Ideally, these subtypes correlate with clinical features, such as the aggressiveness of a tumor or response to drugs, and thus can be used to guide treatment. Early successes in defining such subtypes include the identification of translocations in leukemias, ERRB2 (HER2) amplification in a subset of breast cancers, and others. Since the introduction in the late 1990s of microarray techniques, there has been an explosion of studies to define subtypes according to gene expression signatures. This work has led to some notable successes; but in many cancers, signatures or clinical correlations identified in one study were not reproduced in other studies.
Hofree et al. introduce a novel approach to stratify patients on the basis of the somatic mutations present in their tumors. Cancer is a disease driven by such somatic mutations, which accumulate in the genome during the lifetime of the individual. Recent advances in high-throughput DNA sequencing technologies now enable whole-genome or whole-exome measurement of somatic mutations. In particular, The Cancer Genome Atlas (TCGA) is using whole-exome sequencing to measure somatic mutations in protein-coding regions of genomes from ~500 samples from each of ~25 cancer types. Similar projects are underway by other groups, including dozens of national consortia under the umbrella of the International Cancer Genome Consortium.
The initial results from these large-scale sequencing studies demonstrated a major impediment to the use of somatic mutations for patient stratification, namely, cancers exhibit extensive mutational heterogeneity, with mutated genes varying widely across individuals. Moreover, an individual cancer sample may have somatic mutations in only a few to a few dozen of the ~21,000 human genes. In other words, if one builds a somatic mutation profile for a sample, where each gene is assigned a 1 or a 0 if the gene is mutated or not mutated, respectively, then the resulting profiles will be sparse, or nearly all 0s . Consequently, comparison or clustering of such mutation profiles will not yield additional information beyond that revealed by direct examination of the handful of commonly mutated genes.
Hofree et al. apply NBS to somatic mutation data from TCGA studies of ovarian carcinoma, endometrial carcinoma and lung adenocarcinoma. On the ovarian and lung cancer data sets, NBS computes subtypes that discriminate the survival time of patients better than can subtypes derived from gene expression data. On the endometrial data set, NBS subtypes are closely associated with histological subtypes. Interestingly, although NBS significantly outperforms microarray-based gene expression for patient stratification, its gain over mRNA-Seq is smaller on the lung and endometrial data sets, suggesting an overall advantage for sequencing data (DNA or mRNA) over microarray data.
Given that driver mutations are by definition directly responsible for cancer, one might anticipate that mutation profiles, or network-smoothed mutation profiles, would provide more functional insights than would gene expression signatures.Not all genes in the NBS subtype networks are well-known cancer genes: on the contrary, some are proposed to be genes containing an unusually high number of random, ‘passenger’ mutations.
NBS makes it possible to derive clinically and biologically meaningful subtypes directly from whole-exome and whole-genome cancer sequencing data sets. As these data sets continue to increase in size and scope, NBS may have a prominent role in cancer research and in precision oncology.
Reference:
Making connections: using networks to stratify human tumors. Nature methods. 2013; 10:1077-1078