Date of Award
Spring 2023
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Biomedical Engineering (ENAS)
First Advisor
Miller-Jensen, Kathryn
Abstract
Advances in single-cell RNA-sequencing (scRNA-seq) technology enable unbiased, whole transcriptome profiling of tissues, which have revolutionized systems-level studies of these complex microenvironments. Computational analyses of scRNA-seq data have the potential to provide a tissue-level snapshot of cell types, their associated cell states, and a map of their interactions with other cells in the tissue. However, choosing the optimal approach to reconstruct biologically interpretable cell communities from the sparse single-cell transcriptomes collected in scRNA-seq data, especially in the context of dynamic biologically processes, remains a challenge. This is especially true when analyzing dynamic immune processes, which include many cell types, cell subsets within those types, and cell activation states that are specific to the tissue context and are often poorly defined. To address these issues, the work presented here therefore demonstrates improved computational strategies for analyzing scRNA-seq data in the context of cancer immunotherapy and wound healing. Tumors and healing wounds are two complex microenvironments which are heavily regulated (and dysregulated) by immune cells, and whose productive and pathological immune responses are oppositely related. By developing computational workflows that improve cell classification, inference and interpretation of cell-cell communication, and data integration, we enhance the biological significance of experimentally testable hypotheses extracted from these scRNA-seq data and explore their validation. We first explored the use of a supervised, neural network-based approach for annotation of single cells by cell type from scRNA-seq. Cell type annotation is a critical data processing step supporting the interpretability of all downstream analyses. We found that our workflow yielded reliable cell type labels for two scRNA-seq datasets describing different phases of murine tissue repair, and that it represented an improvement over traditional clustering-based methods through its discrimination of high likelihood from low likelihood labels. We further used the high likelihood cell type labels to support downstream investigations of mechanisms underlying angiogenesis and apoptosis, which are two critical processes by which homeostasis is restored after wounding. The resulting mechanistic hypotheses were critical to frame further experimentation (explored beyond this thesis) which identified potential therapeutic targets in chronic wounds. Next, we investigated the inference and validation of scRNA-seq-derived patterns in cell-cell communication through a single-cell data-driven study of combinatorial immunotherapy with agonistic CD40 (CD40ag), CSF1R inhibition, and PD-1 blockade (i.e., TTx). We aimed to understand the mechanisms by which TTx overcomes resistance to PD-1 blockade in melanoma, as this resistance constitutes a major clinical challenge. Using scRNA-seq collected from control and TTx-treated murine melanomas and annotated with our neural network-based approach, we implicated specific axes of crosstalk between three major immune cell types – macrophages, DCs, and T cell subsets – in the antitumor response observed with TTx. We further experimentally validated the protein-level production of these predicted signals, including IL-12, TNF, and IFNg, and their sufficiency to underlie tumor regression in mice, in part by using an in-house-developed microwell assay for single-cell secretion profiling. Because TTx failed to yield objective responses in a patient cohort, we then tested the ability of CD40ag and PD-1 blockade in combination with CTLA-4 blockade (i.e., CPI + CD40ag) to drive antitumor immunity in murine melanomas. We found that the resulting improvement in long-term survival was IL-12-independent, necessitating the characterization of additional mechanisms underlying the therapy-induced antitumor response. We further required a computational approach which accounted for cell subsets and cell states within cell types that were changing dynamically in the tissue, particularly in response to CD40ag, and for which a priori annotation was not possible. We therefore employed scRNA-seq and the novel combination of NICHES and Milo for cell-cell communication inference and differential abundance testing, respectively, to propose candidate ligand-receptor interactions driving resistance and response to immunotherapy. We predicted the emergence of multiple axes of macrophage- and DC-T cell ligand-receptor communication 24- and 72-hours post-treatment with CPI + CD40ag, whose ligands were further predictive of tumor regression and long-term survival in other systems. The use of NICHES also revealed insights into previously unexplored mechanisms of macrophage mobilization during the CD40ag-driven antitumor response. Together, the work presented in this thesis uses novel computational approaches to characterize mechanisms of immune response in two complex microenvironments from scRNA-seq, which will ultimately aid the design of better therapeutic interventions for immune modulation. More broadly, our work establishes a foundation for augmenting the biological significance of hypotheses extracted from scRNA-seq across biological contexts and pathologies.
Recommended Citation
Bridges, Kathryn Helen, "Improved computational strategies to investigate mechanisms of immune response in the tumor and wound healing microenvironments from single-cell data" (2023). Yale Graduate School of Arts and Sciences Dissertations. 1032.
https://elischolar.library.yale.edu/gsas_dissertations/1032
