One of the main goals of large-scale in-depth analyses of cancer samples is to uncover the common alterations that define a tumor, in hopes of targeting these changes with directed drug therapies. Historically, we have found that genomic alterations in tumors originating in the same organ or tissue can vary widely, while similar patterns of genomic alterations can be found in tumors with different tissues of origin. There are many well described subtypes that share commonalities between tumor types, but vary substantial to other tumors within that organ. Despite this, we describe tumors based on tissue of origin, then perhaps based on subtype, to identify driver alterations and potential treatment options.
A recent analysis “Emerging landscape of oncogenic signatures across human cancers” (Ciriello et al. Nat Genet. 2013 Sep 26;45(10):1127-1133), has tried an alternate approach to identifying driver alterations, by classifying individual tumors based not on their tissue of origin, but rather their selected functional events. Using samples from The Cancer Genome Atlas (TCGA), they selected >3000 tumors from 12 tumor types, some with multiple subtypes, and integrated the copy number, somatic mutation and methylation data. The thousands of alterations were reduced to a candidate list based on recurrent copy number alterations identified by GISTIC , recurrently mutated genes identified by MuSiC and MutSig, and methylated genes know to be epigenetically silenced in cancer. Copy number changes genes and methylated genes of interest were also required to have concordant mRNA expression levels compared to wild-type cases. Altogether, 479 selected functional events were identified (116 copy number gains, 151 copy number losses, 199 recurrently mutated genes, and 13 epigenetically silenced genes).
The authors began to stratify the >3000 tumors based on the 479 selected functional events. Unsurprisingly, certain tumor types were resoundingly characterized by a high frequency of copy number alterations (C class) (OV, BRCA, LUSC, HNSC) or somatic mutations (S class) (KIRC, COAD, READ, GBM, LAML, UCEC), while other tumor types showed a strong mix of both types of alterations (LUAD, BLCA). As has been described in multiple tumor subtypes previously, when evaluating individual tumors, an inverse relationship was observed between the extremes – tumors had either a large number of somatic mutations or a large number of copy number alterations but never both. Additionally, TP53 mutations were strongly enriched in samples with copy number alterations, consistent with early TP53 mutations causing genomic instability.
Interestingly, when the individual tumors were further classified based on their oncogenic signatures, they were able to fit each individual tumor into a specific group within the M Class (17 subclasses) or C Class (14 subclasses). Each subclass is defined by a core set of alterations common to those tumors (see supplemental figure 5 for details). Key M Class alterations included mutations of ARID1A, CTCF, APC, TP53 and KRAS, while key C Class alterations were initially defined by TP53 mutation and the absence or presence of copy number alterations on chromosome 8. And while some subclasses were dominated by one particular tumor type, others were a mix of tumor types.
These clustering results have a significant clinical impact. Many of the selected functional events are also druggable targets, either alone or in combination. However, none of these alterations are exclusive to one tumor type nor present in 100% of samples in a particular tumor type. As the clustering analysis revealed, tumors described by tissue of origin can be much divided in terms of presentation of selected functional events. Therefore, it is suggested that a tumor be classified in a tissue-independent manner, based on the genetic and epigenetic alterations present. By focusing on the ~500 selected functional events to start, this approach would reduce the complexity of initial results for an individual tumor. Classification of the tumor based on the distinct pattern of those genomic features, would help identify potential targeted therapeutic options. Undoubtedly, more tumors and tumor types would need to be added to further stratify and refine the classes, but this methodology would allow for characterization based on oncogenic signatures, and may yield new insight into the mechanisms of oncogenesis and therapeutic targets.