113 Leveraging the Stabl framework for anti-PD1 response biomarker discovery in head and neck squamous cell carcinoma using the TruTumor ex vivo platform

Le résumé fourni par la source

Background The traditional method of biomarker identification often fails to capture the complexity of the Tumor Immune Microenvironment (TME) response, particularly in small datasets. SurgeCare’s Stabl framework 1 addresses this by using noise injection for reproducible, quantitative, and high-throughput biomarker selection. In this study, we applied the Stabl framework to anti-PD1 response data generated using the Farcast TruTumor ex vivo platform2 to identify a response biosignature.Methods Explants generated from fresh surgically resected HNSCC samples (n=67) collected from the consented patients were cultured with Nivolumab (132 µg/ml) for 72 hours and response evaluated using multiple assays such as cytokine release, histopathology, flow cytometry, and NanoString based mRNA analysis. Selected features were derived from multimodal assay data using Stabl.The Stabl methodology involved preprocessing omics datasets using variance thresholds, missing value filters, median imputation, and z-score standardization, implemented with sci-kit-learn v1.2.1 Five repetitions of 5-fold Monte Carlo cross-validation were done to ensure stratified sampling. Feature selection was conducted using Knockoffs or Random Permutation with Logistic Regression or Random Forest, along with Adaptive Lasso for penalized regression. These methods identified features distinguishing control (RxA) from anti-PD1 treated (RxB) arms. A responder signature was derived from 23 samples (with complete multimodal data) by correlating selected features with tumor cytotoxicity markers, such as reduced tumor content and/or increased cleaved caspase 3 expression.Results From a total of 849 multimodal features, 25 were identified by Stabl as effective in distinguishing RxA from RxB, with a predictive power of 0.68 (AUC). A six-feature responder signature was derived from these 25 features by correlating them with tumor cytotoxicity (Spearman, p ≤ 0.08). The features included M1 macrophage, IL1beta signaling, non-T cell immune cell sub-population, activated cytotoxic T-cells (CD8+GranzymeB+), Perforin, and Granzyme B cytokines. UMAP on RxB data using these features revealed three distinct clusters: C1, C2, and C3.C1 (n=8) had the highest responder rate (87.5%), followed by C2 (n=11, 36.4%) and C3 (n=4, 25%).C1 samples showed increased immune infiltration and high non-T cell levels in control, correlating with enhanced tumor cytotoxicity on treatment. C2 had a moderate response profile with high cytokine release but limited immune infiltration. C3 showed low immune infiltration and high tumor content.Conclusions The complex response phenotypes observed from the TruTumor ex vivo platform that can generate multi-dimensional datasets combined with the robust Stabl framework can lead to considerable derisking of novel therapies at late pre-clinical stage.References Hédou J, Marić I, Bellan G, et al. Discovery of sparse, reliable omic biomarkers with Stabl. Nat Biotechnol. 2024;42:1581–1593.Basak NP, Jaganathan K, Das B, et al. Tumor histoculture captures the dynamic interactions between tumor and immune components in response to anti-PD1 in head and neck cancer. Nat Commun. 2024;15(1):1585.Ethics Approval Donor tissue specimens along with matched blood sample were obtained from consented patients. Institutional Ethics Committee (IEC) from the sample collection centers approved the protocol (protocol # FCB-PROTOCOL-01) and informed consent for participation in the approved study was obtained from every donor.

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