Tracking Peripheral T-Cell Repertoire Changes for Preoperative and Early Ovarian Cancer Diagnosis

Ovarian cancer is the most lethal female cancer. When the disease can be diagnosed at early stage, there is striking survival improvement (five year survival ≥ 90%), compared to late stages (≤ 40%). However, currently no early detection method for ovarian cancer has enough accuracy, and most tumors already progressed to advanced stages at diagnosis. Furthermore, over 70% of the adnexal masses detected on preoperative imaging are found to be benign after pelvic surgery. Current clinical tests rely on serum CA-125 and sonograms to diagnose the ovarian adnexal masses. However, CA-125 is elevated by many common benign conditions; and ultrasound imaging of ovary frequently misses small but malignant lesions. As a result, surgical removal of the lesion and histologic evaluation remains the only gold standard for diagnosis. These limitations dictate an urgent clinical need of a better preoperative diagnostic method with high detection accuracy, to lower the mortality rate, reduce unnecessary surgeries and preserve the life choices for many patients, especially young women at reproductive age planning for pregnancies. Here, we propose a completely different route to detect ovarian cancer signals from the blood T cell repertoire. This is feasible because the T lymphocytes recognize tumor antigens at initial stages, proliferate and alter the peripheral T cell repertoire. Therefore, detection of cancer-associated T cells (CAT) in the blood provides an exciting novel opportunity for non-invasive cancer diagnosis. However, no prior studies have achieved this goal because it is difficult to identify CAT in high-throughput, as most of the cancer antigens remain unknown. To prepare for this task, we developed the software TRUST and iSMART, to obtain antigen-specific TCRs from cancer datasets. These tools have enabled us to produce a large training set of CATs, which allowed us to identify diagnostic TCRs for the ovarian cancer patients. Following this result, we further developed DeepCAT, for pan-cancer prediction using blood TCR sequencing data, and demonstrated over 99% specificity and 86% sensitivity in a pilot study to predict ovarian cancer patients (n=14) from healthy donors (n=176). To develop this approach into a novel ovarian cancer specific biomarker, we have established a biorepository to prospectively collect specimens from patients with benign or malignant ovarian lesions and from healthy donors of similar age span, with related clinical information. In Aim 1, we will generate TCR sequencing data of the new patient samples to develop a novel, TCR-based ovarian cancer predictor using machine learning method. In Aim 2, we will combine this approach with existing clinical tests to obtain a multi-modality biomarker, and independently test it using the samples from the Uterine Lavage cohort led by Dr. Steven Skates. These Aims will be delivered by the PIs and co-investigators with complementary expertise covering gynecological oncology, clinical cohort recruitment, biostatistics, artificial intelligence, immunology and ovarian cancer biomarker development.