henhao
Sign in
NeuroscienceNature
Peer-reviewed journal article

A molecularly defined and spatially resolved cell atlas of the whole mouse brain

Meng ZhangiD, Xingjie PaniD, Won JungiD, Aaron R. HalperniD, Stephen W. EichhorniD, Zhiyun LeiiD, Limor CoheniD, Kimberly A. SmithiD, Bosiljka TasiciD, Zizhen YaoiD, Hongkui ZengiD, Xiaowei ZhuangiD

Abstract

In mammalian brains, tens of millions to billions of cells form complex interaction networks to enable a wide range of functions. The enormous diversity and intricate organization of cells in the brain have so far hindered our understanding of the molecular and cellular basis of its functions. Recent advances in spatially resolved single-cell transcriptomics have allowed systematic mapping of the spatial organization of molecularly defined cell types in complex tissues1-3. However, these approaches have only been applied to a few brain regions1-11 and a comprehensive cell atlas of the whole brain is still missing. Here, we imaged a panel of >1,100 genes in [~]8 million cells across the entire adult mouse brain using multiplexed error-robust fluorescence in situ hybridization (MERFISH)12 and performed spatially resolved, single-cell expression profiling at the whole-transcriptome scale by integrating MERFISH and single-cell RNA-sequencing (scRNA-seq) data. Using this approach, we generated a comprehensive cell atlas of >5,000 transcriptionally distinct cell clusters, belonging to [~]300 major cell types, in the whole mouse brain with high molecular and spatial resolution. Registration of the MERFISH images to the common coordinate framework (CCF) of the mouse brain further allowed systematic quantifications of the cell composition and organization in individual brain regions defined in the CCF. We further identified spatial modules characterized by distinct cell-type compositions and spatial gradients featuring gradual changes in the gene-expression profiles of cells. Finally, this high-resolution spatial map of cells, with a transcriptome-wide expression profile associated with each cell, allowed us to infer cell-type-specific interactions between several hundred pairs of molecularly defined cell types and predict potential molecular (ligand-receptor) basis and functional implications of these cell-cell interactions. These results provide rich insights into the molecular and cellular architecture of the brain and a valuable resource for future functional investigations of neural circuits and their dysfunction in diseases.

PDFDOICodeData
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
509
FWCI
45.14
Field percentile
top 1%
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · max effort · 8 Nov 2026 · from the full text of version 1 (7 Mar 2023) · CC BY 4.0

MERFISH profiling of >1,100 genes in ~8 million cells spanning the whole mouse brain, integrated with scRNA-seq, maps >5,000 clusters into the Allen Common Coordinate Framework.

Earlier whole-brain scRNA-seq surveys (~500,000 to 700,000 cells) likely underestimated diversity and lacked spatial context, and spatial transcriptomic atlases covered only a few regions. Registering this single-cell map of the entire mouse brain to the Allen Common Coordinate Framework (CCF), with imputed transcriptomes, enables region-level composition analysis and ligand-receptor interaction calls constrained by physical proximity rather than co-expression alone.

Key results

  1. ShownMERFISH recovered essentially the full scRNA-seq taxonomy in situ, with most imaged cells confidently labeled at subclass and cluster level and per-gene copy numbers reproducible across animals and consistent with sequencing.

    Detected all 306 subclasses and 5,139 (99%) of 5,200 clusters; 82% and 75% of cells passed subclass (0.8) and cluster (0.5) confidence thresholds; per-gene mean copy number r = 0.992 between replicate animals, r = 0.830 vs whole-brain bulk RNA-seq, r = 0.742 vs scRNA-seq. · MERFISH imaging of the whole mouse brain; Cell classification and registration to the…; Figure legends

  2. ShownNeuronal types are highly diverse and mostly region-specific; hypothalamus, midbrain and hindbrain hold the most neuronal types and highest local type complexity, and excitatory-inhibitory balance differs sharply between regions.

    Brain-wide 43% neurons vs 57% non-neuronal cells; 283 of 306 subclasses are neuronal. Glutamatergic ~61% and GABAergic ~38% of neurons; Glut:GABA ratio ~6:1 to 8:1 in hippocampal formation, isocortex and thalamus vs ~1:14 in striatum; ~1% of neurons co-express glutamate and GABA transporters, mostly in non-telencephalic areas. · Cellular diversity and spatial organization of…; Spatially dependent neurotransmitter and…

  3. ShownNon-neuronal cells are also regionally specialized, most strikingly astrocytes: nearly every astrocyte cluster occupies a distinct territory, and the two largest clusters partition telencephalic from non-telencephalic brain.

    23 non-neuronal subclasses and 99 clusters; of non-neuronal cells, oligodendrocytes 31%, vascular 27%, astrocytes 24%, immune 8%, oligodendrocyte progenitors 6%, other 4%. Among 24 astrocyte clusters, Astro 5113 (47% of astrocytes) lies exclusively in telencephalon and Astro 5108 (39%) in non-telencephalic regions. · Cellular diversity and spatial organization of…; Discussion

  4. ShownClustering cells by local cell-type composition yields spatial modules that largely match CCF major regions, while most subclasses show continuous rather than discrete cluster structure and many form spatial expression gradients across subregion borders.

    16 level-1 and 127 level-2 spatial modules; level-1 diverged from CCF mainly at the midbrain-hindbrain border; level-2 split caudoputamen lateral/medial but missed most isocortical area borders. Intratelencephalic neuron pseudotime vs cortical depth r = 0.797; gradients also in D1/D2 medium spiny neurons, lateral septal complex, tanycytes and from hypothalamus to hindbrain. · Molecularly defined brain regions - spatial…; Spatial gradients of molecularly defined cell…; Figure legends

  5. SuggestedProximity testing against local spatial randomization, plus ligand-receptor upregulation in proximal pairs, predicts several hundred interacting subclass pairs, mostly involving non-neuronal cells; Wnt usage is broad among non-neuronal pairs but region-specific for pairs involving neurons.

    Calls needed 15 μm proximity beyond 1000 local randomizations (adjusted p < 0.05, at least 20 proximal pairs) or 30 μm proximity plus a ligand-receptor pair up ≥2-fold (adjusted p < 0.01). Laminin pairs rose most in endothelial and pericyte contacts with border-associated macrophages; chandelier-CA3 Glut neighbours showed Wnt pair and chandelier Grin2a upregulation. · Cell-type-specific cell-cell interactions and…; Cell-cell interaction analysis; Ligand-receptor analysis and analysis of other…; Discussion

Methods, data and reproducibility

Multiplexed error-robust fluorescence in situ hybridization (MERFISH) imaged >1,100 genes in 10-μm sections from four mice; canonical correlation analysis integration with ~4 million scRNA-seq cells transferred labels and imputed whole transcriptomes, and cells were registered to Allen CCFv3 for spatial-module, gradient and proximity-interaction analyses.

  • Pooled whole-brain MERFISH single-cell profiles (1122 shared genes used for integration) spanning 11 major regions plus fiber tracts and ventricles · mouse (C57BL/6, aged 57-63 days) · n = ~8 million cells from four animals · 10.35077/act-bag (Brain Image Library)
  • MERFISH coronal series of 10-μm sections kept every 200 μm (Animal 1, male, 1124-gene panel) or every 100 μm (Animal 2, female, 1147-gene panel) · mouse (C57BL/6) · n = 67 (Animal 1) and 150 (Animal 2) sections · 10.35077/act-bag (Brain Image Library)
  • MERFISH sagittal series of 10-μm sections every 200 μm from midline to lateral edge (Animal 3, male; Animal 4, male, filling broken regions) · mouse (C57BL/6) · n = 28 sections (25 from Animal 3, 3 from Animal 4) · 10.35077/act-bag (Brain Image Library)
  • Allen Institute whole-brain scRNA-seq reference (companion Yao et al. study) with 5,200 clusters in 306 subclasses, used for panel design, label transfer and imputation · mouse · n = ~4 million cells · Neuroscience Multi-omics Archive, https://nemoarchive.org (companion Yao et al. data)
  • Whole-brain bulk RNA-seq (Kapa mRNA HyperPrep, Illumina NextSeq500, RNA integrity >8) used to benchmark MERFISH per-gene copy numbers · mouse (aged 56-63 days) · n = three whole brains
  • Allen Brain Atlas in situ hybridization images used to check measured and imputed spatial expression patterns · mouse (Allen Mouse Brain Atlas)
  • Allen Mouse Brain CCF version 3 with Allen Reference Atlas Nissl templates and region annotations, used as the registration target · mouse (Allen Mouse Brain CCF reference)
Methods & tools
MERFISH (32-bit, Hamming-distance-4, Hamming-weight-4 codebook with 116 blank barcodes), MERlin (drift correction, pixel-based barcode decoding), Cellpose (DAPI nuclei and polyA soma segmentation), Scrublet (doublet removal, score threshold 0.25), ALLCools SeuratIntegration (CCA co-embedding, label transfer, imputation), Scanpy (normalization, PCA, highly variable genes), METIS k-way graph partitioning (48 integration partitions), Elastix (affine and B-spline registration to CCF), Leiden clustering (spatial modules), Pseudotime / PC1 analysis (spatial gradients), CellChat ligand-receptor database, One-sided Welch's t-test with Benjamini-Hochberg correction
Code
https://github.com/ZhuangLab/MERlin
Data & accessions
https://doi.org/10.35077/act-baghttps://nemoarchive.org/http://atlas.brain-map.org/https://mouse.brain-map.org/
Reagents & environment
C57BL/6 mice aged 57-63 days; encoding probes from Twist Biosciences oligo pool; readout probes (Alexa750, Cy5, Alexa647) from Bio-synthesis/IDT; proteinase K (NEB P8107S); RNase inhibitor (NEB M0314S); MERlin, Cellpose, ALLCools, Scanpy, Elastix.
Experimental conditions
Home-built MERFISH imaging platforms with Bioptechs FCS2 flow chamber and automated fluidics; 16 imaging rounds, five 1.5-μm z-positions per field of view (three or six for Animal 1); Leica CM3050 S cryostat; Illumina NextSeq500; no computing hardware stated.
Reproducibility gaps
Public code covers only MERFISH image analysis (MERlin); no repository is cited for integration, CCF registration, spatial-module, gradient or cell-cell interaction analyses, and software versions are not given.
Limitations (6)
  • Interaction calls may still include false positives where cell types merely co-localize in small structures; shrinking the proximity threshold or 100 μm randomization range would cut these but could discard genuine paracrine interactions. · stated by the authors
  • Requiring ligand-receptor upregulation in proximal versus non-proximal pairs can miss interactions via constantly expressed pairs, and soma-centroid proximity cannot capture long-range synaptic or astrocyte-process contacts; all predictions need experimental validation. · stated by the authors
  • CCF registration retains alignment errors from individual brain differences relative to the averaged CCFv3 template and from section deformations not fully corrected during image alignment. · stated by the authors
  • Non-neuronal diversity is likely underestimated: the scRNA-seq reference placed all microglia in one cluster, although multiple microglial states, some seen at low frequency in healthy adult mice, have been described. · stated by the authors
  • Ligand-receptor scores are computed from imputed profiles (weighted averages of 30 scRNA-seq neighbours), so proximal-versus-distal differences may mirror within-subclass cluster or spatial-gradient structure rather than interaction-driven regulation. · noted in this summary
  • Only four mice contribute (coronal series from one male and one female, sagittal largely from one male), so inter-individual and sex variation in regional composition is unquantified; imputed patterns for off-panel genes were checked against Allen in situ hybridization only visually. · noted in this summary

Who should read:Read this if you need a CCF-registered single-cell reference of mouse brain cell types, want to place scRNA-seq clusters anatomically, or build proximity-based cell-cell interaction inference for imaging spatial transcriptomics.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISYao '23He '24Shi '23Zhang '20Han '23Langlieb '23
85% semantic matchNeuroscience
Zizhen Yao, Cindy T. J. van Velthoven, Michael Kunst+97 more·Nature·Dec 13, 2023· 1,189 citations