Abstract
Human cerebellum encompasses numerous neurons, exhibiting a distinct developmental paradigm from cerebrum. Here we conducted scRNA-seq, scATAC-seq and spatial transcriptomic analyses of fetal samples from gestational week (GW) 13 to 18 to explore the emergence of cellular diversity and developmental programs in the developing human cerebellum. We identified transitory granule cell progenitors that are conserved across species. Special patterns in both granule cells and Purkinje cells were dissected multidimensionally. Species-specific gene expression patterns of cerebellar lobes were characterized and we found that PARM1 exhibited inconsistent distribution in human and mouse granule cells. A novel cluster of potential neuroepithelium at the rhombic lip was identified. We also resolved various subtypes of Purkinje cells and unipolar brush cells and revealed gene regulatory networks controlling their diversification. Therefore, our study offers a valuable multi-omics landscape of human fetal cerebellum and advances our understanding of development and spatial organization of human cerebellum.
Similar content being viewed by others
Introduction
The complex array of neuron types in the cerebellum occupies >80% of total neurons in the human brain1. Integrated neural networks in the different cerebellar regions coordinate various functions including motor, cognition, emotion and language2,3,4,5. Impairment of cerebellar circuits would result in dysmetria of thought or movement6. Reduced grey matter in the cerebellar cortex was also observed in autism spectrum symptom or attention disorder7. The architecture of the cerebellar cortex is relatively conserved throughout the evolution8. Almost all kinds of vertebrates have three layers in the mature cerebellum, namely molecular layer (ML), Purkinje cell layer (PCL) and granule cell layer (GCL)8. Above the white matter, the GCL contains intensive granule cells (GCs), which is covered by a layer of Purkinje cells (PKCs)9. The ML consists of the dendritic trees of PKCs and the parallel fibers of GCs9. Various kinds of interneurons interweave within different layers and harness the signal transduction in the cerebellum10.
During the development, the majority of neurons in the cerebellum are derived from two germinal centers, the ventricular zone (VZ) and upper rhombic lip (URL)11 (Fig. 1a). GABAergic lineages including PKCs and various interneurons are generated from the VZ, as well as Bergmann glia11. PKCs are generated first, then radially migrate and are organized into the PCL11. Excitatory lineages including deep cerebellar nucleus neurons, GCs and unipolar brush cells (UBCs) are derived from the URL12. While the UBC progenitors settle in the URL, proliferative GC progenitors migrate tangentially to cover the surface of the cerebellum, forming the external granular layer (EGL) as the secondary germinal zone12,13,14. Since postmitotic GCs form the inner EGL (iEGL) after inward migration, the separation between GC progenitors in the outer EGL (oEGL) and GCs occurs14. GCs will continue to migrate and pass through PKCs to form the internal granular layer (IGL) which finally develops into the GCL14.
a Schematic overview of the development of major cell types in the human cerebellum (top left) and time points of fetal samples and corresponding biological events (top right). Schematic overview of the multi-omic analyses (bottom). b Visualization of integrated human scRNA-seq data with cell identities color coded. GCPs, GC progenitors. PROs, progenitors. INs, interneurons. c Gene expression patterns of marker genes in integrated scRNA-seq. Grey represents low levels and navy blue represents high levels. d Heatmap showing the marker genes of different cell types. e Gene velocity flow map visualized on UMAP embedding of integrated data. Most naïve sites are labeled by green and pink arrowheads. Streamlines represent the RNA velocity predicting the future transcriptional dynamic state of cells as concluded by blue, yellow, and pink curves. f Contribution of human samples and major cell types from different time points in scRNA-seq data (left). Line graph showing the ratio change of different cell types (right). g Correlations between scRNA-seq data from different time points. h–k Spatial transcriptomic data of coronal slice at GW13 (h) and sagittal slice at GW16 (j) with region identities color coded. Spatial gene expression patterns of marker genes of different cell types at GW13 (i) and GW16 (k). l Correlations between spatial transcriptomic data and scRNA-seq data from different time points.
Several studies have applied single cell technologies to investigate the details of mouse cerebellar development from embryonic to postnatal stages15,16,17,18,19,20,21,22,23,24. However, our understanding of human cerebellar development is relatively limited25,26,27. Here, we adopted single-cell multi-omics approaches, including single-cell RNA sequencing (scRNA-seq), single-cell sequencing assay for transposase-accessible chromatin (scATAC-seq) and spatial transcriptomics, to delineate the developmental landscape of human cerebellum at high spatial-temporal resolution. We identified different GC progenitors which are conserved across species and characterized the species-specific spatial pattern of gene expression within GC lineages under different dimensions. Multidimensional features of PKC subtypes were identified. Gene regulatory networks (GRNs) involved in the diversification of different cell types were generated to facilitate our understanding of cerebellar lineage development.
Results
High-quality multi-omics profiling of the developing human cerebellum
We applied scRNA-seq, scATAC-seq and spatial transcriptomics approaches to obtain high-quality multi-omics profiles of the developing human cerebellum. scRNA-seq covering human GW13–GW18 samples, spatial transcriptomic analysis (Stereo-seq) of GW13 and GW16 samples, and scATAC-seq of GW13 sample were performed (Fig. 1a). A series of published mouse scRNA-seq data were also re-analyzed for the cross-species comparison15,19.
Transcriptional profiles of 72,190 high-quality human cells were obtained with an average capture of 2774 genes/cell (Supplementary Table S1). Microglia, endothelial cells and red blood cells were removed in each dataset. Using Seurat package28, scRNA-seq data of the remaining 62,247 cells were integrated and visualized under the embedding of uniform manifold approximation and projection (UMAP) (Fig. 1b). GABAergic lineages such as interneurons (PAX2+), PKCs (CA8+WNT7B+) and excitatory lineages including UBCs (EOMES+LMX1A+), GC progenitors (ATOH1+LHX9–) and postmitotic GCs (NHLH1+LHX9–TBR1–) were identified (Fig. 1c; Supplementary Fig. S1a). HES1+ progenitors and glia cells which are mainly derived from the VZ were also classified (Fig. 1b, c). More cell type-specific markers were explored using the COSG method29 (Fig. 1d; Supplementary Table S2).
Previous studies have found that ATOH1 is highly expressed in URL-derived excitatory progenitors including GC progenitors in the EGL, while PTF1A is enriched in progenitors emerging from the VZ8. The RNA velocity analysis identified the consistent naïve progenitors as reported (Fig. 1e). Co-expression of G2/M phase markers (e.g., TOP2A) and ATOH1 was observed, while the co-expression of TOP2A and PTF1A was barely detected (Fig. 1c; Supplementary Fig. S1a). It is likely due to the small proportion of PTF1A+ progenitors compared with excitatory progenitors. Furthermore, the highest abundance of PKCs was observed in GW13, which decreased sharply later (Fig. 1f; Supplementary Fig. S1b, c and Table S3). Plausibly, PKC production is already completed around GW1330 (Fig. 1a) while the EGL is rapidly proliferating, producing large numbers of GCs that quickly outnumber PKCs. The ratio of UBCs was found to peak around GW16 (Fig. 1f), consistent with a very recent study26. Each dataset also correlated well with data from adjacent time points (Fig. 1g). These results suggested that our scRNA-seq data generally captured the characteristics of human cerebellar development.
Our spatial transcriptomics data were analyzed at bin 50 resolution (50 × 50 DNA nanoballs (DNBs), diameter is ~35 μm, see Materials and methods). 3609 bins with an average of 2481 genes/bin and 15,880 bins with an average of 1890 genes/bin were profiled in samples collected at GW13 and GW16, respectively (Fig. 1h–k; Supplementary Table S4). Correlation between scRNA-seq and spatial transcriptomic data were calculated based on pseudobulk of each dataset (Fig. 1l). Unsupervised clustering based on binned data suggested a multi-layered annular distribution of different cell types in GW13 spatial transcriptomic data, correlated with the oEGL, iEGL, IZ and VZ regions, respectively (Fig. 1h; Supplementary Fig. S1d). Clear separation of the rhombic lip (RL), EGL, and Purkinje cell layer (PKL) was also detected in GW16 spatial transcriptomic data, as well as interposed nucleus and nucleus embedded in the brainstem (Fig. 1j; Supplementary Fig. S1d). By using ArchR package31, the accessibility of each gene in GW13 scATAC-seq dataset was scored, followed by cell type clustering analysis (Supplementary Fig. S1e–g). We found a comparable resolution of cell type classification by using scATAC-seq dataset solely or integration of scATAC-seq and scRNA-seq datasets (Supplementary Fig. S1e).
Developmental trajectory and transient amplifying progenitors in GC lineages
GCs account for the majority of cerebellar excitatory neurons featured by the expression of NRN11,32. To investigate the development of GCs, we first performed cell cycle scoring of NRN1+EOMES– GC lineage cells. Highly proliferative SFRP1+ATOH1+ GC progenitors in G1 or S/G2/M phase, as well as NHLH1+STMN2+ postmitotic GCs could be distinguished (Fig. 2a; Supplementary Fig. S2a, b). The gene ontology (GO) enrichment analysis confirmed the high activity of axon development in postmitotic GCs compared with progenitors (Supplementary Fig. S2c), suggesting the ongoing of neural differentiation.
a Visualization of GC lineages in different phases of cell cycle (top). Pseudotime trajectory visualized on UMAP embedding of G1 phase GC progenitors and postmitotic GCs (bottom). Blue, orange and pink arrows mark three potential sublineages suggested by the trajectory. b Altered expression patterns of different genes along the pseudotime. An overview (left) and the detailed patterns of several genes (right). c Visualization of the AT+GCP and ND+GCP (left) and violin plot showing the expression patterns of their marker genes (right). d Spatial gene expression patterns showing the separated expressions of ATOH1 and NEUROD1 in both GW13 and GW16 EGL in human cerebellum. e Enrichment plots of GO and KEGG terms in AT+GCP compared with ND+GCP. f GRNs of GC development. g Gene expression patterns of SFRP1 and PRPH in scRNA-seq and spatial RNA-seq data. h Spatial gene expression patterns of CPLX3 and SV2B at GW16 with arrowheads highlighting the enrichment in the EGL (left). Protein interaction network according to STRING (right).
Pseudotime trajectory of GC lineage was then constructed based on G1 phase cells by using Monocle3 package33 (Fig. 2a), with the most naïve point calculated by using CytoTRACE and StemSC packages34,35 (Supplementary Fig. S2d). We noticed that the elevation of NEUROD1 expression tends to be earlier than other markers of postmitotic GCs along the trajectory (Fig. 2b). Co-expression of NEUROD1 and S/G2/M phase markers was also found, which indicated the proliferative feature of these cells (Supplementary Fig. S2a). Further classification distinguished two groups of GC progenitors, which were labeled as ATOH1+ GC progenitors (AT+GCPs) and NEUROD1+ GC progenitors (ND+GCPs). While the ND+GCPs expressed higher level of NEUROD1 and ELAVL2, almost no SOX2 was detected in ND+GCPs compared with AT+GCPs (Fig. 2c). Similarly, distinction of these two GC progenitors could be found in mouse (Supplementary Fig. S2e, f). As reviewed recently, the ND+GCPs resemble the transient amplifying progenitors in mouse, which reside beneath the AT+GCP36. Such an inside-outside separation could be observed especially in our GW13 spatial transcriptomic data, in which clear boundary of iEGL and oEGL was detected (Fig. 2d). Gene set enrichment analysis (GSEA) indicated the activation of chemokines, Ras or Rap1 signaling pathways regarding the morphological changes during neural differentiation in both human and mouse ND+GCPs, suggesting the conserved regulatory pathways across species (Fig. 2e; Supplementary Fig. S2g).
Furthermore, transcription factors enriched in either progenitors or postmitotic cells were selected for the GRN analysis through the integration of scRNA-seq and scATAC-seq data. We leveraged the inferred developmental trajectories and the GRNs to identify key genes which may regulate the cell state transitions during the differentiation of GCs (Fig. 2f). Apart from ATOH1, transcription factors like YBX3 and MEF2C were discovered to be potentially involved in the development of GCs (Fig. 2f). MEF2C could be detected in the GCs of human embryonic cerebellum while its function remains unclear37.
Interestingly, we found a marker PRPH in postmitotic GCs based on the pseudotime trajectory of GC lineages. PRPH showed an almost complementary expression pattern with SFRP1 but not detectable in the most matured GCs (Fig. 2g). Consistent with scRNA-seq data, non-overlapping expression patterns of SFRP1 and PRPH was visualized in both of our spatial transcriptomic data (Fig. 2g). The expression of PRPH was almost constrained in the iEGL and hardly detected in the migrating and IGL GCs (Fig. 2g). PRPH could be used as an iEGL marker which is specifically expressed in the postmitotic GCs. Genes with similar expression patterns were further explored. We found that CPLX3 was uniquely detected in anterior iEGL while SV2B was enriched in iEGL with less specificity (Fig. 2h). PRPH encodes the neuronal intermediate filament protein peripherin whose mutation could result in disruption of neurofilament network assembly38. Both CPLX3 and SV2B were reported to participate in the regulation of neurotransmitter secretion39,40. These proteins seemed to be involved in a protein interaction network according to the STRING database41 (Fig. 2h). We speculated that this protein interaction network was involved in the regulation of GC migration in a region-specific way.
GC sublineages and their multidimensional spatial distribution
Neural circuit analyses have unveiled the functional preference in different cerebellar regions, such as anterior lobes for sensorimotor functions, and lateral posterior lobes for cognition2. Since CPLX3 is specifically expressed in the anterior region (Fig. 2h), three sublineages of GCs labeled in the pseudotime trajectory may also resemble the GCs organized in different regions (Fig. 2a). Further analysis unveiled mutually exclusively expressed gene cohorts represented by BARHL1 and TLX3 in both GC progenitors and postmitotic GCs (Fig. 3a; Supplementary Fig. S3a and Table S5). The anterior-posterior (A-P) distribution of BARHL1 and TLX3 was clearly observed especially in the sagittal section of GW16 (Fig. 3b). Three sublineages recognized in the pseudotime trajectory corresponding to the anterior lobes, posterior lobes and flocculonodular lobe, with both flocculonodular and posterior lobes being TLX3+ in human (Fig. 3c, d). Similar gene expression patterns could be found in both scRNA-seq and spatial transcriptomic data, such as CERKL enriched in the flocculonodular lobe, NTF3 in the posterior lobes and TRH in both flocculonodular and anterior lobes (Fig. 3d). Region-specific gene expression in either EGL progenitors or IGL postmitotic GCs could be observed (Fig. 3d).
a Gene expression patterns of BARHL1 and TLX3 in GC lineages (top). Volcano plot showing the complementarily expressed gene cohorts (bottom). Symbols of almost evenly expressed genes in each subtype are colored blue (anterior) or green (posterior). b Spatial gene expression patterns of BARHL1 in anterior regions and TLX3 in posterior regions in both GW13 (top) and GW16 (bottom). SFRP1 labeled GC progenitors. c Schematic overview of three GC sublineages corresponding to three lobes in the human cerebellum. d Spatial organizations of GC sublineages in EGL and IGL. Genes detected in both EGL and IGL (top). Genes (the first gene corresponds to the outside one) enriched in EGL of each lobe (middle). Genes enriched in IGL of each lobe (bottom). e In situ hybridization results of Tlx3 and Parm1 in mouse at E15.5 and E18.5 adopted from Allen Brain (https://developingmouse.brain-map.org). Arrowheads indicated the detection of Parm1 in anterior but not in posterior lobes of mouse. f Correlation of human (left) or mouse (right) gene cohorts in human and mouse datasets showing species-specific spatial gene expression patterns. In each panel the bottom-left parts are human datasets. Only G1 phase GC lineage cells were calculated. g GRNs showing the potential regulatory programs of anterior (blue) or posterior (green) specific genes (top left) and visualization of accessible regions in different genes (bottom right). h Visualization of four subtypes of GC progenitor cells in scRNA-seq data and the expression of their marker genes (top left). Spatial distributions of PRR35, EBF2 and HEY1 in oEGL of GW13 (top right). Spatial distributions of HEY1, PRR35, EBF2 and GALNTL6 in EGL of GW16 (bottom).
Both a previous report and in situ hybridization data from Allen Brain showed that mouse Tlx3 are highly expressed in the posterior lobes but not in the flocculonodular lobe42,43 (Fig. 3e), challenging the evolutionary conservation of spatially specific gene expression patterns. Surprisingly, gene cohorts related to the distribution along A-P axis discovered in human showed almost no correlation with mouse data and vice versa, suggesting a dramatic cross-species difference (Fig. 3f; Supplementary Fig. S3b–d). For example, PARM1 was enriched in the posterior lobes of human while in the anterior lobes of mouse, exhibiting completely reverse distribution across species (Fig. 3d, e). For the first time such a significant difference was uncovered and the functional interpretation remains to be explored.
Taking advantage of the integrated scRNA-seq and scATAC-seq data, GRNs involved in the A-P distribution of GCs were generated (Supplementary Fig. S3e). Several transcriptional factors seemed to be core regulators during the diversification, such as RORB, RORA, ESRRG in the anterior region and TLX3, ETV5, GLI3 in the posterior region (Fig. 3g). According to the integrated data, GCs in different regions showed preference of accessibility within several marker gene loci, consistent with their expression patterns (Fig. 3g). These genomic regions might be the specific enhancers modulating the region-specific activation of the gene expression in the developing human cerebellum.
Besides the separation along A-P axis, variance in dorsal-medial (DM) and ventral-lateral (VL) axis was also identified in the oEGL of GW13 spatial transcriptomic data. GC progenitors could be divided into four groups according to the expression of EBF2, PRR35, GALNTL6, and HEY1 in the scRNA-seq data (Fig. 3h). PRR35 was specifically expressed in the VL region of the oEGL, while EBF2 and HEY1 seemed to be enriched in the DM region (Fig. 3h; Supplementary Fig. S3f). The DM-VL difference reminded us of the heterogeneity between cerebellar vermis and hemispheres. EN1 plays a vital role in the development of cerebellar vermis44 and it was highly expressed in most of the anterior EBF2+ subtype in our data (Fig. 3h). We speculated that PRR35 and EBF2 were uniquely expressed in the GC progenitors at hemisphere and vermis, respectively. GALNTL6 was enriched in the flocculonodular lobe as reported21. The HEY1+ cells were located in the most posterior region and they might develop into the nodule of the cerebellum.
Additional spatial patterns were also detected, which seemed not to exactly overlap with the characteristics mentioned above. Both NPY and PPP1R17 were expressed in the GALNTL6+ GC progenitor subtype, while NPY was also detected in all PRR35+ cells and PPP1R17 was only detected in part of them (Supplementary Fig. S3g). The biological importance of multidimensional patterns in GC lineages awaits future investigation. To understand the functional implication of various GC sublineages, we referred to the Human Phenotype Ontology database45 to inspect the risk genes associated with the cerebellar diseases. Acrocallosal syndrome featured with cerebellar hypoplasia is associated with GLI3 mutation. GLI3 was highly expressed in the flocculonodular lobe GC progenitors (Fig. 3d), suggesting the important role of specific lobule in the cerebellar development.
In conclusion, we dissected the spatial distribution pattern of genes in the developing human cerebellum along the A-P axis or DM-VL axis. Significant differences in the spatial gene expression pattern along A-P axis between human and mouse was discovered.
UBC lineage development
The UBCs are excitatory interneurons enriched in the median cerebellar cortex and part of flocculus/paraflocculus complex13. Both UBCs and GCs are considered as neuronal lineages generated from WLS+ progenitors in the RL46. Here, we described a newly characterized potential neuroepithelium (labeled as UBCpro-1) which may give rise to classic WLS+CALCB+ UBC progenitors (labeled as UBCpro-2) (Fig. 4a).
a Visualization of the UBC lineages. UBCpro-3cc was short for S/G2/M phase UBCpro-3. b Dot plot showing the expression patterns of marker genes of different cell types in a. Genes used in the following illustration of the compartmentation of the RL are colored bule. c Heatmap showing the high strength of co-expressed gene module containing EOMES and CALCB in URL (left). The RNA count map and fluorescence image of human GW16 cerebellum with URL marked by arrowheads (right). d Spatial gene expression patterns showing the distribution of COL2A1+ cells in URL, while CALCB+ and EOMES+ cells established the VZ and SVZ of URL in both GW13 and GW16 human cerebellums. e Heatmap showing the scaled regulon strength in different UBC progenitors. f Gene expression patterns of CNTNAP5 and LMX1A in different UBC sublineages (left). Gene expression patterns of the reported markers for UBC sublineages (right). g Heatmap showing the correlation coefficients of marker genes for UBC sublineages. h GRN showing the potential regulatory programs of different UBC sublineages.
According to the scRNA-seq data, we found a new cluster of COL2A1+PIK3C2G+ progenitors (UBCpro-1) which expressed high level of stemness-related markers such as SOX2 and CYP26B1 (Fig. 4b; Supplementary Fig. S4a). According to the UMAP embeddings, there seemed to be some relevance between UBCpro-1 and classic WLS+ UBC progenitors (Fig. 4a). We then referred to the spatial transcriptomic data to investigate the distribution of these cells. Firstly, the RL was defined by co-expressed gene module analysis of GW13 data and by the morphology of GW16 data (Fig. 4c; Supplementary Fig. S4b). UBC markers including EOMES, LMX1A and OTX2 were detected in the labeled RL regions47 (Fig. 1i, k; Supplementary Fig. S4c). The compartmentation of the RLVZ and the RLSVZ (subventricular zone of the RL), which was featured by the expression of CALCB and EOMES, respectively47, was also recaptured in our spatial transcriptomic data (Fig. 4d). The inside-out structure of CALCB and EOMES was observed and COL2A1 seemed to be highly expressed in the most inside region (Fig. 4d). Thus, we speculated that the UBCpro-1 was the neuroepithelium which gave raise to classic WLS+CALCB+ UBC progenitors. Regulon-based analysis was also applied and identified key transcription factors, such as ASCL1, ETV3, ELF1 and ATOH1, which may regulate the cell state transition during differentiation of UBC lineages (Fig. 4e).
Mature UBCs could be divided into two subtypes based on the expression of CALB2 or GRM113,48,49. We also identified two potential subtypes of UBCs which could be distinguished by the expression of LMX1A and CNTNAP5 (Fig. 4f). LMX1A+ UBCs were more likely to be the typical GRM1+ UBCs with molecular features as reported previously50 and CNTNAP5+ UBCs were possibly immature CALB2+ UBCs (Fig. 4f). KEGG enrichment analysis indicated that genes highly expressed in the CNTNAP5+ cells, such as KCNQ5 and PLCB1, were associated with cholinergic synapse (Supplementary Fig. S4d) which was found to be functionally related to the CALB2+ UBCs13. Using markers of two UBC lineages (Fig. 4g), GRNs involved in the diversification of UBC lineages were constructed (Fig. 4h). We found EOMES and PBX3 as core regulators controlling the specification of all UBC lineages, while SP5 might be critical for the development of CNTNAP5+ sublineage (Fig. 4h). Integrated analysis of human and mouse scRNA-seq datasets19 identified a very similar expression pattern of SP5 across species, indicating the conserved role of SP5 in regulating the diversification of UBC lineages (Supplementary Fig. S4e). Altogether, our analyses revealed the potential developmental progress of UBC lineages and the underlying gene regulatory logic likely involved in the early cell fate specification.
VZ progenitors and interneuron development
Similar to the role of ganglion eminence in the developing forebrain, the VZ gives rise to multiple types of GABAergic neurons including PKCs and various interneurons8. To characterize the development of GABAergic lineages, VZ-derived progenitors and neurons were selected for further analysis (Fig. 5a). Despite the neuronal lineages, we discovered several non-neuronal cells including OLIG2+PDGFRA+ oligodendrocyte progenitor cells and PIFO+RSPH1+ ependymal cell progenitors51,52 (Fig. 5a, b). Bergmann glia cells which facilitate the migration of PKCs and GCs could also be identified according to the expression of TNC and FAM107A. Moreover, LINC01727 seemed to be a specific marker of Bergmann glia in the cerebellum (Fig. 5b). Interestingly, the major exons in LINC01727 gene are truncated in rodent53 (Supplementary Fig. S5a), suggesting the potential cross-species difference.
a Visualization of different progenitors, GABAergic neurons and glia cells. b Violin plots showing the expression patterns of marker genes for cell types in a. c Pseudotime trajectory from the VZ progenitors to interneurons (left top). Gene expression patterns of PANTR1 and IGFBP5 (left bottom). GO enrichment results of PANTR1+ and IGFBP5+ interneurons (right). d Visualization of three PKC subtypes (top) and dot plot showing corresponding expression patterns of marker genes (bottom). e Multidimensional spatial gene expression patterns of PKCs revealed by GW16 Stereo-seq and corresponding gene expression patterns revealed by scRNA-seq. f Gene expression patterns of PLCB4, CADM1 and CADM2 in PKCs revealed by the scRNA-seq and GW13 Stereo-seq (bottom). In situ hybridization results of Plcb4 and Cadm1 in mouse at E15.5 (top) adopted from Allen Brain (https://developingmouse.brain-map.org). g Schematic overview of PKC developmental progress in mouse (left). Expression of FOXP1 and RORB in human GW13 cerebellum along L-M axis (right). h GRN showing potential regulatory programs of different PKC subtypes (top right). Visualization of unsupervised clusters in scATAC-seq data and scores of their marker genes with blue representing low levels and yellow representing high levels (bottom left). i Visualization of accessible regions of different genes in three PKC subtypes. Specific accessible regions are emphasized by deep grey panel. Blue or red color of gene loci represents the reverse or forward transcriptional direction, respectively.
Since the neurogenesis of PKCs has already been completed in our samples, pseudotime analysis was only applied to the potential interneuron lineages (Fig. 5c; Supplementary Fig. S5b). We noticed that the postmitotic interneurons could be dissected through mutually exclusive expression pattern of PANTR1 and IGFBP5. A recent study indicated that IGFBP5 was a pan-marker of Golgi interneurons during human embryonic cerebellar development54. GO enrichment analysis suggested the shift from active axonogenesis in PANTR1+ cells to active synapse organization in IGFBP5+ cells (Fig. 5c), indicating the maturation progress of the interneurons.
PKC subtypes and their GRNs
PKCs execute critical functions in the cerebellar circuits with a relatively limited cell number55. Mature PKCs could be divided into two subtypes by the alternating strips of Zebrin II (protein encoded by gene ALDOC) expression along the lateral-medial (L-M) axis56. Here we identified three molecularly distinct subtypes of PKCs labeled as PKC1, PKC2 and PKC3 in human (Fig. 5d; Supplementary Fig. S5c). PKC1 was EBF2+IGF1+ cells, corresponding to the reported Zebrin II negative PKCs57,58,59 (Fig. 5d). Therefore, both LSAMP+PCP4L1+ PKC2 and SLA+PBX3+ PKC3 with the expression of FGF3 might be subtypes of Zebrin II positive PKCs (Fig. 5d). Separation of both IGF1-FGF3 and PCP4L1-SLA could be found in our GW16 spatial transcriptomic data (Fig. 5e), suggesting the correspondence between three molecular subtypes of PKCs and their spatial organizations. Furthermore, PKCs could also be dissected multidimensionally (Fig. 5e). The distribution of POSTN and IGSF21 resembled the A-P pattern mentioned in the GC lineages (Fig. 5e). While POSTN was reported to be a neurite outgrowth-promoting factor60, IGSF21 was shown to regulate inhibitory presynaptic differentiation61. Such a difference might implicate the functional diversification or the asynchronized developmental rates between anterior and posterior PKCs. Different members of semaphorin family proteins which regulated cell migration were found to be enriched in different regions (Fig. 5e). Previous study showed that Sema3e and Sema3d mediates endothelial cell repulsion through distinct molecular pathways62. Similar mechanisms may underpin the spatial organization of PKC subtypes.
We also discovered a conserved separation between CADM1 and PLCB4/CADM2 along the dorsal-ventral (D-V) axis, especially in the posterior region during the early development of PKCs (Fig. 5f). The expression of CADM1 seemed to be enriched in the ventral region, which could be detected in both human (GW13) and mouse (E15.5) (Fig. 5f). Such separation became relatively subtle later at GW16 and did not follow the distribution along D-V axis. Both CADM1 and CADM2 serve as membrane proteins engaged in the cell–cell adhesion and are important for the neurite arborization63,64. CADM1 was reported as an autism-associated gene and its knockout would lead to smaller cerebellum and decreased synapse of PKCs63. Our discovery suggested the potential role of CADM1 in the organization of PKCs during the early development and the disturbance of this distribution may also underpin the autism caused by CADM1 mutation.
We also inspected the cross-species similarity of PKCs (Supplementary Fig. S5d–i). As reported in a recent study, mouse Car8+ PKCs could be grouped into four subtypes26 (Fig. 5g). Spatial segregation of FOXP1high and RORB+ PKCs along L-M axis was observed in GW13 spatial transcriptomic data, which was similar between human and mouse (Fig. 5g). However, both ABHD3 (or ETV1) and CDH9 which marked the proposed late-born PKCs were hardly detected in GW13 spatial transcriptomic data (Supplementary Fig. S5j). It remains unclear whether such difference was due to the immature state of late-born PKCs or the cross-species heterogeneity.
Taking advantage of integrated scRNA-seq and scATAC-seq data, we also generated the GRNs and investigated the potential regulatory elements involved in the diversification of PKC lineages (Fig. 5h). PKCs could also be classified into three types in our scATAC-seq data (Fig. 5h). We found that EBF2 might specifically activate SEMA3A through an enhancer distant from the promoter (Fig. 5i), which was consistent with the shapely decreased Sema3a+ PKCs in Ebf2-null mice as reported65. NR2F2 and ZNF664 might also be involved in the differentiation of different PKC subtypes (Fig. 5i).
Discussion
In this study, we employed single-cell multi-omic technologies to systematically investigate the early development of the embryonic human cerebellum. Acquisition of high-quality data facilitated the in-depth understanding of molecular heterogeneity, spatial organization, and developmental trajectories of neuronal lineages in the developing human cerebellum. Based on the integration of scRNA-seq and scATAC-seq, we characterized the GRNs involved in the development of GC lineages. Conserved GC progenitors exhibiting varying degrees of stemness were identified. Beyond stemness, GC lineages could be further distinguished by their spatial characteristics which was validated in our spatial transcriptomics data. Three GC sublineages were matched to the different lobes of the cerebellum. Cross-species heterogeneities of gene cohorts enriched in either anterior (BARHL1+) or posterior (TLX3+) regions were identified, represented by PARM1 with completely reverse expression pattern along A-P axis in humans and mouse. Potential regulators involved in the diversification of anterior and posterior GCs were identified based on GRNs. GC progenitors could also be dissected according to the expression of EBF2, PRR35, GALNTL6, and HEY1, suggesting the complexity of the spatial organization in GC lineages during the development. Our analysis suggested the potential COL2A1+ neuroepithelium of UBC lineages. PKC subtypes could be dissected multidimensionally in our spatial transcriptomics data. Different semaphorin proteins were enriched in different regions, suggesting the distinct molecular pathways involved in the organization of PKC subtypes. Potential regulators of each PKC subtype were also identified.
Morphology analysis of cerebellum has revealed cross-species differences in the development of URL1. Our investigation brought forth additional insights into the aspect of molecular features. Although the general cellular architecture of the developing cerebellum is largely conserved, many differences seem to exist in the gene expression patterns and spatial distribution of certain cell types between humans and mice. For example, the AT+GCPs and ND+GCPs were identified in both species, while human GC progenitors were featured by high expression levels of MGP and NPY (Supplementary Fig. S2h). These genes were almost undetectable in mouse GC progenitors15,19, which could be validated in another dataset26. Such cross-species differences are unlikely to be attributed to the data quality, given that the extremely high expression levels of MGP and NPY were detected in human. Furthermore, though HEY1 was suggested as a critical regulator of GC lineage development27, different expression patterns of HEY1 could be observed between human and mouse. Hey1 was almost co-expressed with Atoh1 in mouse, while it was confined in a small fraction of human ATOH1+ GC progenitors (Supplementary Fig. S2a, f). Whether these differences in progenitors account for the different sizes of human and mouse cerebellums remains to be determined. Since the cerebellar development continue to postnatal 14 months in human and postnatal 20 days in mouse, more detailed temporal analyses covering the neonatal stages are needed to fully reveal the developmental landscapes in different species.
Cross-species differences were also observed concerning the region-specific gene expression patterns. Previous works suggested that several genes might be involved in spatial organization among cells in the cerebellum14. Here we further characterized the distinct gene cohorts in human and mouse GC lineages. We speculate that similar systems comprising various gene modules participate in spatial organization of GCs in both species, but the genes involved are not identical. Different gene modules influence cell distribution in different dimensions. Due to the intersection of these gene modules, unsupervised classification results may not represent genuine biological ontologies. More spatial data covering different species and other experiments are needed to validate this hypothesis. It remains unclear how GC progenitors obtain their spatial identity and whether these identities are intrinsically determined or changed during the tangential migration. Previous studies have shown a transient alternative distribution of Ebf2 in mouse GC progenitors, mimicking the distribution of Zebrin II negative PKCs59. Since we also found the similar patten along A-P axis between BARHL1-TLX3 in GCs and POSTN-IGSF21 in PKCs, the potential cross-talk between GCs and PKCs remains intriguing to be investigated. Furthermore, it was reported that the lobes in the posterior cerebellar hemisphere, which tend to be activated by cognitive tasks, occupy larger size in human than in other non-human primates and rodents66,67. Since PARM1 was found to be an oncogene which can promote proliferation68,69, the possibility that PARM1 with reverse expression pattern promotes the enlargement of posterior cerebellar hemisphere in humans warrants further investigation.
Besides the gene expression patterns revealed by scRNA-seq, we also characterized the GRNs involved in the diversification of GCs along A-P axis based on the integration of scRNA-seq and scATAC-seq. However, it should be noticed that the sequencing depth achieved in our study are still insufficient to fully characterize the regional heterogeneity of GCs. Integration of large-scale Smart-seq data and scATAC-seq data would provide more insights into the diversification of GC lineages. During the lineage development analysis of the RL-derived neurons, we did not identify the Lhx9+Pou3f2+ interposed nucleus neurons or Tbr1+ fastigial nucleus neurons70,71. It may result from the bias during the sampling. Additionally, how the GC progenitors are generated from the RL remains uncertain. VZ-derived PKCs were generally characterized into three subtypes and the multidimensional features were described. Though PKCs in mouse could be grouped into four subtypes26 (Fig. 5g), more molecular subtypes could be classified (Supplementary Fig. S5i). Whether these subtypes represent functionally or spatially distinct PKCs or the consequence of over-clustering requires experimental validation.
In summary, our study generated a large-scale multi-omics atlas of the developing human fetal cerebellum. This comprehensive atlas sheds light on the molecular mechanisms underlying early human cerebellar development and evolution. We foresee that the key regulators identified in this study will be leveraged in vitro to generate desired cerebellar cells for future clinical applications. This dataset will also be important to enhance our understanding of the linkage between molecular variation and cell types in neurodevelopmental disorders.
Materials and methods
Human sample collection
The human clinical tissues of pregnancies at GW13–18 were obtained upon therapeutic termination of pregnancy at Obstetrics and Gynecology Hospital of Fudan University. An informed consent document was signed by the patient before collection of the human sample. The whole experiment was examined by Ethics Committee of Obstetrics and Gynecology Hospital of Fudan University (2020-157).
Brain tissue dissection and cell dissociation
The human cerebellar tissues were dissected in ice-cold normal saline under the dissection microscope. The cerebellums from all samples were collected. Half of the human cerebellum was stored in liquid nitrogen for scATAC-seq. Each tissue for scRNA-seq was dissociated in 500 μL dissociation agent (400 U/mL DNaseI on hibernate E buffer, 10 U/mL papain) at 37 °C on a thermocycler for 15 min. Dissociation was terminated by 500 μL of 10% FBS in Hibernate E buffer. Cells were centrifuged at 4 °C for 5–10 min. After removal of the supernatant, 1 mL HA buffer was added and cells were blown up with straws repetitively to generate monoplast suspension.
scRNA-seq
Cell preparation
For the quality check and counting of single cell suspension, the cell survival rate was generally > 80%. The cells that passed the test were washed and resuspended to a suitable cell concentration of 700–1200 cells/μL for 10X Genomics ChromiumTM. The system was operated on the machine.
Gel bead in emulsion (GEM) creation and thermal cycling
GEMs were constructed for single-cell separation according to the number of cells to be harvested. After GEMs were normally formed, GEMs were collected for reverse transcription in a PCR machine for labeling.
Post cycling cleanup and cDNA amplification
The GEMs were oil-treated, and the amplified cDNA was purified by magnetic beads, and then subjected to cDNA amplification and quality inspection.
Library preparation and quantification
The 3ʹ gene expression library was constructed with the quality-qualified cDNA. After fragmentation, adaptor ligation, sample index PCR, etc., the library was finally quantitatively examined.
Sequencing
Cells were loaded onto the 10X Chromium Single Cell Platform (10X Genomics) at a concentration of 1000 cells/μL (Single Cell 3′ library and Gel Bead Kit v3) as described in the manufacturer’s protocol. Generation of GEMs, barcoding, GEM-RT cleanup, complementary DNA amplification and library construction were all performed as per the manufacturer’s protocol. Qubit was used for library quantification before pooling. The final library pool was sequenced on the Illumina Nova6000 instrument using 150-bp paired-end reads.
Single-cell gene expression quantification
After single-cell sequencing, CellRanger (v7.0.0) was applied to generate the fastq format data, and perform the quality control and read counting of genes with default parameters. Human GRCh38 (hg38) reference genome was chosen to perform the genomic alignment of reads. Seurat (v4.3.0) was used to analyze the gene-cell data matrix. MiQC (v1.1.3)72 was adopted to remove low-quality cells (posterior.cutoff = 0.95, model.slot = “flexmix_model”, model.type = “spline”). The remaining low-quality cells or potential doublets were directly discarded (mitochondrial genes > 20% or features < 800 or features > 6500). DoubletFinder (v2.0.3)73 was then applied and predicted doublets were further removed (parameters were based on cell counts of each data). Hemoglobin genes were removed. Standard Seurat pipeline was applied to each dataset (VariableFeatures = 3000, resolution = 1, PC depended). COSG (v0.9.0) and the function FindMarkers in Seurat, was used to calculate marker genes for each cluster. Cluster identities were manually labeled based on reported marker genes. Besides the neuronal lineages, microglia (CX3CR1+TREM2+), macrophage (FOLR2+VISG4+) and endothelial cells (CLDN5+ITM2A+) were identified. Detailed parameters are provided in our code, which is available on GitHub (https://github.com/NeuroXplorer-XuLab/Multi-omics-human-cerebellum).
Integration of human scRNA-seq data
Non-neural lineage cells were discarded from each dataset before the integration. FindIntegrationAnchors and IntegrateData functions from Seurat were applied to the entire integration process. For the integration of all neural lineages, supervised integration of scRNA-seq data was applied based on major markers selected from each cell type. For the analysis of certain neural lineage, cells were first selected and separated based on time point. Unsupervised integration was then applied. Analysis of interneurons did not separate the datasets because of the low cell counts. Due to the low feature counts in UBC, lower number of anchor features was used.
Mouse scRNA-seq analysis
GEO datasets were acquired from the GEO website (GSE11806819) and are publicly accessible at https://www.ncbi.nlm.nih.gov/geo. We mainly utilized data from E10, E12, E14, E16, E18, P0, P5 and P7. Other datasets of E13–18 were collected from PRJEB2305115 in European Nucleotide Archive. Cells expressing < 800 genes were removed. Standard Seurat pipeline was then applied to each data. Similar unsupervised integration was applied. Pearson correlation coefficient was calculated within each integrated data by cor.test provide by R (v4.2.2). Mouse excitatory lineage cells, as well as GABAergic neurons, from each time point were selected and integrated. Rtn1 was used as the marker of neuronal lineages as well as astrocytes. Different kinds of excitatory neurons were dissected by time periods and then integrated within each lineage for further analysis as shown in Supplementary figures. Integration local inversed Simpson’s index (LISI) was calculated via R package lisi (v1.0)74.
Cell cycle analysis
The Seurat package provided cell cycle-related genes, comprising 43 genes associated with the S phase and 54 genes linked to the G2/M phase (Supplementary Table S7), which were utilized to determine each cell’s stage by executing the CellCycleScoring function within Seurat.
GO enrichment, KEGG pathway, and GSEA analyses
The clusterProfiler package (v4.7.1)75 was employed to discern enriched GO terms and KEGG pathways associated with marker genes. Marker genes with a P-value < 0.05 from prior studies were selected and the enrichGO and enrichKEGG functions were utilized (ont = “BP”, qvalueCutoff = 0.05, pAjustMethod = “BH”). Moreover, the gseGO and gseKEGG functions were implemented to further investigate potential GO and KEGG pathways, using genes calculated via the FindMarker function in Seurat (logfc.threshold = 0 and min.diff.pct = 0).
Predict stemness with CytoTRACE and StemSC
Count data extracted from each dataset was input into the CytoTRACE (v0.3.3) and StemSC (v1.0.1). Subsequently, CytoTRACE and StemSC scores were employed to visualize the stemness of each cell type and to facilitate the construction of pseudotime trajectory in Monocle3.
Trajectory analysis
Splicing-specific count data were computed using Velocyto (v0.17.17)76 for RNA velocity analysis, employing default parameters. The generated loom files were further analyzed with scVelo (v0.2.3)77. Gene selection, normalization and moment estimation as well as RNA velocity estimation were performed (min_shared_counts=30, n_top_genes=2000 and n_pcs=30, n_neighbors=30). RNA velocities were visualized on UMAP using the stream embedding function. Monocle3 was also utilized to infer pseudotime trajectories for the major cell types. A Monocle cell dataset was constructed using data calculated by Seurat.
Nucleus isolation from frozen brain tissue
Frozen human tissue was cut into small pieces and ground in 2 mL of ice-cold homogenization buffer (20 mM Tris, pH 8.0 (Thermo Fisher Scientific), 500 mM sucrose (Sigma), 50 mM KCl (Thermo Fisher Scientific), 10 mM MgCl2 (Thermo Fisher Scientific), 0.1% NP-40 (Roche), 1× protease inhibitor cocktail (Roche), and 1% nuclease-free BSA, and 0.1 mM DTT). To release the nuclei, tissues were homogenized by strokes. A total of 30 μm cell strainer was used to filter nuclei into a 15 mL centrifuge tube. After centrifugation for 5 min at 4 °C, nuclei were obtained and washed twice with 1 mL of ice-cold blocking buffer (1× PBS supplemented with 1% BSA). After another centrifugation, the nuclei were collected in 50 μL of 1× PBS containing 1% BSA and counted with DAPI.
scATAC-seq library preparation and sequencing
We used DNBelab C Series Single-Cell ATAC Library Prep Set (MGI, #1000021878), to prepare scATAC-seq libraries. The transposed single-nucleus suspensions were converted to barcoded scATAC-seq libraries. After procedures including droplet encapsulation, pre-amplification, emulsion breakage, capture bead collection, DNA amplification and purification, indexed sequencing libraries were prepared according to the user guide. We measured the concentrations of sequencing library with Qubit ssDNA Assay Kit (Thermo Fisher Scientific). The library was sequenced using a paired-end 50 sequencing scheme by the BGISEQ-500 platform at China National GeneBank78.
scATAC-seq data processing and construction of GRNs
Raw sequencing reads from the BGISEQ-500 sequencer were filtered, demultiplexed using PISA, and aligned to the hg19 human genome. Fragmented data was further processed with ArchR (v2.0.1)31. Cells with transcription start site enrichment scores below 4 and fragment numbers less than 1000 were removed. Doublet analysis was performed using the addDoubletScores and filterDoublets functions in ArchR. “Iterative LSI” was then executed, utilizing major pc2-pc30 to cluster with Seurat’s FindClusters function at a resolution of 0.1. Gene activity scores were employed to identify distinct cell types based on various marker genes. Clusters were validated using several well-established marker genes as previously described. A total of 227,532 peaks were identified via peak calling using MACS2 (v2.1.1), and peaks were linked to genes by the addPeak2GeneLinks function.
The construction and analysis of GRNs in our study involved a three-step process: (1) examination of snATAC-seq and snRNA-seq datasets; (2) integration of these datasets; and (3) inference of cis-regulatory interactions to define a transcription factor-gene GRN (TF-gene GRN). In terms of single-cell transcriptomic and epigenomic data integration, we employed the “addGeneIntegrationMatrix” function to incorporate the gene expression matrix of snRNA-seq data onto the “geneScoreMatrix” of snATAC-seq data in both wild-type and mutant mice.
The method utilized for the subsequent definition of TF-gene GRN, involving the inference of cis-regulatory interactions, is complex and a separate manuscript is currently being prepared in the laboratory of Dr. Qiang Tu. The insights gained from GRNs of our study serve as preliminary findings and may provide potential directions for future hypothesis generation rather than as definitive conclusions. GRN results were visualized by Cytoscape79.
Identification of marker genes among clusters in scATAC-seq
We use getMarkerFeatures function (useMatrix = “GeneScoreMatrix”, bias = c (“TSSEnrichment”, “log10 (nFrags)”, testMethod = “Wilcoxon”)) and FDR ≤ 0.01 and |log2FC | ≥ 1 to find marker genes among clusters in scATAC-seq.
Stereo-seq library preparation and sequencing
Stereo-seq library preparation and sequencing were adapted according to the standard protocol V1.1 with minor modifications80. Tissue sections were adhered to the Stereo-seq chip, and incubated in –20 °C methanol for 30 min fixation, followed by nucleic acid dye staining (Thermo Fisher, Q10212) and imaging (Ti-7 Nikon Eclipse microscope). For permeabilization, tissue sections were permeabilized at 37 °C for 5 min. The cDNA was purified using AMPure XP beads (Vazyme, N411-03). The indexed scRNA-seq libraries were constructed according to the manufacturer’s protocol. The sequencing libraries were quantified by Qubit ssDNA Assay Kit (Thermo Fisher Scientific, Q10212). DNBs were loaded into the patterned Nano arrays and sequenced on MGI DNBSEQ-Tx sequencer (50 bp for read 1, 100 bp for read 2).
Raw Stereo-seq data processing
Fastq files were produced using an MGI DNBSEQ-Tx sequencer, with read 1 containing CID (1–25 bp) and MID (26–35 bp), while read 2 comprised the cDNA sequences. CID sequences on the first reads were initially mapped to the designed coordinates of the in situ captured chip obtained from the first round of sequencing, permitting a single base mismatch to account for sequencing and PCR errors. Reads with MIDs containing either N bases or over two bases with a quality score below 10 were discarded. CID and MID associated with each read were appended to the respective read headers. Retained reads were then aligned to the reference genome (hg19) using STAR81, and mapped reads with MAPQ > 10 were counted and annotated to their corresponding genes. UMIs with the same CID and gene locus were collapsed, allowing a single mismatch to correct for sequencing and PCR errors. Ultimately, this information was employed to generate a CID-containing expression profile matrix. The entire process was integrated into a publicly available pipeline named SAW, accessible at https://github.com/BGIResearch/SAW.
Stereo-seq data analysis and clustering
X-Y coordinates and overall MID counts were acquired using the st.io.read_gef function with the parameter bin_size=50 in Stereopy (https://github.com/BGIResearch/stereopy), simultaneously preparing the data for subsequent analyses. Filter_cells function was applied to exclude low-quality cells with parameters min_gene=50 and min_n_genes_by_counts=3. The Hotspot package was employed for the identification of informative genes and gene modules. Standard Seurat pipeline was used for further data cleaning and analysis.
Data availability
The sequencing data have been uploaded to China National GeneBank DataBase (https://db.cngb.org/) with human fetal cerebellum data in project CNP0005078 and CNP0002781.
References
Haldipur, P. et al. Spatiotemporal expansion of primary progenitor zones in the developing human cerebellum. Science 366, 454–460 (2019).
Koziol, L. F. et al. Consensus paper: the cerebellum’s role in movement and cognition. Cerebellum 13, 151–177 (2014).
Wagner, M. J., Kim, T. H., Savall, J., Schnitzer, M. J. & Luo, L. Cerebellar granule cells encode the expectation of reward. Nature 544, 96–100 (2017).
Schmahmann, J. D. The cerebellum and cognition. Neurosci. Lett. 688, 62–75 (2019).
Van Overwalle, F. et al. Consensus paper: cerebellum and social cognition. Cerebellum 19, 833–868 (2020).
Roussel, M. F. & Hatten, M. E. Cerebellum development and medulloblastoma. Curr. Top. Dev. Biol. 94, 235–282 (2011).
Stoodley, C. J. The cerebellum and neurodevelopmental disorders. Cerebellum 15, 34–37 (2016).
Hibi, M., Matsuda, K., Takeuchi, M., Shimizu, T. & Murakami, Y. Evolutionary mechanisms that generate morphology and neural-circuit diversity of the cerebellum. Dev. Growth Differ. 59, 228–243 (2017).
Hibi, M. & Shimizu, T. Development of the cerebellum and cerebellar neural circuits. Dev. Neurobiol. 72, 282–301 (2012).
Sotelo, C. Molecular layer interneurons of the cerebellum: developmental and morphological aspects. Cerebellum 14, 534–556 (2015).
Haldipur, P. & Millen, K. J. What cerebellar malformations tell us about cerebellar development. Neurosci. Lett. 688, 14–25 (2019).
Iulianella, A., Wingate, R. J., Moens, C. B. & Capaldo, E. The generation of granule cells during the development and evolution of the cerebellum. Dev. Dyn. 248, 506–513 (2019).
Mugnaini, E., Sekerkova, G. & Martina, M. The unipolar brush cell: a remarkable neuron finally receiving deserved attention. Brain Res. Rev. 66, 220–245 (2011).
Consalez, G. G., Goldowitz, D., Casoni, F. & Hawkes, R. Origins, development, and compartmentation of the granule cells of the cerebellum. Front. Neural Circuits 14, 611841 (2020).
Carter, R. A. et al. A single-cell transcriptional atlas of the developing murine cerebellum. Curr. Biol. 28, 2910–2920.e12 (2018).
Saunders, A. et al. Molecular diversity and specializations among the cells of the adult mouse brain. Cell 174, 1015–1030.e16 (2018).
Peng, J. et al. Single-cell transcriptomes reveal molecular specializations of neuronal cell types in the developing cerebellum. J. Mol. Cell Biol. 11, 636–648 (2019).
Rodriques, S. G. et al. Slide-seq: A scalable technology for measuring genome-wide expression at high spatial resolution. Science 363, 1463–1467 (2019).
Vladoiu, M. C. et al. Childhood cerebellar tumours mirror conserved fetal transcriptional programs. Nature 572, 67–73 (2019).
Wizeman, J. W., Guo, Q., Wilion, E. M. & Li, J. Y. Specification of diverse cell types during early neurogenesis of the mouse cerebellum. Elife 8, e42388 (2019).
Kozareva, V. et al. A transcriptomic atlas of mouse cerebellar cortex comprehensively defines cell types. Nature 598, 214–219 (2021).
Luo, W. et al. Single-cell spatial transcriptomic analysis reveals common and divergent features of developing postnatal granule cerebellar cells and medulloblastoma. BMC Biol. 19, 135 (2021).
Sarropoulos, I. et al. Developmental and evolutionary dynamics of cis-regulatory elements in mouse cerebellar cells. Science 373, eabg4696 (2021).
Khouri-Farah, N., Guo, Q., Morgan, K., Shin, J. & Li, J. Y. H. Integrated single-cell transcriptomic and epigenetic study of cell state transition and lineage commitment in embryonic mouse cerebellum. Sci. Adv. 8, eabl9156 (2022).
Aldinger, K. A. et al. Spatial and cell type transcriptional landscape of human cerebellar development. Nat. Neurosci. 24, 1163–1175 (2021).
Sepp, M. et al. Cellular development and evolution of the mammalian cerebellum. Nature 625, 788–796 (2024).
Zhong, S. et al. Single-cell epigenomics and spatiotemporal transcriptomics reveal human cerebellar development. Nat. Commun. 14, 7613 (2023).
Stuart, T. et al. Comprehensive integration of single-cell data. Cell 177, 1888–1902.e21 (2019).
Dai, M., Pei, X. & Wang, X. J. Accurate and fast cell marker gene identification with COSG. Brief. Bioinform. 23, bbab579 (2022).
ten Donkelaar, H. J., Lammens, M., Wesseling, P., Thijssen, H. O. & Renier, W. O. Development and developmental disorders of the human cerebellum. J. Neurol. 250, 1025–1036 (2003).
Granja, J. M. et al. ArchR is a scalable software package for integrative single-cell chromatin accessibility analysis. Nat. Genet. 53, 403–411 (2021).
Erö, C., Gewaltig, M.-O., Keller, D. & Markram, H. A cell atlas for the mouse brain. Front. Neuroinform. 12, https://doi.org/10.3389/fninf.2018.00084 (2018).
Trapnell, C. et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat. Biotechnol. 32, 381–386 (2014).
Gulati, G. S. et al. Single-cell transcriptional diversity is a hallmark of developmental potential. Science 367, 405–411 (2020).
Zheng, H. et al. StemSC: a cross-dataset human stemness index for single-cell samples. Stem Cell Res. Ther. 13, 115 (2022).
Miyashita, S. & Hoshino, M. Transit amplifying progenitors in the cerebellum: similarities to and differences from transit amplifying cells in other brain regions and between species. Cells 11, 726 (2022).
Leifer, D., Golden, J. & Kowall, N. W. Myocyte-specific enhancer binding factor 2C expression in human brain development. Neuroscience 63, 1067–1079 (1994).
Corrado, L. et al. A novel peripherin gene (PRPH) mutation identified in one sporadic amyotrophic lateral sclerosis patient. Neurobiol. Aging 32, 552.e1–6 (2011).
Janz, R., Goda, Y., Geppert, M., Missler, M. & Südhof, T. C. SV2A and SV2B function as redundant Ca2+ regulators in neurotransmitter release. Neuron 24, 1003–1016 (1999).
Chang, S. et al. Complexin stabilizes newly primed synaptic vesicles and prevents their premature fusion at the mouse calyx of held synapse. J. Neurosci. 35, 8272–8290 (2015).
Szklarczyk, D. et al. The STRING database in 2023: protein–protein association networks and functional enrichment analyses for any sequenced genome of interest. Nucleic Acids Res. 51, D638–D646 (2023).
Shirasawa, S. et al. Rnx deficiency results in congenital central hypoventilation. Nat. Genet. 24, 287–290 (2000).
Lein, E. S. et al. Genome-wide atlas of gene expression in the adult mouse brain. Nature 445, 168–176 (2006).
Cheng, Y. et al. The Engrailed homeobox genes determine the different foliation patterns in the vermis and hemispheres of the mammalian cerebellum. Development 137, 519–529 (2010).
Köhler, S. et al. The human phenotype ontology in 2021. Nucleic Acids Res. 49, D1207–D1217 (2021).
Yeung, J., Larouche, M., Ramirez, M., Robert, R. & Goldowitz, D. Handbook of the Cerebellum and Cerebellar Disorders (eds Mario Manto et al.). 1–19 (Springer International Publishing, 2019).
Yeung, J. et al. Wls provides a new compartmental view of the rhombic lip in mouse cerebellar development. J. Neurosci. 34, 12527–12537 (2014).
Kim, J. A., Sekerkova, G., Mugnaini, E. & Martina, M. Electrophysiological, morphological, and topological properties of two histochemically distinct subpopulations of cerebellar unipolar brush cells. Cerebellum 11, 1012–1025 (2012).
Balmer, T. S. & Trussell, L. O. Selective targeting of unipolar brush cell subtypes by cerebellar mossy fibers. Elife 8, e44964 (2019).
Sekerkova, G., Watanabe, M., Martina, M. & Mugnaini, E. Differential distribution of phospholipase C beta isoforms and diaglycerol kinase-beta in rodents cerebella corroborates the division of unipolar brush cells into two major subtypes. Brain Struct. Funct. 219, 719–749 (2014).
Frederico, B. et al. DNGR-1-tracing marks an ependymal cell subset with damage-responsive neural stem cell potential. Dev. Cell 57, 1957–1975.e9 (2022).
Wang, S.-Z. et al. An oligodendrocyte-specific zinc-finger transcription regulator cooperates with Olig2 to promote oligodendrocyte differentiation. Development 133, 3389–3398 (2006).
Kent, W. J. et al. The human genome browser at UCSC. Genome Res. 12, 996–1006 (2002).
Kwan, A. Y. J. Investigating molecular signatures of GABAergic neuron populations in the developing cerebellum. University of Toronto (Canada) https://hdl.handle.net/1807/108912 (2021).
Apsley, E. J. & Becker, E. B. E. Purkinje cell patterning-insights from single-cell sequencing. Cells 11, 2918 (2022).
Sarna, J. R. & Hawkes, R. Patterned Purkinje cell death in the cerebellum. Prog. Neurobiol. 70, 473–507 (2003).
Chung, S. H., Marzban, H., Croci, L., Consalez, G. G. & Hawkes, R. Purkinje cell subtype specification in the cerebellar cortex: early B-cell factor 2 acts to repress the zebrin II-positive Purkinje cell phenotype. Neuroscience 153, 721–732 (2008).
Croci, L. et al. Local insulin-like growth factor I expression is essential for Purkinje neuron survival at birth. Cell Death Differ. 18, 48–59 (2011).
Badaloni, A. et al. Dynamic expression and new functions of early B cell factor 2 in cerebellar development. Cerebellum 18, 999–1010 (2019).
Matsunaga, E. et al. Periostin, a neurite outgrowth-promoting factor, is expressed at high levels in the primate cerebral cortex. Dev. Growth Differ. 57, 200–208 (2015).
Tanabe, Y. et al. IgSF21 promotes differentiation of inhibitory synapses via binding to neurexin2α. Nat. Commun. 8, 408 (2017).
Aghajanian, H. et al. Semaphorin 3d and semaphorin 3e direct endothelial motility through distinct molecular signaling pathways. J. Biol. Chem. 289, 17971–17979 (2014).
Fujita, E., Tanabe, Y., Imhof, B. A., Momoi, M. Y. & Momoi, T. A complex of synaptic adhesion molecule CADM1, a molecule related to autism spectrum disorder, with MUPP1 in the cerebellum. J. Neurochem. 123, 886–894 (2012).
Truvé, K. et al. Identification of candidate genetic variants and altered protein expression in neural stem and mature neural cells support altered microtubule function to be an essential component in bipolar disorder. Transl. Psychiatry 10, 390 (2020).
Croci, L. et al. A key role for the HLH transcription factor EBF2COE2,O/E-3 in Purkinje neuron migration and cerebellar cortical topography. Development 133, 2719–2729 (2006).
Luo, Y. et al. Lobular homology in cerebellar hemispheres of humans, non-human primates and rodents: a structural, axonal tracing and molecular expression analysis. Brain Struct. Funct. 222, 2449–2472 (2017).
Stoodley, C. J. & Schmahmann, J. D. Functional topography of the human cerebellum. Handb. Clin. Neurol. 154, 59–70 (2018).
Fladeby, C. et al. Human PARM-1 is a novel mucin-like, androgen-regulated gene exhibiting proliferative effects in prostate cancer cells. Int. J. Cancer 122, 1229–1235 (2007).
Charfi, C., Levros, L.-C., Edouard, E. & Rassart, E. Characterization and identification of PARM-1 as a new potential oncogene. Mol. Cancer 12, 84 (2013).
Wu, J. P. H. et al. The transcription factor Pou3f1 sheds light on the development and molecular diversity of glutamatergic cerebellar nuclear neurons in the mouse. Front. Mol. Neurosci. 15, 921901 (2022).
Fink, A. J. et al. Development of the deep cerebellar nuclei: transcription factors and cell migration from the rhombic lip. J. Neurosci. 26, 3066–3076 (2006).
Hippen, A. A. et al. miQC: An adaptive probabilistic framework for quality control of single-cell RNA-sequencing data. PLoS Comput. Biol. 17, e1009290 (2021).
McGinnis, C. S., Murrow, L. M. & Gartner, Z. J. DoubletFinder: doublet detection in single-cell RNA sequencing data using artificial nearest neighbors. Cell Syst. 8, 329–337.e4 (2019).
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods 16, 1289–1296 (2019).
Wu, T. et al. clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innovation 2, 100141 (2021).
La Manno, G. et al. RNA velocity of single cells. Nature 560, 494–498 (2018).
Bergen, V., Lange, M., Peidli, S., Wolf, F. A. & Theis, F. J. Generalizing RNA velocity to transient cell states through dynamical modeling. Nat. Biotechnol. 38, 1408–1414 (2020).
Huang, J. et al. A reference human genome dataset of the BGISEQ-500 sequencer. Gigascience 6, 1–9 (2017).
Shannon, P. et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 13, 2498–2504 (2003).
Chen, A. et al. Single-cell spatial transcriptome reveals cell-type organization in the macaque cortex. Cell 186, 3726–3743.e24 (2023).
Dobin, A. & Gingeras, T. R. Mapping RNA-seq reads with STAR. Curr. Protoc. Bioinf. 51, 11.14.11–11.14.19 (2015).
Acknowledgements
This study was supported by the National Natural Science Foundation of China (32330038, 32271024, 32394030, 31921002, 32061143026), the Ministry of Science and Technology of China (STI2030-Major Projects 2021ZD0202300), and Chinese Academy of Sciences (YJKYYQ20200052).
Author information
Authors and Affiliations
Contributions
Z.X., W.Z., Y.F.L., M.X. and X.D. designed the overall project. Z.X. and D.M. supervised the project. For sample preparation, M.X., M.Y. and Y.X. collected samples; M.X., Z.Z. and K.L. managed samples. For scRNA-seq, D.Z. and X.D. managed the sequencing data; F.Y., Z.Z. and D.Z. analyzed the data. For scATAC-seq, T.P. and C.L. prepared the sequencing; Y.J.L. and Q.T. established the method for constructing GRNs; F.Y. and D.Z. analyzed the data. For spatial transcriptome, F.Y. and Z.Z. prepared the slices; T.P., C.L. and Y.D. prepared the sequencing; F.Y., Z.Z., D.Z. and Y.W. analyzed the data. F.Y. organized figures. F.Y., Z.Z. and D.Z. wrote the manuscript. Z.X. and D.M. edited the manuscript.
Corresponding authors
Ethics declarations
Conflict of interest
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Yang, F., Zhao, Z., Zhang, D. et al. Single-cell multi-omics analysis of lineage development and spatial organization in the human fetal cerebellum. Cell Discov 10, 22 (2024). https://doi.org/10.1038/s41421-024-00656-1
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41421-024-00656-1
