869P - Utilizing H&E images and digital pathology to predict response to buparlisib in SCCHN

Background

This study aimed to evaluate a novel methodology to identify subjects that could derive benefit from Buparlisib treatment in metastatic SCCHN patients. The analysis was focused on image analysis of H&E images to select features associated with improved clinical benefit from paclitaxel+buparlisib.

Methods

BERIL-1 (NCT01852292) was a multicenter, randomized, double-blind, placebo-controlled phase II study evaluating treatment with either buparlisib + paclitaxel or placebo + paclitaxel in adult patients with metastatic SCCHN. H&E stained whole slide images (WSI) were used to develop a model to identify features of the tumor and the tumor immune microenvironment through digital pathology. We then evaluated spatial biomarkers from 145 subjects associated with improvement in efficacy endpoints of Progression Free Survival (PFS) and Overall Survival (OS) between the treatment and control arms.

Results

A deep learning model was developed that can accurately identify and classify tumor, necrotic and stromal areas as well as fibroblast, endothelial and immune cells (plasma, lymphocyte, granulocyte), from H&E images. The accuracy of this model was developed against the ground truth of human pathology analysis of the same images. This analysis demonstrated that a >10% infiltration of TILs (p=0.00058, HR=0.195) as well the heterogeneity of cells in the TME (p=0.015, HR=0.53) are both associated with a survival advantage in patients receiving the combination treatment when compared to placebo. Proximity of granulocytes to tumor cells (p=0.00006, HR=0.32) is also associated with improved survival in patients treated with buparlisib + paclitaxel.

Conclusions

This analysis highlights a novel approach, utilizing the common and cost-effective biomarker of H&E to identify metastatic SCCHN subjects that could derive therapeutic benefit from Buparlisib + paclitaxel. This approach also highlights interesting and novel biological observations that underscore the mechanisms of this therapeutic combination that could lead to studies evaluating novel therapeutic combinations. The results of this analysis can be expanded to the ongoing phase III BURAN study to further optimize and validate this method of identifying subjects for therapeutic intervention.

Clinical trial identification

BERIL-1 (NCT01852292).

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Nucleai LTD & Adlai Nortye.

Disclosure

D. Soulieres, J. Lorch: Non-Financial Interests, Advisory Board: Adlai Nortye USA Inc.. J. Lucas, N. He, K. Dreyer, T. Tang: S. Lu: Financial Interests, Institutional, Financially compensated role: Adlai Nortye USA Inc.. O. Matcovitch-Natan, A. Bart, A. Laniado, S. Rosen Zvi, Z. Rachmiel, J. Kaplan Kerner, H. Yehezkeli: Financial Interests, Institutional, Financially compensated role: Nucleai. A. Gutwillig: Financial Interests, Financially compensated role: Nucleai. L.F.L. Licitra: Non-Financial Interests, Institutional, Advisory Board: Adlai Nortye USA Inc.. All other authors have declared no conflicts of interest.