NEW YORK – Hoping to keep up with an increasing number of clinical assays for DNA repair deficiency, nonprofit Friends of Cancer Research is midway through a project aimed at harmonizing existing tests and developing best practices for new entrants.
Hillary Stires, FOCR’s director of regulatory and research partnerships, presented early data from the group’s effort on Saturday at the Association for Molecular Pathology annual meeting. Comparing a subset of available assays that measure homologous recombination deficiency, or HRD, FOCR found not only that assays vary significantly in their content and methodology, but also that they are far from perfectly concordant in their HRD calls.
The HRD harmonization effort follows in the footsteps of similar FOCR projects, namely the group’s analogous program seeking to align tumor mutational burden tests.
HRD has quickly become a crucial biomarker for guiding the use of PARP inhibitors, and could also inform the use of new drugs on the horizon with similar DNA repair-inhibiting actions.
According to some clinicians, HRD presents an even trickier challenge for harmonization than TMB, as assays differ not only in their technical details, but also in how they fundamentally define HRD.
As with the TMB project, FOCR sought to involve a range of commercial assay developers. Investigators from Sophia Genetics, Thermo Fisher Scientific, MD Anderson, Omniseq, Tempus, DNAnexus, Bionano Genomics, Foundation Medicine, Invitae, NeoGenomics, Amoy Diagnostics, and Guardant Health were listed as coauthors on the poster presented at AMP.
“We know that homologous recombination deficiency is defined by several causes —inactivation of genes in the HR pathway, including BRCA1 and 2 — as well as consequences, or genomic instability — reflected in things like loss of heterozygosity,” but prior to the FOCR effort, methodological variability across HRD assays had not been deeply investigated, Stires said at the meeting.
Phase I of the project led to a publication in The Oncologist this January describing this diversity among assays and proposing common language for the effort’s next steps.
Phase II, which Stires discussed at the meeting, was a first step at assay alignment.
Based on a survey of 20 test makers, FOCR concluded that virtually all assays include BRCA 1 and 2 inactivation, but none interrogate methylation. According to the group, genomic loss of heterozygosity is a feature of about three-quarters of tests, but most assays use multiple factors. The team also learned that both scores and cutoffs for HRD and [loss of heterozygosity] range quite a bit across assays.
Other non-BRCA pathway genes, telomeric allelic imbalance, and large-scale transitions each featured in about half of the surveyed tests.
FOCR investigators then compared assay calls across 11 participating vendors who they supplied with a set of about 340 matched ovarian cancer samples from The Cancer Genome Atlas (TCGA).
Test makers ran their examples through their own HRD pipeline and reported a yes or no result. They also described whether the calls they made were based on HRD causes, like BRCA mutations, or consequences.
According to Stires, there was a wide range of HRD positivity in the samples, from 9 percent to 67 percent with a median of 49 percent and a mean of 44 percent.
To assess assay agreement, the FOCR investigators compared the assay calls in pairs. The collaborators had agreed as a group not to use reference standards, so each test was compared to every other test in the study.
“This analysis doesn’t tell you one test is better than another, but it provides directionality toward concordance,” Stires said.
By her measure, the overall agreement among assays was “modest to high.”
The team calculated concordance scores for each assay pair, with zero representing opposite results and one indicating matching results and averaged them across the samples. Stiles said that group did this not only for HRD but also for concordance based on causes and concordance based on consequences specifically.
Overall, assays were about 75 percent concordant, on average, in their HR status calls, up to 87 percent when assessing causes and just 68 percent when using consequences.
Although the inter-assay agreement on HRD status was variable, Stires said this did not appear to be strongly driven by the factors included in a particular test’s call. This underscores the importance of developing best practices, she said.
The next step for FOCR is to follow this up with a similar study using freshly extracted FFPE samples, which is currently planned to begin in early 2023.
“The value of that dataset is that as we have outcomes data … we’ll also use [that] for some exploratory analyses,” Stires said. This should help resolve any bias that resulted in the group’s first comparison from the fact that many assays were initially trained on TCGA data, as well as allow for some initial assessment of how assays compare not just in their calls, but also in terms of how well they predict patient outcomes.