Tracking plasticity along with proliferation and immune evasion using FOXC1, MKI67 and PDL1 for prediction of therapeutic response and risk for hyperprogressive disease with PD1/PDL1 inhibitors in advanced/metastatic cancers: Pan-tumor validation in clinical trial and real-world cohorts

2627 Background: Accurate prediction of therapeutic response (TR) to treatment with immune checkpoint blockade (ICB) as well as any resultant durable clinical benefit in terms of improved progression-free survival (PFS) and overall survival (OS) is not consistently achieved with existing biomarkers like PDL1 or tumor mutational burden (TMB). We hypothesized that a multi-marker predictive biomarker strategy that tracks plasticity using FOXC1 expression, in parallel to tumor proliferation and immune evasion, using expression of MKI67 and PDL1, respectively, may demonstrate superior TR prediction, and also enable accurate prediction of risk for hyperprogressive disease (HPD). Methods: Pre-treatment tumor RNA-Seq data obtained from training/validation clinical trial cohorts and real-world patients diagnosed with advanced/metastatic non-small cell lung cancer (NSCLC, n=82/28) and melanoma (n=154/121) were analyzed for FOXC1, MKI67 and PDL1 expression, and correlated with overall response rate (ORR), PFS, OS and HPD, the latter defined as time-to-treatment-failure <=2 months post-treatment initiation. Optimized biomarker cutoff values based on model area-under-curve were leave-one-out cross validated and cancer-type specific (CTS) prediction algorithms derived. The unmodified strategy was then validated in the independent, CTS validation datasets. Results: ORR prediction accuracy was confirmed in validation datasets with high accuracy: NSCLC AUC=0.96, OR=9.63 (0.98-94.54, 95%CI) p=0.03 and melanoma AUC=0.91, OR=3.85 (1.73-8.58, 95%CI), p=0.0005. Predicted Responders consistently displayed superior PFS and OS compared to predicted Non-Responders: NSCLC PFS HR=0.51 (0.32-0.82, 95%CI) p=0.007, OS HR=0.44 (0.27-0.73, 95%CI) p=0.003; melanoma PFS HR=0.45 (0.30-0.68, 95%CI) p=0.002, OS HR=0.43 (0.26-0.71, 95%CI) p=0.002. Patients at risk for HPD were identified with high accuracy: NSCLC AUC=0.96, OR=9.67 (2.95-31.73) p<0.0001; melanoma AUC=0.98, OR=44.36 (11.73-167.74, 95%CI) p<0.0001. Conclusions: Tracking multiple dimensions of cancer biology including plasticity (using FOXC1), as opposed to tracking immune evasion alone, proved to be a superior and accurate pan-cancer, tissue-agnostic predictor of TR to ICB therapy in advanced/metastatic tumors in terms of ORR, PFS, OS as well as HPD risk prediction. This approach merits further testing in prospective clinical trials.