Development of a methylated gene panel for detection of cell-free, metastatic prostate cancer DNA in plasma

Purpose: We sought to develop a highly sensitive, reproducible cell-free methylated DNA assay for accurately detecting metastatic prostate cancer (mPCa) in plasma. Experimental Design: We identified a new 10-marker panel for accurate detection of PCa in the TCGA prostate adenocarcinoma (PRAD) 450K DNA methylation data. Marker sensitivity and specificity were verified in two independent public prostate cancer datasets (GSE112047, GSE76938). A quantitative multiplex methylation-specific PCR (QM-MSP) assay incorporating these markers was validated in archival benign and prostate cancer tissues from the United States (US, 61 cancer, 26 benign) and South Africa (SA, 80 cancer, 24 benign). We used a newly developed cell-free multiplex methylation-specific PCR (cf-MMSP) assay for the 10-marker panel. Intra-assay and intra-operator reproducibility assays tested its robustness. cf-MMSP performance was assessed in a Training set (40 benign, 40 cancer), and an independent Test set (19 benign, 20 cancer) of plasma. Shapley value regression was used to rank performance of the 10 genes in the cf-MMSP assay. We selected a 5-gene panel based on their high performance . Statistical methods used included the Youden index threshold for ROC analysis and Kruskal-Wallis tests. Results: In tissues, the 10-gene panel detected all stages of prostate cancer: In tissue from the US (n=87), we achieved ROC AUC = 1.00 (CI. 0.99 - 1.00), sensitivity of 98.36% (CI. 91.28 - 99.91) and specificity of 100.00% (CI. 87.13 - 100.00); in SA samples (n=104), ROC AUC was 0.99 (CI. 0.99 - 1.00), sensitivity of 98.28% (CI. 90.86 - 99.91), and specificity of 100.00% (CI. 86.20 - 100.00). cf-MMSP intra-assay test showed low CVs (1.70-5.70% across copy levels) and strong linearity (R² = 0.93). The intra-operator test showed excellent reproducibility, with an ICC of 0.99. In plasma, in the training set, the 10-gene panel showed ROC AUC of 0.89 (CI. 0.81-0.97), sensitivity of 75.00% (CI. 59.81 - 85.81) and specificity of 93.88% (CI: 83.48 - 97.90) in distinguishing mPCa from benign. In the test set, the assay performed with ROC AUC of 0.91 (CI. 0.80 -1.00), sensitivity of 85.00% (CI. 63.96 - 94.76), and specificity of 93.88% (CI. 83.48 - 97.90). Using a standardized Shapley value cutoff of 0.07, we identified the five best-performing genes- HAAO, DLEC1, FBN1, TMEM155 and RTDR1. In the training set of plasma, the refined 5-gene panel subset showed ROC AUC of 0.92 (CI. 0.84 - 0.99), sensitivity of 90.00% (CI. 76.95 - 96.04) and specificity of 90.00% (CI. 76.95 - 96.04). In the test set, this 5-gene panel achieved ROC AUC of 0.92 (CI. 0.84 - 0.99), with a sensitivity of 90.00% (CI. 80.00 - 100.00) and a specificity of 84.20% (CI. 53.00 - 100.00). Conclusions: The 5-gene cf-MMSP panel showed strong performance indices in detecting mPCa in plasma. The assay shows potential for broadly accessible early detection and disease monitoring for molecular management of mPCa.