NEW YORK (GenomeWeb) – Methods to infer genome-wide mutational load without having to sequence the whole genome could serve as easier and cheaper ways to identify patients with highly mutated cancers who might be more likely to respond to immunotherapies, according to two presentations at last week's annual meeting of the American Association for Cancer Research in New Orleans.
As immunotherapy rapidly gains ground in the oncology community, the race is on to identify molecular markers or signatures that can best predict which patients will most likely benefit from these drugs.
One deceptively simple predictive biomarker that has emerged is a tumor's mutational burden, or mutational load. Patients whose tumors have many mutations — and corresponding high numbers of tumor neoantigens — appear more likely to respond to immunotherapies, both CTLA4 and PD1 inhibition, while those with few mutations appear less likely to respond, though important exceptions to the rule exist.
And while some cancers, like melanoma, are characterized as a group by high mutation rates, studies are increasingly showing that small subsets of almost any cancer type may have high mutational loads that predispose them to respond to immunotherapies. This provides a rationale for the potential clinical utility of assessing mutational load across cancers to identify likely responders to these drugs.
Measuring mutational load across the whole genome or exome is a cost-prohibitive and resource-intensive proposition, but two studies presented at AACR provide some early evidence that it is possible to infer the mutational landscape of the entire cancer genome by sequencing a much smaller panel of genes.
In one study, presented by Artur Veloso of the Novartis Institutes for Biomedical Research, investigators tackled the question of how mutations within a small genomic region — just hundreds of genes — might correlate with genome-wide mutation burden.
Veloso and his colleagues used mutation data from The Cancer Genome Atlas and compared exome-wide mutation frequency to numbers of mutations in just 315 genes. This comparison demonstrated a strong positive correlation between the total mutation burden and the mutational load within this subset of genes, he said.
With this encouraging finding in hand, he and his co-investigators went on to try to use gene panels to distinguish highly mutation-prone samples from non mutation-prone ones. First, they derived a cutoff point or threshold from the TCGA exome data — greater or fewer than 181 non-synonymous exome-wide mutations — that best distinguished microsatellite-unstable samples from microsatellite-stable samples with a 95 percent true positive rate.
A statistical analysis revealed that the 315-gene panel had excellent power to discriminate high and low mutation samples, with an area under the receiver operating characteristic curve ranging from 0.85 to 0.97 across indications, Veloso reported. However, he stressed, an optimal threshold for identifying high mutation load samples may vary by cancer type.
The group further explored how different panels — commercial panels, internal panels, and also simulated panels of different sizes — compared in their predictive ability. Overall, he said, the larger the panel, the closer it appears to recapitulate genome-wide mutation burden. The group did see, though, that once a panel gets to about 1,000 genes, the performance starts to plateau, and even 500 genes give a very good performance.
"Around 500 genes is a nice sweet spot," Veloso said. "You really don't need more genes than that."
The group also found that professionally curated panels were much more predictive than random panels, most likely due to their focus on commonly mutated genes.
The study didn't directly address patient response to immunotherapy, so the implications for diagnostics are oblique. However, Veloso said, the results at least demonstrate the feasibility of using mutation burden in a cancer gene panel as a biomarker for genome-wide mutation load or microsatellite instability, which numerous studies have now linked to immunotherapy response.
A second study, led by Foundation Medicine bioinformatics director Garrett Frampton, also looked at tumor mutation load, comparing Foundation Medicine's 315-gene FoundationOne panel to the whole tumor genome.
A spokesperson for the company told GenomeWeb that Foundation Medicine is pursuing several initiatives in the immuno-oncology space, and that analysis of tumor mutational burden is currently available for physicians who have used FoundationOne, but as a research-use only service.
In their study, Frampton and colleagues first used TCGA data to establish a correlation between mutation frequency in Foundation's panel and the whole exome. Frampton said his group is now collecting data from subsequent work to optimize a FoundationOne-based tumor mutation burden prediction method, and expects to publish the results soon.
In their initial work, he and his colleagues also looked more deeply at mutation load across common tumor types, identifying recurrent somatic mutations — such as loss of function mutations in mismatch repair genes (MSH2, MSH6, MLH1 and PMS2), DNA replication genes (POLD1, POLE), and in TP53 — that were individually associated with higher tumor mutation load.
According to Frampton, the data provide support for future use of FoundationOne to asses mutation burden as a marker for patient response to immunotherapies, either approved drugs or drugs in clinical trials. The study's broad characterization of mutational load across a variety of cancer types also supports an expansion of the patient population that could be eligible for these treatments, he said.
"The take home message for us here is really that nearly every different type of cancer, even those … where the overall tumor mutation burden for the disease tends to be low, has some cases with a high tumor mutation burden [so] there could be patients with a huge array of tumor types that may benefit from immunotherapy," he said.