NEW YORK (GenomeWeb) – Using a novel algorithmic approach, a Harvard Medical School team believes that it can advance a test to predict PARP inhibitor response using data from standard targeted sequencing panels being implemented increasingly to guide molecularly targeted drugs.
Investigators published early results from their experiments with the method in Nature Genetics this week, describing it as using a "likelihood-based approach" that allows the detection of a previously defined homologous repair deficiency (HRD) signature from much smaller groups of mutations.
PARP inhibitors are currently approved for several indications, and are being explored in numerous clinical trials for others. But so far, the only sanctioned companion diagnostic markers in the space are BRCA mutations. The US Food and Drug Administration has endorsed Myriad Genetics' BRACAnalysis CDx test to guide treatment with multiple drugs in several indications, as well as Foundation Medicine’s FoundationOne CDx for treatment of BRCA 1/2 positive ovarian cancer with rucaparib.
But evidence has been available for several years that BRCA mutations are not the only cause of the HRD that seems to be at the heart of how these therapies work. As a result, it's likely that there are individuals without these mutations who could also respond to such drugs but are not receiving them.
As PARP therapies have moved into indications beyond breast cancer, diagnostics firms are trying to advance tools to capture more potential responders. Myriad for example, offers a broader test called myChoice HRD in addition to its approved BRCA companion diagnostic. But so far the firm's HRD assay has not shown itself predictive enough to be required as a companion test.
Despite this, the firm still continues to pursue CDx status for myChoice HRD, most recently announcing that it had submitted the first module of its premarket approval application to the FDA this month seeking approval of the test for alongside GlaxoSmithKline's niraparib (Zejula) as a fourth-line option.
MyChoice HRD, though not limited to BRCA, still focuses on specific genomic targets, described by Myriad as "genomic scars," including heterozygosity loss, telomeric allelic imbalance, and large-scale state transitions.
In contrast, the approach the Harvard group is working to advance is an algorithmic method that is independent of specific gene mutations or features, and instead combines signals from the overall mutational landscape to classify tumors as HR deficient or not.
The group's work is based on evidence from prior studies using genome-wide sequencing, which found that HRD-like signals can be present in tumors without BRCA alterations.
Although data on response to PARP drugs in tumors with this broader category of HRD-ness is limited so far, there has been some early evidence that overall HRD may be predictive. Used in ovarian cancer, for example PARP inhibitors have shown effectiveness regardless of BRCA 1/2 mutation, reflective of widespread deficiency in the HR pathway in these tumors. Other data has linked genome-wide HRD signatures to chemotherapy response.
According to Peter Park, a Harvard professor of biomedical informatics and a senior author on the Nature Genetics study, one stumbling block in advancing mutation-agnostic HRD signatures to the clinic has been a requirement for whole-exome or whole-genome sequencing.
But with smaller targeted sequencing panels now increasingly being used to identify patients eligible for drugs that target mutations in genes like EGFR, ALK, BRAF, and others, Park and his colleagues hoped that they could make the strategy much more translatable by finding a way to translate exome-wide HRD analysis to smaller sections of the genome.
Park and his colleagues began by training an algorithm they call SigMA (Signature Mutational Analysis) on thousands of whole tumor genomes, in order to glean a previously determined genome-wide HRD signature called 'signature 3' (Sig3) from a much smaller DNA parcel.
According to the authors SigMA combines elements of other signal analysis approaches with "new measures for associating mutations to signatures," replacing what the group says is an error-prone "spectrum decomposition step" with a "clustering step, using the rich resource of existing WGS data that inform us of the co-occurring signatures and their relative contributions to a given tumor type and its subtypes."
"When the mutation count is low, this is a more stable approach for inferring a combination of signatures present in a sample than performing a linear decomposition directly," the team argued.
To test the accuracy of their model, Park and colleagues measured SigMA's performance against 730 breast cancer samples that had previous whole-genome sequencing, simulating what the algorithm would yield if applied in two widely used targeted panels: Memorial Sloan Kettering's MSK-IMPACT and Foundation Medicine's FoundationOne.
Despite the fact that it used far fewer genes, the SigMA model correctly identified 163 of 221 samples with known genome-wide HR deficiency in the MSK-IMPACT panel, a 74 percent accuracy rate. Using the Foundation panel, accuracy was a bit lower at 68 percent.
According to the group, the SigMA approach can also be tuned to lower the false positive rates at the cost of lower sensitivity. But even at a false positive rate cutoff of 1 percent, the authors still calculated that the use of SIgMA would double the number of breast cancer patients determined to have HR deficiency compared to BRCA 1/2 status in this cohort.
To further gauge SigMA's performance the researchers then applied the algorithm to another 878 breast tumor samples from patients who had been tested using the MSK-ACCESS panel. The test identified 202 cases (23 percent of the tumor samples) as bearing the mark of HR deficiency. Using the more stringent cutoff point, 121 cases would pass selection, the group wrote.
For other tumor types that present with fewer mutations overall, Park and his team wrote, this approach to predicting HRD is more challenging, since there simply might not be enough mutations present in a smaller sequencing panel to assess whether Sig3 is present or not.
Nonetheless, the team did study other cancer samples, and were able to detect the HRD signal at various rates — about 38 percent in ovarian cancers and 5 percent in esophageal carcinomas, for example.
Park argued that even if the approach identifies relatively few individuals, as long as these cases are responsive to PARP inhibition, the use of the test would have a positive impact.
However, as the number of publicly available fully sequenced genomes grows, the algorithm could be trained on more tumor types to detect a greater variety of genetic mutations.
This question of response still remains largely unanswered, though. To add some data in this vein, the group also performed experiments on 383 tumor cell lines from 14 cancer types treated with four PARP inhibitors. Breast cancer cell lines identified by the SigMA algorithm as being Sig3 HRD-positive responded better to the olaparib than cells that were Sig3-negative. A similar effect was observed in breast cancer cell lines treated with three other drugs and in other tumor types.
To check that the observed effect was specific to PARP inhibitors and not due to some drug-agnostic feature, the investigators also challenged the cell lines with other drugs. The PARP inhibitors remained among those with the best efficacy, along with DNA double-strand break-inducing agents, DNA cross-linkers, and anthracycline chemotherapy.
Whether or not the approach actually identifies greater numbers of patients who respond to these drugs will require, at the very least, retrospective studies in which the team can assay patient samples that have corresponding outcome results from PARP trials. Ideally, prospective validation would then follow.
Park said that response to the initial study has been positive, and he and his group are in contact with several potential collaborators to apply SigMA in this vein.