Cx43 carboxyl terminal domain determines AQP4 and Cx30 endfoot organization and blood brain barrier permeability

The neurovascular unit (NVU) consists of cells intrinsic to the vessel wall, the endothelial cells and pericytes, and astrocyte endfeet that surround the vessel but are separated from it by basement membrane. Endothelial cells are primarily responsible for creating and maintaining blood–brain-barrier (BBB) tightness, but astrocytes contribute to the barrier through paracrine signaling to the endothelial cells and by forming the glia limitans. Gap junctions (GJs) between astrocyte endfeet are composed of connexin 43 (Cx43) and Cx30, which form plaques between cells. GJ plaques formed of Cx43 do not diffuse laterally in the plasma membrane and thus potentially provide stable organizational features to the endfoot domain, whereas GJ plaques formed of other connexins and of Cx43 lacking a large portion of its cytoplasmic carboxyl terminus are quite mobile. In order to examine the organizational features that immobile GJs impose on the endfoot, we have used super-resolution confocal microscopy to map number and sizes of GJ plaques and aquaporin (AQP)-4 channel clusters in the perivascular endfeet of mice in which astrocyte GJs (Cx30, Cx43) were deleted or the carboxyl terminus of Cx43 was truncated. To determine if BBB integrity was compromised in these transgenic mice, we conducted perfusion studies under elevated hydrostatic pressure using horseradish peroxide as a molecular probe enabling detection of micro-hemorrhages in brain sections. These studies revealed that microhemorrhages were more numerous in mice lacking Cx43 or its carboxyl terminus. In perivascular domains of cerebral vessels, we found that density of Cx43 GJs was higher in the truncation mutant, while GJ size was smaller. Density of perivascular particles formed by AQP4 and its extended isoform AQP4ex was inversely related to the presence of full length Cx43, whereas the ratio of sizes of the particles of the AQP4ex isoform to total AQP4 was directly related to the presence of full length Cx43. Confocal analysis showed that Cx43 and Cx30 were substantially colocalized in astrocyte domains near vasculature of truncation mutant mice. These results showing altered distribution of some astrocyte nexus components (AQP4 and Cx30) in Cx43 null mice and in a truncation mutant, together with leakier cerebral vasculature, support the hypothesis that localization and mobility of gap junction proteins and their binding partners influences organization of astrocyte endfeet which in turn impacts BBB integrity of the NVU.


Results
Distribution and expression of Cx43 and Cx30 in the different transgenic mice. In this study we have examined brains from 4 transgenic mouse lines in which astrocyte gap junction expression has been altered, as illustrated in Fig. 1a. Immunostaining sections with the endothelial cell marker PECAM 1 (CD31) revealed that both Cx43 (Fig. 1b) and Cx30 (Fig. 1c) are expressed throughout the brain, with Cx43 being particularly abundant in proximity to brain vasculature. As considered in more detail below, perivascular location of these connexins overlapped to some extent.
Leakiness of the blood brain barrier in mice with astrocytes lacking Cx43 or its carboxyl terminal domain. Previous studies reported the observation that the volume of tissue damage in the middle cerebral artery occlusion (MCAO) ischemic stroke model was greater in the Cx43 (Δ/−) mouse 13 and that in the dKO mouse brain high pressure perfusion (HPP) led to greater vascular leakage of HRP than in the wildtype mouse 8 . However, the extent of the difference was not quantified. To explore the relation between Cx43 expression and blood brain barrier permeability in a greater detail, we quantified the impact of both Cx43 truncation and deletion using the HPP model. As illustrated in Fig. 2a, microhemorrhages were readily detectable in brains of mice of all genotypes after high pressure perfusion. Whereas density of infarcted regions did not differ between Cx43 (+/+) /Cx30 (−/−) and Cx43 (+/−) brains (as well as between Cx43 (+/+) /Cx30 (+/+) : Suppl. Fig. S1), the number of micro-hemorrhages per area was higher for Cx43 (Δ/−) mutants compared to Cx43 (+/+) /Cx30 (−/−) and to Cx43 (+/−) mice (Fig. 2b). With regard to the sizes of infarcts, areas were larger in the Cx43 (+/−) /Cx30 (+/+) and lowest in dKO brains (areas ~ 2000 to 4000 μm 2 , Fig. 2c), corresponding to radii ranging from about 25-35 μm (shapes appeared roughly spherical). These findings indicate that in the absence of Cx43 or with Cx43 lacking Scientific Reports | (2021) 11:24334 | https://doi.org/10.1038/s41598-021-03694-x www.nature.com/scientificreports/ its carboxyl terminal domain, the blood brain barrier is more vulnerable to damage, although infarct sizes are smaller than their controls.
Structured illumination microscopy (SIM) imaging of Cx43, AQP4 and AQP4ex in the perivascular domain. In order to compare resolution of conventional confocal images with those obtained using SIM, we focused on regions of the vessel wall at increasing magnification (Suppl. Fig. S2). In initial experiments, we immunostained for Cx43 in transgenic mice in which astrocytes were visualizable due to expression of yellow fluorescent protein (YFP) driven by GFAP promoter (mGFAP-Cre/RCE:loxP; RCE:loxP reporter mice bred with the mGFAP-Cre line) enabling us to distinguish astrocyte endfeet from cell bodies and from other components of the vascular wall (Suppl. Fig. S2A); resolution of Cx43 in the vessel wall was lost when the widefield image was zoomed to very high magnification (Suppl. Fig. S2B). When the widefield image in A was deconvolved and displayed at higher resolution, localization of Cx43 to astrocyte endfeet became apparent (Suppl. Fig. S2C). When Image in B was deconvolved using neighboring z-plane acquisition, a single gap junction plaque became resolvable (Suppl. Fig. S2D). SIM reconstruction of the image in D revealed further details of irregular topology, with subregions that were not visible without SIM in D (Suppl. Fig. S2E). Similarly, AQP4 puncta around vessels could be resolved in single plane widefield deconvolved images (Suppl. Figs. S2f, g), where sub-200 nm details were   Fig. S2h). The extended AQPex isoforms were also resolvable as small aggregates using isoform specific antibodies described in Methods (Suppl. Figs S2i-k). At highest magnification, 3D SIM revealed stripe-like patterns of particles (Suppl. Fig. S2k). As discussed below, AQP4ex isoforms have recently been shown to be located primarily in astrocyte endfeet, anchoring other isoforms 17 .
Distribution of Cx43, AQP4 and AQP4ex in endfeet of transgenic mice with Cx43 mutations in astrocytes. SIM images of immunolabeling with Cx43 and AQP4 antibodies revealed the presence of both proteins in endfeet of each genotype, except that Cx43 was not detected in astrocytes in GFAPCre::Cx43 fl/fl mice (Fig. 3). Figure 3a shows example images that correspond with data displayed in Figs. 4 and 5, while Fig. 3b shows example images for data in Fig. 6. Although at this magnification differences among genotypes at the level of perivascular domains are not obvious (Fig. 3a,b), 3D SIM analyses of Cx43 and AQP4 particle sizes revealed differences as described below.
Sizes of particles were similar in all genotypes (Fig. 6c). (d) Ratio of AQP4ex to total AQP4. In order to determine whether AQP4ex was preferentially localized in the perivascular region, we calculated the ratios of AQP4ex/total AQP4 in perivascular regions of each mutant. Our findings indicated that the ratio of AQP4ex/total AQP4 densities was highest in the Cx43 (−/−) compared to the other genotypes (Fig. 7a). In contrast, the relative size of particles was negatively correlated with Cx43 expression, with Cx43 (+/+) and Cx43 (+/−) having the lowest AQP4ex/AQP4 particle size ratios and dKO and Cx43 (Δ/−) having the highest ratios (Fig. 7b). have shown in previous studies that Cx43 truncation leads to destabilization of GJ plaques in astrocytes 6 , and we hypothesized that this would lead to remodeling of the endfoot domain with respect to other connexin proteins involved in function of the neurovascular unit. To test this hypothesis, we performed confocal microscopy on vascular domains in cortex. Figure 8a shows representative fluorescence micrographs with dual labeling for Cx43 and Cx30. When overlap was quantified for the Cx43 (+/−) brain, we found that Mander's coefficients for overlap of both Cx43 and Cx30 with total staining were about 0.2-0.35, while in the case of the truncation mutant (Cx43 (Δ/−) ) the coefficients were significantly higher (> 0.6) (Fig. 8b). Note that the Mander's coefficients for Cx43 and Cx30 overlap quantified for the wild-type (WT: Cx43 (+/+) /Cx30 (+/+) ) brain (Suppl. Fig. S3) was not significantly different from those recorded for the Cx43 (+/−) (WT m1 = 0.35 ± 0.05, n = 6; Cx43 (+/−) m1 = 0.24 ± 0.1, n = 4, t-test p = 0.27, and WT m2 = 0.38 ± 0.04, n = 7; Cx43 (+/−) m2 = 0.28 ± 0.07, n = 4, t-test p = 0.19). This finding indicates that the mobile Cx43 isoform is more highly interactive, consistent with our finding that the mobile mutants can readily penetrate GJ plaques 7 .

Discussion
In this study we have compared the perivascular organization of gap junctions and AQP4, the channels that are primarily responsible for water distribution in the astrocyte network, using transgenic mouse strains in which Cx43 abundance is reduced or its cytoplasmic binding domain was deleted. This perivascular structure, or glia limitans, is formed by astrocyte endfoot processes closely enwrapping the vessel wall. Both AQP4 and Cx43 are localized to astrocyte endfeet soon after birth (by P0 in mice), whereas Cx30 appears later (at about P12 in mouse) 8,18,19 . The four mouse genotypes examined here include those in which Cx43 expression is normal but Cx30 is deleted (Cx43 (+/+) /Cx30 (−/−) ), Cx43 (+/−) or Cx43 (Δ/−) heterozygotes, where one allele is absent and the other is either WT Cx43 or one in which the coding region is exchanged with a truncated isoform (termed Cx43 (Δ/−) mice), and mice in which Cx30 is deleted and Cx43 ablation is targeted to astrocytes (Cx30 (−/−) /GFAPCre::Cx43 fl/fl , also termed dKO). Upon examination using endothelial CD31 immunostaining to localize distribution of Cx30, Cx43 and AQP4 within endfeet, we found that all vessels were macroscopically similar ( Fig. 1), and immunolabeling was grossly similar except for absence of Cx43 in the GFAP-Cre::Cx43 fl/fl endfeet. Even higher magnification using 3D structured illumination microscopy (SIM) revealed no striking disorganization of the vessels or the endfeet (Fig. 3). This lack of gross alteration in vessel morphology is consistent with reports from others that brains were www.nature.com/scientificreports/ anatomically normal, with no gross difference in vascular morphology in the dKO (Cx30/GFAP::Cx43 fl/fl ) 8,9 or in the Cx43 (Δ/−) mouse cortex 13 . Despite grossly normal vascular morphology, previous electron microscopy revealed that astrocyte endfeet were edematous in the dKO 8,9 ; moreover, oligodendrocytes were found to be vacuolated, which was interpreted as reflecting disruption of the gap junction mediated connections between astrocytes and oligodendrocytes 9 , the socalled panglial syncytium 20 . In our examination of HRP extravasation in the high-pressure perfusion model, we found that both Cx43 deletion and truncation increased vulnerability of the neurovascular unit to similar extent (Fig. 2), with number of micro-hemorrhages approximately double that in the respective controls. Sizes of most hemorrhages were similar, except that the Cx43 (+/−) brains showed somewhat larger areas of extravasation than the truncation brain and the infarct size in Cx43 (+/+) was larger than in the dKO. Therefore, these studies reveal that incidence of stroke vulnerability inversely correlates with Cx43 abundance and intact carboxyl terminal binding sites, conditions under which mobility and distribution of Cx43 and other nexus proteins are altered 7 .
The initial astrocyte-targeted Cx43 knock-out mouse 21 displayed larger volume of injury after middle cerebral artery occlusion 22 , accelerated spreading depression, and motor impairments 21,23 . The residual coupling among astrocytes was abolished after crosses with a mouse deficient in the other major astrocyte connexin, Cx30 24 ; while this has been taken as evidence that only Cx43 and Cx30 contribute substantially to astrocyte intercellular coupling, Cx26 is also normally expressed in at least in some astrocytes 25 and is syntenic with Cx30, so that Cx30 null mice also lack Cx26 expression 26 . The Cx30/GFAPCre::Cx43 fl/fl mice (dKO) have depressed K + buffering capacity and lowered threshold for epileptiform events 24 , and they display more severe pathology when subjected to ischemia/reperfusion injury paradigms 27 . Together, these studies provide extensive evidence that astrocyte connexins play an important role in maintaining ion homeostasis and minimizing damage in the CNS.
The Cx43 truncation mouse was generated in order to evaluate the role of the Cx43 carboxyl terminal domain in systems physiology. The surprising phenotype included pathologically elevated epidermal permeability, which has been attributed to loss of interaction with the tight junction protein ZO-1 10 . Studies on cardiac tissue from these mice have revealed higher propensity for ventricular arrhythmias during reperfusion and increased size and reduced number of GJ plaques in the myocardium (intercalated discs) 12 . This result is the opposite of our finding of increased number and reduced GJ plaque size in astrocyte endfeet (Fig. 4). This may call into question  Figure 7. Ratio of AQP4ex/AQP4 density and size in perivascular brain regions of the Cx43 genotypes. (a) Ratios of particle densities of AQP4ex to AQP4 shows that dKO has higher ratio than the other genotypes.
(b) Ratios of particle sizes shows that Cx43 (+/+) and Cx43 (+/−) are similar but the size of AQP4 is smaller than AQP4ex for the other groups. Data are presented as mean ± SEM (ANOVA followed by Tukey's post hoc test: *p < 0.05 and **p < 0.01).  28 . Studies of the consequences of loss of the Cx43 carboxyl terminus in stroke reported that occlusion of the left coronary artery followed by 4 h reperfusion led to more severe damage in Cx43 (Δ/−) than in the WT heterozygote hearts 14 . In the brain, vessel permeation of India ink was similar in both Cx43 (+/+) and Cx43 (Δ/−) heterozygotes, but larger infarcts were seen in Cx43 (Δ/−) mouse brains when the middle cerebral artery occlusion model was performed 13 . Although previous studies have also shown that blood brain barrier vulnerability is higher in mice lacking astrocyte Cx43 or with the truncated construct, and that perivascular endfeet in these brains appear edematous, the organization of Cx43 in the astrocyte endfeet has not been examined. To determine more precisely how the expression of Cx43 and presence of its carboxyl terminal binding sites impact organization of proteins in the astrocyte endfoot we used 3D structured illumination microscopy (SIM) to achieve moderately high resolution of perivascular protein complexes in brain sections from these mice. Gap junctions can be most unambiguously detected in freeze fracture electron micrographs, where plaques containing even a small number of intramembrane particles can be identified, each of which is believed to be a gap junction channel, see 29 . Histograms of particle sizes of Cx43 GJ plaques detected in our studies show minimal detectable size of about 8000 nm 2 , corresponding to a circular plaque with ~ 50 nm radius that would contain ~ 30 GJ channels within the plaque if interparticle spacing is ~ 14 nm. When genotypes were compared, we found that density of perivascular Cx43 plaques was lowest in WT and Cx43 (+/−) mouse brains (~ 6 compared to ~ 8/10 µm 2 in the truncation mouse), whereas sizes were smaller in the Cx43 (Δ/−) (~ 0.23 vs ~ 0.25 µm diameter) than in the other genotypes, so that there was little difference in total Cx43 at the endfoot (~ 2.8 to 3.2% coverage of total perivascular area).
Like gap junctions, AQP4 forms supramolecular clusters decorating astrocyte plasma membranes. These so-called Orthogonal Arrays of Particles (OAPs) are formed by assemblies of AQP4 tetramers highly concentrated at the astrocyte endfeet. In freeze fracture studies, the lattice corresponds to 6.4 nm center to center spacing [30][31][32] . Using antibodies recognizing all AQP4 isoforms or only AQP4ex, we achieved resolution of particle www.nature.com/scientificreports/ size comparable to that with Cx43, with smallest bin size of 8000 nm 2 in the resulting histograms, similar to that reported by 33 , corresponding to < 500 square arrays. We found that in the perivascular endfeet, AQP4 particle density was directly proportional to Cx43 expression level (Cx43 (+/+) > Cx43 (+/−) > Cx43 (Δ/−) > dKO). Particle sizes were slightly higher for Cx43 (+/−) heterozygotes and lowest for dKO. Coverage of the endfoot was considerably higher for Cx43 (+/+) (~ 7.5%) than for the others and was lowest in the dKO (~ 3.2%). Compared to Cx43 plaques, AQP4 particles were about 50% more abundant but similar in size. AQP4 is distinctive among aquaporin water channels in its expression as several distinct isoforms through two mechanisms that modulate protein translation: Leaky scanning, in which a weak initiation codon may be skipped, generating alternate M1 and M23 AQP4 isoforms 31 , and translational read-through, in which translation continues beyond a stop codon, resulting in longer N-termini, termed AQP4ex to denote the extension 17,34 . Using antibodies recognizing only the AQP4ex isoforms, we determined distribution of particles in the perivascular region and were able to compare to overall AQP4. As in the Cx43 plaque and AQP4 OAP studies, we achieved a resolution limit of 8000 nm 2 . AQP4ex density was found to be higher in Cx43 (+/+) and Cx43 (+/−) than in Cx43 (Δ/−) and the dKO, whereas particle sizes were similar in each group, so that endfoot coverage was slightly lower in the dKO than in WT (ranging from about 3.5-4.5%). Sizes of particles formed of AQP4ex were previously reported to be larger than for total AQP4 35 , which was confirmed in our measurements in dKO and Cx43 (Δ/−) but not in the Cx43 (+/+) or Cx43 (+/−) groups.
Orthogonal particle arrays have long been known to be present in astrocyte membranes 36 and have been used for identification of astrocytes in freeze fracture replicas. AQP4 isoforms comprise these OAPs and show profound polarization in expression at the endfeet [37][38][39] . The polarized distribution has been attributed to the presence of scaffolding molecules to which AQP4 binds with high affinity, primarily dystrophin/syntrophin (see 40 ), and one role of the readthrough extension of AQP4ex may be to anchor this isoform in the endfoot through interactions with syntrophin and other scaffolds 17 . Recent studies using mice in which AQP4ex or syntrophin isoforms were deleted have revealed that AQPex expression appears to reciprocally localize α-syntophin at the endfoot and that deletion of both syntrophin isoforms leads to loss of AQP4 at the endfoot 17,41 . An additional finding from the study on syntrophin deletion was that endfoot AQP4 OAPs were smaller, while Cx43 plaques increased in size 41 . In addition to preferential adhesion to scaffolding proteins as a mechanism that enhances endfoot expression of AQP4 and especially the AQP4ext isoform, it has recently been reported that local translation of certain astrocyte proteins occurs in endfeet 42 . Regulatory proteins that direct translation of individual AQP4 isoforms have been discovered (see 43 ), raising the possibility that local translation may bias expression in specific cellular compartments. Combined with selective retention by cytoskeleton-linked scaffolds, this could lead to both stable and dynamic polarization of the endfoot to respond to local demands for water flux.
The perivascular endfoot gap junction plaques include both Cx30 and Cx43, as evidenced by cryo-immunogold electron microscopy 44 . Although FRAP studies indicate that these connexins can intermix in transfected cells 7 , there are no reports indicating whether these channels are segregated or intermixed in vivo. To test this hypothesis, we used super-resolution microscopy to determine in brain tissue whether Cx30 and Cx43 are co-localized within plaques in astrocyte endfeet, rather than abundance, because in the cerebellum Cx30 is reportedly higher than in cortex or hippocampus 45 and in thalamus Cx43 expression is rather low 46 . Analysis of colocalization of Cx30 with truncated Cx43 (Cx43 (Δ/−) ) in perivascular domains in these mice revealed much higher colocalization than in the Cx43 (+/−) controls. This result is consistent with our photobleach recovery experiments, which showed that Cx30 is much more penetrant into Cx43 (Δ/−) GJ plaques than into Cx43 (+/+) GJs 7 .
Together, AQP4 and GJ channels connect astrocytes into a water-equilibrating network, where the GJ-connected glia extend from arteriolar astrocyte endfeet that take up water, glucose and other solutes, across brain parenchyma, where water balance is maintained and energy substrates are delivered, to venous perivascular astrocyte endfeet where efflux of excess water and metabolites occurs into the circulation. Location, expression and function of AQP4 and GJs are interdependent, as indicated by our initial studies of AQP4 siRNA effects on Cx43 in astrocytes 47 and confirmed by more recent examination of transgenic mice in which either AQP4 4,41,48 or Cx43 49 is manipulated. Such interdependence likely provides a major mechanism by which water and solute homeostasis is achieved, and disruption of normal distribution of these two proteins impacts brain edema and BBB permeability 50 .
Interestingly, OAP alterations have been observed in different pathological conditions such as brain tumors, ischemia and Alzheimer's disease [51][52][53] . Moreover, the interest in OAP studies has recently increased since the discovery of their involvement in the pathogenesis of Neuromyelitis Optica (NMO), a CNS autoimmune channelopathy whose autoantibodies are only able to attack their antigen AQP4 if arranged in OAPs [54][55][56] . Astrocyte endfeet edema has been observed in association with BBB leakage in the dystrophic MDX mouse model 57 during hypoxia 58 or stroke 59 . Thus, in dKO mice, it may directly affect the endothelial barrier properties as well. In particular, the regulations taking place in astroglial endfeet inducing BBB phenotype might be severely compromised.
The permeability barrier between blood and cerebrospinal fluid is established by tight junctions in endothelial cells and maintained by paracrine factors released by astrocyte endfeet 60 . Whether the reported increase in BBB leakage is associated with alteration in endothelial cell tight junctions, as suggested by a Reviewer, could not be resolved at the moment. Supplementary data showing the analysis of the SIM images of the endothelial tight junction protein claudin-5 in the four genotypes indicated that only the size but not the density of claudin-5 particles was smaller in the dKO compared to Cx43 (+/−) and to Cx43 (+/+) /Cx30 (−/−) (Suppl. Fig. S4). However, no significant difference in terms of claudin-5 particle size was detected between Cx43 (Δ/−) and the other genotypes (Suppl. Fig. S4). Given that both dKO and Cx43 (Δ/−) mice showed increased BBB leakage compared to their controls (Fig. 2), and that only dKO displayed smaller claudin-5 particles, we could not establish a correlation between these events. Additional studies are necessary to resolve the extent to which Cx43 carboxyl terminus affects the expression and organization of vascular tight junctions. Cells intrinsic to the brain vasculature (endothelial Scientific Reports | (2021) 11:24334 | https://doi.org/10.1038/s41598-021-03694-x www.nature.com/scientificreports/ cells, smooth muscle and pericytes) express gap junctions and vascular properties might be affected by global deletion of Cx30 or knock-in of the truncated deletion. Although less of an issue in endothelial cells, which in brain express little or no Cx43 or Cx30 61,62 , some pericytes express these connexins 63 and manipulations in transgenic mice could potentially affect vascular permeability through action on these cells. However, we found that organizational properties of the Cx43 null astrocyte endfeet and BBB permeability were similar with and without Cx30, indicating that glial Cx43 is responsible for the reported alterations.
In conclusion, these functional and super-resolution imaging studies of perivascular regions of transgenic mice support the hypothesis that Cx43 and its carboxyl terminal domain contribute to structural organization of the astrocyte endfoot. In the absence of Cx43, or when its carboxyl terminus is lacking, the blood brain barrier is more fragile and sizes and distribution of AQP4 and its extended isoform are altered. Thus, as in cultured cells where exogenous expression of fluorescent Cx43 isoforms has revealed impact of its carboxyl terminus on lateral mobility and interaction with other proteins 7 , immobile Cx43 in gap junction plaques between endfeet appears to provide a framework that contributes to establishing and maintaining polarity of astrocyte processes. BBB permeability assay. For in vivo brain high hydrostatic pressure perfusion (HPP), we used a solution which included albumin as an agent to increase viscosity/hydrostatic pressure [ 8 ; method derived from 64 and modified to include elevated hydrostatic pressure 65 ]. In brief, mice were anesthetized with a cocktail of xylazine (20 mg/ kg)-ketamine (100 mg/kg); deeply anesthetized animals were intraperitoneally (i.p.) injected with norepinephrine (NE: 20 µg/kg) and 5 min later, the thoracic cavity opened to expose the heart for perfusion. The right ventricle was severed and a horseradish peroxidase (HRP: 3.5 mg/ml) Krebs bicarbonate buffered solution containing NE (2 µg/kg) and serum albumin (40 mg/ml) was perfused for 2 min into the left ventricle using a syringe pump at a rate of 2.5 ml/min. This was followed by a 2 min perfusion with HRP-Krebs bicarbonate buffered solution (albumin-and NE-free) and then by a 4 min perfusion with 4% paraformaldehyde solution. Brains were removed and post-fixed for 1 h in 4% PFA prior to cryopreservation in serial sucrose (10-30%) solutions. OCT-embedded brains from five Cx43 (+/+) /Cx30 (−/−) , five dKO, six Cx43 (Δ/−) and seven Cx43 (+/−) mice were cryosectioned for HRP staining (using DAB) and four of these used for Cx43, Cx30, AQP4 and AQPex immunohistochemistry.

Methods
Peroxidase staining and image analysis. To localize peroxidase activity in HRP perfused brains, the Sigma Fast DAB (3,3′-diaminobenzidine tetrahydrochloride; Sigma-Aldrich) reagents were used according to the manufacturer's instructions. Briefly, one DAB tablet and one urea hydrogen peroxide tablet were dissolved in 14.5 ml ultrapure water and 0.5 ml enhancer (0.3% CoCl 2 ) solution added to the DAB reaction solution. Slides containing brain sections were incubated in DAB reaction solution for several min until color developed. Reaction was stopped by washing the slides in PBS which were then mounted in mounting medium for analysis. Bright-field images of 5 Cx43 (+/+) /Cx30 (−/−) KO, 5 dKO, 6 Cx43 (+/−) and 7 Cx43 (Δ/−) brains stained with DAB were acquired using an inverted Nikon microscope equipped with 4 × objective and Metafluor software. Brain section images were evaluated for possible differences in the number, size and density of micro-hemorrhages (DAB positive areas) among the genotypes. For each mouse, quantification was performed using ImageJ software 66,67 to measure area infiltrated and density of microhemorrhages in DAB-reacted sections on a complete set of sections throughout a brain hemisphere.
Imaging acquisition. For confocal microscopy, a Leica inverted DMi8 SP8 microscope and Leica LASX software were used for image acquisition and analysis (Leica Microsystems CMS GmbH). A 63X NA = 1.4 Oil PLAPO, WD = 0.14 mm WHITE objective immersion liquid with a refractive index of 1.5 was used. All confocal images were collected using 594 and 488 nm lasers for excitation and a pinhole diameter of 1 Airy unit. A minimum of four ROIs were acquired per slice and processed for analysis using Image-J.

Structured Illumination imaging (SIM).
For SIM, images of Cx30, Cx43, and AQP4 and its extended AQP4ex isoforms in immunostained brain sections were acquired from each genotype and analyzed using a Nikon ECLIPSE Ti-E microscope and Nikon Elements software, equipped with a oil-immersion Nikon 100X widefield Objective with a refractive index of 1.5, and a Zeiss Elyra S1 with 63X NA1.4 Objective, with 5 grating positions and 40 nm pixels on an EMCCD camera were used. Excitation of the Alexa Fluor488 and Alexa Fluor594 dyes was achieved with 488 and 561 nm laser beams, and the emitted fluorescence was collected through GFP and RFP emission filters, respectively, using an EMCCD camera (Andor iXon3 DU897).
Image analysis. To assess the expression of Cx43, AQP4 and AQP4ex particles localized along blood vessels in the brain cortex, SIM images were acquired from perivascular regions selected randomly within each section. The number and sizes of the particles were determined from mice of each genotype using a plugin ("Analyze Particles") of FIJI-ImageJ. For that, SIM stacked images were converted to monochrome (8-bit) images, thresholded and particles outlined. The number and size (µm 2 ) of outlined Cx43, AQP4 particles obtained for each set of images were then exported and statistically analyzed in GraphPad Prism software.
As is illustrated In Suppl. Fig. S5, log-transformed values of particle sizes were normally distributed; means of log-transformed data or geometric means of non-transformed data sets were used to compare particle sizes and relative density (particles/µm 2 ).
Colocalization of membrane proteins. To evaluate the degree of co-localization of Cx43 and Cx30 puncta at the perivascular domains, brain sections of Cx43 (+/−) and Cx43 (Δ/−) mice were immunostained with anti-Cx43 and anti-Cx30 antibodies. Z-stack images were acquired using a Leica inverted DMi8 SP8 microscope confocal microscope and images analyzed using a plugin ("Co-localization Threshold") of Fiji-ImageJ. For that, regions of interest (ROIs) were placed on each Cx43 immunofluorescent spot located along a blood vessel after splitting the two merged channels (488 and 594 nm). The Mander's correlation coefficient with Costes' auto-threshold detection was used to estimate the fraction of Cx30 fluorescence (594 nm channel) that co-localized with Cx43 (488 nm channel) (m1) and the fraction of Cx43 that colocalized with Cx30 (m2) within the ROIs throughout the z-stacked images.
Statistical analyses. Statistical analyses were performed using GraphPad Prism 7.0 and 9.2 software. Results are presented as the mean ± SEM. Shapiro-Wilk normality test was performed to evaluate the data distribution. Statistical significance was evaluated using unpaired Student's t test or ANOVA followed by Tukey's post hoc with multiple statistical comparisons between groups. Differences were defined as follows: *p < 0.05, **p < 0.01, and ***p < 0.001.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.