Cell State Network-Directed Therapy

ABSTRACT—Cell State Network-Directed Therapy Drug resistance is a significant challenge in cancer therapy and has historically been addressed from “one step behind”, whereby drug(s) become progressively ineffective, resistance mechanism(s) are studied, and then different drug(s) are used. While genetic and other resistance pathways are well-appreciated, so-called cell state

plasticity is increasingly understood to be a major determinant of resistance. Cell state is typically defined by a transcriptome pattern, and plasticity is mediated largely by dynamic epigenetic transitions that mimic developmental or tissue homeostasis programs. The different cell states along with the transitions between them

comprise a cell state network. Because different cell states can have unique drug sensitivities and resistances, and cells can change state dynamically through the network, cell state networks provide flexible resistance mechanisms. This proposal seeks to design therapies based on cell state network dynamics, such that drug

sensitive states are promoted while limiting transitions to drug-resistant states. However, an agnostic experimental screening approach for drug combinations is extremely challenging, even considering just two drugs and the hundreds of FDA-approved anti-cancer compounds, not to mention inter-patient heterogeneity,

doses, and timing/sequence. Computational models that predict how tumor cell populations respond to drug combinations could help fill this gap, but so far this has been a challenging problem, even for modern machine learning methods. Our preliminary work shows combining knowledge of cell state transition network dynamics

and single drug dose responses could enable computational prediction of how varied drug combinations influence cell population growth dynamics, and how a sufficient and attainable set of data enables unique inference of cell state networks. These motivate our proposal to predict and test “cell state network-directed”

therapies, with particular focus on glioblastoma multiforme (GBM), a brain tumor with poor survival and few treatment options. We propose four Aims to study 3 different patient-derived xenograft models of GBM with in vivo-like tumor-chip systems: Aim 1. Develop Glioma Cell State Network Models; Aim 2. Determine How

Microenvironmental Factors Alter Cell State Networks; Aim 3. Determine How Glioma Cell State Networks Respond to Single Drugs; and Aim 4. Evaluate Model-Predicted Combination Therapies in Cell Culture and Tumor Chips. We will combine computational modeling with state-of-the-art single cell RNAseq, spatial

transcriptomics, and flow cytometry to characterize cell state networks in gliomas and their dynamics, and how they respond to a variety of glioma-relevant chemotherapy drugs. We will then use these models to propose efficacious regimens that not only control growth but favorably modulate cell state transitions, and test them

using in-vivo like tumor chips. Furthermore, we will study how spatial arrangements of cell types and co-cultures with primary neurons influence glioma behavior. Although the application is GBM, cell state networks are a universal feature of cancer so findings here may have widespread significance across human cancers.