Multiomic spatial landscape of innate immune cells at human central nervous system borders

Sankowski, Roman; Süß, Patrick; Benkendorff, Alexander; Böttcher, Chotima; Fernandez-Zapata, Camila; Chhatbar, Chintan; Cahueau, Jonathan; Monaco, Gianni; Gasull, Adrià Dalmau; Khavaran, Ashkan; Grauvogel, Jürgen; Scheiwe, Christian; Shah, Mukesch Johannes; Heiland, Dieter Henrik; Schnell, Oliver; Markfeld-Erol, Filiz; Kunze, Mirjam; Zeiser, Robert; Priller, Josef; Prinz, Marco

doi:10.1038/s41591-023-02673-1

Download PDF

Article
Open access
Published: 20 December 2023

Multiomic spatial landscape of innate immune cells at human central nervous system borders

Nature Medicine volume 30, pages 186–198 (2024)Cite this article

22k Accesses
6 Citations
165 Altmetric
Metrics details

Subjects

Abstract

The innate immune compartment of the human central nervous system (CNS) is highly diverse and includes several immune-cell populations such as macrophages that are frequent in the brain parenchyma (microglia) and less numerous at the brain interfaces as CNS-associated macrophages (CAMs). Due to their scantiness and particular location, little is known about the presence of temporally and spatially restricted CAM subclasses during development, health and perturbation. Here we combined single-cell RNA sequencing, time-of-flight mass cytometry and single-cell spatial transcriptomics with fate mapping and advanced immunohistochemistry to comprehensively characterize the immune system at human CNS interfaces with over 356,000 analyzed transcriptomes from 102 individuals. We also provide a comprehensive analysis of resident and engrafted myeloid cells in the brains of 15 individuals with peripheral blood stem cell transplantation, revealing compartment-specific engraftment rates across different CNS interfaces. Integrated multiomic and high-resolution spatial transcriptome analysis of anatomically dissected glioblastoma samples shows regionally distinct myeloid cell-type distributions driven by hypoxia. Notably, the glioblastoma-associated hypoxia response was distinct from the physiological hypoxia response in fetal microglia and CAMs. Our results highlight myeloid diversity at the interfaces of the human CNS with the periphery and provide insights into the complexities of the human brain’s immune system.

Brain endothelial GSDMD activation mediates inflammatory BBB breakdown

Article 17 April 2024

Simultaneous single-cell three-dimensional genome and gene expression profiling uncovers dynamic enhancer connectivity underlying olfactory receptor choice

Article Open access 15 April 2024

Whole-cortex in situ sequencing reveals input-dependent area identity

Article Open access 24 April 2024

Main

The CNS interfaces are physical, immunological and molecular barriers ensuring the structural integrity of the CNS parenchyma while facilitating waste disposal¹. Recent evidence identified CNS interfaces as anatomical sites for pathogen invasion, tumor dissemination and neurodegeneration^2,3,4. In line with the diverse immunological functions, CNS interfaces host more diverse immune populations than the CNS parenchyma (PC)^4,5,6. The mouse perivascular space (PV), leptomeninges (LM), choroid plexus (CP) and dura mater (DM) contain myeloid cells, lymphoid cells and dendritic cells (DCs), whereas the PC largely contains only microglia^7,8. Notably, non-parenchymal PV, LM, CP and DM macrophages represent the main immune cells at CNS interfaces^8,9. Collectively these cells are called CAMs or border-associated macrophages (BAMs)^{10,11,12,13,14,15}. Despite their important functions^11,16, their complexity in human brains remains unexplored.

For many years, it was thought that all CAMs originated from short-lived bone-marrow-derived monocytes that are continuously replaced postnatally¹⁷. Recent evidence established a common prenatal yolk sac-derived origin of CAMs^9,18,19,20. CAMs and microglia were shown to be long-lived and self-renewing^9,21. Despite these common origins, adult microglia and CAMs show distinct transcriptional phenotypes. Namely, mouse microglia mainly express Hexb, Tmem119 and P2ry12, whereas CAMs characteristically show Mrc1, Lyve1 and Ms4a7 expression^7,8,10,22,23.

Recently, we described that while microglia and LM CAMs are present prenatally, PV CAMs only arose postnatally following the establishment of the PV space²⁴. Notably, direct cell–cell interactions between CAMs and smooth muscle cells was crucial for the postnatal colonization of the PV space. These findings point to a pronounced integration of CAMs within their niches and major developmental plasticity.

While individual human CNS interface specimens have been initially profiled, an integrated side-by-side transcriptomic, proteomic and cross-species comparisons have only been conducted for the brain PC²⁵. Also, a spatially resolved transcriptomic analysis of the pathophysiologically important CAM niche is still lacking. Furthermore, little is known about the development, fates and turnover of human CAMs. Here, we applied single-cell RNA sequencing (scRNA-seq), cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq), mass cytometry and high-resolution spatial transcriptomics to generate a comprehensive molecular census of the immune compartment at the human CNS interfaces during fetal development, adulthood and pathology.

Results

Homeostatic immune-cell diversity at human CNS interfaces

We profiled the immune compartment at the human CNS interfaces, with state-of-the-art molecular techniques (Fig. 1a). A total of 11,166 CD45⁺ cells from human PC/PV space, LM, CP and DM were enriched by fluorescence-activated cell sorting (FACS) and analyzed using droplet-based scRNA-seq (Extended Data Fig. 1a,b). Cell types were classified using a combination of Azimuth²⁶ and published CAM and microglia gene sets^7,8,27. We found diverse myeloid subsets, including CAMs, microglia, DCs and monocytes, and several lymphoid subsets, including CD4⁺, CD8⁺ and proliferating T lymphocytes²⁸ (Fig. 1b, Extended Data Fig. 2a–c and Supplementary Table 1). CAMs (C19 and C26) mostly consisted of CP, LM and DM-derived cells (Fig. 1c,d, and Supplementary Tables 2 and 3). In line with evidence from mice^8,27, Kolmer cells co-clustered with activated microglia in C22 (Fig. 1c and Supplementary Table 2). Thus, our dataset contained diverse myeloid and lymphoid subsets.

**Fig. 1: Molecular census of immune cells in human CNS border regions under homeostasis.**

A high degree of vascularization of the CNS interfaces makes the distinction between intravascular and tissue-resident cells challenging. Hence, we conducted histological validation for the main cell-type markers in the LM, the interface predominately included in this study. We detected cells expressing the pan-myeloid marker Iba1 and the T cell marker CD3 outside of blood vessels in considerable numbers (Fig. 1e). Previous work has identified homing markers for myeloid (for example AREG and PLAUR) and lymphoid cells (for example CD69)²⁹. We therefore assessed the expression of these gene modules in our data. Notably, the myeloid-homing module was elevated in DM-associated monocytes (C20) (Fig. 1c,f). This corroborates evidence of continuous monocyte-derived macrophage (MoMΦ) engraftment in the murine DM⁸. The lymphocyte-homing module was elevated in T cells of C8 and C21 (Fig. 1f). This analysis confirmed the presence of diverse tissue-resident immune cells in the LM.

Next, cluster marker analysis of the different myeloid populations found distinct markers for CAMs (C19 and C26: MRC1 and F13A1), Kolmer cells (C22: BHLHE41, APOE, SPP1 and TTR) and the DM-associated MoMΦ (C20: AREG, PLAUR and CD44) (Fig. 1g and Extended Data Fig. 2d). Notably, APOE and SPP1 in Kolmer cells represent so-called ‘disease-associated microglia’ (DAM) markers^8,30. In summary, we detected a broad transcriptional overlap of CAMs markers across the analyzed human CNS interfaces. Additionally, Kolmer cells showed a microglia-like signature and DM-associated MoMΦ-expressed myeloid-homing genes.

Comparison of gene signatures in murine and human CAMs

Next, we investigated the main factors distinguishing the analyzed anatomical compartments. Multifactorial factor analysis of 4,096 MΦ from 22 samples using MOFA2 (ref. ³¹) identified seven latent factors (Extended Data Fig. 3a and Supplementary Table 4). While factors 1 and 2 were evenly distributed across all compartments, factors 3 and 7 were over-represented in DM. Gene Ontology analysis identified enrichment of the term leukocyte migration in factor 3 and extracellular matrix organization in factor 7 (Extended Data Fig. 3b). Factors 4 and 5 contained macrophage-enriched genes as top genes, including MRC1, STAB1 and LYVE1 (Extended Data Fig. 3a). Notably, factor 4 was over-represented in CP, PC/PV and LM, whereas factor 5 was over-represented in CP, DM and LM. Gene Ontology analysis identified enrichment of major histocompatibility complex (MHC) II-associated terms in latent factor 5 suggesting higher immune activation in CP, LM and DM but not PC/PV (Extended Data Fig. 3b). In summary, PC/PV-derived myeloid cells show an attenuated expression of antigen-presentation-associated genes.

Species comparison between human CAMs and microglia with a published mouse dataset⁷ showed conserved expression of MRC1, F13A1, STAB1 and CD163 in CAMs across species (Extended Data Fig. 3c and Supplementary Table 5). Microglia displayed conserved expression of P2RY12, SLC2A5, SELPLG and SPP1. Divergently regulated genes included APOE and AXL that seemed upregulated in murine CAMs and in human microglia. Conversely, MAFB and MERTK seemed upregulated in human CAMs and murine microglia. In summary, CAMs and microglia expressed evolutionary conserved gene sets with notable differences between mice and humans.

Multiomics reveal diversity of steady-state human CAMs

For further validation, we combined three high-dimensional technologies, scRNA-seq, CITE-seq and time-of-flight mass cytometry. The combination of scRNA-seq and mass cytometry previously enabled us to identify and characterize a hitherto unappreciated spectrum of human microglial states^32,33.

First, we enriched CAMs for deeper characterization using CD206 (encoded by MRC1), a conserved pan-CAM marker. Then, 1,962 CD45⁺CD206⁺CD3⁻CD19⁻CD20⁻ cells from 12 patients were FACS-sorted into multiwell plates and analyzed using the high-sensitivity mCEL-Seq2 protocol³⁴ (Fig. 2a). The dataset consisted of CAMs, type 2 conventional DCs (cDC2) and MoMΦ (Fig. 2b and Extended Data Fig. 4a,b). CAMs (C0 and C1) were distributed across all analyzed compartments with enrichment of C0 in PC/PV, CP and DM, and C1 in LM (Fig. 2c and Supplementary Table 6). cDCs (C2) were enriched in LM and PC/PV, whereas MoMΦ (C3) were enriched in CP and DM, underscoring continuous myeloid-derived engraftment⁸. Gene Ontology enrichment analysis showed MHC-II-associated terms in CAMs and cDC2, chemokine and cytokine activity terms in CAMs and fibronectin binding in MoMΦ (Fig. 2d and Supplementary Table 7). In summary, we found a transcriptional spectrum of MHC-II^low in C0 and MHC-II^high CAMs in C1. These analyses confirmed CD206 protein expression in CAMs, cDC2 and MoMΦ.

**Fig. 2: Characterization of human CAMs during homeostasis by single-cell sequencing and CITE-seq.**

Next, we assessed broader surface marker profiles of the immune cells analyzed in Fig. 1 using CITE-seq with 134 cell surface markers. After quality control, we obtained transcriptional and proteomic profiles for 8,161 cells. CAMs (C26 and C19) showed CD304 (encoded by NRP1), CD88 (encoded by C5AR1) and CD169 (encoded by SIGLEC1) as markers (Fig. 2e and Supplementary Table 8). Differential protein expression analysis between the distinct myeloid cell types found separate markers for DM-associated MoMΦ (CD44, CD41 (encoded by ITGA2B) and CD36), CAMs (CD71 (encoded by TFRC), CD304 (encoded by NRP1) and CD169 (encoded by SIGLEC1)), microglia (CX3CR1, podoplanin (encoded by PDPN) and CD39 (encoded by ENTPD1)) and Kolmer cells (CD9, CD49d (encoded by ITGA4) and CD72 (Fig. 2f). Additionally, mass cytometry confirmed and expanded protein markers for monocytes (C9: CLEC12A), CAMs (C7: CD206), microglia (C1 and C3: P2RY12) and cDCs (C11: CCR5 and IRF8) (Extended Data Fig. 4c–f and Supplementary Table 9). Thus, our findings provide a combined assessment of gene and protein expression in human CAMs.

Spatial organization of human CAMs

Next, we examined CAMs in their anatomical niches using multifluorescence confocal microscopy and spatial transcriptomics through in situ sequencing (ISS) and the Nanostring CosMx technology.

We used collagen IV as a marker for basal lamina to confidently assign anatomical locations^9,10,12,24. CD206 and CD163 were homogeneously coexpressed with Iba1 by CAMs across anatomical compartments, but not by microglia and Kolmer cells (Fig. 3a,b and Extended Data Fig. 5). Notably, CD169 was only expressed in up to two-thirds of CD206⁺ cells with the highest expression in perivascular CAMs. S100A6 and CD1C were very lowly expressed in the analyzed compartments. Hierarchical clustering based on average expression of the above-mentioned proteins confirmed the phenotypic similarity of microglia and Kolmer cells (Fig. 3c). Thus, we found enhanced expression of CD169 in PV CAMs.

**Fig. 3: Spatial profiling of human CAMs in situ and their cellular interactions.**

For spatial transcriptome analysis, we applied ISS, a technique for optical detection of genes at a single-cell resolution³⁵. The gene panel included cell-type markers and genes identified above with scRNA-seq (Supplementary Table 10). We analyzed a cortex specimen containing two cortical cross-sections separated by an LM layer (Fig. 3d). Azimuth-based cell-type annotation revealed distinct cellular compositions between the cortex and LM (Fig. 3d). Spatial neighborhood enrichment analysis³⁶ showed that LM CAMs preferentially co-occurred and interacted with fibroblasts and endothelial cells (Fig. 3e,f). For cell–cell interaction analysis of human PC/PV and LM specimens, we used the Nanostring CosMx technology with a 1,000-plex gene panel (Supplementary Table 11). We observed an overlap between transcriptionally similar CAMs and microglia, with CD74 being the main receptor for CAM-directed neighborhood-to-cell interactions (Fig. 3g,h). In summary, spatial transcriptomics highlighted spatial interactions of CAMs.

Microenvironment shapes developing human CAM phenotypes

Myeloid cells enter the human CNS by post-conception week (pcw) 4.5 (ref. ³⁷). Although fetal human microglia have previously been profiled^38,39, comprehensive single-cell analyses of developmental human CAMs are unavailable. Hence, we conducted single-nucleus RNA-seq on 59,053 cells from 32 frozen human PC/PV, LM and CP samples between pcw 7 to 23 and postnatal PC/PV and LM specimens. Additionally, we integrated our immune-cell transcriptomes with published data (Fig. 4a). We FACS-sorted NeuN⁻Olig2⁻ nuclei and obtained 59,053 nuclei from major neuroectodermal, structural and immune-cell types (Fig. 4b,c and Supplementary Table 12). For comparative analysis, the immune cells were integrated with published transcriptomes from human first-trimester PC/PV and adult control CP^2,39. This approach yielded an anatomically dissected dataset of 13,807 immune-cell transcriptomes spanning the period of CNS myeloid cell engraftment through adulthood. We mainly found myeloid cells with a comparatively large proportion of proliferating cells (Fig. 4d and Extended Data Fig. 6a,b). Notably, the proliferating C9 contained two subpopulations expressing either microglia or CAM genes (Extended Data Fig. 6a). Tracking cluster composition chronologically revealed microglia and CAM developmental dynamics. Before pcw 10, microglia were mainly C1 and C2 (Fig. 4e). C3, C4 and C5 became the majority after birth. For LM and CP CAMs, both before and after birth, the main clusters were C0 and C10. Kolmer cells (C8) were not observed before pcw 20, potentially due to under-sampling. Differential gene expression analysis highlighted LYVE1, CD163 and F13A1 in CAMs (C0, C10 and C13) (Extended Data Fig. 6c). Kolmer cells (C8) mainly expressed PADI2 and SORL1, also found in adult microglia (C4 and C5). Notably, first-trimester microglia (C1, C2 and C7) expressed chemokines and the DAM genes SPP1 and APOE. Similar DAM gene module expression was previously observed³⁹ (Extended Data Fig. 6d,e). Notably, even when excluding the Braun et al. data, a statistically significant time-dependent reduction in the DAM signature within microglia was evident, unlike in CAMs or Kolmer cells (Extended Data Fig. 6f). In summary, CAMs and microglia exhibited distinct transcriptional profiles early on, with the so-called DAM microglia signature inversely correlating with brain maturation.

**Fig. 4: Region-dependent transcriptional dynamics of human CNS border macrophages during development.**

We utilized cell–cell interaction analysis to uncover molecular mediators shaping CNS myeloid cell phenotypes. We merged the integrated immune-cell dataset with the non-immune-cell types from Fig. 4b and ran NICHES⁴⁰ to infer interactions toward CAMs, microglia and Kolmer cells, respectively. To avoid sampling issues, we binned prenatal time points and compared them to the postnatal ones. Differential analysis identified transferrin receptor (TFRC) among the top prenatal receptors in all three cell types (Fig. 4f). CITE-seq analysis confirmed prenatal CD71 (encoded by TFRC) upregulation in these cells (Fig. 4g and Extended Data Fig. 6g–i). Postnatal cells upregulated immune mediators, including HLA-DR. Gene Ontology analysis identified hypoxia-related terms in prenatal CAMs and Kolmer cells and autophagy in CAMs and microglia (Fig. 4h and Supplementary Table 13). These results corroborate the importance of TFRC in hypoxia⁴¹. Postnatal CAMs were involved in vascular biological processes. In conclusion, we highlight the fetal brain milieu influencing gene and protein expression in brain myeloid cells with CAMs, microglia and Kolmer cells displaying a response to physiological prenatal hypoxia.

Compartment-dependent turnover rates of CNS myeloid cells

Turnover of CAMs by circulating cells has been investigated in mice^9,10,24 but not in humans. Hence, we established in situ fate mapping in human brain tissues from 15 sex-mismatched peripheral blood stem cell transplantation (PBSCT) recipients⁴². This analysis involved quantifying Y chromosome-positive (Y⁺) immune cells and genome-wide single-nucleus transcriptome profiling (Fig. 5a). Pseudotime analysis⁴³ of intravascular and homing monocytes (C7 and 20 from Fig. 1b) revealed sequential transcriptional changes during myeloid cell engraftment (Fig. 5b). This was marked by an upregulation of chemokine and myeloid homing (CXCL8, AREG and PLAUR)²⁹ and iron-scavenging genes (TFRC, FTL and FTH1) (Fig. 5c and Supplementary Table 14).

**Fig. 5: Compartment-specific turnover of human CNS border macrophages.**

Next, we quantified bone-marrow-derived engrafted cells at the CNS interfaces of female patients who received sex-mismatched PBSCT to treat hematological diseases (Supplementary Table 20). Notably, we observed donor-derived Y⁺Iba1⁺ cells within all examined CNS interfaces and the brain PC (Fig. 5d). Correlation analysis between the percentage of Y⁺Iba1⁺ cells and all Iba1⁺ cells in PC/PV, CP, LM and DM clearly demonstrated a time-dependent increase of Y⁺Iba1⁺ cells (Fig. 5e). Of note, T₅₀, the time to reach a 50% exchange rate of Y⁺Iba1⁺ cells, displayed significant variability across compartments ranging from 51.52 d (95% confidence interval 31.9–97.1) in the CP to 264.55 d (95% confidence interval 145.9–563.9) for the brain PC (Fig. 5f). Notably, the T₅₀ exhibited substantial dispersion, particularly within the parenchyma, suggesting influence from interindividual characteristics. Given CD206 expression in nearly all CAMs whereas SIGLEC1 was only present in a subset (Fig. 3b), we wondered whether Y⁺CD206⁺ and Y⁺SIGLEC1⁺ cells were found at different rates. Indeed, at late post-transplantation time points (218, 453 and 1,018 d) Y⁺SIGLEC1⁺ cells were present in approximately half of the Y⁺Iba1⁺ and Y⁺CD206⁺ cells (Fig. 5g,h).

For a deeper profiling of engrafted cells, we conducted single-nucleus mRNA profiling from formalin-fixed paraffin-embedded (FFPE) tissues of the above-mentioned three post-transplantation samples and age-matched controls. We FACS-sorted NeuN⁻Olig2⁻ nuclei from PC/PV and adjacent LM sections, yielding 9,035 single-nucleus transcriptomes of the major brain, vascular and immune-cell types (Extended Data Fig. 7a,b). Subclustering 363 myeloid cells identified two adjacent clusters (C3 and C0) enriched in control and transplanted patients, respectively (Fig. 5i). Both clusters included C1QA and C1QB among top markers, indicative of microglia and tissue-resident macrophages (Extended Data Fig. 7c and Supplementary Table 15). The control-enriched C3 expressed the microglia-enriched genes BHLHE41 and P2RY12, whereas the post-transplantation-enriched C0 showed upregulation of CD163 and PLTP, implying an activated phenotype (Fig. 5j).

Given the potential comorbidities in PBSCT patients, we analyzed the immune phenotype of resident microglia and CAMs using in situ fate mapping analysis of PC cells (Fig. 5k). Notably, 90% of Y⁺ cells were double-positive for Iba1, but only about 20% of Y⁺ cells expressed microglia core markers P2RY12, GLUT5 and TMEM119 (Fig. 5l). When analyzing the percentage of Y⁺ cells among all Iba1⁺, P2RY12⁺, TMEM119⁺ and GLUT5⁺ cells in the same three cases, we found no major differences between brain-resident and engrafting cells (Extended Data Fig. 7d). In summary, we described patterns of human brain myeloid-derived cell engraftment and evidenced compartment-specific engraftment rates across different CNS interfaces. High-dimensional profiling was constrained by the scarcity of these regions, we revealed that cells engrafting in the parenchyma maintained an activated phenotype.

Context-dependent CAM signatures in human patients with glioblastoma

Glioblastoma are malignant brain tumors with a complex immune microenvironment^33,44. We profiled 11,681 CAMs from 21 anatomically dissected glioblastoma and control samples using scRNA-seq, CITE-seq and spatial transcriptomics (Fig. 6a). Alongside normal brain cell types, tumor samples contained tumor-associated macrophages (TAMs)^33,44 in C2 and C4 (Fig. 6b,c and Supplementary Table 16). C2 cells expressed microglia genes, whereas C4 showed macrophage gene expression, corresponding to microglia TAMs (mgTAMs) and monocyte-derived TAMs (moTAMs), respectively (Extended Data Fig. 8a). Additionally, the tumor-enriched C15 contained homing monocytes resembling transitory moTAMs (Tr. moTAMs)⁴⁴ (Extended Data Fig. 8b and Supplementary Table 16). Label transfer of our clusters onto published data⁴⁴ confirmed Tr. moTAMs in C15, mgTAMs in C2 and hypoxic and lipid moTAMs in C4 (Extended Data Fig. 8c). Notably, so-called SEPP1^hi moTAMs were indistinguishable from control CAMs in C17, mirroring their high SELENOP expression (previously known as SEPP1)⁴⁴ (Fig. 1d, Extended Data Fig. 8d and Supplementary Table 17).

**Fig. 6: Multimodal analysis reveals common activation programs of CAMs and microglia in human glioblastoma.**

Cluster analysis assessed PC/PV and LM contributions to each cluster. Besides intravascular lymphoid cells, LM contained cDC (C21) and Tr. moTAM (C15) (Fig. 6d and Supplementary Table 18). TAMs (C2 and C4) were enriched in PC/PV, suggesting some CAM involvement in the glioblastoma microenvironment.

To dissect distinct transcriptional variations within the glioblastoma immune microenvironment, we focused on tumor-associated myeloid cells and employed MOFA2. Among TAMs, latent factors 1, 2 and 5 were expressed (marker genes NRP1, FTL and HIF1A-AS3) (Fig. 6e and Supplementary Table 19). Notably, PC/PV and LM CAMs were primarily characterized by latent factor 1, whereas latent factor 3 (marker genes CLEC12A and CD36) explained the variability in Tr. moTAMs. Microglia variability was mainly explained by latent factor 6 (marker gene HIF1A-AS3). Notably, latent factor 4 (marker genes CCL4, IL1B and ICAM1) was found in LM-derived cells, highlighting the role of the LM as gateway for engraftment. Gene Ontology analysis of latent factors yielded terms associated with the cell matrix and vasculature for latent factor 1, hypoxia for factors 2, 4 and 5, and immune activation for factors 3 and 4 (Fig. 6f). Considering the influence of hypoxia on TAM transcriptional profiles, we analyzed their distribution within the tumor. ISS on a tumor section spanning several mm² identified nine hypoxic regions with focal VEGFA expression amid cellular tumor regions (Fig. 6g). Spatial trajectory analysis from the periphery to the hypoxic center of area nine unveiled decreasing PLP1 and EGFR expression, followed by a focal peak of VEGFA at the rim and FTL expression at the core (Fig. 6h). Corresponding with FTL expression in MOFA2 latent factor 2, more moTAM were in hypoxic area 9 than the surrounding tumor (Fig. 6i).

To further dissect the perinecrotic area, we employed Nanostring CosMx technology. This revealed a significant rise of moTAMs and Tr. moTAMs toward the necrosis (Fig. 6j). Mirroring ISS, spatial trajectory analysis demonstrated the downregulation of neuroectodermal genes and the upregulation of myeloid genes at the perinecrotic area (Extended Data Fig. 8e,f).

To identify protein markers, we conducted differential expression analysis of the CITE-seq data. Tumor-associated microglia downregulated CX3CR1 but upregulated CD112, CD58 and CD33 (Fig. 6k). Tumor-associated CAMs downregulated CD163 and upregulated CD304 (encoded by NRP1). NRP1 as one of the markers of MOFA2 latent factor 1, was also upregulated in TAMs and microglia, but not Tr. moTAMs. Mass cytometry with additional markers, demonstrated unchanged levels of CD206 but reduced CD169, MS4A7 and CD64 in glioblastoma-associated CAMs (Extended Data Fig. 8g–j). Glioblastoma-associated microglia reduced P2Y12 and increased HLA-DR, CD11c and TGFβ expression.

In summary, glioblastoma contained varied myeloid populations, abundant in hypoxic and necrotic tumor areas. Within microglia, CAMs and TAMs we observed CD304 (encoded by NRP1) upregulation, a feature absent in Tr. moTAMs. LM TAMs displayed enhanced chemokine and integrin expression, potentially linked to LM’s role as gateway to the tumor. Our comprehensive spatial profile deepens the understanding of the glioblastoma-associated myeloid compartment.

Discussion

This study provides a comprehensive exploration and functional analysis of human CNS interfaces during development, homeostasis and glioblastoma (Extended Data Fig. 9). Different scRNA-seq protocols, CITE-seq, mass cytometry and high-resolution spatial transcriptomics unveiled diverse immune-cell interactions within human PC/PV, LM, CP and DM. CAMs consistently exhibit a core macrophage signature (MRC1, LYVE1 and F13A1). CP contains microglia-like Kolmer cells, whereas transitory monocytes predominate in CP and DM, sustaining myeloid-derived engraftment. Histological validation confirms varied CD169 (encoded by SIGLEC1) expression across compartments. Cross-species comparison underscored evolutionary CAM marker conservation in humans and mice. High-resolution spatial transcriptomics implied CAMs crosstalk within their niche. Investigation of in vivo engraftment dynamics at CNS interfaces revealed varying engraftment rates across compartments.

Comparing fetal and adult CNS myeloid cells uncovered distinct populations. Fetal CAMs and microglia were already transcriptionally distinct at pcw 5. Fetal CAMs, microglia and Kolmer cells expressed attenuated immune-mediator proteins and heightened hypoxia markers (CD71). Likewise, glioblastoma-associated LM and PC/PV CAMs exhibited hypoxia-responsive traits, with FTL and NRP1 (encoding CD304) upregulation. High-resolution spatial analysis highlighted a differential distribution of myeloid population across glioblastoma regions.

In summary, our study significantly advances our understanding of human CNS myeloid cells: (1) providing comprehensive transcriptomic and proteomic profiles across human microglia and CAMs along with spatial profiling of the LM and PC/PV niches; (2) offering direct evidence of in vivo peripheral myeloid cells engraftment into human CNS parenchyma and interfaces with concurrent transcriptomic profiling of these cells; (3) highlighting distinct gene and surface protein expression patterns between fetal and postnatal CAMs; and (4) conducting a spatially resolved profiling of the PC/PV and LM glioblastoma microenvironments.

Unlike microglia^10,45 few studies have explored immune cells at human CNS interfaces. Previous comprehensive transcriptomic profiling of CNS interfaces occurred in mice^7,8. Limited CAM populations are reported in separate human and murine PV^4,5,6 and DM⁴⁶ datasets. We have recently shown that PV CAMs are postnatally derived from LM CAMs, implying functional similarity²⁴. Functionally, PV CAMs have been linked to cerebrovascular scavenging and dysfunction^47,48,49,50 supported by our findings of enhanced CD169 scavenging receptor expression in PV.

Distinct CAMs and microglia populations are present in first-trimester human brains. Fetal hypoxia seemed pivotal in shaping myeloid cell phenotypes. Notably, glioblastoma-associated CAMs and TAMs exhibited genes linked to metal iron homeostasis and angiogenesis. Disrupted iron metabolism is a hallmark of cancer-associated hypoxia and anticancer defense^3,41. Previous research has shown fetal-like progenitor cell states in glioblastoma⁵¹. We and others reported hypoxia-responsive states in glioblastoma-associated^32,33,52 and multiple sclerosis-associated⁵³ microglia, highlighting a potentially generic reactive myeloid cell state.

Assessing in vivo engraftment of bone-marrow-derived cells into the adult human CNS is challenging. While microglial engraftment is established in mice^42,54,55, insights into human CNS interface engraftment is lacking. We quantified Y⁺Iba1⁺ donor-derived MΦ in female PBSCT autopsy cases showing varying time-dependent engraftment rates. Transcriptional analysis indicated sustained activation in engrafted cells. Our findings demonstrate human CNS interface myeloid cell engraftment, suggesting PBSCT as a potential CNS-wide myeloid cell replacement strategy for disorders associated with myeloid cell abnormalities¹⁰.

Despite several strengths, our study has limitations. The rarity of CAMs at CNS interfaces necessitated enrichment strategies for adequate cell numbers. Tissue scarcity was another limiting factor, particularly for DM, CP and fetal specimens. Transcriptional convergence of moMΦ and CAMs under pathological conditions renders these cell types difficult to distinguish⁷. We addressed these issues through a semi-supervised cell classification approach using published transcriptional signatures^27,32,44; however, lineage-tracing in transgenic mice surpasses this approach^7,9,22. Additionally, our study is limited by the inherent biases in scRNA-seq⁵⁶.

To conclude, we comprehensively profiled human CNS myeloid cells, focusing on CAMs during homeostasis and disease. We detail the transcriptional spectra of human CAMs and moMΦ states, surface protein profiles, species comparison and perform various validation experiments. Furthermore, we uncover distinct transcriptional profiles of fetal CAMs and identify some similarities to glioblastoma-associated CAMs, driven by different pathophysiological contexts. Notably, myeloid cell engraftment at human CNS interfaces opens the doors to potential CAM replacement therapies for diseases rooted in CAM dysfunction.

Methods

Prospective tissue collection

Experiments on human tissue samples were performed in accordance with the Declaration of Helsinki. Tissues from different sources were analyzed in the present study. The analyzed anatomical compartments include LM (n = 55), PC/PV (n = 52), CP (n = 23), and DM (n = 8). Overall, 61 donors were female and 41 were male. An overview of the sample characteristics is provided in Supplementary Table 20. In short, single-cell and single-nucleus RNA-seq analyses of fresh or fresh-frozen postnatal samples were conducted under the oversight of the local Research Ethics Committee of the University Freiburg Medical Center under the protocol numbers 472/15 and 253/17. Patients or their legal guardians provided written informed consent before tissue collection. Analysis of fetal tissues from the University of Freiburg Medical Center and the Human Developmental Biology Resource (HDBR) were conducted under the oversight of local Research Ethics Committee of the University Freiburg Medical Center under the protocol number 253/17 and the National Research Ethics Service in the United Kingdom. The embryonal and fetal samples were collected under the supervision of specialists and lay persons. Tissue for the HDBR is donated voluntarily after providing informed consent from collaborating clinics in the United Kingdom. Analyses of the adult autopsy tissues were conducted under the oversight of local Research Ethics Committee of the University Freiburg Medical Center under the protocol numbers 10008/09 and 472/15 and the local committees associated with the National Institutes of Health (NIH) bio banks with written informed consent provided by the patients or their legal guardians.

Radiologically healthy or tumor tissues obtained at University of Freiburg Medical Center were placed in ice-cold PBS after surgical removal. Macroscopically, control tissues from focal epilepsy, primary and secondary brain neoplasm surgeries were selected based on radiological appearance and >2 cm distance from the lesion. Control cases that passed pathological examination were included in the study. To assess glioma-associated molecular changes, IDH-wild-type glioblastoma, CNS World Health Organization grade 4 were included (Supplementary Table 20). For single-nucleus RNA-seq, fresh-frozen biobanked tissues were used. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publication^44,57,58. The study was not designed to detect sex differences and no analyses were performed regarding this question. Sex was derived from records.

Immunofluorescence

Data collection and analysis were performed blind to the conditions of the experiments. FFPE sections were blocked and permeabilized with PBS containing 5% normal donkey serum and 0.5% Triton-X 100 for 1 h at room temperature (RT). Primary antibodies were incubated overnight with combinations of Iba1 (WACO), Iba1 (Synaptic Systems), MRC1 (also known as CD206) (Abnova), CD1C (Abcam), CD163 (Sigma-Aldrich), S100A6 (Sigma-Aldrich), SIGLEC1 (also known as CD169) (Sigma-Aldrich) and collagen IV (Sigma-Aldrich,). Secondary antibodies were added as follows: Alexa Fluor 488 1:500 dilution (Thermo Fisher Scientific), Alexa Fluor 568 1:500 dilution (Thermo Fisher Scientific), Alexa Fluor 647 1:500 dilution (Thermo Fisher Scientific) and Alexa Fluor 647 (Jackson ImmunoResearch Laboratories) 1:500 dilution for 2 h at RT. Nuclei were counterstained with DAPI (Carl Roth) when necessary. Images were taken using conventional fluorescence microscopes (Olympus BX-61 and Keyence BZ-9000) and confocal pictures were taken with a Leica TCS SP8 (Leica). Image quantification was conducted in Adobe Photoshop. Outlier values were identified using Grubbs’s test in R and removed.

Tissue dissection for single-cell suspensions and flow cytometry

Before flow cytometry, tissues were digested in Accumax (Sigma-Aldrich) for 30 min at RT with orbital shaking at 800 r.p.m. LM were removed from PC samples and digested separately. All subsequent processing steps were performed on ice. After digestion, tissue samples were suspended in ice-cold Hank’s balanced salt solution (Thermo Fisher Scientific) containing 10 mM glucose (Thermo Fisher Scientific) and 10 mM HEPES (Thermo Fisher Scientific) and mechanically dissociated using glass shearing with a 10-ml Potter–Elvehjem pestle and glass tube homogenizer (Merck). Larger tissue debris were removed by filtering through a 70-µm cell strainer (BD Bioscience). After centrifugation, cell pellets were cryopreserved in fetal calf serum:DMSO (9:1; Merck) until further processing. To minimize batch effects, experiments were conducted in a blocked manner with tissues from different regions and control and diseased tissues processed together using commercial multiplexing kits (10x Genomics, see below for details). Also, where possible several control tissues were processed on the same day. Single-cell sorting was performed on a MoFlo Astrios (Beckman Coulter). Anti-CD45 (clone HI30, APC, BD Bioscience) antibodies were used for droplet-based single-cell RNA-seq (Extended Data Fig. 1a). The following antibodies were used for Cel-Seq2 sorting: anti-CD45 (1:100 dilution, clone HI30, APC, BD Bioscience), anti-MRC1/CD206 (1:400 dilution, clone 15-2, APC-Cy7, BioLegend), anti-CD3 (1:100 dilution, clone SP34-2, PE-Cy7), anti-CD11b (1:800 dilution, clone M1/70, eBioscience), anti-CD19 (1:100 dilution, clone SJ25C1, PE-Cy7, BioLegend) and anti-CD20 (1:400 dilution, clone 2H7, PE-Cy7, BioLegend). Before surface staining, Fc receptors were blocked using Human TruStain FcX (BioLegend). DAPI staining was used for dead cell removal.

Single-nucleus preparations from frozen tissues for flow cytometry

For single-nucleus RNA-seq from fresh-frozen tissues, we utilized an adaptation of the Frankenstein community protocol (www.protocols.io/view/frankenstein-protocol-for-nuclei-isolation-from-f-5jyl8nx98l2w/v3). All steps were performed at 4 °C. Briefly, a tissue piece the size of a grain of rice was mechanically dissociated in Nuclei EZ Lysis Buffer (Merck) using a pellet pestle (Merck). The resulting homogenate was incubated on ice and filtered with a 70-µm cell strainer (Merck). Subsequently, cells were centrifuged at 500g for 5 min and incubated for 5 min with Nuclei EZ Lysis Buffer, centrifuged again and incubated for 5 min in nuclei resuspension buffer (PBS supplemented with 1% BSA solution (Miltenyi Biotec) and 0.2 U μl⁻¹ RNase inhibitor (New England Biolabs)). After two wash steps with nuclei resuspension buffer, nuclei were incubated for 20 min with a master mix containing DAPI (10 µg ml⁻¹), anti-NeuN (clone 1B7, Alexa-647, Novus Biologicals) and anti-Olig2 (clone 211F1.1, Alexa-488, Merck). Single DAPI⁺NeuN⁻Olig2⁻ nuclei were sorted on a MoFlo Astrios (Beckman Coulter) or BD FACSAria III machines (BD Bioscience) to enrich for non-neuroectodermal cell types (Extended Data Fig. 1a). To minimize batch effects, experiments were conducted in a blocked manner with tissues from different regions and control and diseased tissues processed together using commercial multiplexing kits (10x Genomics, see below for details).

Surface protein profiling

CITE-seq was conducted using a commercially available human antibody cocktail (TotalSeq-B Human Universal Cocktail, v.1.0, BioLegend). The lyophilized cocktail was dissolved in Cell Staining Buffer (BioLegend) following the manufacturer’s instructions. After Fc receptor blocking, cell suspensions were incubated with the dissolved antibody solution for 30 min and washed twice.

Sample multiplexing using lipid-conjugated cell-multiplexing oligonucleotides

Sample multiplexing with the 3′ CellPlex kit (10x Genomics) was applied for cost efficiency to pool up to 12 samples per 10x reaction. Cell and nucleus suspensions were incubated on ice for 20 min followed by three washes.

Fixed RNA profiling

Nuclei were extracted from FFPE tissues using the demonstrated protocol supplied by the manufacturer (10x Genomics, CG000632, Rev A). Briefly, three 50-µm sections were cut from FFPE blocks, deparaffinated with xylene (Merck) and rehydrated with an ethanol dilution row. After a wash step with PBS, 100 µl dissociation mix (1 mg ml⁻¹ Liberase (Merck) in RPMI medium (Merck)) was added, and the samples were mechanically dissociated with a pellet pestle followed by enzymatic digestion at 37 °C for 30 min on a radial shaker (800 r.p.m.). Then the sample was triturated with a pipette and passed through a 30-µm filter. Cells were washed with tissue resuspension buffer (0.5 ml) and incubated with an antibody master mix containing anti-NeuN (clone 1B7, Alexa-647, Novus Biologicals) and anti-Olig2 (clone 211F1.1, Alexa-488, Merck). After two additional washes, single DAPI⁺NeuN⁻Olig2⁻ nuclei were sorted on a MoFlo Astrios machine (Beckman Coulter). From each sample, 200,000 nuclei were sorted into LoBind tubes (Eppendorf) and pooled into a control and transplanted sample, respectively.

The following steps were conducted following the 10x Genomics protocol entitled ‘Chromium Fixed RNA Profiling Reagent kits for Single-plexed Samples’ (CG000477 | Rev D). Briefly, sorted nuclei were resuspended in quenching buffer and centrifuged. The sorted pellet was resuspended in hybridization buffer containing Human WTA Probes BC001 (10x Genomics). After 20 h of hybridization at 42 °C the nuclei were repeatedly washed and counted using a hemocytometer. Up to 40,000 nuclei were loaded per reaction. Unused nuclei were stored at −80 °C in storage buffer (0.1 volume Enhancer in Post-Hyb Resuspension Buffer and 10% glycerol). A total of two pooled control and three pooled post-transplantation libraries were prepared from the same nuclei pools due to low yields.

10x Genomics droplet-based single-cell/single-nucleus library preparation

Up to 40,000 sorted cells or nuclei per reaction were loaded on a Chromium controller (10x Genomics). Complementary DNA amplification and library preparation were performed according to the user guide for the Chromium Next GEM Single Cell 3′ Reagent kits v.3.1 (CG000204 or CG000390). Additionally, the 3′ Feature Barcoding kit (10x Genomics) was used for library preparation of multiplexed and CITE-seq samples. The libraries were sequenced on a NextSeq 550 or NextSeq 1000/2000 machines (Illumina) with a sequencing depth appropriate to reach 20,000 reads per cell. The targeted sequencing depth for multiplexing and feature barcoding/CITE-seq libraries were 5,000 reads per cell for each modality. Transcriptome alignment to the GENCODE human genome release 33 was performed with CellRanger v.7.1.0 on a Linux workstation. The CellRanger multi workflow was applied for sample demultiplexing, transcript and surface protein quantification.

Fixed RNA-profiling libraries were generated using the protocol for single-plexed samples (CG000477). Reference probe set alignment and quantification was conducted using CellRanger v.7.1.0 on a Linux workstation with the CellRanger multi workflow.

Integration and analysis of the 3′ mRNA 10x single-cell and single-nucleus transcriptome data

Data collection and analysis were performed in an unsupervised manner, but not blind to the conditions of the experiments. Filtered counts matrices were loaded with Seurat v.4.3.0 (ref. ²⁶). Cells with at least 500 and fewer than 4,000 detected genes and below 20% mitochondrial transcripts were included. Doublet detection and removal were achieved using a combination of the scDblFinder v.1.10.0 and SingleCellExperiment v.1.18.1 packages.

The control 10x dataset consisted of samples with only transcriptome data and samples with transcriptome and CITE-seq information. The glioblastoma data contained transcriptome and CITE-seq information. We used the Azimuth algorithm²⁶ to impute missing CITE-seq information. To this end, data from a reference experiment containing brain PC, LM and CP cells were normalized and scaled with 10,000 most-variable features using the SCTransform Seurat function. The other three control datasets were aligned to this reference using the FindNeighbors and FindTransferAnchors Seurat functions. Next, cell-surface-marker expression values were imputed using the MapQuery Seurat function.

For multimodal mapping, the transcriptome data from different experiments were merged into one Seurat object, then normalized and scaled on 10,000 most-variable features. Then, the different experiments within the Seurat object were integrated using the Harmony R package v.0.1.1 with ‘experiment’ as integration variable⁵⁹. In the next step, the surface receptor data were normalized using a centered log ratio transformation as the normalization method and scaled on all available features. Dimensionality reduction was achieved using the RunPCA Seurat function. Then, multimodal mapping was conducted using the FindMultiModalNeighbors Seurat function with the top 30 components of the transcriptome and top 18 components of the surface receptor data. UMAP embedding was generated from the resulting weighted nearest-neighbor graph. Next, cell clusters were identified using the FindClusters Seurat function with the algorithm parameter set to the smart local moving algorithm and otherwise default settings.

Analysis of the fixed mRNA-profiling single-nucleus data

A total of 9,035 nuclei with above 50 and below 1,000 transcripts were imported into a Seurat object. Doublet detection and removal were achieved using a combination of the scDblFinder v.1.10.0 and SingleCellExperiment v.1.18.1 packages. The data were normalized and scaled using the SCTransform function. After dimensionality reduction with RunPCA, UMAP embedding, nearest-neighbor identification and clustering were conducted from 30 principal components with default parameters.

mCEL-Seq2 single-cell RNA amplification and library preparation

The following antibodies were used for FACS sorting: anti-CD45 (clone HI30, APC, BD Bioscience), anti-CD206 (also known as MRC1, clone 15-2, APC-Cy7, BioLegend), anti-CD3 (clone SP34-2, PE-Cy7, BD Bioscience), anti-CD19 (clone SJ25C1, PE-Cy7, BioLegend) and anti-CD20 (clone 2H7, PE-Cy7, BioLegend). On a MoFlo Astrios machine, CD45⁺CD206⁺Lin⁻ (CD3, CD19 and CD20) cells were sorted into 384-well plates (Bio-Rad Laboratories) (Extended Data Fig. 1b). scRNA-seq was conducted using the mCEL-Seq2 protocol^34,60 on a mosquito nanoliter-scale liquid-handling robot (SPT Labtech). Eight libraries with 192 cells each were sequenced per lane on an Illumina HiSeq 3000 sequencing system (pair-end multiplexing run) at a depth of ~130,000–200,000 reads per cell.

Fastq files were aligned using STAR v.2.7.10a with default parameters to the human GENCODE human genome release 33 (ref. ⁶¹). The left read contains the barcode information; the first six bases represented the unique molecular identifier (UMI) followed by six bases with the cell specific barcode. A poly-T stretch comprised the remainder of the left read that was therefore not used for quantification.

Analysis of the mCEL-Seq2 scRNA-seq data

Data collection and analysis were performed in an unsupervised manner, but not blind to the conditions of the experiments. Overall, 17 mCEL-Seq2 libraries were sequenced and after quality control, 2,490 cells were analyzed. Cells with at least 200 and fewer than 4,000 detected genes and below 20% mitochondrial transcripts were included. Doublet detection and removal were achieved using a combination of the scDblFinder v.1.10.0 and SingleCellExperiment v.1.18.1 packages. Data integration was performed using the Seurat v.4 algorithm running the multimodal reference mapping workflow with 10,000 features²⁶. Briefly, data from each library were normalized and scaled using the SCTransform Seurat function. Then, each library was aligned to multimodal human peripheral blood mononuclear cell (PBMC) data as a reference using the FindNeighbors and FindTransferAnchors Seurat functions. We used PBMCs as a reference to confidently distinguish intravascular cells that are commonly present in human samples that cannot be perfused. This was necessary, as human cDC2 are expected in the CD206 gate due to known high MRC1 expression in these cells⁴⁴. The top 50 components of the supervised principal component dimensionality reduction were used to calculate the UMAP embedding and identify the nearest neighbors for cluster assignment. Clusters were calculated with the resolution parameter set to 0.3.

Re-ordering of clusters

The default cluster ordering of Seurat is based on cluster size with the largest cluster first. To sort the clusters based on transcriptional similarity, we performed hierarchical clustering of the average gene expressions in each cluster. Similarly, cell types were ordered based on transcriptional similarity.

Cell doublet identification and exclusion

Doublets were excluded using the scDblFinder package v.1.10.0. Briefly, the counts slot of the Seurat object was transformed into a SingleCellExperiment object, the scDblFinder function was run on the new object and the original Seurat object was filtered for cells classified as ‘singlet’.

Cell type and state classification based on gene module expression scoring

Cell-cycle scoring and quantification of cell type and functional gene expression modules were conducted based on published gene signatures²⁸ (Supplementary Table 21). Gene module scores were calculated using the UCell R package v.2.0.1 based on the Mann–Whitney U statistic⁶². Briefly, a cell type or cell state-associated gene list was passed to the ScoreSignatures_UCell function. The resulting cell-wise scores were added to the Seurat object using the AddMetadata Seurat function. Cell-cycle modules were scored in Seurat using the CellCycleScoring function, which is a wrapper for the AddModuleScore function. Low-quality cells were identified based on the percentage of mitochondrial across all genes and the expression of the KCNQ1OT1 gene⁴³.

Dendrograms showing the similarity of the identified cell types were prepared using Ward’s method for hierarchical clustering⁶³.

Reference-based cell-type assignment

Cell-type assignment was performed using the Azimuth algorithm²⁶. Briefly, newly generated data were normalized and scaled using the SCTransform Seurat function. Then, they were aligned to multimodal human PBMC data as a reference using the FindNeighbors and FindTransferAnchors Seurat functions. The use of PBMCs as a reference was primarily chosen to confidently assign intravascular cells that are commonly present in human samples that cannot be perfused. The cell-type assignments were added to the original Seurat objects using the MapQuery Seurat function. Tissue-resident cell types not present in the dataset, such as CAMs or microglia were manually added based on the gene module expression of published marker genes (Supplementary Table 21)^7,27,32,44. Also, transitory cells traversing from the blood vessels into the tissue were classified based on the expression of published cell-type homing-gene expression modules (Supplementary Table 21)²⁸. To this end, the cluster-wise expression of the respective gene module calculated using the UCell package was assessed and the respective cell type was assigned accordingly.

Cell-type assignment in the fetal tissues was conducted using reference mapping with a subset of the Braun et al. dataset for PC/PV and Yang et al. dataset for CP cells^2,39.

Differential gene expression analysis

Differentially expressed genes were determined using the FindAllMarkers Seurat function with default settings. For side-by-side comparisons of clusters or conditions were achieved by running the FindMakers Seurat function with logfc.threshold = 0.01 and min.pct = 0.01.

Cluster enrichment analysis

Hypergeometric testing with the phyper base R function was used to calculate enrichment of a given condition in a cluster based on the overall number of cells from this condition in the dataset. The calculated value stands for the probability that number n or more cells from a given condition are found in a cluster by chance. 0.05 was chosen as cutoff for statistical significance. The Benjamini–Hochberg method was used for multiple testing correction of the calculated P values for all conditions. The code for Marimekko plots was modified from R. Scavetta.

MOFA2 latent factor analysis

Latent factors underlying the transcriptional differences between conditions were identified using the MOFA2 algorithm³¹. To this end, we adopted the ‘integration of a time-course single-cell RNA-seq dataset’ workflow. mCEL-Seq2 as specified under the following vignette: raw.githack.com/bioFAM/MOFA2_tutorials/master/R_tutorials/scRNA_gastrulation.html. Briefly, the cell types of interest were extracted from the Seurat object (CAMs, microglia and transitory monocytes (clusters 9 and 21) from the control and CAMs, microglia, TAMs and Trans. moTAMs from the tumor dataset, respectively). The data were normalized and rescaled. Then, the MOFA object was created from the RNA slot of the Seurat object and grouped by the compartments for the control data and composite variables consisting of the compartment and cell types for the tumor data. The convergence mode was set to fast. The number of latent factors was set to 10. After convergence of the MOFA model, the variance explained was visualized using the plot_variance_explained MOFA2 function. The weights for each latent factor were extracted using the get_weights function. The genes with the top 100 weights were extracted and comparative Gene Ontology analysis between the latent factors was performed using the compareCluster function from the clusterProfiler R package v.4.4.4 (ref. ⁶⁴). The factor-wise Gene Ontology terms were visualized using the dot-plot function of the clusterProfiler package.

Species comparison between human and mouse

For cross-species analysis, a previously published control CAMs dataset⁷ was analyzed using Seurat v.4 in an analogous way as the human data in the present study. Differentially expressed genes were separately assessed between CAMs and microglia in mice and humans using the FindMarkers Seurat function. Orthologous genes between human and murine samples were obtained using the biomaRt v.2.52.0 R package. For visualization, a scatter-plot shows the correlation of the average log₂ fold change values between both datasets with positive values showing genes differentially expressed in CAMs and negative values for microglia genes. Human log₂ fold change values are shown on the x axis and the murine ones are shown on the y axis.

Pseudotime analysis

Pseudotime analysis was performed using the StemID analysis within the RaceID v.0.2.4 and FateID v.0.2.2 R packages^34,43. We followed the vignette published by the package authors. Briefly, the gene the counts object of the Seurat object was transformed into a RaceID object. Then a lineage graph was calculated. Based on the suggested links a trajectory was chosen. Genes with at least one UMI in at least ten cells were included in the trajectory. Genes with a correlation coefficient of 0.85 or above were summarized into gene expression modules and modules of at least five cells were included in the analysis.

Gene Ontology enrichment analysis

Gene Ontology term analysis was conducted using the clusterProfiler v.4.4.4 R package⁶⁴. Cluster marker genes were obtained using the FindAllMarkers Seurat function. The top 25 of these markers were transformed into entrez IDs and passed to the compareCluster function of the clusterProfiler package.

Data analysis and visualization

Data collection and analysis were performed in an unsupervised manner, but not blind to the conditions of the experiments. Data analysis was conducted using the R programming language v.4.0.2. The tidyverse R package was used for data processing and visualization (CRAN.R-project.org/package=tidyverse).

Cell–cell interaction analysis

Cell–cell interaction analysis between different cell types in the single-nucleus RNA-seq samples was performed using the NICHES R package v.1.0.0 (ref. ⁴⁰). following a published vignette at github.com/msraredon/NICHES/blob/master/vignettes/01%20NICHES%20Spatial.Rmd. Briefly, the combined immune-cell object containing published fetal PC/PV and postnatal CP immune cells^2,39 was merged with the non-immune single-nucleus data from Fig. 5b. To account for the abundance of neuronal cell types in the non-immune dataset, it was downsampled to up to 500 distinct cells per cell type. The resulting Seurat object was passed to the RunNICHES function and the SystemToCell interactions were analyzed using the fantom5 ligand–receptor database. These data consist of the interactions between the respective cell types and all other cell types in the data. We were particularly interested in the differential interactions of pre- and postnatal cell macrophage subsets. To this end we ran differential expression analysis of ligand–receptor pairs using the FindMarkers R function and visualized to top differentially expressed pairs using volcano plots with the EnhancedVolcano R package v.1.14.0.

In situ analysis using CARTANA technology

Four frozen OCT-embedded control and two glioblastoma samples were cryosectioned (10-μm thickness) and placed onto SuperFrost Plus glass slides (Thermo Fisher) and shipped on dry ice to CARTANA (part of 10x Genomics) for processing.

Samples were fixed (with 4% formaldehyde) permeabilized (with 0.1 mg ml⁻¹ ± pepsin in 0.1 M HCl (P7012 Sigma-Aldrich)) before library preparation. For tissue section mounting, Slow Fade Antifade Mountant (Thermo Fisher) was used for optimal handling and imaging.

For library preparation, chimeric padlock probes (targeting directly RNA and containing an anchor sequence as well as a gene-specific barcode) for three predefined panels (CNS Glia, Immune General and Immune Oncology) as well as a custom panel of 50 genes (Supplementary Table 10 for all genes) were hybridized overnight at 37 °C, then ligated before the rolling circle amplification is performed overnight at 30 °C using the HS Library Preparation kit for CARTANA technology and following manufacturer’s instructions. All incubations were performed in SecureSealTM chambers (Grace Biolabs). Before final library preparation, optimal RNA integrity and assay conditions were assessed using Malat1 and Rplp0 housekeeping genes only using the same protocol.

Quality control of the library preparation was performed by applying anchor probes to detect simultaneously all rolling circle amplification products from all genes in all panels. Anchor probes are labeled probes with Cy5 fluorophore (excitation at 650 nm and emission at 670 nm).

All samples passed the quality control and were sent to CARTANA Sweden, for in situ barcode sequencing, imaging and data processing. Briefly, adaptor probes and sequencing pools (containing four different fluorescent labels: Alexa Fluor 488, Cy3, Cy5 and Alexa Fluor 750) were hybridized to the padlock probes to detect the gene-specific barcodes, through a sequence-specific signal for each gene-specific rolling circle amplification product. This was followed by imaging and performed six times in a row to allow for the decoding of all genes in the panel.

Raw data consisting of ×20 or ×40 images from five fluorescent channels (DAPI, Alexa Fluor 488, Cy3, Cy5 and Alexa Fluor 750) were each taken as a z stack and flattened to two dimensions using maximum intensity projection. After image processing and decoding, the results were summarized in a csv file and gene plots were generated using MATLAB.

Image segmentation of the ISS data was performed with the docker image of the Baysor tool (v.0.4.2.)⁶⁵, using the following modified parameters for the configuration file: min-molecules-per-cell,10; scale, 50; scale-std, 50%; and min-molecules-per-segment, 2. The Baysor analysis was also run using a prior segmentation performed in ImageJ Fiji. The segmentation was performed with the watershed algorithm on an DAPI image indicating nuclei position, using local thresholding when needed. The parameter value used for the prior segmentation confidence was 0.8.

Analysis of the in situ sequencing data

The cell counts matrix and the cell coordinates obtained from image segmentation analysis with Baysor were loaded into the SPAtial Transcriptomic Analysis R package (SPATA2) v.0.1.0. with the initiateSpataObject_CountMtr function. Tissue segmentation of anatomical areas in control samples was anatomically drawn in the SPATA2 viewer after calling the createSegmentation function. For hypoxic areas in the tumor, we visualized in the gene set RCTM_CELLULAR_RESPONSE_TO_HYPOXIA in the SPATA2 viewer. Then, the segments were manually drawn around areas with high expression of the gene set. Similarly, spatial trajectory analysis was performed using the createTrajectories function and manually drawing a trajectory from an area of cellular tumor into the hypoxic area. The top ten up- and downregulated genes along the trajectory were identified using the assessTrajectoryTrends and extracted using the filterTrajectoryTrends functions. Visualization of the resulting smoothed gene expressions was conducted using the plotTrajectoryHeatmap function. Representative genes were visualized using the plotSurfaceComparison function.

Cell-type classification was conducted using label transfer via the Azimuth algorithm. Briefly, the SPATA2 object was converted into a Seurat object using the transformSpataToSeurat function. The data were normalized and scaled on all available genes. For the control samples, LM and PC/PV cells were separately classified with single-nucleus RNA-seq datasets from the LM and PC/PV. The reference datasets were subset to the genes contained in the ISS panel and nuclei with more than ten transcripts were kept. Then, the ISS dataset was integrated with the reference dataset using the FindTransferAnchors Seurat function with the dims parameter set to 1:10. Cell-type labels were transferred using the TransferData with the dims parameter set to 1:10. For the glioblastoma samples, the strategy was slightly adjusted. To account for the predominance of tumor cells in the reference single-nucleus sequencing dataset, it was downsampled to up to 200 distinct cells per cell type. Then we ran the FindTransferAnchors and TransferData Seurat functions with dims set to 1:30. The cell-type classifications were controlled based on differentially expressed genes and manually adjusted, if needed.

Spatial neighborhood analysis

Spatial neighborhood enrichment analysis was performed using the Squidpy algorithm v.1.3.0 (ref. ³⁶) as previously described (www.sc-best-practices.org/spatial/neighborhood.html). Neighborhood enrichment scores of the respective cell types were calculated using the nhood_enrichment function. Subsequently, the interaction_matrix function was used to obtain spatial interaction scores of the cell types. Plotting was performed using Scanpy v.1.9.3 (ref. ⁶⁶) functionality.

CosMx tissue processing, image acquisition, cell segmentation and data analysis for 1,000-plex RNA profiling

FFPE tissue blocks with control and tumor samples were sectioned to consecutive 5-µm slices using a microtome and shipped to NanoString Technologies for processing. Sample processing, staining, imaging and cell segmentation were performed as previously described⁶⁷. Briefly, tissue sections were placed onto VWR Superfrost Plus Micro slides for optimal adherence. Slides were then dried at 37 °C overnight, followed by deparaffinization, antigen retrieval and proteinase-mediated permeabilization (nanostring.com/products/cosmx-spatial-molecular-imager/single-cell-imaging-overview). Then,1 nM RNA-ISH probes were applied for hybridization at 37 °C overnight. After a stringent wash, a flow cell was assembled on top of the slide and cyclic RNA readout on CosMx was performed (16-digit encoding strategy). After all cycles were completed, additional visualization markers for morphology and cell segmentation were added, including pan-cytokeratin, CD45, CD3, CD298/B2M and DAPI. Twenty-four 0.985 mm × 0.657 mm fields of view (FOVs) were selected for data collection in each slice. The CosMx optical system has an epifluorescent configuration based on a customized water objective (×13, NA 0.82) and uses widefield illumination with a mix of lasers and light-emitting diodes (385 nm, 488 nm, 530 nm, 590 nm and 647 nm) that allow imaging of DAPI, Alexa Fluor 488, Atto-532, Dyomics Dy-605 and Alexa Fluor 647, as well as removal of photocleavable dye components. The camera was a FLIR BFS-U3_200S6M-C based on the IMX183 Sony industrial CMOS sensor (pixel size 180 nm). A three-dimensional multichannel image stack (nine frames) was obtained at each FOV location, with a step size of 0.8 µm. Registration, feature extraction, localization, decoding of the presence individual transcripts and machine-learning-based cell segmentation (developed upon Cellpose) were performed as previously described⁶⁷. The final segmentation mapped each transcript in the registered images to the corresponding cell, as well as to subcellular compartments (nuclei, cytoplasm and membrane), where the transcript is located.

Tissue segmentation and trajectory analysis was conducted similarly to the ISS data described above using the SPATA2 R package. Also, cell-type classification was conducted with the reference single-nucleus sequencing datasets subset to the genes contained in the 1,000-plex gene panel.

Spatially resolved cell–cell interaction analysis

Spatially resolved cell–cell interaction analysis between different cell types was performed on the GeoMx data using the NICHES R package v.1.0.0. following a published vignette at github.com/msraredon/NICHES/blob/master/vignettes/01%20NICHES%20Spatial.Rmd. Briefly, missing gene expression values were imputed using the Adaptively-thresholded Low Rank Approximation algorithm on the Seurat object⁶⁸. The resulting Seurat object was passed to the RunNICHES function and the NeighborhoodToCell interactions were analyzed using the fantom5 ligand–receptor database. These data consist of the interaction between the respective cell types in the data and their neighboring cells. The differentially expressed interactions were visualized.

Human sex-mismatched PBSCT autopsy cases

All human brain autopsy samples were derived from the case archive of the Institute of Neuropathology, University of Freiburg. Brains and DM of autopsy cases were transferred into 4% paraformaldehyde within less than 48 h after death and fixed for at least 1 week. After fixation, representative tissue from several brain regions of the left hemisphere was dissected and embedded in paraffin. Among the 15 female patients with a history of male donor-derived PBSCT, paraffin-embedded samples the frontal cortex and cerebellum were available in 14 cases and the hippocampus was available in 13 cases. Hippocampal samples derived from nine patients contained parts of the CP. In nine cases, LM were amenable for analysis as they were well preserved and confounding neuropathological diagnosis was absent. In five female sex-mismatched PBSCT patients and four control cases, the DM adjacent to the frontal branch of the medial meningeal artery was resected and processed for further histological analysis.

Chromogenic in situ hybridization of human brain samples

To investigate CAM engraftment, IHC and CISH labeling the Y chromosome in samples of female sex-mismatched PBSCT patients was performed using 10-µm thick sections as previously described⁴². Briefly, IHC was carried out using Liquid Permanent Red Substrate-Chromogen (Agilent Dako) for antigen visualization. Subsequently, CISH was performed using the ZytoDot CISH Implementation kit (ZytoVision) according to the manufacturer’s instructions with the following modifications. Sections were incubated in EDTA at 95 °C for 15 min and treated with pepsin solution at 37 °C for 6 min. After dehydration, 12 µl of ZytoDot CEN Yq12 digoxigenin-linked probe (ZytoVision) was added for at least 20 h at 37 °C. After washing and blocking steps, sections were incubated in mouse anti-digoxigenin antibody solution, treated with HRP-conjugated anti-mouse antibody at 37 °C for 30 min and bound to DAB at 37 °C for 45 min. Nuclei were counterstained with hematoxylin.

Engraftment analysis

Sections were analyzed using a MikroCam II with a UPlan FLN ×40/0.75 NA objective on a BX40 microscope (Olympus). To scan for the Y chromosome the focus plain was carefully moved through the whole nucleus of each cell analyzed.

First, the detection rate of Y chromosomes was determined for each CAM compartment in n = 3 male control samples. For each compartment and marker, several FOVs were analyzed. The mean Y chromosome detection rate for each CNS myeloid cell niche and marker was calculated.

Next, samples derived from female sex-mismatched PBSCT cases were analyzed. In each compartment, at least 50 CAMs per patient were analyzed. Myeloid cell engraftment rates were calculated by dividing the resulting percentage of Y⁺ cells in each compartment by the corresponding Y chromosome detection rate previously determined in male control tissue.

T₅₀, the duration after PBSCT required for a turnover of 50% of CAMs by donor-derived cells, was inferred by a linear regression analysis between log₁₀-transformed time interval after PBSCT and Pearson’s correlation coefficient. The values for the confidence intervals were added based on the estimated upper and lower bounds of the correlation coefficient.

Assessment of marker regulation in donor- and host-derived cells

To investigate expression of CAM markers in donor-derived cells of the LM and PV space, three patients with long survival after PBSCT and consecutively high numbers of engrafting donor-derived cells were chosen. In these patients, at least 100 perivascular and leptomeningeal Y⁺ cells pooled from cortical, hippocampal and cerebellar samples were assessed for Iba1 (Abcam, clone EPR 16588), CD206 (Abnova, clone 5C11) and Siglec1 expression. For microglia quantification, cortical samples were assessed for Iba1 (Abcam, clone EPR 16588), P2RY12 (Sigma-Aldrich, polyclonal), TMEM119 (Abcam, polyclonal) and GLUT5 (Sigma-Aldrich, polyclonal).

Intracellular barcoding for mass cytometry

Percoll-isolated myeloid cells were fixed with fixation/stabilization buffer⁶⁹ (Smart Tube) and frozen at −80 °C until analysis by mass cytometry. Cells were thawed and subsequently stained with premade combinations of six different palladium isotopes: 102 Pd, 104 Pd, 105 Pd, 106 Pd, 108 Pd and 110 Pd (Cell-ID 20-plex Pd Barcoding kit, Fluidigm). This multiplexing kit applies a 6-choose-3 barcoding scheme that results in 20 different combinations of three Pd isotopes. After 30 min staining (at RT), individual samples were washed twice with cell staining buffer (0.5% bovine serum albumin in PBS, containing 2 mM EDTA). All samples were pooled together, washed and further stained with antibodies.

Antibodies

Anti-human antibodies (Supplementary Table 9) were purchased either preconjugated to metal isotopes (Fluidigm) or from commercial suppliers in purified form and conjugated in-house using the MaxPar X8 kit (Fluidigm) according to the manufacturer’s protocol. For surface and intracellular staining, after cell barcoding, washing and pelleting, the combined samples were resuspended in 100 µl antibody cocktail against surface markers (Supplementary Table 9) and incubated for 30 min at 4 °C. Then, cells were washed twice with cell staining buffer. For intracellular staining, the stained (non-stimulated) cells were then incubated in fixation/permeabilization buffer (Fix/Perm Buffer, eBioscience) for 60 min at 4 °C. Cells were then washed twice with permeabilization buffer (eBioscience). The samples were then stained with antibody cocktails against intracellular molecules in permeabilization buffer for 1 h at 4 °C. Cells were subsequently washed twice with permeabilization buffer and incubated overnight in 2% methanol-free formaldehyde solution. Fixed cells were then washed and resuspended in 1 ml iridium intercalator solution (Fluidigm) for 1 h at RT. Next, the samples were washed twice with cell staining buffer and then twice with ddH₂O (Fluidigm). Cells were pelleted and kept at 4 °C until CyTOF measurement.

CyTOF measurement

Cells were analyzed using a CyTOF2 upgraded to Helios specifications, with software v.6.5.236. The instrument was tuned according to the manufacturer’s instructions with tuning solution (Fluidigm) and measurement of EQ four element calibration beads (Fluidigm) containing 140/142Ce, 151/153Eu, 165Ho and 175/176Lu served as a quality control for sensitivity and recovery. Directly before analysis, cells were resuspended in ddH₂O, filtered (20 µm Celltrix, Sysmex), counted and adjusted to 3–5 × 10⁵ cells ml⁻¹. EQ four element calibration beads were added at a final concentration of 1:10 of the sample volume to be able to normalize the data to compensate for signal drift and day-to-day changes in instrument sensitivity.

Samples were acquired with a flow rate of 300–400 events s⁻¹. The lower convolution threshold was set to 400, with noise reduction mode on and cell definition parameters set at event duration of 10–150. The resulting flow cytometry standard (FCS) files were normalized and randomized using the CyTOF software’s internal FCS-Processing module on the non-randomized (‘original’) data. Settings were used according to the default settings in the software with time interval normalization (100 s per minimum of 50 beads) and passport v.2. Intervals with fewer than 50 beads per 100 s were excluded from the resulting fcs file.

Mass cytometry data processing and analysis

Cytobank was used for initial manual gating on intact single cells. Nucleated single cells were manually gated by DNA intercalators ¹⁹¹Ir/¹⁹³Ir and event length. For de-barcoding, Boolean gating was used to deconvolute individual sample according to the barcode combination. For gated cells from different individuals, expression levels of each marker were assessed and visualized in dot plots and/or histograms. After de-barcoding, each sample was exported as individual .fcs file from Cytobank. We performed the visualization and clustering analysis of the data using the k-nearest-neighbor density-based algorithm X-shift on the VorteX Clustering Environment (github.com/nolanlab/vortex/).

Data analysis and visualization

Data analysis and visualization was mainly conducted in the R v.4.2.0 programming environment using the tidyverse package suite v.2.0.0. and specialized visualization packages, including ComplexHeatmap v.2.12.1 (refs. ^70,71,72,73).

Statistics and reproducibility

No statistical method was used to predetermine sample size. The experiments were not randomized. The investigators were blinded during the analysis of microscopy analyses. For single-cell and single-nucleus RNA-seq, cells with fewer than 500 and more than 4,000 detected genes and more than 20% mitochondrial transcripts were excluded. The presented CITE-seq data only contain cells with more than 50 counts per cell. For Cel-Seq2 data, cells with fewer than 200 and more than 4,000 detected genes and more than 20% mitochondrial transcripts were excluded. Furthermore, clusters with biologically uninformative low-quality cells were excluded after evaluation. The exclusion criteria were in line with previously applied strategies for Cel-Seq2 data^32,43. For fixed RNA-profiling samples, cells with fewer than 50 and more than 1,000 detected genes were excluded. ISS data only contain cells with five or more counts per cell.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

The processed data for this project are available under GSE245311. The raw sequencing files are access-restricted and can be accessed at the European Genome–phenome Archive under the accession number EGAS50000000030 (https://ega-archive.org/studies/EGAS50000000030). Access to the data will require a Data Transfer Agreement. The raw data for the mass cytometry experiments can be found under flow repository ID FR-FCM-Z6S6. Published counts data²⁶ for reference mapping of the immune cells were downloaded from atlas.fredhutch.org/data/nygc/multimodal/pbmc_multimodal.h5seurat. Published counts data² for control human CP single-nucleus RNA-seq samples were downloaded under the accession code GSE159812. Published counts³⁹ data for reference mapping and comparative analyses for prenatal immune cells were downloaded from github.com/linnarsson-lab/developing-human-brain/. Published counts data⁴⁴ for comparative analyses of the glioblastoma samples were downloaded from www.brainimmuneatlas.org/data_files/toDownload/filtered_feature_bc_matrix_HumanGBMciteSeq.zip and their metadata are at www.brainimmuneatlas.org/data_files/toDownload/annot_Human_TAM_DC_Mono_citeSeq.csv. Source data are provided with this paper.

Code availability

The computer code for this project can be found and explored at www.brain-immunity.de.

References

Alves de Lima, K., Rustenhoven, J. & Kipnis, J. Meningeal immunity and its function in maintenance of the central nervous system in health and disease. Annu. Rev. Immunol. 38, 597–620 (2020).
Article CAS PubMed Google Scholar
Yang, A. C. et al. Dysregulation of brain and choroid plexus cell types in severe COVID-19. Nature 595, 565–571 (2021).
Article CAS PubMed PubMed Central Google Scholar
Chi, Y. et al. Cancer cells deploy lipocalin-2 to collect limiting iron in leptomeningeal metastasis. Science 369, 276–282 (2020).
Article CAS PubMed PubMed Central Google Scholar
Yang, A. C. et al. A human brain vascular atlas reveals diverse mediators of Alzheimer’s risk. Nature https://doi.org/10.1038/s41586-021-04369-3 (2022).
Winkler, E. A. et al. A single-cell atlas of the normal and malformed human brain vasculature. Science 375, eabi7377 (2022).
Article CAS PubMed PubMed Central Google Scholar
Vanlandewijck, M. et al. A molecular atlas of cell types and zonation in the brain vasculature. Nature 554, 475–480 (2018).
Article CAS PubMed Google Scholar
Jordão, M. J. C. et al. Single-cell profiling identifies myeloid cell subsets with distinct fates during neuroinflammation. Science 363, eaat7554 (2019).
Article PubMed Google Scholar
Van Hove, H. et al. A single-cell atlas of mouse brain macrophages reveals unique transcriptional identities shaped by ontogeny and tissue environment. Nat. Neurosci. 22, 1021–1035 (2019).
Article PubMed Google Scholar
Goldmann, T. et al. Origin, fate and dynamics of macrophages at central nervous system interfaces. Nat. Immunol. 17, 797–805 (2016).
Article CAS PubMed PubMed Central Google Scholar
Prinz, M., Masuda, T., Wheeler, M. A. & Quintana, F. J. Microglia and central nervous system-associated macrophages—from origin to disease modulation. Annu. Rev. Immunol. 39, 251–277 (2021).
Article CAS PubMed PubMed Central Google Scholar
Kierdorf, K., Masuda, T., Jordão, M. J. C. & Prinz, M. Macrophages at CNS interfaces: ontogeny and function in health and disease. Nat. Rev. Neurosci. 20, 547–562 (2019).
Article CAS PubMed Google Scholar
Prinz, M., Jung, S. & Priller, J. Microglia biology: one century of evolving concepts. Cell 179, 292–311 (2019).
Article CAS PubMed Google Scholar
Herz, J., Filiano, A. J., Smith, A., Yogev, N. & Kipnis, J. Myeloid cells in the central nervous system. Immunity 46, 943–956 (2017).
Article CAS PubMed PubMed Central Google Scholar
Prinz, M., Erny, D. & Hagemeyer, N. Ontogeny and homeostasis of CNS myeloid cells. Nat. Immunol. 18, 385–392 (2017).
Article CAS PubMed Google Scholar
Colonna, M. & Butovsky, O. Microglia function in the central nervous system during health and neurodegeneration. Annu. Rev. Immunol. 35, 441–468 (2017).
Article CAS PubMed PubMed Central Google Scholar
Munro, D. A. D., Movahedi, K. & Priller, J. Macrophage compartmentalization in the brain and cerebrospinal fluid system. Sci. Immunol. 7, eabk0391 (2022).
Article CAS PubMed Google Scholar
Aguzzi, A., Barres, B. A. & Bennett, M. L. Microglia: scapegoat, saboteur, or something else? Science 339, 156–161 (2013).
Article CAS PubMed PubMed Central Google Scholar
Gomez Perdiguero, E. et al. Tissue-resident macrophages originate from yolk-sac-derived erythro-myeloid progenitors. Nature 518, 547–551 (2015).
Article PubMed Google Scholar
Kierdorf, K. et al. Microglia emerge from erythromyeloid precursors via Pu.1- and Irf8-dependent pathways. Nat. Neurosci. 16, 273–280 (2013).
Article CAS PubMed Google Scholar
Schulz, C. et al. A lineage of myeloid cells independent of myb and hematopoietic stem cells. Science 336, 86–90 (2012).
Article CAS PubMed Google Scholar
Tay, T. L. et al. A new fate mapping system reveals context-dependent random or clonal expansion of microglia. Nat. Neurosci. 20, 793–803 (2017).
Article CAS PubMed Google Scholar
Masuda, T. et al. Novel Hexb-based tools for studying microglia in the CNS. Nat. Immunol. 21, 802–815 (2020).
Article CAS PubMed Google Scholar
Mrdjen, D. et al. High-dimensional single-cell mapping of central nervous system immune cells reveals distinct myeloid subsets in health, aging, and disease. Immunity 48, 380–395 (2018).
Article CAS PubMed Google Scholar
Masuda, T. et al. Specification of CNS macrophage subsets occurs postnatally in defined anatomical niches. Nature https://doi.org/10.1038/s41586-022-04596-2 (2022).
Siletti, K. et al. Transcriptomic diversity of cell types across the adult human brain. Science 382, eadd7046 (2023).
Article CAS PubMed Google Scholar
Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587 (2021).
Article CAS PubMed PubMed Central Google Scholar
Sankowski, R. et al. Commensal microbiota divergently affect myeloid subsets in the mammalian central nervous system during homeostasis and disease. EMBO J. 40, e108605 (2021).
Article CAS PubMed PubMed Central Google Scholar
Kowalczyk, M. S. et al. Single-cell RNA-seq reveals changes in cell cycle and differentiation programs upon aging of hematopoietic stem cells. Genome Res. 25, 1860–1872 (2015).
Article CAS PubMed PubMed Central Google Scholar
Travaglini, K. J. et al. A molecular cell atlas of the human lung from single-cell RNA sequencing. Nature 587, 619–625 (2020).
Article CAS PubMed PubMed Central Google Scholar
Keren-Shaul, H. et al. A unique microglia type associated with restricting development of Alzheimer’s disease. Cell 169, 1276–1290 (2017).
Article CAS PubMed Google Scholar
Argelaguet, R. et al. MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data. Genome Biol. 21, 111 (2020).
Article PubMed PubMed Central Google Scholar
Sankowski, R. et al. Mapping microglia states in the human brain through the integration of high-dimensional techniques. Nat. Neurosci. 22, 2098–2110 (2019).
Article CAS PubMed Google Scholar
Friedrich, M. et al. Tryptophan metabolism drives dynamic immunosuppressive myeloid states in IDH-mutant gliomas. Nat. Cancer 2, 723–740 (2021).
Article CAS PubMed Google Scholar
Herman, J. S., Sagar & Grün, D. FateID infers cell fate bias in multipotent progenitors from single-cell RNA-seq data. Nat. Methods 15, 379–386 (2018).
Article CAS PubMed Google Scholar
Qian, X. et al. Probabilistic cell typing enables fine mapping of closely related cell types in situ. Nat. Methods 17, 101–106 (2020).
Article CAS PubMed Google Scholar
Palla, G. et al. Squidpy: a scalable framework for spatial omics analysis. Nat. Methods 19, 171–178 (2022).
Article CAS PubMed PubMed Central Google Scholar
Monier, A. et al. Entry and distribution of microglial cells in human embryonic and fetal cerebral cortex. J. Neuropathol. Exp. Neurol. 66, 372–382 (2007).
Article PubMed Google Scholar
Kracht, L. et al. Human fetal microglia acquire homeostatic immune-sensing properties early in development. Science 369, 530–537 (2020).
Article CAS PubMed Google Scholar
Braun, E. et al. Comprehensive cell atlas of the first-trimester developing human brain. Science 382, eadf1226 (2023).
Article CAS PubMed Google Scholar
Raredon, M. S. B. et al. Comprehensive visualization of cell–cell interactions in single-cell and spatial transcriptomics with NICHES. Bioinformatics 39, btac775 (2023).
Article CAS PubMed Google Scholar
Torti, S. V. & Torti, F. M. Iron and cancer: more ore to be mined. Nat. Rev. Cancer 13, 342–355 (2013).
Article CAS PubMed PubMed Central Google Scholar
Shemer, A. et al. Engrafted parenchymal brain macrophages differ from microglia in transcriptome, chromatin landscape and response to challenge. Nat. Commun. 9, 5206 (2018).
Article CAS PubMed PubMed Central Google Scholar
Grün, D. et al. De novo prediction of stem cell identity using single-cell transcriptome data. Cell Stem Cell 19, 266–277 (2016).
Article PubMed PubMed Central Google Scholar
Pombo Antunes, A. R. et al. Single-cell profiling of myeloid cells in glioblastoma across species and disease stage reveals macrophage competition and specialization. Nat. Neurosci. 24, 595–610 (2021).
Article CAS PubMed Google Scholar
Masuda, T. et al. Spatial and temporal heterogeneity of mouse and human microglia at single-cell resolution. Nature 566, 388–392 (2019).
Article CAS PubMed Google Scholar
Wang, A. Z. et al. Single-cell profiling of human dura and meningioma reveals cellular meningeal landscape and insights into meningioma immune response. Genome Med. 14, 49 (2022).
Article CAS PubMed PubMed Central Google Scholar
Pires, P. W. et al. Improvement in middle cerebral artery structure and endothelial function in stroke-prone spontaneously hypertensive rats after macrophage depletion. Microcirculation 20, 650–661 (2013).
Article CAS PubMed PubMed Central Google Scholar
Faraco, G. et al. Perivascular macrophages mediate the neurovascular and cognitive dysfunction associated with hypertension. J. Clin. Invest. 126, 4674–4689 (2016).
Article PubMed PubMed Central Google Scholar
Santisteban, M. M. et al. Endothelium-macrophage crosstalk mediates blood-brain barrier dysfunction in hypertension. Hypertension 76, 795–807 (2020).
Article CAS PubMed Google Scholar
Drieu, A. et al. Parenchymal border macrophages regulate the flow dynamics of the cerebrospinal fluid. Nature 611, 585–593 (2022).
Article CAS PubMed PubMed Central Google Scholar
Neftel, C. et al. An integrative model of cellular states, plasticity, and genetics for glioblastoma. Cell 178, 835–849 (2019).
Article CAS PubMed PubMed Central Google Scholar
Darmanis, S. et al. Single-cell RNA-seq analysis of infiltrating neoplastic cells at the migrating front of human glioblastoma. Cell Rep. 21, 1399–1410 (2017).
Article CAS PubMed PubMed Central Google Scholar
Absinta, M. et al. A lymphocyte–microglia–astrocyte axis in chronic active multiple sclerosis. Nature 597, 709–714 (2021).
Article CAS PubMed PubMed Central Google Scholar
Bennett, F. C. et al. A combination of ontogeny and CNS environment establishes microglial identity. Neuron 98, 1170–1183 (2018).
Article CAS PubMed PubMed Central Google Scholar
Xu, Z. et al. Efficient strategies for microglia replacement in the central nervous system. Cell Rep. 32, 108041 (2020).
Article CAS PubMed Google Scholar
Sankowski, R., Monaco, G. & Prinz, M. Evaluating microglial phenotypes using single-cell technologies. Trends Neurosci. 45, 133–144 (2022).
Article CAS PubMed Google Scholar
Schirmer, L. et al. Neuronal vulnerability and multilineage diversity in multiple sclerosis. Nature 573, 75–82 (2019).
Article CAS PubMed PubMed Central Google Scholar
Velmeshev, D. et al. Single-cell genomics identifies cell type–specific molecular changes in autism. Science 364, 685–689 (2019).
Article CAS PubMed PubMed Central Google Scholar
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods 16, 1289–1296 (2019).
Article CAS PubMed PubMed Central Google Scholar
Hashimshony, T. et al. CEL-Seq2: sensitive highly-multiplexed single-cell RNA-seq. Genome Biol. 17, 77 (2016).
Article PubMed PubMed Central Google Scholar
Kaminow, B., Yunusov, D. & Dobin, A. STARsolo: accurate, fast and versatile mapping/quantification of single-cell and single-nucleus RNA-seq data. Preprint at bioRxiv https://doi.org/10.1101/2021.05.05.442755 (2021).
Andreatta, M. & Carmona, S. J. UCell: robust and scalable single-cell gene signature scoring. Comput. Struct. Biotechnol. J. 19, 3796–3798 (2021).
Article CAS PubMed PubMed Central Google Scholar
Ward, J. H. Hierarchical grouping to optimize an objective function. J. Am. Stat. Assoc. 58, 236–244 (1963).
Article Google Scholar
Wu, T. et al. clusterProfiler 4.0: a universal enrichment tool for interpreting omics data. Innovation 2, 100141 (2021).
CAS PubMed PubMed Central Google Scholar
Petukhov, V. et al. Cell segmentation in imaging-based spatial transcriptomics. Nat. Biotechnol. 40, 345–354 (2022).
Article CAS PubMed Google Scholar
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).
Article PubMed PubMed Central Google Scholar
Stringer, C., Wang, T., Michaelos, M. & Pachitariu, M. Cellpose: a generalist algorithm for cellular segmentation. Nat. Methods 18, 100–106 (2021).
Article CAS PubMed Google Scholar
Linderman, G. C. et al. Zero-preserving imputation of single-cell RNA-seq data. Nat. Commun. https://doi.org/10.1038/s41467-021-27729-z (2022).
Böttcher, C. et al. Human microglia regional heterogeneity and phenotypes determined by multiplexed single-cell mass cytometry. Nat. Neurosci. 22, 78–90 (2019).
Article PubMed Google Scholar
R Core Team R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2013).
Wickham, H. et al. Welcome to the Tidyverse. J. Open Source Softw. 4, 1686 (2019).
Article Google Scholar
Gu, Z. Complex heatmap visualization. iMeta https://doi.org/10.1002/imt2.43 (2022).
Gu, Z., Eils, R. & Schlesner, M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics 32, 2847–2849 (2016).
Article CAS PubMed Google Scholar

Download references

Acknowledgements

We thank J. Bodinek-Wersing, T. El Gaz, E. Barleon, S. Wundt and A. Frömming for their excellent technical support. We thank the Freiburg Single-cell Omics Platform. We thank D. Grün and Sagar for their help with the mCEL-Seq2 experiments. We thank D. Pfeifer, the Department of Internal Medicine I, the Medical Center-University of Freiburg and the Laboratory for Molecular Diagnostics and Genome Analysis for their crucial support. We thank the Lighthouse Core Facility for their support with cell sorting, FACS and confocal microscopy. The Lighthouse Core Facility is funded in part by the Medical Faculty, University of Freiburg (project nos. 2021/A2-Fol; 2021/B3-Fol) and the German Research Foundation (DFG; project no. 450392965). Human embryonic and fetal material was provided by the Joint MRC/Wellcome Trust (grant no. MR/006237/1) HDBR (www.hdbr.org). Human tissues were obtained from the NIH Neurobiobank at Mount Sinai. R.S. is supported by the IMMediate Advanced Clinician Scientist-Program, Department of Medicine II, Medical Center, University of Freiburg and Faculty of Medicine, University of Freiburg, funded by the Bundesministerium für Bildung und Forschung (Federal Ministry of Education and Research), 01EO2103. Furthermore, R.S. is supported by the Else Kröner Fresenius Foundation Fritz Thyssen Foundation and the German Research Foundation (SFB-1479 project ID 441891347). P.S. is a member of the KFO5024–505539112 (A04) and the research training group GRK2162 (270949263/GRK2162) funded by the DFG and is supported by the University Hospital Erlangen (IZKF project J94, IZKF clinician scientist programme). G.M. is a Marie Skłodowska-Curie fellow. M.P., J.P., C.B. and R.Z. are supported by the DFG (CRC/TRR167 project ID 259373024 ‘NeuroMac’). M.P. is further supported by the Sobek Foundation, the Ernst-Jung Foundation, the Klaus Faber Foundation, the Novo Nordisk Prize, the German Research Foundation (SFB 992 project ID 192904750, SFB 1160, SFB-1479 project ID 441891347, TRR 359 project ID 491676693 and Gottfried Wilhelm Leibniz Prize) and by the DFG under Germany’s Excellence Strategy (CIBSS EXC-2189 project ID 390939984). We acknowledge the assistance of the BCRT Flow Cytometry Laboratory (Charité, Universitätsmedizin Berlin). J.P. received additional funding from the DFG (PR 683/8-1 and SFB/TRR265 B04) and the UK DRI (Programme Award). R.Z. was supported by SFB-1479, project ID 441891347, CIBSS, EXC-2189, project ID 390939984 and ERC Advanced grant (101094168 AlloCure). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. We received no specific funding for this work. M.P. thanks Prof. Adriano Aguzzi, Zurich, an outstanding teacher and mentor, a pioneer in neuropathology and neurodegenerative research, a great supporter for many years and an eternal inspiration for fascinating experiments.

Author information

These authors contributed equally: Roman Sankowski, Patrick Süß.

Authors and Affiliations

Institute of Neuropathology, Faculty of Medicine, University of Freiburg, Freiburg, Germany
Roman Sankowski, Patrick Süß, Alexander Benkendorff, Chintan Chhatbar, Jonathan Cahueau, Gianni Monaco, Adrià Dalmau Gasull, Ashkan Khavaran & Marco Prinz
Department of Molecular Neurology, Friedrich Alexander University Erlangen-Nürnberg, University Hospital Erlangen, Erlangen, Germany
Patrick Süß
Neuropsychiatry Unit and Laboratory of Molecular Psychiatry, Charité, Universitätsmedizin Berlin and DZNE, Berlin, Germany
Chotima Böttcher, Camila Fernandez-Zapata & Josef Priller
Institute for Transfusion Medicine and Gene Therapy, Medical Center-University of Freiburg, Freiburg, Germany
Gianni Monaco
Department of Neurosurgery, Faculty of Medicine, University of Freiburg, Freiburg, Germany
Jürgen Grauvogel, Christian Scheiwe, Mukesch Johannes Shah, Dieter Henrik Heiland & Oliver Schnell
Department of Gynecology, Obstetrics, and Perinatology, Faculty of Medicine, University Hospital, Freiburg, Germany
Filiz Markfeld-Erol & Mirjam Kunze
Department of Internal Medicine I, Faculty of Medicine, Medical Center-University of Freiburg, Freiburg, Germany
Robert Zeiser
Signalling Research Centres BIOSS and CIBSS, University of Freiburg, Freiburg, Germany
Robert Zeiser & Marco Prinz
Department of Psychiatry and Psychotherapy, School of Medicine and Health, Technical University of Munich, Munich, Germany
Josef Priller
University of Edinburgh and UK DRI, Edinburgh, UK
Josef Priller

Authors

Roman Sankowski
View author publications
You can also search for this author in PubMed Google Scholar
Patrick Süß
View author publications
You can also search for this author in PubMed Google Scholar
Alexander Benkendorff
View author publications
You can also search for this author in PubMed Google Scholar
Chotima Böttcher
View author publications
You can also search for this author in PubMed Google Scholar
Camila Fernandez-Zapata
View author publications
You can also search for this author in PubMed Google Scholar
Chintan Chhatbar
View author publications
You can also search for this author in PubMed Google Scholar
Jonathan Cahueau
View author publications
You can also search for this author in PubMed Google Scholar
Gianni Monaco
View author publications
You can also search for this author in PubMed Google Scholar
Adrià Dalmau Gasull
View author publications
You can also search for this author in PubMed Google Scholar
Ashkan Khavaran
View author publications
You can also search for this author in PubMed Google Scholar
Jürgen Grauvogel
View author publications
You can also search for this author in PubMed Google Scholar
Christian Scheiwe
View author publications
You can also search for this author in PubMed Google Scholar
Mukesch Johannes Shah
View author publications
You can also search for this author in PubMed Google Scholar
Dieter Henrik Heiland
View author publications
You can also search for this author in PubMed Google Scholar
Oliver Schnell
View author publications
You can also search for this author in PubMed Google Scholar
Filiz Markfeld-Erol
View author publications
You can also search for this author in PubMed Google Scholar
Mirjam Kunze
View author publications
You can also search for this author in PubMed Google Scholar
Robert Zeiser
View author publications
You can also search for this author in PubMed Google Scholar
Josef Priller
View author publications
You can also search for this author in PubMed Google Scholar
Marco Prinz
View author publications
You can also search for this author in PubMed Google Scholar

Contributions

R.S., P.S., A.B., C.B., C.F.Z., G.M., J.C., C.C. and A.D.G. conducted experiments and analyzed the data. J.G., C.S., M.J.S., D.H.H., O.S., F.M.E., M.K. and R.Z. performed tissue acquisition, provided reagents and edited the paper. R.S., P.S., J.P. and M.P. supervised the project and wrote the paper.

Corresponding authors

Correspondence to Roman Sankowski or Marco Prinz.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Medicine thanks Jo van Ginderachter and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Anna Maria Ranzoni, in collaboration with the Nature Medicine team.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 FACS gating strategies.

a. Representative FACS plots of cell suspensions for 10x single-cell analysis (top figure panel). Cells were gated i. on singlets followed by ii. collection of DAPI- and CD45+ cells. The middle figure panel shows the gating strategy for single-nucleus RNA sequencing. The bottom figure panel shows the gating strategy for single-nucleus fixed RNA profiling. Cells were sorted into 1.5 ml tubes. b. For the enrichment of CD206+ cells, i. singlets were selected followed by ii. exclusion of DAPI + CD3 + CD19 + CD20+ cells. iii-iv-. CD45 + CD206+ cells were sorted into 384-well plates.

Extended Data Fig. 2 Overview of cell type-enriched gene expression signatures and cell cycle scores.

a. Cell type assignment of the dataset using the Azimuth human PBMC reference dataset. b. UMAP visualizations color coded for the expression of gene modules enriched in the indicated cell types. The color coding represents gene module expression scores for each cell using the Mann-Whitney U statistic. c. UMAP visualizations color coded for cell cycle scoring. The color coding represents gene module expression scores for each cell using the Mann-Whitney U statistic. The module score classification is based on published gene expression signatures²⁷.d. UMAP visualizations color coded for the expression of the published gene module²⁹. The color coding represents module enrichment scores for each cell using the Mann-Whitney U statistic.

Source data

Extended Data Fig. 3 Conserved and distinct signatures of tissue resident myeloid cells in CNS interfaces between mouse and man.

a. Heatmap visualization of latent factors determined by multiomic factor analysis (MOFA)2 analysis of macrophages across the indicated compartments. The color scale indicates the percentage of variance explained across the compartments. For enhanced readability, the respective values are indicated in each heat map tile. Selected top genes for each latent factor are presented to the right of the heat map. PC/PV: parenchyma/perivascular space, CP: choroid plexus, LM: leptomeninges, DM: dura mater. b. Dotplot showing gene ontology term enrichment analysis across the top 100 differentially enriched genes in the MOFA2 latent factors from panel a. The dot size indicates the gene ratio of genes differentially expressed in each cluster over the genes in the indicated gene ontology terms. The color coding of the dot indicates the Benjamini–Hochberg adjusted p value based on a one-sided Fisher’s exact test.c. Dot plot depicting the cross-species comparison of differentially expressed orthologue genes between CAMs and microglia in humans and mice. On the x-axis, adjusted average log2-fold changes from the comparison of murine MHC-II^low and MHC-II^high are signed positively if they are upregulated in the former and negatively if they are upregulated in the latter. The y-axis contains the adjusted average log2-fold changes of the human counterparts. The log2-fold changes were calculated using two-sided unpaired Wilcoxon Rank-Sum tests followed by Bonferroni correction for multiple testing. Orthologue genes in the top right and bottom left quadrants are differentially expressed in the same direction. Orthologue genes in the top left and bottom right quadrants are showing opposite directions of differential expression. Orthologue genes positioned along the axes are only upregulated in one species. The color coding indicates the directionality of the gene expression across species. Blue genes were upregulated in mice and humans; red genes were downregulated in mice and humans; green genes were upregulated in humans and downregulated in mice; violet genes were differentially regulated in humans, but not in mice; orange genes were differentially regulated in mice but not in humans.

Source data

Extended Data Fig. 4 Cell type gene module expression of mCEL-Seq2 data and validation with mass cytometry.

a. UMAP visualizations color coded for the expression of gene modules enriched in the indicated cell types. The color coding represents module enrichment scores for each cell using the Mann-Whitney U statistic. b. UMAP visualization color coded for the expression of published myeloid (left) and lymphoid cell homing gene modules²⁸. The color coding represents module enrichment scores for each cell using the Mann-Whitney U statistic. c. Spider visualization of 1,999 CD45+ cells analyzed using mass cytometry. The plots are color coded for K-nearest-neighbor density-based X-shift algorithm-based cluster assignment. The indicated cell type assignment is based on the expression of published cell type markers³¹. The cells labeled as ‘others’ did not show CD45 expression. Numbers indicate the different clusters. d. Spider presentation of the cells from panel f color coded for the compartment the cells were extracted from. The right bottom panel is showing a Marimekko chart of the distribution of compartments per cluster. CP: choroid plexus, LM: Leptomeninges, PV/PC: perivascular space and parenchyma. e. Spider visualization of the cells from panel f color coded for the expression of selected genes. The color scale represents Pearson’s residuals from a regularized negative-binomial regression. f. Single-cell heatmap of the protein expression in each cluster depicted in panel f. The color scale represents Pearson’s residuals from a regularized negative-binomial regression.

Source data

Extended Data Fig. 5 In-situ validation of immunohistochemistry markers identified in the scRNA-Seq data.

Filled arrowheads indicate double-positive cells, empty arrowheads indicate single-positive cells. Scale bars correspond to 100 µm. At least 3 images per patient were analyzed.

Extended Data Fig. 6 Single-nucleus and single-cell profiling of fetal and postnatal tissues.

a. UMAPs color coded for the expression of cell type-enriched gene modules. The color coding represents gene module scores using the Mann-Whitney U statistic. b. UMAPs color-coded for cell cycle scoring calculated using the Mann-Whitney U statistic. The underlying gene modules were previously described²⁷. c. Heatmap of the average expression of the 7 top markers in the indicated cell types. The color scale indicates the z-score. The dendrogram represents the hierarchical clustering based on Euclidean distances. d. UMAP color-coded for the expression of the published antigen-presentation and DAM modules^7,29. The color coding represents gene module scores using the Mann-Whitney U statistic. e. UMAP (top) and Marimekko chart (bottom) color-coded for the dataset of each cell. The Marimekko chart represents the contribution of each dataset to the respective cluster. Asterisks indicate statistical significance from one-sided hypergeometric tests with Benjamini–Hochberg adjustment for multiple testing. **p < 0.01; ***p < 0.001. Note that for enhanced readability, asterisk are only indicated up to C17. f. Dot-line plots showing the average expression of the macrophage (top), microglia (middle) and DAM modules per cell type across the developmental stages. To avoid pseudo-replication, gene expression was averaged for each cell type and patient. Each dot represents one donor. The lines indicate linear regression results with the confidence intervals displayed as shaded areas. Pearson correlation coefficient and p value are given at the top of each plot. g. UMAP of 4,332 FACS-sorted CD45+ cells from the PC/PV, LM and CP of a fetus from pcw 23 and adult controls. The color coding indicates Seurat clusters. The cell-type assignment is based on published datasets^2,38. h. UMAP (top) and Marimekko chart (bottom) color-coded for the anatomical compartment each cell was derived from. The Marimekko chart represents the contribution of each compartment to the respective cluster. The postnatal samples are from 5 individuals with 2 samples each from the PC/PV, CP and LM. i. Equivalent of figure panel h color-coded for developmental timepoints.

Source data

Extended Data Fig. 7 Profiling of engrafting myeloid cells.

a. UMAP visualization of 9,035 color-coded for Seurat v4 clustering results. Cell type assignment was conducted based on published gene expression signatures. Marimekko chart of the contribution of control and transplanted patient to each cluster. Asterisks indicate the results of statistical testing using one-sided hypergeometric tests. Adjustment for multiple testing was done using the Benjamini–Hochberg method. **p < 0.01; ***p < 0.001. b. Single-cell heatmap depicting the gene expression of the top 7 cluster marker genes per cluster of the cells shown in panel a. The genes are shown at the left-hand side of the heatmap. Color-coding of the cluster is consistent with panel a. The color scale represents Pearson’s residuals from a regularized negative-binomial regression. c. Single-cell heatmap depicting the gene expression of up to top 10 cluster marker genes per cluster of the myeloid cell cluster from Fig. 4 i. d. Comparative analysis of the percentage of Y+ positive cells among all identified Iba1 + , P2RY12 + , TMEM119+ and GLUT5+ cells in the cortex. Positivity for the pan-myeloid marker Iba1 was considered as the maximum achievable percentage of engrafting myeloid cells. The lines connect the dots between results from the same patients. The indicates p values were calculated using paired t-tests.

Source data

Extended Data Fig. 8 Validation of control vs GBM-associated immune cells assessed by single-cell RNA-Seq and mass cytometry.

a. UMAP color-coded for the expression of the indicated modules scores. Module scores are based on the Mann-Whitney U statistic. The cell cycle module genes are published²⁷. b. UMAP color-coded for DAM and myeloid cell homing module expression^28,29. Module scores are based on the Mann-Whitney U statistic. c. Marimekko chart showing the integration results of the present dataset with published data⁴³. Cluster assignments of the present data were transferred on the Antunes et al dataset. Asterisks indicate statistical significance from one-sided hypergeometric tests with Benjamini–Hochberg adjustment for multiple testing. *p < 0.05; **p < 0.01; ***p < 0.001. d. Single-cell heatmap showing the expression of the top 20 cluster markers with selected genes shown on the left. Cluster colors are consistent with panel b. The color scale represents Pearson’s residuals from a regularized negative-binomial regression. e. Spatial plot of hypoxia (red) and adjacent necrosis (orange) analyzed with Nanostring CosMX. The white arrow indicates a spatial trajectory shown in panel f. The shown data is representative of 8 analyzed fields from 2 glioblastoma samples. f. Spatial trajectory heatmap showing the smoothed gene expression along the trajectory from figure panel e with representative genes on the right. g. Spider visualization of mass cytometry data from 2,439 CD45⁺ glioblastoma-derived cells color-coded for clusters. Cell type assignment is based on published cell type marker expression. CD45^- cells are labeled as ‘others’. h. Spider presentation (top) and Marimekko chart (bottom) showing the distribution of diagnoses per cluster. i. Protein expression single-cell heatmap of the cells in panel e. The color scale represents Pearson’s residuals from a regularized negative-binomial regression. j. Dot-line plot depicting the expression of selected proteins between tumor-associated and control CAMs in C7 from panel e (top) and microglia from C1 (bottom). The indicated p values were calculated using unpaired two-sided t-tests with Benjamini–Hochberg multiple testing adjustment. Each symbol represents one cell. Medians are indicated.

Source data

Extended Data Fig. 9 Graphical abstract of the present study.

The identified markers were chosen based on the combined findings of the modalities.

Supplementary information

Reporting Summary

Supplementary Tables 1–21

Supplementary Table 1. Cell metadata of the control CD45⁺ dataset. Supplementary Table 2. Results of the one-sided hypergeometric testing of the cluster-wise compartment contributions. Supplementary Table 3. Top 20 cluster marker genes of the control CD45⁺ dataset. Significance was calculated using two-sided unpaired Wilcoxon rank-sum tests followed by Bonferroni correction for multiple testing. Supplementary Table 4. Top 100 genes per latent factor identified by MOFA2 analysis. Supplementary Table 5. Genes contained in the species comparison between human and mouse cells in Extended Data Fig. 3c. Statistical testing was performed by two-sided hypergeometric tests followed by adjustment for multiple testing using the Benjamini–Hochberg method. Supplementary Table 6. Results of the one-sided hypergeometric testing of the cluster-wise cell type contributions of cells from the CD45⁺CD206⁺Lin⁻ gate of CNS border regions analyzed using mCEL-Seq2. Statistical testing was performed by two-sided hypergeometric tests followed by adjustment for multiple testing using the Benjamini–Hochberg method. Supplementary Table 7. Cluster marker genes of cells from the CD45⁺CD206⁺Lin⁻ gate of CNS border regions analyzed using mCEL-Seq2. Significance was calculated using two-sided unpaired Wilcoxon rank-sum tests followed by Bonferroni correction for multiple testing. Supplementary Table 8. Cluster marker profiles for CITE-seq data of control cells from Fig. 2e. Significance was calculated using two-sided unpaired Wilcoxon rank-sum tests followed by Bonferroni correction for multiple testing. Supplementary Table 9. Mass cytometry antibody panel. Supplementary Table 10. ISS gene panel. Supplementary Table 11. Nanostring CosMx gene panel. Supplementary Table 12. Top 20 cluster marker genes of the fetal single-nucleus sequencing dataset. Significance was calculated using two-sided unpaired Wilcoxon rank-sum tests followed by Bonferroni correction for multiple testing. Supplementary Table 13. Up to top 100 marker genes for pre and postnatal CAMs, microglia and Kolmer cells. Significance was calculated using two-sided unpaired Wilcoxon rank-sum tests followed by Bonferroni correction for multiple testing. Supplementary Table 14. Genes visualized in the pseudotime analysis heat map in Fig. 6c. Supplementary Table 15. Up to top 10 cluster marker genes of myeloid cells analyzed with single-nucleus RNA profiling in FFPE samples. Significance was calculated using two-sided unpaired Wilcoxon rank-sum tests followed by Bonferroni correction for multiple testing. Supplementary Table 16. Results of the one-sided hypergeometric testing of the cluster-wise contributions of cells from control and tumor tissues. Supplementary Table 17. Top 20 cluster marker genes of the tumor dataset. Statistical testing was performed by two-sided hypergeometric tests followed by adjustment for multiple testing using the Benjamini–Hochberg method. Supplementary Table 18. Results of the one-sided hypergeometric testing of the cluster-wise contributions of cells from LM and PC/PV. Supplementary Table 19. Top 100 genes per latent factor identified by MOFA2 analysis in Fig. 6e. Supplementary Table 20. Characteristics of the patient samples analyzed in the present study. Supplementary Table 21. Marker genes for cell types and other gene expression modules.