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

Astrocytes, once viewed as passive space-filling neural support cells, are now appreciated as highly polarized and specialized cells that play a major role in dynamic control of neuronal activity. Astrocyte processes extend to each of the other neural components, creating distinct functional compartments: The interfaces between individual astrocytes and heteroglial contacts with oligodendrocytes, the cradle around the synapse, and the endfoot surrounding endothelial cells in the neurovascular unit. A remarkable feature of astrocytes is that at each of these regions, gap junctions are present and produce a pan-glial syncytium enabling long range signaling within anatomically defined areas; gap junctions in astrocytes provide heterocellular, heterotypic gap junctions with oligodendrocytes, the formation of both reflexive (or “autaptic”) junctions with itself (see¹) and homotypic junctions with other astrocytes². Gap junctions consist of core channel proteins, the connexins (primarily Cx43 and Cx30 in astrocytes), to which are attached scaffolding, signaling and cytoskeletal proteins, assembling into what we have termed the Nexus³. Intercellular gap junction channels are the primary mechanism by which astrocytes communicate among themselves and with oligodendrocytes, and the other Nexus components likely act to modulate intercellular communication and provide the additional functions proposed for gap junctions, including adhesion, cell migration/process motility, and maintenance of both gap and tight junctions in neighboring cells.

The astrocyte endfoot serves as the gateway for delivery of water, ions and metabolites into the brain parenchyma, and location of channels and transporters for this function is highly polarized to face the perivascular space. AQP4 at the perivascular endfoot and intercellular gap junction channels connect astrocytes into a water distribution network that extends from the perivascular space throughout the parenchyma. This flux drives bulk flow of water, ions and small molecules, in a process termed glymphatic circulation that is driven by cardiovascular pulsation and is crucial to delivery of nutrients and removal of debris. Water uptake and delivery throughout brain parenchyma are highly interdependent, as evidenced by enhanced gap junction coupling resulting from AQP4 deletion⁴. Our studies using fluorescence recovery after photobleaching (FRAP) to determine mobilities of gap junction plaques with respect to other membrane proteins have revealed that gap junctions formed of Cx43 are ordinarily highly stable, whereas paired Cx30 connexons rearrange rapidly⁵. Truncation of the cytoplasmic carboxyl terminus or mutagenesis of cytoplasmic cysteine residues converts Cx43 behavior to that of mobile Cx30⁶. Stability of the Cx43 plaque was also shown to dictate distribution and mobility of other membrane proteins, lipids and cytoplasmic Nexus components⁷, leading to the hypothesis that Cx43 plaques serve as structural determinants of endfoot organization.

Consistent with this hypothesis, transgenic mice lacking astrocyte connexins display numerous abnormalities, including leaky blood–brain-barrier and edematous astrocyte endfeet with reduced AQP4 and beta-dystroglycan expression⁸, and abnormal white matter with vacuolated oligodendrocytes⁹. Moreover, mice have been generated in which full-length Cx43 has been replaced with Cx43 mutant truncated at amino acid 258, immediately upstream of the region we have implicated in conferring junction stability. Homozygous Cx43(K258stop) mice are not viable due to severe edema, attributed to loss of the epithelial junctional barrier¹⁰; mice heterozygous for truncated Cx43 (Cx43^K258/−) show defective migration of neuronal precursors¹¹, altered intercalated disc structure and localization¹² and predisposition to cardiac arrhythmias and larger infarct size after cardiac or cerebral ischemia^13,14. It is noteworthy that the location of this Cx43 mutation coincides with a frame-shift mutation responsible for occulodentodigital dysplasia (ODDD), a syndrome with major neurological impairment^15,16.

These prior studies indicate that astrocyte endfeet and the gap junctions that connect them are essential for the proper molecular organization of the neurovascular unit and for its barrier function. To test the extent to which the immobile Cx43 gap junction plaques impose structural order that enables barrier formation, we have used a model of high-pressure perfusion to evaluate whether transgenic mice with altered connexin expression or not possessing the Cx43 carboxyl terminus scaffolding-binding domain exhibit differential susceptibility of brain vasculature to rupture and have used super-resolution microscopy to examine organization of endfoot proteins. Our findings provide evidence that presence of Cx43 and its carboxyl cytoplasmic binding domain are important determinants of structural aspects of AQP4 aggregates and Cx43 gap junction plaques. We propose that through its linkage to scaffolding and other Nexus components, cytoskeletal elements and endfoot markers, Cx43 provides stability and overall structural organization for the astrocyte endfoot.

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 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 cortex¹³.

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 oligodendrocytes⁹, the so-called panglial syncytium²⁰. 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 altered⁷.

The initial astrocyte-targeted Cx43 knock-out mouse²¹ displayed larger volume of injury after middle cerebral artery occlusion²², 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, Cx30²⁴; 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 astrocytes²⁵ and is syntenic with Cx30, so that Cx30 null mice also lack Cx26 expression²⁶. The Cx30/GFAPCre::Cx43^fl/fl mice (dKO) have depressed K⁺ buffering capacity and lowered threshold for epileptiform events²⁴, and they display more severe pathology when subjected to ischemia/reperfusion injury paradigms²⁷. 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-1¹⁰. 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 the widely accepted role of the carboxyl terminal PDZ binding domain in corralling Cx43, delimiting plaque size²⁸. 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 hearts¹⁴. 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 performed¹³.

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, see²⁹. Histograms of particle sizes of Cx43 GJ plaques detected in our studies show minimal detectable size of about 8000 nm², 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 µm² 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^30–32. Using antibodies recognizing all AQP4 isoforms or only AQP4ex, we achieved resolution of particle size comparable to that with Cx43, with smallest bin size of 8000 nm² in the resulting histograms, similar to that reported by³³, 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 isoforms³¹, 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 nm². 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 AQP4³⁵, 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 membranes³⁶ 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^37–39. The polarized distribution has been attributed to the presence of scaffolding molecules to which AQP4 binds with high affinity, primarily dystrophin/syntrophin (see⁴⁰), and one role of the readthrough extension of AQP4ex may be to anchor this isoform in the endfoot through interactions with syntrophin and other scaffolds¹⁷. 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 size⁴¹. 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 endfeet⁴². Regulatory proteins that direct translation of individual AQP4 isoforms have been discovered (see⁴³), 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 microscopy⁴⁴. Although FRAP studies indicate that these connexins can intermix in transfected cells⁷, 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 hippocampus⁴⁵ and in thalamus Cx43 expression is rather low⁴⁶. 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^(+/+) GJs⁷.

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 astrocytes⁴⁷ and confirmed by more recent examination of transgenic mice in which either AQP4^4,41,48 or Cx43⁴⁹ 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 permeability⁵⁰.

Interestingly, OAP alterations have been observed in different pathological conditions such as brain tumors, ischemia and Alzheimer’s disease^51–53. 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^54–56. Astrocyte endfeet edema has been observed in association with BBB leakage in the dystrophic MDX mouse model⁵⁷ during hypoxia⁵⁸ or stroke⁵⁹. 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 endfeet⁶⁰. 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 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 connexins⁶³ 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 proteins⁷, immobile Cx43 in gap junction plaques between endfeet appears to provide a framework that contributes to establishing and maintaining polarity of astrocyte processes.

Grant support: R21NS116892 and NS092466 (DCS and ES), 1S10D1218, 1S10D18218 and P30CA013330 (Einstein Analytical Imaging Facility), EU project H2020-MSCA-ITN 722053 EU-GliaPhD (CS), BMBF projects 16GW0182 CONNEXIN and 01DN20001 CONNEX (CS). We gratefully acknowledge Drs Antonio Frigeri and Grazia Paola Nicchia (University of Bari, Italy) for the gift of anti-AQP4ext antibody, the technical assistance of Jian Pan (NYMC) with cryosections, and Dr. Raddy Ramos assistance with Cx43^(Δ/−) colony establishment and breeding at the animal facility at New York Institute of Technology College of Osteopathic Medicine.

D.C.S. and E.S. conceptualize the study; K.M., G.S., C.S., P.G. helped setting up the experiments; A.C., R.S., A.T., L.M., P.G., G.S., E.S. performed experiments; A.C., R.S., E.S., D.C.S. analyzed the data; D.C.S. lead the writing in consultation with AC, RS, ES; KM, GS, CS provided critical feedback and helped writing the paper. All authors reviewed the manuscript.

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

The authors declare no competing interests.

David C. Spray, Email: David.spray@einstenmed.edu

Eliana Scemes, Email: escemes@NYMC.edu.

The online version contains supplementary material available at 10.1038/s41598-021-03694-x.