Single-Cell Approaches Define the Murine Leptomeninges: Cortical Brain Interface as a Distinct Cellular Neighborhood Composed of Neural and Non-neural Cell Types

scRNA-seq identifies cell types within the meningeal:brain interface of the adult murine cerebral cortex

To ask about cells that comprise the cortical brain surface interface, we initially performed single-cell RNA sequencing (scRNA-seq) and used two different dissection protocols to obtain a global overview of the cells at this interface (see Materials and Methods). In one approach, we isolated cortical surface-associated cells, including the leptomeninges, by dissecting the top of the cortex, adhering the ventral surface to a coverslip, and then liberating single cells enzymatically (Extended Data Fig. 1-1A). In the second approach, we enriched for bone-associated cells such as the dura by removing the skull, dissecting the associated connective tissue, and enzymatically dissociating single cells. For both dissections, we used FACS to select for viable cells and then isolated and sequenced single cells using the 10x Genomics Chromium platform. In total, we performed four separate runs for the cortical surface and two runs for bone-associated cells, using 6–10 mice per run.

The resultant transcriptomes were analyzed using a previously described pipeline (see Materials and Methods for details). After filtering the datasets for low-quality cells with few expressed genes or high mitochondrial proportions and cell doublets, we obtained 810–2,709 single-cell transcriptomes in each of the six independent datasets (9,436 cells total). We then merged the transcriptomes from the bone-associated datasets and those from the cortical surface datasets separately. We used genes with high variance to compute principal components as inputs for projecting cells in two dimensions using UMAPs, and clustering was performed using a shared nearest neighbors-cliq (SNN-cliq)-inspired approach built into the Seurat R package at a range of resolutions (Hao et al., 2024). Since there was some segregation between the four different cortical surface datasets, we normalized the merged dataset using one iteration of Harmony batch correction (Korsunsky et al., 2019; Extended Data Fig. 1-1B). We then used gene expression overlays for well-validated marker genes to define cell types.

In both preparations, this analysis (Fig. 1A,B; Extended Data Fig. 1-1B–E) identified Pdgfra-positive mesenchymal cells and vasculature-associated cells including endothelial cells, pericytes, and VSMCs. We also identified peripheral immune populations including macrophages, T cells, and B cells. Since mice were not perfused prior to single-cell isolation, these could be either blood-borne or tissue-localized. Some cell types were specific to one or the other isolation. For the cortical surface analysis (Fig. 1A; Extended Data Fig. 1-1A–C), we identified CNS cell types including glial cells, microglia, a small number of neurons, and choroid plexus cells. In contrast, the bone-associated preparation (Fig. 1B; Extended Data Fig. 1-1D,E) did not include CNS cell types or microglia but did include Schwann cells derived from peripheral cranial nerves (Carr et al., 2019; Bastedo et al., 2024).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

scRNA-seq identifies distinct populations of cortical brain interface versus bone-associated cells. Also see Extended Data Figures 1-1, 1-2, and 1-3. A, B, UMAP visualization of all transcriptomes identified in the cortical surface-associated (CSA; A) or cortical bone-associated (BA; B) datasets as analyzed by scRNA-seq. Shown are the mergers of four independent surface-associated runs (A) and two independent bone-associated runs (B). The merged CSA dataset in (A) was batch corrected using one iteration of Harmony. Transcriptionally distinct clusters are color coded, and plots are annotated for cell types corresponding to these clusters, as determined by analysis of well-characterized marker genes. VSMCs, vascular smooth muscle cells; Lepto, leptomeningeal cells; Dural, dural cells. C, D, Violin plots showing expression of Pdgfra, Pdgfrb, and Col1a1 in leptomeningeal (C; colored blue) and dural (D; colored blue) cells relative to the Pdgfra-negative pericytes (colored red) and VSMCs (colored green). Dots represent expression levels in individual cells. E, F, Pdgfra-positive cell transcriptomes were subsetted from the surface-associated and bone-associated datasets shown in A and B and merged together. Panel E shows the cluster UMAP, where transcriptionally distinct clusters are colored and numbered and F shows the dataset of origin for each transcriptome. CSA, cortical surface associated; BA, bone associated. G, Gene ontology was performed on the differentially expressed genes identified in the comparison between leptomeningeal and dural cells as defined in F (>1.3-fold average change, padj < 0.05; Extended Data Figs. 1-2 and 1-3). Shown are selected categories (y-axis) and the adjusted p value (x-axis). The peach-colored bars are categories enriched in leptomeningeal cells and the turquoise enriched in dural cells. H, Single-cell heatmap of select mRNAs that were differentially expressed (Extended Data Fig. 1-2) in the comparison between the leptomeningeal cells in cluster 0 and the dural cells in cluster 1 from the UMAPs in E and F. Every row represents expression in an individual cell, and levels are color coded as per the adjacent key. I, Violin plots showing relative expression levels of two mRNAs, Ptgds and Igfbp7, that were identified in the differential gene expression analysis as highly enriched in leptomeningeal (colored red) versus dural (colored turquoise) cells, respectively. Dots represent expression levels in individual cells. J, Transcriptional signatures for the putative leptomeningeal and dural cells were defined using the 50 most differentially expressed mRNAs in the leptomeningeal versus dural cell comparison (Extended Data Fig. 1-2). Shown is a UMAP as in E highlighting cells expressing a transcriptional signature score of >0.5 for one or the other cell type.

We observed several distinct types of Pdgfrb-positive cells within these datasets. In both cortical surface- and bone-associated datasets, there were Pdgfrb-positive pericytes and VSMCs that were negative for Pdgfra and positive for either the pericyte markers Rgs5 and Kcnj8 or the VSMC markers Myh11 and Acta2 (Fig. 1C,D; Extended Data Fig. 1-1F,G). These vasculature mural cells are likely associated with meningeal blood vessels and parenchymal vasculature (Vanlandewijck et al., 2018). More importantly for our analysis, we also identified cells positive for Pdgfrb, Pdgfra, and Col1a1 that included putative dural meningeal cells in the case of the bone-associated dataset and leptomeningeal cells in the case of the cortical surface-associated dataset (Fig. 1C,D; Extended Data Fig. 1-1F,G). Since there is increasing evidence that the family of cells expressing these mRNAs, which includes fibroblasts, are heterogeneous (Carr et al., 2019; Storer et al., 2020; Li et al., 2024), then from hereon we have called these dural cells, leptomeningeal cells, or when their precise identity or location is unknown, Pdgfra-positive mesenchymal cells. Thus, our findings validated the dissections and identified likely meningeal mesenchymal cells (Vanlandewijck et al., 2018; DeSisto et al., 2020; Pietilä et al., 2023).

Dural meninges and leptomeninges have distinct transcriptional signatures

We further analyzed the potential meningeal cells by subsetting, merging, and reanalyzing Pdgfra-positive cells from the bone-associated and cortical surface-associated datasets. We performed one iteration of Harmony batch correction to ensure integration without overcorrection (Korsunsky et al., 2019). UMAP-based cluster visualization (Fig. 1E) and analysis of the dataset of origin (Fig. 1F) identified two major clusters of Pdgfra-positive mesenchymal cells. Cluster 1 included cells almost exclusively from the bone-associated (BA) dataset, while cluster 0 included cells predominantly from the cortical surface-associated (CSA) dataset. We interpret these as bone-associated dural cells and cortical surface-associated leptomeningeal cells, respectively.

We directly compared these two populations using several approaches. First, we performed Pearson’s correlation analysis of averaged gene expression. The two mesenchymal cell populations were quite distinct with r = 0.844 (Extended Data Fig. 1-1H). Second, we performed differential gene expression analysis (Extended Data Fig. 1-2) and identified 1,239 differentially expressed genes (average log2 fold change ≥1.3, adj p value <0.5, expression in at least 10% of cells). A total of 872 genes were enriched in the putative dural cells and 367 in the putative leptomeningeal cells. GO analysis (Kolberg et al., 2023; Fig. 1G; Extended Data Fig. 1-3) showed that the dural cells were enriched for terms like extracellular matrix organization, skeletal system development, and ossification, consistent with their role as skull stromal cells. In contrast, the leptomeningeal cells were enriched for terms like active transmembrane transporter activity, cell junction organization, and cell–cell signaling, consistent with their role as brain interface cells. Heatmaps and violin plots (Fig. 1H,I; Extended Data Fig. 1-1I) confirmed enrichment for these mRNAs. The leptomeningeal cells were highly enriched for expression of the transmembrane transporter Slc22a6, the basal lamina component Lama1, the connexins Gjb2 and Gjb6, and the prostaglandin D2 synthase Ptgds, which is associated with cerebral spinal fluid leak (Kondabolu et al., 2011). Conversely, the dural cells were enriched for Igfbp7, Clec11a, Alpl, Ostn, H19, and Mgp, all of which are implicated in osteochondrogenesis (Zhang et al., 2018; Sato et al., 2021; Liu et al., 2023). We used the top 50 differentially expressed dural versus leptomeningeal genes to define transcriptional signatures that readily distinguished these two populations (Extended Data Fig. 1-2). The leptomeningeal signature (Fig. 1J, blue) included genes such as Ptgds, Slc22a6, Gjb6, Gjb2, Bmp6, and Lama1, while the dural signature (Fig. 1J, red) included genes such as Matn4, Igfbp7, H19, Clec11a, and Smoc2. Thus, both dura and leptomeninges are composed of Pdgfra-positive mesenchymal cells, but the dural cells are more analogous to bone stromal cells while the leptomeningeal cells have a transcriptional profile consistent with a brain interface role.

Brain interface astrocytes are transcriptionally distinct from cortical gray matter astrocytes

The brain interface is composed of leptomeningeal cells and closely associated CNS astrocytes. We therefore asked if the brain interface-associated astrocytes were also a transcriptionally distinct population using scRNA-seq. To do so, we examined the cortical surface scRNA-seq dataset. Since this contained relatively few astrocyte transcriptomes, we merged it with several additional astrocyte scRNA-seq datasets. First, we extracted cortical astrocyte transcriptomes from two recently published scRNA-seq datasets, one including total postnatal day 60 (P60) adult cortex cells and the other P60 cortical gray matter cells (GEO GSM8072112; GSM8072111; Dennis et al., 2024; Extended Data Fig. 2-1A,B). Second, we took advantage of the recent finding that astrocytes of the glia limitans superficialis are enriched for Myoc and Gfap mRNAs (Hasel et al., 2025). We therefore subsetted and included Myoc-expressing, Gfap-enriched astrocytes from a previously published total brain astrocyte scRNA-seq dataset (Hasel et al., 2021; GEO GSE148611; Extended Data Fig. 2-1C,D). We merged all of these astrocyte transcriptomes with the astrocyte transcriptomes we had obtained in the cortical surface analysis (the dataset in Fig. 1A) and performed batch correction using Harmony.

This analysis (Fig. 2A,B) identified four astrocyte clusters in three distinct groups. One group of clusters (0 and 3) likely include cortical white matter astrocytes, since analysis of the dataset of origin showed they were almost completely composed of astrocytes from the total cortex dataset and did not include gray matter transcriptomes. The gray matter transcriptomes were instead in cluster 1, which also included transcriptomes from the total cortex and thus likely includes cortical parenchymal astrocytes. The final cluster (2) was largely composed of Myoc-enriched whole-brain astrocytes together with cells from the cortical surface dataset. All cells expressed pan-astrocyte markers such as Slc1a3 (GLAST) and Cxcl14, and, as predicted, cluster 2 was enriched for Myoc and Gfap (Fig. 2C; Batiuk et al., 2020). From hereon, we have called these cluster 2 cells putative border astrocytes, since Myoc and Gfap are enriched in astrocytes at the brain surface.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Identification of transcriptionally distinct border astrocytes by scRNA-seq analysis. Also see Extended Data Figures 2-1, 2-2, and 2-3. A, B, Astrocyte transcriptomes were subsetted from four different murine scRNA-seq datasets: cortical surface-associated cells as in Figure 1A, P60 whole cortex and P60 cortical gray matter from Dennis et al. (2024) (GEO GSE255405; Extended Data Fig. 2-1A,B), and whole-brain astrocytes from Hasel et al. (2021) (GEO GSE148611). In the latter case, only astrocytes enriched for Myoc and Gfap were extracted (Extended Data Fig. 2-1C,D). These astrocyte transcriptomes were merged and batch corrected using one iteration of Harmony. The resultant merged dataset is shown as UMAP visualizations with the transcriptionally distinct clusters numbered and color coded in A and datasets of origin shown in B. C, UMAPs as in A and B overlaid for expression of two pan-astrocyte mRNAs, Slc1a3 and Cxcl14, and two brain interface-enriched mRNAs, Myoc and Gfap. The putative border astrocytes are circled. Expression levels are color coded as per the adjacent keys. D, Single-cell heatmap of select mRNAs that were differentially expressed (Extended Data Fig. 2-2) in the comparison between the putative border astrocytes (cluster 2 in panel A) and cortical gray matter astrocytes (cluster 1 in panel A). Every row represents expression in an individual cell, and levels are color coded as per the adjacent key. E, Gene ontology was performed on the differentially expressed genes identified in the comparison between border astrocytes and cortical gray matter astrocytes (>1.3-fold average change, p adj < 0.05; Extended Data Figs. 2-2, 2-3). Shown are selected categories (y-axis) and the adjusted p value (x-axis). F, A transcriptional signature for the putative border astrocytes (cluster 2 in A) was defined using the 20 most differentially expressed mRNAs in the border versus cortical gray matter astrocyte comparison (Extended Data Fig. 2-2). Shown is a UMAP as in A with cells expressing a border astrocyte transcriptional signature score >0.5 highlighted in purple.

To better understand the putative cluster 2 border astrocytes, we directly compared them with cluster 1 cortical parenchymal/gray matter astrocytes using differential gene expression analysis. A total of 575 mRNAs differed significantly between the two groups (average log2 fold change ≥1.3, adj p value <0.5, expression in at least 10% of cells; Extended Data Fig. 2-2). Cluster 2 border astrocytes were enriched for 381 mRNAs including Myoc, Gfap, Fbxo2, Id1, Vim, and Agt, as illustrated in a heatmap (Fig. 2D; Extended Data Fig. 2-2). Gene ontology analysis of these differentially enriched border astrocyte genes (Extended Data Fig. 2-3) identified terms that included cellular detoxification and astrocyte end foot (Fig. 2E). In contrast, cluster 1 gray matter astrocytes were enriched for 194 mRNAs, including Gria2, Igfbp2, and Slc1a3 (Fig. 2D; Extended Data Fig. 2-2) and gene ontology identified terms such as postsynapse and amino acid transmembrane transporter (Fig. 2E; Extended Data Fig. 2-3). We used the 20 most differentially expressed genes to define a specific transcriptional signature for the putative Myoc-positive border astrocytes that included Myoc, Gfap, Fbxo2, Slc38a1, and Fxyd6 (Fig. 2F; Extended Data Fig. 2-2).

Visualization of leptomeningeal:astrocyte interactions at the cortical brain interface

One caveat of the scRNA-seq data is that it does not provide spatial information about the interface cells. We therefore performed two complementary morphological studies, lineage tracing plus immunostaining and single-cell spatial transcriptomics. For the lineage tracing analysis, we initially used a well-characterized mouse line where CreERT2 is driven from the endogenous Pdgfra locus (Chung et al., 2018). We crossed these to another mouse line carrying a floxed TdTomato transgene in the Rosa26 locus (Madisen et al., 2010); when these crossed Pdgfra-CreERT2;R26RtdT mice are exposed to tamoxifen, TdTomato will be expressed in and specifically tag Pdgfra-positive mesenchymal cells. Pdgfra-positive OPCs will also be genetically tagged, but these are readily distinguished by their location, morphology, and transcriptional profiles. We therefore exposed crossed mice to tamoxifen for 5 d, isolated the cortex and overlying skull 10 d later, decalcified the skull, and analyzed sagittal sections from layer 1 through to the top of the skull (Fig. 3A). This analysis (Fig. 3B) showed that as predicted both the dura and the leptomeninges were TdTomato-positive and that the leptomeninges were only two to three layers thick in this region. Moreover, immunostaining for GFAP (Fig. 3B,C) confirmed that it was highly enriched in the border astrocytes (Fig. 2C) and demonstrated that the TdTomato-positive leptomeningeal cells were immediately adjacent to GFAP-positive border astrocytes. Similar findings were made by immunostaining sections for PDGFRα protein and GFAP (Extended Data Fig. 3-1A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Lineage tracing and immunostaining identify the leptomeninges and the dura at the cortical brain interface. Also see Extended Data Figure 3-1. A, Schematic showing the region of the cortex and overlying skull that were analyzed in the lineage tracing and immunostaining studies. B, Ten-week-old Pdgfra-CreERT2;R26RtdT mice were treated with tamoxifen and tissues collected 10 d after the final administration. The brain and skull were decalcified, and sagittal cortical sections were immunostained for GFAP (white) and counterstained with DAPI (blue). Shown is a representative confocal image of the brain interface and skull showing TdTomato (red), GFAP (white), and DAPI (blue). The boxed region is shown at higher magnification to the right without the DAPI. Lepto, leptomeninges; Astro, border astrocytes. C, Representative confocal image of a sagittal wild-type mouse brain interface section prepared as in B, immunostained for GFAP (white) and counterstained with DAPI (blue). Lepto, leptomeninges. D, Violin plots showing the relative expression of Gjb6, Gjb2, Lama1, Tmem119, and Matn4 in the leptomeningeal (colored red) versus dural (colored turquoise) cells shown in Figure 1E,F. Dots represent individual transcriptomes. E, Ten-week-old Tmem119CreErt2;R26RtdT mice were treated with tamoxifen and tissues collected 10 d later. The brain and skull were decalcified and sagittal cortical sections immunostained for GFAP (white) and counterstained with DAPI (blue). The left panel shows a low magnification image of the skull, dura, and cortical interface showing TdTomato (red) and DAPI (blue). The space between the leptomeninges and dura is a processing artifact. The two right panels are high magnification confocal images of the same section showing TdTomato (red) and GFAP immunostaining (white) in the dura and skull (top right) and the brain interface (bottom right). Both top and bottom right panels come from the same field of view. Lepto, leptomeninges; Astro, border astrocytes. F, Representative images of sagittal sections through the decalcified skull and cortex, immunostained for GFAP (white) and MATN4 (yellow) and counterstained with DAPI (blue). The left panel shows a low magnification image of the skull, dura and cortical interface. The space between the leptomeninges and dura is a processing artifact. The two right panels are high magnification confocal images of a similar section from the same brain showing MATN4 (yellow) and GFAP (white) immunostaining in the dura and skull (top right) and the brain interface including the leptomeninges (bottom right). Both top and bottom right panels come from the same field of view. Lepto, leptomeninges; Astro, border astrocytes. G, High magnification confocal image of a sagittal adult cortical interface section immunostained for GFAP (white) and Laminin (yellow) and counterstained with DAPI. Note that the Laminin immunoreactivity is localized between the border astrocytes and the leptomeningeal cells. Lepto, leptomeninges; Astro, border astrocytes. H, High magnification confocal image of a sagittal adult cortical interface immunostained for GFAP (white) and CX26 (yellow) and counterstained with DAPI. The hatched line delineates the border between the leptomeninges and the dura. Lepto, leptomeninges; Astro, border astrocytes. I, High magnification confocal image of a sagittal adult cortical interface section immunostained for GFAP (white) and CX30 (yellow) and counterstained with DAPI. Lepto, leptomeninges; Astro, border astrocytes. For all panels, scale bars are 10 µm except for the leftmost images in E and F, where scale bars are 100 μm.

We performed a similar analysis with a Tmem119-CreERT2;R26RtdT mouse line that has been widely used to label microglia (Kaiser and Feng, 2019), based upon our finding that Tmem119 is expressed in both the dura and the leptomeninges (Fig. 3D). As predicted, following tamoxifen treatment both the dura and leptomeninges were TdTomato-positive in the Tmem119-CreERT2;R26RtdT mice as were microglia located in the cortical parenchyma (Fig. 3E). As seen with the Pdgfra-CreERT2-based lineage tracing, the leptomeninges were only several cell layers thick at this point above the cortex. Moreover, immunostaining for GFAP showed that the GFAP-positive border astrocytes were closely juxtaposed to the TdTomato-positive leptomeningeal cells (Fig. 3E).

Both the dura and leptomeninges express Pdgfra and Tmem119, and thus these lineage tracing approaches do not distinguish the two compartments. We therefore immunostained for proteins predicted to be differentially expressed in the transcriptional data. For the dura, we immunostained decalcified sagittal cortical sections that included the skull for MATN4, which is highly enriched in the putative dural cells (Fig. 3D). For the leptomeninges, we immunostained similar sections for Laminin (Lama1), or the gap junction proteins Connexin 26 (Gjb2) or Connexin 30 (Gjb6), all of which are highly enriched in the leptomeningeal cells (Fig. 3D; DeSisto et al., 2020; Pietilä et al., 2023). Of these, Gjb2 is also expressed at low levels in border astrocytes and Gjb6 is widespread in astrocytes (Extended Data Fig. 3-1B–D). In all cases, we double-labeled sections for GFAP to visualize the border astrocytes. As predicted, immunoreactivity for the extracellular matrix protein MATN4 was not detectable in the leptomeninges but was observed in the dura (Fig. 3F). In contrast, Laminin, CX26/Gjb2, and CX30/Gjb6 were detectable in the leptomeninges but not the dura (Fig. 3G–I). Laminin immunoreactivity was specifically associated with the basal lamina between the GFAP-positive border astrocytes and leptomeningeal cells, while CX26 and CX30 immunoreactivity were present in a punctate pattern in the leptomeningeal cells immediately adjacent to GFAP-positive border astrocytes. As predicted (Extended Data Fig. 3-1B–D), CX30 mRNA was also detected in border astrocytes, as was CX26 at apparently lower levels (Lynn et al., 2011; Nagy et al., 2011; Huang et al., 2021). These studies thus identify lineage tracers for the leptomeninges, validate our findings from the scRNA-seq datasets, and highlight the close interactions between border astrocytes and leptomeningeal cells at the cortical brain interface.

Single-cell spatial transcriptomics defines the brain interface as closely associated leptomeningeal cells, border astrocytes, and border macrophages

The lineage tracing and immunostaining studies are limited by their ability to only analyze several marker genes/proteins at a time. To obtain a more global overview of the cortical brain interface, we performed single-cell multiplexed in situ gene expression analysis using the Xenium platform (Janesick et al., 2023; Willis et al., 2025). We did this with two different probesets, one a custom 480 gene probeset optimized to distinguish different mesenchymal cell populations, including those of the meninges, and the second a standard Xenium brain probeset with a custom 100 gene add-on composed of probes allowing better resolution of non-neural brain cell types (see Materials and Methods; Extended Data Fig. 4-3).

We used these probesets to analyze the leptomeninges and the adjacent cortical layer 1 in coronal forebrain sections from freshly frozen adult murine brain tissue (Fig. 4A; Extended Data Fig. 4-1A,B). For sections analyzed using the mesenchymal probeset, we also included the dura. We then performed Xenium-based multiplexed in situ gene expression analysis as described previously (Dennis et al., 2024; Willis et al., 2025) using one of the two probesets. For the mesenchymal probeset, we analyzed six coronal rostral forebrain sections at the level of the lateral ventricles from four different mice (Extended Data Fig. 4-1A). For the brain probeset, we ran two rostral coronal forebrain sections, one each from two different adult mouse brains (as in Extended Data Fig. 4-1A). We augmented this with a recently published Xenium dataset obtained using the same brain probeset to analyze three somewhat more caudal coronal sections of the adult cortex from three different mice (Willis et al., 2025; GEO GSE266689, sections GSM8647390, GSM8647391, GSM8647392; Extended Data Fig. 4-1B). In all cases we defined an ROI that spanned the midline and included cortical layer 1, the top of adjacent layer 2 plus the overlying leptomeninges (Fig. 4A). After initial processing of data from individual sections, we excluded cells where the detected genes and total transcript counts per cell were ±2.5 or more standard deviations from the mean in either parameter since these were likely cellular fragments or cellular doublets. We then merged cells from different sections/animals that had been analyzed using the same probeset.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Single-cell multiplexed in situ gene expression analysis with two different probesets identifies the spatial location of cortical interface and layer 1 cell types. Also see Extended Data Figures 4-1, 4-2, and 4-3. Coronal adult mouse cortex sections at the levels shown in Extended Data Fig. 4-1A,B were analyzed by Xenium-based single-cell multiplexed in situ gene expression analysis with either a brain-targeted probeset targeting 347 genes or a mesenchymal probeset targeting 480 genes (Extended Data Fig. 4-3). The ROI that was analyzed is shown in A. Datasets from different sections and conditions were then merged and cell types identified by marker gene expression. A, Schematic of a coronal section through the adult cortex in the midline region showing the ROI that was analyzed (outlined with hatched black lines). B, C, UMAP cluster visualization of the merged transcriptomes resulting from analysis with the brain probeset (B) or the mesenchymal probeset (C), annotated for cell types. Each dot represents a single cell. VSMC, vascular smooth muscle cell; Oligos, oligodendrocytes; SST, somatostatin; Pvalb, parvalbumin; VIP, vasoactive intestinal peptide. D, E, Spatial plots of the midline and adjacent cortical interface and layer 1 from the ROI of representative sections analyzed using the brain (D) or mesenchymal (E) probesets, showing all cell types except neurons, color coded as per the legend. The boxed regions are also shown in enlarged views. L1 denotes cortical layer 1. Lepto, leptomeninges; B. Ast, border astrocytes. F, G, Spatial plots as in D and E from the same representative sections analyzed using the brain (F) or mesenchymal (G) probesets showing the leptomeninges and all of the neurons that were identified, color coded. L1 denotes cortical layer 1. Low magnification spatial plots in D–G were generated using Seurat with D and F showing the same ROI and E and G the same ROI. Each dot represents the centroid of one cell.

With the mesenchymal probeset, we ultimately obtained 29,644 total cellular transcriptomes from the merged ROIs of six sections/four brains and with the brain probeset, we obtained 20,371 transcriptomes from merged ROIs of one section each from five brains. With both probesets, transcriptomes from the different sections/brains were well integrated, as shown on UMAPs (Fig. 4B,C; Extended Data Fig. 4-1C; Extended Data Fig. 4-2A). With two exceptions, both probesets defined the same cell types, as determined by analysis of a panel of marker genes (see Materials and Methods; Fig. 4B,C; Extended Data Fig. 4-1C,D; Extended Data Fig. 4-2A,B). The exceptions were dural cells, which as predicted were only present in the mesenchymal probeset sections, and interneurons (discussed further below).

Spatial plots identified the locations of the different identified cell types, with similar spatial plots obtained on tissue sections from different mice (Fig. 4D–G; Extended Data Fig. 4-2C). The brain interface was composed of leptomeningeal cells, macrophages, and border astrocytes that were tightly packed and interspersed with vasculature mural cells and endothelial cells that were associated with penetrating blood vessels. Layer 1 was instead relatively cell sparse, and included astrocytes, oligodendrocytes, OPCs, and vasculature-associated cells. As predicted, the only neurons in layer 1 were interneurons (Fig. 4F,G; Extended Data Fig. 4-2C). The brain probeset defined these as expressing Lamp5 or Vip (Fig. 4B,F; Extended Data Fig. 4-2C), consistent with previous immunocytochemical analyses (Schuman et al., 2019; 2021). The mesenchymal probeset instead defined two interneuron clusters that differed in their spatial locations (from hereon called upper or lower interneurons; Fig. 4C,G). The border of layer 1 was clearly delineated by layer 2 excitatory neurons and a few Somatostatin- or Parvalbumin-expressing interneurons that were included in the ROI (Fig. 4F,G; Extended Data Fig. 4-2C), again as previously reported (Tremblay et al., 2016; Lim et al., 2018; Yao et al., 2021).

We used these datasets to analyze the brain interface and layer 1 cellular neighborhoods in greater detail. Consistent with the immunostaining analyses (Fig. 3), the leptomeninges at the brain interface were a continuous layer of cells that were lined by border astrocytes on the CNS side (Fig. 5A). Resident macrophages were scattered along this interface. These are likely analogous to the previously described border macrophages (Jordão et al., 2019; Kolabas et al., 2023; Pietilä et al., 2023). The dural cells were located on the skull side of these interface cells (Fig. 5B), as predicted. They were transcriptionally distinct from the leptomeningeal cells, expressing mRNAs such as Matn4 and H19 but not the leptomeningeal mRNA Gjb2 (Fig. 5C,D).

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Single-cell spatial transcriptomic analysis defines the cortical brain interface structure and cell types. Also see Extended Data Figure 5-1. Coronal adult mouse cortex sections at the levels shown in Extended Data Fig. 4-1A,B were analyzed by Xenium-based single-cell multiplexed in situ gene expression analysis with either a brain probeset targeting 347 genes or a mesenchymal probeset targeting 480 genes (Extended Data Fig. 4-3). The ROI and UMAPs of the resultant merged datasets are shown in Figure 4A–C. A, Spatial plots of the cortical interface analyzed with the mesenchymal (MP, left) or brain (BP, right) probesets showing leptomeningeal cells (pink), border astrocytes (yellow), border macrophages (dark blue), and dural cells (green, mesenchymal probeset only). B, Spatial plot of the cortical interface and midline analyzed with the mesenchymal probeset showing dural (green) and leptomeningeal (pink) cells. C, High-resolution Xenium Explorer image of the cortical interface region analyzed with the mesenchymal probeset (MP) showing expression of H19 mRNA (orange dots) and Gjb2 mRNA (bright blue dots) relative to the leptomeningeal cells (pink) and dural cells (green). Nuclei of other cell types (white) are also shown. D, Spatial plots of the dural mRNA H19 (top) or the leptomeningeal mRNA Gjb2 (bottom) in all cells at the cortical interface, midline, and layer 1 analyzed using the mesenchymal probeset (MP). Relative mRNA expression levels are coded as per the adjacent keys. E, High-resolution Xenium Explorer images of the cortical interface region analyzed with the mesenchymal (MP, top) or brain (BP, bottom) probesets showing the spatial arrangement of leptomeningeal cells (pink), border astrocytes (yellow), border macrophages (blue), and dural cells (green, mesenchymal probeset only). Nuclei are also shown (white/gray). F, Spatial plots of the cortical interface and layer 1 analyzed with the mesenchymal (MP, top) or brain (BP, bottom) probesets showing border astrocytes (yellow) and parenchymal astrocytes (brown). G, Spatial plots of Myoc and Gfap mRNA expression in all cells of the full ROI analyzed using the brain probeset (BP). Relative expression levels are coded as per the adjacent keys. H, High-resolution Xenium Explorer image of the cortical interface analyzed with the mesenchymal (MP, left) or brain (BP, right) probesets showing expression of Myoc mRNA (light blue dots) relative to the leptomeningeal cells (pink) and border astrocytes (yellow). Nuclei of other cell types (white) are also shown. I, Spatial plots of the cortical interface and layer 1 analyzed with the mesenchymal (MP, left) or brain (BP, right) probesets showing border macrophages (blue) and microglia (tan). J, High-resolution Xenium Explorer image of the cortical interface and layer 1 analyzed with the mesenchymal probeset (MP) showing expression of Mrc1 mRNA (bright blue dots) and Sall1 mRNA (bright green dots) relative to the border macrophages (dark blue), microglia (tan), leptomeningeal cells (pink), and dural cells (darker green). Nuclei of other cell types (white) are also shown. Low magnification spatial plots in A, B, D, F, G, and I were generated using Seurat and show part of the ROI centered on the cortical midline region from one representative section each, with each dot representing the centroid of one cell. High-resolution spatial plots in C, E, H, and J were generated using Xenium Explorer. Scale bars, 10 μm.

A higher resolution analysis highlighted the close association between the different brain interface cell types (Fig. 5E). The cortical surface leptomeninges were 2–3 cells wide, with the lowest layer immediately adjacent to the border astrocytes. The border macrophages were interspersed within this leptomeningeal layer. Each of these interface-associated cell types differed transcriptionally from similar cells within other regions of the meningeal/cortical layer 1 tissue. Specifically, the border astrocytes were transcriptionally distinct and spatially segregated from their parenchymal astrocyte counterparts in layer 1 (Fig. 5F) and, consistent with the scRNA-seq, were enriched for Myoc and Gfap (Fig. 5G,H). The border macrophages were also transcriptionally and spatially distinct from the parenchymal microglia (Fig. 5I). The border macrophages were closely associated with the leptomeningeal cells, with the occasional macrophage in the parenchyma where they were associated with vasculature cells and were presumably present in blood vessels.

To confirm these macrophage/microglia cell type assignments, we performed differential gene expression (Extended Data Fig. 5-1) on the macrophages and microglia from the cortical surface-associated scRNA-seq data (Fig. 1A). This analysis showed that surface macrophages versus microglia were enriched for Mrc1 versus Sall1, respectively, as previously reported (Jurga et al., 2020; Paolicelli et al., 2022). Since probes for both of these genes were present in the mesenchymal probeset, we analyzed their expression spatially. This analysis confirmed that the cells identified as border macrophages were positive for Mrc1 while those identified as microglia were specifically positive for Sall1 (Fig. 5J).

Distinguishing the arachnoid and pial leptomeningeal layers

These analyses identified the leptomeninges but did not distinguish arachnoid and pial cells. To do this, we took advantage of two previously published brain surface scRNA-seq datasets characterizing various meningeal populations (Pietilä et al., 2023; GEO GSE227713; see Materials and Methods for details). We ran these datasets through our pipeline, identified Pdgfra-positive mesenchymal cell transcriptomes, and then subsetted and reanalyzed these on their own. This analysis identified several distinct meningeal mesenchymal cell populations as previously described (Extended Data Fig. 6-1A). We merged these with the dural and leptomeningeal cell subset described here (from Fig. 1E) and used two iterations of Harmony to correct for differences in read-depth (Extended Data Fig. 6-1B–D). This analysis identified two leptomeningeal cell clusters (0 and 4) expressing Slc22a6 and Gjb6 and two mRNAs expressed in the developing leptomeninges, Crabp2 and Aldh1a2 (Fig. 6A–C; Extended Data Fig. 6-1E). Both clusters included intermingled cells from the cortical brain surface and Pietilä et al. datasets (Fig. 6A,B). Notably, cluster 0 was enriched for Lama1 and cluster 4 for Ppp1r1a (Fig. 6D), identifying them as potential pial versus arachnoid cells, respectively.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Distinguishing the arachnoid and pial leptomeningeal layers using scRNA-seq and single-cell spatial transcriptomics. Also see Extended Data Figures 6-1 and 6-2. A–D, scRNA-seq-derived mesenchymal cell transcriptomes from Pietilä et al. (2023) (GEO GSE227713; shown in Extended Data Fig. 6-1A) were merged with the meningeal cell transcriptomes from Figure 1E and batch effects minimized using two iterations of Harmony. A UMAP of the total merged, annotated dataset is shown in Extended Data Figure 6-1B. A, B, UMAP visualizations of the leptomeningeal cells from the merged total meningeal cell scRNA-seq dataset (Extended Data Fig. 6-1B) showing the potential pial and arachnoid cell clusters (A) and their datasets of origin (B). C, UMAP as in A, overlaid for two pan-leptomeningeal mRNAs, Slc22a6 and Gjb6. Gene expression levels are color coded as per the adjacent keys. D, Violin plots showing relative expression of two mRNAs, Lama1 and Ppp1r1a, in cluster 0 versus 4 (the potential pial versus arachnoid cells) in the UMAP shown in A. Dots represent individual transcriptomes. E, Spatial expression plots of mRNAs differentially enriched in the cluster 0 putative pial cells from A (also see Extended Data Fig. 6-2), in the leptomeningeal cells identified by Xenium using either the mesenchymal probeset (MP) or the brain probeset (BP; as shown in Fig. 4F,G). Also shown is a similar spatial plot for the pial mRNA Lama1. Relative mRNA expression levels are coded as per the adjacent keys. F, Spatial expression plots of mRNAs differentially enriched in the cluster 4 putative arachnoid cells from A (also see Extended Data Fig. 6-2), in the leptomeningeal cells identified by Xenium using either the mesenchymal probeset (MP) or the brain probeset (BP; as shown in Fig. 4F,G). Also shown is a similar spatial plot for the arachnoid mRNA Ppp1r1a. Relative mRNA expression levels are coded as per the adjacent keys. G, UMAP cluster visualization of the leptomeningeal transcriptomes from the Xenium brain probeset data (Fig. 4B), shown at higher resolution and annotated for potential pial versus arachnoid cells. The transcriptionally distinct clusters (4, 6, and 8) are color coded as per the adjacent key. H, Violin plots showing relative expression levels of two pial mRNAs, Lama1 and Cyp1b1, and two arachnoid mRNAs, Slc47a1 and Dpp4, in the potential arachnoid versus pial leptomeningeal cells as annotated in G. Purple and green denote arachnoid and pial cells, respectively. Each dot corresponds to expression in an individual cell. I, Spatial plots of the cortical interface and midline showing spatial distribution of the cluster 6 pial cells (green) and the clusters 4 and 8 arachnoid cells (turquoise and purple, respectively), as annotated in G. Note that the pial but not arachnoid cells extend down the midline. J, High-resolution Xenium Explorer image of the cortical interface (all panels) and midline (left panel) analyzed with the brain probeset (BP) showing expression of the arachnoid mRNAs Cdh1 (red dots), Dpp4 (yellow dots), and Slc47a1 (pink dots) relative to the arachnoid and pial cells (turquoise and green, respectively) as annotated in G. Nuclei of other cell types (white) are also shown. Note that the pial cells extend down the midline (left panel). Low magnification spatial plots in E, F, and I were generated using Seurat, and each dot represents the centroid of one cell. High-resolution spatial plots in J were generated using Xenium Explorer. Scale bars, 10 μm.

We compared the potential pial and arachnoid cells by differential gene expression analysis (Extended Data Fig. 6-2). This analysis identified 555 differentially expressed genes (average log2 fold change ≥1.3, adj p value <0.5, expression in at least 10% of cells), with 454 and 101 enriched in the potential pial versus arachnoid cells, respectively. Since a subset of these mRNAs were targeted in the Xenium probesets (Extended Data Figs. 4-3, 6-2), we asked about their spatial expression profiles in brain interface leptomeningeal cells. Notably, spatial plots showed that the putative pia-enriched mRNAs, including Coch, Lamc3, Gjb2, Igfbp4, and Cyp1b1, were expressed in leptomeningeal cells that extended down the midline (Fig. 6E). In contrast, mRNAs enriched in the putative arachnoid cells, such as Bnc2, Cdh1, Dpp4, and Rspo2, were excluded from the midline (Fig. 6F). The pial marker Lama1 and arachnoid marker Ppp1r1a displayed similar differential spatial localization (Fig. 6E,F).

We asked whether these gene expression patterns corresponded to transcriptionally distinct leptomeningeal populations within the Xenium analysis, as seen in the scRNA-seq. Higher resolution UMAP analysis of the brain probeset data identified two groups of leptomeningeal cells (Fig. 6G). One of these (cluster 6) included potential pial cells that extended down the midline and were enriched for Lama1, Cyp1b1, Gjb2, and Igfbp4 (Fig. 6H,I). The other population (clusters 4 and 8) included potential arachnoid cells that were excluded from the midline and were enriched for Dpp4, Slc47a1, Cdh1, and Opn3 (Fig. 6H–J). Similar results were obtained with a higher resolution analysis of the mesenchymal probeset data (Extended Data Fig. 6-1F–H), identifying pial cells that extended down the midline that were enriched for Lamc3, Bmp2, Lama1, and Cyp1b1, and arachnoid cells that were excluded from the midline and were enriched for Bnc2 and Dpp4. Thus, the Xenium and scRNA-seq analyses define two leptomeningeal cell populations that have locations and transcriptional signatures consistent with their identification as pia versus arachnoid cells.

Brain interface and layer 1 cells comprise different cellular neighborhoods

We next compared the cellular environments of the brain interface and the adjacent cortical layer 1. To do so, we used the mesenchymal and brain probeset spatial transcriptomic datasets to define the cortical layer 1 cellular environment. In both cases, spatial plots showed that layer 1 was relatively cell sparse and included scattered microglia, astrocytes, OPCs, and oligodendrocytes as well as vasculature-associated endothelial and mural cells (Fig. 7A,B). As indicated above, both probesets identified interneurons but not excitatory neurons in layer 1, as compared with layer 2, which was densely packed with both excitatory and inhibitory neurons (data not shown). The brain probeset identified a more abundant Lamp5-expressing interneuron population localized throughout layer 1 and a second less abundant Vip-expressing population largely located closer to the layer 2 border (Fig. 7C). As predicted, this analysis also identified interneurons in layer 2 but not layer 1 that expressed Sst or Parvalbumin (Fig. 7C; Lim et al., 2018; Yao et al., 2021).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Single-cell spatial transcriptomic analysis defines distinct proximal cellular neighborhoods for the cortical brain interface and layer 1. Coronal adult cortical sections were analyzed by Xenium-based single-cell multiplexed in situ gene expression analysis with either a brain probeset targeting 347 genes or a mesenchymal probeset targeting 480 genes (Extended Data Fig. 4-3). The ROI and UMAPs of the resultant merged datasets are shown in Figure 4A–C. A, Spatial plots of the cortical interface and layer 1 analyzed with the mesenchymal (MP, left) and brain (BP, right) probesets showing oligodendrocytes (pink), microglia (tan), parenchymal astrocytes (dark brown), leptomeninges (dark pink), OPCs (orange), border astrocytes (yellow), dural cells (green, mesenchymal probeset only), and border macrophages (dark blue). B, Spatial plots of the cortical interface and layer 1 analyzed with the mesenchymal (MP, left) and brain (BP, right) probesets showing pericytes (turquoise), endothelial cells (red), VAMCs (blue), dural cells (green, mesenchymal probeset only), and leptomeninges (dark pink). C, Spatial plot of the cortical interface and layer 1 analyzed with the brain probeset (BP) showing the leptomeninges (dark pink) and interneurons expressing Lamp5 (lime green, Lamp5IN), Vip (blue-green, VIPIN), Somatostatin (gray, SSTIN), or Parvalbumin (black, PvalbIN). D, Proximity analysis showing the relative strength of statistically significant interactions between different cell types located in the cortical brain interface and adjacent layer 1, as analyzed from the datasets obtained using either the brain probeset (left) or the mesenchymal probeset (right). The nodes are color coded and annotated for cell type, and the weight of the line indicates the strength of the interaction. Black lines indicate significant interactions based on a permutation test, comparing our data to a null distribution created from the random permutation of cell labels with fixed positions (see Materials and Methods for details). Green lines indicate interactions that were not statistically significant. Excitatory neuron-1 and excitatory neuron-2 denote transcriptionally distinct excitatory neuron clusters. Cell centroid information was used to perform this analysis, with a proximal interaction being defined as one where a cell centroid was within 70 µm of another cell centroid. This unsupervised analysis identifies two neighborhoods, the cortical brain interface and layer 1, as well as a partial neighborhood centered around the layer 2 excitatory neurons at the boundary of layer 1. Oligos, oligodendrocytes. Low magnification spatial plots (A–C) were generated using Seurat, and each dot represents the centroid of one cell. The proximity maps in D were generated using Giotto.

We then more quantitatively defined cells in close proximity to each other within the brain interface/layer 1 environments using cellProximityEnrichment in Giotto v. 4.0.2 (described in detail in Dries et al., 2021) to analyze the single-cell spatial transcriptomic data. We defined cells as being proximal if their cell centroids were within 70 µm of each other and termed each proximal association an interaction. For both the mesenchymal and brain probeset data, this analysis identified two distinct cellular neighborhoods (Fig. 7D). One of these included the brain interface cells, the leptomeningeal pia and arachnoid cells, border astrocytes, and border macrophages, all of which were located in very close proximity to each other. Also in the vicinity were the VSMCs and, in the case of the mesenchymal probeset, the dura. In contrast, the brain interface cells were largely not significantly associated with other layer 1 cell types, although spatial plots did detect the occasional neural cell close to the interface (Fig. 7A).

The second proximal neighborhood included the cortical layer 1 and layer 2 cells other than border astrocytes: various interneurons, OPCs, oligodendrocytes, microglia, parenchymal astrocytes, excitatory neurons at the layer 1/2 border, as well as the endothelial cells and pericytes associated with smaller vessels and capillaries. With the exception of interactions involving layer 2 excitatory and inhibitory neurons, these interactions were largely not as robust as those between the border interface cells, consistent with the relatively sparse cellular density in layer 1 (Schuman et al., 2021). Thus, the cellular environments of the brain interface and cortical layer 1 are distinct, and the brain interface environment is likely predominantly determined by the border astrocytes, border macrophages, and leptomeningeal cells.

Leptomeningeal cells express ligands predicted to act on both border astrocytes and macrophages

We next defined ligands made within the brain interface that might influence the biology of this cell-dense neighborhood. To do this, we used the scRNA-seq datasets (Fig. 1A,B) and a previously published ligand–receptor database (Toma et al., 2020). We defined a ligand mRNA as expressed if it was detectable in at least 5% of the relevant cell type. We identified 65 leptomeningeal ligands, 48 border macrophage ligands, and 54 border astrocyte ligands (Extended Data Fig. 8-2). Of these, 15 were expressed by all three cell types, including ligands such as Bmp6, Efnb1, Hbegf, Il18, Vegfb, and Gas6. Fifteen were expressed by leptomeninges and border astrocytes, including Ctf1 and Gdf11, 12 were expressed by leptomeninges and border macrophages, including Wnt5a, Igf2, and Tgfb1, while one, Sema4d, was expressed by border astrocytes plus macrophages. The remaining ligands were expressed by only one of the three cell types; leptomeninges expressed 13 distinct ligands including Bmp4, Bmp5, Bmp7, Sema3a, and Igf2, border macrophages expressed 9 distinct ligands including Osm, Tnf, and Il1b, while border astrocytes specifically expressed 13 ligands, including Fgf1, Tgfb2, and Wnt7a, and the semaphorins Sema3b, Sema4g, and Sema5b.

To ask if these ligands might influence the brain interface neighborhood, we used the same scRNA-seq data and ligand–receptor database to define receptors expressed by each interface cell type. This analysis defined 102, 95, and 79 receptors expressed by the leptomeninges, border astrocytes, and border macrophages, respectively (Extended Data Fig. 8-3). We used this information to predict which ligands might be biologically active. This modeling (Fig. 8A,B; Extended Data Fig. 8-4) predicted 78 ligand–receptor interactions within the brain interface neighborhood. Most ligands (52) had receptors on all three brain interface cell types, but the remainder were more specific.

• Download figure
• Open in new tab
• Download powerpoint

Figure 8.

Leptomeningeal cells express ligands predicted to regulate border astrocytes and border macrophages. Also see Extended Data Figures 8-1 to 8-4. A–D, Leptomeningeal cells, border astrocytes, and border macrophages were analyzed for their expression of ligand and ligand receptor mRNAs using the scRNA-seq data (Extended Data Figs. 8-2, 8-3, and 8-4). Ligand or receptor mRNAs were included if they were expressed in ≥5% of the relevant cell type. The data were used to generate predictive models of the bioactive ligand environment within the brain interface neighborhood. Each box includes a ligand known to bind to a corresponding receptor expressed by at least one of the brain interface cell types. A, B, All of the ligands made by the brain interface cells that have a receptor expressed by at least one of the interface cell types. In A, these are color coded to show the cells that express these ligands, while in B, they are color coded to show the cells that express the relevant receptor and thus are predicted to respond to the ligands. C, D, Models with ligands expressed by the leptomeningeal cells that have corresponding receptors on either border macrophages (C) or border astrocytes (D). E, Dot plot showing expression levels of Igf2, Sema3a, Bmp6, and Bmp7 mRNAs in leptomeningeal cells, border astrocytes, and border macrophages, as determined from the scRNA-seq data. The size of the dot indicates the percentage of cells detectably expressing the mRNA, and the color indicates relative expression level, coded as per the adjacent keys. F, Spatial plots of Bmp7, Bmp6, Igf2, Sema3a, and Igf1r mRNAs in all brain interface and layer 1 cells analyzed used the brain probeset (BP), all shown on the same representative section. Relative mRNA expression levels are coded as per the adjacent keys. G, High-resolution Xenium Explorer images of the cortical interface region analyzed with the brain probeset (BP) showing expression of Bmp6 (light pink dots), Igf2 (blue dots), Bmp7 (green dots), Igf2 (blue dots), and Sema3a (white dots) mRNAs relative to the leptomeningeal cells (pink), border astrocytes (yellow), and border macrophages (dark blue). The same field of view is shown in each panel, and arrows denote the same leptomeningeal cell detectably coexpressing multiple ligand mRNAs. Nuclei of other cell types (white) are also shown. H, High-resolution Xenium Explorer images of the cortical interface region analyzed with the brain probeset (BP) showing expression of Igf2 (blue dots) and Igf1r (yellow dots) mRNAs relative to the same cell types as in G. Nuclei of other cell types (white) are also shown. Low magnification spatial plots in F were generated using Seurat and the high-resolution spatial plots in G and H with Xenium Explorer. Scale bars, 10 μm.

We also more specifically modeled ligands made by leptomeningeal cells that could influence their neighboring border astrocytes and macrophages (Fig. 8C,D; Extended Data Fig. 8-4). Most leptomeningeal ligands were predicted to act on both neighboring cell types. Analysis of the scRNA-seq data showed that some of these ligand mRNAs were highly enriched in the leptomeningeal cells relative to other neural and non-neural cell types, including Igf2, Bmp6, Bmp7, and Sema3a (Fig. 8E; Extended Data Fig. 8-1A). Since probes for these same ligands were present in the Xenium panels, we analyzed their spatial expression. Spatial plots showed that all four ligands were highly enriched and coexpressed in leptomeningeal cells at the brain interface (Fig. 8F,G). There was also scattered expression of these ligands in some layer 1 cells (Fig. 8F), consistent with the scRNA-seq. Furthermore, the Xenium panels included probes for Igf1r, one of the main receptors for Igf2. Spatial plots for Igf1r confirmed that it was broadly expressed within the brain interface and adjacent layer 1 and that Igf1r-positive border astrocytes and macrophages were adjacent to Igf2-expressing leptomeningeal cells (Fig. 8F,H). Consistent with these findings, all four of these leptomeningeal ligands have been shown to directly affect astrocytes and macrophages (Ishikawa et al., 1995; Choe et al., 2014; Aluganti Narasimhulu and Singla, 2020) and cultured meningeal cells secrete all of these ligands (Ishikawa et al., 1995; Ohe et al., 1996; Niclou et al., 2003; Choe et al., 2012; Segklia et al., 2012). Thus, leptomeningeal cells express ligands that are predicted to directly influence their macrophage and border astrocyte neighbors and perhaps play a role in regulating the brain interface.