Compass: FBA-Powered Modeling of Single-Cell RNA-Seq Characterizes Cell-to-Cell Metabolic Heterogeneity and Reveals Novel Therapeutic Targets in Autoimmunity
Conference on Constraint-Based Reconstruction and Analysis (COBRA)
2018
5th Conference on Constraint-Based Reconstruction and Analysis (COBRA 2018)
General Submissions
New Frontiers 2
Sunday, October 14, 2018 - 4:30pm to 4:45pm
The rapid advance in single-cell RNA-Sequencing (scRNA-Seq) is one of the most exciting recent developments in biomedical research. By comprehensively quantifying the transcriptome of individual cells, rather than the average over many cells, scRNA-Seq allows, for example, to discover rare cell sub-types and to retrace intermediate steps in lineage development. However, scRNA-Seq requires tailor-made computational methods that take advantage of its novel properties while accounting for its unique technical limitations (e.g., dropouts), and its immense data magnitude, which is fast approaching millions of cells per experiment (Wagner et al., Nature Biotechnology 2016). Here, we answer these challenges in the realm of cellular metabolism. We present COMPASS, an FBA algorithm for comprehensive characterization of single-cell metabolic states. COMPASS capitalizes on the magnitude of scRNA-Seq data by modeling cells as points in a high-dimensional metabolic space, while mitigating scRNA-Seq data sparsity by smoothing kNN neighborhoods. The resulting metabolic space is directly interpretable in mechanistic terms. It allows data-driven characterization of metabolic heterogeneity among cells and associating metabolic programs with phenotypes of interest. To demonstrate COMPASS, we study T helper 17 (Th17) cells, which are remarkably plastic and mediate both autoimmunity and immune tolerance. A COMPASS characterization of Th17 metabolic heterogeneity recovered known associations of Th17 effector states with metabolic programs and predicted novel metabolic intervention targets. We show that the glycolytic shift in pathogenic Th17 is associated with extensive remodeling of anabolic pathways, which depends on pyruvate dehydrogenase (PDH). We then implicate the polyamine pathway, which has been scarcely studied in autoimmunity, in Th17 pathogenicity. In vivo deletion of either PDH or polyamine synthesis led to considerably better clinical outcome in a murine model of multiple sclerosis (MS). The clinical significance of our findings is supported by differential abundance of polyamines in the blood of healthy human donors compared with MS patients.