Single-cell epigenomics and spatiotemporal transcriptomics reveal human cerebellar development

Abstract

Human cerebellar development is orchestrated by molecular regulatory networks to achieve cytoarchitecture and coordinate motor and cognitive functions. Here, we combined single-cell transcriptomics, spatial transcriptomics and single cell chromatin accessibility states to systematically depict an integrative spatiotemporal landscape of human fetal cerebellar development. We revealed that combinations of transcription factors and cis-regulatory elements (CREs) play roles in governing progenitor differentiation and cell fate determination along trajectories in a hierarchical manner, providing a gene expression regulatory map of cell fate and spatial information for these cells. We also illustrated that granule cells located in different regions of the cerebellar cortex showed distinct molecular signatures regulated by different signals during development. Finally, we mapped single-nucleotide polymorphisms (SNPs) of disorders related to cerebellar dysfunction and discovered that several disorder-associated genes showed spatiotemporal and cell type-specific expression patterns only in humans, indicating the cellular basis and possible mechanisms of the pathogenesis of neuropsychiatric disorders.

Fig. 1. Temporal-spatial molecular diversity of single cells from the developing human cerebellum.

a, b Visualization of seventeen major classes of RNA-seq data (a) and thirteen classes of ATAC-seq data (b) in the developing human cerebellum using UMAP. Each dot represents a single cell, and cells are laid out to show similarities. Each cell color represents the cell type (RL, rhombic lip; VZ, ventricular zone; eCN, excitatory cerebellar nuclei cell; UBC, unipolar brush cell; OPC, oligodendrocyte precursor cell). c, d 10x Genomics Visium data showing the spatial distribution of different clusters in the GW12 (c) and GW17 (d) cerebellum. The expression of known markers is shown using the same layout on the right. e Images of DAPI staining showing the regions of TF-seqFISH in the GW19 cerebellum. TF-seqFISH plots showing the spatial distribution of different cell types in the GW19 cerebellum. Panels 1, 2, and 3 show the RL, VZ and cerebellar cortex, respectively. TF-seqFISH plots showing the gene patterns in the GW19 cerebellum using the same layout on the right. Scale bar, GW19, 500 μm. f Sankey plot showing the correlation between scRNA-seq, scATAC-seq, TF-seqFISH and 10x Genomics Visium Data in GW17. g Heatmap showing the spatial-specific modules for the GW12 cerebellum. h Gene patterns of each module shown in the same layout in (c). i Gene Ontology analysis of spatial-specific modules showing the KEGG pathways or biological processes in the GW12 cerebellum. Dots show the numbers of genes in each module, and the scale bar shows the -log(P-value) for the GO terms. Hypergeometric test.

Fig. 2. Cell diversity and regulon-typed cell-type specificity in the developing human cerebellum.

a Velocity visualization of the RL lineage and VZ lineage in the developing human cerebellum. b Scatterplot of all genes for correlation with the conserved differentiation network across the RL lineage (red plot) and VZ lineage (blue plot). c 10x Genomics Visium data showing the spatial distribution of RL and VZ in the GW12 cerebellum. The expression of region markers is shown using the same layout on the right. d Heatmap showing the expression level and identity of genes in RL progenitors and VZ progenitors. Specific gene expression in each type is shown on the right of the heatmap panel. e Velocity visualization of the UBC, eCN, granule cell and VZ neuron lineages using the same layout as in Fig. 2a. Cells from other lineages are colored gray. f Regulon-target modules showing the lineage trajectory in the developing cerebellum from progenitors to different cell types. RL lineage (top), VZ lineage (bottom). g A dendrogram of regulons on the top for each cell cluster in the main lineage in the cerebellum except for microglia, meninges, T cells, Schwann cells, endothelial cells and other neurons, showing the lineage trajectory in the developing cerebellum. The TFs shown at each branching of the dendrogram are representative of subjacent groups of regulons. h Regulon E2F2 patterns shown in the UMAP plots using the same layout as in Fig. 2a, black, no expression; yellow, relative expression. i Gene expression of RNA (red) and ATAC (blue) across pseudotime for E2F2. The shadow represents the 95% confidence interval around the fitted curve. j, k 10x Genomics Visium data showing the gene E2F2 spatial expression in the GW12 (j) and GW17 cerebellum (k). l Cicero coaccessibility in the region surrounding the E2F2 gene is shown for different cell types. m Regulon-target modules showing the differentiation point between the cerebellar cortex and nuclei in the RL lineage. n Feature Plots of Regulon MEIS3, FOXN3, OLIG3 and MSX1 in the RL lineage. o GO terms of the targets of the cortex or nuclei modules. Hypergeometric test. p Regulon-target modules showing the differentiation point between neurons and glia in the VZ lineage. q Feature Plots of Regulon SOX4, TFAP2A, SOX2 and HES5 in the VZ lineage. r GO terms of the targets of the VZ neuron or glial modules. Hypergeometric test.

Fig. 3. Regulatory networks of Purkinje cell development.

a Cell lineage relationships of progenitors and Purkinje cells analyzed in the developing human cerebellum in ATAC-seq using the same layout as in Fig. 1b (ATAC). Monocle recovered a branched single-cell trajectory beginning with progenitors and terminating at Purkinje cells. Each dot represents a single cell; the color of each cell represents the cell type (left) and pseudotime (right). Cells from other lineages are colored gray. b Smoothed pseudotime-dependent accessibility curves of VZ progenitor and Purkinje cells generated by negative binomial regression and scaled as a percent of the maximum accessibility of each site. The top 10000 highly expressed sites with pseudotime-dependent accessibility are shown. c Regulon PTF1A, FOXP1, BCL11B, RORA, RORB and FOXP4 patterns shown in the UMAP plots using the same layout as in Fig. 1b (RNA), black, no expression; yellow, relative expression. The graphs on the left show the motif sequences of the regulon (left); gene expression of ATAC (blue) across pseudotime for the above genes (middle), the shadow represents the 95% confidence interval around the fitted curve; 10x Genomics Visium data showing the gene spatial expression in the GW12 cerebellum. d Heatmap showing the expression level and identity of marker genes in Purkinje cell subtypes. e The pseudo-time and subtype-specific gene patterns in GW12 spatial data. f Trajectory analysis of nine subtypes of Purkinje cells in the developing human cerebellum using UMAP and FeuturePlots showing PCP4, RORA and RORB expression. Each dot represents a single cell, and cells are laid out to show similarities. Each cell color represents the cell type. g RNAscope images of RORA and RORB in the GW16 cerebellum. Scale bar, 500 μm (top), 100 μm (bottom). The experiments were repeated three times independently with similar results. h Gene ontology analysis of target genes of Regulon RORB showing the predicted function of target genes of Regulon RORB in Purkinje cells in the developing human cerebellum.

Fig. 4. Dynamics of neurogenesis of granule cells in the developing human cerebellum.

a Scatter plots depicting highly expressed peaks annotated by mapping to the hg19 human genome in granule cells. b Gene expression of RNA (green) and ATAC (black) across pseudotime for HEY1, the shadow represents the 95% confidence interval around the fitted curve. c Regulon HEY1 patterns shown in the UMAP plots using the same layout as Fig. 1b (RNA), black, no expression; yellow, relative expression. d GO terms of the target genes of regulon HEY1. Hypergeometric test. e Heatmap showing the differences in developing stages of granule cells. Gene ontology analysis showing the biological functions of different stages of granule cells (right). Hypergeometric test. f, g Scatter plots (f) and gene ontology (g) analysis showing the differentially expressed genes between early- (red plot) and late-stage (blue plot) proliferating granule cells. Hypergeometric test. h Immunofluorescence images of CALB1 and PAX6 in GW16, GW21 and GW27. Scale bar, GW16, 500 μm (left), 200 μm (right, top), 100 μm (right, bottom); GW21, 1000 μm (top), 300 μm (left, bottom), 100 μm (right, bottom). GW27, 1000 μm (top), 300 μm (left, bottom), 100 μm (right, bottom). The experiments were repeated three times independently with similar results. i–k Spatial-specific gene modules showing the gene patterns from the RL to EGL differentiation process (i, j). Modules 2 and 6 are related to the RL and EGL regions, respectively. GO terms showing the KEGG pathways in this process (k). Hypergeometric test. l–n Spatial-specific gene modules showing the gene patterns from the EGL to IGL differentiation process (l, m). Modules 13 and 14 were related to the EGL and IGL regions, respectively. GO terms showing the KEGG pathways in this process (n). Hypergeometric test. o 10x Genomics Visium data showing the spatial distribution of four subclusters of granule cells and the expression of different markers in the GW17 cerebellum.

Fig. 5. Specific genes expressed in the developing human cerebellum.

a The chart shows the predicted mouse age of the developing human cerebellum. The blue triangle shows the real age and predicted age of the developing mouse cerebellum. The red dot shows the predicted mouse age of human data (GW12: E15, GW14: E16, GW16-20: E18, GW24: P0, GW27: P2). b tSNE visualization of human (GW12-27) and mouse (E15-P4) neural lineage cell types analyzed using CCA and color-coded based on CCA joint clusters (top middle) and cell type. The top left graph shows human cell types, and the top right graph shows mouse cell types, with each dot representing a single cell. River plots comparing cell type assignments for humans and mice with CCA joint clusters. c Heatmap showing the conserved genes, mouse-specific genes and human-specific genes between the human (GW12-27) and mouse cerebellum (E15-P4). d Pairwise comparison of gene expression in Purkinje cells between humans and mice. The red dots (human) and blue dots (mouse) indicate differentially expressed genes across species. e Vlnplots of RORB, CA8, FOXP1 and CNTNAP2 expression in Purkinje cells between humans and mice. f 10x Genomics Visium data showing RORB, CA8, FOXP1 and CNTNAP2 expression in the GW17 cerebellum. g Immunofluorescence images of RORB and CALB1 in human GW16 and mouse P0. Scale bar, 300 μm (left) and 100 μm (right) for human GW16; 300 μm (left) and 100 μm (right) for mouse P2. The experiments were repeated three times independently with similar results. h Vln plots of ARHGAP11B, ASPM, SGOL2, NMU and MXD3 expression in progenitors in the human cerebellum. i 10x Genomics Visium data showing ARHGAP11B, ASPM, SGOL2, NMU and MXD3 expression in the GW12 cerebellum. j Overexpression of ARHGAP11B at E11.5 promotes cerebellar cortex folding observed at P2 in mice. Scale bars, 500 μm. k Quantification of the area and perimeter of the mouse cerebellar cortex at P12. **P < 0.01, P value(area)=0.0024, P value(perimeter)=0.0015, two-sided t-test, n = 3 biologically independent animals. mean ± s.e.m.

Fig. 6. Diseases in the human cerebellum.

a Aggregated expression of disease-associated genes in each cell type (coding region, left; noncoding region, right). b Expression patterns of selected disease-associated genes in each cell type in the ATAC dataset. c Normalized ATAC-seq profiles of SPTBN2 and PLEKHG4 in the cerebellum, with all cell types showing activation of SPTBN2 and PLEKHG4.