(P 418) A SYSTEMS BIOLOGY APPROACH FOR CHARACTERIZING TUMOR HETEROGENEITY AND CELL STATE-SPECIFIC THERAPEUTIC VULNERABILITIES FROM SINGLE NUCLEUS RNA-SEQ PROFILING OF DESMOPLASTIC SMALL ROUND CELL TUMORS

Objective: Desmoplastic Small Round Cell Tumors (DSRCT) are rare, aggressive sarcomas affecting children and young adults with less than 20% five-year survival. They are characterized by an established, EWSR1-WT1, oncogenic fusion protein that is not directly actionable with few alternative therapeutic targets. Additionally, they present complex intra-tumoral heterogeneity and stromal infiltration. The co-existence of diverse, aberrant lineages of transformed cells with distinct drug sensitivities poses challenges to its treatment, complicating therapy response assessment and contributing to rapid relapse from inherently resistant cellular states. Novel therapeutic approaches addressing these challenges are urgently needed. Toward this end, we generated single nucleus RNA-seq (snRNA-seq) profiles and leveraged network-based systems biology approaches to discover and target non-oncogene encoded tumor vulnerabilities—Master Regulator (MR) proteins—that represent pharmacologically accessible, mechanistic determinants of DSRCT cell states. Our goal is to prioritize drug combinations targeting intra-tumoral heterogeneity to expand treatment options for patients with DSRCT.

Methods: We analyzed snRNA-seq profiles from fresh-frozen tissue for six DSRCT patient samples (five post-treatment and one pre-treatment) and two matched patient-derived xenograft (PDX) models. After filtering for quality control, the data was normalized, and cells clustered by gene expression with the Louvain algorithm. Cell type annotation was performed with SingleR and transformed tumor cells were identified by inferCNV analysis. DSRCT tumor-specific gene regulatory networks were generated for each sample using the ARACNe algorithm and integrated to generate a DSRCT-specific consensus network. Single cell regulatory protein activity was then inferred based on enrichment of target genes in the differential gene expression signature, with clustering and further analysis of protein activity derived cell states ongoing. We further used the OncoMatch algorithm to identify DSRCT cell lines best recapitulating MR signatures of patient tumors for drug perturbation studies via the PLATE-seq technology to generate drug-perturbed differential protein activity in the DSRCT context. This will allow us to predict drugs that either target actionable MRs (OncoTarget) or reverse the entire tumor checkpoint of most differentially active MRs (OncoTreat) in individual DSRCT cell states for further preclinical validation, including in combination.

Results: We generated snRNA-seq across 86,895 high-quality cells derived from the six DSRCT patient samples. Clustering of cells annotated using singleR successfully identified putative stromal cells (including fibroblasts, endothelial cells, and immune cells) and tumor cells. As is often seen by gene expression, presumed malignant cells segregated by patient sample with limited intermixing, while stromal cells intermixed and segregated by cell type. We validated the transformed nature of presumed tumor cells by leveraging an external snRNA-seq dataset encompassing eight tissue types as a reference for inferCNV analysis, identifying single cell copy number alterations only within the presumed tumor clusters. Analysis of the transformed tumor population by VIPER-inferred regulatory protein activity is ongoing to define the MR proteins driving distinct cell states in DSRCT. Further, analysis of bulk RNA-seq data from four DSRCT cell lines compared to a cohort of 19 DSRCT patient samples and 9 DSRCT PDX samples by the OncoMatch algorithm identified JN-SP as the cell line most significantly enriched in DSRCT patient MRs (p < 10-5 for 19/19 samples). We performed drug perturbation studies using the PLATE-seq technology in this cell line assessing differential protein activity in drug vs. vehicle control treated cells across a library of 372 FDA-approved and late-stage experimental drugs at their EC20 concentration in replicate. This will allow us to prioritize drugs that reverse the activity of cell state-specific MR tumor checkpoints for preclinical validation.

Conclusion: We employ an innovative pipeline for the network-based inference of regulatory protein activity to analyze snRNA-seq data in DSRCT, characterizing intra-tumoral heterogeneity and identifying non-oncogene encoded therapeutic vulnerabilities. The data generated will allow for systematic prioritization of existing drugs that invert MR activity profiles in DSRCT malignant cell states with orthogonal drug sensitivities, with preclinical validation of their subpopulation-specific effects expected to occur in PDX models. This can provide a platform for leveraging transcriptomic data in the development of molecularly driven, rational combination therapy approaches in highly heterogeneous tumors driven by untargetable oncoproteins.