Plasma metabolomic profiling predicts response to neoadjuvant immunochemotherapy in locally advanced NSCLC

Background: While immune checkpoint inhibitors (ICIs) have demonstrated significant benefit in non-small cell lung cancer (NSCLC), the efficacy varies substantially among individuals. Established biomarkers like PD-L1 expression and tumor mutational burden (TMB) offer limited predictive value, underscoring the critical unmet need for more reliable predictive biomarkers. To address this, our study aimed to discover predictive plasma metabolomic biomarkers for treatment response to neoadjuvant immunochemotherapy in patients with locally advanced NSCLC. Methods: Between November 2023 and July 2025, 71 patients with locally advanced NSCLC receiving neoadjuvant immunochemotherapy were prospectively enrolled. According to best overall response (BOR), patients were classified into the good response (GR) group (n=38, partial response) and the limited response (LR) group (n=33, stable or progressive disease). Plasma sampling included 71 baseline samples, 50 samples after cycle one and 39 samples after cycle two. The association between clinicopathological features and clinical response was assessed using multivariate logistic analysis. Untargeted metabolomic profiling of all plasma samples was conducted using liquid chromatography-tandem mass spectrometry (LC-MS/MS) to identify predictive metabolic biomarkers. Results: After neoadjuvant therapy, surgical resection in 22 patients revealed a pathological complete response (pCR) rate of 36.4% (8/22). Multivariate analysis identified squamous cell carcinoma histology and tumor cell PD-L1 expression >50% as predictors of good response to neoadjuvant immunochemotherapy. Plasma metabolomic profiling detected 3,756 metabolites. The GR group exhibited significantly higher levels of glycerol ester of acylcarnitine and hydroxypropionylcarnitine, whereas the LR group was enriched in glycocholic acid, phosphatidylcholine and taurocholic acid. KEGG pathway analysis indicated that the differential metabolites were involved in primary bile acid biosynthesis and cholesterol metabolism. Using machine learning, we integrated the baseline, post-cycle one and post-cycle two plasma metabolomic profiles to develop a 15-signature GLMNet model for pCR prediction, which achieved an area under curve (AUC) of 0.906. Conclusion: Dynamic plasma metabolomic signatures are promising non-invasive biomarkers for predicting outcomes to neoadjuvant immunochemotherapy in NSCLC. These findings provide a rationale for leveraging metabolomics to stratify patients and optimize personalized treatment strategies.