Dissection of immunotherapeutic predictive versus prognostic transcriptional programs identifies SLC22A5-centric carnitine metabolism-driven resistance to anti-PD-(L)1 treatment in non-small cell lung cancer

AIMS: Prognostic and predictive biomarkers are two common biomarker types in clinics, with the former indicating the natural course of cancer regardless of treatment, and the latter determining the response to a specific regimen. Understanding the predictive versus prognostic effect of biomarkers is essential to understand treatment-specific response from the inherent prognosis of cancer. Herein, we aimed to uncover the predictive metabolic signatures specific to immunotherapy resistance by distinguishing the predictive versus prognostic effect of transcriptional programs in advanced non-small cell lung cancer (NSCLC) treated with immunotherapy. METHODS: Clinical and transcriptomic data were collected from two randomized controlled trials, OAK (n = 699, discovery cohort) and POPLAR (n = 192, validation cohort) comparing immunotherapy with chemotherapy. Metabolic transcriptional signature scores were calculated through gene set variation analysis. Cox regression and interaction test were conducted to differentiate the predictive versus prognostic effect. Additionally, lung tumor-bearing murine models were established using Slc22a5-overexpressing (OE) and control Lewis Lung Carcinoma (LLC) cells, and treated with immunotherapy or chemotherapy. The translational potential of an SLC22A5 (Solute Carrier Family 22 Member 5) inhibitor in combination with immunotherapy was assessed in preclinical setting. The tumor microenvironment was analyzed by flow cytometry, immunofluorescence, and Enzyme-Linked Immunosorbent Assay (ELISA) to validate the mechanistic findings. RESULTS: Metabolic transcriptional programs were divided into four categories based on different predictive effects specific to immunotherapy or chemotherapy, among which carnitine metabolism stood out as the most prominent metabolic process contributing to the resistance to immunotherapy. Specifically, SLC22A5 as the only high-affinity carnitine transporter was remarkably upregulated in immunotherapy-resistant patients. The predictive effect of SLC22A5-centric carnitine metabolism for resistance to immunotherapy rather than chemotherapy was independently validated in an external randomized trial. Critically, preclinical models revealed that Slc22a5 overexpression drove resistance to immunotherapy but not chemotherapy, by fostering an immunosuppressive microenvironment characterized by M2 macrophage accumulation and CD8 + T cell exclusion. Furthermore, pharmacological inhibition of SLC22A5 by meldonium reshaped the tumor microenvironment toward a more inflamed state and re-sensitized resistant tumors to immunotherapy. CONCLUSIONS: Our study elucidates the predictive versus prognostic effect of metabolic pathways in advanced NSCLC under immunotherapy. Tumor-intrinsic carnitine metabolism may predict and drive immunotherapy resistance, and targeting SLC22A5-mediated carnitine metabolism could be used to overcome resistance in advanced NSCLC.