Circulating tight-junction proteins are potential biomarkers for blood–brain barrier function in a model of neonatal hypoxic/ischemic brain injury

Postnatal day 7 (PND7) Wistar rat pups were bred in-house at the Laboratory for Experimental Biomedicine of Gothenburg University (parents were sourced from Janvier Labs, Le Genest-Saint-Isle, France) and maintained under normal housing conditions with a 12 h light/dark cycle and free access to water and standard laboratory fodder. Animals of both sexes and different litters were used for the experiments and care was taken to minimise the number of animals used and to maintain an even sex-balance in all experimental and control groups. All experiments were approved by the Gothenburg Committee of the Swedish Animal Welfare Agency (Application nos. 663/17) and performed in accordance with the ARRIVE guidelines. A total of 104 animals were used throughout the study.

PND7 rats were anaesthetised with isoflurane (3.5–5%, Vetmedic, Stockholm, Sweden) in a 50/50 oxygen/nitrogen mixture, placed on their backs and a small incision was made in the neck to gain access to the carotid artery. A suture was placed permanently around the left carotid artery and the incision sealed with Vetbond tissue adhesive (3 M, MN, USA). Surgery typically lasted for 3–5 min. Following surgery, pups were allowed to recover in their home cage together with their mother for 1 h. Subsequently, operated pups were placed in a 36 °C chamber. The chamber was perfused first with humidified air for 10 min followed by 8% oxygen for 1 h and then by humidified air for 10 min. After the hypoxic exposure, pups were returned to their home cages. Control animals was subjected to sham-surgery (anaesthesia and incision) but no hypoxia. During all procedures, animals were monitored for vitals (i.e. breathing and skin-colour); every animal in the study survived the surgery and hypoxia.

For all time-points after HI (i.e. 6 h, 24 h and 5 days) injured and control animals were euthanised with a lethal overdose of pentobarbital. Cerebrospinal fluid was collected from the cisterna magna through glass capillaries as described previously [39] and blood was collected by cardiac puncture with ethylenediaminetetraacetic acid (EDTA)-treated syringes. CSF was checked for blood contamination as previously described [40] and samples discarded when contamination detected (detection limit about 0.2%). Blood samples were centrifuged at 2000xg for five min to separate the plasma. Samples were placed on dry ice after collection and long-term stored in - 80 °C freezer until analysed. Whole brains (excluding the cerebellum and brain stem) were collected and immersed in cold 6% buffered formaldehyde (Histofix; Histolab, Gothenburg, Sweden) at 4 °C for 24 h before processing for paraffin embedding.

The activity of cleaved caspase-3 at 6 h (n = 7) and 24 h (n = 7) after HI was measured using a fluorometric assay based on an earlier study [41]. Whole brain hemispheres where homogenised in cold RNase free phosphate-buffered saline (PBS) and sonicated in cold RNase free PBS containing 2% protease inhibitor cocktail (Sigma-Aldrich, MO, USA) and 10 mM EDTA. Aliquots were centrifuged at 10 000xg for 15 min in 4 °C and some supernatant were used for bicinchoninic acid concentration measurements. For caspase-3 activity, 20 μl supernatant were incubated with 80 µl extraction buffer composed of a buffer base (50 mM Tris, 100 mM NaCl, 5 nM EDTA, 1 mM egtazic acid, pH 7.3) and 0.2% 3-(3-cholamidopropyl)dimethylammonio-1-propanesulfonate, 1% protease inhibitor cocktail, and 1 mM phenylmethylsulfonyl fluoride (PMSF) (Sigma-Aldrich) on a 96 well plate for 15 min in room temperature (RT). 100 µl assay buffer made up of buffer base plus 4 mM dithiothreitol, 1 mM PMSF and 25 µM caspase-3 substrate (Peptides International, KY, USA) were added to the wells before the plate was read for 1 h at 37 °C with 2 min intervals on a SpectraMax Gemini EM microplate reader (Molecular Devices, CA, USA) set to excitation wavelength 380 nm and emission wavelength 460 nm. Endpoint readings were made before and after 10 µl of 10 µM free 7-amino-4-methylcoumarin (AMC) (Peptides International) and the V_max was calculated from the linear part of the curve, caspase-3 activity was expressed as pmol AMC/min·mg caspase-3.

Plasma- and CSF-samples were analysed using pre-coated ELISA kits for tight-junction proteins CLDN5 (Nordic BioSite. Stockholm, Sweden), OCLN (Cusabio, Wuhan, China), and ZO-1 (Cusabio) as per the manufacturer’s instructions. Plasma was diluted 20 times and CSF 10 times. In short, standards, CSF and plasma-samples were diluted in sample diluent buffer and incubated on ELISA-plates pre-coated with the antibody. After incubations with a biotinylated secondary antibody, horseradish peroxidase (HRP)-avidin, 3,3′,5,5′-Tetramethylbenzidine substrate, and a stop-solution the optical density was determined with a Spectramax Plus microplate reader (San Jose, CA, USA) set to 450 nm with 540 nm wavelength-correction (OCLN and ZO-1) or 450 nm (CLDN5). The protein concentration was determined from the resulting standard-curve. CSF and plasma from the same animals were analysed for both CLDN5 and OCLN, n = 7–8 for all time-points. Due to some differences between ELISA-plates, all data were normalised to the median of time-matched controls analysed on the same plate.

The blood–brain barrier permeability was measured using radiolabelled sucrose as described by our group earlier [39]. For all time points after HI; 6 h (n = 9), 24 h (n = 8), 5 days (n = 9), injured and PND7 and PND12 control animals (n = 6 for each age) were injected intraperitoneally (i.p.) with two µCi ¹⁴C‐sucrose (American Radiolabelled Chemicals, MO, USA) in saline (100 µl injection volume). Thirty min later, they were euthanised with a lethal overdose of pentobarbital. Blood was collected through cardiac puncture using a heparinised syringe and centrifuged at 2000xg for five min to separate plasma. Choroid plexuses were removed and whole cerebellum and brain stem as well as left and right hippocampus, cortex and striatum/thalamus were dissected and collected into pre-weighed scintillation vials and then re-weighed. 500 µl of Solvable (PerkinElmer, MA, USA) was added to all samples and they were incubated overnight in a 40 °C oven to dissolve the tissue. After checking that all tissues were solubilised, samples were left to cool down to RT, mixed with 10 mL Ultima Gold scintillation cocktail (PerkinElmer) and left for 60 min in darkness. The radioactivity in each sample was determined by liquid scintillation counting in a Tri-Carb 4910TR (PerkinElmer) and calculated as cpm/mg sample after background corrections. Brain/plasma sucrose concentration ratios were used as a measurement of BBB-permeability as previously described after correcting for residual blood space in brain [42]. These concentration ratios were calculated as a measure of BBB permeability in each region and ratios in the left (injured) hemisphere was compared to the right hemisphere as previously outlined [15].

BBB-disruption after HI was also tested using injections of Evans blue (EB) dye, a dye that binds to albumin in the blood and thus should be regarded as a high-molecular marker opposed to sucrose [43]. 4% EB dissolved in PBS were injected i.p. (4 µl per g body weight) 6 h post-HI (n = 3) and in control animals (n = 3). After 1 h, animals were euthanised with a lethal overdose of pentobarbital and transcardially perfused with saline and 6% buffered formaldehyde. Whole brains (excluding the cerebellum and brain stem) were collected and immersed in cold 6% buffered formaldehyde at 4 °C for 24 h before they were embedded in 4% agarose and cut in 100 µm thick sections in a Leica 1200 VT vibratome (Leica Biosystems, Wetzlar, Germany).

Sections were mounted in water-based CC/Mount (Sigma) and imaged at 680 nm, the wavelength in which emitted fluorescence from EB peaks [44]. EB-extravasation into the brain were quantified in cortical micrographs from injured animals (n = 3) and controls (n = 3). Images were segmented via thresholding, creating binary images of EB + -area which were measured and calculated as a percentage of the entire image area.

Entire hemispheres of CLDN5-stained fluorescent paraffin-sections of brains collected five days after HI and controls (n = 5 per group) were imaged with a tiling and stitching function. Two levels (700 µm apart) at mid-hippocampal level were imaged per animal and analysed with an in-house developed macro for the Fiji-build [45] of ImageJ [46] that utilises difference of Gaussian to eliminate all background while preserving all vessel information to accurately measure the area of blood vessels in an image. Briefly, entire CLDN5-stained brain hemispheres were imaged in a fluorescent microscope (Additional file 1: Figure S1a); the process is shown in a smaller selection of the image (Additional file 1: Figure S1b) for clarity. A copy of the image was subjected to Gaussian blur with sigma = 10 (Additional file 1: Figure S1c). The blurred image was subtracted from the original and threshold applied with Fiji’s “analyse particle”-tool to filter out any eventual debris so only marked blood vessels remained (Additional file 1: Figure S1d). The resulting vessel-image were then superimposed on the original image (Additional file 1) to confirm accurate vessel labelling. Blood vessel area was quantified in both injured and uninjured hemispheres by first outlining a region of interest (ROI) delineating the entire cortex and hippocampus and measure the total tissue area. Then the area of marked blood vessels within the ROI was determined and the percentage of blood vessel area of the total area in the hemisphere was calculated. Averages were calculated from the two mid-hippocampal levels per animal. Investigators were blinded to treatment groups during analysis.

Paraffin-embedded brains were cut in seven µm thick coronal sections at six levels and 40 sections apart with a microtome, starting at what corresponds to approximately -2.5 mm from bregma in an adult rat. For 3, 3′-diaminobenzidine (DAB) immunohistochemistry (IHC), sections were deparaffinised by 30 min incubation at 65 °C followed by xylene, and decreasing gradients of ethanol (100% to 70%), and rinsed in dH₂O. Antigens were retrieved by boiling in citric buffer (10 mM, pH 6) before endogenous peroxidases were blocked with 3% H₂O₂. Unspecific binding was blocked by incubating sections in serum-free protein block (Aglient Dako, CA, USA) for 1 h in room-temperature (RT) followed by 4 °C overnight incubation with primary antibodies (the used antibodies were directed against platelet endothelial cell adhesion molecule (CD31) microtubule- associated protein-2 (MAP-2), CLDN5, and OCLN, diluted in PBS/0.05% Tween20 (see Table 1). After incubation with the appropriate biotinylated secondary antibodies (Vector Laboratories CA, USA) for 1 h at RT, the staining was enhanced by treatment with Vectastain Elite ABC HRP kit (Vector Laboratories). Finally, sections were dehydrated in gradients of ethanol (70–100%) followed by xylene and mounting in Pertex xylene-based mounting media (Histolab). For fluorescent IHC; deparaffinization, antigen retrieval, blocking, and antibody-incubations were performed as described above before mounting with ProLong Gold Antifade with or without 4′,6-diamidino-2-phenylindole/DAPI (ThermoFisher, MA, USA). Between all staining steps, sections were washed three times with PBS/0.05% Tween20 (except for after blocking). DAB-stained sections were imaged and photographed with a BX60 microscope equipped with a TH4-200 light-source using the cellSens software (Olympus, Tokyo, Japan) and fluorescently stained sections were examined with a Zeiss Axio Imager.Z2 equipped with Colibri 7 LED-light-source and a MRc AcioCam using the ZEN Blue software (Zeiss, Oberkochen, Germany).

CLDN5 immunorecativity was quantified in brightfield micrographs of entire brain sections from all time points after HI. For each image, separate ROI:s were drawn around the left and right hemisphere and the images were, similarly to the EB-studies above, segmented via thresholding into binary images with CLDN5-positive areas marked. CLDN5 immunoreactivity was calculated as a percentage of the entire hemisphere area.

To test correlation of tight-junction protein levels and degree of brain injury, brains, CSF and blood plasma were collected from HI-animals (n = 12) 24 post-HI. The plasma and CSF were analysed for CLDN5 and OCLN with ELISA as described above while the brains were embedded in paraffin and sectioned to assess the brain injury. Grey matter tissue loss in the injured hemisphere was determined in brightfield-micrographs of coronal brain-sections stained for the neuron- and dendrite-marker Microtuble-associated protein 2 (MAP2). The images were analysed in ImageJ by delineating regions of interests encompassing the entire injured or uninjured hemispheres and measuring the MAP2 positive immunoreactivity in each hemisphere by investigators blinded to which groups and animals the images belonged to. The percentage of tissue loss in each level were calculated from the MAP2-positive area with this formula: (MAP2_uninjured − MAP2_injured)/MAP2_uninjured × 100 [47]. In all animals, the analysis was performed at six levels encompassing the entire brain and the mean tissue loss of all levels was used in the correlation analysis.

Statistical analyses were made using GraphPad Prism version 8.00 for Windows (GraphPad Software, CA, USA). We used one-way ANOVA with Dunnett’s multiple comparison test, and Pearson’s correlations. The Benjamin-Hochberg method (FDR 0.1) was used to control for multiple correction problems when multiple t-tests were conducted. Specific tests are stated in each Figure legend. Principal component analysis was made using Qlucore Omics explorer software (Lund, Sweden) where the built-in statistics module was used to test differences between sexes on variables (unpaired t-test). Images were processed in the Fiji build [45] of ImageJ [46], figures were designed in Affinity Photo and Designer (Serif Europe, West Bridgford, United Kingdom). The variance of the data in the text of the results-section is presented as mean ± SD.

In this model of neonatal HI, the combination of left carotid artery ligation and global hypoxia produces brain injury and tissue loss in the left hemisphere [36]. To confirm injury in all animals the activity of caspase-3, a hallmark of apoptosis, was measured in homogenates from both injured (left) and uninjured (right) hemisphere of HI and control animals (Additional file 2). Virtually no caspase-3 activity was detected in either the uninjured hemispheres of HI-animals nor in hemispheres of the controls while a significant increase in caspase-3 activity was seen in the injured hemisphere of HI-animals 6 h after HI compared to the uninjured hemisphere (p = 0.0344) or control animals (p = 0.028). Caspase-3 activity was further increased in the left hemisphere at 24 h after HI compared to 6 h after HI (p = 0.0087), controls (p = 0.0027) as well as the uninjured hemisphere (p = 0.0060). The range of caspase activity for the 6 h and 24 h post-HI groups were 3–77 and 10–950 pmol AMC/min x mg protein, respectively.

Levels of CLDN5, OCLN and ZO-1 were measured in CSF and plasma with ELISA at 6, 24 h and 5 days after HI and all time-points were compared with a control group collected and analysed at the same time and on the same ELISA-plate. CLDN5 and OCLN proteins were detected in all samples with levels ranging between ~ 41 to 1300 pg/ml (plasma), ~ 100 to 2400 pg/ml (CSF) for OCLN and ~ 2 to 30 ng/ml (plasma), ~ 30 to 53 ng/ml (CSF) for CLDN5. Elevated CLDN5-levels were detected in CSF (Fig. 1d) at 24 h post-HI (p = 0.0082), while plasma-concentrations (Fig. 1c) were higher than controls at 6 h (p = 0.0427). There were no difference at later times between HI and controls. Similarly, OCLN concentration in CSF (Fig. 1b) was raised at 6 h (p = 0.0026) after HI while the levels were higher in plasma (Fig. 1a) of HI-animals at 24 h (p = 0.0285). Measured values (mean ± SD) for CLDN5 and OCLN in blood plasma and CSF are available in Additional file 3. ZO-1 was only detectable in 47% of CSF samples. We found great variably in levels between plasma with as high as 2000 pg/mL in one animal but no differences were measured across groups at any time point after HI (Additional file 4).

Principal component analysis (PCA) was performed (dimension reducing) as means of discriminate analysis with the input of all above measured variables for OCLN and CLDN5. HI and control animals were grouped into three distinct groups at 6 h and 24 h after HI (Fig. 2a) whereas at 5 days post-HI (Fig. 2b), no clustering of control vs HI animals was apparent. To test effect of sex on levels of TJ-proteins we performed unpaired t-test between male and female animals pooling all samples within all injured and within all control animals, adjusting for time as a factor. Male rats (n = 10) showed significantly higher levels of OCLN in plasma (p = 0.035) and CLDN5 in CSF (p = 0.036) than female rats (n = 11), control animals (males n = 7, females n = 9) had no sex-differences in plasma OCLN (p = 0.91) or CSF CLDN5 (p = 0.19) (Fig. 2c). In HI-animals, the average amount of OCLN in plasma were ~ 700 pg/ml for males and ~ 400 pg/ml for females, for CLDN5 in CSF the numbers were ~ 36 ng/ml for males and ~ 26 ng/ml for females.

To determine changes in BBB function over time after HI we performed measurements of BBB-permeability in different brain regions (i.e. hippocampus, cortex, and striatum/thalamus) by quantifying the permeability for ¹⁴C-labelled sucrose across the BBB. We previously showed [15] that the BBB in the uninjured hemisphere is not altered. We confirmed this (Additional file 5) for the 6 h (controls n = 5, HI n = 5) and 5-day time-points (controls n = 5, HI n = 5) corroborating with caspase-3 activation (Additional file 2). Increased BBB permeability occurred in the ipsilateral hemisphere at 6 h post-HI in the cortex (1.10 ± 0.05, p = 0.0007), hippocampus (1.07 ± 0.08, p = 0.0277), and striatum/thalamus (1.05 ± 0.05, p = 0.0111) and was also significantly higher at 24 h in both hippocampus (1.12 ± 0.12, p = 0.0486) and cortex (1.18 ± 0.28, p = 0.0474) while the striatum/thalamus appeared unaltered (1.01 ± 0.07, p = 0.7624) (Fig. 3a). Mentionable is that these concentration ratios are probably somewhat affected by the edema occurring in the injured hemisphere after the insult suggesting that the magnitude of BBB permeability increase is likely marginally higher than what these ratios reflect. The concentration ratios 5d post-HI was significantly lower in the injured cortex (0.88 ± 0.12, p = 0.0191) as well as in the entire injured hemispheres (0.91 ± 0.10, p = 0.0303) (Fig. 3b).

BBB disruption following HI were confirmed via assessment of EB-bound albumin extravasation into the brain parenchyma in brain sections. Animals injected with EB at 6 h post-HI showed fluorescence in the cortex of the injured hemispheres (Fig. 3d) while the uninjured hemispheres showed weak fluorescence restricted to blood vessels (Fig. 3e). Control animals injected with EB and not perfused showed distinct blood vessels-restricted fluorescence (Fig. 3f) while animals not injected with EB showed no signal. Extravasated EB encompassed 51.1 ± 8.4% of the total area in images from the injured hemisphere, significantly compared to images from the uninjured hemisphere (6.3 ± 1.9%, p = 0.0002) and controls (5.8 ± 1.6%, p = 0.0002) where the EB was restricted to blood vessels.

Since the cerebrovascular area, the effective surface area for exchange between blood and brain, can affect measurements of BBB-permeability we developed an in-house written macro for blood vessel analysis of CLDN5 (as a vascular marker) immunolabelled sections. We specifically wanted to estimate cerebrovascular area at later times after injury since there is loss of brain tissue and potentially blood vessels. Sections from brains 5 days post-HI were used to calculate the area of blood vessels in the brain. Two levels were analysed per animal and results averaged (Fig. 3c). For this analysis, the hippocampus and cortex results were combined in each hemisphere. In control animals 2.01 ± 0.64% of the brain area was comprised of vessels while in HI animals vessels comprised 1.53 ± 0.37% of the uninjured hemisphere and 2.05 ± 0.61% in the injured hemisphere. No significant differences were detected between control animals and either hemisphere of injured animals (p > 0.05).

Given that there are reports of changes in TJ-protein immunoreactivity following hypoxia/ischemia and hypoxia alone [28, 48], we performed double immunofluorescent labelling of CLDN5 (Fig. 4a) together with blood vessel marker CD31 (platelet endothelial cell adhesion molecule) in paraffin brain-sections collected at all time points after HI (time chosen given BBB changes). In control animals, we found robust immunoreactivity of TJ-proteins in vessels in all brain regions examined including the cortex, hippocampus, and striatum/thalamus, while no labelling was detectable in parenchyma of the brain. Likewise, in animals after HI we found immunolabelling of blood vessels across all brain regions including MAP2-negative regions with no apparent changes compared to control animals. CLDN5 immunoreactivity was quantified in DAB-developed sections for CLDN5 only. No significant differences in CLDN5-coverage were seen between HI (n = 9) and control animals (n = 6) (p > 0.05), or between the injured and uninjured hemisphere in HI-animals (p > 0.05), at any time-point (n = 3 per time point) when CLDN5-immunoreactivity was quantified in entire stained brain hemispheres (Fig. 4b).

The correlation between circulating TJ-protein levels and brain injury severity were investigated 24 h after HI by analysing plasma- and CSF-levels of TJ:s and simultaneously quantifying the loss of grey matter in the brain of the same animal. This time-point was chosen based on the earlier ELISA-results (Fig. 1). The average brain tissue loss in the HI-group (n = 12) varied from 24.7 to 59.9% (one representative level is shown in Fig. 5b and d, respectively). By using the Pearson correlation on the levels of circulating TJ-proteins and the tissue loss percentages (Fig. 5a), the levels of CLDN5 in CSF was found to significantly (p = 0.016) correlate with the severity of the grey matter tissue loss (r = 0.702). No correlation was observed between the loss of grey matter and levels of CLDN5 in plasma (r = -− 0.001) nor OCLN in CSF (r = 0.060) or plasma (r = 0.039).