Fig 2: Microglia form direct purinergic contact with cells in the NVU that regulate CBF. (a) 3D reconstruction of in vivo two-photon Z-stacks down to 600 µm below the dura mater in the cerebral cortex of CX3CR1tdTomato mice. Note contacting microglia (arrowheads) at meningeal (m), penetrating (p), and first- to third-order capillaries. Scale bar, 50 µm. (b) Microglial processes (arrowheads) dynamically contact different segments of the vascular tree (visualized by i.v. FITC-dextran). Scale bar, 20 µm. (c) Microglial processes are extended beyond the perivascular glial endfeet and form direct contact with smooth muscle cells (arrows) at the level of penetrating arteries. (d) CLSM images show microglia (P2Y12R, cyan) contacting endothelial cells (tomato lectin, magenta) in the cerebral cortex. (e) EM images show microglia (m, P2Y12R-immunogold labeling, cyan) directly contacting endothelial cells (e, magenta) and pericytes (p, purple). (f) Frequency of vessels receiving microglial contact, and microglial process coverage of endothelial cell surface. (g) EM images show microglia (m, P2Y12R-immunogold labeling, cyan) directly contacting pericytes (p, purple). (h) 83.4 ± 1.4% of pericytic cell bodies are contacted by microglial processes. (i) 3D reconstruction of electron tomogram shows clustering of anti-P2Y12R-immunogold on microglial processes (m) directly contacting the endothelium (e) of an arteriole/postcapillary venule. The left two panels are conventional EM images of the same area on the adjacent ultrathin section. The right panels show a tomographic virtual section and 3D reconstruction of the direct contact. (j) Unbiased anatomic analysis reveals enrichment of endothelial mitochondria (TOM20+, green), at sites of microglial contacts (P2Y12R+, cyan). ****, P < 0.0001, Mann–Whitney U test. See analysis details in Fig. S1 d. (k) EM images show microglia (m, P2Y12R-immunogold, cyan) directly contacting endothelial cells (e, magenta) in human neocortex. z1–z3 panels show the contact on three consecutive ultrathin sections; arrows mark the edges of direct membrane contact. (l) CLSM image shows microglia (P2Y12R, cyan) contacting the cell body of an astrocyte (GFAP labeling, green) and astrocytic endfeet (arrowheads). (m) EM images show direct contact between microglial (m, cyan) and astrocytic (a, green) processes. e, endothelial cell, magenta; lum, lumen. (n) CLSM images in human neocortex reveal P2Y12R+ microglial processes (cyan) contacting perivascular astrocytes (GFAP, green) on astrocyte endfeet (white arrowheads) and endothelial cells (tomato-lectin, magenta, white arrows), with astrocytic endfeet directly touching the endothelial monolayer (empty arrowheads). (o) CLSM image and fluorescent intensity plots show microglial process (m) contacting the endothelial layer (e) within the astrocytic layer (a). (p) To reveal 3D connections of individual microglial cells, a CLSM maximum-intensity plot was generated. Microglia (cyan) contact several microvessels (lectin, magenta; AQP4, green) and neurons (Kv2.1, ochre) simultaneously. For appropriate visualization of neuronal cell bodies, Kv2.1 is shown in yellow pseudocolor only with the maximal diameter planes included. Scale bars: (c) 50 µm; (d) 3 µm; (e) 2 µm; (i) 200 nm; (j) 2 µm; (k) left, 1 µm, and z3, 400 nm; (l) 10 µm; (m) 200 nm; (n) 10 µm; (o) 5 µm; (p) 10 µm.

Fig 3: Microglia form direct contacts with cells in the NVU, but microglia depletion does not disrupt cerebal perfusion or metabolism. (a) 3D reconstruction of a high-resolution CLSM Z-stack shows microglia (P2Y12R, cyan) contacting both the cell body of an astrocyte (GFAP labeling, green, arrow) and astrocytic endfeet (arrowheads) ensheathing a capillary. Pericytes are visualized by anti-PDGFRb labeling (blue). (b) Microglia (CX3CR1, magenta) are able to form direct contact by their cell body with astrocytes (GFAP labeling, green) in the cerebral cortex. Scale bar, 10 µm. (c) CLSM image shows P2Y12R-positive microglial process (cyan) contacting perivascular AQP4-positive astrocyte endfeet (green) and also extends to the endothelial layer (lectin, magenta) where astrocytic coverage is not present (arrows). z1–z3 panels show the contact area on three consecutive confocal sections. Scale bar, 3 µm. (d) The process of semiautomated unbiased analysis of fluorescent intensity area for the graph presented in Fig. 1 j is depicted. White dashed lines represent the outer and the inner profiles, based on the outline of the endothelial cell. P2Y12R intensity was measured along the outer profile and TOM20 intensity along the inner profile, starting from the arrow. The intensity values are plotted (right) along the perimeter of the vessel. Contact site (marked by the gray column in the plots) was defined automatically. Scale bar, 2 µm. (e) CLSM image and fluorescent intensity plots show microglial process extending beyond perivascular astrocytic endfeet to interact with the endothelium. The fluorescent intensity profile plot (measured along the 3.5-µm long white arrow) clearly shows the presence of the microglial process under the astrocytic endfeet. (f) CLSM image and fluorescent intensity plots show microglial processes interacting with GFAP- and AQP4-positive astrocytes in the human brain. The fluorescent intensity profile plot (measured along the 15-µm-long white arrow) clearly shows the presence of the microglial process between the endothelium and the endfeet of perivascular astrocytes. (g) Schematic summary of contacts formed between microglia (m; cyan) and different cell types of the NVU. Neurons (n; red), astrocytes (ast.; green), pericytes (p; purple), endothelial cells (e; pale crimson), and vascular smooth muscle cells (s.m.; dark crimson) are shown. (h) Characterisation of CX3CR1tdTomato mice. Parenchymal tdTomato-positive cells coexpress Iba1 and P2Y12R in the cerebral cortex. Cell nuclei stained with DAPI appear in yellow pseudocolor in the merged middle image. Scale bars, 25 µm (left), 10 µm (middle). (i) Feeding C57BL/6J mice with a diet containing PLX5622 results in an almost complete (97%) elimination of resident microglia as evidenced by the numbers of P2Y12R-positive cells in the cerebral cortex. Scale bar, 100 µm. n = 12 control and n = 9 depleted mice per group; ****, P < 0.0001 control versus depleted, unpaired t test with Welch’s correction. (j and k) HMPAO-SPECT and FDG-PET images of control and microglia-depleted mice. Proportion of measured and injected HMPAO activity (Inj. Act.) and standard uptake values (SUVs) of FDG are shown. Atlas-based ROI analysis (j) shows no significant differences between the normalized regional uptake values (k) of the two groups. n = 5 and 5 mice. Data are expressed as mean ± SEM.

Fig 4: Blood vessel localization of AQP4X, AQP4, and α-syntrophin is altered in AQP4x mutants. (A) Depiction of astrocyte endfoot contacting the blood vessel wall (CD31). α-syntrophin is anchored to AQP4 via the C-terminus of AQP4x in the endfoot. Quantification of endfoot localization was performed along a line extending from the vessel luminal wall out at least 4.5 μm into parenchymal space. (B, E, H) Loess curve fit of fluorescent intensities of AQP4x (B), AQP4 (E) α-syntrophin (H), and along a line from the vessel luminal wall toward the parenchyma as diagrammed in (A) for each genotype shown. Shading depicts the edges of the 95% confidence interval. Significance testing with linear mixed models reveals a significant effect of genotype in AQP4x, α-syntrophin, and AQP4 channels (p-values = 1.862e-9, 2.049e-09, and 2.128e-10, respectively). Modeling also indicated a highly significant interaction effect of genotype and distance for the three channels AQP4x, α-syntrophin, and AQP4 (p-values = 2.2e-16, 2.2e-16, and 2.2e-16, respectively). (C, F, I) Boxplot depicting AQP4x levels shows significant changes across the genotypes in both the perivascular and parenchymal regions for AQP4x, a-syn, and AQP4, respectively. Post-hoc p-values generated from the Wilcox test are shown, with *** representing a p-value of 2.2e-16. (D, G, J) Boxplot of perivascular intensity of a given line minus parenchymal intensity of a given line. Post-hoc p-values generated from the Wilcox test highlight significant changes in localization, with *** representing a p-value of 2.2e-16. Color code as in Figure 1.

Fig 5: Generation of an obligate AQP4 readthrough line and characterization of AQP4x production. (A) CRISPR/Cas9 construction of the AllX mutant mouse line. (B) Reference and founder sequence showing conversion of the TGA stop codon to a TGG (encoding tryptophan). (C) Litter distribution among the genotypes follows the Mendelian ratio. n (litters) = 13 Aqp4 (KO) line, 36 AllX line, and 32 NoX line. (D) Weight distribution at P30 between AllX and WT mice, n = 10 WT, 9 AllXHom, 22 AllXHet. (E) Western blot showing AQP4 and AQP4x expression, n = 3 per genotype, 1 global AQP4 KO mouse as a specificity control for the antibody. Total brain lysates probed with anti-AQP4 and anti-AQP4x antibodies show a significant increase in AQP4x compared to AQP4 in the AllX variants. AQP4 expression is enhanced in the no-readthrough variants and decreases in the AllX variants. The anti-AQP4 antibody exhibits two non-specific bands at 40 and 50 kDa, as evidenced by their presence in the global AQP4 knockouts (asterisk). (F) Quantification of Western blot as box plots (log10 scale) of AQP4 and AQP4x total and the relative ratio of AQP4x to AQP4 quantifying Western blot using ImageJ, normalized to 50 kDa band and wildtype. WT AllX and WT NoX merged and analyzed using the Wilcox test. (G) BN-PAGE with anti-AQP4 antibody showing orthogonal array particles and tetramer expression. (H) Quantification of orthogonal array particle signal from (G). n = 3–4 mice per genotype. One-way ANOVA, with post-hoc P-values from Dunnett's test.