OvaPrint-A Cell-free DNA Methylation Liquid Biopsy for the Risk Assessment of High-grade Serous Ovarian Cancer

Abstract

Purpose: High-grade serous ovarian carcinoma (HGSOC) is the most lethal epithelial ovarian cancer (EOC) and is often diagnosed at late stage. In women with a known pelvic mass, surgery followed by pathologic assessment is the most reliable way to diagnose EOC and there are still no effective screening tools in asymptomatic women. In the current study, we developed a cell-free DNA (cfDNA) methylation liquid biopsy for the risk assessment of early-stage HGSOC.

Experimental design: We performed reduced representation bisulfite sequencing to identify differentially methylated regions (DMR) between HGSOC and normal ovarian and fallopian tube tissue. Next, we performed hybridization probe capture for 1,677 DMRs and constructed a classifier (OvaPrint) on an independent set of cfDNA samples to discriminate HGSOC from benign masses. We also analyzed a series of non-HGSOC EOC, including low-grade and borderline samples to assess the generalizability of OvaPrint. A total of 372 samples (tissue n = 59, plasma n = 313) were analyzed in this study.

Results: OvaPrint achieved a positive predictive value of 95% and a negative predictive value of 88% for discriminating HGSOC from benign masses, surpassing other commercial tests. OvaPrint was less sensitive for non-HGSOC EOC, albeit it may have potential utility for identifying low-grade and borderline tumors with higher malignant potential.

Conclusions: OvaPrint is a highly sensitive and specific test that can be used for the risk assessment of HGSOC in symptomatic women. Prospective studies are warranted to validate OvaPrint for HGSOC and further develop it for non-HGSOC EOC histotypes in both symptomatic and asymptomatic women with adnexal masses.

Figure 1.

Sample summary and study workflow. Figure shows all samples used in the study. Samples are subdivided by sample type (tissue, cfDNA) and diagnosis (HGSOC, normal/benign, other EOC). The assay type for each sample (RRBS and hybridization probe capture) is shown. Samples excluded are indicated.

Figure 2.

OvaPrint feature discovery using RRBS in HGSOC and normal tissue. A, Left pie chart shows the number of HGSOC, normal ovarian, and normal fallopian tissues analyzed by RRBS; right pie chart shows the breakdown by stage of all HGSOC samples. B, UMAP showing separation of HGSOC from normal fallopian and normal ovarian tissue using beta values from 1,677 DMRs found during feature discovery. C, Heat map showing separation of HGSOC from normal fallopian and normal ovarian tissue using beta values from all 1,677 regions found during feature discovery; annotation bars show HGSOC stage, HGSOC cluster, and sample group. D and E, Beta values from an external dataset published by Bosquet and colleagues (ref. ; GEO Accession: GSE133556). We calculated mean beta value for each sample across 383 hypermethylated and 389 hypomethylated DMRs that overlapped at least one EPIC array probe. The boxplots show beta values in HGSOC and normal fallopian tubes in D (Wilcoxon P value < 0.0001 for both comparisons). Mean Δβ for each DMR across all tumor and all normal samples from GSE133556 correlates with the Δβ we observed in our RRBS samples (E).

Figure 3.

HGSOC DNA methylation profiles more closely resemble fallopian tube tissue rather than ovarian tissue. UMAPs representing genome-wide methylation profiles of HGSOC and normal fallopian and ovarian tissues. The UMAPs are generated from genome wide methylation profiles of each sample at 2,231,688 CpGs. UMAPs are colored by sample type (A) and tumor stage (B).

Figure 4.

Performance of risk assessment classifier for HGSOC versus benign pelvic masses. A, ROC curve indicating the performance of OvaPrint to discriminate between benign pelvic masses and HGSOC. AUC with 95% CI is annotated. B, Boxplot showing OvaPrint scores of HGSOC and benign samples. Each point represents one sample; points are colored by risk category (blue = high, green = moderate, red = low); the red band indicates the moderate risk band.

Figure 5.

HGSOC versus normal as evidence for screening utility. Sample performance visualized by UMAPs and box plot. UMAPs of all HGSOC and benign samples run on OvaPrint using beta values from the top 25 most important regions (A). Each sample is colored by sample group (benign, HGSOC, normal). B, Boxplot showing OvaPrint scores of HGSOC, benign, and normal samples. Each point represents one sample; points are colored by risk category (blue = high, green = moderate, red = low); the red band indicates the moderate risk band.

Figure 6.

Generalizability of OvaPrint in non-HGSOC EOC histotypes. Boxplots of OvaPrint scores from cfDNA samples with non-HGSOC histotypes of EOC. A shows high-grade and low-grade tumors; B shows all borderline cases. Each point represents one sample; boxplots and points are colored by tumor grade in A.