2779 - Radiopharmaceutical Cell Surface Target Selection

S. C. Callahan¹, and S. G. Zhao²; ¹Department of Human Oncology, University of Wisconsin School of Medicine and Public Health, Madison, WI, ²Department of Human Oncology, University of Wisconsin Hospitals and Clinics, Madison, WI

Purpose/Objective(s): Radiopharmaceutical Therapies (RPTs) are an emerging modality in solid malignancies, particularly in the metastatic setting. While there are many potential cell surface targets, the optimal targets for radiopharmaceutical payloads are unknown. We therefore developed an approach to predict the best cell surface targets for RPT development. Materials/

Methods: The ideal RPT target gene would both be expressed on the cell surface and be independently associated with radiosensitivity. To develop a model for radiosensitivity, we utilized RNA-seq data from the Cancer Cell Line Encyclopedia (CCLE) paired with published radiosensitivity data for 524 human cancer cell lines. We then used Elastic-Net regression, a regularized regression method that is a linear combination of the LASSO and Ridge methods, to train a radiosensitivity model on the z-scored radiotherapy AUC values from the clonogenic survival assays. Values for a and ? were optimized using a 10-fold nested cross validation approach in which features were selected based on the correlation of rank-normalized gene expression with radiotherapy AUCs. The resulting final model was then applied to rank-normalized gene expression data from The Cancer Genome Atlas (TCGA) to predict radiation sensitivity on over 9000 human cancer samples. The resulting TCGA radiation sensitivity predictions were then correlated with the expression cell surface targets in clinical trials.
Results: We found that our elastic net regression model’s predicted radiotherapy AUC was significantly correlated with the true radiotherapy AUC in the CCLE dataset (p < 0.0001) in cross-validation. We then predicted the radiotherapy AUC for all samples in the TCGA dataset using this signature. Correlating these predicted AUCs with the measured expression of clinical cell surface targets demonstrates that our signature can find numerous genes that are correlated with radiation sensitivity. In particular, we find that DLL3 expression is significantly correlated with predicted radiation sensitivity (p < 0.0001). DLL3 is an inhibitory Notch pathway ligand known to be overexpressed in many high-grade neuroendocrine tumors. There has been increasing interest in DLL3 as a clinical candidate gene for targeted therapies (e.g. ADCs and BiTEs), with preclinical models and early clinical trials focusing on small-cell lung cancer (SCLC) and neuroendocrine prostate cancer (NEPC).
Conclusion: We have trained an elastic-net regression model to predict genes whose expression is correlated with radiation sensitivity, with the aim of identifying cell surface targets for RPT development. Applying this model to TCGA data has revealed numerous candidate genes, the top candidate being DLL3, which is overexpressed in neuroendocrine tumors such as SCLC and NEPC. This work supports the development of DLL3 as a potential RPT target.