Blood-Brain Barrier Breakdown after Embolic Stroke in Rats Occurs without Ultrastructural Evidence for Disrupting Tight Junctions

The term blood-brain barrier (BBB) relates to the ability of cerebral vessels to hold back hydrophilic and large molecules from entering the brain, thereby crucially contributing to brain homeostasis. In fact, experimental opening of endothelial tight junctions causes a breakdown of the BBB evidenced as for instance by albumin leakage. This and similar observations led to the conclusion that BBB breakdown is predominantly mediated by damage to tight junction complexes, but evidentiary ultrastructural data are rare. Since functional deficits of the BBB contribute to an increased risk of hemorrhagic transformation and brain edema after stroke, which both critically impact on the clinical outcome, we studied the mechanism of BBB breakdown using an embolic model of focal cerebral ischemia in Wistar rats to closely mimic the essential human pathophysiology. Ischemia-induced BBB breakdown was detected using intravenous injection of FITC-albumin and tight junctions in areas of FITC-albumin extravasation were subsequently studied using fluorescence and electron microscopy. Against our expectation, 25 hours after ischemia induction the morphology of tight junction complexes (identified ultrastructurally and using antibodies against the transcellular proteins occludin and claudin-5) appeared to be regularly maintained in regions where FITC-albumin massively leaked into the neuropil. Furthermore, occludin signals along pan-laminin-labeled vessels in the affected hemisphere equaled the non-affected contralateral side (ratio: 0.966 vs. 0.963; P = 0.500). Additional ultrastructural analyses at 5 and 25 h after ischemia induction clearly indicated FITC-albumin extravasation around vessels with intact tight junctions, while the endothelium exhibited enhanced transendothelial vesicle trafficking and signs of degeneration. Thus, BBB breakdown and leakage of FITC-albumin cannot be correlated with staining patterns for common tight junction proteins alone. Understanding the mechanisms causing functional endothelial alterations and endothelial damage is likely to provide novel protective targets in stroke.


Introduction
Ischemic stroke still represents one of the leading causes of death worldwide [1]. On the cellular level, the ischemia-induced breakdown of the blood-brain barrier (BBB), which is known to contribute to an increased risk of hemorrhagic transformation and brain edema with perilous outcome, was often linked to a putative disruption of tight junctions in the endothelial layer of cerebral vessels [2]. However, this putative mechanism of breakdown has never been convincingly proven by ultrastructural analyses in conjunction with physiological conditions.
Originally, the observation that cerebral vessels hold back hydrophilic molecules from entering the brain was first described by Paul Ehrlich in 1885 [3]. In 1900, Max Lewandowski drew the revolutionary conclusion that the morphological correlate of this barrier function indeed must be the capillary wall [4], which was further specified by the groundbreaking paper from Reese and Karnovsky in 1967, who linked this property to belts of endothelial tight junctions in cerebral capillaries [5]. These complexes consist of three major transmembrane protein families comprising occludin, claudins and junction associated proteins (JAMs), as well as several cytoplasmic proteins including the zonula occludens (ZO) protein family (ZO-1, ZO-2 and ZO-3), providing the link to the actin cytoskeleton [2,6,7]. Whereas this general composition is also known for epithelial tight junctions of peripheral organs, some proteins were found to be specific for brain endothelial cells, such as cingulin, claudin-3 and -5 [8,9].
Under physiological conditions cerebral endothelial cells ensure the outstanding cerebral barrier function, which is maintained by the critical interaction of pericytes, astrocytes, neurons and the extracellular matrix as parts of the neurovascular unit (NVU) [7]. In this context, astrocytes were originally thought to induce BBB characteristics in cerebral endothelial cells [10][11][12][13], but also pericytes were shown to enhance their barrier function [14,15]. In the concert of cells constituting the NVU the establishment and maintenance of the BBB represents a permanently regulated process, which begins in early development and lasts throughout ageing [16][17][18]. The criticial impact of proper endothelial function is exemplarily demonstrated by experimental disruption of endothelial tight junctions, which resulted in generation of epileptiform activity by failure of cellular homeostasis in the NVU [19].
During ischemic stroke, alterations in cerebral blood flow by acute vessel occlusion result in energy failure of the NVU, thereby causing mitochondrial damage, release of toxic oxygen species and intracellular edema, which critically impact on the clinical outcome [20][21][22][23][24]. Furthermore, upregulation of the matrix metalloproteinases (MMP)-2 and -9, as well as of the vascular endothelial growth factor (VEGF) are closely associated with BBB disruption [25,26]. In the past, this event was predominantly linked to tight junction failure, consequently leading to an extravasation of blood-borne molecules along the paracellular route [2,27,28]. Indeed, a variety of studies reported on alterations in the expression or localization of critical tight junction proteins, such as occludin, claudin-5 and ZO-1 in diverse models of in vitro hypoxia as well as of global and focal cerebral ischemia in rodents (e.g., [26][27][28][29][30]). Despite these efforts, a conclusive concept of BBB breakdown during ischemia is currently lacking. The present situation is thus characterized by a discrepancy between the well known clinical impact of the BBB integrity during ischemic stroke and its poorly understood mechanisms, which might impede further therapeutic interventions [31,32]. Concerning translational issues, the used tracers for visualizing BBB leakage were recently accounted for the risk to inherit a methodological bias due to own intrinsic effects. For instance, Evans blue as probably the most frequently applied agent [25,[33][34][35][36][37] was found to inhibit the uptake of glutamate [38,39] as one of the key mediators during the ischemic cascade [20]. Furthermore, in reference to failures of more than 1,000 neuroprotective approaches in stroke [40], a critical discussion came up addressing the comparability of routinely used rodent models due to their preponderant artificial techniques for vessel occlusion [41,42].
The present study aimed at the ultrastructural analysis of tight junction integrity after embolic middle cerebral artery occlusion (eMCAO) in rats, which is currently believed to best mimic the human pathophysiology [43]. To investigate the fate of endothelial tight junctions in regions of ischemia-related BBB breakdown, we focused on areas exhibiting fluorescein isothiocyanate (FITC)labeled albumin (FITC-albumin) extravasation. Multiple fluorescence labeling was utilized to localize endothelial tight junction proteins, while the ultrastructural integrity of tight junctions and routes for tracer extravasation were addressed by electron microscopy.

Study setup and surgical procedure
Experiments involving animals were performed according to the European Communities Council Directive (86/609/EEC) after protocol approval by local authorities (Landesdirektion Leipzig, Germany, reference numbers TVV 02/09 and 34/11). Generally, efforts were made to minimize the total number of animals and suffering of animals, which were housed in a temperature (21-22uC) and humidity (45-60%) controlled room with 12 hours of light/dark cycle and free access to food and water.
Overall 13 male Wistar rats weighing 322633 g, bred by Charles River (Sulzfeld, Germany), were subjected to right-sided eMCAO as originally described by Zhang and coworkers [44] with some minor modifications. Briefly, after surgical exposure of right-sided cervical arteries, a polyethylene (PE) tube with tapered end and a maximum outside diameter of 0.4 mm was introduced into the external carotid artery and moved forward through the internal carotid artery up to the origin of the middle cerebral artery (about 1.6 cm from carotid bifurcation). A blood clot (length: 560.5 cm) -prepared from blood collected in a PE tube at the previous day and allowed to clot on a warming pad (37uC) for 2 hours followed by overnight storage at 4uC -was injected with about 40 ml of saline. After catheter removal and ligation of the external carotid artery stump, the wound was closed. PE tubes were further inserted into the femoral vein to allow future administration of FITC-albumin and femoral artery to control vital parameters during surgery. Generally, the surgical procedure was done in anesthesia using 2.0 to 2.5% isoflurane (Isofluran Baxter, Baxter, Unterschleißheim, Germany; mixture: 70% N 2 O/ 30% O 2 ; vaporisator: VIP 3000, Matrix, New York, USA). To avoid anesthesia-related cooling during surgery, the body temperature was adjusted to 37.0uC by a thermostatically controlled warming pad with rectal probe (Fine Science Tools, Heidelberg, Germany). After surgery, animals spent time on a commercial warming pad (37.0uC) until recovery from anesthesia. Sufficient eMCAO induction was ensured by functional assessment using Menzies score (basically ranging from 0 'no apparent deficit' to 4 'spontaneous contralateral circling'; [45]), whereas animals had to show a functional relevant deficit during observation period as indicated by a score of at least 2. Post-surgical pain control was ensured by metamizol-or paracetamol-enriched (Novaminsul fon-ratiopharm, Paracetamol-ratiopharm, ratiopharm, Ulm, Germany) drinking water.
Four or 24 hours after ischemia onset, FITC-albumin (20 mg/ 1 ml saline; Sigma, Taufkirchen, Germany) was intravenously administered to allow fluorescence-based localization of BBB breakdown, indicated by FITC-albumin leakage [46]. After a circulation period of usually 1 hour, animals were deeply anesthetized using CO 2 and transcardially perfused with either 200 ml saline (n = 5, for immunohistological analyses by fluorescence microscopy at 25 hours), or 200 ml saline followed by 200 ml of a phosphate-buffered fixative containing 4% paraformaldehyde and 0.5% glutaraldehyde (n = 4 each, for ultrastructural analyses by electron microscopy at 5 and 25 hours). After extraction of the brains the samples for fluorescence microscopy were immediately snap frozen in isopentane on dry ice.

Fluorescence microscopy and quantification of tight junction integrity
For fluorescence microscopy, the brains were coronally cut in 10 mm sections on a cryotome (Leica Microsystems, Wetzlar, Germany). Before starting the staining procedure, the samples were post-fixed with Zinc-formalin for 5 minutes at room temperature followed by several rinses with 0.1 M phosphatebuffered saline, pH 7.4 (PBS). Unspecific binding of the antibody was inhibited by a blocking step applying 5% of normal goat serum (NGS) and 0.3% of Triton X-100 in PBS for 20 min. The primary antibodies mouse anti-occludin (Antibodies-online, Aachen, Germany; 1:200), rabbit anti-claudin-5 (Abcam, Cambridge,UK; 1:200), rabbit anti-laminin (Dako, Hamburg, Germany; 1:200) and guinea pig anti-GFAP (Synaptic Systems, Göttingen, Germany; 1:200) were allowed to incubate over night at 4uC in PBS containing 0.5% NGS. After several washing steps, the sections were reacted with highly purified fluorochromated antibodies (Dianova, Hamburg, Germany; diluted in PBS containing 0.5% NGS) specifically recognizing IgGs from the host species of primary antibodies for 2 hours. Subsequently, sections were washed several times in PBS and coverslipped with fluorescence mounting medium (Dako, Hamburg, Germany). Omitting the primary antibodies in control experiments resulted in the expected absence of staining. The specimens were analyzed with an Olympus BX51 fluorescence microscope (Olympus, Hamburg, Germany) equipped with a digital camera.
For quantification of tight junction integrity, the ratio of tight junction-positive vessels with respect to the total number of vessels was captured in the area of FITC-albumin leakage, as indicator of BBB breakdown, and compared to the corresponding region on the contralateral hemisphere. Therefore, we applied double immunofluorescence staining for occludin as marker for endothelial tight junctions and pan-laminin as a general marker for cerebral vessels [47]. To ensure exclusive analysis of ischemiaaffected areas on the ipsilateral hemisphere depicting BBB breakdown, we focused on regions showing a clear FITC-albumin extravasation into the neuropil. Low power (106 objective) magnification was used to count the number of occludin-positive vessels in relation to the total number of laminin-immunopositive vessels in usually 8 non-overlapping optic fields per animal and hemisphere, resulting in an amount of about 600 to 1,000 vessels per hemisphere. After calculating the ratio per optic field, a mean of ratios was build per animal and hemisphere. Finally, the global mean of the ipsilateral hemisphere was compared to the contralateral hemisphere, originating from n = 5 animals for each.

Ultrastructural analysis
For electron microscopy the brains were consecutively cut in 60 mm sections on a vibratome (Leica Microsystems, Wetzlar, Germany) in cooled PBS. The immunohistochemical conversion of FITC into an electron-dense reaction product was performed according to our previous study [45]: After several washing steps in 0.1 M Tris-buffered saline, pH 7.4 (TBS), the sections were blocked with TBS containing 2% bovine serum albumin (TBS-BSA) for 30 min and then incubated with peroxidase-conjugated anti-fluorescein IgG (Dianova, Hamburg, Germany; 1:2000 = 0.5 mg/ml) in TBS-BSA for 2 hours. Next, the tissue was rinsed with TBS twice and with the substrate buffer (0.05 M Tris, pH 7.6) followed by staining with diaminobenzidine (DAB) as chromogen.
After several washing steps in PBS the sections were stained with 0.5% osmium tetroxide in PBS for 30 min. After 5 washing steps the sections were dehydrated using 30%, 50% and 70% of ethanol. Subsequently, the tissue was incubated with 1% uranyl acetate in 70% ethanol for 1 hour. Following further dehydration using 80%, 90%, 96%, 100% ethanol and finally propylene oxide the sections were incubated in Durcupan (Sigma Aldrich, Steinheim, Germany) and embedded between coated microscope slides and cover slips followed by polymerization at 56uC for 48 hours. After microscopical identification of areas exhibiting FITC-albumin leakage demarked by DAB staining the cover slips were removed and the respective areas were transferred on blocks of resin followed by another step of polymerization at 56uC for 48 hours. Afterwards, the blocks of resin were trimmed down to the areas intended for ultrastructural analysis which were relocated on semi-thin sections and then ultra-thin sections of 60 nm thickness were prepared on an ultramicrotome (Leica Microsystems, Wetzlar, Germany). Finally, sections were transferred on formvar-coated grids and stained with lead citrate for 6 min. Ultrastructural analysis was performed using a Zeiss EM900 transmission electron microscope and a Zeiss SIGMA electron microscope equipped with a STEM detector, respectively (Zeiss NTS, Oberkochen, Germany).

Statistical analysis
The obtained data on tight junction integrity were processed with IBM SPSS Statistics 20 (IBM Corp., New York, USA). Thereby, the Wilcoxon test was utilized to check for statistical significance between means of ratios. In addition, results were further stressed by adding the Monte Carlo simulation with 10,000 samples and a confidence interval of 99%. Generally, a P,0.05 was considered as statistical significant.

Identification and characterization of areas showing BBB breakdown 25 hours after ischemia onset
According to our previous study [46], the applied model of eMCAO resulted in alterations of the vascular architecture with a clear extravasation of FITC-albumin in respective areas, primarily located in the right striatum. In contrast, BBB breakdown was not observed on the contralateral hemisphere as the respective areas here were devoid of FITC-albumin leakage. To distinguish leakage of the tracer into the vascular wall or adjacent perivascular spaces from leakage into the neuropil, we applied double fluorescence labeling of vascular basement membranes by combined immunostaining for laminin and astroglial GFAP (Fig. 1A).
For ultrastructural analysis and identification of areas showing FITC-albumin leakage, we made use of a peroxidase-coupled antifluorescein IgG and DAB, which served as permanent and electron dense chromogen. To prove the specificity of this technique on glutaraldehyde-treated sections for electron microscopy, control stainings were performed on consecutive vibratome sections. The specificity of the staining was proven by restriction of the DAB signal to the right striatum affected by eMCAO and the absence of any labeling on the contralateral hemisphere (Fig. 1B). To precisely focus further ultrastructural analysis to areas exhibiting extravasation of the tracer, respective areas were identified by light microscopy after embedding of the tissue into resin (Fig. 1C). Importantly, this protocol allowed confinement of ultrastructural analysis to regions of BBB breakdown and their corresponding control areas.

Expression of critical tight junction proteins in areas of FITC-albumin extravasation
To further assess whether belts of tight junctions are altered in areas of FITC-albumin leakage critical tight junction constituents were analyzed by double fluorescence labeling. In contrast to our own expectations, visualization of the essential transmembrane tight junction constituents occludin and claudin-5 revealed presence and regular morphology of both antigens in control areas as well as in areas of BBB breakdown (Fig. 2). Presence of established and still continuous belts of tight junctions in vessels showing bright signals of FITC-albumin in the perivascular and juxtavascular area suggests that extravasation of FITC-albumin does not necessarily depend on changes of staining patterns reflecting a decomposition of endothelial tight junctions in the applied model of eMCAO in rats.
No difference in the distribution of occludin-positive vessels in areas of FITC-albumin leakage compared to control areas To investigate the question of whether or not there is a statistical significant difference in the expression of critical tight junction proteins between areas of FITC-albumin extravasation and respective contralateral regions, double fluorescence labeling was applied by use of an antibody for laminin as a pan-vessel marker and the critical transmembrane protein occludin to demark tight junction complexes. Using low power (106 objective) magnification localization of the tracer beyond the glial basement membrane ensured precise confinement of analysis to areas showing distinct FITC-albumin leakage into the brain parenchyma. Here, we compared the number of occludin-and lamininpositive vessels in areas of BBB breakdown and their corresponding control areas (Fig. 3A). As shown in Fig. 3B, nearly identical ratios were found for the ischemia-affected (0.96660.010) and the contralateral hemisphere (0.96360.003), impressively failing statistical significance (Wilcoxon test, P = 0.500; Monte Carlo simulation, P = 0.630).

Ultrastructural evidence for vascular leakage in areas of morphologically intact tight junctions
Due to apparent limitations of fluorescence microscopy to elucidate minor changes of tight junction complexes, electron microscopy was applied to reveal ultrastructural alterations responsible for vascular leakage. Using flat-embedded vibratome sections stained with DAB to demark the applied tracer, we were able to precisely identify regions of BBB breakdown and their corresponding control areas by initial light microscopy (Fig. 1C). On the contralateral hemisphere, ultrastructural analysis elucidated intact belts of tight junctions within a smooth endothelial layer showing no ultrastructural alterations. The ensheathing basement membrane was found to be continuous and the adjacent neuropil appeared unaffected (Fig. 4A). In areas of FITC-albumin extravasation, the juxtavascular parenchyma often displayed edema of astrocytic endfeet, in conformity with Ito and coworkers [48], and also cellular debris. On the abluminal side of the vascular basement membrane DAB grains regularly indicated extravasation of the applied tracer. However, in these segments of the vascular tree, the tight junctions were still found to be intact, thus sealing the endothelial layer as shown by fluorescence microscopy before (Fig. 4B-D). and GFAP (red) in combination with applied FITCalbumin (green) reveals areas of ischemia-related BBB breakdown. By application of both, an antibody detecting astrocytes (GFAP, red) and an antibody for pan-laminin (blue) which visualizes vascular as well as glial basement membranes [47] the observed leakage can clearly be demonstrated to reach the brain parenchyma proper. Therefore, FITCalbumin is detectable within all the three compartments of the neurovascular unit, represented by the vascular wall (1 st compartment), the perivascular space (2 nd compartment) and the adjacent neuropil (3 rd compartment), delineated by astocytic endfeet (red) forming the glia limitans [54]. (B) To prove specificity of the applied reagent used for conversion of extravasated FITC-albumin into a permanent labeling by DAB for ultrastructural analysis, we exemplarily performed control stainings on vibratome sections, which were not further processed for electron microscopy. The low magnification of a coronary section clearly demonstrates the specificity of the applied reagent and its reaction product (DAB, brown). These sections were counterstained with hemalaun (blue). Areas of FITC-albumin leakage are clearly confined to the striatum. Higher magnification reveals a general leakage of FITC-albumin into the neuropil (brown background). In perivascular and juxtavascular areas the DAB staining appears to be more intense. (C) To confine ultrastructural analysis to areas with BBB breakdown, we identified areas of FITC-albumin extravasation after embedding in resin on coated microscope slides. These areas were selectively processed for ultrastructural analysis by electron microscopy. In corresponding control areas no FITC-albumin extravasation was observed. Please note, the general brown tissue background is a consequence of the embedding procedure using osmium tetroxide and uranyl acetate, which clearly can be distinguished from DAB staining as indicated on the left. doi:10.1371/journal.pone.0056419.g001 Ultrastructural evidence for transcellullar, not paracellular leakage As evidence for leakage through altered endothelial tight junctions was lacking, a more detailed comparison between the vasculature on the control hemisphere and areas of FITC-albumin leakage on the ipsilateral hemisphere was performed to search for signs of transcellular leakage. In control areas, the endothelium appeared not to be affected by the experimental procedure. The nuclei exhibited the typical distribution pattern of euchromatin and heterochromatin, but hardly any transcellular vesicles. Adjacent cells within the neurovascular unit appeared normal and extravasation of the tracer was not observed (Fig. 5A). In contrast, endothelial cells within areas of BBB breakdown often showed a ruffled vascular surface with dramatically increased numbers of cytoplasmic vesicles. As described above, tight junction complexes concomitantly appeared to be intact (Fig. 5B). Around vessels showing the described alterations grains of DAB were regularly observed in adjacent perivascular spaces or within the neuropil. Therefore, we propose that the contribution of leaky belts of tight junctions to the failure of the endothelial BBB characteristics is rather questionable in the model of eMCAO. More likely, it is the result of a failure in suppressing vesiclemediated leakage through the endothelium itself.

Evidence for leakage across disintegrated endothelium
In addition to the finding of an enhanced vesicle transport through the endothelial layer, a disintegration of endothelial cells was often observed allowing direct extravasation of FITC-albumin into perivascular spaces and the adjacent neuropil. Here, the luminal plasma membrane frequently appeared to be discontinuous. Thus, the former barrier for blood-borne molecules towards the brain parenchyma proper is often constituted by endothelial debris and remnants of the vascular basement membrane only (Fig. 6).

Equal ultrastructural conditions in the early stage after ischemia onset
Finally, we applied electron microscopy at 5 hours after ischemia onset to capture the condition in this early stage of ischemia, and to allow comparison to the previously described 25hour time point. Thereby, an equal situation was noted, clearly demonstrating regularly established tight junctions in the presence of endothelial damage and FITC-albumin leakage, as well as enhanced vesicle trafficking within the morphologically altered endothelium (Fig. 7).

Discussion
Despite their enormous clinical relevance, the underlying mechanisms for ischemia-induced brain edema and hemorrhagic transformation at the endothelial level are still poorly understood. Consequently, the regulation of BBB integrity has achieved high priority rating in today's stroke research [31]. In the past, promising concepts of stroke treatment developed by experimental research using different animal models often failed to be transferred into the clinical setting, such as NXY-059 [42,49]. Since the ways to reproducibly mimic conditions of cerebral ischemia in experimental models are limited, investigators are often forced to make use of artificial and unphysiological ways to decrease blood flow in respective cerebral vessels [41]. Thus, frequently used techniques vary from electrocoagulation or ligation to the internal occlusion of the vascular lumen by inserted filaments [50]. Furthermore, the applied tracers to investigate a putative leakage of the endothelial barrier such as dextrans [51,52] or the widely utilized Evans blue [25,[34][35][36][37] generally comprise the risk to produce bias due to immunological side effects or modulation of ischemic key mediators [38,39,53], thereby disturbing the physiological environment of the NVU and its constituents. Moreover, investigation of cellular mechanisms and intracellular signaling cascades predominantly requires use of in vitro experiments, which are useful to study endothelial function, but not pathophysiological events of the complete NVU including pericytes, astrocytes, neurons and components of the extracellular matrix, which act in concert to maintain the barrier function of the cerebral vasculature [7,23,[54][55][56].
In order to minimize artificial alterations, we used an embolic model of ischemic stroke in rats, which is believed to best mimic the human pathophysiology [43]. Furthermore, BBB leakage was assessed by injected FITC-labeled albumin which is described to show the same pattern of extravasation as compared to traditional tracers, such as rat IgG with nearly double molecular weight [46]. By limiting the circulation period of FITC-albumin to 60 minutes prior to our observation time points at 5 and 25 hours after ischemia onset, we made sure to precisely capture BBB alterations in this distinct time window. Although ischemia-related BBB breakdown is reported to follow a complex and dynamic pattern, detection of FITC-albumin leakage assured investigation of affected areas in a temporally and locally specific manner. Thereby, FITC-albumin enabled a clear-cut identification of areas with BBB breakdown as known precondition for brain edema and hemorrhage in the clinical situation [57], and the application of double fluorescence labeling for transmembrane tight junction constituents allowed simultaneous analyses of endothelial belts of tight junctions.
Although alterations of tight junctions have been described in several models of hypoxia and stroke before [26,28,30,58,59], we were not able to correlate areas of FITC-albumin leakage with alterations in the expression pattern of the transmembrane proteins occludin and claudin-5 in this model of focal cerebral ischemia (Fig. 2). This finding was confirmed by a quantitative analysis based on the presence of essential tight junction proteins in areas of FITC-albumin extravasation compared to control regions on the contralateral hemisphere. In detail, immunolabeling of the transmembrane protein occludin demarked endothelial belts of tight junctions and laminin-immunoreactivity served as a vascular basement membrane marker [47], allowing a ratio of occludin-positive vessels to be calculated. Importantly, these ratios did not differ when comparing areas of FITC-albumin leakage and the respective control regions (Fig. 3), indicating not only a lack of qualitative but also of quantitative changes.
To rule out that possible alterations of the endothelial tight junctions remain undetected due to limitations of fluorescence microscopy, electron microscopy was applied to analyze ultrastructural changes within the endothelial layer. As previously described [46], we used the nearly background-free immunohistochemical conversion of FITC-albumin by peroxidase-coupled anti-fluorescein into the electron-dense chromogen DAB to specifically identify areas of BBB breakdown. In fact, by embedding of DAB stained vibratome sections under coated cover glasses, we were able to identify respective areas and the exact Figure 6. Evidence for leakage across structurally altered endothelium. In addition to the observation of a dramatic increase of the vesicle density, we often observed a disintegration of the whole endothelial layer. In areas exhibiting BBB breakdown the cellular surface of the endothelium was frequently found to be ruffled or discontinuous. Therefore, the vascular wall was often shown to consist of endothelial debris and basement membranes, only (brackets). Thus, in contradiction to a variety of studies our data strongly suggest a transendothelial leakage pattern of affected vessels. DAB grains indicating extravasation of FITC-albumin are demarked by arrow heads. E = endothelial cells; L = vascular lumen; asterisk = cellular debris in the lumen of the vessel; arrow heads = DAB grains; arrow = tight junction. doi:10.1371/journal.pone.0056419.g006 location on the contralateral hemisphere before preparation of ultra-thin sections for electron microscopy. Conversion of FITCalbumin into DAB ultrastructurally resulted in a typical granular staining, which was strictly confined to sites of extravasation of the tracer. In line with our findings based on fluorescence microscopy, at both 5 and 25 hours after ischemia onset the tight junction complexes were regularly established although DAB grains clearly indicated extravasation into perivascular spaces or the adjacent neuropil (Fig. 3). These results are clearly contradicting studies describing ultrastructural alterations of tight junctions in diverse models of hypoxia and stroke [28,60]. However, the traditional link between presence of tight junctions and BBB integrity as well as the variety of valuable data demonstrating altered expression patterns of tight junction proteins in models of hypoxia and stroke [26,[27][28][29][30]59] may have constituted a bias affecting interpretation of ultrastructural images. Thus, questionable alterations were addressed as intercellular openings or tight junction loss [28,60]. On the other hand, previous data also showed molecular alterations of tight junction proteins, mediated as for instance via changes in the extra-/intracellular calcium level or phosphorylation of proteins, finally leading to variant functional characteristics [6,9]. Moreover, concerning molecular formation a recent report identified changes of the extracellular loop 2 of claudin-5 as underlying mechanism for different phenotypes of tight junctions [61]. Taking these findings into consideration, there might be ischemia-related changes in the organization of tight junctions which are not detectable by fluorescence labeling or electron microscopy. This perspective is supported by a previous study showing paracellular tracer leakage but no change in tight junction-associated proteins in an in vitro model of endothelial cells [62]. Indeed, the dramatic endothelial damage observed in this study powerfully illustrates that staining patterns of typically used tight junction proteins cannot be correlated with BBB integrity.
However, Reese and Karnovsky who first described endothelial tight junctions as 'morphological correlate of a BBB' also Figure 7. Evidence for FITC-albumin leakage across disintegrated endothelium in the early stroke phase. Finally, the same pattern of extravasation and endothelial damage was found at 5 hours after ischemia induction (A-D), as shown at 25 hours before. While the tight junctions remain, the rest of the endothelial cells may be constituted of debris, only (A and B). Often, the vascular basement membrane is exposed to the vascular lumen (B). Furthermore, electron dense vesicles carrying FITC-albumin can often be observed in the endothelium and the adjacent basement membrane (C) where the content is found to be deployed (D). E = endothelial cells; L = vascular lumen; arrow heads = DAB-filled vesicles; arrow = tight junction. doi:10.1371/journal.pone.0056419.g007 considered in the same paper 'paucity of transcytotic vesicles' as a mechanism to prevent penetration of hydrophilic molecules into the brain [5]. It is interesting to note that an early study from Ito and coworkers in 1980 investigated the effect of hypertension after cerebral ischemia, and thereby, already suggested an upregulation of the transcellular vesicle transport to result in a transendothelial leakage pattern [63]. Moreover, the concept of a predominant transendothelial route was also mentioned by Dietrich and coworkers in 1987 in a model of photochemically induced cerebral ischemia [64]. However, this reasonable perspective has been widely neglected in the literature.
As we were also not able to demonstrate loss of tight junctions, we further analyzed the areas of tracer extravasation for differences in the distribution of transcytotic and pinocytotic vesicles compared to respective control areas. Indeed, we found a remarkable increase of these vesicles in areas of FITC-albumin leakage while control areas exhibited no or scarcely any vesicles at all (Fig. 5). Furthermore and supporting the hypothesis of a transcellular leakage, we frequently observed a distintegration of the endothelial layer itself, thereby allowing unhindered extravasation of blood-borne molecules at different time points (Fig. 6 and Fig. 7). Given that the alterations revealed in this study were that striking, it is reasonable for us to assume that they may be primarily responsible for the observed BBB breakdown after ischemic stroke. Therefore, our study provides robust evidence that the pathophysiological mechanism leading to hemorrhagic transformation and brain edema in stroke does not necessarily depend on ruptures of endothelial tight junctions in areas of ischemia-related BBB dysfunction. We rather suggest, that in addition to the established paracellular leakage pattern, an endothelial dysfunction leading to an extravasation over the whole surface of affected endothelial cells is likely to critically contribute to the formation of brain edema and its well-known clinical implications. This view can be substantiated by recent data proposing that enhanced vesicle trafficking may be related to an upregulation of caveolin-1 under hypoxic conditions [65]. Thus, effects on the expression of tight junction constituents do not necessarily represent the cause for BBB breakdown but may rather be a consequence of a dysfunctioned endothelial layer which allows transcellular leakage. Although our data need to be complemented by future studies focusing on time points beyond the first day after ischemia onset, since BBB alterations were described to maintain up to days or weeks [66,67], the present findings might contribute to a better understanding of the NVU under ischemic conditions. In this context, our study may demonstrate that targeting tight junctions alone is inappropriate to address BBB function. Furthermore, the obtained results of a predominantly dysfunctional endothelium may pave the way towards novel neuroprotective strategies by targeting the endothelial layer.