WASHINGTON (GenomeWeb) – At the American Association for Cancer Research's annual meeting on Monday, researchers gathered to hear presentations on the power of pan-cancer genomics studies.
Though much of cancer research can be said to take a narrow view of each disease or tumor type by itself — or even focusing on one individual patient at a time — the pan-cancer analyses showed the benefits of zooming out and taking a wide view of cancer as a whole.
Peter Van Loo of The Francis Crick Institute in the UK presented his group's work looking at the genetic archeology of more than 2,700 cancers across more than 30 cancer types. He called the cancer genome "an archeological record of tumors past," noting that somatic changes can be studied to infer a tumor's evolutionary history.
For example, Van Loo said, a researcher can count the number of mutations in a given tumor to determine the timing of copy-number gains, and see when a certain chromosome was duplicated, leading to a specific problem that eventually leads to cancer. An analysis looking at the timing of copy-number gains across each cancer type can also show which gains happen late in the tumor's development as opposed to being present from the beginning. Most tumor types gain a majority of their mutations later on in their development, Van Loo said, whereas other cancers like ovarian and breast cancer have a substantial subset of gains that happen early on in tumor evolution.
After building a timeline of mutational development for various cancer types, Van Loo and his colleagues saw something that may help partially explain why late-stage disease is so much harder to treat than early-stage cancer. Driver mutations that occur early in a cancer's development can generally be explained by one or a combination of genes in a set of 54. Later tumor evolution, however, requires more than 100 genes to explain. As time advances in a tumor's evolution, Van Loo explained, the tumors have many different pathways they can choose to use to grow and develop. Such a wider variety of pathways and possible mutations makes it harder to pinpoint a strategy for treatment in later disease.
The team also attempted to use these mutational timelines to pinpoint the emergence of each cancer type's most recent common ancestor cell. They found that in most cancers, the most recent common ancestor emerges one to two years before diagnosis. In some cancers, the most common ancestor can emerge up to five years before diagnosis. The researchers further found that whole-genome duplications can occur 10 years before diagnosis, depending on the cancer type. In certain lung cancers, these duplications can happen up to 20 years before a patient is diagnosed.
Finally, Van Loo's team studied the development of intra-tumor heterogeneity in 1,900 cancer samples for which they had the power to detect subclones. Looking at the fraction of single nucleotide variants and the diffraction of subclonal copy-number changes, they found that each cancer type tells its own story — liver cancer, for example, has a small amount of subclonal SNVs and indels, and shows more subclonal copy-number changes, whereas pancreatic cancer shows a lot of subclonal point mutations and very few subclonal copy-number changes.
Underlying the clonal expansion is a positive selection of driver mutations, he added. Intra-tumor heterogeneity is pervasive across cancers, and subclones contain driver mutations that are under positive selection.
Gad Getz, of the Massachusetts General Hospital Cancer Center, detailed his team's work on using the power of multiple modeling tools to uncover driver mutations in different cancers.
His team concentrated on looking at mutational burden, clustering of mutations, and impact on cell function to find genomic elements such as non-coding genes, core promoters, 3' and 5' UTRs, enhancers, long non-coding RNAs, promoters of lncRNAs, small RNAs, and miRNAs. They built what Getz called meta-cohorts, combining cohorts of cancer patients grouped together by tumor type such as adenocarcinoma or lymphoma, and so on, and also by tissue type.
The researchers analyzed these cohorts with multiple modeling tools, some of which look at mutational burden, some of which group different mutations based on the their consequences to tumor development, and others of which look at clustering of mutations to determine their importance. Combining these tools to do a meta-analysis of each meta-cohort improved sensitivity and specificity in finding candidate mutations of significance, Getz said.
Indeed, the group had a number of significant hits per genomic element. In 3' UTRs, for example, the team saw ALB mutations in liver and digestive tract cancers, FOXA1 mutations in prostate cancer, NFKBIZ mutations in lymphomas, SFTPB mutations in lung cancer, and TOB1 mutations in carcinomas. Expression data showed that mutations in NFKBIZ were subject to copy gains and increased expression in certain lymphomas, suggesting that it's a true driver gene. Likewise, TOB1 was found to be amplified in the focal region in breast cancer, suggesting it's a true driver. The researchers found similar mutations in each genomic element they studied, many of which were suggested to be true driver mutations by the expression data.
Even with 3,000 whole genomes to study, Getz said, researchers are still far from completing the landscape of drivers, especially non-coding drivers. Despite this, however, they were still able to find mutations using their pan-cancer method that other analyses would have missed. Further understanding of the functional role of non-coding regions is needed to improve methodologies, he added.
Finally, Johns Hopkins researcher Alexander Baras detailed his work comparing small and large gene panels to analyze tumor mutational burden.
Total mutational burden has been shown to correlate with response to immune checkpoint inhibition, he said. With that in mind, Baras and his team used de-identified sequences to look at the non-silent total somatic mutational burden, comparing the power of whole-exome sequencing, large gene panels, and smaller targeted panels to determine mutational burden.
What they found is that hotspot panels are able to detect more mutations per megabase sequenced than smaller targeted panels. In fact, they compare favorably to exome sequencing data from The Cancer Genome Atlas.
The team characterized tumor mutational burden in sequencing data from 14,000 samples tested on large gene panels and smaller targeted panels, and stratified tumor mutational burden by the number of mutations in hotspot regions. Tumors that have no hotspot mutations tend to have few samples with high mutational burden, Baras said. Of the 14,000 samples, 2,000 did have high mutational burden, but only 130 or so were detected by smaller targeted panels.
The smaller panels just don't have enough power to detect high mutational burden, Baras concluded. But while exome sequencing is the gold standard for determining total mutational burden, large sequencing panels can be used to estimate tumor mutational burden in diverse tumor types and are very appropriate in the context of clinical testing, he said.