Abstract
Over-activation of the epidermal growth factor receptor (EGFR) is a hallmark of glioblastoma. However, EGFR-targeted therapies have led to minimal clinical response. While delivery of EGFR inhibitors (EGFRis) to the brain constitutes a major challenge, how additional drug-specific features alter efficacy remains poorly understood. We apply highly multiplex single-cell chemical genomics to define the molecular response of glioblastoma to EGFRis. Using a deep generative framework, we identify shared and drug-specific transcriptional programs that group EGFRis into distinct molecular classes. We identify programs that differ by the chemical properties of EGFRis, including induction of adaptive transcription and modulation of immunogenic gene expression. Finally, we demonstrate that pro-immunogenic expression changes associated with a subset of tyrphostin family EGFRis increase the ability of cytotoxic T-cells to eradicate tumor cells. Our study provides a framework that considers each agent's unique and often unknown polypharmacology to prioritize compounds that induce clinically favorable molecular responses.
Introduction
Glioblastoma (GBM) remains one of the most challenging cancers to treat, with over-activation of the Epidermal Growth Factor Receptor (EGFR) serving as a primary oncogenic driver. Despite the development of numerous EGFR inhibitors (EGFRis), clinical trials have consistently failed to show significant benefit. While the blood-brain barrier is often cited as the primary obstacle, this study investigates the role of polypharmacology—the phenomenon where drugs interact with multiple targets beyond their intended one. These "off-target" effects can create a complex landscape of cellular responses, potentially inducing adaptive resistance and hindering therapeutic efficacy.
Methods: Multiplex Single-Cell Chemical Genomics
To map the transcriptional landscape of EGFR inhibition, the researchers utilized sci-Plex, an ultra-high-throughput single-cell RNA sequencing method. This approach enabled the simultaneous screening of thousands of individual cells treated with various drug-dose combinations across multiple GBM models.
To analyze this high-dimensional data, the team developed Multi-resolution Variational Inference (MrVI), a deep generative model that isolates biological variation from technical noise. MrVI allowed the researchers to group drugs into Transcriptional Drug Classes (TDCs) based on the actual gene expression programs they activated, rather than their chemical structure alone.
Key Insights: Beyond Proliferation
A central finding of the study is that a drug's impact on cell proliferation does not necessarily reflect its global transcriptional effect. Some drugs effectively halt division but trigger strong adaptive resistance programs.
The Shared EGFR Module and Patient Survival
The researchers identified a "Shared EGFR module"—a 48-gene signature triggered by nearly all EGFRis. Analysis of clinical data from The Cancer Genome Atlas (TCGA) revealed that patients whose tumors naturally express this signature have significantly worse overall survival. This suggests that the module represents a general stress response or survival mechanism that cancer cells use to resist treatment.
Tyrphostins and Immunogenic Priming
Perhaps the most significant discovery was the unique effect of the tyrphostin family of inhibitors (e.g., Tyrphostin 9, AG490). Unlike other EGFRis, these compounds upregulated the Antigen Processing and Presentation Machinery (APM), including MHC Class I molecules. This polypharmacological effect makes tumor cells more "visible" to the immune system. Functional assays confirmed that tumor cells treated with tyrphostins were more effectively killed by cytotoxic T-cells.
Results: Mapping the Response Landscape
The application of MrVI categorized the 72 compounds into 16 distinct TDCs. Notably, third-generation inhibitors like osimertinib were found to be potent inducers of adaptive resistance signatures. In contrast, drugs like CUDC-101 induced unique transcriptional programs distinct from the standard EGFR inhibition profile.
Discussion and Future Directions
This study provides a framework for evaluating drugs based on their comprehensive molecular impact rather than just their primary target. While the results in patient-derived models are promising, future work must determine if these immunogenic effects persist in the complex brain microenvironment and how specific EGFR mutations influence the initial cellular state. Ultimately, this approach paves the way for "rationally designed" combination therapies that inhibit oncogenic drivers while simultaneously priming the immune system for a more effective anti-tumor response.
Summary updated