Abstract

Background: Pan-cancer biomarkers of homologous recombination deficiency (HRD) are an area of unmet clinical need due to efforts to develop second generation PARP inhibitors across solid tumor cancers. An RNA-based method has the potential to overcome limitations of DNA scar-based signatures, which are less robust across indications and represent historical, static measures of HRD.

Methods: We developed a 1660 gene expression-based logistic regression signature using Tempus xR to predict HRD status in solid tumors, including BRCA-cancers (pancreatic, prostate, ovarian, and breast) and 35 other cancers. Training labels were defined with DNA data (Tempus xT) with BRCA1/2 biallelic loss as positive and wildtype (WT) in 14 homologous recombination repair (HRR) genes as negative. Data were split into training (N ~ 100k), evaluation for threshold selection (N~25k), and validation (N~25k).

Results: Sensitivity was 84% (ovarian), 82% (breast, pancreatic, prostate), and 54% for other cancers. Predicted HRD (HRD-RNA) prevalence was 27% in ovarian, 22% in breast, 14% in prostate, 8.9% in pancreatic and 5.4% in all other cancers. HRD-RNA prevalence was higher in HRR-biallelic altered tumors compared to HRR WT (31% vs. 11% in BRCA-cancers; 11% vs. 4.7% in other cancers). Among CCNE1-amplified ovarian tumors, thought to be mutually exclusive with HRD, HRD-RNA prevalence (12%) was lower than gwLOH-based calling (20%), highlighting improved specificity of this RNA-based approach. The model showed equivalent performance to cancer-specific models trained individually on pancreatic, ovarian, and prostate cancer, demonstrating generalizability. Model gene features were positively enriched for hallmark pathways associated with DNA damage and repair including E2F family of transcription factors, G2/M checkpoint, MYC targets, mitotic spindle, HRR, and DNA damage response pathways. Negative enrichment was seen in the epithelial to mesenchymal transition pathway. The association of ssGSEA scores of these hallmark pathways with HRD-RNA status was consistent across most individual cancers. Similar associations of these pathways with other models of HRD have been shown in publicly available data (PMID: 39073402). We also observed cancer-specific expression associated with HRD-RNA status. For example, BRCA2 expression is positively associated with HRD-RNA in ovarian cancer and negatively associated in most other cancers, namely in prostate cancer. Notably, there is a strong positive relationship with the oxidative phosphorylation pathway ssGSEA scores in ovarian cancer, consistent with reports that HRD results in a shift from glycolytic to oxidative metabolism (PMID: 32400970).

Conclusion: The RNA signature detects HRD within and beyond BRCA1/2 biallelic loss pan-cancer. Similar pathways were associated with HRD-RNA status as reported in DNA-based methods of HRD, highlighting the biological validity of an RNA-based HRD signal.

VIEW THE PUBLICATION