Election 2016: Voting on Variants

Sequencing of cancer genomes has become affordable and accessible. We have transitioned in the last decade from focused sequencing of recurrently mutated “hotspots” in gain-of-function oncogenes (e.g., BRAF^V600E, KRAS^G12D) to unbiased interrogation of all coding genes using hybrid capture–based next-generation sequencing (NGS). Gene panels, or targeted capture methods that interrogate thousands of coding exons in hundreds of genes, are increasingly deployed in oncology clinics, within and beyond academic medical centers. Experts in inherited disorders utilize whole-exome sequencing in children to diagnose disorders that do not comport with known genetic causes. Although numerous questions remain about the clinical utility, cost effectiveness, and appropriate payer reimbursement for NGS assays, it is hard to imagine retreating to a day when we interrogated a few single base positions in metastatic, life-threatening cancers. NGS is here to stay.

A by-product of the greater insights provided by less biased genomic characterization is unexplained observation: a first step in discovery. It is underappreciated that so-called high-prevalence recurrent mutations in a given cancer type are actually relatively rare when all variants are considered. We have by now discovered and catalogued thousands of these variants of unknown significance (VUS). These alleles, rare yet recurrent, are increasingly backlogged without further characterization as we move from hundreds to thousands of tumors sequenced (1, 2). Functional sequelae of certain variants (e.g., nonsense mutations) can usually be deduced, but most variants are missense mutations with unknown effects on protein function, stability, or localization. Their biological impact is difficult to predict computationally, despite intense in silico efforts to do so. At some point, good old-fashioned functional characterization is required. Understanding rare variants is especially relevant for the majority of patients whose tumors lack canonical driver oncogene mutations. What genetic pathway drives these tumors? How can we personalize care for these patients?

Kim and colleagues (3) have constructed a systems approach to identify cancer mutations affecting cellular transformation and gene expression. They first chose significantly mutated genes from >5,000 tumors representing 27 cancer types. They focused their functional assay on 474 observed mutations in 178 genes; three quarters of the alleles were found only once in the >5,000 tumors, making these truly rarely mutated genes. These alleles were placed, accompanied by barcodes, into a sensitized, nontransformed human cell line. Lentiviral transduction was performed in arrayed fashion, and the cells were pooled to modulate the number and distribution of known oncogenic alleles (Fig. 1). These pools were then injected subcutaneously into immunosuppressed mice. If and when tumors formed, the persisting transduced alleles were profiled. Many of the alleles passing selection were of RAS genes, including a rare KRAS^D33 allele. Additional transforming alleles were found in NFE2L2, AKT1, PIK3CB, and POT1. Some pools of cells transduced with 110 unique alleles in aggregate did not form a single tumor after 18 weeks, indicating that spontaneous tumors appear rarely in this system. One specific and surprising finding was the ability of NFE2L2 variants to transform the target embryonic kidney cell line, as it was believed that activation of the KEAP1/NRF2 pathway was required of cancers arising in a highly oxidative environment (4).

To orthogonally characterize the mechanism of oncogenic VUS in their system, the authors used an inexpensive, bead-based assay to assess the effects of each mutation on 978 mRNA transcripts with high discriminative potential. This genomic signature provided both complementary and independent information about specific variants' effects on cellular physiology. For example, KRAS alleles both were transforming in vivo and affected gene expression in directionally predictable ways. Other, nontransforming alleles, such as those in isocitrate dehydrogenase genes, can likely be classified as change/gain-of-function, based on their transcriptional similarity to activating alleles.

The distribution of frequency of specific mutations in genes, across all cancers, is striking and specific. A handful of genes are recurrently mutated at very high frequencies in a multitude of different cancer types. Well-recognized members of this class include TP53 and RAS genes. Many of these mutations are also commonly found in benign proliferations that do not progress to cancers at a high rate [TP53 is observed at very high frequencies even in phenotypically normal, sun-exposed skin (5); BRAF is observed in benign epidermal nevi (6)], strongly suggesting that they are relatively easily acquired and require partners to fully disrupt normal differentiation. An additional subset of genes are recurrently mutated at substantial prevalence in specific cancers, such as activating NOTCH mutations in hematologic malignancies or isocitrate dehydrogenase in astrocytomas, suggesting a tissue-specific differentiation pathway that is deregulated by the mutation. The very rare mutations targeted in Kim and colleagues' study appear to be phenomenologically distinct. They may arise in human cancers at such low rates for one or more of several plausible reasons.

First, the mutations may be permitted only in certain cell types or at certain developmental stages, restricting the window of time or space in which their acquisition is beneficial. Second, they may represent weak or conditional alleles, requiring cooperation with other common or rare somatic mutations to promote tumorigenicity. Third, they may require a specific, relatively rare germline genetic background in which to produce a pro-oncogenic effect or to avoid a deleterious one. Fourth, they may not increase proliferation de novo, lessening their likelihood of co-acquiring other mutations in a tumor. A particular strength of the Kim and colleagues approach is that it crisply and immediately distinguishes between these possibilities, while opening roads to further dissect them mechanistically. For example, alleles produced in this system are unlikely to require a second rare mutation or a rare genetic background, given the timing of the tumor formation and the cell type that was transformed. They are also more likely to induce proliferation, founding cell populations that may acquire additional somatic events or epigenetic alterations. Loss-of-function mutations are not modeled in this positive selection approach.

It was not obvious that a pooled strategy, in such a short time frame, would be informative. Generalizing the cellular context in which a mutation occurs likely lowers sensitivity but increases specificity. Thus, one of the most exciting features of this demonstration is the potential to expand such screens in the context of germline activating mutations. Such studies would begin to model cancer mutation interactions that have thus far remained experimentally elusive and also begin to place loss-of-function mutations in tumor suppressors into context.

The specific mutations in PIK3CB and POT1 detected in this study imply that a larger set of alleles in these genes may promote oncogenesis than currently recognized, providing immediate assistance to clinicians tailoring therapies. Most importantly, the system by which Kim and colleagues interrogate hundreds of mutant alleles gives us two powerful tools to interrogate thousands more. Pooled barcoded cDNA in vivo screens and the L1000 transcriptional profile are each platforms accessible to the community at large. Predictable future iterations may include customization of cell type for specific disease (e.g., melanocytes for modeling variants found in BRAF, NRAS, NF1 wild-type melanoma) or tailored to reflect a specific cancer subtype or differentiation state.

No potential conflicts of interest were disclosed.