Integration of multimodal single-cell data to characterize T cell phenotypes, gene regulation, and disease associations

Abstract

T cells are heterogeneous immune cells with a spectrum of functional phenotypes—cell states—critical to processes ranging from infection to autoimmunity. Traditionally, T cells have been partitioned into mutually exclusive states demarcated by a few predefined surface proteins, transcription factors, T cell receptors, or cytokines. However, this approach uses a limited set of markers, often from a single modality. Furthermore, in practice, states are dynamic, continuous, and may even co-exist in the same cells. New technologies such as CITE-seq enable high-throughput multimodal measurements from the same single cells. Leveraging CITE-seq data from multiple cohort studies, I used an integrative single-cell approach to more precisely define T cell states, measure their associations with diverse phenotypes, and characterize the regulatory architecture mediating these states’ effects on gene expression. First, I developed a statistical strategy using canonical correlation analysis to integrate single-cell mRNA and surface protein expression. I used this approach to analyze >90,000 T cells from rheumatoid arthritis (RA) synovium and demonstrated that T cell states thought to be general markers of RA—such as peripheral helper T cells—were specifically expanded in certain disease subtypes. Next, I analyzed 500,089 memory T cells from a steady-state Peruvian tuberculosis disease (TB) progression cohort. I found that people who progressed to active TB had reduced abundance of a TH17-like state that produced both IL-17 and IFNγ in response to Mycobacterium tuberculosis peptide antigens. Finally, I modeled the state-dependent gene-regulation architecture of these memory T cells. Using a single-cell Poisson mixed effects model, I comprehensively characterized memory T cell expression quantitative trait loci (eQTLs), showing that one third varied with continuous states such as cytotoxicity or regulation. This work demonstrates the importance of multimodal single-cell analysis to more precisely define memory T cell states associated with clinical variables, demographics, and gene regulation. Association of these cell states with disease outcomes and the colocalization of state-specific eQTLs with disease-associated loci suggest these states are important in disease etiology. These statistical approaches are broadly applicable to many cell types and contexts and motivate further interrogation of functional states through a single-cell lens.