Abstract 5403: Predicting cancer cell response to TEAD auto-palmitoylation inhibitor using bulk RNA-seq data and a random-forest based algorithm

Abstract

Abstract TEAD transcription factors are the major effectors of the Hippo-YAP/TAZ pathway essential in controlling organ size and maintaining tissue homeostasis. Upon coactivator YAP/TAZ binding, TEAD transcription factors are activated to induce expression of genes involved in cell proliferation and survival. It has been shown that auto-palmitoylation is required for TEAD interaction with YAP/TAZ and hence activation of transcriptional activity. Potent small molecule TEAD auto-palmitoylation inhibitors have been reported (Tang et al, 2021, Mol Cancer Ther.). These TEAD inhibitors disrupt YAP/TAZ-TEAD protein interaction, suppress TEAD transcriptional activity, and selectively block proliferation of NF2-deficient mesothelioma in vitro and inhibit NF2 mutant xenograft tumor growth in vivo. Although genetic alterations of pathway components (such as, NF2) leading to YAP/TAZ constitutive nuclear localization and TEAD activation have been reported in a variety of human malignancies, these alterations are infrequent in most cancers, and increased accumulation of nuclear YAP has been reported in cancers (e.g., hepatocellular carcinoma) that do not harbor pathway mutations. Utilizing in vitro efficacy data from cell line screens, gene expression (RNA-Seq) datasets, and a custom bioinformatics data-processing and normalization pipeline, we sought to elucidate gene expression patterns by using a random-forest based classifier. Our classifier, which aims to predict the response (efficacy) of treatment using normalized gene expression of selected genes, was originally trained on the public bulk RNA-seq pan-cancer in vitro dataset available in the Cancer Cell Line Encyclopedia (CCLE) from the Broad Institute, as well as a privately obtained TEADi response data for corresponding CCLE cell lines. Binary response labels (responder/non responder) for each cell line were obtained by thresholding efficacy measures. The classifier - evaluated using cross-validation - achieved an average AUC of 0.80. Then from our screen of 50 patient-derived Chinese liver cancer (CLC) cell models (Qiu et al, 2019, Cancer Cell), both as single agent and in combination with mTOR inhibitor everolimus, we found that TEAD inhibitors (TEADi) were efficacious in several of these liver cancer models. This allowed us to verify the classifier without any re-training on an independent dataset (CLC). In this independent validation scenario, the classifier achieved the AUC of 0.82. Having established efficacy of predictions on both in vitro datasets, we re-trained the classifier using both CCLE and CLC datasets as input to achieve maximum predictive performance, and we predicted, post-hoc, response status of 99 responders out of 2056 patient-derived xenograft models. The algorithm showed early promise by retrospectively classifying a model with in vivo efficacy as the most likely to respond. Citation Format: Adam Kurkiewicz, Sebastian Y. Müller, Oliver Gibson, Sara Castellano, Germán Stark, Pol A. Vecino, Shuirong Zhou, Tracy T. Tang. Predicting cancer cell response to TEAD auto-palmitoylation inhibitor using bulk RNA-seq data and a random-forest based algorithm. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5403.