Landscape of the Radiosensitizing Properties of Targeted Agents from the NCI CTEP Portfolio across Genomically Diverse 3D Tumor Models

Abstract

Clinical translation of molecular targeted radiosensitizers remains challenging. To improve the robustness of preclinical data, we sought to broadly examine across a wide spectrum of genomic cancer features the radiosensitizing properties of targeted drugs from the NCI Cancer Therapy Evaluation Program (CTEP). As part of a U01 Consortium, we adapted a robotic high-throughput platform for testing ionizing radiation (IR) plus drug combinations in 3D cultures derived from non-small cell lung carcinoma (NSCLC), head/neck (HNSCC), and pancreatic cancers. All cell lines were annotated, mycoplasma free, and re-authenticated. Signal windows were established for all 3D growth formats. CTEP drugs were added in four concentrations 24 hours prior to IR exposure of 384-well plates. Five days after IR, cell viability was assessed, an endpoint previously benchmarked using colony formation. Each data point was based on 4-6 technical replicates with 2 biological repeats. A multistep statistical method combining Wilcoxon signed-rank and Spearman rank-order correlation tests together with adjusted Benjamini-Hochberg corrections for multiple comparisons was used to identify radiosensitization hits. Select validation experiments of hits were carried out. We screened 73 3D tumor models with 34-42 CTEP drugs each plus IR (n = 10,744 cell line/drug concentration combinations). IR/drug patterns were comparable for the three cancer types. DNA damage response (DDR) inhibitors constituted the most active class of radiosensitizers with DNA-PKcs, PARP, and ATR being the top targets. In contrast, the majority of the other agents radiosensitized only a minority of models. Highlighting NSCLC, out of 1,120 cell line/drug combinations, 328 (29.3%) were associated with statistically significant radiosensitization. PARP inhibitors, of particular interest for clinical translation in this space, demonstrated relatively broad radiosensitization for talazoparib (62%) and olaparib (44%) but less so for veliparib (28%), which correlated with differential PARP trapping potentials. Radiosensitization was evident even for tumor genotypes associated with radioresistance such as mutant KRAS, KEAP1, and STK11. Interestingly, single-agent activity of olaparib was more pronounced in 11/30 3D cultures compared to 2D growth conditions but with only 2/11 having known homologous recombination defects. Lastly, for HPV negative HNSCC (735 combinations), there was radiosensitization in subsets of models for IAP, ATR, and AKT targets, which are under clinical investigation. We present a unique resource for examining radiosensitizing drug properties in 3D cultures. The observed radiosensitization patterns strongly argue for the use of larger panels of clinically relevant tumor models rather than a few non-representative cell lines. Our data highlight the urgent need for predictive biomarkers to guide IR/drug combinations, even for some DDR inhibitors.