Approaches for engineering transcriptional state of fibroblasts and other differentiated cell types

The overarching goal of this proposal is to transform Perturb-seq into a platform for rationally engineering cellular transcriptional states. We are motivated by ongoing cell atlas projects, which have revealed remarkable diversity in the form of hundreds of distinct cell types, but also highlighted that even greater heterogeneity exists at the level of transcriptional states within tissues and across disease contexts. However, these observational studies leave open the questions of how these diverse transcriptional states are established and whether they confer specific functional properties. Answering these questions in vivo is labor-intensive due to the rarity and number of states. To address this gap, we pursue a complementary approach, developing a combination of experimental and computational tools that extend Perturb-seq to enable systematic reconstruction of these states in vitro using combinatorial genetic perturbations. We will focus on fibroblasts, a cell type with extensive transcriptional heterogeneity linked to their diverse roles in tissue homeostasis, wound healing, and disease. Our preliminary data from CRISPRa Perturb-seq screens identified perturbations that induced physiological levels of relevant transcription factors and that elicited gene programs resembling in vivo fibroblast states, supporting the feasibility of our method. We will develop three complementary approaches to transform Perturb-seq into a tool for engineering cell states: Aim 1 will establish an iterative selection strategy suitable for reconstructing states when a cell surface marker is known. We apply it to realizing an interesting "universal" fibroblast state in vitro using ordered combinations of CRISPRi/a perturbations. Aim 2 will develop and apply multiomic Perturb-seq to test the hypothesis that perturbation-induced chromatin remodeling can predict synergistic gene combinations genome-wide, yielding a biologically grounded approach for choosing which combinations of perturbations to prioritize. Aim 3 will produce INNsight, an invertible neural network that learns the manifold of reachable fibroblast transcriptional states, removes technical confounders, and enables rational experiment design from Perturb-seq data. The innovation of our approach lies in the use of CRISPRa to perform nuanced perturbations, our development of novel technological extensions of Perturb-seq, our focus on developing principles for efficient, biased exploration of possible combinatorial perturbations, and the development of INNsight, a model that can provably disentangle independent sources of transcriptional variation. The project is significant because it will (1) enable in vitro models of in vivo transcriptional states, facilitating the study of their regulation and functional properties; (2) enhance our understanding of the transcriptional heterogeneity of fibroblasts, a major current area of research in cancer and autoimmune disease; and (3) establish principles for the rational engineering of transcriptional state in differentiated cell types that are directly relevant to engineering cell therapies. Project Number: 1R01HG014186-01 | Fiscal Year: 2025 | NIH Institute/Center: National Human Genome Research Institute (NHGRI) | Principal Investigator: Thomas Norman | Institution: SLOAN-KETTERING INST CAN RESEARCH, NEW YORK, NY | Award Amount: $789,739 | Activity Code: R01 | Study Section: Genomics, Computational Biology and Technology Study Section[GCAT] View on NIH RePORTER: https://reporter.nih.gov/project-details/11103923