Abstract 387: Glycoproteomics as a powerful liquid biopsy-based predictor of checkpoint-inhibitor treatment response

Introduction: Protein glycosylation is the most abundant and most complex form of post-translational protein modification. Glycosylation profoundly affects protein structure, conformation, and function. The elucidation of the potential role of differential protein glycosylation as biomarkers has so far been limited by the technical complexity of generating and interpreting this information. We have recently established a novel, powerful platform that combines ultra-high-performance liquid chromatography coupled to triple quadrupole mass spectrometry with a proprietary machine learning and neural-network-based data processing engine that allows, for the first time, high-throughput, highly scalable interrogation of the glycoproteome. Experimental Procedures: Using this platform we interrogated 413 individual glycopeptide (GP) signatures derived from 69 abundant serum proteins in pretreatment blood samples from a cohort of 36 individuals (11 females, 25 males, age range 28 to 90 years) with metastatic malignant melanoma treated either with nivolumab plus ipilimumab (12 patients) or pembrolizumab (24 patients). Progression-free survival (PFS) data with follow-up of up to 3.7 years (median: 0.8 years) were used as clinical endpoint phenotype against which the predictive power of differential abundance of GPs was assessed. PFS data were analyzed using Cox Proportional Hazards models, and Kaplan Meier curves were generated for GP markers that showed statistically significant differential abundances using an FDR-adjusted p-value of ≤0.1 as a cutoff. Summary of Results: We identified 27 GPs with abundance differences at FDR p≤0.1, and among them 8 at p≤0.001. Using the latter 8 markers, we created a multivariable model for PFS by generating leave-one-out cross-validation (LOOCV) scores and determining an optimized cutoff value for these scores using Harrel9s concordance index. Dichotomizing the LOOCV scores using this cutoff value demonstrated the model to yield a hazard ratio of 9.2 at a p-value of 10-5 for separating treatment responders and non-responders (70% vs. 0% PFS, respectively, at 18 months based on LOOCV score above/below cutoff), as compared to a hazard ratio of 1.5, p=0.5 for PDL1 expression. Conclusions: Our results indicate that glycoproteomics holds a strong promise as a response predictor to checkpoint inhibitor treatment that appears to significantly outperform other currently pursued biomarker approaches in this context. Citation Format: Gege Xu, Rachel Rice, Hector Huang, Klaus Lindpaintner, Jillian M. Prendergast, Karl Normington, Dennie Frederick, Genevieve M. Boland, Daniel Serie. Glycoproteomics as a powerful liquid biopsy-based predictor of checkpoint-inhibitor treatment response [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 387.