NEW YORK – Researchers at the University of North Carolina have developed a 20-gene model for gauging whether patients' tumors are truly KRAS-dependent and are likely to respond to KRAS G12C inhibitors.
Although the predictive model hasn't yet been clinically tested, researchers led by Chad Pecot, an associate professor of medicine at the UNC oncology division, are hoping that drug companies developing KRAS inhibitors will evaluate the gene signature's ability to identify best responders to their KRAS inhibitor in clinical trials.
Pecot's team described their efforts to develop a gene signature, dubbed K20, in PLOS Computational Biology this month. They're hopeful that if the model's predictive abilities are further validated by drugmakers and other researchers, it could help doctors better identify patients most likely to benefit from KRAS G12C inhibitors and could even improve response rates among patients with KRAS G12C mutations.
Pecot noted that his team's research was spurred by the modest lung cancer response rates observed in clinical studies of the two approved KRAS G12C inhibitors, Amgen's Lumakras (sotorasib) and Mirati Therapeutics' Krazati (adagrasib), compared to other lung cancer targeted therapies.
"If you look at other oncogene-driven lung cancers such as EGFR-mutant, ALK fusion-positive, ROS1-rearranged, and RET-positive, the response rates with these newer inhibitors typically run about 60 percent to 70 percent," he said. "But when the response rates for sotorasib and adagrasib started coming in right around 40 percent, that was not a surprise to me, but it is disappointing to people who take care of lung cancer patients."
KRAS mutations are different from other targetable mutations in lung cancer, he explained. "Just having a KRAS mutation in a tumor, even if it's a known activating mutation such as KRAS G12C, that doesn't mean the tumor is dependent on that mutation," Pecot noted. "That's a huge difference than pretty much any other oncogene-driven cancer."
The study by Pecot's team builds on previous research that has shown that not all KRAS-mutant tumors are necessarily dependent on KRAS for proliferation and growth. That earlier research also yielded a gene-expression signature for determining whether tumors were KRAS dependent.
While the first-generation of KRAS inhibitors, Lumakras and Krazati, have demonstrated modest response rates among lung cancer patients, some of the next-generation drugs from companies like Loxo Oncology, Genentech, and Innovent are already demonstrating better activity in early-stage trials. As the KRAS inhibitor market becomes more crowded, however, there is a growing need for better tools to identify best responders to these different options.
Pecot's team developed the K20 model to better identify potential responders based on molecular profiles of cell lines from DEMETER2, an RNA interference (RNAi) screening dataset that combines data from the Broad Institute Project Achilles, Novartis Project DRIVE, and a large breast cancer dataset published in Cell in 2016. They then validated the model's ability to determine KRAS dependency in a dataset of KRAS G12C-mutant lung cancer cell lines treated with a G12C inhibitor and using data from The Cancer Genome Atlas (TCGA).
The K20 model, described in the PLOS Computational Biology paper, combines 20 features, including expression of 19 genes and KRAS mutation status, to determine if a tumor is KRAS-dependent. The researchers tested the model across 14 tumor types and in samples with and without KRAS G12C mutations.
Pecot's team concluded in their paper that the identification of a KRAS G12C mutation alone was "not sufficient to justify treatment with KRAS G12C inhibitors," in lung cancer cell lines. However, the K20 model was able to predict KRAS dependency in lung cancer cell lines harboring KRAS G12C mutations that were treated with a G12C inhibitor, ARS1620.
The researchers also evaluated KRAS dependency in tumor types with typically high rates of KRAS mutations, including lung, pancreatic, and colorectal tumor types. In colorectal cancers, the K20 model identified three of 93 tumors with KRAS mutations that were predicted to be resistant to KRAS inhibition. In pancreatic cancers, they found the K20 model could identify which tumors, such as those with copy number aberrations, may be more sensitive than others. In lung cancer, the K20 model identified a proportion of KRAS wild-type tumors that may have been sensitive to KRAS inhibition.
Researchers noted that these different tumor types were characterized by varying mutations and immune subtypes that affected KRAS dependency. In lung cancer, for instance, STK11, p53, and EGFR mutations played a role in KRAS dependency, while in pancreatic cancer tumors with immunogenic or aberrantly differentiated endocrine exocrine subtypes had higher KRAS dependency scores.
"The best predictor in this model of whether a tumor is KRAS dependent is the presence of a KRAS mutation, and the second strongest one was actually the expression level of KRAS," Pecot said. "The higher the level of KRAS expression, the higher the dependency the tumor had on KRAS. Because of that model, we actually found that some tumors that don't even have a KRAS mutation, but if they have high expression of KRAS wild type, they will be dependent on KRAS signaling."
Pecot noted that his team was unable to evaluate the model in clinical samples from patients treated with KRAS G12C inhibitors due to the lack of publicly available data in these patients. "In the end, what will really make this kind of a model powerful is if the companies that are evaluating KRAS inhibitors would test whether the K20 signature better predicts that the drugs will work compared to simply looking at whether there's a KRAS G12C or other KRAS mutation," he said. "There are certainly big pharma companies that already have the KRAS mutational status and the response data, so they'll be the ones capable of testing [the K20 model] out first."
He also hopes that other academic groups and clinical trial investigators will test out the model's predictive power.
Because of the modest response rates in clinical trials of KRAS inhibitors, Pecot believes a more nuanced method is needed to identify patients for treatment with these drugs. He recalled how EGFR inhibitors initially showed low responses in lung cancer patients, but after narrowing the best responder population to only EGFR-mutant patients, the drugs became a mainstay of lung cancer treatment.
"It's the same story unfolding in that we're not seeing the kind of responses we're typically used to seeing when a mutation is present," he explained. "KRAS [inhibitors] are going to have to follow a more refined type of precision medicine [approach] where it can't just be the presence of the mutation. We know that just because there's a KRAS mutation doesn't mean that there's KRAS dependency, and we need something beyond just the mutation alone."