Single-Cell Analysis of Blood-Brain Barrier Response to Pericyte Loss

Abstract

Rationale:

Pericytes are capillary mural cells playing a role in stabilizing newly formed blood vessels during development and tissue repair. Loss of pericytes has been described in several brain disorders, and genetically induced pericyte deficiency in the brain leads to increased macromolecular leakage across the blood-brain barrier (BBB). However, the molecular details of the endothelial response to pericyte deficiency remain elusive.

Objective:

To map the transcriptional changes in brain endothelial cells resulting from lack of pericyte contact at single-cell level, and to correlate them with regional heterogeneities in BBB function and vascular phenotype.

Methods and Results:

We reveal transcriptional, morphological and functional consequences of pericyte absence for brain endothelial cells using a combination of methodologies, including single-cell RNA sequencing, tracer analyses and immunofluorescent detection of protein expression in pericyte-deficient adult Pdgfb^ret/ret mice. We find that endothelial cells without pericyte contact retain a general BBB-specific gene expression profile, however, they acquire a venous-shifted molecular pattern and become transformed regarding the expression of numerous growth factors and regulatory proteins. Adult Pdgfb^ret/ret brains display ongoing angiogenic sprouting without concomitant cell proliferation providing unique insights into the endothelial tip cell transcriptome. We also reveal heterogeneous modes of pericyte-deficient BBB impairment, where hotspot leakage sites display arteriolar-shifted identity and pinpoint putative BBB regulators. By testing the causal involvement of some of these using reverse genetics, we uncover a reinforcing role for angiopoietin 2 at the BBB.

Conclusions:

By elucidating the complexity of endothelial response to pericyte deficiency at cellular resolution, our study provides insight into the importance of brain pericytes for endothelial arterio-venous zonation, angiogenic quiescence and a limited set of BBB functions. The BBB-reinforcing role of ANGPT2 is paradoxical given its wider role as TIE2 receptor antagonist and may suggest a unique and context-dependent function of ANGPT2 in the brain.

Graphical Abstract

INTRODUCTION

Pericytes (PCs) and vascular smooth muscle cells (VSMCs), collectively referred to as vascular mural cells, form together with the endothelial cells (ECs) the cellular component of the blood vessel wall. While VSMCs surround arteries and veins and, depending on vessel type, regulate vascular tone and diameter, PCs are typically embedded within the basement membrane (BM) of blood capillaries, where they establish direct contacts with ECs.¹ PC morphology, distribution and density vary between organs and vascular beds. The central nervous system (CNS) microvasculature is considered to have a high PC density compared to peripheral organs,² suggesting a specific importance of PCs in the CNS. Indeed, studies suggest that PCs are critically required for the establishment of fully functional blood-brain and blood-retina barriers (BBB, BRB).^3–6 Developmental absence or reduction of PCs leads to increased endothelial transcytosis and misexpression of molecular transporters and leukocyte adhesion molecules.^3–6,7 A number of CNS diseases have been associated with PC loss or damage, including diabetic retinopathy,^8–10 stroke,¹¹ Alzheimer’s disease,^12–14 amyotrophic lateral sclerosis¹⁵ and aging.¹⁶ However, whether PC damage contributes causally, and how and to what extent pericytes regulate the functional state of ECs in the pathogenesis of these conditions, are still not clear.

Platelet-derived growth factor-B (PDGFB) signaling through PDGF receptor-beta (PDGFRB) is essential for PC recruitment during developmental angiogenesis and postnatal survival in mice.^17,18 PC-deficient adult mice can however be generated through partial PDGFB/PDGFRB loss-of-function approaches. The most extensively PC-deficient brain microvasculature compatible with adult life observed so far occurs in mice carrying a homozygous deletion of the PDGFB heparan sulfate proteoglycan-binding motif (Pdgfb^ret/ret mice).^19,20 These mice show increased BBB permeability, dysregulated CNS immune homeostasis, and defective para-vascular fluid transport, and they model the human genetic disease primary familial brain calcification.^3,7,21–26 Here, we characterize the EC consequences of PC loss in Pdgfb^ret/ret mice using single-cell RNA-sequencing (scRNAseq) in combination with high-resolution imaging and genetic analyses of candidate molecular mediators of permeability changes. Our data show that PCs regulate EC identity and function in multiple ways. We identify two distinct modes of BBB breakdown in Pdgfb^ret/ret mice and uncover an unexpected role of angiopoietin 2 (ANGPT2) at the BBB. Our data illustrate the power of scRNAseq to decode EC heterogeneity and provide a resource for the further analysis and interpretation of microvascular abnormalities in the adult brain.

METHODS

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request. The single cell raw sequencing data and also processed counts table of our study are available from the NCBI Gene Expression Omnibus (GSE146007). Previously published wt EC scRNAseq dataset used in this study is also available in the GEO repository (GSE98816).

A complete description of the methods and reagents is provided in the Supplemental Material section.

RESULTS

A major proportion of the Pdgfb^ret/ret cortical microvasculature is devoid of pericyte contact.

Constitutive Pdgfb knockout mice (Pdgfb^−/−) die perinatally displaying >95% loss of brain PCs.^17,27 By comparison, Pdgfb^ret/ret mice¹⁹ show 70–80% brain PC loss and survive postnatally (Figure 1A).³ EC hyperplasia was reported in Pdgfb^−/− embryos²⁸ but we concluded that overall EC numbers were normal in the adult Pdgfb^ret/ret brain using immunofluorescent (IF) staining of EC nuclei by ERG (erythroblast transformation specific-related gene) antibodies (Online Figure IA). However, ECs formed a sparser network of vessels with increased diameter and numbers of ECs per vessel length in these mice (Figure 1A insets a,b and Online Figure IB).

Figure 1. Vascular abnormalities in Pdgfb^ret/ret mice.

A, IF and Pdgfrb-EGFP analysis of sagittal sections from adult brain cortex. Insets a-c illustrate vascular morphologies and cellular composition. Scale bars 25 μm. B, Quantification of tip cells per field as percentage of the total EC number (n=5, 6 fields per mouse). C, Representative confocal images of PC IF, asterisk marks the cell soma. Scale bar 20 μm.D, Quantification of PC:EC ratio (Pdgfb^ret/+ n=7 and Pdgfb^ret/ret n=6, 6 fields per mouse). E, Quantification of PC length in cerebral cortex. Left graph illustrates the length range in all measured PC (n=5 mice, Pdgfb^ret/+ n=78 PC and Pdgfb^ret/ret n=24 pericytes). Right graph shows mean PC length per mouse (n=5). In D,E, data was normally distributed and significance evaluated using unpaired two-tailed t-test with Welch’s correction. In B,D,F,G,J, data was unevenly distributed and significance evaluated using non-parametric Mann Whitney test. In B and D, the data are presented as geometric mean ± geometric SD, and in B, as mean ± SD.

We found numerous angiogenic sprouts with distinct tip cells²⁹ in adult Pdgfb^ret/ret brains, whereas these were rarely seen in age-matched wildtype (WT, Pdgfb^+/+) or heterozygous (Pdgfb^ret/+) controls (Figure 1A inset c). By IF for ERG and PECAM1 (platelet endothelial cell adhesion molecule 1), we calculated the tip cell frequency to ≈0,6% of the total EC number in Pdgfb^ret/ret brain (Figure 1B). Labeling with EdU (ethynyl deoxyuridine) showed that the tip cells were not cycling (Online Figure IC,D), confirming previous observations in the developing mouse retina.²⁹ Neither overall cell proliferation, nor EC proliferation or death, the latter judged by cleaved caspase-3 IF, were significantly increased in Pdgfb^ret/ret cerebral cortex (Online Figure ID,E). The unchanged turnover of ECs suggests that the observed angiogenic sprouting in Pdgfb^ret/ret brains is either abortive (sprouts fail to establish patent capillary connections) or reflects a balanced formation and regression of vessel branches.

To assess PC morphology and density, we used IF analysis for aminopeptidase N (ANPEP) and two transgenic reporters: chondroitin sulfate proteoglycan 4 promoter-driven dsRED (Cspg4-dsRED) and PDGF receptor-beta promoter-driven green fluorescent protein (Pdgfrb-GFP)^30,31 in combination with Hoechst nuclear staining and EC detection by IF for ERG and PECAM1. These analyses showed that while residual PC in Pdgfb^ret/ret were elongated, the PC:EC ratio in Pdgfb^ret/ret cerebral cortex was ≈8-fold lower than in Pdgfb^ret/+ littermates (1:54 vs. 1:7) (Figure 1C-E) and >10-fold lower than separately bred Pdgfb^+/+ mice, where the PC:EC ratio was estimated to 1:4. The slight reduction in PC density in Pdgfb^ret/+ compared to Pdgfb^+/+ does not correlate with any obvious differences in vascular morphology and function, however, according to our previous analyses.³

Pdgfb^ret/ret capillaries with residual PC contact showed a normalized diameter (Figure 1C), suggesting a direct effect of PCs on capillary diameter. While in Pdgfb^ret/ret each PC contacted an extended capillary length, a major proportion (≈75%) of the capillary length remained devoid of visible PC contact. In contrast, both Pdgfb^ret/+ and WT mice showed PC contact along ≈100% of the brain capillary vessel length (Figure 1A), confirming previous observations.³²

Although less severely than PCs, also VSMCs were affected in Pdgfb^ret/ret brains. Pdgfb^ret/ret arteries and veins maintain VSMCs coverage but we observed a reduction in the alpha-smooth muscle actin (αSMA a.k.a. ACTA2) staining of larger arteries (Online Figure IF), as well as a shift from circumferential to longitudinal orientation and reduced vessel coverage by VSMC at these sites.

Endothelial transformation and skewed arterio-venous zonation in Pdgfb^ret/ret mice.

We applied scRNAseq analysis using fluorescence-activated cell sorting (FACS) and the SmartSeq2 protocol³³ to investigate how PC deficiency affects EC gene expression. We first generated Pdgfb^ret/+ and Pdgfb^ret/ret mice carrying Claudin 5 (Cldn5)-GFP and Cspg4-DsRed, transgenic reporters for EC and PC, respectively.^31,34 Messenger RNA contamination between ECs and PCs is frequently observed in public scRNAseq datasets,^35,36 likely reflecting that the two cell types, firmly adjoined by a common BM, are difficult to separate during tissue disintegration. We therefore used the double transgenic reporters to deselect contaminated cells during FACS.³⁷ Pure brain EC (Cldn5-GFP⁺, Cspg4-DsRed^-) from 3-month-old mice were sorted into 384-well-plates and RNA-sequenced (Figure 2A).

Figure 2. Transcriptomic changes in EC devoid of PC contact.

A, Schematic overview of EC isolation for scRNAseq. B, Dot plot of top markers for 9 Pagoda2 EC clusters. Dot size illustrates percentage of cells presenting transcript sequence count(s) and color illustrates the average expression level in the cells within a cluster. C, The 9 EC clusters from B visualized in a UMAP landscape (left), UMAP distribution by genotype (right). D, Genotype proportions in each cluster. Total proportion is shown in the right column. E, tSpace projections of all EC labeled by cluster assignment (left) and genotype (right). F,G, IF analysis of A-V marker expression as indicated. F inset a illustrates that EC devoid of PC lack MFSD2A expression. Inset b shows that EC with residual PC contact retain MFSD2A expression. Scale bar 25 μm. G, IF staining against the capillary/venous markers SLC16A1 and TFRC. PODXL (podocalyxin-like) stains all EC. Arrowheads point to veins and arrows to arterioles. Scale bar 40 μm.

1,187 quality-controlled single cell transcriptomes from Pdgfb^ret/+ and Pdgfb^ret/ret mice were obtained and compared to scRNAseq data reported previously from WT mice of the same genetic background.^34,37 A total of 2,389 EC transcriptomes distributed into 9 clusters using the Pagoda2 package (Figure 2B).³⁸ Based on previously established markers for arterio-venous (A-V) zonation in the adult mouse brain,³⁴ some clusters were identified as arterial/arteriolar (#8 and 9), capillary (#4 and 6) or venous (#3) EC (Figure 2C). Gene Ontology (GO) analysis of cluster-enriched transcripts showed significantly different association with terms for biological processes such as vascular development (cluster #9), cell migration (cluster #8), molecular transport (cluster #6) and cell adhesion (cluster #3), implying functional differences related to A-V position (Online Figure IIA and Online Table I). Two clusters (#5 and #7) were characterized by high expression of immediate early genes (IEG) and interferon response genes (IRG), respectively, and the enriched transcriptomes of these two clusters were associated with GO terms encompassing metabolic processes (#5) and response to virus (#7), respectively (Online Figure IIA and Online Table I). We consider it likely that the IEG response was established during the experimental procedure as a result of cell stresses induced by tissue digestion. While we do not have an explanation for the underrepresentation of Pdgfb^ret/ret cells in the IRG cluster, it is possible that it is caused by stochastic exposure to double-stranded RNA released from lysed cells.^34,39,40 Similar clusters have been observed in scRNAseq datasets from other organs⁴¹, and they will not be discussed further herein. Finally, two clusters were enriched with transcripts that encode proteins associated with angiogenesis (#2) and growth factor signaling (#1). Because cluster #1 represents the strongest deviation from normal, we named this cluster “transformed”. The enriched transcriptomes of both clusters #1 and #2 were associated with GO terms for developmental processes and cell migration (Online Figure IIA and Online Table I), similar to the arterial cluster #9 in spite of differences in clustering and specific gene expression (see further below).

In a UMAP (uniform manifold approximation and projection) landscape, the arterial/arteriolar and venous clusters were located at opposite ends, separated by the capillary/venular clusters (Figure 2C). The transformed and angiogenic clusters (#1 and #2) were on the side of this trajectory, but closer to the capillary-venous clusters. The relative distribution of cells of the three genotypes (Pdgfb^ret/ret, Pdgfb^ret/+ and Pdgfb^+/+) in the different clusters showed that cluster #1 was almost completely composed of, and cluster #2 dominated by, Pdgfb^ret/ret cells (Figure 2C,D and Online Figure IIB). Conversely, the microvascular clusters (capillaries #4, #6 and arterioles #8) were underrepresented with Pdgfb^ret/ret cells (Figure 2D). The contribution of Pdgfb^ret/ret cells to the arterial and venous clusters was instead roughly proportional to their overall representation in the dataset (≈25%), suggesting that the arterial and venous EC phenotypes remain comparably intact in the Pdgfb^ret/ret brain (Figure 2D). Capillary EC distributed into two clusters (#4 and #6) with different proportions of Pdgfb^ret/+ and Pdgfb^+/+ cells, however, with similar marker expression and position in UMAP (Figure 2B,C). The cluster #4 vs. 6 assignment may therefore reflect a technical batch effect or minor quantitative gene expression differences provided by the two control genotypes. The most differentially expressed genes for each cluster are shown in Figure 2B. Lists of enriched genes for clusters #1–9 are provided in Online Table II. The gene expression patterns of all genes across the clusters are available gene-by-gene as bar-plot data at https://betsholtzlab.org/Publications/RetECscRNAseq/database.html.

To define nearest-neighbor relationships between cells, we next applied trajectory space (tSpace) analysis, which recapitulates dynamic and spatial relationships better than UMAP in multiple organisms and processes.^42–44 tSpace defined three principal branches in our data, representing arterial, venous and angiogenic ECs, all joined by a centrally located population of capillary/microvascular cells (Figure 2E). The tSpace branches correspond roughly to the different corners of the UMAP landscape but provide clearer separation of mature vs. transformed/angiogenic subsets of EC.

Taken together, these analyses suggest that a major proportion (albeit not all) of the Pdgfb^ret/ret brain microvascular ECs display a substantially transformed gene expression. In addition to the upregulated expression of molecules involved in angiogenesis and growth factor signaling, the cells of clusters #1 and #2 show decreased expression of certain transporters (Mfsd2a, Tfrc, Slc16a1, Slc38a3) (Figure 2B), suggesting that PC contact is required for ECs to assume and/or maintain certain specific BBB functions. Nevertheless, the phenotypic transformation of EC that we observe in the absence of PCs does not reflect a general loss of BBB markers. Many established BBB markers, including the tight junction molecule claudin 5 (Cldn5), the glucose transporter Glut 1 (Slc2a1), other influx transporters (Slco1c1, Slco1a4, Slc40a1), efflux transporters (Abcb1a a.k.a. P-glycoprotein) and the BBB-specific transcription factor Zic3, were all normally expressed (Online Figure III). Out of 64 CNS-specific EC markers⁴⁵ regulated by Wnt,⁴⁶ 20 were downregulated and 2 upregulated in Pdgfb^ret/ret cells (Online Table III), but these changes were modest (2-fold or less). Moreover, up- and downregulation, respectively, of venous (Slc38a5 and Nrp2) and capillary (Mfsd2a, Slc16a1, Slc1a1) markers³⁴ suggested venous skewing of the EC identity rather than loss of BBB functions (Online Figure IVA). To deepen the analysis of A-V zonation, we assessed 64 top venous and arterial markers of brain ECs³⁴ across the 9 clusters, confirming that Pdgfb^ret/ret ECs are venous-shifted with several venous markers upregulated including Nr2f2 and Icam1 (Online Figure IVB). Although less conspicuously, some arterial markers were also upregulated, including Bmx and Gkn3, possibly suggesting a mixed A-V identity in the Pdgfb^ret/ret microvasculature (Online Figure IVC).

To confirm the aberrant marker expression and disturbed A-V zonation in the Pdgfb^ret/ret brain vasculature at the tissue level, we analyzed protein and transgenic reporter expression (Figure 2F,G and Online Figure IIC,D). Combined MFSD2A^34,47 and ANPEP^48,34 IF analysis showed that MFSD2A was lost specifically in EC devoid of PC contact (Figure 2F), suggesting that PCs control EC expression of MFSD2A directly through para- or juxtacrine signals. This conclusion was also supported by Mfsd2a-CreERT2-induced Ai9 reporter⁴⁹ expression, which labeled 84% of the capillary length in controls compared to 32% in Pdgfb^ret/ret (Online Figure IIC,D). Similarly, expression of the capillary-venous markers solute carrier family 16 member 1 (SLC16A1) and transferrin receptor C (TFRC)^3,34 were focally lost in Pdgfb^ret/ret brain microvessels (Figure 2G). Taken together, scRNAseq and tissue imaging show disturbed EC differentiation and A-V zonation in the Pdgfb^ret/ret brain, implicating a critical role for PC in the establishment of proper vascular hierarchy in this organ.

Endothelial heterogeneity hallmarks the Pdgfb^ret/ret microvasculature.

We next focused our attention to the cells of the transformed and angiogenic clusters (#1 and #2) dominated by Pdgfb^ret/ret cells. A comparison between all Pdgfb^ret/ret and all control (Pdgfb^ret/+ and Pdgfb^+/+) cells revealed almost 1,000 differentially expressed genes (Online Table IV; the top 25 up- or downregulated genes are shown in Online Figure V). Although a few cells in clusters #1 and #2 originated from Pdgfb^ret/+ and Pdgfb^+/+ controls, the Pdgfb^ret/ret cells distinguished from these by being uniquely enriched with a number of transcripts, including fibroblast growth factor-binding protein-1 (Fgfbp1), connective tissue growth factor (Ctgf), bone morphogenetic protein (Bmp) −4 and −6 and Angpt2 (Figure 3A). Intriguingly, high expression of these genes was seen also in the Pdgfb^ret/ret cells present in the venous (#3) and capillary (#4) clusters (Figure 3A). Fgfbp1 encodes a secreted protein that binds FGFs and promotes signaling through FGFRs,^50,51 and it has been functionally implicated in embryonic angiogenesis in the chick⁵² and BBB development in the mouse.⁵³ Fgfbp1 expression is high in mouse embryonic brain EC (our unpublished data) but negligible in adult brain EC.^34,36 Our results therefore suggest that the absence of PC causes a molecular transformation of capillary EC, including loss of markers for brain capillary specialization and concomitant gain of venous markers, as well as upregulation of transcripts normally expressed during development (Fgfbp1) or in pathological angiogenesis (Angpt2).

Figure 3. Pdgfb^ret/ret EC heterogeneity.

A, ScRNAseq bar plots showing the expression of selected genes that are upregulated in clusters #1 and #2. B, Upper panel: Re-clustering of #1 and #2 results in 5 subclusters (i-v). Lower panel: Dot plot showing enriched genes for subclusters i-v. Dot size illustrates percentage of cells presenting transcript sequence count(s) and color illustrates the average expression level in the cells within a cluster. C, Bar plots of representative markers for each subcluster. D, IF visualization of tip cell (asterisk) in cerebral cortex. Arrows point at low-MCAM expressing PC. Scale bar 20 μm. E, Heatmap of i-1-enriched markers across i-1 and i-2 cells sorted by SPIN. F, GO analysis of subcluster i-1-enriched genes. Terms for biological process are shown. G-J, tSpace projections. G, Supervised (left) and unsupervised (right) mutual nearest neighbor (MNN) alignment of EC by clusters. H-J, Cell distribution and gene expression in MNN-aligned graphs, as indicated. Art- arterial, Ven- venous, Ang- angiogenic.

To explore the putative heterogeneity of the cells of clusters #1 and #2, we re-clustered them into 5 subclusters (i-v in Figure 3B). Subcluster i, comprising 90 cells, showed expression of several genes proposed as markers for tip cells, including Mcam and Angpt2 (Figure 3C).^36,54–57 It also contained Lamb1-positive cells. IF analysis of MCAM and LAMB1 confirmed the specific expression of these two proteins in tip cells (Figure 3D). MCAM is expressed also by PCs¹, but at substantially lower level compared to tip cells (Figure 3D).

Not all cells in subcluster i were Mcam and Lamb1 positive. To assess putative further heterogeneity in subcluster i, we sorted the transcriptomes uni-dimensionally using the SPIN algorithm (sorting points into neighborhoods),⁵⁸ resulting in two groups, i-1 and i-2, consisting of 7 and 83 cells, respectively (Figure 3E). i-1 matched the expected tip cell profile, including expression of Mcam and Lamb1 along with previously suggested tip cell markers Cxcr4,⁵⁶ Pgf,⁵⁹ Kcna5, Sirpa, and Smoc2.⁴⁵ GO analysis of i-1 markers showed association with actin filament assembly and filopodia (Figure 3F) as expected for tip cells.²⁹ The number of i-1 cells constituted 1.2% of the total number of analyzed Pdgfb^ret/ret EC, i.e. roughly comparable to the tip cell frequency counted in Pdgfb^ret/ret brain (0.6%).

Subcluster ii was enriched with capillary markers³⁴ (e.g. Mfsd2a and Tfrc) and dominated by Pdgfb^ret/ret cells (Figure 3C). Subclusters iii and iv were enriched for venous markers⁶⁰ (e.g. Nr2f2, Bmp4) with iv displaying IEGs on top (Figure 3B), presumably reflecting gene activation during tissue dissociation, as discussed above. Subcluster v was positive for arteriolar EC markers⁶⁰ (Gja4, Bmx, Gkn3, Sema3g) (Figure 3B). Thus, the EC heterogeneity revealed through subcluster analysis appears to reflect in part a “shadow” A-V identity among the angiogenic and transformed ECs present in Pdgfb^ret/ret brain. For complete lists of enriched genes for each subcluster, see Online Table V.

As a complement to the supervised assignment of A-V shadow identity using selected A-V marker distribution in subclusters, we also analyzed the relationships between subclusters i-v and their mature counterparts (global clusters #3–9) using unsupervised mutual nearest neighbor (MNN) alignment⁶¹ and visualization in tSpace (Figure 3G-J and Online Figure VIA,B). MNN alignment provides batch correction, i.e. it subtracts global differences between genotypes, such as Fgfbp1 and other genes hallmarking the Pdgfb^ret/ret “batch” (Figure 3G,H). With this analysis, the transformed arterial cells stayed at more proximal locations in the arterial tSpace branch (weaker shadow identity) compared to the transformed venous cells, which distributed more distally into the venous branch (stronger shadow identity) (Online Figure VIA). The angiogenic EC were still positioned within their own branch after MNN alignment, with cluster i-1 cells (tip cells) preferably at distal position (Figure 3I,J and Online Figure VIB). In conclusion, both supervised cluster annotation (Figure 3C) and unsupervised MNN alignment (Figure 3G-J and Online Figure VIA,B) suggested a heterogenous A-V shadow identity with venous preference among the transformed EC, and the presence of angiogenic EC with a distinguishable tip cell transcriptome.

Endothelial phenotypes correlate with BBB dysfunction.

The BBB leakage observed in Pdgfb^ret/ret appears to be caused in part by endothelial transcytosis.³ Systemically administered tracers between 1–200 kDa in size accumulate progressively in the Pdgfb^ret/ret brain over 2–16 hours.³ This accumulation is, however, uneven across the brain, which might reflect local differences in residual pericyte contact and/or different modes of BBB impairment in different vessel types along the A-V axis. We therefore set out to explore the correlation between gene expression and BBB dysfunction, focusing on subclusters iii, iv and v, which were dominated by ECs from Pdgfb^ret/ret (Figure 3C). ANGPT2 is known to destabilize blood vessels through its antagonistic function at the tyrosine kinase receptor TIE2 (reviewed in⁶²). MFSD2A on the other hand has a BBB-stabilizing role (its loss leads to increased brain endothelial transcytosis^63,64), We therefore anticipated that subclusters iii and iv, in which Angpt2 was high and Mfsd2a low, represented the ECs with the most extensively impaired BBB. However, by correlating the leakage of systemically administered 70 kDa tetramethylrhodamine (TMR)-dextran with ANGPT2 and MFSD2A expression in tissue sections, we instead found that the most extensive (hotspot) TMR-dextran leakage occurred at sites where both ANGPT2 and MFSD2A were low, i.e. corresponding to the gene expression pattern of subcluster v (Figure 3C). Thus, in addition to a broadly observable vesicular accumulation of TMR-dextran (Figure 4A) in ECs, consistent with transcytosis,³ we detected hotspot leakage sites (Figure 4B). Neither type of leakage was observed in controls. Whereas areas with detectable transcytosis displayed MFSD2A expression (Figure 4A), only about 10% of the hotspot leakage sites had detectable MFSD2A (Figure 4B,C). ANGPT2 was barely detectable by IF staining in control brains, but strongly and heterogeneously expressed in Pdgfb^ret/ret vessels, including but not limited to tip cells (Figure 4A,B). However, ANGPT2 was low or undetectable at most hotspot leakage sites (Figure 4B), and conversely, most sites with high ANGPT2 staining did not display hotspot leakage (Figure 4B). Quantifications revealed that only ≈11% of the hotspot leakage sites showed detectable ANGPT2 (Figure 4C). ANGPT2 expression was mainly observed in vessels devoid of mural cells, suggesting that it depends on local rather than systemic cues (Figure 4D). Plasmalemma vesicle-associated protein (PLVAP) was upregulated at the mRNA level (Plvap) in Pdgfb^ret/ret brains in subcluster v cells (Figure 3C) and an irregular PLVAP staining could be detected at a majority (59%, Figure 4C) of the hotspot leakage sites (Figure 4E). PLVAP is normally found in fenestrated blood vessels, including the capillaries of the brain’s circum-ventricular organs, as exemplified by the organum vasculosum of the lamina terminalis and choroid plexus in the lateral ventricle, but otherwise not in the healthy BBB (Online Figure VIIA). Additionally, the leukocyte adhesion molecule, intercellular adhesion molecule 1 (Icam1) mRNA was upregulated in the Pdgfb^ret/ret dominated subclusters iii, iv, and v (Figure 3C), a finding supported by ICAM1 protein detection by IF. In adult control brain, ICAM1 was detectably expressed mainly in veins and arteries (Figure 4F, and data not shown), whereas in Pdgfb^ret/ret, it was broadly upregulated in PC-deficient capillaries (Figure 4F and Online Figure VIIB,C). ICAM1 staining was observed at most (92%, Figure 4C) of the hotspot leakage sites. We also found extravasated CD45-positive leukocytes in Pdgfb^ret/ret brains (but not in controls) often clustered at hotspot leakage sites (Online Figure VIID). Finally, we observed abnormal and irregular patterns of the adherens and tight junction proteins VE-cadherin (CDH5) and CLDN5 at hotspot leakage sites devoid of mural cells (Figure 4G,H), suggesting that EC junctions are altered at these sites.

Figure 4. Hotspot leakage sites in Pdgfb^ret/ret brains.

A,B, Confocal images of cerebral cortex of 70 kDa tetramethylrhodamine (TMR)-dextran (red) injected mice, together with IF staining for the indicated proteins A, Inset depicts MFSD2A^low vasculature with increased TMR-dextran positive vesicles (transcytosis). Scale bar 20 μm. B, Examples of TMR-dextran positive hotspot permeability sites (arrowheads). Inset shows ANGPT2-positive tip cell negative for TMR-dextran. Scale bar 20 μm. C, Quantification of colocalization between TMR-dextran hotspots and indicated markers (n=3). D, Upregulated ANGPT2 expression where PC contact is absent. Scale bar 40 μm. E, Illustration of PLVAP staining at TMR-dextran positive hotspot permeability sites. Scale bar 20 μm. F, The majority of TMR-dextran positive hotspot permeability sites are MFSD2A-negative and ICAM1-positive. Scale bar 40 μm. G, Adherens junctions stained for CDH5 are thin and regular in controls (all EC stained by anti-glucose transporter type 1 (SLC2A1)), but discontinuous and irregular at TMR-dextran positive hotspot leakage sites. Adherens junctions at non-hotspot permeability sites are irregular but continuous (middle panel). ANPEP visualizes PC. Scale bar 10 μm. H, Tight junctions visualized using anti-CLDN5 staining are continuous in Pdgfb^ret/+ (all EC visualized by anti-PECAM1), but discontinuous and irregular at TMR-dextran positive hotspot leakage sites. Tight junctions in non-hotspot leakage sites in Pdgfb^ret/ret are continuous (middle panel). Scale bar 10 μm.

Collectively, these results suggest two distinct modes of BBB dysfunction in Pdgfb^ret/ret. On the one hand hotspot leakage sites devoid of PC clearly visible by 70 kDa TMR-dextran extravasation; these sites were characterized by an arteriolar EC expression profile (subcluster v in Figure 3C), disturbed junctions, low or missing expression of MFSD2A, upregulated ICAM1 expression, extravasation of leukocytes, and ectopic PLVAP expression. On the other hand, the previously characterized increase in vesicular transcytosis across the EC in Pdgfb^ret/ret brain;³ this activated transport leads to widely distributed tracer leakage in Pdgfb^ret/ret brains and seem to correlate with increased expression of a large number of mRNAs and proteins, including Fgfbp1, Ctgf, Angpt2, and Bmp4.

In order to provide mechanistic insight into the BBB leakage in Pdgfb^ret/ret brain, we chose Mfsd2a, Fgfbp1, and Angpt2 for further functional interrogation using reverse genetics. Mfsd2a was selected for study because it was downregulated in transformed EC in Pdgfb^ret/ret and has been shown to suppress EC transcytosis in the intact BBB.^63,64 Fgfbp1 was selected because it is the most strongly upregulated gene in Pdgfb^ret/ret brain EC and has previously been implicated in BBB regulation.^52,53 Angpt2, finally, was selected because it is upregulated in Pdgfb^ret/ret transformed EC and is known to enhance vascular permeability by antagonizing ANGPT1 signaling through TIE2.⁶²

Loss of Mfsd2a or Fgfbp1 do not mediate major BBB permeability in Pdgfb^ret/ret mice.

We compared the distribution of two tracers with different sizes, 1 kDa Alexa Fluor 555-(A555)-cadaverine (Figure 5A,B) and 70 kDa TMR-dextran (Figure 5C) following intravenous injection and circulation in adult awake Mfsd2a^−/− or Pdgfb^ret/ret mice. If MFSD2A mediates PC-suppressed EC transcytosis, we had expected to see similar or higher tracer leakage in Mfsd2a^−/− than in Pdgfb^ret/ret mice. However, under the experimental conditions used, neither tracer showed significantly increased accumulation in Mfsd2a^−/− compared to Mfsd2a^+/+ brains, in marked contrast to the strong leakage observed in Pdgfb^ret/ret brains (Figure 5A-C). MFSD2A has been shown to be a critical lipid transporter at the BBB, responsible for the transport of the essential omega-3 fatty acid docosahexaenoic acid (DHA) into the brain,^47,65,66 and DHA levels in Mfsd2a^−/− brains are markedly reduced compared to controls.⁴⁷ In spite of the reduced MFSD2A levels observed in Pdgfb^ret/ret brain vessels, lipidomic analysis did not reveal detectable changes in lipid composition in Pdgfb^ret/ret brains (Figure 5D), suggesting that the residual MFSD2A expression observed in these mice is sufficient to maintain DHA transport. Together, these observations demonstrate that the increased permeability of the Pdgfb^ret/ret brain vasculature is caused by other mechanisms than loss of MFSD2A.

Figure 5. Lack of role of Mfsd2a and Fgfbp1 in Pdgfb^ret/ret BBB leakage.

A,B, Intravenously administered 1 kDa Alexa Fluor (A555) cadaverine permeability after two hours circulation in adult Mfsd2a^+/+, Mfsd2a^−/− and Pdgfb^ret/ret mice. A, Graph shows quantification of extravasated tracer in the brain (Mfsd2a^+/+ n=7, Mfsd2a^−/− n=5 and Pdgfb^ret/ret n=7). Lower panel shows whole brains photographed after tracer circulation. B, Three-dimensional reconstructions of confocal image z-stacks of cerebral neocortex vasculature IF stained as indicated. Extravasated A555-cadaverine accumulates in neurons.¹⁷ Scale bar 20 μm. C, 70 kDa TMR-dextran accumulation into brains after 16-hour circulation. Lower panel illustrates whole brains photographed after tracer circulation. (Mfsd2a^+/+ n=4, Mfsd2a^−/− n=5 and Pdgfb^ret/ret n=3). D, Lipidomic analysis of aged Pdgfb^ret/ret and Pdgfb^ret/ret brains (Pdgfb^ret/+ n=7, Pdgfb^ret/ret n=6). DHA- docosahexaenoic acid, PC-phosphatidylcholin, PE- phosphatidylethanolamine, PS- phosphatidylserine. Data in A,C were normally distributed and significance was evaluated using One-way ANOVA with Bonferroni’s multiple comparison test. Significance in D was evaluated using unpaired two-tailed t-test with Welch’s correction. Data are presented as geometric mean ± geometric SD. n.s.=not significant.

To address if the second candidate, FGFBP1, mediates or protects against the increased BBB permeability in Pdgfb^ret/ret mice, we generated compound constitutive Fgfbp1^−/−;Pdgfb^ret/ret mice and studied their vasculature in adult brains. Examining these mice for BBB permeability, we could not detect any increased leakage in single Fgfbp1^−/− mice (Pdgfb^ret/+ background), nor could we detect increased leakage in double Fgfbp1^−/−:Pdgfb^ret/ret vs. single Pdgfb^ret/ret mutants (Online Figure VIII). From these data, we conclude that FGFBP1 does not cause BBB leakage in Pdgfb^ret/ret mice, nor does it protect against it.

An unexpected role for Angpt2 at the BBB.

Although most of the strong sites of ANGPT2 expression observed in Pdgfb^ret/ret mice did not co-localize with hotspot permeability sites (Figure 4B,C), Angpt2/ANGPT2 expression was generally increased in the EC of Pdgfb^ret/ret brains (Figure 3A,4B,D and Online Figure VIID), particularly in EC devoid of PC contact (Figure 4D). To analyze if increased Angpt2/ANGPT2 expression contributes to the BBB defects observed in Pdgfb^ret/ret mice, we induced deletion of Angpt2 in Pdgfb^ret/ret background. We generated Pdgfb^ret/ret:Cdh5-CreERT2:Angpt2^flox/flox mice and induced deletion of Angpt2 (Pdgfb^ret/ret;Angpt2^iECKO) by tamoxifen administration at 1 month of age, followed by analysis at 3 months of age. Contrary to our expectation, double Pdgfb^ret/ret;Angpt2^iECKO mutants were not protected against tracer leakage, but instead showed a marked increase in the brain accumulation of 1 kDa A555-cadaverine in comparison with single Pdgfb^ret/ret mice (Figure 6A). IF staining confirmed the efficient depletion of ANGPT2 in Pdgfb^ret/ret;Angpt2^iECKO mutants (Figure 6B).

Figure 6. Role of Angpt2 at the BBB.

A, Quantification of A555-cadaverine extravasation into brain tissue (Ctrl n=8, Pdgfb^ret/ret n=4, Angpt2^iECKO;Pdgfb^ret/ret n=4, control genotypes provided in Methods). B, Three-dimensional reconstruction of confocal image z-stacks of cerebral neocortex vasculature depicted by IF staining as indicated in Ctrl, Pdgfb^ret/ret and Angpt2^iECKO;Pdgfb^ret/ret mice. Extravasated A555-cadaverine accumulates in neurons. Scale bar 20 μm. C, CLDN5 distribution is irregular in both Pdgfb^ret/ret and Angpt2^iECKO;Pdgfb^ret/ret brains. Scale bar 20 μm. D, Quantification of the A555- cadaverine extravasation into brain tissue (Ctrl n=4, Angpt2^−/− n=5, control genotypes are provided in Methods). E, Representative images of the vascular mural cell coverage in cerebral cortex (n=3). Scale bar 40 μm. F, Representative images of the CDH5 and CLDN5 junctional distribution in EC in cerebral cortex (n=3). Insets in Angpt2^+/+ illustrate continuous and overlapping CDH5 and CLDN5, whereas insets in Angpt2^−/− panel show continuous CDH5 but discontinuous CLDN5 (arrows). Scale bar 20 μm. Data in A and D was normally distributed and significance evaluated using One-way ANOVA with Bonferroni’s multiple comparison test. Data are presented as geometric mean ± geometric SD.

The aggravated tracer leakage in the Pdgfb^ret/ret;Angpt2^iECKO mice was unexpected in view of the reported role for ANGPT2 in vessel destabilization. To further address this paradoxical result, we analyzed constitutive Angpt2 knockout mice (Angpt2^−/− mice), which demonstrated that also single Angpt2^−/− mice displayed abnormal brain vascular morphology and increased leakage of A555-cadaverine tracer (Figure 6D). The mural cell abundance and zonation judged by pericyte and VSMC marker expression was normal in Angpt2^−/− mice, pointing to a different mechanism(s) of BBB disruption than in Pdgfb^ret/ret mice (Figure 6E). We therefore focused our attention to EC junction morphology and the expression of markers of adherens junctions (CDH5) and tight junctions (CLDN5). While the distribution of both proteins confirmed the expected abnormalities in EC shapes, the Angpt2^−/− vessels displayed specific alterations in the distribution of CLDN5, which was both weaker and discontinuous along the cell-cell junctions, something that was not observed for CDH5 (Figure 6F). In contrast, control mice showed largely overlapping and continuous CDH5 and CLDN5 staining along the EC junctions. These data point to an important role for ANGPT2 in the development of CLDN5-containing tight junctions in the brain vasculature, as well as in maintenance of the integrity of these junctions during pericyte deficiency.

DISCUSSION

Our analysis provides detailed insight into how PC-deficiency changes the identity and behavior of brain EC in multiple ways with consequences for BBB integrity.

We found that the molecular A-V identity in EC was skewed in the PC-deficient brain. Out of 64 markers for brain venous EC, more than half were upregulated in Pdgfb^ret/ret EC. Normally, a continuum of gene expression changes is observed along the A-V axis, including nested gene expression patterns of mRNAs coding for transcription factors, transporters and other protein classes.³⁴ This A-V zonation is likely a prerequisite for proper positioning of vascular functions along the A-V axis, such as molecular transport and immune cell trafficking. The mechanisms establishing endothelial zonation in the brain and elsewhere are generally not known, but our current findings suggest a role for PCs in this process.

We also addressed how PC-deficiency affects organotypic specialization in the brain EC, i.e. the molecular specification of the BBB phenotype. BBB function is impaired in the pericyte-deficient state,^3–5 but it has remained unclear how the PC-deficient BBB compares to other states of BBB-deficiency. Inhibition of Wnt signaling causes profound downregulation of BBB genes.⁴⁶ Our present results show that PC-deficiency have comparably modest effects on the BBB. Most major BBB markers were normally expressed in Pdgfb^ret/ret mice, and where differences were observed, they seemed to reflect a venous-shifted rather than dismantled BBB. The lipid transporter MFSD2A is an interesting exception, as its expression in brain ECs requires direct PC contact. However, the reduced MFSD2A expression observed in Pdgfb^ret/ret mice was sufficed to maintain a normal brain lipid composition, and it did also not explain the BBB leakage, which was far more extensive in Pdgfb^ret/ret mice than in Mfsd2a^−/− under our experimental conditions.

The finding of angiogenic sprouts and typical tip cells was surprising considering the general vascular rarefaction in the adult Pdgfb^ret/ret brain. Analysis of EC cycling, apoptosis and total cell number suggested that the PC-deficient vessels were proliferatively quiescent. This provided us with an opportunity to molecularly characterize tip cells in absence of the concomitant vascular expansion that typically occurs during developmental and tumor angiogenesis. Indeed, we identified tip cells with a distinct gene and protein expression profile. While only a small number of tip cells was obtained, the depth of the SmartSeq2 technique provided meaningful information. To define markers unique for tip cells of the angiogenic sprout, we compared the tip (i-1) cells with the remaining cells (i-2) of the angiogenic cluster, which presumably include stalk cells. Of 112 enriched genes, 32 remained after multiple test correction. Clearly, this list is conservative. Moreover, because tip and stalk cells are known to dynamically switch position during angiogenic sprouting,⁶⁷ some lag in transcriptional changes between tip cells and stalk cells would be expected, as also indicated by tSpace analysis, which positioned i-1 cells preferably at the distal part of the angiogenic branch. Nevertheless, the 32-gene list of tip cell- vs. stalk cell-enriched transcripts includes Mcam and Lamb1, as well as previously proposed markers (Cxcr4,⁵⁶ Pgf,⁵⁹ Kcna5, Sirpa, and Smoc2⁴⁵) along with several novel and unannotated transcripts, including micro-RNAs without known function. The erythropoietin receptor (Epor) mRNA was among the most distinct tip cell markers in our dataset, which may be relevant in relation to the suggested roles for erythropoietin in tumor angiogenesis.⁶⁸ Intriguingly, many of the tip cell markers, including Epor, are shared with rare capillary-resident angiogenic precursors recently described in lymph nodes and other tissues.⁴² The significance of this similarity remains to be investigated.

Previous analyses have established transcytosis as a major mode of tracer transport across the BBB in PC-deficient mice.^3,5 Our present analysis revealed extensive EC heterogeneity and a putative second mode of BBB impairment in Pdgfb^ret/ret brains: hotspot leakage. By correlating mRNA profiles in the single-cell data with BBB permeability and protein expression patterns, we found that hotspot leakage sites had unique morphological and molecular characteristics, including the abnormal distribution of EC junctional molecules. Hotspot sites were hallmarked by low MFSD2A and low ANGPT2 expression. They also displayed high ICAM1 and detectable PLVAP, features of possible relevance for the observed focal extravasation of leukocytes.

The low ANGPT2 expression at the hotspot leakage sites in Pdgfb^ret/ret brain was surprising as ANGPT2 is conceived mainly as a TIE2 receptor antagonist, promoting vascular destabilization and pro-inflammatory responses.^{62, 69–75} However, ANGPT2 has also been shown to exert context-dependent TIE2 agonistic activity, particularly in lymphatic endothelium or lymphatic-like vessels.^76–78 While we had anticipated that the upregulated expression of ANGPT2 in Pdgfb^ret/ret mice exacerbated the BBB impairment, our data implicated the opposite, because double Pdgfb^ret/ret;Angpt2^iECKO mice showed an aggravated rather than a rescued BBB function. That ANGPT2 exerts a protective rather than destabilizing function at the BBB was further supported by data from single Angpt2^−/− mice, which showed increased tracer leakage, abnormal vascular morphology and weak irregular junctional staining for CLDN5. If this reflects an agonistic role of ANGPT2 at the TIE2 receptor specifically in a BBB microenvironment remains to be investigated. TIE2 agonistic functions of ANGPT2 have been demonstrated in lymphatic EC, where they were linked to low expression of the TIE2 phosphatase VEPTP (PTPRB).⁷⁷ However, our scRNAseq data showed normal Ptprb expression in Pdgfb^ret/ret EC (https://betsholtzlab.org/Publications/RetECscRNAseq/database.html).

Another proposed mechanism for ANGPT2-TIE-2 agonistic function involves translocation of ANGPT2-TIE2-integrin complexes from the BM to cell junctions, resulting in their stabilization and agonistic signaling.⁷⁹ Whether a specific molecular composition of the brain vascular BM prevents binding to ANGPT2-TIE2-integrin complexes and instead promotes a junctional location with resulting stabilization and reinforcement of CLDN5-containing tight junctions remains an intriguing possibility for further study.

It is increasingly appreciated that primary vascular dysfunction may cause or accelerate neurological disease.^{80, 81} Conversely, neurodegenerative diseases may negatively affect brain vascular function.^82–84 Among the emerging links between neurodegenerative and cerebrovascular diseases, PC have received special attention.⁸⁵ PC loss or dysfunction correlate with a wide range of neurological and neurodegenerative conditions, including diabetic retinopathy,⁸ Alzheimer’s and other neurodegenerative diseases,^{12–15, 86–89} aging¹⁶ and stroke.⁹⁰ Mouse models of PC deficiency involving loss-of-function mutations in the Pdgfb or Pdgfrb genes or PC ablation by diphtheria toxin⁹¹ display impaired BBB function,^3–5 reduced oxygen supply to the brain^{91, 92} and myelin degradation.⁹³ Hypomorphic Pdgfb mice also mimic primary familial brain calcification in humans^{23, 24} and display neurotoxic astrocyte activity and altered behavior and cognition suggestive of neuronal damage.²⁵ While all these studies point to a role for PC in overall brain function, it has remained unclear how PC deficiency affects their nearest neighbors, the ECs, at a molecular level. We undertook the present study using the most extensive, yet adult viable, mouse model of primary PC deficiency known, namely Pdgfb^ret/ret mice,^{3, 19, 20} in which ≈85% of the brain PC are lost, causing ≈75% of the capillary length to be devoid of visible PC contact. Because a the Pdgfb^ret/ret brain vasculature is sufficiently functional to support general brain functions and adult life, these mice are particularly suitable for analysis of how the EC phenotype changes when PC contact is missing. Our data show that the EC response to PC loss is complex, as may be expected also in aforementioned diseases. Our herein reported data and the appended online database https://betsholtzlab.org/Publications/RetECscRNAseq/database.html should guide further studies of pathogenic vascular states and their consequences in brain disease.

NOVELTY AND SIGNIFICANCE.

What Is Known?

• The blood-brain barrier (BBB) is a semipermeable layer of endothelial cells that selectively regulates solutes crossing into the extracellular fluid of the central nervous system where neurons reside.

• The BBB depends on signals from surrounding non-endothelial cell types, including the pericyte.

• Mice with low numbers of pericytes show increased passage of blood-borne molecules across the BBB.

• Several brain diseases have been associated with pericyte loss or damage and BBB disruption.

What New Information Does This Article Contribute?

• This study provides detailed information about how brain endothelial gene expression is reprogrammed when pericytes are missing.

• The expression of growth factors and pro-inflammatory mediators is increased, and the molecular profile is shifted from capillary to venous, while the BBB identity remains largely intact.

• Activated angiogenic sprouting without concomitant cell proliferation allowed identification of a specific molecular profile of endothelial tip cells.

• A new mode of BBB impairment induced by pericyte deficiency was uncovered, i.e. focal hotspot leakage sites with distinct gene and protein expression including the absence of angiopoietin 2 (Angpt2).

• Angpt2 null mice show a specific defect in claudin 5-containing tight junctions, suggesting that Angpt2 signaling reinforces endothelial cell-cell contacts when they are challenged by pericyte loss.

Pericytes are obligatory cellular constituents of blood micro-vessels, but their function(s) remain poorly understood. Pericytes are particularly numerous in the brain and play a poorly defined role in the development and regulation of the BBB,. Numerous cerebral diseases have been associated with pericyte loss or damage, but whether this is causally pathogenic or a secondary effect is not clear. Using a mouse model with extensive brain pericyte hypoplasia, we studied how the absence of pericytes influences gene expression and function in brain endothelial cells. We found an extensive endothelial heterogeneity, including specific alterations in the expression of molecular transporters, a venous-shifted identity, the activation of angiogenic sprouting, and two distinct modes of BBB disruption, one widespread, involving vesicular transcytosis, and one focal, involving disrupted tight junctions. The latter (hotspot leakage sites) showed lowered expression of Angpt2, which first appeared paradoxical, since Angpt2 is primarily an endothelial destabilizing signal outside of the brain. However, Angpt2 knockout mice demonstrated disrupted claudin 5-containing tight junctions in the brain, suggesting that Angpt2 is necessary for proper BBB development, and that reduced Angpt2 expression triggers BBB disruption when pericytes are missing.

Nonstandard Abbreviations and Acronyms:

ACTA2

alpha-smooth muscle actin

ANGPT2

angiopoietin 2

ANPEP

aminopeptidase N

A-V

arterio-venous

BBB

blood-brain barrier

BM

basement membrane

BMP

bone morphogenetic protein

BRB

blood-retina barrier

CDH5

vascular endothelial cadherin

CLDN5

claudin 5

CNS

central nervous system

CSPG4

chondroitin sulfate proteoglycan 4

CTGF

connective tissue growth factor

DHA

docosahexaenoic acid

EC(s)

endothelial cell(s)

EdU

ethynyl deoxyuridine

ERG

erythroblast transformation specific-related gene

FACS

fluorescence-activated cell sorting

FGFBP1

fibroblast growth factor-binding protein 1

GFP

green fluorescent protein

GO

gene ontology

ICAM1

intercellular adhesion molecule 1

IEG

immediate early genes

IF

immunofluorescence

IRG

interferon response genes

MFSD2A

major facilitator superfamily domain containing 2

MNN

mutual nearest neighbor

PC(s)

pericyte(s)

PDGF-B

platelet-derived growth factor B

PDGFRb

platelet-derived growth factor receptor beta

PC:EC

pericyte:endothelial cell ratio

PECAM1

platelet endothelial cell adhesion molecule 1

PLVAP

plasmalemma vesicle-associated protein

scRNAseq

single cell RNA sequencing

SLC16A1

solute carrier family 16 member 1

SPIN

sorting points into neighborhoods

TEK

tyrosine kinase receptor TIE2

TMR

tetramethylrhodamine

TFRC

transferrin receptor C

tSpace

trajectory space

UMAP

uniform manifold approximation and projection

VSMCs

vascular smooth muscle cells

WT

wild type