Spatial Biomarker Deep Learning Model Predicts Response to PI3K Inhibition in Head and Neck Cancer

Background: Buparlisib, combined with paclitaxel, improved survival in BERIL-1 trial patients with recurrent/metastatic head and neck squamous cell carcinoma (R/M HNSCC). However, predictive biomarkers of benefit remain undefined. Objective: To evaluate whether spatial biomarkers extracted from hematoxylin and eosin (H&E) slides using artificial intelligence (AI) can predict overall survival benefit from buparlisib. Methods: Whole-slide H&E images from BERIL-1 trial patients were analyzed using a deep learning model trained to segment tissue compartments and classify cell phenotypes. Three predefined spatial features were evaluated: tumor-infiltrating lymphocyte density, tumor microenvironment heterogeneity, and granulocyte fraction in the tumor invasive margin. Cox proportional hazards model assessed biomarker-treatment interactions. Results: Of 158 trial participants, 144 had available slides. High tumor-infiltrating lymphocyte density (>10%) was associated with significantly improved overall survival with buparlisib versus placebo (HR, 0.25 (95% CI, 0.01-0.64; p = 0.002)), as were high tumor microenvironment heterogeneity (HR, 0.47 (95% CI, 0.27-0.80; p = 0.005)) and granulocyte enrichment in the tumor invasive margin (HR, 0.51 (95% CI, 0.30-0.88; p = 0.01)); within-arm proximity analysis showed higher granulocyte-tumor-cell proximity correlated with improved overall survival on buparlisib (HR, 0.32 (95% CI, 0.18-0.58; p < 0.001)). AI-derived spatial metrics outperformed CD3 immunohistochemistry. Among oropharyngeal tumors, HPV-positive cases were more frequent in patients with high tumor-infiltrating lymphocytes. Conclusions: AI-extracted spatial features from H&E slides were associated with overall survival benefit from buparlisib in R/M HNSCC. These scalable biomarkers support image-based patient selection strategies and are being prospectively evaluated in the BURAN phase 3 trial.