109 Development of a novel immuno-metabolic spatial signature to predict response and resistance to immunotherapy in NSCLC patients

Le résumé fourni par la source

Background Immune checkpoint blockade therapies (ICB) have led to durable benefits in a subset of patients in non-small cell lung cancer (NSCLC). Developing predictive signatures of response and resistance to ICB requires a thorough understanding of the tumor biology and the tumor microenvironment. In this study, we spatially profiled the tumor-immune contexture and metabolic activity of first-line ICB treated patients to gain further understanding of the functional cell types, states, and interactions to predict treatment sensitivity and resistance. Methods We profiled the tumor microenvironment using an ultra-deep multiplexed immunofluorescence (mIF) panel of 45 proteins two independent retrospective cohorts of NSCLC patients treated with ICB (n=117, 72 with clinical outcome). Utilizing Nucleai’s deep learning multiplex imaging analysis pipeline, we identified distinct tumor phenotypes based on metabolic and immune activity. More than 1000 spatial features were calculated combining cell types, phenotypic state, spatial context, and cell-cell interactions. We built a multivariate model of clinical benefit from spatial features in a training set (n=53) and validated the model on unseen samples (n=19). Model performance was evaluated using ROC analysis, as well as association with clinical endpoints to therapy (progression free survival PFS, overall survival OS). Results 271,193 cells were segmented from 117 tissue cores of NSCLC, and assigned to 13 cell types by known marker expression profiles. We identified cell subtypes, by clustering positive prediction probabilities of metabolically and immunologically relevant markers. Five distinct tumor clusters and phenotypes were identified by the differential expression of CD44, G6PD, Hexokinase-1, HLA-A, PD-L1, and oxidative phosphorylation proteins. ICB-related PFS was significantly associated with the predominant tumor clusters in each patient. Patients in the CD44-high tumor cluster had a prolonged PFS vs the G6PD and Hexokinase-1 high tumor clusters (median PFS of 23.2, 3.0 and 3.3 months respectively, p=0.02). The multivariate ICB outcome prediction model incorporated spatial features associated with response under univariate analysis within each tumor cluster. Performance evaluation of the model demonstrated high AUC (0.84, p=0.007) and compared to the predicted resistance group, the predicted responders group had higher clinical benefit rate (81.8% vs. 25, p=0.02), as well as higher median PFS (13.7 vs 2.3 months, p=0.0007) and median OS (Not reached vs. 9.3 months, p=0.001). Conclusions Taken together, our study provides an mIF-based predictive model for ICB response in NSCLC, based on immuno-metabolic profiles of the tumor. This work underscores the importance of immuno-metabolic profiling of the tissue for building accurate predictive models for treatment outcomes. Ethics Approval Informed Written Consent was obtained for this study from study participants. The study has University of Queensland Human Research Ethics approval.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.