Detection of high-grade serous ovarian carcinoma using adaptive immune response and T cell repertoire (201)

Abstract

<b>Objectives:</b> Identifying high-grade serous ovarian cancer (HGSOC) at its early or preinvasive stage is paramount in improving ovarian cancer (OC) mortality. In contrast to the conventional approaches relying on assessing tumor-released proteins, or imaging assessment, we hypothesized that it is feasible to identify cancer signals from the adaptive immune system. It has been reported that shared sequence similarity in the hypervariable CDR3 region of the T cell receptors (TCR) can be used as a surrogate for shared antigen-specificity. Under this observation, clustering similar TCRs has proven powerful in identifying antigen-specific T cell groups. Our objective was to demonstrate a distinct T cell immune repertoire signature for HGSOCs. <b>Methods:</b> Targeted TCR repertoire sequencing using the immunoSEQ platform was performed on blood samples from cancer patients and healthy controls in the public domain. The CDR3 regions from these patients were used as training data to develop a computational framework to systematically investigate TCR repertoires in cancer patients. ImmunoSEQ platform was performed on blood samples from 45 non-treated HGSOCs, ten non-HGSOCs, 31 benign cysts, and 34 healthy controls collected under an IRB-approved protocol. This data was used as a proof-of-principle analysis of our computational framework to differentiate ovarian cancers from benign cysts and healthy controls. <b>Results:</b> Our training data consisted of 20 million cancer-associated TCR sequences and 500,000 non-cancer TCR extracted from 50 healthy donors. Pooled TCRs from the above samples were partitioned into 5,000 clusters. The centroid of each cluster represents an antigen-specific TCR signature, referred to as ‘repertoire functional unit' (RFU). Each RFU can be perceived as a ‘gene' of the TCR repertoire, which is responsible for recognizing a class of unknown antigens. Although their antigen-specificities remain uncharacterized, the relative abundance of RFUs in each repertoire can reveal important information of the disease signatures, including ovarian cancer. With this transformation, each repertoire sample was converted into a 5,000 vector of the frequency of the same RFUs, disregarding the large diversity of the original CDR3 sequences. An unsupervised principal component analysis (PCA) of the sample-by-RFU matrix was able to differentiate HGSOCs from benign and normal controls (p=2.5, 10-6, one-way ANOVA). We next performed a proof-of-principal analysis by using the RFUs as predictors for a cohort of 120 HGSOC and control patients collected from our biore-pository. We showed that a simple machine learning model using different numbers of selected RFUs achieved high prediction accuracies ranging from 97.3% to 100%. <b>Conclusions:</b> This result suggested that the HGSOC tumor significantly altered the blood TCR repertoire composition, causing a sufficient shift in the RFU composition that dominated the variation between cancer and non-cancer samples. Proof-of principle analysis demonstrates a strong and distinctive immune repertoire signature of HGSOCs with high prediction accuracy.