Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2021 Oct;598(7879):214-219.
doi: 10.1038/s41586-021-03220-z. Epub 2021 Oct 6.

A transcriptomic atlas of mouse cerebellar cortex comprehensively defines cell types

Velina Kozareva #  1 , Caroline Martin #  1 , Tomas Osorno  2 , Stephanie Rudolph  2 , Chong Guo  2 , Charles Vanderburg  1 , Naeem Nadaf  1 , Aviv Regev  1 , Wade G Regehr  2 , Evan Macosko  3   4
Affiliations

A transcriptomic atlas of mouse cerebellar cortex comprehensively defines cell types

Velina Kozareva et al. Nature. 2021 Oct.

Erratum in

Abstract

The cerebellar cortex is a well-studied brain structure with diverse roles in motor learning, coordination, cognition and autonomic regulation. However, a complete inventory of cerebellar cell types is currently lacking. Here, using recent advances in high-throughput transcriptional profiling1-3, we molecularly define cell types across individual lobules of the adult mouse cerebellum. Purkinje neurons showed considerable regional specialization, with the greatest diversity occurring in the posterior lobules. For several types of cerebellar interneuron, the molecular variation within each type was more continuous, rather than discrete. In particular, for the unipolar brush cells-an interneuron population previously subdivided into discrete populations-the continuous variation in gene expression was associated with a graded continuum of electrophysiological properties. Notably, we found that molecular layer interneurons were composed of two molecularly and functionally distinct types. Both types show a continuum of morphological variation through the thickness of the molecular layer, but electrophysiological recordings revealed marked differences between the two types in spontaneous firing, excitability and electrical coupling. Together, these findings provide a comprehensive cellular atlas of the cerebellar cortex, and outline a methodological and conceptual framework for the integration of molecular, morphological and physiological ontologies for defining brain cell types.

PubMed Disclaimer

Conflict of interest statement

A.R. is a founder and equity holder of Celsius Therapeutics, an equity holder in Immunitas Therapeutics and until 31 August 2020, was an SAB member of Syros Pharmaceuticals, Neogene Therapeutics, Asimov and ThermoFisher Scientific. From 1 August 2020, A.R. has been an employee of Genentech. For E.Z.M., the research herein was not done in his capacity as an MGH employee.

Figures

Fig. 1
Fig. 1. Comprehensive transcriptional profiling of cell types across the mouse cerebellum.
a, Experimental design with lobe-based sampling and profiling. AN, ansiform lobule; COP, copula pyramidis; CUL, culmen; F, flocculus; PF, paraflocculus; PRM, paramedian lobule; SIM, simple lobule. b, UMAP visualization of 611,034 nuclei (after profile quality control and annotation; Methods), coloured by cell type identity. ODC, oligodendrocyte; OPC, oligodendrocyte precursor cell. c, Dendrogram indicating hierarchical relationships between cell subtypes (left) with a paired dot plot (right) of scaled expression (exp.) of selected marker genes for cell type identity. Background text colours correspond to cell types in b. d, Allen Brain Atlas expression staining for selected gene markers of canonical cell populations, indicating cerebellar layer localization. GCL, granule cell layer; ML, molecular layer; PCL, Purkinje cell layer.
Fig. 2
Fig. 2. Characterization of spatial variation and patterning in neuronal and glial cell types.
a, Plot indicating neuronal and glial clusters that have lobule enrichment patterns that are significantly different from the cell type population as a whole (Pearson’s chi-squared, FDR < 0.001 indicated by dashed line). x-axis shows −log10-transformed q values; y axis shows log2-transformed maximum lobule enrichment across all 16 lobules (Methods). Clusters with high correlation (Pearson correlation coefficient > 0.85) in lobule enrichment values between replicate sets and maximum lobule enrichment > 2 are labelled (Methods, Extended Data Fig. 3a). b, Dot plot of scaled expression of selected gene markers for PC clusters. c, Regional enrichment plots indicating average lobule enrichment for aggregated Aldoc-positive PC subtypes (left) and aggregated Aldoc-negative subtypes (right). Regions labelled as in Fig. 1a. A-Aldoc, anti-Aldoc. d, Regional enrichment plots indicating lobule enrichment for PC clusters. e, Dot plot of scaled expression of selected gene markers for granule cell clusters. f, Regional enrichment plots indicating lobule enrichment for three spatially significant granule cell clusters. g, Dot plot of scaled expression of selected gene markers for Bergman glial clusters. h, Regional enrichment plot indicating lobule enrichment for the Mybpc1-positive Wif1-positive Bergmann glial cluster.
Fig. 3
Fig. 3. Cross-cluster continuity among select neuronal populations, including unipolar brush cells.
a, Gene expression ordered by pseudotime (Methods) for two of the top differentially expressed genes between two clusters within the MLI1 type (left, MLI1_1 and MLI1_2), and between two cell types (right, MLI1 and MLI2 cell types). Curves represent logistic fits estimated via nonlinear least squares; maximum slope values (m) are indicated. Differences in magnitude of m values correspond well with visually distinctive molecular continuity versus discreteness. b, Empirical cumulative distributions of log10(m) values (curve fit as in a) for top differentially expressed (DE) genes among aggregated combinations of MLI, Golgi cells and UBC clusters. c, t-distributed stochastic neighbour embedding (t-SNE) visualizations of UBCs (n = 1,613), coloured by pseudotime loading (top left), and log-normalized expression of canonical gene markers for all UBCs (Eomes), ON UBCs (Grm1, Plcb4), and OFF UBCs (Calb2, Plcb1). d, Pseudotime rank ordered gene expression for canonical markers associated with ON UBCs (Grm1, Plcb4) and OFF UBCs (Calb2, Plcb1). Cells are coloured by pseudotime loading. Curves indicate logistic fits estimated as in a; maximum slope values (m) are indicated. e, Left, cell-attached recordings of spiking responses and whole-cell recordings of currents evoked by brief glutamate puffs in ON (top), biphasic (middle) and OFF (bottom) UBCs. Right, heat map of the currents recorded in all UBCs, sorted by the magnitude and time course of charge transfer. f, Left, schematic describing whole-cell recordings obtained from UBCs evoked by pressure application of glutamate (left traces, black), the mGluR1 agonist DHPG (middle trace, red) and the mGluR2 agonist LY354740 (right trace, blue) using three pipettes placed within 20 μm of the recorded cell. Evoked currents from each application are summarized in the correspondingly coloured plots. Representative UBC recordings are shown in Extended Data Fig. 5.
Fig. 4
Fig. 4. Two molecularly and electrophysiologically distinct populations of MLIs.
a, Dendrogram of gene expression relationships (computed as in Fig. 1c) among MLI and PLI populations (left), paired with dot plot (right) of selected gene markers. b, Representative image of smFISH expression of Sorcs3 (purple) and Nxph1 (green) within the molecular layer (n = 16 slides sectioned from 3 mice). Scale bar, 20 μm. c, The percentages of Sorcs3+, Nxph1+ and Sorcs3+Nxph1+ cells in the inner third, middle third and distal third of the molecular layer (ML) (total cells counted shown; n = 16 slides sectioned from 3 mice). Data are mean ± s.d. d, UMAP visualization of cerebellar interneuron developmental trajectory, coloured by developmental age of collection (left) and by pseudotime (right) (Methods). e, UMAP visualization of Sorcs3, Nxph1 and Fos expression across MLI development. f, Four two-photon images of representative basket and stellate-like MLI1 and MLI2 neurons in cerebellar slice. Insets show the fluorescent fill of the cell body (red) with the smFISH signals superimposed (top), and the smFISH signal only (bottom; Sorcs3, purple; Nxph1, green). Scale bars, 30 μm (black) and 10 μm (white). n = 23 MLI1 cells, 20 MLI2 cells. g, Scatter plots of firing rate (left) and membrane resistance Rm (right) of molecularly identified MLI1 (purple, n = 23 cells) and MLI2 neurons (green, n = 20 cells), as well as MLIs whose molecular identity was not ascertained (grey, n = 55 cells). Corresponding distributions at the right. h, Left, Mean input–output curves of MLI1 (purple) and MLI2 neurons (green). Shaded area denotes s.e.m. n = 22 MLI1 cells, 20 MLI2 cells. ***P < 0.001, generalized linear mixed effect model (Supplementary Table 3). Right, representative traces of MLI1 (purple) and MLI2 (green) for 20, 40 and 60 pA current injections. i, t-SNE visualization of expression of Gjd2 across MLI and PLI cell types. j, Representative image of smFISH expression of Sorcs3 (purple), Nxph1 (green), and Gjd2 (red), in the molecular layer (n = 16 slides sectioned from 3 mice). Scale bar, 20 μm. k, The percentages of MLI1s and MLI2s expressing Gjd2 in the molecular layer (total cells counted shown; n = 16 slides sectioned from 3 mice). Data are mean ± s.d. l, The percentages of all cells (grey), MLI1s (purple) and ML2s (green) in which spikelets were observed. m, Example voltage clamp recordings in the presence of synaptic blockers show spikelets in MLI1 (purple), but no spikelets in MLI2 (green). Firing rates, Rm values, input–output curves and the numbers of cells with spikelets were all significantly different for MLI1 and MLI2 (Supplementary Table 3).
Extended Data Fig. 1
Extended Data Fig. 1. Summary and quality control analyses for nuclei sampling.
a, Bar graph showing the number of cells contributed by each individual per region across the dataset of 611,034 cerebellar nuclei (post-quality control, 6 total individuals, 16 regions). b, Violin plot showing distribution of log10(nUMI) across the regions profiled. Cell numbers: n = 23,364 for region I, n = 29,513 for region II, n = 24,344 for region III, n = 27,680 for region CUL, n = 80,166 for region VI, n = 35,622 for region VII, n = 21,299 for region VIII, n = 47,118 for region IX, n = 20,423 for region X, n = 38,227 for region AN1, n = 72,135 for region AN2, n = 58,142 for region PRM, n = 42,518 for region SIM, n = 14,816 for region COP, n = 20,836 for region F, n = 54,831 for region PF. Box plots centred on median and bounded by interquartile range. c, Violin plot of log10(nUMI) per profile across the 18 cell types identified. The relative median values here are consistent with known differences in cell size; for example, Purkinje cells have the highest median number of UMIs. Cell numbers: n = 16,717 for astrocytes, n = 17,498 for Bergmann glia, n = 591 for choroid, n = 5,333 for fibroblasts, n = 2,271 for mural cells, n = 6,142 for stalk cells, n = 243 for ependymal cells, n = 3,989 for Golgi cells, n = 477,176 for granule cells, n = 280 for macrophages, n = 1,296 for microglia, n = 32,716 for MLI1, n = 10,608 for MLI2, n = 13,363 for oligodendrocytes, n = 2,443 for PLIs, n = 2,121 for polydendrocytes, n = 16,634 for Purkinje cells, n = 1,613 for UBCs. Box plots centred on median and bounded by interquartile range. d, Bar graph of alignment scores (Methods) calculated across replicates for each lobule, after performing LIGER integration (across sex) (Methods) for each regional subset. Subsets sampled from the final set of 611,034 high-quality nuclei profiles. These analyses represent examples of expected replicate alignment when using the described pipeline. Note that lobule COP is excluded as it did not include representation from male and female replicates. e, Visualizations of representative cell type analyses, indicating high alignment across replicate sets (granule is UMAP, all others t-SNE). Replicate sets were designated as in Methods (cluster regional composition test and lobule enrichment). f, Plot indicating probability of sufficiently sampling 10 very rare populations (prevalence 0.15%) as a function of total number of cells profiled in experiment (probability estimated as in Methods). Number of high-quality nuclei profiled here (611,034) and corresponding probability are indicated.
Extended Data Fig. 2
Extended Data Fig. 2. Characterization and annotation of cerebellar subtypes.
a–k, Visualizations of all individual cell type analyses of Purkinje (a), granule (b), UBC (c), MLI or PLI (d), and Golgi (e) neurons, as well as Bergmann (f), astrocyte (g), OPC/ODC (h), endothelial (i), choroid (j), and macrocytic (k) glial populations, labelled by cluster designations. Granule is UMAP, all others t-SNE. l–n, Dot plots of scaled expression of selected marker genes for individual cell type analyses not displayed in the main figures: ODC and OPC (l), Golgi (m) and astrocyte (n).
Extended Data Fig. 3
Extended Data Fig. 3. Additional analyses of spatial variation in neuronal and glial subtypes.
a, Scatter plot of inter-replicate correlation (Pearson) for lobule enrichment scores calculated for replicate sets individually, across each cluster (clusters ordered by decreasing correlation). Two replicate sets were designated for each major cluster analysis by aggregating the individuals with the highest representation for each lobule into a single replicate (and similarly for the individuals with second highest representation). High inter-replicate correlation indicates consistent lobule enrichment for subtypes. r = 0.85 is indicated. b, Allen Brain Atlas expression staining for two selected PC markers representing clusters with their respective lobule enrichments indicated; Drd3 in the flocculus (Purkinje_Aldoc_2), and Gpr176 in lobule VI (Purkinje_Aldoc_1). C, coronal section; S, sagittal section. c, Allen Brain Atlas expression staining for three selected GC markers representing clusters with their respective lobule enrichments indicated; Rasgrf1 in the anterior lobules (Granule_1), Gprin3 in the posterior lobules (Granule_2), and Galntl6 in the nodulus (Granule_3). d, Left, lobule enrichment plot indicating enrichment of UBCs in posterior lobules of the cerebellum, particularly lobules IX (uvula) and X (nodulus). Right, Allen Brain Atlas expression staining for UBC marker Eomes, showing enrichment in lobules IX, X and VII. e, Allen Brain Atlas expression staining for Wif1 (a Bergmann_2 cluster marker), indicating expression enriched in lobules VI, X (nodulus), and IX (uvula).
Extended Data Fig. 4
Extended Data Fig. 4. Integrative analysis of human and mouse cerebellar interneuron profiles.
a, c, e, g, UMAP representation of the integrative analyses of UBC (1,613 mouse; 3,893 human) (a), MLI/PLI (45,555 mouse; 14,971 human) (c), Golgi (3,989 mouse; 1,059 human) (e), and granule (119,972 mouse; 130,335 human) (g) cells, coloured by species (top), or joint cluster (bottom, for MLI and PLI, Golgi, and granule only). b, d, f, Heat maps showing expression of selected genes in UBC (b), MLI and PLI (d), and Golgi (f). Profiles are segregated both by species and cluster. UBC profiles are ordered by iNMF factor loadings for factor corresponding to OFF UBCs. h, Left, dot plot showing expression of selected genes in granule clusters, within human (red) and mouse (blue). Right, proportional representation of lobule dissections across the granule clusters. Granule cluster numbers approximately correspond to the mouse-only clusters shown in Fig. 2e.
Extended Data Fig. 5
Extended Data Fig. 5. Recordings of UBCs.
Representative recordings of UBCs from Fig. 3f. In cells in which glutamate evoked primarily an inward current (UBC16), there was a very large mGluR1 component and a small mGluR2 component. The opposite was true for UBCs in which glutamate evoked primarily an outward current (UBC2). For intermediate cells such as UBC9 and UBC7, mGluR1 and mGluR2 components were both prominent.
Extended Data Fig. 6
Extended Data Fig. 6. Developmental trajectory analysis of MLI1 and MLI2 neurons.
a, UMAP visualizations showing expression of canonical marker genes of cerebellar interneuron development (Tfap2b, Ascl1, Neurog1, Neurog2 and Pax2), and of mature PLIs (Klhl1). b, Allen Developing Mouse Brain Atlas expression staining of Tfap2b across four developmental time points, showing that MLI progenitors begin to enter the molecular layer around P14. c, Interneuron developmental trajectory coloured by annotated clusters. d, Heat map showing the loading of Monocle3-defined gene modules across undifferentiated interneuron precursors, MLI1 and MLI2. Cells are ordered in each block by pseudotime rank. e, UMAP visualizations showing expression of selected top-loading genes from the indicated Monocle3-defined modules. f, Dot plot showing the transient, MLI2-specific expression of three activity-regulated genes (Fos, Junb and Fosl2) and two genes whose expression persists in adulthood (Fam135b and Sorcs2).
Extended Data Fig. 7
Extended Data Fig. 7. Additional imaging of MLI1 and MLI2 neurons.
a, Representative image of smFISH expression of Sorcs3 (purple) and Nxph1 (green) at low magnification. Scale bar, 100 μm. The distal, middle and inner molecular layer sublayer boundaries are designated with a dotted line; the thinner dashed line marks the outer boundary of cortical folds, and the thicker dashed line indicates the location of the Purkinje layer (n = 16 slides sectioned over 3 mice). b, MLIs were imaged and identified as in Fig. 4. Examples of MLI1s (left) and MLI2s (right) are shown for cells located in the distal third (top), the middle third (middle) and the inner third (bottom) of the molecular layer. n = 23 MLI1 cells and 20 MLI2 cells.
Extended Data Fig. 8
Extended Data Fig. 8. Further comparison of electrical properties of MLI1 and MLI2 neurons.
a–c, Measurements of: capacitance (a), the amplitude of currents through hyperpolarization and nucleotide-gated (HCN) channels, Ih, (b) and the action potential width (c) in recordings of subsequently identified MLI1 (purple) and MLI2 (green) neurons. The properties are plotted as a function of position within the molecular layer. Density plots summarize the properties for all cells (grey), MLI1 (purple) and MLI2 (green). a, Capacitance was determined by measuring the responses to a −10-mV voltage step, and integrating the capacitive current as shown (purple and green shaded area for MLI1 and MLI2, respectively). These traces from cells in the inner third of the molecular layer show the large difference in resistance (Rm = −10 mV ISS), in which ISS is the steady state current in response to the voltage step (black continuous line). Red dashed line denotes the baseline current. There was a significant difference in the capacitance of MLI1 and MLI2 neurons (P = 2.03 × 10−4). b, The amplitude of Ih was determined by measuring the responses to a −30-mV step, and evaluated as shown. Measured Ih was significantly larger for MLI2 (P = 4.24 × 10−6), and the difference was particularly notable for MLIs in the inner third of the molecular layer. c, The action potential width was measured as shown and there was no significant difference for MLI1 and MLI2 neurons. d, The presence or absence of spikelets is shown as a function of position in the molecular layer is summarized for all MLIs, MLI1 and MLI2, with ‘+’ indicating the presence of spikelets. e, Example recordings are shown for six MLI1 neurons (left, purple) and six MLI2 neurons (right, green). Number of cells and statistical tests are summarized in Supplementary Table 3.
Extended Data Fig. 9
Extended Data Fig. 9. Molecular variation in Golgi interneuron clusters.
a, UMAP visualization of expression of selected genes in the two Golgi interneuron clusters, showing variation in many of the same genes that define the MLI1/MLI2 distinction (Sorcs3, Gjd2 and Nxph1). b, Four representative Golgi cells from an smFISH experiment for Lgi2 (Golgi cell marker, green), Sst (Golgi_2 marker, purple), and Gjd2 (higher in Golgi_1, red). Scale bar, 50 μm. n = 77. c, Quantification of mean smFISH pixel intensity for the genes Sst and Gjd2, across 77 cells. The four representative cells in b are labelled with the corresponding Roman numeral. Spearman correlation (ρ) is shown.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Klein AM, et al. Droplet barcoding for single-cell transcriptomics applied to embryonic stem cells. Cell. 2015;161:1187–1201. doi: 10.1016/j.cell.2015.04.044. - DOI - PMC - PubMed
    1. Macosko EZ, et al. Highly parallel genome-wide expression profiling of individual cells using nanoliter droplets. Cell. 2015;161:1202–1214. doi: 10.1016/j.cell.2015.05.002. - DOI - PMC - PubMed
    1. Zheng GXY, et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 2017;8:14049. doi: 10.1038/ncomms14049. - DOI - PMC - PubMed
    1. Apps R, Hawkes R. Cerebellar cortical organization: a one-map hypothesis. Nat. Rev. Neurosci. 2009;10:670–681. doi: 10.1038/nrn2698. - DOI - PubMed
    1. Tasic B, et al. Shared and distinct transcriptomic cell types across neocortical areas. Nature. 2018;563:72–78. doi: 10.1038/s41586-018-0654-5. - DOI - PMC - PubMed

Publication types

MeSH terms

-