Skip to main content

Dana-Farber Team Developing MicroRNA Liquid Biopsy to Diagnose Ovarian Cancer

Premium

SAN FRANCISCO (GenomeWeb) – Researchers from the Dana-Farber Cancer Institute and Brigham & Women's Hospital are developing a liquid biopsy assay that analyzes circulating microRNA to try and come up with a better method for diagnosing early-stage ovarian cancer.

Dipanjan Chowdhury, chief of the division of radiation and genome stability at Dana-Farber, said that there is a critical need for better tools to catch ovarian cancer early. When diagnosed at stage I, he said, the five-year survival rate is 90 percent. Unfortunately, nearly 70 percent of women are diagnosed at stage III or IV, when the five-year survival rate drops to just 30 percent, he said. "That's really striking," and points to the idea that "we have the therapeutics to cure it, we're just not detecting it early enough," he added.

In a study published recently in eLife, Chowdhury's team, which has studied microRNAs and their role in DNA repair in cancer, collaborated with researchers at Brigham & Women's Hospital who had access to a large cohort of ovarian cancer samples.

The researchers used next-generation sequencing and neural networks to identify a set of seven microRNAs that could distinguish between women with ovarian cancer and those without cancer.

The next step, Chowdhury said, is to validate those microRNAs prospectively on larger numbers of samples to better understand the test's sensitivity for detecting early-stage cancer. For that, he said the researchers plan to leverage banked samples from two large cohorts: the National Institute's of Health's Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial and Partners HealthCare's biobank. Chowdhury said the team has access to longitudinal samples from 600 patients who were diagnosed with ovarian cancer and participated in the NIH PLCO trial, as well as from 500 ovarian cancer patients who donated samples to the Partners Biobank, including some who donated samples years before being diagnosed.

The goal is to make sure that the test can identify those women who had ovarian cancer and also to "see if we can detect the disease years before they were diagnosed," Chowdhury said.

Currently, ovarian cancer is diagnosed either through ultrasound or by analyzing levels of the CA125 protein, but both of those methods have high false-positive rates and are not very good at detecting early-stage cancer.

In the recent study, the team analyzed samples from 179 women from three prospective studies of ovarian cancer and age-matched healthy controls. They performed small RNA sequencing on the samples and randomly assigned subjects to either a training or testing set.

Next, the researchers applied machine learning tools and other statistical analyses to the microRNA-seq data to create an algorithm that could distinguish ovarian cancer from benign tumors, non-invasive tumors, and healthy controls. Ultimately, they found that the most robust approach was a neural network model that analyzed a set of 14 microRNAs. The neural network model was able to analyze interactions between the microRNAs and distinguish between cancer and non-cancer with an area under the curve (AUC) of 0.93 for the training set samples and 0.90 for the testing set samples. The researchers also confirmed that the model worked equally well on older and younger patients as well as for early and advanced disease.

For 120 subjects, there was data on serum levels of CA125. The neural network model significantly outperformed CA125 as a biomarker for cancer, and most notably, had significantly fewer false positives.

For a diagnostic test, Chowdhury said that the team planned to use qPCR to evaluate the miRNAs, rather than NGS, since qPCR would be more cost effective for such a small panel. "Ultimately, for a diagnostic, you have to have something that's easily doable," he added.

The researchers validated the qPCR version on 120 additional samples, recalibrating the algorithm to run on qPCR data. They also found that reducing the neural network from 14 miRNAs to seven improved its performance. After developing the final "locked-down" model, the team evaluated 325 samples — including both healthy controls and ovarian cancer samples.

In a final test of clinical performance, the group calculated the assay's positive predictive value to be 91.3 percent with a negative predictive value of 78.6 percent and an AUC of 0.85.

Chowdhury said that the group hopes that within the next two to three years it will complete the two larger studies on the banked samples from the PLCO and Partners biobanks, but that will largely depend on available funding. The plan is to run those trials as if they are retrospective, since there are longitudinal samples.

In addition, he said that he is also studying individuals with BRCA mutations who are at a higher risk of ovarian cancer. Depending also on a woman's family history, those with BRCA mutations are sometimes recommended to have preemptive surgery.

"Not every BRCA-positive woman goes on to develop ovarian cancer, but right now their only option is surgery," Chowdhury said. The goal, he said, is to develop a miRNA-based test to see whether among women who are already at a higher risk for ovarian cancer, there is a way to further stratify them to enable better decision making about who should have surgery and who should just be monitored. "This is very important, particularly if you're young and want to have children," he said.

The test could serve to both stratify those higher-risk patients and to monitor those initially deemed at lower risk. Though, he added that such a panel may have a slightly different miRNA composition than the one described in the eLife study.

A miRNA-based liquid biopsy could also serve to monitor patients post-surgery or after chemotherapy, in order to identify if there is residual disease or to predict relapse earlier.

Such applications have been pursued broadly in the liquid biopsy field, but most have so far focused on analyzing circulating tumor DNA, Chowdhury said. While such approaches have proven to be effective for lung and other cancers, there isn't a specific test for ovarian cancer, he said. "We wanted to look at this new class of molecules," he said.

Chowdhury also said that miRNAs could potentially serve as a biomarker to screen the general population. Such a test could be as common as a cholesterol test, he said. It would avoid invasive biopsies and should be relatively cheap, if it is based on qPCR.