A spatially-resolved multiomic cell atlas reveals gene regulatory networks underlying cell specification in the developing human heart

Introduction Cardiogenesis requires the differentiation and coordinated function of a large variety of cell types. Congenital heart disease (CHD), affecting 1% of live births[1], and several adult-onset heart diseases result when these processes go awry[2–4]. To understand the cellular and molecular mechanisms at play we must define the repertoire of cell types in the developing heart as well as the cellular niches within which they interact. Single-cell RNA sequencing is a powerful means of defining cell populations, however existing studies of the human foetal heart[5,6] have been hampered by low cell numbers which limits detection of rare but important cell types. To create an improved atlas we increased the number of cells, increased the age range of foetal hearts sampled, and generated spatial transcriptomic data to map the defined cells to multicellular niches of co-located and interacting cells. To better infer the transcription factors (TFs) governing cell identity using gene regulatory network (GRN) inference we also generated paired single-cell resolution RNA and chromatin accessibility profiles using Assay for Transposase-Accessible Chromatin (ATAC) sequencing. Methods We performed paired single-nuclei RNA and ATAC sequencing (10X Multiome) and single-cell RNAseq (10X 3’ RNAseq) on 23 human foetal hearts ranging from 4 to 20 post-conception weeks. In addition we generated spatial transcriptomic data (10X Visium) from these hearts. We validated cell types defined in single-cell data using protein-level immunofluorescence imaging. Results Our healthy reference atlas includes 300k cells and nuclei. Using their gene expression profile we resolved 46 fine-grained cell types (FIGURE 1). We used cell2location[7], a machine learning classifier trained on the atlas to map the defined cell types to the spatial data (FIGURE 2) and detected niches of co-locating cells, such as a macrophage population in the walls of the great vessels predicted to interact with mural cells. We were able to define location-specific transcriptional signatures such as cardiomyocytes of each chamber (as well as the compact and trabeculated myocardium) and smooth muscle cells specific to great vessels vs coronary arteries. Using a combination of RNA and ATAC data we inferred cell-specific GRNs, which implicated novel TFs such as MSX2 in sinoatrial node pacemaker cells and the enrichment of CHD genes amongst cardiomyocyte-specific GRNs. In the spatial data we showed that gene modules associated with different forms of CHD enrich in specific anatomical regions, such as the outflow tract. Discussion: Together these data constitute the most comprehensive cardiac developmental cell atlas to date. It sheds new light on the diversity of cell states, their niches, and their vulnerability to the genetic causes of CHD.