Background Immune-related adverse events (irAEs) resulting from immune checkpoint inhibitors (ICIs) can substantially affect patient quality of life and treatment trajectory. Currently, there are no reliable pre-treatment biomarkers for predicting the development of irAEs; hence, there is a clinical need for irAE predictive biomarkers.

Methods Plasma samples were obtained at baseline from 426 non-small cell lung cancer (NSCLC) patients treated with ICIs as part of an ongoing multi-center clinical trial (NCT04056247; approved by local IRB committees from each site) with irAE-related information. Proteomic profiling of plasma samples was performed using the SomaScan® assay (SomaLogic Inc.), enabling deep coverage of approximately 7000 proteins in each sample. A machine learning-based model was developed to predict significant irAEs arising up to 3 months from treatment initiation; significant irAEs were defined as irAEs with CTCAE grade ≥3 or irAEs that induced treatment discontinuation. Using the model, we identified a set of plasma proteins, termed Toxicity Associated Proteins (TAPs), that serve as indicators of irAEs depending on their plasma level in the individual patient. Bioinformatic analysis was performed to decipher the biology underlying immune-related toxicity implied by the TAPs.

Results Overall, 60 patients experienced significant irAEs at early onset; 197 patients had low grade irAEs, irAEs at late onset or AEs that are not immune-related; and 169 patients did not display any adverse event. A computational model was generated to predict significant irAEs, showing a strong correlation between the predicted probability of significant irAEs and the observed rate of such events (R²= 0.92; p-value <0.0001), implying good prediction capabilities. The prediction was based on a set of 449 TAPs. Interestingly, nearly half of these TAPs were previously identified as proteins associated with clinical benefit from ICI therapy, suggesting a close relationship between irAEs and clinical benefit, in accordance with previous reports. A detailed examination of the TAPs revealed some key findings. Patients who experienced irAEs had a larger number of TAPs related to neutrophils, inflammation, and cell death resistance, while the number of lymphocyte-related TAPs was low in these patients. Patients who did not experience irAEs displayed higher levels of extracellular matrix-related proteins.

Conclusions We describe a novel computational model for predicting significant irAEs in patients with NSCLC based on proteomic profiling of pre-treatment plasma samples. The TAPs provide insights into the biological processes underlying irAEs. Early prediction of irAEs could enable personalized management plans and mitigation strategies to reduce the risk of irAEs in NSCLC.

Ethics Approval Participants gave informed consent before taking part. Institutional Review Board of the following institutes gave ethical approval for this work: Asklepios Kliniken GmbH; Rambam Medical Center; Hadassah Hebrew University Medical Center; Meir Medical Center; Emek Medical Center; Kaplan Medical Center; Rabin Medical Center Davidoff Cancer Centre; Shamir Medical Center; Bnai Zion Medical Center; Roswell Park Comprehensive Cancer Center; Asklepios Kliniken GmbH; Sheba Medical Center; Cheltenham General Hospital; Aberdeen Royal Infirmary Grampian NHS; Barzilai Medical Center; Sunderland Royal Hospital; Shrewsbury and Telford Hospital.