NEW YORK (GenomeWeb) – In the third and final set of papers out this week from The Cancer Genome Atlas, researchers examined the molecular clustering of tumors, how key oncogenic processes contribute to tumor development, how certain signaling pathways are altered in cancer, and more.
TCGA has been publishing this week a collection of 29 papers, with the last set of papers appearing today in Cell, dubbed the Pan-Cancer Atlas. Papers appearing earlier in the week focused on cancer development and tumor subtypes.
"This project is the culmination of more than a decade of groundbreaking work," NIH Director Francis Collins said in a statement. "This analysis provides cancer researchers with unprecedented understanding of how, where, and why tumors arise in humans, enabling better informed clinical trials and future treatments."
Among the papers published today are three "flagship" papers that describe some of the effort's core findings.
In the first such paper, TCGA Research Network investigators performed molecular clustering on the full set of 10,000 TCGA specimens hailing from 33 cancer types. Using the iCluster algorithm and data on aneuploidy, DNA hypermethylation, mRNA, microRNA, and proteins from the tumor samples, the researchers identified 28 molecular subtypes, more than they uncovered in their initial pan-cancer analysis in 2014.
A third of these clusters were largely homogenous for a certain tumor type, while the others were more heterogeneous. Most of those mixed clusters followed particular cell-of-origin or organ system patterns, such as pan-GI or pan-squamous. These findings suggested to the researchers that similarities among these cancers could inform patient treatment approaches.
"Tumor location has been the primary method for determining treatment for a given cancer patient," co-first author Katherine Hoadley from the University of North Carolina-Chapel Hill said in a statement. "This study helps us get a better understanding of the relationship across and within different tumor types. If tumors are genetically diverse within an organ, we should rethink the way we treat them."
In the second paper, TCGA Research Network investigators described the three main processes that contribute to cancer development: the effect of somatic and germline mutations; the influence of the tumor genome and epigenome on the transcriptome and proteome; and the relationship between the tumor and the microenvironment.
They examined the influences of these three components within the full set of TCGA tumors to find that their interplay is complex. The researchers noted that the interactions between driver genes and the transcriptome depend on the context and that certain deregulated oncogenic processes are gene- rather than driver-dependent.
But these interactions also indicated that combining different immunomodulatory treatments could target a wider range of these processes and that tumors should be evaluated within their larger environment. This, the researchers said, would mean that "drastic changes in clinical practice and drug development" are needed.
With the third milestone paper, the TCGA Research Network investigators characterized how 10 key signaling pathways are affected by somatic alterations in 33 different cancer types. Those pathways included the cell signaling, Myc, Notch, and beta-catenin/WNT pathways.
By drawing on mutation, copy-number, mRNA expression, and DNA methylation data from more than 9,000 tumors, the researchers found that 89 percent of tumors have a driver mutation in at least one of these pathways and that 57 percent of tumors have at least one pathway alteration that could be targeted by an existing drug. The researchers further could stratify these tumors into 64 subtypes based on changes to their signaling pathways.
Additionally, they noted that a number of these potentially actionable alterations co-occurred, meaning that combination therapies might be useful. However, they added that developing combination therapies has been challenging due to the lack of safety data, clinical trials, and the slow uptake of multi-panel gene tests.
A handful of companion papers were also published today in Cell.
For instance, Washington University in St. Louis' Li Ding and her colleagues examined whether pathogenic germline variants were present among the more than 10,000 cancer cases in the set. They uncovered 853 pathogenic or likely pathogenic variants in 8 percent of those cases.
Most of these predisposition variants were marked by a loss of heterozygosity or biallelic two-hit events and were expressed at a low level. At the same time, though, 33 variants affected the activating domains of oncogenes and ramped up their expression.
In another paper from Ding and her colleagues, they characterized oncogenic driver genes and mutations within the TCGA tumor set. Using more than two dozen computational tools, they identified nearly 300 cancer driver genes within more than 9,000 tumor exomes. These genes and other cancer-linked mutations were present across tumors hailing from different organs and cell types. They further identified more than 3,400 putative missense driver mutations.
Overall, they noted that 57 percent of tumors harbor oncogenic events that are potentially actionable. However, they noted that the gene list they developed is constrained by its emphasis on point mutations and small indels.
Also using the TCGA data, Chan Soon-Shiong Institute of Molecular Medicine at Windber's Hai Hu and his colleagues developed a standardized dataset that folds in clinical outcomes. This TCGA Pan-Cancer Clinical Data Resource (TCGA-CDR) includes data from across the 11,000 TCGA samples on four clinical endpoints: overall survival, progression-free interval, disease-free interval, and disease-specific survival.
This resource, Hu and his colleagues said, could power either pan-cancer or disease-level translational studies, though they noted it has a few limitations.
Meanwhile, MD Anderson Cancer Center's Han Liang and his colleagues in another paper describe their analysis of enhancer expression across the TCGA samples. They reported that cancer tissues, as compared to matched normal tissues, exhibited global enhancer activation. Additionally, they reported that global enhancer activity was positively associated with aneuploidy, though not with mutations.
The researchers also combined eQTL, mRNA co-expression, and Hi-C data to develop a computational approach to infer causal enhancer-gene interactions. With it, they uncovered enhancers associated with clinically actionable genes, including one located 140 kilobases downstream of the immunotherapy target PD-L1, which could be alternative drug targets.
Lastly, Maciej Wiznerowicz from the Poznan University of Medical Sciences in Poland and his colleagues applied a one-class logistic regression machine-learning approach to transcriptomic and epigenetic data from TCGA to tease out two indices of stemness. One, dubbed mDNAsi, reflects epigenetic features, while the other, mRNAsi, reflects gene expression. These could then be translated into stemness scores that could be used to stratify tumors based on whether they appear de-differentiated. The stemness indices also reflect potential drug targets, the researchers noted.
All the Pan-Cancer Atlas papers are available through a portal at Cell Press.