Liquid-biopsy-derived glycoproteomic profiling as a novel means for noninvasive diagnosis of ovarian cancer.

Abstract

e17604 Background: Ovarian cancer (OC) is the fifth- leading cause of cancer-related deaths among women, and the most lethal gynecological cancer. Currently available biomarkers, including CA-125 and HE4, show suboptimal diagnostic performance for differentiating among benign and malignant pelvic tumors, and the early recognition of OC. Differentiation of benign and malignant pelvic tumors is required for proper patient triaging and management, yet non-invasive methods remain a largely unmet medical need. Methods: We applied a novel platform for characterizing peripheral blood glycoproteomic biomarkers, combining liquid-chromatography/mass spectrometry (LC-MS) with artificial intelligence/neural networks (AI-NN) for the targeted quantification of serum protein glycosylation at intact glycopeptide level to analyze serum samples from 296 treatment-naïve women with histopathology-confirmed diagnosis of either benign (n = 151) or malignant (n = 145) tumors, and from 55 healthy control subjects, procured from a commercial biobank. Using data-dependent acquisition, a panel of 683 serum glycosylated and non-glycosylated peptides, representing 71 serum proteins, was interrogated. Samples were randomly divided into a training and a testing set for multivariable analysis. Data processing was performed using PB-Net, an in-house-developed high-throughput peak integration software. Raw data were normalized, processed by statistical analysis, and applied to machine learning models. Results: Comparison of glycopeptide abundances among patients with OC and benign pelvic tumors yielded 428 statistically significantly differentially expressed glycopeptides/peptides (at FDR < 0.05). A subpanel of these markers used to generate a score for predicting OC yielded areas under the receiver-operating-characteristic of 0.955 and 0.894 in the training and testing sets, respectively. The predicted probability of malignancy increased with cancer stage, and probability distributions were similar across training and test sets. Applying the model to healthy subjects, not utilized in training, resulted in few misclassifications and a spread nearly equivalent to that of the benign tumor cases. This indicates the signature robustly predicts malignancy and severity of disease. Conclusions: Our novel approach exhibited impressive levels of accuracy for the noninvasive differentiation of benign and malignant pelvic masses, compared with existing biomarkers and algorithms, thereby demonstrating the utility of glycoprotein profiles as a powerful, noninvasive new diagnostic modality.