The role of pericytic laminin in blood brain barrier integrity maintenance

Laminin, a major component of the basement membrane, plays an important role in blood brain barrier regulation. At the neurovascular unit, brain endothelial cells, astrocytes, and pericytes synthesize and deposit different laminin isoforms into the basement membrane. It has been shown that laminin α4 (endothelial laminin) regulates vascular integrity at embryonic/neonatal stage, while astrocytic laminin maintains vascular integrity in adulthood. Here, we investigate the function of pericyte-derived laminin in vascular integrity. Using a conditional knockout mouse line, we report that loss of pericytic laminin leads to hydrocephalus and BBB breakdown in a small percentage (10.7%) of the mutants. Interestingly, BBB disruption always goes hand-in-hand with hydrocephalus in these mutants, and neither symptom is observed in the rest 89.3% of the mutants. Further mechanistic studies show that reduced tight junction proteins, diminished AQP4 expression, and decreased pericyte coverage are responsible for the BBB disruption. Together, these data suggest that pericyte-derived laminin is involved in the maintenance of BBB integrity and regulation of ventricular size/development.

the time course of this hydrocephalic phenotype, we analyzed these mice at two different developmental stages: Embryonic day (E) 15.5 and postnatal day (P) 2. No hydrocephalic embryos/pups were found at these time points (not shown), suggesting that this phenotype occurs at later stages. It should be noted that the hydrocephalic phenotype, unlike the muscle pathology, was not observed in all PKO mice. Statistics showed that 10.7% of the PKO mice were hydrocephalic, whereas only 0.5% of the littermate controls were hydrocephalic (Supplementary Table  1). These results suggest that PDGFRβ + cell-derived laminin is involved in the formation of hydrocephalus either directly or indirectly. In addition, survival analysis has demonstrated that the PKO-h mice usually die within 2 months after birth (Fig. 1d). Although the PKO-nh mice have a longer life span (Fig. 1d), they usually die within 4 months probably due to respiratory failure caused by congenital muscular dystrophy 26 .

PKO-h but not PKO-nh mice show BBB disruption.
To evaluate vascular integrity of the PKO mice, we first performed immunohistochemistry against IgG. Although IgG was absent in control and PKO-nh mice at 1-2 months, it was detected in the brains of age-matched PKO-h mice (Fig. 2a). Quantification showed a significantly higher level of IgG in PKO-h brains (Fig. 2b), suggesting breakdown of BBB in these mice. To further test this hypothesis, we injected FITC-Dextran intravenously and Evans blue intraperitoneally into 1-2-month-old control and mutant mice, and examined their leakage into brain parenchyma 12 hours later. Consistent with IgG data, substantially higher levels of FITC-Dextran (Fig. 2c) and Evans blue (Fig. 2d) were found in PKO-h brains, compared to the controls or PKO-nh brains, strongly indicating that the BBB integrity is compromised in PKO-h mice at 1-2 months. Furthermore, we also examined BBB integrity at E15.5 and found no leakage in PKO mice ( Supplementary Fig. 1a). Given that the barrier property of BBB is already formed by E15.5 17,29 , these results suggest that BBB barrier property formation during early development is not affected by the loss of PDGFRβ + cell-derived laminin. Consistent with the BBB permeability data, we also detected hemoglobin in 1-2-month-old PKO-h but not control or PKO-nh brains (Fig. 2e,f). Altogether, these results suggest that the vascular integrity is indeed disrupted in PKO-h mice.
To further investigate whether BBB disruption in PKO-h mice is due to the lack of pericytic laminin directly or caused by hydrocephalus indirectly, we generated an inducible knockout line by crossing the NG2-CreER TM transgenic mice with our laminin γ 1 flox/flox mice. We administered tamoxifen into these inducible mice at various developmental stages, and tried to identify and analyze animals that do not develop hydrocephalus. This experiment, however, was not successful due to the long turnover rate of laminin. We treated mice with tamoxifen at either neonatal or embryonic stages consecutively for 5 days, and laminin synthesized at early embryonic stage was still detectable two months after treatment (not shown). These data suggest that the inducible system is unable to ablate laminin function, owing to its extremely long turnover rate. Laminin expression is abrogated in PDGFRβ + cells in both PKO-h and PKO-nh mice. To determine whether the observed CNS phenotype is due to loss of laminin in PDGFRβ + cells, we first performed lineage-tracing experiments by crossing the Pdgfrβ -Cre + mice with the Ai14 reporter line, which harbors a floxed STOP sequence before reporter gene tdTomato (TdT). In the resulting Ai14:Pdgfrβ -Cre + mice, TdT co-localized with PDGFRβ expression in the brain (Fig. 3a), suggesting that PDGFRβ + cells are indeed targeted for recombination. Next, immunohistochemistry revealed strong laminin γ 1 staining in vessels from control brains (Fig. 4a). In both PKO-h and PKO-nh mice, however, laminin γ 1 expression was slightly reduced (Fig. 4a). Western blotting revealed slightly but significantly lower levels of laminin γ 1 in PKO-h and PKO-nh brains (Fig. 4b), suggesting that laminin expression is decreased in these mice. Furthermore, we isolated PDGFRβ + cells (mostly pericytes due to sequential filtering) directly from control, PKO-h, and PKO-nh brains, and examined their expression of laminin (Fig. 4c). Laminin γ 1 expression was found in PDGFRβ + cells isolated from control brains but not in those isolated from PKO-h or PKO-nh brains (Fig. 4d), suggesting that laminin γ 1 expression is abrogated in PDGFRβ + cells from both PKO-h and PKO-nh mice. In addition, we also isolated astrocytes and brain microvascular endothelial cells from these mice and examined their laminin expression. Consistent with previous reports [13][14][15][16] , astrocytes predominantly express laminin α 1 and α 2, whereas brain microvascular endothelial cells mainly produce laminin α 4 and α 5 ( Supplementary Fig. 2). The same expression pattern and comparable expression levels of these laminin α chains were observed in PKO-h and PKO-nh cells ( Supplementary Fig. 2), suggesting that loss of PDGFRβ + cell-derived laminin does not affect laminin expression in astrocytes or brain microvascular endothelial cells.
It has been shown that PDGFRβ labels both pericytes and VSMCs 30 . Consistent with these reports, TdT co-localized with both pericyte marker Desmin (Fig. 3b) and VSMC markers SMA and SM22α (Fig. 3c) in our lineage-tracing experiment. To further determine whether the CNS phenotype is due to loss of laminin in pericytes or VSMCs, we also generated a VSMC-specific conditional knockout line (laminin γ 1 flox/flox :Transgelin/ SM22α -Cre + , termed SKO hereafter) by crossing the laminin γ 1 flox/flox mice with the Transgelin/SM22α -Cre + line 12 . The specificity of the Transgelin promoter has been validated in a variety of studies [31][32][33] . The SKO mice are born at expected Mendelian ratio with no gross abnormality, although they have a lower blood pressure 12 . Immunohistochemical analysis revealed negligible levels of IgG and hemoglobin in SKO brains (Fig. 5a). In addition, no differences in FITC-Dextran (Fig. 5b) and Evans blue (Fig. 5c) levels in brain parenchyma were found between SKO and control mice. These data suggest that the CNS phenotype (hydrocephalus and BBB breakdown) observed in PKO-h mice is due to lack of laminin expression in pericytes rather than VSMCs.   Tight junction protein expression is reduced in PKO-h but not PKO-nh mice. To investigate the mechanisms that are responsible for the compromised vascular integrity in PKO-h mice, we first examined tight junction protein expression in endothelial cells. CD31 staining revealed comparable vessel density between control and PKO (both PKO-h and PKO-nh) brains (Fig. 2a), suggesting that the observed abnormalities are less likely due to defects in angiogenesis or differences in vessel density. Next, we examined the expression of tight junction proteins. Consistent with previous data, ZO-1 expression co-localized well with CD31 in 1-2-month-old control mice (Fig. 6a). In age-matched PKO-h mice, however, ZO-1 level was substantially decreased (Fig. 6a). In contrast to the PKO-h mice, PKO-nh mice showed ZO-1 expression to a level comparable to that in control brains (Fig. 6a). Western blot revealed similar results and quantification showed the reduction of ZO-1 in PKO-h brains was statistically significant (Fig. 6b). Like ZO-1, claudin-5 was substantially decreased in PKO-h but not PKO-nh brains (Fig. 6c), suggesting loss of tight junction proteins in PKO-h mice. In addition, we also examined the structure of tight junctions by electron microscopy and found no obvious defects. In control and PKO-nh brains, endothelial cells usually extended stretches that overlapped with each other, forming long junctions (Fig. 6d). In PKO-h brains, however, endothelial stretches tended to grow into each other and short junctions were more frequently observed (Fig. 6d). Altogether, these data suggest that the biochemical components of tight junctions are affected in PKO-h mice.

Astrocytic expression of AQP4 is reduced in PKO-h but not PKO-nh mice. Previous studies have
shown that astrocyte polarity, characterized by expression of aquaporin-4 (AQP4) predominantly in endfeet, also contributes to vascular integrity 22,[34][35][36] . Therefore, we also examined AQP4 expression in the control and PKO brains. At 1-2 months, AQP4 signal was predominantly found in the abluminal side of CD31 + vessels in control mice (Fig. 7a), suggesting that astrocytes are highly polarized. In age-matched PKO-h mice, however, AQP4 expression was significantly reduced and most CD31 + vessels were no longer covered by AQP4 (Fig. 7a). In contrast to PKO-h mice, PKO-nh mice demonstrated high level of AQP4 staining (Fig. 7a). Its distribution, however, showed a patchy pattern and was no longer along the abluminal side of CD31 + vessels (Fig. 7a). Consistent with these immunohistochemical data, western blotting revealed a significant reduction of AQP4 expression in PKO-h but not PKO-nh brains (Fig. 7b), suggesting that astrocytic expression of AQP4 is severely compromised in PKO-h mice (both expression level and distribution pattern) and moderately affected in PKO-nh mice (only distribution pattern).
Pericyte density/coverage is altered in PKO-h but not PKO-nh mice. Accumulating evidence suggests that pericyte coverage negatively correlates with BBB permeability 17,24,34,37 . To investigate if pericyte coverage is affected in PKO-h mice, we examined PDGFRβ expression by immunohistochemistry. In 1-2-month-old control brains, PDGFRβ co-localized with CD31 + vessels, covering most capillaries (Fig. 8a). In age-matched PKO-h brains, however, PDGFRβ expression was dramatically reduced, leaving many CD31 + vessels uncovered (Fig. 8a). In PKO-nh mice, PDGFRβ demonstrated an expression pattern similar to that in control (Fig. 8a). Western blot analyses revealed a significant decrease of PDGFRβ intensity in PKO-h but not PKO-nh brains (Fig. 8b). Consistent with these data, pericyte coverage was substantially decreased in PKO-h but not PKO-nh brains (Fig. 8c). Similar result was found when desmin was used as a marker for pericytes in this study (not shown). In addition, similar levels of pericytes were found around CD31 + vessels in control and PKO brains at E15.5 ( Supplementary Fig. 1b), suggesting that pericyte migration during early development is not affected in the PKO mice.

Discussion
In this study, we generated the PKO mice by crossing the laminin γ 1 flox/flox mice with the Pdgfrβ -Cre + line, and categorized them into two groups (PKO-h and PKO-nh) based on the size of their ventricles. We found that PKO-h but not PKO-nh mice showed signs of BBB breakdown and micro-hemorrhage. Consistent with this observation, decreased tight junction proteins, reduced pericyte coverage and compromised astrocyte polarity were detected in PKO-h but not PKO-nh brains. The phenotypes found in PKO-h mice, including hydrocephalus, BBB disruption, and muscular dystrophy, closely mimic the symptoms of dystroglycanopathies, a heterogeneous group of disorders characterized by defective glycosylation of dystroglycan 38 . When properly glycosylated, dystroglycans directly connect the extracellular matrix with cytoskeleton, allowing communications between cells and their environment. Defective glycosylation abolishes appropriate interactions between cells and the extracellular matrix and disrupts tissue organization 39 , leading to muscular dystrophy and brain deformity. Many genes in dystroglycan signaling pathway, including POMT1 40 and FKTN 41 , have been found to contribute to the pathogenesis of dystroglycanopathies. Our results suggest that laminin, a ligand for dystroglycan, is involved in the pathogenesis of dystroglycanopathies. This is, to the best of our knowledge, the first direct evidence linking laminin to hydrocephalus.
It should be noted that laminin expression is absent in PDGFRβ + cells from all PKO mice, whereas only 10.7% show brain abnormality (hydrocephalus and BBB breakdown). The absence of brain abnormality in the rest 89.3% of PKO mice could be due to an incomplete penetrance of the CNS phenotype, which can be easily explained by the C57BL/6J × FVB mixed genetic background. Accumulating evidence shows that the genetic background heavily affects the penetrance and/or expressivity of certain phenotypes. For instance, about 0% TGFβ 1 null animals survive to birth on a C57BL/6J/Ola background and the number increases to 80% on a NIH/ Ola background 42 . Another example is that transgenic mice expressing G93A human SOD1 gene have a more severe phenotype in ALR, NOD.Rag1KO, SJL or C3H background than in B6, B10, BALB/c and DBA lines 43 . In addition, considerable variations in phenotypes due to different genetic background have also been reported in fibronectin −/− and EGFR −/− mice 44,45 . Interestingly, although only 10.7% of the PKO mice show brain abnormality (hydrocephalus and BBB disruption), all of them demonstrate a severe muscle deficit 26 . This observation suggests different penetrance for the brain phenotype and muscle phenotype, which may be explained by the fact that brain pericytes and muscle pericytes have distinct biochemical properties and functions. It has been shown that pericytes in the brain and muscle have different developmental origins 24 , which may confer biochemical and functional differences to these two populations and thus result in different penetrance/phenotypes in PKO mice. To test this possibility, Cre lines that specifically target brain pericytes and/or muscle pericytes are required. Unfortunately, such mouse lines are currently not available.
The fact that BBB breakdown always goes hand in hand with hydrocephalus suggests that BBB disruption may be secondary to hydrocephalus. There is evidence showing that impairment of BBB integrity is in association with hydrocephalus 18,46,47 . For example, 3,6-diaminoacridine hydrochloride 48 and lanthanum chloride 49 were able to traverse the endothelial tight junctions and infiltrate into brain parenchyma in hydrocephalic mice. In addition, structural changes in endothelial cells were observed in both hydrocephalic rats 50 and hydrocephalic human patients [50][51][52] . Moreover, BBB-related transporters, including P-glycoprotein, were altered in hydrocephalic rats 53,54 . However, due to technical difficulties (e.g., long turnover rate of laminin), we are unable to experimentally demonstrate that the BBB breakdown in a small percentage (10.7%) of PKO mice is secondary to or caused by hydrocephalus.
Compared to the phenotype of astrocytic laminin deficiency (BBB breakdown and age-dependent intracerebral hemorrhage) 22,23 , loss of pericytic laminin leads to a milder phenotype in vascular integrity. This mild phenotype may be explained by the anatomical structure of the BBB and laminin isoforms expressed by these cells. At the BBB, pericytes are sandwiched between endothelial cells and astrocytic endfeet 24 , and covered by laminin-containing BM synthesized by all three cell types. Previous studies reported that astrocytes produce laminin α 1 and α 2, whereas endothelial cells make laminin α 4 and α 5 [13][14][15][16] . Our data showed that pericytes, like endothelial cells, predominantly synthesize laminin α 4 and α 5. Thus, loss of pericytic laminin could be compensated by endothelial laminin, leading to a mild phenotype; while loss of astrocytic laminin can not be compensated by laminin synthesized by endothelial cells or pericytes, resulting in a severe phenotype. This explanation is consistent with the phenotype of laminin α 4 deficient mice. The laminin α 4 −/− mice developed hemorrhage in embryonic and neonatal stages, but were recovered in adulthood 21 . The recovery may be explained by the contribution of pericyte-derived laminin α 4. However, since the PKO mice were not in a pure background, we cannot exclude the possibility that the mixed genetic background plays a role in the reported phenotype.
Consistent with the important roles of laminin in vascular integrity regulation, accumulating evidence shows that laminin receptors also actively regulate vascular permeability. For example, binding of laminin to integrin α 1β 1 and α 6β 1 promotes endothelial cell differentiation and vessel stabilization 55 . Blocking integrin β 1, on the other hand, decreased tight junction protein expression and increased vascular permeability 56 . In addition, deleting either subunit of integrin α vβ 8, which can bind laminin, vitronectin, and collagen IV 57-59 , leads to defective vascularization and intracerebral hemorrhage 60,61 . Furthermore, conditional knockout of integrin α v in astrocytes or neural progenitors results in severe hemorrhage 62 . Although hemorrhage was absent in mice with integrin β 8 deficiency in neural progenitors, BBB breakdown was identified in these mice 63 . Altogether, these data suggest that laminin, by binding to its receptors, contributions to the regulation of vascular integrity. In addition to laminin, other extracellular matrix proteins and their receptors also participate in the regulation of vascular integrity. For example, inactivation of fibulin-1 leads to massive hemorrhage and perinatal lethality 64 . Loss of nidogen-1 results in structural changes in the basement membranes at brain capillaries 65 . Next, recurrent multifocal hemorrhage was identified in Col4a1 +/Δex41 mice, which generate mutant Col4a1 lacking exon 41 that accumulates within cells 66 . Further studies showed that accumulation of this mutant protein in endothelial cells or pericytes, but not astrocytes leads to hemorrhage 66 . Altogether, these studies strongly suggest that extracellular matrix proteins and their receptors play critical roles in vascular integrity regulation.

Materials and Methods
Animals. Laminin γ 1 flox/flox mice were crossed with the Pdgfrβ -Cre + and Transgelin (SM22α )-Cre + transgenic lines to generate PKO (laminin γ 1 flox/flox :Pdgfrβ -Cre + ) and SKO (laminin γ 1 flox/flox :Transgelin/SM22α -Cre + ) mice, respectively. For PKO mice, those with and without hydrocephalus were defined as PKO-h and PKO-nh, respectively. In addition, Laminin γ 1 flox/flox mice were crossed with NG2-CreER TM to generate inducible knockout mice. For lineage tracing experiment, Ai14 reporter line was crossed with the Pdgfrβ -Cre + line and the resulting Ai14:Pdgfrβ -Cre + mice were used for experiments. The PKO mice used in this study were in a C57BL/6J × FVB mixed background, and both sexes were used. All mice were maintained in the animal facility at the University of Minnesota with free access to water and food. Experimental procedures were in accordance with the NIH guide for care and use of animals and were approved by the Institutional Animal Care and Use Committee at the University of Minnesota.

Primary Cell Isolation.
Pericyte. Primary brain pericytes were isolated as described previously with modifications 22,67 . Briefly, brains were collected under aseptic conditions and meninges were removed under dissection microscope. Next, the brains were minced with a blade and triturated. The tissue solution was then incubated with 0.1% collagenase at 37 °C for 1 hour, and centrifuged at 700 g for 8 minutes. The pellet was resuspended in 17% sterile dextran solution, followed by centrifugation at 6,000 g for 20 minutes at 4 °C. Blood vessel-containing pellet was washed in DMEM and sequentially filtered through 100 μ m and 40 μ m cell strainers. Microvessels trapped on 40 μ m cell strainer were further digested with 1 mg/ml collagenase/dispase (Roche) for 4 hours with constant shaking at 37 °C. Then, the cells were stained with anti-CD31-APC (Biolegend, 102509), anti-CD45-FITC (Biolegend, 103108), and anti-PDGFRβ -PE (eBioscience, 12-1402), and subjected to FACS. Sorted pericytes (PE + APC − FITC − ) were grown in Pericyte Medium (ScienCell, 1201) and used for immunocytochemistry.
Brain Microvascular Endothelial Cells. Brain microvascular endothelial cells were isolated using the same protocol. Briefly, brains were processed similarly as described above. After staining, endothelial cells (PE − APC + FITC − ) were isolated by FACS and grown in endothelial medium.
Astrocytes. Primary astrocytes were isolated from adult mouse brains as described previously with minor modifications 68,69 . Briefly, brains were collected and minced as described above, followed by incubation with papain (20 U/ml) for 90 minutes at 34 °C. Then protease inhibitor solution was added and the tissue was triturated with 5 ml serological pipettes. After passing through 70 μ m cell strainer, the cell suspension was subjected to myelin removal using Myelin Removal Beads II (Miltenyi Biotec, 130-096-733) according to the manufacturer's instruction. Next, the cell suspension was sequentially added to petri dishes pre-coated with CD45, CD31, O4, and CD90 antibodies to remove microglia/macrophages, blood vessels, OPCs, and neurons, respectively. Cells were allowed 15 minutes at room temperature to attach to each petri dish. Unbound cells were then transferred to the next petri dish. After the last petri dish, unbound cells (astrocytes) were collected and cultured in DMEM supplemented with 10% FBS and 1% Penicillin/Streptomycin.  22,23 . Then, 20 μ m-thick z-stacks were captured using a LSM 710 confocal microscope. Imaging processing was performed using ImageJ and Adobe Photoshop CC. All immunohistochemistry images are projections of the entire 20 μ m-thick sections and all immunocytochemistry images are individual slices. For IgG and Hemo quantification, mean signal intensity per field was used. Sections from control and knockout mice were placed in the same slides and processed in parallel. At least three random images from each section and at least three sections (more than 300 μ m apart from each other) per sample were analyzed for this analysis. Five animals were used for quantifications. Pericyte Coverage Analysis. Pericyte coverage analysis was performed as described previously 17,34,70 .
Briefly, brain sections were immunostained with CD31 to label blood vessels and PDGFRβ or NG2 or Desmin to label pericytes as described above. The areas covered by these signals were measured by ImageJ (area measurement tool). Pericyte coverage was quantified as the percentage (%) of PDGFRβ + or NG2 + or Desmin + pericyte area covering CD31 + capillary area per field. At least three random images from each section and five sections (more than 300 μ m apart from each other) per sample were analyzed for this analysis. Five animals were used for quantifications.
Histology. H&E staining was performed according to standard protocols.
Ventricle Size Quantification. It has been shown that the ventricle-to-brain ratio is linearly related to ventricle size and it provides a measurement of hydrocephalus severity 27,28 . Thus, the ventricle-to-brain ratio was quantified and used to reflect the size of ventricles as described previously with minor modifications 27,28 . Specifically, brains were subjected to sequential sectioning, and sections at the level of striatum and anterior commissure were selected for quantification. The ventricle-to-brain ratio was obtained by dividing the maximal width of the lateral ventricles with the maximal brain width. BBB Permeability Assays. For adults. sterile Evans blue solution (2% in saline, 10 μ g per g of body weight) or 4 kD-FITC-Dextran (25 mg/ml, 50 μ l) was injected intraperitoneally or intravenously into control, PKO-h, PKO-nh, and SKO mice, respectively. Twelve hours later, these mice were transcardially perfused with 50 ml saline, and the brains were removed and divided into two hemispheres. The right hemispheres were homogenized in 800 μ l saline followed by sonication and centrifuge (13200 rpm for 20 minutes at 4 °C). For FITC-Dextran, the supernatant was collected and read in a fluorescent plate reader (Biotek) at 485/528 nm. For Evans blue, 400 μ l supernatant was collected and mixed with equal amount of 50% trichloroacetic acid, followed by incubation over night at 4 °C. After centrifuge (13200 rpm for 20 minutes at 4 °C), the supernatant was collected and read in a spectrophotometer (Biotek) at 620 nm. Each sample was measured in triplicates and the average of these technical replicates was used as one biological replicate. Six animals per group were used for quantification. The results were normalized to the controls.
For P2 pups. P2 pups were anesthetized and 10 μ l of 4-kD FITC-Dextran (25 mg/ml) was injected into the left ventricle with a Hamilton syringe. After 5 minutes of circulation, brains were collected and fixed in 4% PFA overnight at 4 °C. The brains were then incubated in 30% sucrose for 24 hours and frozen in OCT. 20 μ m-thick sections cut with a cryostat were used for immunohistochemistry.
For embryos. Embryonic BBB permeability assay was performed according to a published protocol 71 . Briefly, a Cesarean section was performed on pregnant females. Each E15.5 embryo was injected with 5 μ l FITC-Dextran (4 kD, 25 mg/ml) into the liver using a Hamilton syringe, while they were still attached via the umbilical cord to the mother's circulation. After 3 minutes, the heads were collected and fixed in 4% PFA overnight at 4 °C. The heads were then incubated in 30% sucrose for 24 hours and then frozen in OCT. 20 μ m-thick sections cut with a cryostat were used for immunohistochemistry. Western Blot. Brain samples were homogenized in lysis buffer on ice. Equal amount of proteins was loaded and separated in SDS-PAGE and then transferred to PVDF membranes (Millipore). The membranes were probed with anti-laminin γ 1 (Thermo, RT-795, 1:500), Claudin-5 (Thermo, 35-2500, 1:500), ZO-1 (Thermo, 61-7300, 1:500), AQP4 (Millipore, AB3594, 1:500), PDGFRβ (Cell Signaling, 3169, 1:500), and anti-actin (BD, 612657) antibodies over night at 4 °C. Next, the membranes were incubated with appropriate secondary antibodies at room temperature for 1 hour. Proteins were visualized by ECL (PerkinElmer, NEL104001EA) and ChemiDoc Imaging System (Bio-Rad). The density of bands was quantified using Quantity One and normalized to actin. The expression levels of these proteins in PKO-h and PKO-nh brains were normalized to that in control brains. Five animals per group were used for quantification. Electron Microscopy. Electron microscopy was performed as described previously 22 . Briefly, age-matched control, PKO-h, and PKO-nh mice were anesthetized and perfused with PBS followed by fixing buffer (0.1 M sodium cacodylate buffer with 2% paraformaldehyde and 2% glutaraldehyde). Next, the striatum was carefully dissected out and fixed in fixing buffer overnight. The tissue was then post-fixed in 1% osmium tetroxide and 1% K-ferrocyanide, en bloc stained with 2% uranyl acetate, and embedded in resin. Ultra-thin sections were cut on a Reichert-Jung Ultracut E microtome. After post-stained with 2% uranyl acetate and 1% lead citrate, the sections were examined and photographed using JEOL100CXII at 80 KV.

Statistics.
Results are shown as mean ± S.D. Student's t-test, performed by SPSS Statistics or Microsoft Excel, was used to analyze differences between two groups. Sample number (n) represents biological replicates.