Single-cell atlas of the human brain vasculature across development, adulthood and disease

1The brain vasculature is important for both the proper functioning of the normal brain as well as for a variety of vascular-dependent CNS pathologies such as brain tumours, brain vascular malformations, stroke and neurodegenerative diseases–. A better understanding of the underlying cellular and molecular mechanisms and architecture of the vasculature during brain development, in the healthy adult brain, as well as in vascular-dependent brain diseases, has broad implications for both the biological understanding as well as the therapeutic targeting of the pathological brain vasculature–. Vascular growth and network formation, involving endothelial cells (ECs) and other cells of the neurovascular unit (NVU), are highly dynamic during brain development, almost quiescent in the healthy adult brain and reactivated in a variety of angiogenesis-dependent brain pathologies, including brain tumours and brain vascular malformations,,–, thereby activating ECs and perivascular cells (PVCs) of the NVU and other tissue-derived cells (hereafter collectively referred to as PVCs). However, it is unclear which molecular signalling cascades are reactivated and how they regulate brain tumour and brain vascular malformation vascularization and growth.

2The CNS vasculature has unique features such as the blood–brain barrier (BBB) and the NVU–. During development, various CNS-specific and general signalling pathways drive CNS angiogenesis,,,–. The brain vasculature also displays an arteriovenous (AV) endothelial hierarchy similar to peripheral vascular beds–. Developmentally regulated signalling axes in ECs are thought to contribute to the establishment of CNS-specific properties as well as AV specification of the endothelium in the healthy adult brain and to their alteration in disease,. Over the past years, single-cell transcriptome atlases of human peripheral organs–, the human brain vasculature–, and the mouse brain and peripheral vasculature, were established. Nevertheless, a landscape of the human brain vasculature at the single-cell level across fetal development, adulthood and various vascular-dependent diseases with a focus on the brain vascular endothelium is lacking. Here we created a comprehensive molecular atlas of the human brain vasculature using single-cell RNA sequencing (scRNA-seq) analysis in the developing, adult/control and diseased human brain (Fig. 1a, Extended Data Fig. 1 and Supplementary Methods). We identified extensive heterogeneity among ECs as well as common hallmarks across a spectrum of multiple brain pathologies, including commonly regulated angiogenic signalling pathways that significantly overlap with the fetal signalling axes, altered AV specification and CNS specificity, upregulation of MHC class II signalling and strong EC–EC/EC–PVC communication networks.

3We constructed a human brain vasculature single-cell atlas (sc-atlas) using samples from fetal as well as adult control (undiseased atlas) and diseased brains, including adult brain vascular malformations and brain tumours (diseased atlas). We acquired freshly isolated cells (both fluorescence-activated cell sorting (FACS)-sorted ECs and unsorted ECs and PVCs; Supplementary Tables 1–4) from 8 individual fetuses– and from 61 adult brain samples (from 61 individual patients), covering adult temporal lobe (TL) controls and adult vascular-dependent pathologies, including brain vascular malformations, namely, brain AV malformation (AVM),, and brain tumours, notably, lower-grade glioma (LGG),, glioblastoma (GBM),,, lung cancer brain metastasis (MET), and meningioma (MEN) (Fig. 1a, Extended Data Fig. 1a,b and Supplementary Tables 1–4). Brain tissue samples were dissociated into single-cell suspensions, which were either FACS-sorted for ECs (CD31+CD45−) or processed as unsorted samples to examine all cells of the NVU (Fig. 1a). Single-cell transcriptomes were collected using the 10x Genomics Chromium system and analysed. CD31+CD45− ECs showed consistent expression of classical endothelial markers, such as CD31, VWF and CLDN5, while not expressing PVC markers (Supplementary Fig. 1a–o), thereby confirming the purity of EC isolations. In summary, 606,380 single cells, including 304,016 sorted ECs and 302,364 unsorted ECs and PVCs, passed the quality-control criteria (Fig. 1a and Supplementary Tables 3–5). The number of sorted ECs analysed here substantially exceeds the number of ECs analysed using scRNA-seq, or single-nucleus RNA-seq, in previous studies, and we directly compared single-cell transcriptomics of sorted ECs from the vasculature of the fetal and adult brain and of various brain pathologies. Notably, we report higher numbers of sorted ECs and of unsorted ECs and PVCs in the different brain entities compared with previous studies–,, enabling us to assess EC heterogeneity across development, adulthood and disease at a high resolution.

4To address the role of the endothelium within the brain NVU across different entities, fetal, adult/control and pathological unsorted EC and PVC transcriptomes from 31 patients were analysed (Fig. 1a and Supplementary Fig. 2a–f). We identified 18 major brain cell types, including all known vascular, perivascular and other tissue-derived cell types in the human brain. The detected cell type distributions within the NVU differed between the fetal, adult control and pathological brain samples (Supplementary Figs. 2e–g and 3a–n and Supplementary Tables 10–16). Key signatures and differentially expressed genes (DEGs) were validated using bulk RNA-seq, RNAscope, spatial transcriptomics, immunofluorescence (IF) and imaging mass cytometry (IMC) (Fig. 1a).

5We next compared ECs in the sorted samples across entities and found that ECs from different entities exhibited prominent transcriptomic heterogeneity (Extended Data Fig. 1c,f) as well as distinct gene expression signatures (Fig. 1b–f, Extended Data Fig. 1d,e,g,h and Supplementary Fig. 5). We defined major EC signatures, including a human fetal CNS (and peripheral) signature characterizing CNS and periphery-specific markers of the fetal vasculature (Extended Data Fig. 1g and Supplementary Table 6), a human fetal/developmental CNS/brain signature revealing properties of the developing and mature human brain vasculature, and a pathological signature of the diseased brain vasculature including a brain vascular malformation and a brain tumour signature (Fig. 1d–f, Supplementary Fig. 5a–c and Supplementary Table 25). The fetal and pathological brain EC signatures revealed differential expression of the well-known angiogenic markers PLVAP and ESM1,,–, which we confirmed in the fetal and diseased brain entities using IF analysis,,– (Fig. 1g–n and Extended Data Fig. 2).

6Although all of the entities revealed distinct EC markers, some EC markers were conserved across two or more entities (such as ADIRF, EGR1, PLPP1 and ANGPT2) (Fig. 1b, Extended Data Fig. 1e,h and Supplementary Tables 8 and 9).

7We further assessed the differences in ECs across developmental stages and in pathological conditions (Supplementary Tables 7 and 25). DEGs between the fetal and adult/control stage and between the adult/control and pathological brain showed developmental and pathology-specific gene and pathway enrichments (Supplementary Fig. 5d–f), providing insights into functional specialization of the human brain vasculature across development, homeostasis and disease (Supplementary Fig. 5). Using various approaches, including statistical regression, we found no evidence that age and sex– are confounders of our findings (Supplementary Tables 13 and 14). The top differentially regulated pathways in both fetal versus adult/control as well as in pathological versus adult/control brain EC signatures belonged to five main groups, including development and neurovascular link, cell–cell/extracellular-matrix-related processes, immune-related processes, angiogenesis and metabolism (Supplementary Fig. 5d–f). Notably, of the 1,409 differentially regulated pathways in pathology versus adult/control brain ECs, more than half (997) also showed differential regulation in fetal versus adult/control brain ECs (Fig. 1o and Supplementary Fig. 5f), highlighting the importance of developmental pathways in vascular-dependent brain pathologies. Bulk RNA-seq analysis confirmed the scRNA-seq findings, including the dysregulated pathways across pathologies (Supplementary Fig. 6 and Supplementary Table 26). Together, these data indicate that signalling axes driving vascular growth during fetal brain development are silenced in the adult control brain and reactivated in the vasculature of brain tumours and brain vascular malformations and that common dysregulated pathways are observed in the pathological brain vascular endothelium across diseases.

8To further address EC heterogeneity across different brain entities at the single-cell level, we pooled, integrated and batch-corrected (using RPCA)–, clustered and visualized all fetal (21,512), adult/control (76,125) and pathological (145,884) sorted brain EC transcriptomes from 43 patients (Fig. 2a and Supplementary Figs. 7a–e and 9). Brain vascular ECs are organized along the human brain AV axis, referred to as AV zonation,–. Endothelial clusters were biologically annotated using DEGs across entities, and we identified 44 EC subclusters (Fig. 2e, Supplementary Fig. 7 and Supplementary Table 18) that were arranged according to AV zonation, which we grouped into 14 major EC clusters for further downstream analysis (Fig. 2a and Supplementary Fig. 7a,h).

9We characterized AV zonation markers, with arterial (subclustered into large artery, artery, arteriole) and venous (subclustered into large vein, vein, venule) clusters located at the opposite ends of the uniform manifold approximation and projection (UMAP), separated by major capillary clusters (subdivided into capillary and angiogenic capillary) in the fetal, adult and pathological brains (Fig. 2a,e, Supplementary Figs. 7 and 10 and Supplementary Tables 18 and 19), providing unprecedented transcriptional resolution by AV zonation–.

10While we confirmed differential expression of known marker genes of AV specification,, we also identified AV-zonation markers that have not to our knowledge been identified previously in the human brain: LTBP4 (large arteries); ADAMTS1 (arteries); VSIR, AIF1L, CD320 and others (arterioles); SLC38A5, BSG, SLC16A1 and SLCO1A2 (capillaries); JAM2, PRCP, PRSS23 and RAMP3 (venules); PTGDS, POSTN and DNASE1L3 (veins); CCL2 (large veins); and PLVAP, ESM1 and CA2 (angiogenic capillaries) (Fig. 2e, Supplementary Fig. 7i and Supplementary Tables 18 and 19). PLVAP and ESM1 were among the top markers of the angiogenic capillary cluster and we confirmed PLVAP and ESM1 expression in diseased brain entities and in the fetal brain, indicating its role in developmental and pathological vascular growth,. Indeed, PLVAP and ESM1 exhibited RNA and protein expression in human fetal brain and human brain vascular malformation/tumour ECs on the basis of RNAscope and IF analysis (Extended Data Fig. 2).

11We also assigned EC clusters outside AV zonation, notably (proliferating) stem-to-endothelial cell transdifferentiating (stem-to-EC) clusters and (proliferating) endothelial-to-mesenchymal transition (EndoMT) clusters (Fig. 2e and Supplementary Fig. 7a,i), for which we identified specific molecular markers. We found proliferating ECs (such as TOP2A and MKI67) in the fetal (3.54%), adult (0.4%) and pathological brains (1.37%) (Fig. 2e, Supplementary Fig. 7i and Supplementary Table 19). We identified EndoMT clusters expressing both mesenchymal (such as APOE, ACTA2 and TAGLN) and endothelial markers (Fig. 2e, Supplementary Fig. 7i and Extended Data Fig. 2qi–bii,gii–jii′). Notably, we observed two subsets of EndoMT ECs (proliferating EndoMT and EndoMT) that were increased in certain pathologies (MEN > LGG > GBM). Proliferating EndoMT ECs expressed both EndoMT and proliferation markers (for example, ACTA2 and MKI67) (Fig. 2e, Supplementary Fig. 7i and Supplementary Tables 8 and 9). Using RNAscope, IF and IMC, we found ACTA2+CD31+CLDN5+ co-expressing but PDGFRβ− (suggesting no pericyte identity) ECs across pathologies (Extended Data Fig. 2qi–bii,gii–jii′,kii–qii5), indicating the presence of EndoMT ECs in the diseased vasculature. In GBMs and METs, we observed stem-to-EC clusters that expressed classical EC markers (such as CD31, CLDN5, CDH5 and VWF) to a lower level compared with other EC clusters, as well as some markers of (tumour) stem cells (Fig. 2e, Extended Data Fig. 3 and Supplementary Fig. 7i), suggesting ECs undergoing stem-to-EC transdifferentiation. In GBMs, we identified a stem-to-EC cluster expressing the GBM stemness markers SOX2, PTPRZ1, POUR3F2 and OLIG1,, and EC markers, (Supplementary Fig. 7i and Extended Data Fig. 3a–i,ai–ni′). In METs, we noted a previously undescribed stem-to-EC population that co-expressed EC markers and stem cell markers of lung cancers (for example, SOX2, EPCAM, CD44 and SFTPB) (Extended Data Fig. 3n–v,gi–zi′ and Supplementary Fig. 7i). In GBMs and METs, we identified groups of stem-to-ECs that co-expressed stemness (for example, SOX2, PTPRZ1, EPCAM1 and SFTPB) and proliferation markers (for example, MKI67, BEX1, HMGB2 and UBE2C) (Fig. 2e and Extended Data Fig. 3). To validate the stem-to-EC clusters in GBM and MET, we used double immunostaining for EC and stemness markers. In GBM, we found SOX2+CD31+ and PTPRZ1+CD31+ co-expressing ECs, whereas, in MET, we observed EPCAM+CD31+ and SFTBP+CD31+ co-expressing ECs (Extended Data Fig. 3). The confirmation of tumour stemness marker enrichment in a subset of tumour ECs suggests the presence of stem-to-ECs in GBM and MET vasculature.

12We next addressed the distributions of EC clusters between the fetal, adult/control and pathological brains. Capillaries accounted for around 60.5% of ECs, arterial ECs accounted for 18.2% and venous ECs accounted for 16.2%, in agreement with previous studies,. We further uncovered previously unrecognized EC heterogeneity across a wide range of human brain tissues (Fig. 2b–d, Supplementary Fig. 10 and Supplementary Tables 10–16). Angiogenic capillary proportions were significantly higher in the fetal brain and in brain tumours (GBM > MET > MEN > LGG), illustrating their angiogenic capacity,,,, whereas brain vascular malformations (AVM) revealed significantly elevated proportions of venous clusters, indicating their venous character, (Fig. 2c,d, Supplementary Fig. 10 and Supplementary Table 12). We next evaluated whether AV-zonation markers were conserved across species–,, (Supplementary Fig. 4). Although the overall structure of AV zonation was conserved between human and mouse, the number of conserved AV-zonation genes in the different AV compartments was low. Accordingly, we found the highest proportion of human-specific AV-zonation markers in small > large-calibre and venous > arterial vessel ECs (Fig. 2f, Supplementary Fig. 4zxxix,zxxx and Supplementary Table 17), and we validated these human-specific markers referring to the Human Protein Atlas (HPA),– (Supplementary Fig. 8).

13In the brain NVU, mapping of our dataset to the freshly isolated mouse dataset revealed high transcriptomic similarity between species for ECs and PVCs (Supplementary Fig. 4zxvii–xvix). We further observed that neurons and astrocytes showed the greatest transcriptional divergence (Fig. 2f and Supplementary Table 17), while ECs and oligodendrocytes displayed the highest percentage of human-specific markers, in agreement with previous studies–. These species-specific differences along AV-zonation suggest fundamental disparities in brain vascular properties, indicating the necessity to directly study sorted/enriched ECs and PVCs of the human brain vasculature at the single-cell level.

14As EC clusters reside in close proximity to each other along the AV tree, we next inferred cell–cell communication pathways,. Differential analysis revealed increased cellular cross-talk among EC clusters in pathological and fetal ECs, highlighting a key role for angiogenic capillaries. Angiogenic capillaries displayed upregulation of several signalling pathways, including the five above-mentioned groups in both the diseased and fetal (versus adult/control) brain (Fig. 2g–k and Supplementary Fig. 11), highlighting this cluster as a major signalling mediator within brain EC networks.

15Failure of proper AV specification in brain vascular malformations and formation of tortuous arteries and veins in brain tumours has been reported,, but AV specification in brain pathologies and fetal (brain) development remains poorly understood. We ordered ECs along a one-dimensional transcriptional gradient using Monocle and TSCAN to examine the AV axis in the different entities. Whereas arterial and venous markers peaked at opposite ends, capillary markers showed peaks in the mid-section throughout all entities (Fig. 3b,f,j, Extended Data Fig. 4b,f,j,n,r,v and Supplementary Fig. 13III), indicating that in silico pseudospace and pseudotime recapitulate in vivo anatomical topography of EC clusters in the human brain vasculature,. We observed AV zonation throughout the fetal, control and pathological brains, but observed a partial alteration of EC ordering along the AV axis in disease (Fig. 3b,c,f,g,j,k and Extended Data Fig. 4).

16We defined an AV signature comprising genes revealing significant expression gradients of ECs along the AV axis (Fig. 3d,h,l and Extended Data Fig. 4d,h,l,p,t,x). The seamless zonation continuum was recapitulated in all entities but again showed alteration in pathologies. While AV markers revealed a clear distinction between AV compartments in the fetal and adult/control brain, showing specific markers of large arteries (for example, VEGFC, FBLN5), arterioles (such as LGALS3, AIF1L), capillaries (for example, SLC35A5, MFSD2A), angiogenic capillaries (such as ESM1, ANGPT2), venules (for example, JAM2 and PRCP) and large veins (such as SELE and SELP), some zonation markers showed a less-specific presence in pathologies (Figs. 2 and 3d,h,l and Extended Data Fig. 4d,h,l,p,t,x).

17Whereas almost all fetal and pathological ECs were quite similar to temporal-lobe EC clusters, small-calibre vessel ECs were more different compared with their healthy temporal lobe counterparts (Extended Data Fig. 4zv–zviii), probably pertaining to a higher vulnerability of small-calibre vessel ECs to alterations in the local tissue microenvironment.

18To further address lineage relationships in AV specification, we referred to RNA velocity, and diffusion map, revealing vectors from multiple EC clusters towards the angiogenic capillary cluster in angiogenic entities (tumours > fetal brain > vascular malformations), whereas, in vascular malformations, we observed vectors from various EC clusters towards venous clusters (Fig. 3a,e,i, Extended Data Fig. 5 and Supplementary Fig. 13). These results indicate that RNA velocity can suggest timeline relationships among human brain vascular ECs,.

19We next addressed whether EC markers of AV clusters were conserved between vascular beds or expressed in a more tissue-specific manner. While we identified multiple conserved markers for large arteries and large veins, capillaries were more tissue/entity specific, indicating a more pronounced transcriptional heterogeneity of the capillary bed across the different brain tissues. Accordingly, capillaries showed more tissue-specific markers than large-calibre vessels (Fig. 3m–p), indicating a higher susceptibility of capillary ECs to the local tissue microenvironment.

20We next examined CNS-specific properties distinguishing brain ECs from ECs outside the CNS,. Bulk RNA-seq analysis in mice revealed a BBB-enriched transcriptome, but how the human brain EC CNS properties differ at the single-cell level and whether they are heterogeneous across developmental stages and in disease remains largely unclear. The development of the human fetal BBB (occurring between gestational week 8 and 18),, is controversial,. We therefore studied human fetal BBB development at a high resolution. Referring to a human/mouse BBB signature (Supplementary Table 20 and Extended Data Fig. 6a–j), we observed an increased BBB signature expression with increased gestational age (Extended Data Fig. 6b–j), in agreement with a previous study. Along the AV compartments, the BBB signature revealed a higher expression in small- versus large-calibre vessels across developmental stages (Extended Data Fig. 6d), in agreement with previous findings and probably pertaining to the susceptibility of capillaries to the local microenvironment.

21Next, to address molecular differences of CNS and peripheral ECs at the single-cell level, we defined a human adult and fetal CNS signature (Fig. 4a–f, Supplementary Fig. 14l–q and Supplementary Table 20). These include known BBB (MFSD2A and CLDN5) and capillary markers (CA4, and SPOCK2), as well as novel genes enriched in the CNS vasculature such as SPOCK3, BSG and CD320 (Supplementary Fig. 14b,e and Supplementary Table 20). The adult and fetal CNS signatures showed elevated expression with increasing gestational age and in small- versus large-calibre vessels across developmental stages, reminiscent of the BBB signature expression pattern (Fig. 4g and Extended Data Fig. 6d).

22CNS properties were observed in the fetal and adult brains, whereas alteration of the CNS signature and concomitant acquisition of the peripheral signature was seen in pathologies (Fig. 4a–f and Supplementary Figs. 14 and 15). Comparing the CNS signature between pathological and adult/control brain ECs, we found downregulation of SLC2A1, which is dysregulated in neurodegenerative conditions; the lipid transporter MFSD2A, which is expressed in brain ECs and restricts caveolae-mediated transcytosis at the BBB,,; and the BBB marker CLDN5 (Fig. 4d–f and Supplementary Table 20), therefore suggesting BBB alteration in pathologies. The CNS signature was highest in the temporal lobe, followed by intra-axial primary brain tumours and fetal brain (LGG > fetal brain > GBM), brain vascular malformations, intra-axial secondary brain tumour MET and extra-axial brain tumour MEN, whereas the peripheral signature followed an inversed pattern (Fig. 4e,f and Supplementary Fig. 14p,q).

23We next addressed CNS-specific properties along the AV axis. In the fetal and adult brain, the CNS signature was mainly expressed by small-calibre vessels, while the peripheral signature was predominantly present in large arteries and large veins. We observed a similar pattern in pathological brains with, however, a notable decrease in cells expressing the CNS signature (most pronounced for angiogenic capillaries > capillaries), paralleled by an increase in cells expressing the peripheral signature predominantly for angiogenic capillaries and large-calibre vessels (Fig. 4g and Supplementary Fig. 14h).

24We next examined how the CNS, BBB and peripheral signatures changed along the AV axis in each pathological entity. The CNS and BBB signatures were downregulated in every pathology with a similar pattern to that described above and reaching the highest baseline values of CNS specificity at the capillary and arteriole levels, with the capillaries being the cluster mostly affected by pathologies (Fig. 4g and Supplementary Fig. 14h), probably pertaining to the influence of the local microenvironment for small-calibre vessels. The peripheral signature was upregulated in disease, peaking for AVM > MEN > MET > GBM and lower expression for LGG, predominantly affecting large-calibre vessels and angiogenic capillaries (Fig. 4h and Supplementary Figs. 14 and 15). These data indicate that CNS ECs acquire CNS-specific properties during fetal-to-adult transition and take on a peripheral EC signature in disease conditions.

25The CNS and BBB signatures are tightly linked to a functional BBB in vivo and BBB dysfunction affects the CNS properties of ECs. We therefore investigated the human and mouse BBB dysfunction modules, with the latter being upregulated in CNS ECs after various disease triggers in the mouse brain, shifting CNS ECs into peripheral EC-like states. We found that the human and mouse BBB dysfunction modules were upregulated in human brain tumours and brain vascular malformations as well as in the fetal brain (Supplementary Fig. 16a–n), probably due to pathways related to BBB dysfunction. Both the human and mouse BBB dysfunction modules were highest in AVM > GBM > MET > MEN (Supplementary Fig. 16a–n). The BBB dysfunction modules expression along the AV axis revealed enrichment in large-calibre vessels and angiogenic capillaries, mimicking the peripheral signature expression, again indicating that pathological CNS ECs take on a peripheral endothelial gene expression (Supplementary Fig. 16h–n). Comparison to BBB dysfunction modules in human Alzheimer’s disease, Huntington’s disease and AVMs revealed some overlap with the human and mouse BBB dysfunction modules (Supplementary Fig. 16o–s and Supplementary Table 21), indicating common and distinct features among brain diseases.

26We confirmed decreased expression of the CNS-signature genes SPOCK3, BSG, CD320, PPP1R14A and SLC38A5 in all brain tumours and brain vascular malformations (Fig. 4h–x and Extended Data Figs. 6k–t′4, 9 and 10) using IF and IMC, thereby highlighting the alteration of CNS properties in the diseased human cerebrovasculature.

27We identified EC populations expressing the MHC class II genes CD74, HLA-DRB5, HLA-DMA, HLA-DPA1 and HLA-DRA in pathological CNS tissues. This antigen-presenting signature, indicating possible immune functions of human brain ECs, prompted us to investigate the heterogeneity of MHC class II transcripts between tissues at the single-cell level.

28scRNA-seq identified endothelial MHC class II expression in peripheral human and mouse tissues,, but assessment of MHC class II expression in human brain vascular beds at the single-cell level is lacking. To assess MHC class II gene expression across brain development and disease, we defined a human MHC class II signature including MHC class II receptors (Fig. 5d and Supplementary Table 22). The MHC class II signature was upregulated in pathologies, and low in the fetal brain (Fig. 5a–f and Extended Data Fig. 9), in agreement with a previous study. We found that the MHC class II signature was highest in AVM > MEN > MET, followed by LGG > GBM, the temporal lobe and the fetal brain, grossly following the peripheral signature expression gradient (Fig. 5f). We examined MHC class II signature expression patterns according to AV zonation. Whereas, in the fetal brain, only very few ECs (large arteries and arterioles) expressed the MHC class II signature, mainly large-calibre vessel (large arteries and large veins) ECs expressed a signature of genes involved in MHC class II-mediated antigen presentation in the adult brain (Fig. 5a–c,g).

29The MHC class II signature was upregulated in all pathologies according to the pattern described above (Fig. 5a–c,g–z). We observed a partial overlap of the MHC class II and peripheral signatures and of the BBB dysfunction module with a common predominance in large-calibre vessels, but a more widespread/stronger expression of the peripheral signature and BBB dysfunction module in angiogenic capillaries consistent with previous findings. These data suggest that pathological CNS ECs upregulate MHC class II receptors in brain tumours and brain vascular malformations.

30We confirmed enrichment of MHC class II genes including CD74 and others in the pathological human cerebrovasculature using IF, IMC and RNAscope (Fig. 5h–w and Extended Data Figs 10, 11, 12r and 13).

31Spatial transcriptomics confirmed spatial co-localization of MHC class II ligands and receptors on ECs in the temporal lobe and in GBM (Supplementary Fig. 18). MHC class II signalling seems to be mediated mainly by APP, COPA and MIF ligands and the CD74 receptor in AV clusters (Supplementary Fig. 17), and APP–CD74, COPA–CD74 and MIF–CD74 have been described as ligand–receptor pairs,.

32Single-cell transcriptomics of unsorted ECs and PVCs offers the opportunity to address cellular cross-talk and the role of ECs within the NVU. To address cell–cell interactions across entities, we constructed ligand–receptor interaction maps,. In the majority of entities, ECs were at the centre of the network displaying numerous interactions with other ECs and PVCs (Extended Data Fig. 12a–i and Supplementary Fig. 19), indicating a crucial role of ECs in NVU function and EC–PVC cross-talk. In fetal and adult/control brains, ECs showed most interactions with fibroblasts, pericytes and astrocytes (Extended Data Fig. 12a–f). In brain pathologies, ECs displayed increased interaction numbers as well as increased interactions with immune cells (Extended Data Fig. 12g–i and Supplementary Figs. 19 and 20). Intercellular signalling pathways were substantially increased in fetal and pathological ECs (and PVCs) (Extended Data Fig. 12j,k and Supplementary Fig. 20a–g). Cell–cell communication analysis predicted upregulation of signalling pathways in the developing versus control brain as well as diseased versus control brain NVUs, including similar pathways as observed among EC clusters in EC–EC networks (Fig. 2i–k, Extended Data Fig. 12l–n and Supplementary Fig. 20k–n). Intercellular cross-talk analysis predicted a key role for the ECs within the fetal, adult/control and diseased brain NVU signalling networks (Extended Data Fig. 12l–n and Supplementary Fig. 20e–g).

33During fetal brain development and in brain pathologies, we observed upregulation of ligands and receptors on ECs and PVCs as well as of the corresponding pathways, which partially overlapped with the ones in EC–EC cross-talk (Extended Data Fig. 12l–n), suggesting that these ligand–receptor interactions contribute to brain EC–PVC signalling.

34On the basis of our observation of MHC class II signalling in EC–EC communication, we next addressed MHC class II signalling in EC–PVC intercellular cross-talk, which predicted elevated MHC class II signalling (predominantly in microglia and macrophages, ECs and tumour cells/oligodendrocytes) in brain pathologies (Extended Data Fig. 12o,p,q and Supplementary Fig. 20h–j). Spatial transcriptomics confirmed spatial co-localization of MHC class II ligands and receptors on ECs and PVCs (Supplementary Fig. 18) in the temporal lobe and in GBM, while IMC illustrated physical proximity between MHC class II-expressing ECs and microglia/macrophages across all entities (Extended Data Figs. 12r and 13). Notably, the APP–CD74, COPA–CD74 and MIF–CD74 ligand–receptor pairs that were predicted to mediate MHC class II signalling in EC–EC interactions were also predicted ligand–receptor pairs in the developing, adult and diseased NVU (Supplementary Fig. 20h–j), with ECs notably strongly expressing CD74 (Supplementary Fig. 19e,j,o,t,y). These data suggest involvement of MHC class II in EC–immune cell interactions and indicate that the APP–CD74, COPA–CD74 and MIF–CD74 ligand–receptor pairs contribute to NVU signalling.

35Here we generated a large-scale single-cell molecular atlas of the developing fetal, adult/control and diseased human brain vasculature at a very high resolution, using scRNA-seq, composed of 606,308 freshly isolated endothelial, perivascular and other tissue-derived cells covering a substantial diversity of human brain tissue.

36We have provided molecular definitions of human brain cell types and their differences by brain developmental stage and pathology, thereby unravelling organizational principles of ECs and PVCs composing the human brain vasculature. Our experimental methodology relies on transcriptional profiles of human cerebrovascular cells generated from fresh human neurosurgical resections and fresh fetal abortions, reducing the likelihood of transcriptional alterations associated with post-mortem tissue acquisition (Supplementary Discussion).

37Our human vascular brain atlas provides a basis for understanding the organizing principles and single-cell heterogeneity of universal, specialized and activated endothelial and PVCs with broad implications for physiology and medicine, and serves as a powerful publicly available reference for the field.

38Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

39Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-024-07493-y.