Glioblastoma (GBM) is an aggressive brain cancer with limited treatment efficacy and a poor prognosis (median survival 14-17 months) despite standard-of-care (SOC) treatment (surgery, radiation, temozolomide). GBM tumors exhibit significant intratumoral heterogeneity, with diverse malignant tumor-cell states that respond differently to treatment. This heterogeneity complicates treatment strategies based on bulk genomic, epigenomic, and transcriptomic analyses, as these markers might not be present across all tumor cells. SOC treatment may select for or induce drug-resistant states, leading to recurrence. Current precision medicine approaches, focusing on mutation-based or bulk transcriptional profiling, often fail to capture this complexity and mechanistic drivers of cell-state biology. This study introduces a new framework to model and characterize non-genetic tumor-cell states during GBM progression, from disease onset to recurrence, including the understudied post-SOC treatment period. The framework leverages single-nucleus multi-omic analysis (snRNA-seq, snATAC-seq) of a primary tumor biopsy, time-series PDXs, and a matched recurrent tumor to achieve this.

Previous research has established that GBM tumors are complex ecosystems with diverse malignant tumor cell states and intratumoral heterogeneity impacting treatment response. Studies utilizing single-cell RNA sequencing (scRNA-seq) have highlighted this heterogeneity, but traditional bulk-level analysis obscures the underlying complexity and mechanistic insights into master regulators. Precision medicine approaches have primarily focused on mutation-based profiling or bulk transcriptional profiling, neglecting the crucial role of cell-state vulnerabilities and the dynamic shifts in tumor composition during and after SOC treatment. Therefore, there's a critical need for models that accurately reflect this complexity and identify state-specific vulnerabilities.

The study employed a multi-modal single-cell approach using samples from a single GBM patient. The samples included: a primary tumor biopsy, four patient-derived xenografts (PDXs) split into control and treatment groups (harvested at 24 and 72 hours post-treatment), and a recurrent tumor obtained at autopsy. Single-nucleus RNA sequencing (snRNA-seq) and single-nucleus Assay for Transposase-Accessible Chromatin sequencing (snATAC-seq) were performed on all samples. The snRNA-seq data was analyzed using the Systems Genetics Network Analysis (scSYGNAL) platform to identify regulons (sets of co-regulated genes) and transcriptional programs (clusters of regulons with similar activity profiles). This identified five transcriptional network states in the primary tumor. snATAC-seq data was analyzed using ArchR to identify transcription factors (TFs) with differential binding activity. The intersection of scSYGNAL and snATAC-seq results yielded a consensus set of seven TFs. PDXs were used to model treatment response, with comparisons made between primary tumor, untreated and treated PDXs, and the recurrent tumor using SNN clustering, differential expression analysis, enrichment testing, and cell-to-cell comparison. Longitudinal categorization of tumor cells revealed distinct dynamic behavior of network modules over time. The Open Targets platform was used to identify potential drugs targeting identified TFs and regulons. Cell hashing was used to demultiplex pooled single-cell samples, and doublet prediction was conducted using both experimental (cell hashing) and computational methods (DoubletDecon). Quality control was applied to both snRNA-seq and snATAC-seq data before normalization and analysis. PCA and UMAP were used for dimensionality reduction and visualization.

The PDX models successfully recapitulated several phenotypic states observed in the primary and recurrent tumor, demonstrating their utility as in vivo models of GBM progression. scSYGNAL analysis revealed five transcriptional network states in the primary tumor, each characterized by distinct regulon activity profiles and enriched for specific biological processes. The study identified a consensus set of seven TFs (AR, TEAD1, RUNX1, RORA, EBF1, ZEB1, and TCF4) from both snRNA-seq and snATAC-seq analyses, highlighting their potential role in GBM regulation. Longitudinal analysis revealed distinct dynamic behavior of regulons and cell states over time, including the identification of "selected against," "selected/induced," and "transient" regulons. The Open Targets platform helped identify potential drugs that could target these regulons and cell states during different stages of tumor evolution. For instance, galunisertib could target PRRX1 in "selected/induced" populations, and CDK4/6 inhibitors could be used against IKZF2. For transient states, targeting SP100, TGF-β pathway, or TTK were potential therapeutic avenues. Finally, the study identified RFX3 and RFX7, involved in ciliogenesis, as potential targets in the recurrent tumor, highlighting the value of multi-omic data integration.

This study demonstrates a novel framework for understanding GBM progression at the single-cell level, bridging pre-clinical models with potential clinical applications. The use of PDXs as patient avatars showed strong correlation with the primary and recurrent tumors, indicating the potential for generalizable predictions. scSYGNAL identified a detailed regulatory network and transcriptional programs, providing mechanistic insights beyond traditional gene expression analysis. The integration of snATAC-seq data further strengthened the identification of key regulators. The longitudinal analysis allowed for the identification of treatment-induced cell-state transitions, revealing potential therapeutic targets for concurrent, adjuvant, or recurrent settings. While this study focuses on a single patient, the results suggest that a similar approach could be applied to other patients to tailor treatment based on their individual tumor's characteristics.

This study establishes a robust framework for precision medicine in GBM using a comprehensive, single-cell multi-omic approach. The patient-specific modeling, incorporating primary tumor, treated PDXs, and recurrent tumor, provides an unprecedented level of detail on GBM progression and identifies potential therapeutic vulnerabilities. Future work should focus on validating the predicted drug-target pairings in larger patient cohorts and further investigate the use of continuous PDX models. This framework holds significant promise for advancing personalized GBM treatment strategies.

This study is limited by its focus on a single patient, limiting the generalizability of the findings. The validation of predicted drug-target pairings requires further experimental testing in relevant model systems. The availability of limited biopsy samples prevented the testing of predictions in a patient-specific manner, highlighting the need for continuous maintenance of PDX models. Further research is needed to establish the consistent accuracy of PDX models for broader clinical use. Many identified regulators may not be directly druggable requiring alternative targeting methods.