PALB2 and CHEK2 variants contribute to familial breast and ovarian cancer risk

Credit: Povozniuk/Getty

Around 4% of breast cancer cases are consistent with inheritance of a high-risk pathogenic germline variant (PGV). BRCA1 and BRCA2 account for about 2% of breast cancer incidence. But other rare variants also contribute risk. In this issue, Woodward and colleagues examine the contribution of PALB2 and CHEK2_1100delC PGVs to familial breast and ovarian cancer. More than 3,000 women with histologically confirmed diagnosis of invasive or in situ breast cancer or epithelial nonmucinous ovarian cancer referred for genetic testing underwent germline testing of BRCA1, BRCA2, PALB2, and CHEK2_1100delC for PGVs. About 1,500 women without a breast cancer diagnosis served as controls. Sequencing uncovered 35 (1.1%) PGVs in PALB2 and 44 (1.4%) in CHEK2_1100delC in breast and/or ovarian cancer cases. In contrast, only 3 (0.2%) and 5 (0.3%) PGVs were found in controls for PALB2 and CHEK2_1100delC, respectively. Nearly a third of tumors in patients with PGVs in PALB2 (8/25) were triple negative (TNT). When the researchers then considered all patients with PALB2-associated breast cancers known to their service, they found that the TNT phenotype occurred in 28% of cases and an ER-positive, HER2-negative phenotype occurred in 35%. A grade 3 phenotype occurred in more than three-quarters of individuals (76%), regardless of receptor type. The researchers calculated Manchester scores for all affected women. They found that the likelihood of a PALB2 PGV increased with increasing Manchester score, but rates of CHEK2_1100delC did not. Altogether the findings indicate that PALB2 and CHEK2_1100delC PGVs account for nearly 2.5% of familial breast and ovarian cancer risk. The authors conclude that detecting PALB2 and CHEK2_1100delC PGVs is critical for accurate breast and ovarian cancer risk assessment. —V. L. Dengler, News Editor

InpherNet accelerates causative gene detection for monogenic disease diagnosis

Using genome sequencing to diagnose monogenic diseases can be a time-consuming task for clinicians. Sequencing can produce large numbers of candidate causative genes, and when knowledge about a gene’s pathogenicity is lacking, clinicians spend valuable time hunting through the literature for indirect evidence that a given gene is a plausible hypothesis for causing a patient’s disease. Yoo and colleagues present a network-based machine learning gene-ranking method called InpherNet. InpherNet predicts causative genes using variant information and four sources of indirect evidence: phenotypes associated with orthologs, paralogs, members of the same functional pathway, and expression colocalized interaction partners. The researchers based the network on human, mouse, and zebrafish Ensembl gene sets in combination with nine ontology sources, including species-specific phenotype databases such as Human Phenotype Ontology, cross-species pathway data from Reactome, and potential human protein–protein interaction data from BioGRID. To train the candidate gene classifier, the researchers constructed synthetic patients by adding randomly selected pathogenic variants from ClinVar for a known OMIM disease to a 1000 Genomes Project genome. They then associated the patient with a subset of disease-associated phenotypes. In total, the researchers trained InpherNet on more than 2,500 synthetic patients with nine phenotypes and 300 candidate genes per patient. The team tested InpherNet on actual patients with prediagnosis phenotypes and clinician-verified monogenic diagnoses whose causative genes were given incorrectly low rankings by other tools. InpherNet outperformed existing ranking approaches. For example, InpherNet correctly ranked the causative gene 1, at the top, in over three times as many cases as the comparable tool, Phrank. The authors conclude that InpherNet accelerates identification of a causative gene where direct information is insufficient by leveraging indirect evidence from human and model organism data. —V. L. Dengler, News Editor