Kevin Elias

MD

Brigham and Women's Hospital

Director, Gynecologic Oncology Laboratory

PLCO (Learn more about this study)

2017-0023

Aug 3, 2017

Phase 3 validation of a serum miRNA neural network for early detection of ovarian cancer

While five-year survival rates exceed 90% for women with Stage I epithelial ovarian cancer (EOC), this contrasts sharply with the 25-30% five-year survival rates for women with advanced stage disease, which represents most cases. These statistics present a strong rationale for screening. The recent recognition that most high grade EOCs may begin in the distal fallopian tube rather than the ovarian surface prompted us to search for markers specific for a tubal origin. Using next-generation sequencing (NGS) to compare the expression of non-coding RNAs in immortalized benign fallopian tube secretory epithelial cell (FTSEC) lines versus high grade ovarian cancer cell lines, we discovered a unique microRNA (miRNA) signature that distinguished the EOC cells from their tubal precursors. Chief among these were members of the mir-200 family, which have been reported as elevated in cases of EOC. Using in situ hybridization, we identified that mir-200 members are highly expressed not only in ovarian tumors, but also in pre-invasive ovarian cancer precursor lesions called tubal intraepithelial carcinomas (TICs). We then tested the hypothesis that these miRNAs might circulate in patient serum. In our “Phase I” study, we again used NGS and identified circulating mir-200 members in samples from 100% of 60 women coming to surgery for a pelvic mass. For a Phase II study, we aimed to combine mir-200 with other circulating miRNAs identified by the sequencing platform into a reproducible algorithm for ovarian cancer diagnosis. To our original 60 samples, we added sequencing data from another 120 samples drawn from a pre-operative pelvic mass study. These samples were selected per specific diagnostic categories: serous cystadenoma, serous borderline tumor, endometrioma, Stage I/II serous ovarian cancer, Stage III/IV serous ovarian cancer, Stage I/II clear cell/endometrioid ovarian cancer, and Stage III/IV clear cell/endometrioid ovarian cancer. Specimens from healthy controls were also available. We used global sensitivity analyses to identify those miRNAs most contributing to the ability to discriminate benign from malignant masses and constructed a neural network algorithm for predicting the presence of EOC, dividing our samples into training and testing sets. We then validated this signature using qPCR. The final algorithm used a neural network analysis incorporating seven miRNA members with a ROC 0•97 (95%CI 0•91-1•00). In Phase 2b, we validated the neural network on an independent publicly available dataset from 454 patients with a wide range of diagnoses and confirmed that our miRNA neural network was highly predictive for EOC (ROC 0.92) with an overall sensitivity of 75% and specificity of 100%. The results to date have been submitted in a manuscript and are under review (at Lancet Oncology). We now propose a Phase III study to assess the potential diagnostic lead time for the algorithm as a screening test, utilizing the longitudinal nature of the PLCO trial to examine samples obtained months to years before a clinical EOC diagnosis.

Specific Aims
Aim 1. Using the penultimate sample before an ovarian cancer diagnosis, we will evaluate the miRNA neural network we have previously constructed for its utility in ovarian cancer detection.

Aim 2. Assuming a positive result for Aim 1 in this study, we will proceed with Aim 2. For those patients with multiple samples before EOC diagnosis, we will assess the longitudinal trajectory of serum miRNA to understand the possible diagnostic lead time for EOC. We also comprehensively profile miRNA expression in the serum samples using a panel of 752 miRNAs to look for additional biomarkers which might refine or inform our original neural network construction.