Overview of the Brain Cell AtlasThe resource, which is also provided as an interactive web portal, includes 11.3M human cells from 14 main regions and 30 subregions of the brain (Supplementary Fig. 1), while the mouse data include 15M cells. Single-cell RNA sequencing and single-nucleus RNA sequencing data of the brain were searched through literature and the single-cell database23, covering over 1,800 published datasets deposited in Gene Expression Omnibus (GEO)24, the UCSC browser25, ArrayExpress26, Allen Brain Map (https://portal.brain-map.org/) and Synapse (https://www.synapse.org/) (Supplementary Fig. 2). The resource covers 70 human studies of 6,577 samples (Supplementary Fig. 3a–c), along with 103 mouse studies of 25,710 samples. The metadata were manually curated and raw counts were collected (Methods, Supplementary Fig. 2 and Supplementary Table 1) in a consistent manner. Two well-established datasets were used as refs. 10,11 to infer cell type labels in other datasets using reference-based machine learning algorithms. The adult ref. 10 contains 3.3 M nuclei from tissues of four post-mortem healthy adults aged from 29 to 60 years across the whole brain, while the fetal ref. 11 contains 1.6 M cells from the first-trimester developing brain tissues.The human brain datasets were sorted into four types based on the sample source: adult (8,062,832 nuclei and cells), fetal (2,203,728 cells), organoids (861,169 cells) and brain tumor (234,295 cells) (Fig. 1a), while 94.8% cells were sequenced with 10x Chromium (Supplementary Fig. 3a) covering a time span from 6 gestational weeks (GW) to over 80 years old (Fig. 1b). In total, 46.4% of the fetal cells were from the first trimesters (0–12 GW), while postnatal cells were mainly (68.3%) from 40–80-year-old donors. As for sex in adult data, cells from female, male and unknown sex take up 24.6%, 70.7% and <5%, respectively. The sex for most (91.7%) cells from the fetal samples were undetermined, leaving a female-to-male ratio of 1.3:1 in the rest (Fig. 1c). For disease status, 74.1% were healthy samples and 3.1% were unspecified, while disease samples were dominated by Alzheimer’s disease (AD) followed by epilepsy, gliomas (glioblastomas, oligodendroglioma, astrocytoma, mixed glioma and so on), amyotrophic lateral sclerosis (ALS), major depressive disorder (MDD), autism spectrum disorder (ASD), dementia, Parkinson’s disease (PD) and multiple sclerosis (MS) (Fig. 1d). As for brain regions, the resource covers major cerebral cortex regions (frontal lobe, parietal lobe, occipital lobe and temporal lobe), cerebellum, brain stem (midbrain, pons and medulla oblongata) and the limbic system (hippocampus, thalamus, hypothalamus and amygdala) (Fig. 1e,f). Most cells or nuclei were collected from the hippocampus (13.1%), followed by prefrontal cortex (11.1%), occipital lobe (10.3%) and basal ganglia (9.4%) (Fig. 1g).Fig. 1: Statistics of the Brain Cell Atlas.a, Circular plot showing the proportions of the four primary sample types in the Brain Cell Atlas: adult (n = 8,062,832), fetal (n = 2,203,728), organoid (n = 861,169) and tumor (n = 234,295). A fraction of nonadult postnatal samples ranging from 0 to 20 years old were included in the adult brain data (see b). b, Bar plots showing the distribution of cell numbers across various age groups in human samples, spanning 6 to 39 GW in fetal samples and from 0 to over 80 years of age in adults. N/A indicates that age information is not available in the original publication. c, Stacked bar plot showing the proportions of donor sex in adult and fetal samples. N/A denotes samples with unavailable sex information. d, Histogram showing the cell counts categorized by donor status in both healthy and diseased conditions. ‘Gliomas’ include glioblastomas, oligodendroglioma, astrocytoma, mixed glioma and so on. ‘Others’ include carcinoma and mild cognitive impairments. N/A represents that medical condition information is not accessible. e, Anatomical depiction of the main regions where samples were collected in the Brain Cell Atlas. f, Hierarchical representation of anatomical structures in the adult brain, with line thickness reflecting cell proportions in each region. g, Histogram showing the cell counts per region in the adult brain data. h, Dot plot showing the cell markers derived from adult brain data, along with the top two relevant markers for each region. i, UMAP visualization of the adult brain data achieved by label transfer of reference data, showcasing cell type proportions across different brain regions. j, Stacked bar plot showing the cell type distribution in different brain regions. The color codes are the same as those listed in i.Source dataAs an integrative resource, a consensus cell type annotation of all the adult data was derived from seven well-established reference-based machine learning methods (Methods) as well as an in-house built hierarchical annotation workflow (scAnnot). This general cell type annotation based on the eight machine learning methods may help with the selection of target data for specific analysis. The consensus cell type annotation resulted in 32 primary clusters on the Uniform Manifold Approximation and Projection (UMAP) visualization (Fig. 1i), while cell type-specific differentially expressed genes (DEGs) can be derived (Methods and Fig. 1h). The cell type composition across brain regions indicates the regional specificity and heterogeneity (Fig. 1j). For instance, upper-layer and deep-layer intratelencephahlic neurons are more abundant in cortex regions than hippocampus.Atlas-level hierarchical cell type annotation with scAnnotTo achieve a multigranularity cell type annotation, we present scAnnot, a hierarchical cell annotation workflow based on the Variational Autoencoder model from single‐cell ANnotation using Variational Inference (scANVI27) (Methods and Fig. 2a). Although 45 out of the 70 datasets have their cell type annotations available (Supplementary Fig. 3b), the lack of consensus annotations hinders data integration of the resource. The cell types in the brain appear in a hierarchical manner of different granularities, which cannot be considered in the well-established reference-based machine learning methods. scAnnot trains machine learning models at different resolutions (granularities) and applies these models in a hierarchical structure. Using the adult reference of 31 primary cell types at the first-level of annotation, scAnnot selects 200 feature genes for each cell type-trained machine learning model with different hyperparameters (Supplementary Fig. 4). Then, it predicts the harmonized latent space of the cells, based on which the cell type labels can be inferred.Fig. 2: Atlas-level hierarchical cell type annotation.a, Schematic diagram illustrating the scAnnot tool (created with BioRender.com). This hierarchical classification model, based on scANVI, categorizes cells into cell types. The first level of classification groups cells into broad cell types, while the second level further classifies cells into more specific types within each broad category. The algorithm can be performed iteratively in classifying the cell types. b, Heatmap demonstrating the prediction accuracy for the first-level cell types. The rows represent the cell types reported in the publications, while the columns represent the scAnnot-predicted cell types. The color intensity represents the accuracy of the predictions, with darker colors indicating higher accuracy. c, Bar plot showing training and validation accuracies for the second layer of cell types. The x axis represents the broad cell types, while the y axis indicates the prediction accuracy. The blue and orange bars represent the training and validation accuracies, respectively. d, UMAP visualizing the reported cell types in the published data, with colors indicating the reported cell types. e, River plot illustrating the transition between reported and predicted first-level cell types. The left side represents the reported cell types, while the right side displays the scAnnot-predicted first-level cell types. f, UMAP visualization of the first-level scAnnot-predicted cell types. g, Stacked violin plot depicting the expression levels of select feature genes in the published data used by the scAnnot tool. The y axis represents the expression level, while the x axis denotes the gene names. The rows represent different cell types. h, UMAP visualization of the second-level scAnnot-predicted cell types.Source dataThe annotation accuracy can be assessed by the confusion matrix28 between the reported cell type labels and the scAnnot-predicted labels (Fig. 2b,c). Most cell types can be predicted with high accuracy (above 93%), and the average accuracy is 98%. The second-level (with a finer granularity) cell type annotation achieves accuracies of 90% and 83% on the training and validation sets, respectively. Both visual inspection on UMAP and quantitative evaluation of Silhouette score29 indicate that the integrated data are not much affected by the batch effect (Supplementary Fig. 5). The classification accuracy of the subpopulations in each cell type ranged from 50% to 100%, with Splatter cluster being the least discriminative, as expected (Fig. 2c).The primary cell types labels inferred by scAnnot are consistent with published annotation (Fig. 2d), while scAnnot further divides the intratelencephalic (IT) population into upper-layer intratelencephalic, deep-layer intratelencephalic and some miscellaneous (Fig. 2e,f). These cell clusters annotated by scAnnot can be confirmed by the feature gene expression (Fig. 2g). The hierarchical classification approach can further identify subpopulations at the second-level annotation with finer granularity (Fig. 2h).Potential NPCs in adult hippocampusMost single-cell data of the adult human brain generated from previous studies involve only a few samples on a specific experimental protocol or technology, resulting in disagreement over neurogenesis cell type definitions14,18,19,20. Taking advantage of the large-scale data in the Brain Cell Atlas, we investigated the potential existence of rare NPCs in the adult hippocampus. According to the machine-learning-based annotation, we selected data from five independent human studies including adults14,15,18,30, children14, infants14 and fetuses13. To facilitate the cell type annotation by cross-species comparison, the data were integrated with mouse data across all development stages31 (Fig. 3a,b) according to orthologous genes. Considering that the adult human data are dominated by mature neurons, we integrated adult human data with fetal and mouse data, which cover the complete neurogenic trajectory.Fig. 3: Integrating Brain Cell Atlas data to explore the existence of AHN.a, UMAP visualization of the integrated data after Harmony integration, colored by different studies. The Hochgerner dataset is a mouse dataset. b, Overview of the age distribution and cell count of the samples included in the studies used in a. Left: the distribution of sample ages. Right: the number of cells for each dataset. c, UMAP plot colored by the annotated cell types determined through marker gene expression. d, UMAP plots illustrating the expression levels of the key marker genes. The color from dark to bright represents the expression level from low to high. e, Dot plot showing the marker gene expression across different cell types. f, Volcano plot displaying the DEG results (two-sided Wilcoxon test) in putative NPCs compared to all others. The x axis indicates the log2 fold change in gene expression, while the y axis represents the negative logarithm (base 10) of the adjusted P values. The red dots represent significant DEGs (Benjamini–Hochberg-adjusted P value <0.05 and |logFC| >0.5), while the blue dots represent nonsignificance. The horizontal dashed line is the cutoff of the P value. The vertical dashed lines are the cutoff of the logFC. g, Bar plot showing the proportion of cells expressing the conserved cross-species NPC marker genes12 in different cell types. h, Violin plot showing the expression of the conserved cross-species NPC markers across different age groups in the NPCs in human datasets. The child group was removed due to only three cells annotated as NPCs. i, UMAP visualization of NPC gene module score, where brighter colors represents higher scores. j, Line chart showing the relative percentages of cells that change with the NPC gene module score. The x axis shows the NPC gene module score (Methods). For each NPC gene module score, the y axis shows the relative percentages of cells, which is the number of cells divided by the total number of putative NPCs. k, Confocal image of colocalization of NPC marker ASCL1 (green) and proliferative marker MKI67 (red) within the DG of healthy adult humans (n = 2 specimens). Scale bars, 20 μm (overview) and 10 μm (magnification). GCL, granule cell layer.Source dataAfter data integration using the Harmony32 program (Methods), the UMAP visualization shows well-mixed data from the six studies, while the data distribution shows a complete landscape of neurogenesis as well as the enriched mature neurons in the adult samples (Supplementary Fig. 6). The cell clusters were annotated according to the well-established cell type marker genes (Methods and Fig. 3c), including (1) MKI67 and TOP2A for NPCs, (2) DLX2 and SOX11 for neuroblast cells, (3) SOX11 and PROX1 for immature glutamatergic cells, (4) PROX1 and PLEKHA2 for glutamatergic neurons, (5) SLC17A7 and COL5A2 for CA neurons, (6) GAD1 and GAD2 for GABAergic neurons, (7) GFAP and AQP4 for astrocytes, (8) FLT1 and ENG for endothelial cells, (9) PDGFRA and OLIG1 for oligodendrocyte precursor cells (OPCs), (10) OLIG2 and SOX10 for newly formed oligodendrocytes (NFOLs) and (11) MOG and MAG for oligodendrocytes12,19 (Fig. 3d,e). According to proliferative markers (MKI67 and TOP2A) (Fig. 3d,e and Supplementary Fig. 7a,b), only a small proportion of cells (33 cells) in the adult hippocampus as well as 95 fetal and 494 mouse cells can be defined as putative NPCs. When hippocampal tissue sections from male adult macaques aged 7 and 15 years were stained (Supplementary Fig. 7c), we observed coexpression of progenitor cell markers (ASCL1 and SOX2) and glial fibrillary acidic protein (GFAP) in the dentate gyrus (DG). Furthermore, immunostaining of brain sections from individuals aged 6, 7 and 15 years revealed that SOX2+MKI67+ cells were detected in the subgranular zone (SGZ) of adult macaques (Supplementary Fig. 7d). The DEGs (Methods and Supplementary Table 2) of putative NPCs show some well-established neural progenitor markers12, including TOP2A, HMGB2, PBK and UBE2C (Fig. 3f and Supplementary Fig. 7e–h). Furthermore, reference-based machine learning methods based on the mouse data31 as the reference (Methods) also confirmed these putative NPCs (Supplementary Fig. 8).Additionally, we investigated NPC gene module score (Methods) analysis to validate putative NPCs on the basis of conserved cross-species NPC markers (TOP2A, HMGB2, PBK, UBE2C, RRM2, CDCA3, CCNA2 and TPX)12. Each gene in the NPC gene module score is expressed in approximately half of the putative NPCs but not in the other cell types (Fig. 3g and Extended Data Fig. 1a,b). These genes are lower expressed in adult and infant hippocampus than in fetuses (Fig. 3h). Putative NPCs attained the highest NPC gene module scores against other cell types, indicating that they exhibit the strongest signal of coexpression of conserved cross-species NPC markers (Fig. 3i and Extended Data Fig. 1c). The number of cells coexpressing two or more of these genes decreased sharply in adults, suggesting that putative NPCs in adults may exhibit distinct transcriptional signatures than the ones in fetuses (Fig. 3j and Supplementary Table 3).Trajectory analysis can also facilitate progenitor identification according to the developmental order of cells inferred from gene expression33 or RNA splicing status34. We extracted these putative NPCs together with cells in the two bifurcating directions, which are astrocytes and glutamatergic neurons, for trajectory inference (Extended Data Fig. 2a). Both pseudotime analysis and RNA velocity (Extended Data Fig. 2b,c and Supplementary Fig. 9) confirmed the trajectories starting from putative NPCs bifurcating to astrocytes and mature neurons. In adult humans, the DEGs (Extended Data Fig. 2d) of these putative NPCs against other mature cell types validate their cell identity, consistent with the DEGs of putative NPCs derived from fetal humans (Extended Data Fig. 2e,f). Gene Ontology (GO) enrichment analyses showed that putative NPCs mainly participate in neural precursor cell proliferation (NES, FABP7, ASCL1 and KDM1A), cell cycle regulation (TOP2A, MKI67, UBE2C, CENPF and TPX2), DNA replication (CCNA2, BRCA2 and CDK2), nuclear division (CENPF, SMC4 and CDC25C) and chromosome segregation (SMC1A, PLK1, MAD2L1 and AURKB) (Extended Data Fig. 2g).For experimental validation, we performed multiple immunostaining assays using antibodies against proliferating neural progenitor marker ASCL1 (ref. 35), along with the proliferative marker MKI67. Immunostaining shows that MKI67 colocalized with ASCL1 in the hippocampal DG of healthy adult humans, suggesting the existence of proliferative NPCs (Fig. 3k).Identification of PCDH9
high microglia across brain regionsThe unprecedented scale of the resource also facilitates the exploration of cell type diversity. A microglia population with a high level of PCDH9 expression was identified from the integrated data of 43 samples, covering 511,872 cells. These samples were obtained from four studies of adult human prefrontal cortex and hippocampal regions17,18,30,36, providing 12 well-annotated primary cell types (Fig. 4a). Zooming in the microglia cells from the primary cell types, we characterized a novel population of microglia with high PCDH9 expression (Fig. 4b). We next confirmed the existence of PCDH9+IBA1+ microglia in healthy adult brains across the prefrontal cortex and hippocampus by double immunofluorescence staining of the corresponding proteins (Extended Data Fig. 3a,b). Furthermore, leveraging DEGs identified in the 12 distinct microglial states by Sun et al.37 as a point of reference, our gene set scoring analysis demonstrated that microglia (PCDH9high) were positioned between a state of homeostasis and inflammatory II (Extended Data Fig. 4a,b). In addition to the microglial markers (APBB1IP (ref. 30), TBXAS1, SPP1 (ref. 38), LPCAT2 (ref. 39), P2RY12 (ref. 40) and SLCO2B1 (ref. 41)), the microglia (PCDH9high) population also exhibits high expression of immune-related genes (SPTLC2, CTTNBP2 (ref. 42), PEAK1 and APP) (Fig. 4c and Extended Data Fig. 4c,d), indicating a functional discrimination in modulating immune responses against other microglia cells. The microglia cluster highly expresses nonhomeostatic marker APOE43, colony-stimulating factor 1 receptor (CSF1R) and phagocytosis receptor MERTK43 (Supplementary Table 4).Fig. 4: Integrated datasets across brain regions yield a subtype of PCDH9-high expressing microglia.a, UMAP plot showing the four datasets integrated by harmony integration, with the colors representing the annotated cell type. The red dashed outline indicates microglia and microglia (PCDH9high) populations. b, UMAP plot illustrating a subset highlighted in red circles within a, comprising microglia and microglia (PCDH9high) cells. c, Dot plot showing the expression marker genes for microglia and microglia (PCDH9high). d, Bar plot showing the enriched GO terms in microglia and microglia (PCDH9high) clusters. The analysis was based on Enrichr, using two-sided Fisher’s exact test with Benjamini–Hochberg correction for multiple comparisons. e, GSEA visualization of microglia (PCDH9high) participating in axon guidance and endocytosis pathways. The analysis was based on clusterProfiler, using a two-sided hypergeometric test with Benjamini–Hochberg correction for multiple comparisons. f, Triple immunostaining of SPP1 (red), PCDH9 (gray) and MAG (green) confirmed the phagocytosis of myelin debris by microglia (PCDH9high) in the hippocampal DG (n = 2 specimens). Scale bars, 20 μm (overview) and 10 μm (magnification). g, Triple immunostaining of SPP1 (red), PCDH9 (gray) and MAG (green) confirming the phagocytosis of myelin debris by microglia (PCDH9high) in the prefrontal cortex (n = 2 specimens). The yellow arrowheads indicate SPP1+ PCDH9+ MAG+ microglia. Scale bars, 20 μm (overview) and 10 μm (magnification).Source dataThe GO terms of microglia (PCDH9high) against microglia (Supplementary Table 5) indicated that microglia might be engaged in synapse pruning, regulation of dopamine metabolic process and positive regulation of cytokine production. Yet, microglia (PCDH9high) might be involved in axon guidance, axonogenesis, nervous system development, neuron projection guidance and regulation of neuron projection development (Fig. 4d). Gene set enrichment analysis (GSEA) confirmed the involvement of microglia (PCDH9high) in axon guidance and endocytosis pathways (Fig. 4e). SPP1, defined as a molecular signature in axonal tract-associated microglia21,44, exhibits notable expression level in microglia (PCDH9high), suggesting that they may congregate around axon tracts. Surprisingly, myelin proteins (MBP, MAG and MOG) were detected in microglia (PCDH9high) (Extended Data Fig. 5a). Multiple immunostaining showed that PCDH9 colocalizes with the typical microglial activation gene SPP1 in the prefrontal cortex and hippocampus region of healthy adult human brains (Extended Data Fig. 5b). Additionally, immunostaining showed that PCDH9+SPP1+ microglia are intermingled with myelin-associated glycoprotein (MAG) in the prefrontal cortex and hippocampus region (Fig. 4f,g and Extended Data Fig. 5b), suggesting that microglia (PCDH9high) may engulf myelin debris, potentially contributing to the maintenance of physiological axon myelination.As SPP1 is known to be related to disease-associated microglia (DAM)22,45, we investigated the relationship between this microglia (PCDH9high) population and DAM. Although microglia (PCDH9high) cells express high SPP1, they show different expression patterns for the DAM activation genes (TREM2, APOE, TYROBP, CST7 and LPL). Activated microglia (PCDH9high) cells exhibit elevated expression of lysosomal-associated genes, along with phagocytic phenotypes (Extended Data Fig. 4e,f), but demonstrate a limited association with DAM signatures (Extended Data Fig. 4g). Taken together, microglia (PCDH9high) might concentrate around the axon, correlating with immune cell activation, lysosomal activity and phagocytic processes.Regional microenvironment drives microglial heterogeneityThe same cell population may demonstrate different gene regulatory patterns in different microenvironments, and understanding such a niche difference may help the development of in vitro cell culture protocols or technologies46. Single research group datasets may be limited in size or confounded in experimental design, hindering the understanding of microenvironment-related differences, especially cross-brain region effects. In large-scale atlas data, a gene covarying exclusively with brain regions, not sequencing batches or studies, is likely to be region specific rather than batch specific (Supplementary Fig. 10). Under this assumption, we performed differential expression analysis for the above-mentioned microglia (PCDH9high) population across two brain regions, prefrontal cortex and hippocampus (Fig. 5a and Supplementary Table 4).Fig. 5: Regional transcriptional identities of microglia (PCDH9high).a, PCA plot showing microglia (PCDH9high) cells according to the hippocampus and prefrontal cortex. The cells were colored by brain region. b, Volcano plot showing that there were 1,469 DEGs between hippocampus and prefrontal cortex (adjusted P value <0.05). The horizontal dashed line is the cutoff of the P value. The vertical dashed lines are the cutoff of the logFC. c, Bar plot showing enriched GO terms in the microglia (PCDH9high) cells from hippocampus and prefrontal cortex. d, KEGG pathway enrichment analysis of microglia (PCDH9high) cells from hippocampus and prefrontal cortex. e, Common and specific transcriptional features of microglia (PCDH9high) in the hippocampus and prefrontal cortex. f, GSEA result of gene sets from hippocampal microglia (PCDH9high). g, GSEA result of gene sets from microglia (PCDH9high) in the prefrontal cortex. The analysis in c and d was based on clusterProfiler, using a two-sided hypergeometric test with Benjamini–Hochberg correction for multiple comparisons.Source dataRegional characteristics of the microglia (PCDH9high) show that complement components (C1QA and C1QC) are highly expressed in the prefrontal cortex, while microglial cell activation gene DLG1 (ref. 47) is highly expressed in the hippocampal region (Fig. 5b). Enriched GO analysis revealed advanced phagocytic scavenging capacity (C3 and PLCG2), antigen presentation (major histocompatibility complex (MHC) class II genes) and immune response (CD74, TLR2 and SYK) in microglia (PCDH9high) of the prefrontal cortex, while hippocampal microglia (PCDH9high) exhibits associations with the regulation of excitatory synapse plasticity (NRXN1, GRIK2 and HOMER1) (Fig. 5c). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis validates the biological process of microglia (PCDH9high) regional heterogeneity (Fig. 5d), with genes involved in phagocytosis and modulating synaptic plasticity enriched specifically in the prefrontal cortex and hippocampus regions, respectively (Fig. 5e and Supplementary Table 5). Consistently aligning with the GO term, KEGG enrichment and GSEA (Fig. 5f), hippocampal microglia (PCDH9high) exhibit a high-level expression of glutamate receptors (GRIA2, GRIK2 and GRIK3), suggesting its potential involvement in bidirectional interactions with excitatory neurons (Extended Data Fig. 6a,b). Compared to the hippocampus, these analytical outcomes substantiate a heightened proinflammatory and phagocytic state of microglia (PCDH9high) in the prefrontal cortex (Fig. 5g). Pearson correlation demonstrates a positive trend between the inflammatory cytokines tumor necrosis factor (TNF), interleukin-1α (IL1A) and glutamate ionotropic/metabotropic receptors (excluding GRIA3) (Extended Data Fig. 6c), indicating a potential regulation by glutamatergic neurons on the activation, phagocytic activity and phenotypic differentiation of microglia (PCDH9high).To further elucidate the bidirectional neuronal–microglial (PCDH9high) communication, we employed the CellChat program48 to explore potential ligand–receptor interactions (Fig. 6a–d and Supplementary Table 6). Overall, 60 pathways (980 genes) were involved in building the cell–cell communication network of the neural cell niches, including 45 conserved pathways, 12 prefrontal cortex-specific pathways and 3 hippocampal-specific pathways (Fig. 6e). As shown in Fig. 6f and Extended Data Fig. 7a,b, the distribution of cells in two-dimensional (2D) space shows changes in the interaction strength of outgoing and incoming signaling between the microglia (PCDH9high) cells in prefrontal cortex and hippocampus. Furthermore, the hippocampal microglia (PCDH9high) shows distinctive signaling alterations, characterized by the specific changes in the neuregulin (NRG), cell adhesion molecule (CADM), neuronal growth regulator (NEGR) and laminin pathways (Fig. 6g).Fig. 6: Cell–cell communication of the neurogenic niches in the prefrontal cortex and hippocampus.a, Circle plot showing the number of interactions and the strength of interactions among different cell types. The red (blue) colored edges represent increased (decreased) signaling in the hippocampus compared to the prefrontal cortex region. b, Circle plot showing the differential interaction strength among different cell types. The red (blue) colored edges represent increased (decreased) interaction strength in the hippocampus compared to the prefrontal cortex region. c, Circle plot displaying the number of interactions and the strength of interactions between any two cell groups in the prefrontal cortex region. The number of lines represents the number of interactions, and the thickness of the lines is proportional to the strength of the interactions. d, Circle plot displaying the number of interactions in the hippocampus. The number of lines represents the number of interactions, and the thickness of the lines is proportional to the strength of the interactions. e, Stacked bar plot showing the overall information flow of each signaling pathway. The vertical dashed line indicates the position where the sample accounts for 50% of the overall information flow. f, Scatter plot showing dominant senders and receivers in a 2D space, showing the prefrontal cortex (left) and the hippocampus region (right). g, Scatter plot demonstrating the signaling changes associated with microglia (PCDH9high) cell groups in the prefrontal cortex and hippocampus. h, Dot plot displaying the expression of significant ligand–receptor pairs in the NRG, CADM, NEGR and laminin pathways from all senders to hippocampal microglia (PCDH9high). P values are computed from a one-sided permutation test according to CellChat. i, Dot plot showing the expression of significant ligand–receptor pairs in the NRG, CADM, NEGR and laminin pathways from hippocampal microglia (PCDH9high) to cell receivers. P values are computed from a one-sided permutation test according to CellChat. Commun.Prob., communication probability.Source dataIn the hippocampus, ten cell senders, which secrete ligands, interact with the microglia (PCDH9high) cell population via the NRG, CADM, NEGR and laminin pathways mediated by multiple ligand–receptor pairs (Fig. 6h). Nine cell receivers interact with microglia (PCDH9high) when it acts as a signal sender: oligodendrocyte precursor cells, oligodendrocytes, NFOLs, astrocytes, vascular leptomeningeal cells, fibroblasts, endothelial cells, GABAergic neurons and glutamatergic neurons (Fig. 6i and Extended Data Fig. 7c). Although the functional role of most ligand–receptor pairs in microglia remains elusive, some ligand–receptor pairs expressed on hippocampus-enriched cell types (Extended Data Fig. 7d). LAMB1–ITGB8 may serve as a specific ligand–receptor pair for glutamatergic neuronal-to-microglial (PCDH9high) communication, whereas LAMA1–SV2B, LAMA2–(ITGAV + ITGB8) and LAMA2–(ITGA7 + ITGB1) might function as specific ligand–receptor pairs for microglial (PCDH9high)-to-neuronal communication. These differential ligand–receptor pairs suggest that the microglia (PCDH9high) population selectively prunes glutamatergic neurons. Collectively, we present a cell–cell communication network in the prefrontal cortex and hippocampus, enhancing the understanding of the neuronal–microglial crosstalk pathways.
A brain cell atlas integrating single-cell transcriptomes across human brain regions
