1241P - Decoding the glycan code: Pioneering early detection of non-small cell lung cancer through glycoproteomics

Background

Non-small cell lung cancer (NSCLC) stands as a prevailing cause of cancer-related death worldwide, largely due to the late-stage diagnosis that diminishes the prospects of cure. Abnormal protein glycosylation, a recognized hallmark of cancer, offers a promising avenue for the exploration of novel biomarkers. Recent advancements in AI-supported analytical technologies have opened possibilities for a thorough examination of cancer-specific glycoforms. The objective of this study is to identify circulating glycoproteomic biomarkers for the early detection of NSCLC, with the potential to enhance patient outcomes and revolutionize diagnostic approaches.

Methods

With a novel high resolution and high throughput glycoproteomic profiling platform combining liquid-chromatography/mass-spectrometry (LC-MS) and artificial-intelligence (AI)-powered data processing, we assessed 694 glycopeptide (GP) and non-glycosylated peptide quantification transitions in pre-treatment peripheral blood serum samples from 372 NSCLC, 148 COPD, and 494 healthy individuals. The full sample set was split into training and validation loops, with a held-out test set. Cross-validated elastic net-regularized logistic regression was performed to develop a multivariate classifier to detect NSCLC.

Results

A total of 340 statistically significantly (FDR < 0.05) differentially abundant glycopeptides (GPs)/peptides when comparing NSCLC and healthy controls, and 50 biomarkers were significantly differentially expressed between healthy controls and NSCLC samples, COPD and NSCLC samples, but not between healthy controls and COPD samples. A multivariate model achieved 85.5% specificity and 84.2% sensitivity to detect all stages of NSCLC with an AUC of 0.875 in a held-out test set. Notably, it performed well in detecting both early (stages I-II) and late stages of NSCLC (stages III-IV) with sensitivities of 81.8% and 87.2%, respectively.

Conclusions

We established a novel platform combining LC-MS and AI data processing, and identified comprehensive serum glycoproteomic profiles which allow the further clinical translations including development of effective non-invasive NSCLC screening and early detection tools.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

InterVenn Biosciences.

Funding

InterVenn Biosciences.

Disclosure

K. He: Financial Interests, Personal, Advisory Board: Mirati, Iovance, BMS, AstraZeneca, Lyell, BeiGene; Financial Interests, Personal and Institutional, Coordinating PI: mirati, OncoC4; Financial Interests, Personal and Institutional, Steering Committee Member: Iovance, Lyell; Financial Interests, Personal and Institutional, Local PI: BMS; Financial Interests, Personal, Steering Committee Member: BeiGene; Financial Interests, Institutional, Local PI: Amgen, GSK, Genentech. G. Absenger: Financial Interests, Personal, Advisory Board: Amgen, AstraZeneca, Bristol Myers Squibb, Janssen, Lilly, Merck Sharp & Dohme, Novartis, Pfizer, Takeda, Sanofi, Roche. C. Pickering, C. Guerrier, C. Dhar, G. Xu, J. Hartness, R. Rice, F. Schwarz, D. Serie, D. Hommes: Financial Interests, Personal, Full or part-time Employment: InterVenn. L. Brcic: Financial Interests, Personal, Advisory Board: Invitae, Eli Lilly, AstraZeneca, Bristol Myers Squibb, Janssen, Lilly, Merck, Novartis, Pfizer, Takeda, MSD, Roche.