Deep Profiling of Transcription Factor Activities for Precision Cancer Immunotherapy

Abstract

Cancer arises from dysregulated transcriptional programs and is dependent on the deregulation of a core set of master transcription factors (TFs) that control global gene expression landscapes. Typically, tumor-driving TFs exhibit expressions and activities that are highly cell-state-specific, and therefore represent promising therapeutic targets that can be leveraged for precision cancer treatments. Here, we present a systematic high-throughput screening pipeline (i.e., Synthetic Transcription-factor Activity Responsive, STAR) that quantitatively guides the design of synthetic promoters with the highest cell-state specificity of interest. Implementing paired screenings in the context of ovarian carcinoma and melanoma, we identified synthetic promoters having up to 483-fold activity differences between tumor and normal cell states. We further exploited deep learning to build interpretable models that predicted up to ~74% of the expression and specificity driven by independent test promoters. Model-guided promoter engineering designed new promoters with up to ~ 36% increases in tumor-specificity from the highest experimentally measured counterparts. Taken together, these analyses demonstrated that our STAR pipeline can capture promoter-specific transcriptional behaviors across cell identities and reveal molecular insights underlying the design principles of tumor-specific synthetic promoters.