Introduction

Ischemic stroke ranks among the most common causes of mortality and adult disability in industrialized countries [38]. Astrup et al. [5] observed in 1977 that the reduction of cerebral blood flow to under a critical limit created dysfunctional regions in the brain, as the occlusion of a cerebral artery causes the abrupt deprivation of oxygen and nutrients leading to impaired ionic balance, neuronal depolarization, and massive cell death [1, 27, 43]. The region in which blood flow drops to below 20 % of preocclusion values is designated as the core of the lesion. Basic tissue metabolism cannot be maintained and cell loss is fast and irreversible [21]. In the surrounding penumbra, blood flow ranges between 25 and 50 % of preocclusion values [17, 20]. The damaged tissue there remains in a critical state for hours and days, during which a further reduction of blood flow leads to cell death but an increase to functional recovery and repair [39]. Consequently, an evolvement of tissue damage in the ischemic penumbra is observed, which can be separated in three phases. First, the acute phase, which takes place in the first minutes upon stroke and is characterized by massive cell loss in the core region. Subsequently, the subacute phase lasts for about 6–24 h and is characterized by an expansion of tissue damage from the core into the surrounding healthy tissue of the penumbra. Finally, the delayed injury phase, which lasts for days to weeks and is characterized by secondary phenomena, such as vasogenic edema and inflammation [21]. The temporal and spatial progression of ischemic cell damage and its dependence on blood flow suggest that modulators of blood vessel integrity affect the magnitude of damaged tissue and the speed at which damage occurs.

Key elements in stroke are brain resident mononuclear phagocytes termed microglia, which are part of the innate immune system and are considered as orchestrators of the inflammatory response in the CNS [61]. Under physiological conditions these cells exhibit a small cell body bearing thin and ramified processes that are continuously palpating the brain parenchyma [9, 51]. Upon ischemic stroke microglia become activated and adopt a de-ramified morphology with retracted processes and enlarged cell bodies. Several studies have demonstrated the involvement of microglia in brain ischemia [22, 23] but their contribution to the progression of stroke remains under debate. The transplantation of microglia into a lesioned area reduced the infarct volume and improved functional recovery [28, 29]; however, less invasive loss-of-microglia-function studies such as the genetic knock-out of CX3CR1 led to a reduction of the infarct volume by 56 % [12].

Putatively, the analysis of different microglia subpopulations and a diverging impact on the breakdown of the blood–brain barrier (BBB) [67] cause these antagonistic observations, and in order to explore the mechanism, dynamics, and the orchestration of microglia-dependent events in ischemic stroke, we performed kinetic investigations and intravital two-photon laser-scanning microscopy of CX3CR1+/GFP mice subjected to reversible middle cerebral artery occlusion (MCAO). In a time frame corresponding to the subacute to early delayed phase of stroke, we observed the engulfment of blood vessels by microglia in the penumbra, which culminated in the phagocytic degradation of blood vessels and the active breakdown of the BBB leading to the conclusion that microglia actively participate in the transfer of stroke-induced injury in healthy neighboring tissue by the disassembly of blood vessels and the resulting decrease in blood flow in the ischemic penumbra.

Materials and methods

Middle cerebral artery occlusion (MCAO)

MCAO was performed by application of the intraluminal suture technique using C57BL/6J or B6.129P-CX3CR1tm1Litt/J mice (The Jackson Laboratory, Bar Harbor, ME, USA), referred to in the text as CX3CR1+/GFP (heterozygous, no phenotype) or CX3CR1GFP/GFP (homozygous, CX3CR1 knock-out) mice, respectively. Previously, C57BL/6J and heterozygous animals have been shown to exhibit similar infarct sizes [13]. Briefly, after a midline incision at the neck, a filament with a silicon-coated tip (Doccol, Sharon, MA, USA) was advanced from the external carotid artery into the lumen of the internal carotid artery until it blocked the origin of the left middle cerebral artery. MCAO was verified by a drop in laser Doppler flow (Perimed, Järfälla, Sweden) of at least 80 % of the baseline. Sham animals were obtained by inserting the filament into the internal carotid artery without insertion into the middle cerebral artery. The filament was removed 1 h after MCAO in order to allow reperfusion. Animal procedures were performed under the supervision of authorized investigators in accordance with the European Union normative for care and use of experimental animals. Mice were anesthetized by isoflurane (AbbVie, Ludwigshafen, Germany) applied in oxygen (3 l/min) and received a subcutaneous injection of Rimadyl (Pfizer, Berlin, Germany) as an analgesic. Body temperature was maintained at 37.5 ± 0.5 °C by a homeothermic blanket. Finally, animals were placed for 2 h in a recovery chamber maintained at 33 °C after subcutaneous administration of 500 µL of 0.9 % saline. Only animals that exhibited a score ≥1 in neurological evaluations [36] were used for experimental analyses.

Intravital microscopy

Mice were anesthetized with 1.5 % isoflurane (AbbVie) applied in oxygen/nitrous oxide (2:1, 3 l/min) using a face mask and then tracheotomized and continuously respirated with a Harvard Apparatus advanced safety respirator (Hugo Sachs, March-Hugstetten, Germany). The cerebral cortex was exposed by careful removal of the skull bone of the left hemisphere. The brain was superfused with isotonic Ringer solution that was continuously exchanged by a peristaltic pump. An agarose patch (0.5 % in 0.9 % NaCl solution) was installed on the exposed brain surface to reduce heartbeat and breathing artifacts. The depth of anesthesia was controlled by continuous CO2 measurements of exhaled gas and recorded with a CI-240 Microcapnograph (Columbus Instruments, Columbus, OH, USA). Prior to preparation the vasculature was stained by intravenous injection of rhodamine–dextran 1.5 h before imaging (Sigma, Hamburg, Germany). Experiments were performed using a specialized two-photon laser-scanning microscope (TriMscope; LaVision BioTec, Bielefeld, Germany) equipped with a Ti:Sa laser (MaiTai, SpectraPhysics, Darmstadt, Germany) set at 850 nm and a synchronously pumped optical parametric oscillator (APE, Berlin, Germany) set at 1,100 nm for simultaneous imaging of two colors. XYZ stacks were typically collected within a scan field of 300 × 300 μm at 512 × 512 pixel resolution and a z-plane distance of 2 μm at a frequency of 400 or 800 Hz. The penumbra region in two-photon imaging studies was defined by preliminary MCAO studies (ligation time 1 h, n = 16). One day after reperfusion mice were killed and the location and size of the infarct analyzed by cresyl violet staining. MCAO yielded reproducible results and allowed prediction of the localization of core and penumbra of stroke. Further, mice were scored for motor deficits and only animals that exhibited a score ≥1 in neurological evaluations [36] were used for two-photon imaging.

Immunofluorescence

Mice were anesthetized with ketamine (Ratiopharm, Ulm, Germany) plus xylazine (Bayer Vital, Leverkusen, Germany) and were killed by transcardial perfusion with 4 % paraformaldehyde (PFA) at 1, 3, 6, 12, 24, and 72 h post reperfusion. These mice received an intravenous injection of 100 µl of 20 mg/ml Evans blue (Sigma) 30 min before the perfusion. Brains were removed, successively transferred into 15 and 30 % sucrose solution, frozen in cooled isopentane (Applichem, Gatersleben, Germany), and subsequently cryosectioned. Brain sections (20 µm) were incubated overnight with rabbit anti-caveolin-1 (1:250, clone D46G3, Cell Signaling Technology, Cambridge, UK), rat anti-CD31 (clone MEC 13.3, 1:150, BD Biosciences, San Jose, CA, USA), rat anti-CD31/PECAM1 (1:100, Dianova, Hamburg, Germany), rabbit anti-CD68 (1:200, AbD Serotec, Oxford, UK), mouse anti-claudin-5 (1:200, Invitrogen, Carlsbad, CA, USA), rabbit anti-Glut1 (1:100, Merck Millipore, Darmstadt, Germany), goat anti-EGFL7 (1:100, Santa Cruz, La Jolla, CA, USA), rabbit anti-Iba-1 (1:1,000, Wako Chemicals, Richmond, VA, USA), rat anti-ICAM-1 (1:100, clone YN1/1.7.4, BioLegend, San Diego, CA, USA), mouse anti-MAP2 (1:750, Sigma), rat anti-neutrophil (imp-R14, 1:250, Abcam, Cambridge, UK), mouse anti-NF160 (1:300, Sigma), mouse anti-NG2 (1:500, Merck Millipore), rabbit anti-PDGF-B (kindly provided by Carl-Henrik Heldin, LICR, Uppsala University, Sweden), goat anti-podocalyxin (1:200, R&D Systems, Minneapolis, MN, USA), mouse anti-SMA (1:200, Sigma), rabbit anti-VEGF (1:100, Abcam), or rabbit anti-vWF (1:1,000, Dako, Hamburg, Germany) primary antibodies followed by incubation with Alexa Fluor 488-conjugated goat anti-rabbit, Alexa Fluor 568-conjugated goat anti-mouse, anti-rabbit or anti-rat, Alexa Fluor 647-conjugated goat anti-rabbit (Invitrogen), or Alexa 594-conjugated donkey anti-goat (Jackson, Baltimore, PA, USA) secondary antibodies at a dilution of 1:1,000. Cell nuclei were counterstained with 0.5 µg/ml DAPI (Sigma) for 5 min at RT. Images were captured using an SPE confocal microscope (Leica, Mannheim, Germany). 3D reconstruction and analysis were performed using Imaris 8 (Bitplane, Zurich, Switzerland) or ImageJ software v1.41 (National Institute of Health, Bethesda, MD, USA).

In order to analyze BBB breakdown and blood leakage, murine IgG in the brain parenchyma was measured in the cortical stroke region. Deeply anesthetized mice were perfused with 4 % PFA and brain sections prepared on a vibratome (Thermo). Slides were incubated overnight with Alexa Fluor 488-conjugated goat anti-mouse antibody (1:250; Invitrogen) to stain for murine IgG. The next day neurons were identified by rabbit anti-NeuN (1:500; Abcam) as a primary and Alexa Fluor 568-conjugated goat anti-rabbit as a secondary antibody (1:1,000, Invitrogen). Slides were mounted in Immu-Mount (Thermo) and images obtained by an inverted TCS-5 confocal laser-scanning microscope (Leica).

The association of microglia with blood vessels was quantified in fluorescence microscopy images using the Leica Microscope Imaging Software by counting and comparing cells associated and not associated with blood vessels. In order to measure endocytosis of blood vessel components by microglia, immunohistochemical colocalization of the blood vessel marker CD31 and the microglia marker Iba-1 was quantified using the Imaris 8 software (Bitplane). Thresholds were chosen according to signal intensities and analyzed data represent the percentage of the signal that colocalized above the threshold for each time point.

Magnetic resonance imaging (MRI)

CX3CR1GFP/GFP and wild-type mice underwent MCAO and were killed 6 or 24 h post reperfusion by transcardial perfusion with 4 % PFA solution. The MR contrast agent Gadovist (Bayer) was injected by intravenous injection at 0.2 mmol/kg in all animals, 15 min before being killed. Brains were removed, post-fixed in 4 % PFA and, for MR examinations, processed in 4 % agarose to prevent movement within the scanner. The samples were placed in a mouse whole body coil (Bruker, Ettlingen, Germany) of a dedicated small animal ultra-high-field MR scanner (ClinScan 7 Tesla). Imaging sequences and parameters were chosen as follows: T1-weighted imaging (slices 12, TR 500 ms, TE 9 ms, averages 2, voxel size 0.073 × 0.073 × 0.7 mm3, acquisition time 5:38 min), diffusion-weighted imaging (DWI, slices 7, TR 8,000 ms, TE 60 ms, averages 1, voxel size 0.234 × 0.234 × 1.0 mm3, acquisition time 16:16 min) and T2*-weighted imaging (slices 14, TR 983 ms, TE 4 ms, averages 2, voxel size 0.156 × 0.156 × 0.5 mm3, acquisition time 6:17 min). The volumes (mm3) of diffusion restriction after MCAO (infarcted areas are displayed as hypointense on maps of the apparent diffusion coefficient, ADC), of local bleeding (displayed as hypointense on maps of T2*-weighted imaging), or the breakdown of the BBB (displayed as hyperintense on T1-weighted imaging) were determined using the Osirix software (Osirix-Foundation, open-source).

Gene expression

BV2 cells [19] were cultured under standard conditions (5 % CO2 and 95 % humidity) in Dulbecco’s modified Eagle’s medium (DMEM, Gibco, Karlsruhe, Germany) supplemented with 10 % fetal calf serum and 1 % penicillin/streptomycin. Cells were serum deprived 1 day prior to the experiment and the next day trypsinized, plated on poly-l-lysine (Sigma)-coated dishes, and treated for 6 h with 0.1 or 0.5 % albumin (Sigma), 0.01 or 0.5 % fibrinogen (Merck Millipore), or 5 % fetal calf serum (Sigma). Total RNA was isolated using the RNeasy Mini Kit (Qiagen, Hilden, Germany) and DNAse I treatment (Roche, Mannheim, Germany) performed to avoid contaminations with genomic DNA. The quality of total RNA preparation was confirmed using a Nanodrop 2000c spectrophotometer (Thermo) and cDNA obtained using the Superscript III first strand synthesis system together with random hexamer primers (Invitrogen) according to the manufacturer’s guidelines. Primers for quantitative reverse transcriptase polymerase chain reactions (qRT-PCRs) were designed using the Beacon Designer 8 software (Premier Biosoft, Palo Alto, CA, USA) and were subsequently tested for efficiency and specificity. qRT-PCR was performed using the iQ SYBR Green supermix and the iCycler iQ5 device (both BioRad, Hercules, CA, USA). Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and elongation factor 1 alpha (EF-1 alpha) were used as a housekeeping genes and relative changes in gene expression were determined using the ΔΔCt method [37].

qRT-PCR analyses of matrix metalloprotease 9 (MMP-9) were performed in duplicate using the primers AAGTCTCAGAAGGTGGAT and AATAGGCTTTGTCTTGGTA yielding an amplicon size of 106 bp, a SYBR Green PCR kit (BioRad), and a Lightcycler 480 PCR system (Roche). Absolute copy numbers of MMP-9 were normalized to peptidylprolyl isomerase A (PPIA) as a housekeeping gene. Statistical analyses were performed as described below.

Trans-well migration assays in vitro

Migration of BV2 cells was measured using modified Boyden chambers (8 µm pore size, BD Biosciences) coated with poly-l-lysine. Then 1.5 × 105 cells were seeded in 200 µl medium in the upper chamber, while the bottom well contained 600 µl of medium supplemented with albumin (0.1 and 0.5 %), fibrinogen (0.01 and 0.5 %), or fetal calf serum (5 %). Migration was carried out for 6 h and cells were fixed with 4 % PFA. Cells that had migrated to the lower side of the filter were stained with Hoechst 33258 (Sigma) and counted upon image acquisition with an inverted microscope (Leica).

Flow cytometry analyses (FACS)

Mice were deeply anesthetized and perfused with PBS 1 day post reperfusion. Subsequent to the removal of the cerebellum, each brain was separated into an ipsilateral and a contralateral hemisphere. Tissue was dissociated using the “neuronal tissue dissociation kit (P)” (Miltenyi Biotec, Bergisch Gladbach, Germany) according to the manufacturer’s instructions. Cells were enriched by a Percoll (Sigma) gradient and prepared for FACS using anti-CD11b-Pe-Cy7 (eBioscience, San Diego, CA, USA), anti-CD45.2-PE (eBioscience), anti-CD68-FITC (Acris Antibodies, San Diego, CA, USA), or anti-CD86 V450 (BD Biosciences) antibodies. Measurements were performed using a FACSCanto II (BD Biosciences) device and data was analyzed using FlowJo v9.6.2 software (Tree Star, Ashland, OR, USA). CD45.2 intermediate and CD11b high cells were considered as microglia.

Statistical analysis

Data were analyzed using Prism5 software (Graphpad, San Diego, CA, USA). Data are presented as mean ± SEM of at least three independent experiments. Statistical analysis of the data was conducted using a non-parametric test (Mann–Whitney or Kruskal–Wallis tests) followed by a Dunn’s multiple comparison test. Alternatively, data are presented as mean ± SD and were compared between experimental groups with Wilcoxon Mann–Whitney rank sum tests with p values adjusted for each parameter for multiple comparisons by Bonferroni adjustment. Statistical analysis was performed with the Sigma Plot 11 statistical software package (Systat Software Inc., San Jose, CA, USA).

Results

Microglia associate with blood vessels after MCAO

BBB disruption and microglial activation are prominent pathological features of stroke; however, little attention has been paid to the dynamics and kinetics of microglial activation and the implications for blood vessels. In order to address this issue, we applied a mouse model in which the knock-in of GFP into the CX3CR1 gene labels microglia and peripheral macrophages under heterozygous conditions (CX3CR1+/GFP) and which serves as a microglia-loss-of-function model under homozygous conditions (CX3CR1GFP/GFP). CX3CR1+/GFP mice were subjected to transient MCAO with an occlusion time of 1 h, which was determined as a sufficient time frame to induce cortical damage (Fig. 1a). GFP expression was restricted to Iba-1-positive microglia (Fig. 1b–d). Imaging studies were performed at a tissue depth of 80 µm or deeper as measured from the tissue surface to reduce imaging artifacts due to the opening of the skull. Brain sections of MCAO-treated mice revealed that the core of the stroke was devoid of microglia and neurons (Fig. 1e), but in the ischemic penumbra, a close association of microglia (Iba-1), blood vessels (CD31), and neurons (MAP2) was observed (Fig. 1f–h). In a time course analysis taken 1–72 h post reperfusion, microglia adjacent to CD31-positive blood vessels displayed hypertrophic characteristics, an elongated shape, and thicker and retracted processes after 1 h (Fig. 1i). The number of cells that displayed an elongated morphology and enlarged cell bodies steadily increased over time (3 h) and microglia began to align with blood vessels in the ischemic hemisphere (Fig. 1j). At 6 h post reperfusion, the ongoing association of microglia with blood vessels in the penumbra resulted in an uneven distribution of microglia in the perilesional tissue (Fig. 1k). After 12 h, perivascular microglia exhibited the characteristic morphology of activated cells and displayed enlarged cell bodies with shortened thick processes that were in direct contact with blood vessels (Fig. 1l). At 24 h post reperfusion an increased number of microglia displayed spherical cell bodies (Fig. 1m). After 72 h a decreased number of blood vessels was observed (Fig. 1n). Quantification of the amount of blood vessel-associated microglia over time displayed a steady increase in perivascular microglia reaching a 2.5-fold increase 24 h and a 3.1-fold increase 72 h post reperfusion as compared to sham-operated mice (Fig. 1o). Costaining of microglia (Iba-1) and smooth muscle cells (SMA; Suppl. Fig. 1a–d) or pericytes (NG2; Suppl. Fig. 1e–h) in the ischemic penumbra 24 h post reperfusion showed that perivascular microglia did not stain positive for either of the latter markers.

Fig. 1
figure 1

Microglia associate with the vasculature upon MCAO. a Cresyl violet stained murine brain sections 24 h after MCAO. Box indicates the region analyzed in subsequent sections. Bar represents 1 mm. b–d Brain sections derived from CX3CR1-GFP mice were immunostained for Iba-1 to verify co-labeling of microglia by both markers. Bar represents 20 µm. e–h Murine brain sections 24 h after MCAO. Infarct area including core and penumbra are labeled with dotted lines. Iba-1-positive microglia (e, f, and h) as well as MAP2-positive neurons (g, h) are present in the penumbra but not in the core of the infarct area. Scale bar represents 500 µm (e) and 10 µm (h). i–n Brain sections derived from wild-type mice subsequent to transient MCAO at various time points after reperfusion. Microglia cells were identified by Iba-1 and blood vessels by anti-CD31 immunostaining, cell nuclei were counterstained by DAPI. Bar represents 25 µm. o Quantification of blood vessel-associated and non-associated microglia revealed a steady increase in the amount of perivascular microglia over time post reperfusion (n ≥ 3)

Full size image

Microglia form perivascular clusters upon MCAO

In order to study the dynamics of microglia association with blood vessels upon MCAO, intravital two-photon microscopy was performed in anesthetized CX3CR1+/GFP mice using a cranial window drilled into the skull to analyze the cerebral cortex. At 24 h after reperfusion, evenly distributed, highly ramified microglia were observed in control mice (Fig. 2a). In contrast, microglia in the penumbra of MCAO-treated animals displayed enlarged cell bodies and formed perivascular clusters (Fig. 2b). Time-lapse recordings taken 24 h after reperfusion illustrated “resting microglia”, which scanned the brain parenchyma with thin and long processes in control animals (Fig. 2c) but identified activated microglia with enlarged cell bodies and shorter processes in the penumbra (Fig. 2d). As a matter of fact, neighboring groups of microglia protruded their processes towards rhodamine–dextran-labeled blood vessels (Fig. 2e). Individual microglia fully enwrapped small blood vessels as displayed in a 3D reconstruction of a perivascular microglia cell in the penumbra (Fig. 2f).

Fig. 2
figure 2

Microglia form perivascular clusters and engulf blood vessels upon stroke. a Two-photon images of the cerebral cortex of CX3CR1+/GFP control (sham operated) or b MCAO animals in vivo 24 h post reperfusion. Blood vessels were visualized by perfusion with rhodamine–dextran. Insets represent gray scale pictures of the same figures to illustrate microglia morphology (n ≥ 10). Bar represents 20 µm and 5 µm (inset). c Time-lapse recordings identified resting microglia actively palpating the tissue in control animals, while d MCAO led to an activation of microglia as displayed by a migratory phenotype and e an active association of microglia with rhodamine–dextran-positive blood vessel (n ≥ 10). Bars represent 5 µm. f Microglia enwrapped blood vessel in the penumbra 24 h post MCAO as determined by double-immunostaining of Iba-1 (microglia) and CD31 (ECs) followed by orthogonal optical sectioning and a 3D reconstruction. Bars represent 5 µm

Full size image

Blood-derived proteins activate microglia

The engulfment of blood vessels in the ischemic penumbra by microglia raised the question of what are the molecular mechanisms that lead to the recruitment and activation of the local microglia pool. Serum albumin and fibrinogen have been described as proteins leaking into the parenchyma upon BBB breakdown [2, 70]; therefore, we analyzed whether or not these proteins attract microglia upon MCAO. The administration of albumin and fibrinogen to microglial BV2 cells in trans-well migration assays in vitro increased cell migration in a dose-dependent manner (Fig. 3a). Further, albumin and fibrinogen caused an activation of BV2 microglia cells as determined by the upregulation of inflammatory proteins such as the cytokines IL-1β, IL-6, IL-10, TNF-α, the chemokines MIP-1α and MCP-1, and the chemokine receptor CX3CR1 (Fig. 3b–h). In order to verify the activation of microglia in vivo, CD45.2-intermediate and CD11b-high microglia were isolated from wild-type, sham- or MCAO-treated animals 24 h after surgery and analyzed by FACS. Interestingly, the populations of cells positive for the phagocytosis marker CD68 (16.5-fold) or the co-stimulatory molecule CD86 (55-fold) were elevated in microglia from the ipsilateral hemisphere as compared to those from the contralateral site or control (Fig. 3i, j). Data suggest extravasating serum proteins such as albumin and fibrinogen contribute to the recruitment and activation of microglia subsequent to MCAO.

Fig. 3
figure 3

Extravasating blood serum proteins activate microglia. a Trans-well migration assays of microglia cells revealed a dose-dependent stimulation of microglia migration by blood serum albumin, fibrinogen, or fetal calf serum, which served as a positive control. Results are plotted as compared to the migration of unstimulated microglia (mean ± SEM, n = 4, *p < 0.05). b–f Likewise, albumin and fibrinogen caused an activation of microglia cells as determined by qRT-PCR of inflammatory proteins such as b IL-1β, c IL-6, d IL-10, and e TNF-α, the chemokines f MIP-1α and g MCP-1, or h the chemokine receptor CX3CR1 (mean ± SEM, n = 5, *p < 0.05). i, j CD45.2-intermediate and CD11b-high microglia were isolated from the brain of control (sham-operated) mice or the contra- as well as the ipsilateral hemisphere of MCAO-treated animals 24 h after surgery. FACS analyses revealed that the phagocytosis marker i CD68 and j the co-stimulatory molecule CD86 were upregulated 16.5-fold or 55-fold, respectively

Full size image

Phagocytosis of blood vessels by perivascular microglia

The above findings raised the questions of the functional relevance of the association of microglia with blood vessels. Therefore, brain sections of the contralateral (Fig. 4a) or ipsilateral hemisphere of MCAO-treated animals (Fig. 4b) were stained for Iba-1 (microglia) and CD31 (blood vessels). Subsequently, immunostaining and microscopical analysis were performed and orthogonal optical sections were three-dimensionally reconstructed. Perivascular microglia were barely seen in the control situation (Fig. 4a′) but most microglia were found to be associated with blood vessels in the ipsilateral hemisphere upon MCAO (Fig. 4b′). Importantly, the association of microglia with blood vessels was not restricted to individual cells but was observed for almost all microglia. Furthermore, the micromorphology of microglia was different comparing cells derived from the contralateral (Fig. 4c) and the ipsilateral hemisphere (Fig. 4d). In the latter, individual perivascular microglia displayed intracellular vesicles containing CD31-positive particles, suggesting a partial or full uptake of blood vessel-derived endothelial cells (EC). A time course analysis revealed a steady increase in CD31-positive intracellular vesicles in perivascular microglia (Fig. 4e–g). Quantification yielded a massive increase in internalization 24 h (70-fold) and 72 h (115-fold) post reperfusion as compared to sham-operated mice (Fig. 4h). Immunohistochemical staining of the phagocytosis marker CD68 verified that intracellular vesicles were phagosomes (Fig. 4i). To verify that intracellular particles carried EC material, we additionally applied the brain vasculature-specific EC markers Glut-1 (Suppl. Fig. 2a–c), caveolin-1 (Suppl. Fig. 2d–f), claudin-5 (Suppl. Fig. 2g–i), and podocalyxin (Suppl. Fig. 2j–l), which colocalized specifically with CD31 in brain vessels. Indeed, intracellular vesicles in microglia (Iba-1) also stained positive for all four endothelial markers, verifying that blood vessel components were internalized by microglia (Suppl. Fig. 3a–l). Further, no colocalization between perivascular microglia and smooth muscle cells (Suppl. Fig. 1a–d) or pericytes was observed (Suppl. Fig. 1e–h), showing that not these cells but ECs were internalized. Whether microglia phagocytized EC or dying neurons was analyzed by staining for Iba-1 and a combination of the neuronal marker MAP2 plus the neurofilament marker NF160 that marks neuronal processes and filaments. However, no perivascular microglia phagocytizing neuronal debris could be observed (Suppl. Fig. 1i–l), indicating that indeed EC components were taken up by perivascular microglia in the ischemic penumbra.

Fig. 4
figure 4

Perivascular microglia phagocytize blood vessels upon MCAO. a Brain slices of the contralateral or b ipsilateral hemisphere of MCAO-treated animals were stained for Iba-1 (microglia) and CD31 (ECs). Orthogonal optical sections revealed individual microglia enwrapping and partially digesting blood vessels after stroke. a′, b′ 3D reconstructions indicate an intimate spatial relationship of microglia and blood vessels upon MCAO in the penumbra. Bars represent 20 µm. c, d Higher magnifications of a and b with d displaying CD31-positive vesicles in perivascular microglia suggesting phagocytosis of EC components as indicated by arrows. Bars represent 10 µm. e–g Double-staining of microglia for Iba-1 and CD31 (ECs) revealed a steady increase in the amount of CD31-positive intracellular vesicles over time post reperfusion (arrowheads and asterisks). Bars represent 10 µm. h Quantification of colocalization (yellow, intracellular particles) yielded a 70-fold increase after 24 h and a 115-fold increase 72 h post reperfusion as compared to sham-operated mice. i Staining of microglia (Iba-1) for the phagocytosis marker CD68 showed that perivascular microglia were phagocytically active 24 h post reperfusion. Bars represent 10 µm

Full size image

Perivascular microglia-mediated blood vessel disintegration was followed in real time and in vivo by intravital two-photon microscopy of heterozygous CX3CR1+/GFP mice. At 4 h post reperfusion, we detected the activation of microglia with extensions of individual cells converging towards adjacent blood vessels in the cerebral tissue (Suppl. Movie 1). Long-term measurements starting 4 h post reperfusion and running for approximately 4.5 h revealed the localized and simultaneous recruitment of microglia to blood vessels as soon as 8 h post reperfusion in the ischemic penumbra (Suppl. Movie 2). This involved movements of microglia processes as well as cell bodies (Suppl. Movies 3, 4). At 24 h post reperfusion perivascular microglia started to phagocytize rhodamine–dextran-positive blood vessel components (Suppl. Movie 5), which eventually led to the disintegration of individual tubes (Suppl. Movie 6). Data suggest an MCAO-induced association of microglia with blood vessels, which led to a partial or full phagocytosis of neighboring blood vessel ECs.

Perivascular microglia contribute to the breakdown of the BBB

The findings above raised the question of whether the phagocytosis of EC components affects the respective blood vessels. Under physiological conditions, ECs of healthy vessels are found in a resting mode referred to as phalanx cells. Upon an insult that disrupts the integrity of the EC layer, phalanx cells become activated and may form tip and stalk cells, which display a high degree of plasticity and allow the damaged blood vessel to repair itself through a process termed angiogenesis. Once the damage is fixed, ECs re-enter a quiescent mode [7]. A novel marker expressed in activated ECs is the protein epidermal growth factor-like protein 7 (EGFL7) [49, 60], which we employed here as a marker for the activated endothelium [50] as an alternative to the vascular endothelial growth factor (VEGF). Immunohistochemical staining of brain slices for EGFL7 and the EC activation marker VEGF 24 h post reperfusion showed that both proteins colocalized in brain ECs in the ischemic penumbra (Suppl. Fig. 4a–c). Upregulation of further activation markers such as the platelet-derived growth factor-B (PDGF-B), the von Willebrand factor (vWF), or the intercellular adhesion molecule-1 (ICAM-1) [24, 64, 68] verified that indeed ECs of brain vessels in the ischemic penumbra were activated 24 h post MCAO (Suppl. Fig. 4d–l). Brain slides taken from control animals (Fig. 5a–d) or MCAO-treated animals (Fig. 5e–h) were immunohistochemically triple-stained for claudin-5 (blood vessels; Fig. 5a, e), EGFL7 (activated blood vessels; Fig. 5b, f), and Iba-1 (microglia; Fig. 5c, g). Data revealed that perivascular microglia upregulated EGFL7 in adjacent blood vessels 24 h post MCAO (Fig. 5h′). This indicates that ECs in these blood vessels were activated, which is indicative of disintegration of the BBB [69].

Fig. 5
figure 5

Perivascular microglia stimulate blood vessel plasticity upon MCAO. Brain slices of MCAO-treated animals were immunohistochemically stained for claudin-5 (ECs), EGFL7 (EC plasticity), and Iba-1 (microglia) on the contralateral (a–d) and ipsilateral (e–h) hemisphere 24 h post insult. Cell nuclei were counterstained with DAPI. d, h Perivascular microglia induced an EC-specific upregulation of the blood vessel plasticity marker EGFL7, indicative of EC activation and BBB breakdown in the penumbra. Bars represent 25 µm. d′, h′ Display higher magnifications of the boxed regions in the corresponding figures. Bars represent 10 µm

Full size image

In order to determine whether or not perivascular microglia induced a breakdown of the BBB, various hallmarks of this event were analyzed. At 24 and 72 h post reperfusion the expression of MMP-9, a protein described to be involved in stroke BBB turnover [4, 11], increased threefold in the penumbra upon MCAO as compared to sham-treated animals (Fig. 6a). In parallel, staining for extravasated IgG in the cerebral cortex of animals upon MCAO revealed that blood vessels in the ipsilateral hemisphere of the treated animals became much more permissive for blood components such as IgG 24 h after reperfusion as compared to the contralateral hemisphere (Fig. 6b, c). Leakage of IgG occurred in the core and the penumbra as indicated by the diminished immunoreactivity for the neuronal marker NeuN (Fig. 6d). Immunohistochemical analysis and 3D reconstruction revealed that the association of microglia with blood vessels contributed to their disintegration and BBB breakdown as shown by the extravasation of Evans blue in vessels that were still perfused (Fig. 6e, e′). Further, qRT-PCR analysis revealed a threefold increase in the amount of CD11b-positive immune cells in the ipsilateral hemisphere as compared to the contralateral site (Fig. 6f). This is indicative of the activation of the local microglia pool [35] as well as an increased extravasation of immune cells into the penumbra. In order to validate the latter, we immunohistochemically stained for neutrophils, microglia, and blood vessels in the penumbra of MCAO-treated animals at various times post reperfusion (Fig. 6g–i). Activated perivascular microglia and extravasating neutrophils were observed in the ipsilateral hemisphere in the penumbra 24 h post reperfusion (Fig. 6h). At 72 h post reperfusion, activated microglia and neutrophils were still observed in the penumbra but no tubular structures and only fragments of blood vessels could be detected (Fig. 6i). Data indicate that the blood vessels in the ischemic penumbra disintegrated and were subsequently degraded. The physiological relevance of the above findings was analyzed using CX3CR1GFP/GFP mice harboring microglia that are functionally impaired and therefore suited to perform loss-of-microglia-function studies [12]. MCAO was performed for 1 h and animals were killed 6 and 24 h post reperfusion by transcardial perfusion. At 15 min before being killed mice received an injection of 0.2 mmol/kg Gadovist as a contrast agent in order to quantify leakage of blood vessels in the stroke region. Stroke size and extravasation of contrast agent in the brain penumbra were measured by ultra-high-field MRI (Fig. 7a). At 24 h post reperfusion loss of microglia function led to a significant (p < 0.05) reduction in stroke size of 46.8 % (Fig. 7b) as determined on apparent diffusion coefficient (ADC) maps. Furthermore, extravasation of the contrast agent Gadovist (T1-weighted imaging), representing a breakdown of the BBB, was reduced by 50.4 % (Fig. 7c) in CX3CR1GFP/GFP mice as compared to wild-type animals. At 6 h post reperfusion results showed comparable trends but did not reach significance. Data demonstrate that reduced microglia activity correlates with a decrease in stroke size as well as increased blood vessel integrity.

Fig. 6
figure 6

MCAO induces BBB breakdown in the penumbra. a MCAO induced a significant upregulation of the metalloprotease MMP-9 in the penumbra (mean ± SEM, n ≥ 6, *p < 0.05). b–d Extravasation of IgG from leaky blood vessels, 24 h post MCAO in the b contralateral or c ipsilateral cerebral cortex with respect to the vessel occlusion. Bars represent 10 µm. d Double immunostaining using antibodies specific to the neuronal marker NeuN or d′ IgG illustrating neuronal degeneration as well as the leakage of blood vessels in the stroke area. d″ Arrows indicate the colocalization of extravasated IgG and cortical neurons in the penumbra. Bars represent 100 µm. e Microglia (Iba-1), blood plasma extravasation (Evans blue) and cell nuclei (DAPI) were detected by immunohistochemistry of brain sections of MCAO-treated animals 24 post reperfusion. Microglia associated with the still perfused blood vessels (see erythrocytes) during their disintegration, as shown by the faint cloud of Evans blue around the tube. e′ The 3D reconstruction highlights the close interaction of microglia and perfused blood vessel. Bars represent 10 µm. f qRT-PCR analysis revealed a threefold increase in the amount of CD11b-positive immune cells detected in the ipsilateral hemisphere as compared to the contralateral site (mean ± SEM, n ≥ 6, *p < 0.05). Immunohistochemical staining of blood vessels (claudin-5), microglia (Iba-1), neutrophils (Nimp), and cell nuclei (DAPI) of the g contra- or h ipsilateral hemisphere of MCAO-treated animals 24 h post reperfusion revealed microglia activation and association with vascular structures as well as extravasation of neutrophils into the penumbra, which were observed intravascularly (arrows) and extravascularly (arrowheads). i 72 h post reperfusion, microglia were still activated and neutrophils detected in the penumbra but blood vessels were fully degraded. Bars represent 50 µm

Full size image
Fig. 7
figure 7

Magnetic resonance imaging of mouse brains after MCAO. a Apparent diffusion coefficient (ADC) maps of diffusion-weighted imaging (DWI) and T2*-weighted imaging of mouse brains following MCAO (6 or 24 h) in coronal orientation. Hypointense areas represent restricted diffusion of stroke (ADC) as well as local bleeding (T2*) in WT and CX3CR1GFP/GFP mice (arrowheads). b Quantification of stroke size as measured by the volume (mm3) of diffusion restriction on ADC maps or c the volume (mm3) of contrast media extravasation (T1-weighted imaging) revealed a significant reduction of both parameters by approximately 50 % 24 h post reperfusion in CX3CR1−/− mice (mean ± SEM, n = 3, *p < 0.05)

Full size image

Discussion

Intravital two-photon microscopy of CX3CR1+/GFP mice in vivo revealed that microglia and peripheral macrophages associated with blood vessels in the ischemic penumbra upon transient MCAO. Microglia exhibited an altered morphology as compared to control cells with larger cell bodies and extensions that were both shorter and thicker [26]. The somata of perivascular microglia were found to fully engulf individual blood vessels in the perilesional tissue within a time frame of 24 h after the ischemic insult. In a related study, microglia were instantly activated in response to an acute microdisruption of the BBB induced by a laser [51]. Resting microglia of control animals repetitively palpated their environment in the cerebral parenchyma with cellular protrusions but microglia adjacent to the lesion switched from an undirected to a directed movement of their cellular extensions towards the damaged area. This culminated in a shielding of blood vessels by microglial processes. However, this study did not elaborate on the subsequent role of microglia in blood vessel disintegration and the implications of this pathological process for the evolvement of tissue damage into the ischemic penumbra.

The clearance of damaged cerebral tissue by microglia has been shown to be essential for neuroregeneration [48] and neuroplasticity [41]. In the current study, we present evidence that perivascular microglia engulf blood vessels and actively perform phagocytosis of EC or subcellular EC-derived particles, as perivascular microglia displayed CD68- and CD31-positive intracellular vesicles upon MCAO. Engulfed blood vessels were still perfused as indicated by erythrocytes within the vascular tube (Fig. 6e), suggesting that ECs were alive and the phagocytized CD31-positive material was not derived from dead cells. Additionally, perivascular microglia exhibited spherical extensions and cups at the extremity of the cellular processes, which have been implicated in the clearance of cellular material by microglia [63, 65]. The disposal of newborn neurons [62], living neural precursor [8], and complete glioma cells [31] by microglia was previously observed upon engulfment, indicating that perivascular microglia are capable of internalizing and degrading EC.

This raises the question of the contribution of the microglia population to phagocytosis in general as microglia are not the only phagocytic cell type detected upon cerebral ischemia in the brain. As a matter of fact, an influx of circulating monocytes and macrophages has been reported to occur 24 h after the insult and may persist for up to 1 week [18]. Schilling et al. analyzed mice harboring genetically GFP-labeled blood leukocytes for parenchymal leukocyte invasion into the brain in response to stroke at various time points. The study demonstrated that microglia predominantly exerted a phagocytic activity against neuronal cell debris in the penumbra 24 h after transient MCAO, which matches the time frame in which most of our analyses were performed [59]. Nevertheless, Schilling et al. did not emphasize the role of perivascular microglia in the phagocytosis of EC and the physiological role of it remains enigmatic. We suggest that the association of microglia with blood vessels 24 h post MCAO and the succeeding phagocytosis of EC components are indicative of an active breakdown of tight junctions [30] and the BBB by microglia.

Indeed, permeabilization of the BBB following cerebral ischemia has been reported in mammals [6, 33]. The opening of the BBB in the ischemic brain is associated with blood vessel remodeling and angiogenesis in a VEGF-dependent manner [69]. This biphasic event consists of an early phase 4 h after stroke and a late phase occurring 24–48 h after the ischemic insult [32, 47]. The early permeability phase [58] coincides with a peak expression of VEGF, the earliest marker of vascular remodeling [15, 54], in the ischemic penumbra [53]. At 24 h after the insult, EC activation markers such as EGFL7 [50], PDGF-B [3], vWF [66], and ICAM-1 [24] are upregulated and, indeed, we detected the expression of all four markers in blood vessels of the ischemic penumbra at this time point, which is indicative of the early stage of BBB breakdown. In order to test for the disintegration of the BBB, we analyzed for proteins and cells that have been associated with BBB breakdown. Pharmacological and genetic evidence indicated that MMP-9 is required for angiogenesis in response to brain ischemia [34, 45]. Further, microglia are considered to be a major source of the protein in human specimens [56]. Indeed, we found MMP-9 to be upregulated in the penumbra 24–72 h after the insult. Classical markers of BBB breakdown are the release of blood serum components (e.g., serum albumin, fibrinogen, IgG), Evans blue, or neutrophils from blood vessels into the brain parenchyma. All of these hallmarks were observed 24–72 h post MCAO in our model, indicating a breakdown of the BBB at this time point. Remarkably, at 72 h post reperfusion, vascular structures in the penumbra were in part no longer present, indicating that activated microglia contributed to the disintegration and degradation of blood vessels in this area. Intraparenchymal neutrophils were observed 72 h post reperfusion; however, the amount in the penumbra was low as compared to the core (Fig. 6h), which is in agreement with previous publications [14, 25] and supports the role of microglia in blood vessel breakdown.

Evidently, disintegration of blood vessels was not only restricted to the core of the stroke but also occurred in the neighboring tissue. Perivascular microglia were mostly observed in the ischemic penumbra and, therefore, we wondered which molecular cues might attract microglia to local blood vessels. We hypothesized that, upon permeabilization of the vascular wall caused by the ischemic stroke, blood components may diffuse into the cerebral parenchyma. Serum components such as albumin or fibrinogen have been described as major stimulators of microglia [2, 70]. Recently, the deposition of fibrinogen in the cerebral parenchyma has been reported to induce the clustering of microglia along blood vessels as a consequence of BBB disintegration [10]. These fibrinogen deposits have been shown to precede the infiltration of circulating monocytes into the brain parenchyma [16]. Furthermore, fibrinogen-derived peptides were described as chemotactic cues for human monocytes a long time ago [55]. In order to test whether blood serum-derived proteins have the capacity to activate resident microglia, we applied purified albumin and fibrinogen to BV2 [19] and found that both types of proteins induced the migration of microglia cells in a dose-dependent manner. Furthermore, microglia activation markers such as the cytokines IL-1β or TNF-α were significantly upregulated. This is in line with a previous publication that identified fibrinogen as an activator of microglia that led to an increase in cell body size and phagocytic activity [2]. Likewise, incubation of murine N9 microglia cells with serum albumin promoted the secretion of cytokines such as IL-1β and TNF-α [70]. Activation of microglia was also observed upon stroke in vivo and we found the phagocytosis marker CD68 as well as the co-stimulatory molecule CD86 to be upregulated, which is in agreement with the phagocytosis and activation data described above.

Although neurological damage occurs within seconds to minutes upon vessel occlusion, ischemic injury continues for hours and days [46]. Current data imply a slow-burning and irreversible scenario in which ischemic stroke causes a reduction in blood flow in blood vessels of the penumbra leading to hypoxia and upregulation of VEGF in the early subacute phase 4–6 h following stroke [42, 50, 53]. VEGF triggers an angiogenesis response that causes the disintegration of local EC and the extravasation of blood serum components like albumin or fibrinogen, which recruit nearby microglia that phagocytize the now activated EC and support the breakdown of the BBB in the late subacute phase 6–24 h after stroke. Blood flow in the affected blood vessels and in tubes connected to these vessels drops, which leads to an activation of microglia [44]. This process evolves along the vascular network until a larger part of the ischemic penumbra is irreversibly damaged, but only an early restoration of blood flow can stop the evolving tissue damage in stroke [40, 46]. In the final delayed phase of stroke, immune cells such as lymphocytes and neutrophils invade the brain parenchyma and cause secondary inflammatory effects [21].

We present an alternative scenario that is compatible with previous data, but which includes the contribution of microglia and, crucially, allows for the possibility of halting the “slow-burning” damage in the early subacute phase. We found that microglia remove ECs from the process by phagocytosis (Fig. 8) and, as a consequence, the endothelium is irreversibly damaged and blood vessels regress resulting in tissue loss in the penumbra. As a matter of fact, loss-of-microglia-function studies using CX3CR1GFP/GFP animals revealed a reduction in stroke size and blood vessel extravasation by about half, which is in agreement with previous reports [12]. Our model also extends to peripheral macrophages, which cannot be distinguished from microglia using the methods applied in this manuscript. It fits well to the kinetics observed in patients, where brain damage caused by stroke gradually gets worse within the first 24 h post insult [21, 40, 46] and leads to BBB permeability as measured by the invasion of macrophages labeled with ultra-small superparamagnetic iron oxide into the brain parenchyma after 24 h [57]. Furthermore, targeting early BBB disruption has been shown to mediate neurovascular protection upon ischemic stroke [52]. In summary, the data above suggest the inhibition of microglial activation as an approach to decrease tissue damage in patients within the first 24 h post ischemic stroke in order to keep as many blood vessels intact as possible and to favor tissue repair in the ischemic penumbra over injury [39].

Fig. 8
figure 8

Model of stroke-induced and microglia-mediated blood vessel destruction. Upon ischemic stroke, blood vessels in the core are destroyed but are potentially salvable in the penumbra. Untreated, hypoxia in these vessels triggers the upregulation of VEGF 4 h post insult, which activates the endothelium and renders the BBB partially permissive. Release of blood serum components such as fibrinogen leads to an activation of local microglia, which are recruited to blood vessels and contribute to their disintegration by the phagocytosis of EC and EC-components (CD68+/CD31+ microglia). Enlarged vessel damage allows for the extravasation of neutrophils and further blood serum components, which in turn activate more microglia and thereby contribute to the expansion of the damage in the neighboring ischemic penumbra along the tubular network. At 72 h post insult, blood vessel degradation is complete and remaining vessel debris is cleared by microglia and the invading immune cells

Full size image