Decision tree model with tumor immune function genes for predicting response to anti-PD1 therapy in renal cell carcinoma.

Abstract

e17084 Background: No current models can predict prognosis as well as immunotherapy benefits in renal cell carcinoma (RCC). The current study was undertaken to develop and validate a simple decision tree model to predict survival and immunotherapy benefits in patients with RCC. Methods: Analysis on RNA expression in a reported clinical trial dataset identified tumor immune function genes differentially expressed between anti-PD-1 response and non-response RCC. The analysis on these genes in different populations and platforms to develop a Bi-Luo (BL) DTM. The predictive performance of the BL DTM on the prognosis and the immunotherapy outcome in RCC was comprehensively analyzed and validated. Then, the mechanism of BL classification on the immune microenvironment in RCC was extensively analyzed. Results: Immunologic genes expression alteration resulted in the sensitivity for anti-PD-1 therapy in RCC. Twelve tumor immune function genes were enriched based on anti-PD-1 response. Survival analysis and machine learning algorithms on 12 genes established a BL classification based on 545 RCC samples to differentiate patients into low risk (BL1) and high risk of poor survival (BL2) groups ( P < 0.0001). The BL classification was a better independent prognosis indicator ( P = 0.004). TIDE algorithm predicted the response of two subgroups to immune checkpoint blockade, and BL1 was more promising to respond to anti-PD-1 therapy ( P = 0.00003268). Our BL classification was visualized in a 3 genes-DTM (F2RL1, TNFRSF and WHSC1) with a high prediction accuracy (AUC = 0.84) in an external validation cohort. Furthermore, in a small-scale immunotherapy clinical trial dataset, our BL DTM had a high specificity (0.86) to identify the non-response RCC in anti-PD-1 therapy. Gene expression of immune inhibiting factors in BL2 subtype was higher than BL1 subtype (all P < 0.01), but less in tumor mutation burden ( P = 0.009). Moreover, we found that BL2 subtype had higher fractions of Tregs ( P < 0.001), CD8+ T cells ( P = 0.037) and lower fractions of resisting memory CD4+ T cells ( P < 0.001) than BL1 subtype. And most HLA subtype genes, namely HLA-G, HLA-E, HLA-DR, HLA-DQ and HLA-DP, were significantly down-regulated in BL2 subtype (all P < 0.05). Conclusions: Our BL based DTM could predict survival and has potential for identifying RCC patients who could not benefit from anti-PD-1 immunotherapy. Non-response of RCC to anti-PD-1 immunotherapy is strongly related to the down-regulation of the immune response.