WASHINGTON (GenomeWeb) – The first results from the American Association for Cancer Research's Project GENIE (Genomics Evidence Neoplasia Information Exchange) have upheld some of the hopes and expectations for this type of large-scale data-sharing effort — yielding new insights into the landscape of mutations across cancers and how they influence patient response to targeted therapies.
Researchers shared analyses from the project at AACR's annual meeting this week, including some insights from an overview of the landscape of mutations in the 19,000 patients included in GENIE's first public release, which took place this past January.
Charles Sawyers, chairperson of the AACR Project GENIE steering committee, said in an interview that the group is expecting to publish this landscape analysis very soon in more detail. But a few details were discussed during the meeting this week.
"The session [on pan-cancer analyses] started with a focus on these really detailed analyses of whole-exome or genome data, looking for new recurrent alterations that can explain cancers," Sawyers said. "But then we shifted to these four short talks from GENIE … where in contrast, we don't find any new cancer drivers, but we do find things we didn't know before, or didn't understand as well."
One example is the frequencies of certain mutations in the GENIE cohort and the contrast with what has been seen in large cancer sequencing programs like TCGA and ICGC. Individuals in those studies were largely untreated, Sawyers said, whereas the GENIE cohort reflects current clinical practice where most sequenced patients have later-stage disease and have been treated sometimes with multiple therapies.
"As you would expect, we saw that there are different spectrum of mutations that show up," he said. One is estrogen receptor mutations in breast cancer, which are rare in the TCGA dataset, but were relatively frequent in GENIE — about 15 to 20 percent.
"We'd expect that because this is a selection of women with metastatic disease who relapsed on hormonal therapy. So, it's kind of like a positive control that everything is working," Sawyers said.
Another less obvious and more experimental finding was a subset of glioblastoma patients who had much higher tumor mutation loads than analyses like TCGA have found. "The median tumor mutation burden was about the same as TCGA, but there was this group of outliers at the top that weren’t seen in [that study]," he said.
One hypothesis for the difference might be that many patients were likely treated with temozolomide, which would induce DNA damage. "But why would it just be this subset and not the whole population? We have to study that now," Sawyers said. "And the way GENIE is set up, this is data with real clinical records we can go back to and check, so that's how we do that."
In another presentation on GENIE's first 19,000 subjects, investigators discussed a project integrating and reconciling different clinical actionability tools — Vanderbilt-Ingram Cancer Center's My Cancer Genome, MD Anderson's Personalized Cancer Therapy, and Memorial Sloan Kettering's OncoKB.
Overall, the combined knowledgebases identified a therapeutic option for more than 33 percent of the cohort. Approximately 15 percent reflected standard-of-care therapies, and about 8 percent were at the investigational level. The remaining matches were more exploratory.
According to the researchers, the most frequently matching diagnoses at the standard-of-care level were non-small cell lung cancers, breast cancer, and melanoma.
Looking at clinical trial matching, about 84 percent of the cohort matched to some biomarker-driven clinical trial including studies exploring the impact of mutations along an entire cell signaling pathway. Though high levels of patient benefit shouldn’t necessarily be assumed considering the exploratory nature of many of these trials, the analysis speaks to the potential benefit in more broadly screening caner patients for targetable mutations.
For example, Sawyers said, researchers ran a simulation using reported data on recruitment patterns in the the NCI MATCH trial , and were able to exactly predict the number of patients needed to be screened to fill the trial.
"We could fill the MATCH trial from the entire GENIE dataset, no problem, every single cohort," Sawyers said.
"So if you are planning a trial like that, that's very valuable data to help estimate how many patients you might need to screen, and it's also a pretty strong argument to prescreen large numbers of patients as a society in order to accrue trials much faster," he added.
Yet another analysis of the recently released data looked at the landscape of ERBB2 mutations in the GENIE database, comparing the frequency of different mutations to data from an ongoing Phase II basket trial called SUMMIT, which is investigating the pan-HER inhibitor neratinib in HER2- (ERBB2-) mutant solid tumors.
According to the investigators, ERBB2 mutations were present in 2.8 percent of the GENIE cohort — 519 patients.
The tumor types with the highest proportion of ERBB2 mutations were bladder, breast, colorectal, and NSCLC, and among patients with copy number data available, about 11 percent had concurrent ERBB2 amplification, most often in breast cancer, the researchers reported.
More importantly, the frequencies of ERBB2 mutant cancer types observed in the GENIE cohort were generally comparable to the patients enrolled in SUMMIT. At the variant level, S310F/Y mutations were less prevalent in GENIE compared to the neratinib study, but a range of other important mutations seemed to be similarly frequent.
According to the researchers, the results provide direct evidence that basket study enrollment like that used in SUMMIT seems to accurately capture the true landscape of a particular alteration.
According to Sawyers, the results also highlight some additional cancers not recruited to SUMMIT that could benefit from neratinib, for example cholangiocarcinoma.
These types of discoveries are what researchers hoped would come out of the amassing of larger datasets through sharing efforts like GENIE.
Based on yearly rates of sequencing at each of the eight founding institutions, together with the planned addition of new members, representatives from GENIE estimated that the database could grow to more than 100,000 samples within five years.