Pericyte, but not astrocyte, hypoxia inducible factor-1 (HIF-1) drives hypoxia-induced vascular permeability in vivo

Mice were standardly housed at a constant temperature of approx. 22 °C under a 12 h dark/light cycle with unlimited access to water and normal chow (65% carbohydrate, 22% protein, and 12.5% fat). The cages were equipped with nesting material and housing structures. All animal procedures were performed in accordance with Swiss animal protection laws and approved by the University of Zurich Institutional Animal Care guidelines. For clarity, pericyte data are always presented in blue graphs while astrocyte data is green. Cell-specific HIF-1 loss of function (LoF) mice were generated using homozygous floxed HIF-1α^fl/fl mice in which exon 2 is floxed [31] (kindly provided by Dr. Randall Johnson, University of Cambridge, UK). Notably, the HIF-1α exon 2 is critical for the dimerization of HIF-1α with HIF-1β and subsequent HIF-1 transcriptional activity. Thus excision of exon 2 via Cre recombinase prevents both dimerization and transcriptional activation resulting in HIF-1 LoF. Astrocyte-specific HIF-1 LoF animals were generated by crossing these mice with a line expressing a Cre recombinase fusion protein having a modified estrogen receptor binding domain (CreER^T2) under the control of either the glial fibrillary acidic protein (GFAP) [32] or glutamate aspartate transporter (GLAST) [33] promoter to obtain Cre heterozygous floxed mice, namely GFAP-CreER^T2; HIF-1α^fl/fl or GLAST-CreER^T2; HIF-1α^fl/fl respectively (Fig. 1). Both male and female animals were included in the analysis. Pericyte-targeted HIF-1α LoF was obtained by crossing the same floxed animal with a smooth muscle myosin heavy chain promoter (SMMHC)-CreER^T2 line kindly provided by Dr. Stephan Offermanns, Max Planck Institute, Germany (Fig. 1). Cre-mediated recombination in these mice has been shown to occur mainly in microvascular pericytes and to a lesser degree in smooth muscle cells of larger vessels [34]. Other studies confirm the SMMHC-CreER^T2 system localizes mainly to pericytes in the capillary bed [35, 36] a fact confirmed by our observation that HIF-1α LoF did not impact hypoxic water content or Evans Blue leakage in organs outside the CNS (data not shown). Only male littermates were used as SMMHC-CreER^T2 is located on the Y chromosome [34]. It is also important to note that all lines had different genetic backgrounds despite being crossed to the same Cre line. Standard mouse genotyping was performed on DNA isolated from ear biopsies using single or duplex polymerase chain reaction (PCR) with primers against HIF-1α and the specific Cre recombinase (commercially obtained from Microsynth AG, Switzerland). Primer sequences, concentrations and annealing temperatures are indicated in Table 1. The PCRs were standardly run with 35 cycles and an extension temperature of 72 °C.

Animals were randomly assigned to groups by the study director. Subsequent experiments and analyses were performed double-blinded. Tamoxifen was used to induce excision of HIF-1α exon 2 in 8–12 week old adult mice. Briefly, tamoxifen (Sigma-Aldrich, Switzerland) was dissolved in a sunflower oil/ethanol (10:1) mixture at 100 mg/ml. Animals were injected once a day intraperitoneally (50 mg/kg body weight) for 5 consecutive days and experiments started only after 5 subsequent days had elapsed to allow complete recombination (Additional file 1: Fig. S1A). The tamoxifen stock solution was adjusted with oil to a final volume of 40 µl for all injections, control mice were injected with 40 µl sunflower oil/ethanol mix only. Exon deletion was confirmed using genomic DNA isolated from cortical tissue or isolated microvessels. Standard PCR for floxed HIF-1α (35 cycles, 65 °C elongation) was performed on 50 ng of genomic DNA using the following primers (200 nm final concentration): HIF-1α^flox/flox forward 5’-GGGATGAAAACATCTGCTTTGGA-3’ and reverse 5’-TGTGTTGGGGCAGTACTGG-3’ and β-actin forward 5’-CTGGCTCCTAGCACCATGAAG-3’ and reverse 5’-GCCACCGATCCACACAGAGT-3’.

Half brains were homogenized with a dounce homogenizer in ice cold buffer (0.27 M sucrose, 2 mM EDTA pH8, 600 mM KCl, 15 mM NaCl, 15 mM HEPES) supplemented with 1 mM phenylmethansulfonyl fluoride (PMSF, Sigma-Aldrich, Switzerland), 1 mM sodium orthovanadate (NaV, Sigma-Aldrich, Switzerland), and proteinase inhibitor cocktail (Calbiochem, Germany). The cell suspension was layered onto a sucrose cushion (30% w/v sucrose, 2 mM EDTA pH8, 600 mM KCl, 15 mM NaCl, 15 mM HEPES) and centrifuged at 1500 g for 10 min at 4 °C. The cytosolic fraction was collected and the nuclear pellet resuspended in extraction buffer (20 mM HEPES, 400 mM NaCl, 1 mM EDTA pH8) supplemented with 1 mM PMSF, 1 mM NaV and proteinase inhibitor cocktail. After 15 min incubation on ice the nuclear fraction was obtained by centrifugation (15′000 g for 5 min) at 4 °C. Proteins (30 µg) were separated by SDS-PAGE then transferred to nitrocellulose membranes (Amersham™ 0.45 µm, Sigma-Aldrich, Switzerland). After blocking in 5% non-fat dry milk in TBS, membranes were incubated with primary antibodies against Cre (1:1000, MAB3120, Sigma-Aldrich, Switzerland) and β-actin (1:5000, A5441, Sigma-Aldrich, Switzerland). Membranes were then incubated with horseradish peroxidase-conjugated secondary antibodies for 1 h at RT. Bands were detected using a luminescent image analyzer (Fujifilm, LAS-3000, Switzerland).

Mice in their home cages were exposed to either normoxia (21% O₂) or hypoxia (8% O₂) for 96 h in an oxygen-controlled humidified glove box (Coy Laboratories, USA) with unlimited access to food and water (standard chow). Temperature (20–24 °C) and relative humidity (45–60%) were also computer controlled. Post euthanisation, blood samples were sampled by heart puncture and hematocrit was used to confirm the hypoxic impact. As expected, all animals demonstrated similar hemoconcentration that was unaffected by tamoxifen treatment (Additional file 1: Fig. S1B).

All cell culture reagents were obtained from Gibco (Thermo Fischer Scientific, Switzerland) unless otherwise indicated. Astrocyte isolation was performed according to an in-house protocol. In short, isolated cortices were mechanically dissociated then enzymatically digested in HBSS supplemented with 0.25% Trypsin, 10 mM EDTA and 1 mg DNaseI (Roche, Switzerland) at 37 °C for 30 min. The homogenate was separated by gradient centrifugation (22% BSA in PBS, 3000 g) and cells resuspended in DMEM supplemented with 10% horse serum, 1% Penicillin/streptomycin, 1% L-glutamine and 50 µg/ml gentamycin (AppliChem GmbH, Germany). Cells were seeded on collagen IV coated 24 well plates and media changed first after 48 h, then every 5 days until near confluency. Astrocytes were not passaged before use.

Pericytes were isolated as previous [37] with slight modifications. Cortices were homogenized and the tissue digested in DMEM buffer supplemented with 2 mg/ml collagenase-dispase, 10 μg/ml DNase I and 30U/ml papain for 60 min at 37 °C. The digested tissue was then homogenized by passing subsequently through 18 and 21 gauge needles. The homogenate was separated by gradient centrifugation (22% BSA in PBS) at 4000 g for 10 min. The obtained cell pellet was washed and after final centrifugation (700 g for 5 min), pericytes were plated on collagen coated petri dishes in selective media (DMEM containing 20% FBS, 50 μg/ml gentamycin sulfate and 2.5 μg amphotericin B) until confluency. After first passage cells were maintained on uncoated dishes without amphotericin and used at passage 1. Purity of all cultures was ≥ 98% as assessed by immunostaining for standard cell markers GFAP, GLAST, NG-2 and PDGFR-β as previous [37]. Iba1 and CD31 immunostaining further confirmed absence of contaminating microglia or endothelial cells.

Astrocytes and pericytes isolated from GLAST-CreER^T2; HIF-1α^fl/fl and SMMHC-CreER^T2; HIF-1α^fl/fl mice respectively, were treated with oil or tamoxifen (2 µM) during 48 h hypoxic exposure (1% O₂) then harvested in TRIzol® reagent (Thermo Fisher Scientific, Switzerland). As shown previously [37], this time point ensures hypoxic target gene induction in these primary cells without the occurrence of cell death. Total RNA was isolated by PureLink® RNA Mini Kit (Invitrogen, USA) according to the manufacturer’s description. 1 μg of RNA was reverse transcribed (ImProm-II ReverseTranscriptase kit, Promega, Switzerland) and 20 ng cDNA used for Power Sybr® Green qPCR with an ABI 7500 Fast Real-Time PCR System (Applied Biosystems, Switzerland). QPCR primers were used at 100 nM final concentration: Glut1 forward 5’-GGGCATGATTGGTTCCTTCTC-3’ and reverse 5’-CAGGTTCATCATCAGCATGGA-3’; VEGF forward 5’-CGCAAGAAATCCCGGTTTAA-3’ and reverse 5’-CAAATGCTTTCTCCGCTCTAA-3’; β-actin forward 5’-CTGGCTCTAGCACCATGAAG-3’ and reverse 5’-GCCACCGATCCACACAGAGT-3’. All data were normalized to β-actin and fold changes were calculated based on the comparative ΔΔCt method.

At experimental end animals were euthanized by CO₂ inhalation. Brain and other organs were quickly removed. Wet weights were measured then tissues dried at 85 °C for 72 h to obtain dry weight values. Water content was calculated as [(wet – dry weight)/wet weight] × 100% and graphically represented. Due to hypoxia-induced body weight loss, water content was normalized to end body weight in GLAST-CreER^T2; HIF-1α^fl/fl mice [38].

Tracer extravasation was used to assess vascular permeability as previously described [39]. Briefly, mice were anesthetized with isoflurane then injected with a mixture of 450 Da Lucifer yellow (6.25 mM, Sigma-Aldrich, Switzerland) and 70 KDa Texas red dextran (0.01875 mM, Invitrogen, USA) in 0.9% NaCl via the femoral vein. After circulation for 3 min mice were overdosed with ketamine/xylazine and transcardially perfused with ice-cold PBS for 10 min. Brains were harvested, weighed then homogenized in ice cold PBS. After centrifugation (10,000 g for 5 min) fluorescence intensity of the supernatants was measured (excitation 425 nm and emission 525 nm for Lucifer yellow; emission 595 nm and excitation 625 nm for Texas red dextran). Sham animals, injected with 0.9% NaCl only, were included in each experiment to control for background auto fluorescence. Brain fractions were calculated using the formula: (brain raw fluorescent units/ serum raw fluorescent units)/ brain wet weight [mg]. Due to hypoxia-induced body weight loss in GLAST-CreER^T2; HIF-1α^fl/fl mice this formula was corrected for total blood volume of each animal at experimental end: (brain raw fluorescent units/ serum raw fluorescent units) / (brain wet weight [mg] x end blood volume [ml]).

Animals were anesthetized then injected with 1% Evans blue dye (2 μg/g body weight) via the femoral vein. After 45 min circulation time, blood was collected via heart puncture and animals transcardially perfused with ice cold PBS for 10 min. Brains were then collected, rinsed in ice-cold PBS, blotted dry and weighed. Dye retained in the tissue was extracted with formamide (5 μl/mg tissue) for 72 h. Absorbance was measured at 620 nm and expressed as fold change of normoxic Ctrl mice.

Brain sagittal cryosections (20 µm) at 1.35–1.525 mm lateral to the midline were fixed in 4% paraformaldehyde for 10 min then incubated in 5% NGS in PBS for 60 min. Endogenous mouse IgG was blocked for 2 h with unconjugated goat anti-mouse IgG Fab Fragments (1:100, 20 µg/ml, 115,007, Jackson ImmunoResearch, UK). Sections were incubated overnight at 4 °C with antibodies against CD31 (1:1000, 5 µg/ml, 553,370, Millipore, USA), ZO-1 (1:100, 2.5 µg/ml, 61–7300, Invitrogen, USA), claudin-5 (1: 250, Invitrogen, USA), NG-2 (1:200, 5 µg/ml, AB5320, Millipore, USA), PDGFR-β (1:500, 0.4 µg/ml, sc-432, Santa Cruz, USA) or Cre recombinase (1:500, MAB3120, Millipore, USA) followed by secondary antibody for 1 h and counterstained with DAPI. Images were acquired with an epifluorescence microscope (Carl Zeiss, Germany) at 20× magnification spanning three brain regions namely cortex, subcortex and hippocampus. For each group, a minimum of three animals were analysed from independent experiments. From each animal, three individual brain sections were used for analysis. Per section three images were captured for each brain region, yielding 9 images for each brain region per animal. Processing and quantification of the number and diameter of CD31-immunopositive blood vessels was carried out manually using NIH ImageJ. Vascular density was expressed as vessel number per mm². ZO-1 and claudin-5 staining areas were obtained using the threshold method in NIH ImageJ. Tight junction image analysis was performed by calculating the area of overlap of ZO-1/claudin-5 positive areas with CD31-positive endothelial cells. All quantitative analyses were performed blind.

Transgenic animals were transcardially perfused using 0.1 M PBS, 2.5% glutaraldehyde, 2% formaldehyde for 15 min then brains were isolated and fixed for 48 h. Cortices were dissected and cut into approximately 1mm³ pieces using a vibratome. After washing in 0.1 M CaCo and post fixation in 1% osmium in water on ice for 1 h, tissues were contrasted with 1% uranyl acetate overnight. Following dehydration in an alcohol series brains were embedded in Polypropylene, Propp:Epon-Araldite 1:1 for 1 h, followed by Epon-Araldite for 28 h at 60 °C. Ultrathin sections (70 nm) were cut with a Reichert Ultracut E microtome then post stained with lead for 10 min. Sections imaged using a Phillips CM100 transmission electron microscope with a digital Gatan Orius 832 camera (FEI, The Netherlands).

Measurements of intracellular and serum levels of VEGF and transforming growth factor β (TGF-β) were performed using Quantikine ELISA kits (VEGF: MMV00, TGF-β: DB100B, R&D systems, USA) according to the manufacturer’s instructions using 500 µg of brain protein and fivefold diluted serum. Optical density was measured at 450 nm, with wavelength correction at 570 nm (Multiskan RC; Thermo Labsystems, Finland).

Statistical comparisons were performed using GraphPad Prism 5 software (GraphPad, USA) and all data are expressed as mean ± SD of at least 3 independent experiments. Comparisons among different groups were made using one-way ANOVA with Holm-Sidak post hoc test or two-way ANOVA with Tukey's correction where appropriate. P values < 0.05 were considered significant.

To confirm exon 2 excision after tamoxifen treatment standard genotyping was performed on DNA isolated from cortices of mice using duplex polymerase chain reaction (PCR). In all lines only the WT allele (HIF-1α^F, 1000 bp) was detected after oil treatment whereas both WT and exon 2 floxed fragments (HIF-1α^Δ, 300 bp) were amplified after tamoxifen treatment as shown in SMMHC-LoF (Fig. 2A) and GLAST-LoF (Fig. 2A) mice. This outcome was unaffected by hypoxic exposure. Immunofluorescence staining showed that Cre expression co-localizes with PDGFR-β and NG2 in the SMMHC line (Fig. 2B), and GFAP-positive vascular-associated cells in the GLAST line as expected (Fig. 3B). Western blot analysis  confirmed tamoxifen treatment induces translocation of Cre from cytoplasmic to nuclear cortical fractions in the different CreER^T2; HIF-1α^fl/fl mouse lines confirming recombinase activation (Figs. 2C and 3C and Additional file 1: Fig. S2). Finally, loss of HIF-1 activity and function was verified by evaluating hypoxic induction of its target genes in the cellular specific compartments. To this end primary astrocytes and pericytes were isolated and cultured from the mice then exposed to either tamoxifen or oil prior to normoxic or hypoxic exposure. Quantitative real-time PCR showed mRNA levels of key HIF-1 target genes Glut1 (Figs. 2D and 3D) and VEGF (Figs. 2E and 3E) were induced by hypoxia in both oil treated pericytes and astrocytes but significantly abrogated in tamoxifen exposed cells as expected.

Sustained exposure to hypoxia is known to induce endothelial remodelling in different disease paradigms [40] but whether perivascular HIF-1 stabilization influences microvascular stability still needs to be addressed. We performed immunostaining with an anti-CD31 antibody to evaluate if HIF-1 LoF in pericytes or astrocytes modulates vessel characteristics when mice are exposed to 21% or 8% O₂ for 96 h. As expected brain sections from normoxic Ctrl and LoF SMMHC-Ctrl mice showed no apparent differences in vessel structure (Fig. 4A). A clear hypoxia-induced increase in vessel diameter in all regions including cortex, sub-cortex and hippocampus was noted in the hypoxic SMMHC-Ctrl mice whereas, surprisingly, vessels of hypoxic SMMHC-LoF mice looked very similar to normoxic controls (Fig. 4A). Quantitative analysis of mean vessel diameter confirmed hypoxia induced vessel dilation was abrogated in cortex (Fig. 4B) as well as subcortex and hippocampus (Additional file 1: Fig. S3A, B). Notably, total number of cortical vessels and vessel density was largely unaffected (Fig. 4C, D).

HIF-1 LoF also had no effect on normoxic vessel diameter in the GLAST (Fig. 5A) or GFAP astrocyte lines (data not shown) and did not abrogate the hypoxia-mediated vessel dilation seen in the Ctrl animals (Fig. 5A). Quantification confirmed increased mean vessel diameter in cortex (Fig. 5B) as well as subcortex and hippocampal (Additional file 1: Fig. S3C, D) regions of hypoxic GLAST mice. Similar data was obtained using GFAP LoF mice (Additional file 1: Fig. S4A–C). Notably however, HIF-1 LoF had no effect as both the total number of vessels and vessel density remained constant (Fig. 5C, D). Thus pericyte, but not astrocyte, HIF-1 LoF prevents hypoxia-mediated vessel dilation.

Hypoxia-driven vascular remodelling is highly associated with increased vascular permeability in vivo [40]. Unexpectedly, the absence of detectable vascular modifications in both the GLAST and GFAP lines implied little contribution of astrocyte activated HIF-1 pathways to hypoxia-induced vascular changes. In contrast, since blocking pericyte-mediated HIF-1 signalling suppressed vessel dilation it seemed highly likely that BBB integrity was also better maintained in these animals. To directly evaluate barrier functionality we assessed vascular permeability using different functional assays. First, we quantified brain edema by measuring water content (i.e. % wet weight) of whole brain tissue after 96 h normoxic or hypoxic exposure. Significantly increased wet weight in SMMHC-Ctrl mice during hypoxic conditions were appreciably abrogated in the SMMHC-LoF animals (Fig. 6A). In correlation functional permeability assays showed hypoxia substantially increased paracellular flux of both Evans blue (Fig. 6B) and Lucifer yellow (Fig. 6C) tracers in SMMHC-Ctrl animals as expected, but in both cases this effect was abrogated in SMMHC-LoF animals. In complete contrast, and in correlation with the inability of astrocyte HIF-1-LoF to suppress hypoxia-induced vascular dilation, endothelial permeability remained unaffected. Indeed, water content as well as Lucifer yellow and Evans blue flux was constant in both hypoxic GFAP- and GLAST-LoF mice compared to controls (Fig. 6D–F). Thus specifically reducing pericyte but not astrocyte HIF-1 activation suppresses hypoxia-induced BBB permeability. Of note, no permeability changes were measured when analysing the tracer dye Texas Red (Additional file 1: Fig. S5A, B).

Since hypoxia-induced vascular alterations were resolved only in SMMHC-LoF animals we focused on this line in further experiments. To understand how BBB permeability was modulated we first assessed if HIF-1 LoF alters pericyte survival or proliferation. Only few Ki-67 positive pericytes were observed after hypoxia with no difference observed between Ctrl and LoF mice and no proliferation of other cells was detected (Additional file 1: Fig. S6A). Similarly no cell death (measured by TUNEL) was observed at all (Additional file 1: Fig. S6B) suggesting that hypoxic BBB integrity is primarily modulated by HIF-1-driven alterations in pericyte signalling. Several downstream mediators are potential candidates for the mechanism. We focused on VEGF, TGF-β and MMPs as they are HIF-1 targets known to be strong mediators of vascular permeability that can be secreted and/or activate cognate receptors on the endothelium [41]. Analyses of both brain protein lysates and serum samples by ELISA showed increased levels of VEGF and TGF-β during hypoxia as expected but no alterations after pericyte HIF-1 LoF (Fig. 7A-D). No changes in MMP-2 or 9 by either zymography or in situ zymography were detected (data not shown).

Finally we assessed if microvessel tight junction organization was altered by pericyte-targeted HIF-1 LoF. Immunostaining of brain sections from normoxic and hypoxic SMMHC-Ctrl and SMMHC-LoF mice was performed for the tight junction proteins occludin, claudin-5 and ZO-1 (Fig. 8) in combination with the blood vessel marker CD31. Low power ZO-1 images (Fig. 8A) provide an overview of vessel structure and regional TJ expression whereas high power claudin-5 images (Fig. 8B) highlight TJ organization. In control animals (Hx Ctrl) an evident reduction of vessel localized ZO-1 staining was observed with regions of complete loss of signal compared to Nx Ctrls. Although similar disruption of the vessels was seen in Hx LoF mice, ZO-1 expression was more similar to normoxic animals with reduced disrupted or discontinuous staining. High power Z-stack images of claudin-5 staining (green, Fig. 8B) revealed multiple breaks and regions of complete signal loss at vessel walls in hypoxic SMMHC-Ctrl animals (Hx Ctrl) as indicated by arrow heads. In contrast hypoxic SMMHC-LoF mice (Hx LoF) showed considerably improved TJ localization that was more aligned to normoxic animals without disrupted or discontinuous staining. Comparable results were obtained for occludin (data not shown). Quantification of the staining further confirmed that hypoxia-induced ZO-1 and claudin-5 disruption is rescued in Hx SMMHC-LoF mice (Fig. 8C, D). To assess TJ arrangement in even more detail, direct transmission electron microscopy was performed on brain sections from these animals. Results confirmed that pericyte HIF-1 LoF improves tight junction organization during hypoxia (Fig. 8E). Close apposition of two endothelial membranes is apparent in normoxic SMMHC-Ctrl (Nx Ctrl) as indicated by the arrow. This organization is compromised during hypoxia (Hx Ctrl) wherein the uniform pattern of electron dense particles at the TJ becomes more akin to a string of pearls and gaps are clearly evident (arrowheads, Fig. 8E). In contrast hypoxic SMMHC-LoF animals (Hx LoF) display tight junction arrangement very similar to normoxic animals.