Abstract 161: Evaluating the clinical utility of RNA-seq-based PD-L1 expression and cellular deconvolution as alternatives to conventional immunohistochemistry in clear cell renal cell carcinoma

Abstract

Abstract The emergence of immune checkpoint inhibitors (ICI) have revolutionized treatment of many cancer types, including renal cell carcinoma (RCC). Although there are currently three ICI combination regimens approved as standard of care for the first-line treatment of advanced RCC, the utility of PD-L1 detected by immunohistochemistry (IHC) as a predictive biomarker remains somewhat controversial. In this study, we sought to determine the ability to detect PD-L1 expression accurately and sensitively in comparison with IHC as well as whether RNA-seq-based cellular deconvolution potentially correlates with PD-L1 IHC for the development of a deconvolution-based biomarker. Tumors from 103 clear cell RCC (ccRCC) patients were collected to compare sensitivity and clinical utility of RNA-seq and IHC in PD-L1 detection. Additionally, we used gene expression signatures to comprehensively characterize the ccRCC tumor microenvironments (TMEs) from RNA-seq to examine PD-L1 expression levels across varied TMEs. This classification system revealed that the ccRCC samples clustered into an immune-enriched microenvironment with high lymphocyte infiltration and an immune desert microenvironment predominantly composed of malignant cells. For RNA-seq-based cellular deconvolution, a unique and robust machine learning algorithm, named Kassandra, was developed using over 8,000 RNA profiles of sorted cells that outperformed all current cellular deconvolution algorithms. The RNA-seq of ccRCC clinical tumors (n =89) were analyzed by Kassandra. Our RNA-seq analysis identified ccRCC tumors with high (maximum), intermediate and low (minimum) PD-L1 expression, which were then validated with IHC. Among these, 5/5, 10/10 and 87/88 ccRCC tumors with high, medium and low expression, respectively, detected via RNA-seq matched with the PD-L1 levels detected with IHC, suggesting a high degree of concordance (p < 0.01) between the two methods. Notably, the immune-enriched ccRCC clusters had a higher percentage of PD-L1-positive malignant and immune cells. PD1+ CD8 T cells, identified via RNAseq-based cellular deconvolution (Kassandra), most strongly correlated with PD-L1 IHC (p < 0.05). PD-L1 status was predicted with an 0.82 AUC performance score using a regression model. In conclusion, our results indicated that PD-L1 expression detected by RNA-seq as well as the presence of PD1+ CD8 T cells determined by applying Kassandra strongly correlated with the PD-L1 values obtained via IHC, potentially providing a less costly and more robust alternative to the conventional methods. In addition, an increased number of PD-L1-positive tumor and immune cells within the immune-enriched TME may suggest the significance of the TME composition as a predictive factor in response to ICIs. Citation Format: Natalia Miheecheva, Maria Sorokina, Akshaya Ramachandran, Yang Lyu, Danil Stupichev, Alexander Bagaev, Ekaterina Postovalova, Krystle Nomie, Felix Frenkel, Maria Tsiper, Nathan Fowler, Ravshan Ataullakhanov, James J. Hsieh. Evaluating the clinical utility of RNA-seq-based PD-L1 expression and cellular deconvolution as alternatives to conventional immunohistochemistry in clear cell renal cell carcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 161.