Cited by 261Single-cell and spatial transcriptomicsDevelopmental Biology and Gene RegulationCell Image Analysis Techniques

The DOI matches its Crossref record; 138 of the 138 references it cites were found in a registry or an index; 19 of the 112 sources its claims cite were checked for support.

A harmonized atlas of mouse spinal cord cell types and their spatial organization

Open access

Abstract

Single-cell RNA sequencing data can unveil the molecular diversity of cell types. Cell type atlases of the mouse spinal cord have been published in recent years but have not been integrated together. Here, we generate an atlas of spinal cell types based on single-cell transcriptomic data, unifying the available datasets into a common reference framework. We report a hierarchical structure of postnatal cell type relationships, with location providing the highest level of organization, then neurotransmitter status, family, and finally, dozens of refined populations. We validate a combinatorial marker code for each neuronal cell type and map their spatial distributions in the adult spinal cord. We also show complex lineage relationships among postnatal cell types. Additionally, we develop an open-source cell type classifier, SeqSeek, to facilitate the standardization of cell type identification. This work provides an integrated view of spinal cell types, their gene expression signatures, and their molecular organization.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

This paper harmonizes six published single-cell/nucleus RNA-sequencing datasets into a common reference atlas of postnatal mouse spinal cord cell types, identifying 16 major cell types and 69 neuronal populations. It validates combinatorial marker genes in adult tissue, infers embryonic lineage relationships, and provides a two-tiered classifier and web resource, SeqSeek.

Key findings

  • Integration of six studies yielded a final dataset of 52,623 cells/nuclei and 16 major cell types, including 69 neuronal clusters comprising 19,353 neuronal cells/nuclei.
  • Spatial location was the highest level of organization: dorsal and mid/ventral neurons separated before neurotransmitter status, and dorsal clusters were more robust and more readily identified in situ than ventral clusters.
  • In adult lumbar spinal cord, 71% of neurons (2057/2894) were confidently assigned to one of the 69 neuronal clusters, and an additional 9% (266/2894) to pairs of closely related clusters; 79 of 95 tested genes showed reliable expression.
  • Automated classification performed well: coarse cell-type accuracy was 89% for label transfer, 83% for a neural network, and 80% for an SVM; the two-tiered model classified 90% of nuclei from known independent clusters accurately and identified 28% of nuclei from unknown clusters.

Methods

  • Raw sequencing reads from the first six publicly available postnatal mouse spinal cord single-cell datasets were reprocessed uniformly, merged, and integrated with Seurat; results were compared with Harmony, Conos, and LIGER.
  • Clustering was performed in phases for all cells, all neurons, and targeted mid/ventral and motoneuron subsets; cell-type relationships were analyzed by principal-component distances, dendrograms, differential expression, and GO analysis.
  • Validation used high-content RNA in situ hybridization with 95 unique genes in ten overlapping sets on adult lumbar spinal cord sections from N = 3 animals, plus immunofluorescence and WFA-lectin staining; embryonic lineage was inferred by co-integration with an e9.5–e13.5 dataset, and classifiers were tested on an independent 10x single-nucleus dataset from four adult mice.

Limitations

  • Stated: the atlas is limited to RNA expression in postnatal mouse spinal cord and merges multiple time points; in situ validation is limited to adult lumbar spinal cord, so failure to detect some genes may reflect age, region, transient expression, or technical issues.
  • Stated: the analysis is population-based and likely underrepresents true biological variability, including noisy expression and continua of cell types; single-cell/nucleus profiling gives one perspective and may not capture connectivity, electrophysiology, or behavioral function.
  • Evident: classifier training used labels from the harmonized analysis because no gold standard exists, and several original datasets were included in training, so some performance estimates may be artificially high; ventral clusters were less robust and less separable than dorsal clusters.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41467-021-25125-1

Cites

138 of 138 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Integration of six independent studies on single cell spinal cord data reveals the major cell types of the spinal cord. a Summary of the datasets used in this study, including the studies that used single-cell/nucleus RNA sequencing to analyze postnatal mouse spinal cord cell types (colored names above the gray bar) and additional studies that were used for focused aspects of the analysis below the gray bar. The age and technique (cell or nuclei isolation) is represented for each study. b UMAP presentation of the 52,623 cells/nuclei in the final dataset, without integration and colored by the study of origin (colors in the legend). c UMAP presentation of the same 52,623 cells/nuclei in the final dataset, integrated by study and colored by the study of origin (same colors as in (b)). d UMAP presentation of the cells/nuclei in the merged dataset, integrated by study and colored by cell type. e Dot plot of the expression of marker genes for the major coarse cell types. Average expression for each cluster is shown by color intensity and the percent of cells/nuclei in each cluster that expressed each gene is shown by dot diameter.
Fig. 1 Integration of six independent studies on single cell spinal cord data reveals the major cell types of the spinal cord. a Summary of the datasets used in this study, including the studies that used single-cell/nucleus RNA sequencing to analyze postnatal mouse spinal cord cell types (colored names above the gray bar) and additional studies that were used for focused aspects of the analysis below the gray bar. The age and technique (cell or nuclei isolation) is represented for each study. b UMAP presentation of the 52,623 cells/nuclei in the final dataset, without integration and colored by the study of origin (colors in the legend). c UMAP presentation of the same 52,623 cells/nuclei in the final dataset, integrated by study and colored by the study of origin (same colors as in (b)). d UMAP presentation of the cells/nuclei in the merged dataset, integrated by study and colored by cell type. e Dot plot of the expression of marker genes for the major coarse cell types. Average expression for each cluster is shown by color intensity and the percent of cells/nuclei in each cluster that expressed each gene is shown by dot diameter.CC BY · PubMed Central
Fig. 2 Harmonized atlas of 69 populations of spinal cord neurons. a UMAP presentation of 19,353 neuronal cells/nuclei of the postnatal mouse spinal cord, colored and annotated by cell type cluster. b–f The same cells/nuclei, colored by the study of origin (b), by robustness (silhouette) score (c), a neurotransmitter (d), lamina (e), and family (f). I inhibitory, I/Ch inhibitory cholinergic, Ch cholinergic, E/Ch excitatory cholinergic, E excitatory, MN motoneurons, ME mid excitatory, CC Clarke’s Column (*see main text for note on this designation), MI mid inhibitory, VE ventral excitatory, VI ventral inhibitory. Laminae were assigned based on in situ hybridization validation experiments and are colored by the approximate depth from the dorsal surface of the cord (hot pink to violet). See main text for description of neuronal families.
Fig. 2 Harmonized atlas of 69 populations of spinal cord neurons. a UMAP presentation of 19,353 neuronal cells/nuclei of the postnatal mouse spinal cord, colored and annotated by cell type cluster. b–f The same cells/nuclei, colored by the study of origin (b), by robustness (silhouette) score (c), a neurotransmitter (d), lamina (e), and family (f). I inhibitory, I/Ch inhibitory cholinergic, Ch cholinergic, E/Ch excitatory cholinergic, E excitatory, MN motoneurons, ME mid excitatory, CC Clarke’s Column (*see main text for note on this designation), MI mid inhibitory, VE ventral excitatory, VI ventral inhibitory. Laminae were assigned based on in situ hybridization validation experiments and are colored by the approximate depth from the dorsal surface of the cord (hot pink to violet). See main text for description of neuronal families.CC BY · PubMed Central
Fig. 3 Trends in dorsal-ventral organization of spinal cord neuron types. a Dendrogram showing the relationships between the 69 neuronal cell types based on their distance from each other in the 50-dimensional principal component (PC) space. MN motoneuron, IN interneurons (and projection neurons), CSF-cN cerebrospinal fluid contacting neurons, DE dorsal excitatory, DI dorsal inhibitory ME mid excitatory, MI mid inhibitory, VE ventral excitatory, VI ventral inhibitory, center represents a group of 3 cell types located near lamina X–the center of the spinal cord. b Differential gene expression tests (ROC) were used to compare overall gene expression between the dorsal cell types and mid/ventral cell types and significant gene lists were analyzed by gene ontology (GO) term searches with GO DAVID using molecular function and biological process terms, as well as KEGG pathway lists (which are underlined) and the top terms for each cell class are shown. c–f Validation of differentially expressed genes using RNA in situ hybridization (c, e), antibody staining (d), or WFA-lectin staining (f). 20x tiled images, with brightness and contrast adjusted. All images are representative of the pattern observed in at least 3 sections each from N = 3 animals. Scale bars are 200 μm. g Dot plots showing expression of plasticity-related genes in each harmonized cluster, in which dot color intensity corresponds to average expression level (Ave Exp) and dot size corresponds to the percent of each cluster that expressed the gene (% Exp).
Fig. 3 Trends in dorsal-ventral organization of spinal cord neuron types. a Dendrogram showing the relationships between the 69 neuronal cell types based on their distance from each other in the 50-dimensional principal component (PC) space. MN motoneuron, IN interneurons (and projection neurons), CSF-cN cerebrospinal fluid contacting neurons, DE dorsal excitatory, DI dorsal inhibitory ME mid excitatory, MI mid inhibitory, VE ventral excitatory, VI ventral inhibitory, center represents a group of 3 cell types located near lamina X–the center of the spinal cord. b Differential gene expression tests (ROC) were used to compare overall gene expression between the dorsal cell types and mid/ventral cell types and significant gene lists were analyzed by gene ontology (GO) term searches with GO DAVID using molecular function and biological process terms, as well as KEGG pathway lists (which are underlined) and the top terms for each cell class are shown. c–f Validation of differentially expressed genes using RNA in situ hybridization (c, e), antibody staining (d), or WFA-lectin staining (f). 20x tiled images, with brightness and contrast adjusted. All images are representative of the pattern observed in at least 3 sections each from N = 3 animals. Scale bars are 200 μm. g Dot plots showing expression of plasticity-related genes in each harmonized cluster, in which dot color intensity corresponds to average expression level (Ave Exp) and dot size corresponds to the percent of each cluster that expressed the gene (% Exp).CC BY · PubMed Central
Fig. 4 Family structure and in situ validation for adult spinal cord tissue. a UMAP for neuronal cell types Excit-14 through Excit-19. b Dot plot of the distribution of selected marker genes across the 69 neuronal clusters in which dot color intensity corresponds to average expression level (Ave Expression) and dot size corresponds to the percent of each cluster that expressed the gene (% Expressed). c Feature plots of each gene expression pattern in Excit-14 through Excit-19. Expression levels are indicated by color intensity, with the maximum level indicated below each plot. The co-expression of Nmu (red) and Tac2 (green) are shown in the right-most plot, with expression levels cut-off at a maximum of 2.5 to highlight co-expressing cells in yellow. d, e RNA in situ hybridization of selected marker genes Sox5, Col5a2, Tac1, Nmu, Tac2 on an adult mouse lumbar spinal cord section. Cells were assigned to individual excitatory clusters with cluster number identity shown based on marker gene expression. Inset show representative cells of Excit-14 (14*) and Excit-15 (15**) with in situ hybridization for Sox5 (green), Col5a2 (red), Enpp1 (blue). 20x tiled images, with brightness and contrast adjusted. All images are representative of the pattern observed in at least 3 sections each from N = 3 animals. Scale bar is 100 μm in (d) and 25 μm in (e). f Quantification of the cells in adult spinal cord tissue that could be defined using sets of marker genes in situ. The cell types analyzed by each set of genes are shown on the left, the number of cells counted for each set are shown at the base of the bars, and the percent of counted cells are shown for each animal (N = 3, replicates and mean ± standard error) that could be confidently assigned to a single cluster (white bars), or that could be assigned to a single cluster or to pair of closely related clusters (gray). For each set, the coarse criteria for counting total cells are specified in the Methods. Set 4, which includes the Sox5 family clusters, is highlighted in green as an example.
Fig. 4 Family structure and in situ validation for adult spinal cord tissue. a UMAP for neuronal cell types Excit-14 through Excit-19. b Dot plot of the distribution of selected marker genes across the 69 neuronal clusters in which dot color intensity corresponds to average expression level (Ave Expression) and dot size corresponds to the percent of each cluster that expressed the gene (% Expressed). c Feature plots of each gene expression pattern in Excit-14 through Excit-19. Expression levels are indicated by color intensity, with the maximum level indicated below each plot. The co-expression of Nmu (red) and Tac2 (green) are shown in the right-most plot, with expression levels cut-off at a maximum of 2.5 to highlight co-expressing cells in yellow. d, e RNA in situ hybridization of selected marker genes Sox5, Col5a2, Tac1, Nmu, Tac2 on an adult mouse lumbar spinal cord section. Cells were assigned to individual excitatory clusters with cluster number identity shown based on marker gene expression. Inset show representative cells of Excit-14 (14*) and Excit-15 (15**) with in situ hybridization for Sox5 (green), Col5a2 (red), Enpp1 (blue). 20x tiled images, with brightness and contrast adjusted. All images are representative of the pattern observed in at least 3 sections each from N = 3 animals. Scale bar is 100 μm in (d) and 25 μm in (e). f Quantification of the cells in adult spinal cord tissue that could be defined using sets of marker genes in situ. The cell types analyzed by each set of genes are shown on the left, the number of cells counted for each set are shown at the base of the bars, and the percent of counted cells are shown for each animal (N = 3, replicates and mean ± standard error) that could be confidently assigned to a single cluster (white bars), or that could be assigned to a single cluster or to pair of closely related clusters (gray). For each set, the coarse criteria for counting total cells are specified in the Methods. Set 4, which includes the Sox5 family clusters, is highlighted in green as an example.CC BY · PubMed Central
Fig. 5 Co-integration of embryonic and postnatal through adult spinal cord neuronal types. a To reveal the temporal relationship between embryonic and postnatal through adult cell types, the Delile, and harmonized datasets were co-integrated and are shown in a UMAP, colored by dataset (right). b A UMAP of the co-integrated datasets, colored by clusters from the Delile et al. study (bold labels) or the harmonized analysis (regular font labels). Cluster annotations are repeated in cases in which a group of cells from a given cluster are located at a distance from the cluster centroid (ex. for dI4.6). c Feature plots of selected marker genes for the cardinal classes of spinal cord lineages. d Sankey plot of the relationships between embryonic lineages (left) and harmonized cell types (right) showing multiple examples of divergence and convergence.
Fig. 5 Co-integration of embryonic and postnatal through adult spinal cord neuronal types. a To reveal the temporal relationship between embryonic and postnatal through adult cell types, the Delile, and harmonized datasets were co-integrated and are shown in a UMAP, colored by dataset (right). b A UMAP of the co-integrated datasets, colored by clusters from the Delile et al. study (bold labels) or the harmonized analysis (regular font labels). Cluster annotations are repeated in cases in which a group of cells from a given cluster are located at a distance from the cluster centroid (ex. for dI4.6). c Feature plots of selected marker genes for the cardinal classes of spinal cord lineages. d Sankey plot of the relationships between embryonic lineages (left) and harmonized cell types (right) showing multiple examples of divergence and convergence.CC BY · PubMed Central
Fig. 6 Computational classification of spinal cord cell types. a Confusion matrices of the F1 scores for the classification of coarse cell types using label transfer, a support vector machine (SVM), and a fully connected neural network (neural net), (blue = 0; maroon = 1). The actual cell types are in rows and the predicted cell types are in columns in the same order. b Confusion matrices of the F1 scores for the classification of fine neuronal sub-types using label transfer and a fully connected neural network. The actual cell types are in rows and the predicted cell types are in columns, both in the order presented in Table 1. Alternating cell types are labeled. c Model of the two-tiered classification approach in which all cells/nuclei are classified into coarse cell types using label transfer (also including low-quality junk and doublets). Subsequently, all cells/nuclei that were classified as neurons, motoneurons, or doublets by label transfer are further classified into 69 neuronal cell types (also including doublets). d Experimental design for generating an independent set of single nucleus RNA sequencing data. e Distribution plot showing how nuclei from each cluster (rows) were distributed into each of the harmonized cell types (columns), normalized by rows with dark blue = 0.0 fraction; maroon = 1.0 fraction). f Bar plot of the total counts of nuclei that were from known clusters and were correctly classified (81% of total), that were from known clusters and were incorrectly classified (9% of total), that were from unknown clusters but could be identified by their classification (3% of total), or that were from unknown clusters and could not be identified (7% of total). OPC oligodendrocyte precursor cell, progen.1 oligodendrocyte progenitor 1, progen.2 oligodendrocyte progenitor 2, Olig.1 oligodendrocyte 1, Olig.2 oligodendrocyte 2, Periph. peripheral glia, Mening.1 meninges 1, Mening.2 meninges 2, Epend. Ependymal cells, Astro.1 astrocytes 1, Astro.2 astrocytes 2, Endoth endothelial cells, Pericy pericytes, MN = motoneurons; low qual. low quality, MNa motoneurons alpha, PGC preganglionic cell.
Fig. 6 Computational classification of spinal cord cell types. a Confusion matrices of the F1 scores for the classification of coarse cell types using label transfer, a support vector machine (SVM), and a fully connected neural network (neural net), (blue = 0; maroon = 1). The actual cell types are in rows and the predicted cell types are in columns in the same order. b Confusion matrices of the F1 scores for the classification of fine neuronal sub-types using label transfer and a fully connected neural network. The actual cell types are in rows and the predicted cell types are in columns, both in the order presented in Table 1. Alternating cell types are labeled. c Model of the two-tiered classification approach in which all cells/nuclei are classified into coarse cell types using label transfer (also including low-quality junk and doublets). Subsequently, all cells/nuclei that were classified as neurons, motoneurons, or doublets by label transfer are further classified into 69 neuronal cell types (also including doublets). d Experimental design for generating an independent set of single nucleus RNA sequencing data. e Distribution plot showing how nuclei from each cluster (rows) were distributed into each of the harmonized cell types (columns), normalized by rows with dark blue = 0.0 fraction; maroon = 1.0 fraction). f Bar plot of the total counts of nuclei that were from known clusters and were correctly classified (81% of total), that were from known clusters and were incorrectly classified (9% of total), that were from unknown clusters but could be identified by their classification (3% of total), or that were from unknown clusters and could not be identified (7% of total). OPC oligodendrocyte precursor cell, progen.1 oligodendrocyte progenitor 1, progen.2 oligodendrocyte progenitor 2, Olig.1 oligodendrocyte 1, Olig.2 oligodendrocyte 2, Periph. peripheral glia, Mening.1 meninges 1, Mening.2 meninges 2, Epend. Ependymal cells, Astro.1 astrocytes 1, Astro.2 astrocytes 2, Endoth endothelial cells, Pericy pericytes, MN = motoneurons; low qual. low quality, MNa motoneurons alpha, PGC preganglionic cell.CC BY · PubMed Central

Verification

Claims the paper makes

136 Own result · 42 Own interpretation · 38 Not checked yet · 17 Own method · 17 No basis given · 4 Partly supported · accuracy not yet measured

Verify all 93 references — free account
  • It may state two separate claims as one. 111
  • The quote only partly states this claim. 2
  • Probably not a substantial claim. 28
130 of 130
  • The paper relies on this without support.

    There is no existing gold standard for spinal cord cell identities.

    “because there is no existing gold standard for spinal cord cell identities”
  • The paper relies on this without support.

    The harmonized analysis cell type labels are adequate as training labels for machine learning classifiers.

    “each of these models were trained using cell type labels from the harmonized analysis”
  • The paper relies on this without support.

    Perfect performance would require perfect and invariable biological data: discrete cell types that express completely distinct patterns of gene expression and experimental data without natural heterogeneity, doublets, low-quality cells, or other sources of indeterminate data.

    “Perfect performance would require perfect and invariable biological data: discrete cell types that express completely distinct patterns of gene expression and experimental data without natural heterogeneity, doublets, low-quality cells, or other sources of indeterminate data.”
  • The paper relies on this without support.

    Developmental lineage is a powerful and influential framework for categorizing spinal neurons.

    “Developmental lineage has been a powerful and influential framework for categorizing spinal neurons, particularly within the ventral horn, and relies on the combinatorial expression code of transcription factors that specify distinct progenitor domains along the dorsal-ventral axis of the spinal cord.”
  • The paper relies on this without support.

    There is no commonly accepted standard set of spinal cord cell types.

    “In the absence of a commonly accepted standard set of spinal cord cell types, it is impossible to answer this question completely.”
  • The paper relies on this without support.

    The rate of false-negative data in each cell and the requirement for statistical power necessitates analyzing many cells of each type and considering population-level shared patterns.

    “Data is captured from thousands of individual cells, but the rate of false-negative data in each cell and the requirement for statistical power necessitates analyzing many cells of each type and considering population-level shared patterns.”
6 of 6

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 131 references

Research works printed but not linked: 2

  1. Blum, J. A. et al. Single-cell transcriptomic analysis of the adult mouse spinal cord reveals molecular diversity of autonomic and skeletal motor neurons. Nat. Neurosci. 24, 572–583 (2021). · DOI · Paper page
  2. Dubreuil, A., Valente, A., Beiran, M., Mastrogiuseppe, F. & Ostojic, S. Complementary roles of dimensionality and population structure in neural computations. bioRxiv https://www.biorxiv.org/content/10.1101/2020.07.03.185942v1.full?%3Fcollection= (2020). · DOI

Cited by

Sorted by citations

See all 261

Metadata

Authors
Russ, D. E., Cross, R. B. P., Li, L., Koch, S. C., Matson, K. J. E., Yadav, A., Alkaslasi, M. R., Lee, D. I., Le Pichon, C. E., Menon, V., & Levine, A. J.
Affiliations
  • National Institutes of Health (US)
  • University College London (GB)
  • Columbia University Irving Medical Center (US)
Corresponding author
Levine, Ariel J.
Venue
Nature Communications
Published
2021-09-29
Volume
12
Issue
1
Pages
5722-5722
ISSN
2041-1723 (ISSN-L)
OpenAlex
W3203053031
Cited by
255
References
131
Citations by year
2026: 53 · 2025: 49 · 2024: 65 · 2023: 57 · 2022: 30 · 2021: 1
FWCI
12.48
Type
journal article
Language
en
Open access
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • UCL Discovery (University College London) (submittedVersion)
Funders
U.S. Department of Health & Human Services | NIH | National Institute of Neurological Disorders and Stroke (1ZIA NS003153)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Developmental Biology and Gene Regulation
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Biophysics › Cell Image Analysis Techniques
Keywords
mouse spinal cord, cell type atlas, single-cell transcriptomics, cell type classification, neurotransmitter identity, gene expression signatures
MeSH terms
RNA-Seq, Animals, Atlases as Topic, Cell Nucleus, Neurons, Spinal Cord, Mice, Single-Cell Analysis, Transcriptome, Spatial Analysis, Datasets as Topic

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-08, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-08 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review