MSACL 2025 Abstract

Development of an MS1 Peptide Profiling Analytical Workflow and Feasibility Evaluation for Lung Cancer Detection

Honglei Huang (1), Taryn Reid (1), Luke Hankey (1), Nikola Gushterov (1), Devalben Patel (2), Geoffrey Liu (2), Peter Jianrui Liu (1), Daniel Sulzc (1), Junetha Syed Jabarulla (1), Heinrich Roder (1) and Harriet Ferguson (1)
(1) Oxford Cancer Analytics Ltd, Oxford, UK (2) Princess Margaret Cancer Centre, University Health Network, Toronto, ON, Canada

Honglei Huang, PhD (Presenter) Oxford Cancer Analytics

Presenter Bio: A self-motivated and enthusiastic biological research scientist employing omics techniques with interest for understanding basic biomedical questions, such as biomarker discovery and validation for cancer, host-pathogen interactions in infectious disease, molecular signature of organ quality for transplantation.

Relevant Financial Disclosures (within past 24 months, reported on Mar 21, 2025) Other Potential Conflicts Oxford Cancer Analytics / Employee Conflict mitigation NOT REQUIRED.

INTRODUCTION:
Early detection of lung cancer is crucial for improving patient outcomes. This study aims to enhance lung cancer detection through a liquid chromatography-mass spectrometry (LC-MS)-based peptide profiling approach. Conventional proteomics relies on precursor (MS1) and fragment ion (MS2) quantification, but complexity of plasma samples and MS2 duty cycles hinder the accuracy and throughput in a clinical setting. Given the limitations this presents to the clinical adoption of LC-MS based proteomics, MS1-only profiling methods have been proposed as an alternative, improving quantification sensitivity and throughput. To assess the ability of such an approach to detect lung cancer we have conducted a pilot study evaluating the robustness, quantification accuracy, and feasibility of MS1-profiling in distinguishing lung cancer cases from controls using machine learning.

METHODS:
A gold reference standard (GRS) was created by pooling plasma from eight healthy individuals. Neat plasma preparation was compared with affinity-based depletion and perchloric acid precipitation. Protein digestion was performed using a single-port solid-phase sample preparation (SP3) method. Peptide samples were analyzed using an Evosep One–Exploris 480 system in MS1-only mode at 120,000 resolution (22 min per sample).

This protocol was applied to a pilot study including 50 non-small cell lung cancer (NSCLC) cases and 50 age- and sex-matched controls. GRS samples were used to define stable features and included for quality control. Features were quantified using Dinosaur software, with in-house tools used for batch correction and retention time alignment. A balanced random forest model evaluated the ability of stable features (CV < 50%) to differentiate cancer from controls using out-of-bag estimates of ROC AUC.

RESULTS:
Neat plasma preparation was selected for MS1 profiling due to simplicity, minimal sample loss, and the lowest coefficient of variation (CV). Our approach detected 30,000 - 40,000 features in GRS samples, with 4,695 consistently identified across 20 replicates (median CV ~21%). Peptides from over 300 proteins were quantified with high reproducibility. Using a balanced random forest machine learning model, we achieved an out-of-bag ROC AUC of 84% for lung cancer discrimination.

CONCLUSIONS:
We developed a sensitive, high-throughput, quality-controlled MS1-based analytical approach for proteomic feature detection in a pilot lung cancer study. Machine learning demonstrated reasonable performance in distinguishing lung cancer from controls. These results support the feasibility of MS1-based analytics for lung cancer detection, but clinical validation is required before application in high-risk population screening.