Innate Response to Focal Necrotic Injury Inside the Blood-Brain Barrier¹

In mammals, the innate immune system responds rapidly to apoptotic and necrotic cell death to limit damage to tissues and escape of causal pathogens (1, 2). Apoptosis due to programmed cell death is generally noninflammatory whereas factors released from necrotic cells can be proinflammatory in the absence of pathogen-derived innate signals (2, 3, 4). The response to necrotic injury involves neutrophil and monocyte infiltration to suppress infection and initiate repair (5, 6). The role of neutrophils and monocytes in necrotic injury in the cerebral cortex is less well defined (6, 7).

The blood-brain barrier (BBB)³ separates blood leukocytes that would normally be a primary cell type responding to necrotic injury from the brain parenchyma where necrotic cell death might take place in response to infection, toxin action, excitotoxicity, or trauma (8). The BBB is composed of two layers. The first layer is composed of microvascular endothelial cells, which have abundant tight junctions. The second layer is the glia limitans that is formed by glial foot processes (8). The perivascular space between the endothelial cell and astrocyte-derived basement membranes is continuous with the subarachnoid space of the meninges and is populated by perivascular macrophages, some of which express CD11c and behave like immature dendritic cells (9, 10, 11). Large necrotic injuries destroy both neural tissue and the BBB and hence blood leukocytes play a major role in surveillance of these sites for infection and tissue repair (12, 13). Experimental necrotic injuries can be induced behind the intact BBB using a focused laser beam (14, 15).

Microglial cells are long-lived phagocytic cells of the myeloid lineage that populate the parenchyma of the CNS (11, 16, 17). Recent two photon laser scanning microcopy studies have demonstrated dramatic responses of microglial cells to laser-induced focal necrotic injury (14, 15). This process requires endogenous ATP acting through metabotropic P2YRs (14). Astrocyte networks were implicated in this response through experiments with inhibitors of connexin hemichannels, which are known to be involved in ATP-induced ATP release to relay signals (14). However, astrocytes labeled with sulforhodamine 101 showed minimal morphological changes in response to laser injury (15). The involvement of conventional leukocytes in this process has not been investigated.

In this study, we examine the cellular response to necrotic injury outside and inside the BBB using two-photon laser scanning intravital microscopy in the same thinned skull surgical preparation. We visualize blood leukocytes, microglial cells, and astrocytes using different transgenic or knockin mouse strains in which the respective endogenous cell types express GFP. Necrotic injury outside the BBB induces rapid leukocytic infiltration that walls off the necrotic area, essentially forming a granuloma. In contrast, there was no leukocyte infiltration behind the BBB unless direct vascular injury was induced along with the necrotic tissue injury. By 24 h after parenchymal injury behind the BBB, the microglial response also results in cell body translocation and formation of a granuloma-like lesion in which cell bodies surround the injury, but lack granulocytes and monocytes. We also observed a morphological polarization of the astrocyte cytoplasm toward the injury and the formation of a cytoplasmic Ca²⁺ gradient in the astrocyte processes. The astrocyte polarization and Ca²⁺ response is blocked by the same agents, which include ATP and gap junction blockers, that inhibit the microglial motility response. These observations support the model that astrocytes and microglial cells cooperate to generate an effective innate response to necrotic injury that excludes blood leukocytes.

CX₃CR1^+/gfp mice used to visualize ramified microglial cells and perivascular cells (18) were a gift from D. Littman (New York University School of Medicine, Howard Hughes Medical Institute, New York, NY); FVB/N-Tg(glial fibrillary acidic protein (GFAP)-enhanced GFP (eGFP))14Mes/J used to image astrocytes (19) and Tg(Thy-1H-eYFP) mice used to image neurons (20) were obtained from The Jackson Laboratory; Tg(LysM-eGFP) used to image granulocytes, monocytes, and macrophages (21) were a gift from T. Graf (Albert Einstein College of Medicine, Bronx, NY). All mice were housed in specific pathogen-free conditions and treated in accordance with Institutional Animal Care and Use Committee protocols of New York University School of Medicine.

Thinned and open skull intravital window surgeries were performed as previously described (14, 15). Anesthetized mice were maintained on 37°C warming plates and the intravital window was warmed through an objective heater to maintain 37°C. Laser ablation and mechanical injury were performed as previously described (14). The size of injury induced is typically ∼15 μm in diameter. The yellow fluorescent protein (YFP) signal in Tg(Thy1H-eYFP) mice was used to define the depth at which brain parenchyma starts in the barrel cortex. Vascular or perivascular spaces were defined by injection of 655 nm emitting quantum dots (R&D Systems) into the blood or subarachnoid space, respectively. Quantum dots excite at the same Ti-Sapphire wavelength as GFP (22).

Drugs were applied through open skull as previously described (14). Cell-permeant Ca²⁺ chelator BAPTA-AM (23) and cell permeant Ca²⁺ indicator Fluo-4-AM (Molecular Probes) were applied and washed as previously described (24, 25).

GFP, YFP, and quantum dots were excited with a mode locked Tsunami Ti-Sapphire laser tuned to 920 nm (Spectraphysics) connected to a Bio-Rad Radiance multiphoton microscope. Fluo4-AM was excited at 840 nm. Stacks of image planes were acquired using step sizes 1–3 μm to a depth of 100–200 μm using ×40 or ×60 water dipping objectives. The GFP and YFP signals were separated using a cyan fluorescent protein/YFP filter cube (22, 26). Time-lapse images were acquired at 1- to 1.5-min intervals.

Microglial and astrocyte process movements were followed using Volocity software (Improvision) and intensities were quantified using ImageJ (〈http://rsb.info.nih.gov/ij/〉) (14, 22, 26). Single-plane pseudocolor images were rendered with MetaMorph software (Molecular Devices). Cell counts were performed within three-dimensional volumes to calculate cells/mm³. Values of p were determined using two-tailed t test.

To determine the effect of tissue injury outside and inside the BBB, we used an intravital microscopy preparation in which the parietal bone of the skull is thinned to ∼30 μm (27). Two-photon laser scanning microscopy can then be used to image through the thinned bone (Fig. 1,A, blue) into the meninges, which includes the dura and the subarachnoid space (Fig. 1,B), and the cerebral cortex parenchyma (Fig. 1, C and D). When polyethylene glycol-coated quantum dots were injected into the subarachnoid space, they diffused into the perivascular spaces surrounding large vessels in the cerebral cortex (Fig. 1, B–D). This perivascular space is outside the glia limitans.

We imaged blood leukocytes using transgenic mice in which the myeloid-specific lysozyme promoter drives expression of eGFP in granulocytes, monocytes, and macrophages, but not parenchymal microglial cells (14). Initially, we focused on the effects of injury outside the BBB by focusing just beyond the thinned skull into the meninges. Before injury, there were scattered eGFP⁺ cells present in the meninges and lined meningeal blood vessels (Fig. 1^,E). It is possible that the marginated eGFP⁺ cells were a consequence of trauma from the bone thinning process. To induce injury, the laser was tuned to maximum power at 800 nm and held in place for 1 s. This is a nonlinear process rather than simple heating so the volume in which cells are killed is tightly restricted to ∼15-μm diameter circular volume at the laser focus. Focal injury induced in the meninges caused a rapid recruitment of eGFP⁺ cells to the injury site within 30 min (Fig. 1, F and G, and supplemental movie 1).⁴ Many of these cells were drawn from the intravascular pool that was commented on previously. The median speed of eGFP⁺ cell movement during this process was 5.23 ± 1.78 μm/min (range 2.11 to 9.03, n = 30 cells). The speeds were bimodal with populations centered at 7 and 3 μm/min. After a day, the injury was entirely walled off by the eGFP⁺ cells in a granuloma-like structure (Fig. 1 H). Therefore, necrotic injury outside the BBB induces rapid blood leukocyte extravasation and convergence on the injury site.

We next asked whether laser injury in the parenchyma recruited blood or perivascular leukocytes through the BBB. After the skull thinning, there were a few perivascular eGFP⁺ cells (6 per 0.02 mm³, range 2–11, n = 4). These cells were extensively spread, sessile, and displayed probing movements consistent with their identification as macrophages (Fig. 1,I, supplemental movie 1). Elongated eGFP⁺ blood leukocytes were often observed flowing rapidly through small capillaries that were detectable with intravascular quantum dots (supplemental movie 1). The cerebral cortex parenchyma did not contain eGFP⁺ cells (Fig. 1,I). As a control, we also imaged mice subjected to the thinned skull procedure, but without laser injury and the basal pattern did not change over days (Table I). Parenchymal laser injury did not lead to recruitment of eGFP⁺ cells to the injury focus (Fig. 1, J and K, and supplemental movie 1) or at 1 day (Fig. 1,L and supplemental movie 1). Parenchymal laser injury recruited eGFP⁺ cells to the perivascular spaces in the vicinity of the injury between days 1 and 7 and subsided to control levels by day 10 (Table I (n = 4 animals)). Thus, laser injury behind the BBB was weakly detected through an increase in perivascular leukocytes over a period of days, but these cells did not enter the parenchyma.

Do microglial cells in the parenchyma induce a granuloma-like encapsulation of the injury focus in the parenchyma? This is an important question because organotypic culture of brain slices revealed that the soma of microglial cells can move within the tissue (28), while the initial response to focal injury in the intact brain is restricted to the processes of the ramified microglial cells (14). To visualize microglial cells, we imaged cytoplasmic eGFP in CX₃CR1^gfp/+ knockin mice (18). We determined the position of microglial cell bodies over a period of days in the intact CNS using adult CX₃CR1^gfp/+ × Tg (Thy1H-YFP) mice. The Tg(Thy1H-YFP) mice were used to visualize the neural scaffold (27). We imaged CX₃CR1⁺ ramified cells at day 0 and day 3 in the same volume as defined by the highly stable pattern of neural processes and dendritic spines. Among 62 ramified microglial cells examined in three separate fields, none of these cells displayed a positional change of the soma over 3 days (Fig. 2, A and B). The process termini of the ramified microglial cells oscillated with a mean speed of 1.73 μm/min (range 0.22–6.49; n = 8 animals; p < 0.01). Thus, ramified microglial cell bodies were sessile in vivo.

We next examined the impact of focal parenchymal injury on the movement of microglial processes and soma in the first hour and then over longer periods. Sterile focal injuries were created in the parenchyma of CX₃CR1^gfp/+ × Tg(Thy-1H-eGFP) mice with the laser. Red quantum dots were used to mark the vasculature to avoid induction of perivascular injury. As previously reported, the response at 30 min after injury was based entirely on movement of processes, not soma (Fig. 2, C and D, and supplemental movie 2). We quantified the movement of individual microglial processes early in the response and found that they moved with a median rate of 0.82 μm/min (range 0.57–1.97; n = 7 animals). Thus, these tethered processes move ∼6-fold slower than leukocytes converging on an injury site in the meninges.

We had previously reported no microglial soma movement toward the injury site over ∼9 h of imaging (14). However, when we examined microglial cells 24 or 72 h after injury, we found that microglial cells with amoeboid and short ramified processes converged on the injury sites (n = 5 animals) (Fig. 2, E and F, and supplemental movie 2). In contrast, the neural processes remained in place (Fig. 2,F). The bodies of the microglial cells reached the injury site by 24 h and conversion to amoeboid shape peaked at 3 days, when the injury site had lost most of its autofluorescence and was instead brightly marked by GFP⁺ microglial cell bodies, which could be distinguished by movement and fluorescence spectrum from the injury itself (Fig. 2 F). The injury focus was resolved by day 10 (data not shown). This observation suggests that the injured tissue may have undergone phagocytosis by microglial cells. In addition, the activated microglial cells in the region surrounding the injury also relocated. There was a 19% increase in the number of microglial cell bodies in 20–40 μm radius from the injury center (range +14 to +25, n = 4 animals) and a 20% decrease in 40–60 μm radial space (range −15 to −30%; n = 4 animals), strongly supporting the soma convergence toward the injury. There was no evidence that CX₃CR1 cells were recruited from >150 μm from the injury site because there was no difference in the cell number within the 150 μm radius between the time immediately after injury and 3 days later (60 ± 3 cells/0.01 mm³, 58 ± 6 cells/0.01 mm³, n = 4 animals). Thus, while perivascular macrophages and monocytes also express CX₃CR1 there were no changes in cell number in the vicinity of the injury that would suggest recruitment of cells that were not present in the parenchyma at day 0.

We conclude from these studies that microglial cells respond to injury in two phases: an acute phase of process movement (0–30 min) and a slower phase of soma movement (1–3 days).

Astrocytes were examined to determine whether they undergo morphological or physiological responses to injury that are consistent with the proposed role in ATP-dependent amplification of signaling required for the rapid microglial response (14). We were particularly interested in this as a paradigm for tissue cell participation in guiding innate responses of myeloid cells. We used Tg(GFAP-eGFP), whose eGFP signal was detectable up to 100 μm from the surface of parenchyma (19, 28, 29). Using brain slice preparations of this mouse, Kirchhoff and colleagues (29) demonstrated fine filopodia-like branches from the apical astrocyte process that undergo extension/retraction cycles. In our in vivo imaging, these fine processes filled the space in a dense manner generating a “mossy” effect where it was not possible to identify individual processes (146 ± 11 cells/0.01 mm³; n = 7 animals, Fig. 3,A). The most striking aspect of the astrocyte processes is that unlike the microglial processes that surveyed a large volume through active probing of the parenchyma, the astrocyte processes filled the parenchyma to such as extent that only small movements not readily detectable with two photon microscopy through the thinned skull would be necessary to survey all extracellular space in the parenchyma (Fig. 3 A, supplemental movie 3).

The ability to visualize astrocyte processes using a bright cytoplasmic marker led us to reinvestigate the morphological response of astrocyte processes to focal injury in vivo. We determined the astrocyte response to laser damage in Tg(GFAP-eGFP) mice from immediately post injury to 48 h. After laser injury, we observed a 46% (n = 6) decrease in GFP fluorescence in a halo-like zone around the bright, autofluorescent injury site within the first few seconds to minutes following the injury (Fig. 3,B, supplemental movie 3). This reduction in fluorescence was not due to photobleaching because experiments in both CXCR3^gfp/+ and Thy-1-eYFP Tg mice did not display photobleaching in the same region (data not shown). Because it was hard to resolve individual processes, the reduced fluorescence could be due either to a retraction of some processes or a reduction in cytoplasmic volume within the same number of processes. It was clear that some processes remain in contact with the injury site. This result is consistent with the earlier reported decrease in sulforhodamine 101 fluorescence near the laser injury site (15). The same study noted no recovery of sulforhodamine fluorescence. In contrast, we observed postinjury changes in GFP fluorescence associated with thickening of a few processes in halo zone within 5–10 min (Fig. 3, C and D). The formation of these thick processes was maximal by 30 min, which corresponded to the time frame for the microglial processes to reach the injury focus (Fig. 3, E–G). Astrocytes are polarized cells with one major projection corresponding to the apical pole (29). Thus, the process thickening toward the laser injury site is most likely a reorientation of neighboring astrocytes toward the injury. This was followed by a slower phase of recovery between 30 min to 48 h, by which time the GFP signal in the fine processes was fully recovered except where excluded from a tight cuff around the injury site, most likely corresponding to the microglial cell bodies (Fig. 3 G, supplemental movies 3). Using Tg(GFAP-eGFP) to mark astrocyte cytoplasm, we observed a distinct cytoplasmic polarization of astrocytes toward the injury.

A more detailed analysis of this polarization process in response to injury revealed that the large astrocyte processes remain in contact with the injury site based on visible eGFP⁺ strands (Fig. 4, A–F, supplemental movie 3, and data not shown). The enlarged processes displayed irregular, saccadic motions consistent with their viability (n = 3 animals). Because there appeared to be many small viable processes that continued to permeate the halo around the injury it is not clear how a few processes are selected to enlarge. This might be explained by a cell polarity mechanism that can reinforce initial selection of a dominant process.

We next asked whether astrocyte polarization requires a gradient of extracellular ATP as reported for microglial cells (14). We tested pharmacological interventions that had previously been shown to block the microglial response (30, 31, 32, 33). ATP is a putative chemoattractant that requires formation of a gradient. Application of large amounts of ATP through an opening in the skull obscures this gradient (14). The ATP gradient can also be destroyed by degrading endogenous ATP with apyrase (14). Applications of 5 mM ATP, UTP, or CTP to the cortex in vivo before laser injury did not inhibit the initial reduction of eGFP fluorescence around the injury site (n = 3 animals) (Fig. 4, G–J, K–N, and O–R, supplemental movie 3). However, ATP and UTP, but not CTP, abolished the polarization response (Fig. 4, G-J, K–N, and O–R). Therefore, ATP signaling, possibly through P2Y2 (34), was required for the polarization of cytoplasm to astrocyte processes connected to the injury site. We further found that the ATP-degrading enzyme apyrase also blocked polarization (Fig. 4, S–V). Unlike the microglial reaction, which was restored soon after washout of apyrase (14), astrocyte polarization was not restored after washout of apyrase (n = 3, each). A second injury in an adjacent location after apyrase washout progressed normally with intact astrocyte polarization, demonstrating that apyrase was effectively removed. Thus, astrocyte polarization to the injury was dependent upon early events after injury and did not take place when the early responses were blocked. Furthermore, astrocyte polarization as defined above was not necessary for the microglial movement to the injury, which does recover fully after apyrase washout.

We next asked whether astrocyte polarization was dependent upon connexin hemichannels (33). Connexin hemichannels expressed by astrocytes are implicated in release of ATP to relay the signal to adjacent cells such as microglial cells and in the entry of Ca²⁺ into the cytoplasm of the astrocytes to mediate activation-related signals in the astrocyte network (23, 32). The connexin hemichannel inhibitor flufenamic acid (FFA) reversibly blocks microglial process convergence to laser injury (14). We found that FFA abolished astrocyte polarization after laser injury (Fig. 4, W–Z, n = 3). Although the removal of FFA restores microglial process convergence (14), the astrocyte polarization was not restored after FFA washout. However, astrocyte polarization was seen in new injuries after FFA washout (n = 3, each, data not shown). We also tested the effect of BAPTA-AM, an intracellular Ca²⁺ chelator that would block the Ca²⁺ signal, on astrocyte polarization after laser injury. BAPTA-AM blocked the astrocyte polarization response (Fig. 4, AA–DD) and also blocked the microglial response (data not shown). Rapid reversal of BAPTA-AM was not possible because the active form is a highly charged molecule that becomes trapped in the cytoplasm of cells after de-esterification and is expelled slowly following washout. The accumulated evidence indicates that the astrocyte cytoplasmic polarization is a component of an astrocyte response. This response takes place in parallel to the microglial response, but astrocyte polarization of eGFP to the injury was not necessary for microglial movement to this site.

We next asked whether we could detect function of the many astrocytic processes that remained in the halo region. Extracellular ATP triggers astrocytic network signaling through calcium in vitro (35) and ATP injection has been shown to elicit calcium waves in brain slice studies (32). Relative Ca²⁺ concentration in the astrocyte cytoplasm can be detected with Fluo-4 AM applied to the cerebral cortex through an open skull preparation (24). Although Fluo-4AM is taken up by all cells in the cortex the majority of the signal comes from astrocytes, which as described above, fill most of the volume of the brain and are more active in taking up and de-esterifying the dye (24). Fluo-4 AM signal has been shown to be in the astrocytes cytoplasm selectively by sulforhodamine 101 double labeling (36). The experiments were performed in CX₃CR1^gfp/+ mice because this allowed monitoring of the microglial response directly and the GFP⁺ microglial processes occupy so little of the volume of the cortex that they did not interfere with the measurement of the Fluo-4 signal (Fig. 5,A). Immediately after the injury the Fluo-4 fluorescence in the injury site had a similar profile to the GFP fluorescence in the Tg(GFAP-eGFP) mice with strands of signal in the halo <20 μm from the bright autofluorescence at the injury (n = 3 animals) (data not shown). Fluo-4 fluorescence increases specifically near the injury as early as 1 min following injury, preceding the microglial reaction (Fig. 5, B–F, supplemental movie 4) (n = 3 each). The Fluo-4 fluorescence formed a gradient that can be observed to decline >5-fold over a distance of 100 μm from the injury site (Fig. 5, B–F). At the resolution of our imaging acquisition rate and within the field size (300 μm × 300 μm), the Ca²⁺ signaling was not in the form of temporal waves as previously reported in tissue culture (23) or brain slice (32), but in vivo we observed a standing gradient extending over the astrocyte network that remained in place over the duration of microglial movement with maximal increase during the first 60 min (Fig. 5, B–F). Bathing the open cortex with 5 mM ATP (Fig. 5, G–I, supplemental movie 4) or with 300 U/ml apyrase before injury abolished the Ca²⁺ gradient, but the Ca²⁺ gradient was restored with apyrase washout (Fig. 5, K–O) (n = 3, each) (1). The recovery of the Ca²⁺ signal is readily detectable, but lower in magnitude than the signal resulting from a fresh injury. The recovery of microglial reaction is also slower after washout of apyrase than in a fresh injury (data not shown). CTP preincubation, in contrast had no effect on the injury induced Ca²⁺ gradient (data not shown). Thus, ATP is required for Ca²⁺ gradient formation. Fluo-4 AM itself slowed the microglial response by 2-fold, consistent with a mild Ca²⁺ buffering effect of Fluo-4AM. FFA and Fluo-4 AM together appeared to cause toxicity to microglial cells based on loss of GFP fluorescence in CX₃CR1^gfp/+ mice so this combination could not be used to study the role of connexin hemichannels in the generation of the Ca²⁺ gradient (data not shown).

We conclude that focal injury generates a standing Ca²⁺ gradient in the astrocyte network that depends upon extracellular ATP and gap junction hemichannels.

In this study, we have compared the innate cellular responses to focal necrotic injury induced outside or behind the BBB. Outside the BBB LysM-eGFP⁺ cells (mostly neutrophils) rapidly leave the vascular space and surround the injury focus. Behind the intact BBB there was no invasion of the parenchyma by LysM-eGFP⁺ cells, but a robust albeit slower response by microglial cells and astrocytes. Our results support a model first proposed by Davalos et al. (14) in which astrocyte networks guide microglial cells to the injury (Fig. 6, A and B). Although these responses are very distinct in terms of the responding cell populations, the end result of a necrotic lesion fully surrounded by cell bodies of phagocytic cells is similar. Although myeloid cells did not invade the CNS parenchyma, we show that they were recruited in the nearby perivascular areas over days (Fig. 6 C).

There are a number of mechanisms that could account for the exclusion of LysM-eGFP⁺ cells from the brain parenchyma during responses to necrotic injury. One is that the hydrodynamics of blood flow in the brain may be consistent with neutrophil margination in the meninges, but not parenchymal microcirculation. It has been noted that parenchymal microcirculation is very high velocity compared with meningeal microcirculation (37). A second possibility is that the intact glia limitans may not allow blood leukocytes to enter the parenchyma. This is consistent with the observation that more than 1 week after laser injury many LysM-eGFP⁺ cells accumulate only in the perivascular spaces (Fig. 6 C). These cells may have migrated in from the subarachnoid space in response to signal from the resolving necrotic injury. The only barrier that would prevent these cells from entering the parenchyma from the perivascular space is the glia limitans. Finally, the coordinated movement of the microglia and astrocytes may act functionally as a second line BBB in intercepting signals to blood leukocytes that keep these chemotactic signals at a low level in the vasculature and work with the BBB to reduce neutrophil and monocyte entry.

The astrocyte/microglial axis forms a first-line innate immune mechanism in the parenchyma to detect mediators released from dead or injured cells using the ATP signaling system (4, 38). Although leukocytes can also use ATP as a signaling molecule (33, 39, 40), it is not clear whether ATP is used in the leukocyte response in the meninges. The meninges are an important site for infections by viruses and bacteria so understanding this response will be important. We limit our detailed focus on further understanding the collaboration between microglial cells and astrocyte behind the BBB, which serves as a model for tight collaboration of tissue cells with myeloid cells to guide cell migration.

We confirm the rapid movement of microglial projections to focal injury as previously published (14, 15). We extended our intravital observations to days and found the microglial cell bodies directly surround the injury focus as early as 24 h (Fig. 6,A). This more extensive response may facilitate phagocytosis of necrotic material from the injury site and containment of pathogens that might be a cause of cell necrosis. We found that the astrocytes undergo changes in morphology and Ca²⁺ signaling that are fully consistent with their playing a key role in this process, although none of the experiments we have performed directly demonstrate a causal link between astrocyte signaling and microglial responses. Astrocytes polarize their cytoplasm toward the injury site in parallel to the convergence of the microglia, but astrocyte polarization was not needed for microglial process movement toward the injury site (Fig. 6,A). The astrocyte cytoplasm generates a Ca²⁺ gradient toward the injury focus that is well-correlated with the microglial response (Fig. 6 B). Our results establish a strong link in that only astrocytes are known to have gap junction hemichannels that would be blocked by FFA, but it is possible that FFA could have some direct effect on microglial cells. Similarly, ATP, apyrase, and BAPTA-AM may all have effects directly on microglial cells. More specific genetic tools are needed to forge a causal link between the astrocyte Ca²⁺ gradient and microglial responses, but our results here provide motivation for future effort in this direction.

Previous organ and cell culture studies demonstrated a Ca²⁺ wave in the astrocytic network in response to ATP injection in a hemichannel but not gap junction-dependent manner (32, 33). In contrast to these brain slice studies, we report for the first time formation of a standing gradient of cytoplasmic Ca²⁺ in the astrocyte network around the injury site. It is possible that the level of damage in our model is not sufficient to induce Ca²⁺ waves as in an epilepsy model (25). The Ca²⁺ gradient formed around focal injury sites was well-correlated with the rapid response of microglial processes, which is only observed in the in vivo setting. It is not clear how this Ca²⁺gradient is maintained in the extensively interconnected astrocyte network and whether the triggering of the gradient requires a direct astrocyte injury besides the ATP release from the injured cell(s).

The movement of the astrocyte cytoplasm toward the injury site may represent a classical polarization response. Astrocytes have been studied as a model for cell polarization (41). Neural cells are derived from neuroectoderm and thus are epithelial in origin with distinct apical and basolateral surfaces. At baseline, astrocyte calcium activity is lower and the cell bodies and their major process extend in random orientations in the brain parenchyma. Astrocyte polarization toward sites of injury may allow them to contribute to the process of containing the injury, although this polarization was not required for the microglial convergence that leads to full containment of the lesion by microglial cell bodies. Alternatively, the enlargement of these processes toward the injury site may be a response to injury that does not involved polarization machinery in the cell. Other markers of astrocyte polarity could be investigated to resolve this.

Focal laser injury is different from other models like photocoagulation to model stroke or freezing to model larger necrotic injuries in terms of the small and controlled size of injury (∼volume of a single cell) as well as minimal damage to the BBB near the injury (14, 15, 42, 43, 44). Therefore, this study highlights the critical role played by the microglia and astrocytes in the innate response to tissue injury that may include active maintenance of the BBB in the day-to-day trauma-ridden environment faced by mammals. We did not examine larger injuries that span the vasculature and result in loss of the BBB. In larger injury models such as in stroke and trauma, the mechanical breakdown of the BBB will enable initial infiltration of leukocytes with inflammatory tissue damage (6). Even in large necrotic injuries with reperfusion the role of blood leukocytes may be limited (6, 7). Although neutrophils may play a role in reperfusion injury in stroke by damaging the vasculature, our results suggest that they may not enter live brain parenchyma where the astrocyte-microglial reaction is capable of walling off injuries. Indeed, in brain slice studies, activated microglial cells are found all around the edge of the tissue (45).

We thank Guy Shakhar, Dimitrios Davalos, Jason Lieberthal, Wenbiao Gan, Dan Littman, Juan Lafaille, and the members of the Dustin laboratory for helpful discussions. We thank Jianzhong Li for help with Fig. 2 E. We thank Dan Littman and Thomas Graf for their gift of mutant and transgenic mice used in our study.

The authors have no financial conflict of interest.