NEW YORK (GenomeWeb) – Investigators at Rutgers Cancer Institute of New Jersey have developed a statistical method called the loss of heterozygosity-germline inference calculator (LOHGIC), which they claim is able to classify mutations in tumor suppressor genes as somatic or germline and predict gene loss in cancer cells using data from tumor-only deep-sequencing assays.
The method, which is described in a paper published recently in JCO Precision Oncology, is currently being used as part of the institute's precision medicine program to analyze assay results from patients with rare cancers. According to tests described in the paper, LOGHIC returned extremely accurate results when it was used to analyze sequencing data from 64 patients with pathogenic BRCA1/2 mutations.
Specifically, the software predicted germline versus somatic mutation status with high confidence in 88 percent of patients, 66 percent of whom were predicted to have germline variants. Of the 64 patients analyzed, 28 had genetic testing reports, and LOHGIC's predictions corroborated the data in these reports with 93 percent accuracy, 100 percent positive predictive value, and 96 percent true-positive rate. LOH was present in 76 percent of patients determined to have germline mutations and in 63 percent of the patients determined to have somatic mutations.
Additional findings include that germline mutations were found exclusively in tumors in women. Out of 15 female patients whose data was assessed, 87 percent were diagnosed with hereditary breast and ovarian cancer or fallopian tube cancers, while 73 percent had LOH mutants. In a separate dataset of 22 men and women, the researchers predicted LOH in 77 percent of patients all with germline mutations.
"We concentrate on patients with advanced and rare cancers because we feel these are the ones that have the greatest need for this kind of approach," said Shridar Ganesan, associate director for translational science and chief of molecular oncology at Rutgers Cancer Institute and associate professor of medicine and pharmacology at Rutgers Medical School. "Over the past several years we have established a process in which many of the patients here with rare and refractory cancers undergo CLIA-certified sequencing."
Like several other institutions have done, Rutgers established a precision medicine program about five years ago with an eye towards integrating clinical-grade sequencing into patient care. Their process involves profiling patients' genomes using Foundation Medicine's FoundationOne assay (which gained US Food and Drug Administration approval in November), as well as other CLIA-certified assays. Next, systems biologists review the results of the assay as does an internal tumor board comprised of clinicians, basic scientists, systems biologists, and early-phase clinical trial researchers with an eye towards identifying more effective treatment options based on sequencing results.
For the Rutgers researchers, the decision to use the FoundationOne assay, which detects genomic alterations in more than 300-cancer related genes in solid tumors, was a straightforward one, according to Ganesan, who is one of the lead authors on the LOHGIC paper. "When we started five years ago they were basically the only solution out there," he said. There are however other testing platforms available now that Rutgers uses or has at least tested including a tumor assay offered by NantOmics.
One of the benefits of that initial agreement with Foundation Medicine was that it included providing Rutgers researchers with access to lot of the raw sequence data generated from patients, according to Kim Hirshfield, an assistant professor of medicine at Rutgers Medical School and another lead author on the paper. "It's the access to the raw data and exploring that data that has allowed us to arrive at [tools]such as LOHGIC," she said in an interview. It allows "[us] to utilize more of the information about patients than what we see in the reports."
Data from assays such as FoundationOne are fed into LOHGIC which combines information on tumor purity, allele frequency, and gene copy number to distinguish between somatic mutations, which are only found in tumor cells, and germline mutations which are present in both normal and tumor cells. Inherited germline mutations contribute to the development of 5 percent to 10 percent of all cancers many of which have undergone LOH. One of these, hereditary breast and ovarian cancer, or HBOC, syndrome is among the most common familial cancer syndromes, with a prevalence of one in 400 to one in 500 in the general population, according to the researchers.
"When we see an alteration especially in a tumor suppressor gene, it raises multiple questions," Ganesan said. Besides classifying it as a germline or somatic variant, "is it a germline variant has it undergone LOH? … That was an important question that isn't answered by just the report." Equally important is being able to determine whether the mutation in question is a true tumor driver or a simply a bystander, he added.
The answers to these questions are crucial for therapy selection, according to the Rutgers researchers. The paper, which focuses on analyzing BRCA 1/2 mutations, notes that inherited mutations in the BRCA1/2 genes contribute to 1 percent to 5 percent of breast cancers most of which undergo LOH. Studies have shown these types of tumors are sensitive to PARP inhibitors and platinum treatment. Thus, detecting and classifying variants early as well as LOH determination is crucial for making treatment decisions.
But LOH information is not always readily available from existing assays, according to the researchers. By way of example, they note that if a male patient with a known pathogenic germline BRCA2 mutation develops a tumor that is not typically associated with the BRCA2-linked cancers, standard approaches will be unable to determine whether the cancer cells have undergone LOH at BRCA2 and will be sensitive to PARP inhibitors. LOHGIC is designed to make this sort of analysis possible, Ganesan said. "Obviously there are cases where we can't conclude any of these with certainty and there are cases where we can."
Furthermore, "now that we have access to deeper data, it's easier to implement algorithms like [LOHGIC] that can include uncertainty in those measurements," Hossein Khiabanian, an assistant professor of pathology and laboratory medicine at Rutgers Medical School and one of the corresponding authors on the work, said in an interview. "And [as] we add more data, we can include more information about accuracy of all the measurements." Assessing LOH in this manner would have been more challenging with older algorithms developed for earlier sequencing technologies which did not provide the depth and breadth of information that current sequencing instruments can provide, he noted.
The ability to characterize variants based on LOH has implications not just for treatment of the immediate tumors but also for testing patients and family members for predisposition to other kinds of cancers associated with LOH in BRCA1/2, Hirshfield noted. It could also be useful in cases where patients have limited information on their family history. "A lot of those individuals present with breast cancer that often mimics what sporadic breast cancers look like," she said. "If [these] show up in testing … and we suggest it to LOHGIC, we may find out that there is a high probability that they in fact carry what looks like a germline BRCA mutation and that may be the driver for their cancer. And that would then lead us to test that patient using a CLIA-certified test for germline BRCA mutations."
For now, the Rutgers researchers are using LOHGIC simply to understand sequencing results and provide guidance for additional testing but not in direct patient care and treatment, Ganesan stressed. "These interpretations are not directly CLIA-certified," he said. "We have to be very clear that these are in the setting of research and to help us evaluate … and guide patients towards CLIA-certified tests." But in the future, this kind of analysis could become more and more routine, he added.