P2Y1 Receptor-evoked Glutamate Exocytosis from Astrocytes

ATP, released by both neurons and glia, is an important mediator of brain intercellular communication. We find that selective activation of purinergic P2Y1 receptors (P2Y1R) in cultured astrocytes triggers glutamate release. By total internal fluorescence reflection imaging of fluorescence-labeled glutamatergic vesicles, we document that such release occurs by regulated exocytosis. The stimulus-secretion coupling mechanism involves Ca2+ release from internal stores and is controlled by additional transductive events mediated by tumor necrosis factor-α (TNFα) and prostaglandins (PG). P2Y1R activation induces release of both TNFα and PGE2 and blocking either one significantly reduces glutamate release. Accordingly, astrocytes from TNFα-deficient (TNF–/–) or TNF type 1 receptor-deficient (TNFR1–/–) mice display altered P2Y1R-dependent Ca2+ signaling and deficient glutamate release. In mixed hippocampal cultures, the P2Y1R-evoked process occurs in astrocytes but not in neurons or microglia. P2Y1R stimulation induces Ca2+-dependent glutamate release also from acute hippocampal slices. The process in situ displays characteristics resembling those in cultured astrocytes and is distinctly different from synaptic glutamate release evoked by high K+ stimulation as follows: (a) it is sensitive to cyclooxygenase inhibitors; (b) it is deficient in preparations from TNF–/– and TNFR1–/– mice; and (c) it is inhibited by the exocytosis blocker bafilomycin A1 with a different time course. No glutamate release is evoked by P2Y1R-dependent stimulation of hippocampal synaptosomes. Taken together, our data identify the coupling of purinergic P2Y1R to glutamate exocytosis and its peculiar TNFα- and PG-dependent control, and we strongly suggest that this cascade operates selectively in astrocytes. The identified pathway may play physiological roles in glial-glial and glial-neuronal communication.

Ca 2ϩ -independent glutamate release is less well characterized. It may take place by molecular transport of cytosolic glutamate across the plasma membrane through specialized proteins such as channels and transporters. Carrier-mediated release can occur by reversal of high affinity glutamate trans-porters, a process that depends on ionic membrane changes mostly occurring under ischemic/hypoxic conditions (46) or by cystine-glutamate exchange (47). Three types of large plasma membrane channels have been claimed to participate in the nonexocytic release of glutamate as follows: (i) swelling-activated anion channels, whose identity is still unknown (48); (ii) purinergic P2X7 receptors, gated by high concentrations of extracellular ATP (49); (iii) gap-junction hemi-channels formed by hexamers of connexin 43 (50). Opening of connexinhemichannels (51)(52)(53) or of P2X7 receptors (54) would contribute to release of ATP as well, although a Ca 2ϩ -dependent ATP release process has also been described (55).
ATP and other purine nucleotides and nucleosides play an increasingly recognized role in rapid intercellular signaling in both the central and peripheral nervous system (reviewed in Refs. 4 and 56). Two large families of purinergic receptors mediate neural ATP functions, ionotropic P2XR and metabotropic P2YR, operating on different time scales; P2XR generally act within milliseconds and mediate ATP-dependent neurotransmission and other very fast ATP effects (57,58). By contrast, P2YR act on longer time scales and, among other functions, mediate intercellular communication in glial networks (7,8,56,59,60). Here we show that selective activation of the purinergic P2Y1R subtype triggers glutamate exocytosis from cultured astrocytes as well as from acute hippocampal slices (but not from isolated synaptosomes). In both preparations P2Y1Revoked Ca 2ϩ -dependent glutamate release is controlled by TNF␣ signaling via TNFR1 and by prostaglandins.

MATERIALS AND METHODS
Pharmacological Agents-Recombinant rat TNF␣ and sTNFR1 were obtained from R&D Systems Europe (Oxon, UK); anti-TNF␣ antibody (AbTNF␣) was from Genzyme (Cambridge, MA); PD98059 was from Biomol (Plymouth Meeting, PA); tetanus neurotoxin (TeNT) was a gift from C. Montecucco; and BB3103 was a gift from British Biotech (Oxford, UK). Agonists and antagonists of glutamate receptors were from Tocris Cookson Ltd. (Bristol, UK). All other agents were from Sigma, except as otherwise indicated.
Hippocampal Cell Cultures-Astrocyte cultures obtained from newborn rats and mice were prepared as described (33). Once the cultures reached the appearance of confluent monolayers, they were depleted of microglial cells by mechanical shaking and replated at 2 ϫ 10 5 viable cells/ml of Eagle's minimal essential medium (ICN, Milan, Italy) plus 10% fetal calf serum either in 35-mm Petri dishes or on glass coverslips. Cultures were 98% pure in astrocytes, as revealed by immunocytochemistry with monoclonal antibodies against GFAP (1 g/ml; Chemicon, Hofhein, Germany). Experiments were generally performed on confluent monolayers (high density cultures, 40 cells/14,500 m 2 ). Low density cultures were, however, utilized for experiments with co-cultured INS-1 cells (see below). In some experiments the astrocyte cultures were enriched in activated microglia (10 -12% of total cells) as described (33). To prepare mixed neuron-glia cultures, hippocampal pyramidal neurons were obtained from embryonic day 17 rat brains. Hippocampi were dissociated by treatment with trypsin (0.25%) plus DNase (0.05%) in Hanks' balanced salt solution (15 min, 37°C), stopped with 10% fetal bovine serum, and followed by gentle mechanical pipetting. Cells were then plated at a density of 3 ϫ 10 5 onto astrocyte monolayers for mixed cultures. Cultures were then grown in Dulbecco's modified Eagle's medium (supplemented with 10% horse serum and N2 (Invitrogen)) and treated on day 3 with cytosine arabinoside (5 M; 24 h) to prevent further glial proliferation. In some experiments the mixed cultures were subsequently depleted of the neuronal component (61). Briefly, at 7 days in vitro, cultures were exposed to N-methyl-D-aspartate (NMDA, 300 M) for 1 h in a Mg 2ϩ -free medium and utilized 24 h later, after death of all neurons and removal of debris by several pipette washes. For preparing neuron-enriched cultures, hippocampal cells obtained from embryonic day 17 rat brains as above were plated onto ornithine-coated dishes and grown in Dulbecco's modified Eagle's medium supplemented only with N2. About 95% of the cells in this culture were positive for the neuronal marker MAP2 (Sigma). Activated microglia-pure cultures (Ͼ99% positive for the specific marker, isolectin B4) were prepared as described (33) and used in experiments within 12-24 h.
Enzymatic Assay of Endogenous Glutamate Release-Efflux of endogenous glutamate from cell cultures, hippocampal slices, or synaptosomal suspensions was monitored by use of an enzymatic assay as described previously (33). Briefly, either type of preparation was lodged in a cuvette inside a computerized spectrofluorometer (LS50B, PerkinElmer Life Sciences) at 37°C under stirring in a buffer containing (in mM) NaCl 120, KCl 3.1, NaH 2 PO 4 1.25, HEPES-Na 25, glucose 4, MgCl 2 1, CaCl 2 (pH 7.4), added with glutamate dehydrogenase (GDH, 40 units/ml; batch B27516; Calbiochem) and 1 mM NADP ϩ . Glutamate released from the preparations was immediately oxidized by GDH to ␣-ketoglutarate with formation of NADPH and fluorescence emission at 430 nm (excitation light, 335 nm). Release was quantified referring to standard curves constructed with exogenous glutamate and normalized for the protein content of each sample. Agents were added directly in the cuvette through a microsyringe, except those acting intracellularly, which required preincubation (30 min, unless otherwise specified). The effect of Baf A1 in hippocampal slices was assessed using two different experimental protocols as follows: (a) by simply exposing the slices to the drug in the dark for 2 or 6 h; (b) by exposing the slices to the drug for 2 h by having prestimulated the slices with agonists or saline at the beginning of the exposure period.
Astrocyte and INS-1 Cell Cultures for Imaging Experiments-For TIRF and Ca 2ϩ imaging experiments, we used high density astrocyte cultures (see above). To monitor glutamatergic vesicle fusions, astrocytes were transfected with the VGLUT2-EGFP construct as described (37) about 6 days after replating. For experiments using two color imaging with alternating TIRF and epifluorescence (EPI) illumination, we used low density co-cultures of astrocytes and INS-1 insulinoma cells (INS-1 cells; INS-1E cell line obtained from Dr. CB Wollheim). Astrocytes were plated (2.5 ϫ 10 4 cells) on glass coverslips and used 2-3 days after transfection with the VGLUT2-EGFP construct. INS-1 cells were plated either alone (1.5 ϫ 10 5 cells) or on the transfected astrocytes and were used the next day. For experiments based on single cell [Ca 2ϩ ] i measurements with EPI illumination, cells were loaded with 2.5 g/ml X-rhod-1 AM (Molecular Probes) for 40 min at room temperature in a HEPES-KRH buffer containing (in mM) the following: NaCl 120, KCl 3.1, MgCl 2 2, CaCl 2 1.8, NaH 2 PO 4 1.25, HEPES-Na 25, glucose 4 (pH 7.4) and then allowed to de-esterify for 40 min before imaging. For experiments where we combined acridine orange (AO) and X-rhod-1 imaging under dual illumination (TIRF and EPI), X-rhod-1 loading was followed by staining with AO (2 M, 10 min). After several washes, coverslips were mounted in an open perfusion micro-incubator (PDMI-2, Harvard Apparatus) set at 37°C on the stage of the optical recording microscope. During experiments, cells were continuously perfused with HEPES-KRH (1 ml/min), and stimuli were rapidly (2 or 5 s) applied via a software-controlled micro-perfusion fast-step device (100 l/min, Warner Instrument Corp.).
Optical Imaging-A Zeiss Axiovert 200 inverted fluorescence microscope was modified to allow both EPI and TIRF illumination (Visitron Systems). EPI illumination was utilized for single-cell [Ca 2ϩ ] i measurements in either rat or mouse astrocytes or INS-1 cells. X-rhod 1 fluorescence was recorded through a 63ϫ objective lens (Zeiss, Plan Apochromat) and directed through a 590 long pass filter (Zeiss filter set 15) at 50or 100-ms (for INS-1 cells and astrocytes, respectively) intervals by imaging with excitation light at 568 nm generated by a polychromator illumination system (Visichrome, Visitron, Germany). For TIRF illumination, the expanded beam (488/568 nm argon/krypton multiline laser, 20 milliwatts; Laserphysics) was passed through an AOTF laser wavelength selector (Visi-Tech International, UK) synchronized with a SNAP-HQ CCD camera (Roper Scientific) under Metafluor software (Universal Imaging) control and introduced from the high numerical aperture objective lens (Zeiss ␣-plan FLUAR 100ϫ, 1.45 NA). Light entered the coverslip and underwent total internal reflection at the glass-cell interface. In our experimental conditions penetration depth of TIRF illumination was calculated to be 84 nm (37,65). In experiments on high density astrocyte cultures, we imaged VGLUT2-EGFP and AO fluorescence under 488 nm TIRF illumination as described (37,66). The images were acquired at 20 interlaced frames/s. In experiments on low density astrocyte-INS-1 cell co-cultures we imaged both X-rhod-1 and AO fluorescence with a dual shutter. In practice, we combined EPI illumination (568 nm xenon arc lamp) and TIRF illumination (488 nm laser) at 10 interlaced frames/s. Lights were filtered with a beam splitter (Zeiss filter set 24).
Image Analysis-Video images, digitized with MetaFluor, were analyzed with MetaMorph software (Universal Imaging). AO flashes were measured as fluorescence intensity changes (arbitrary units) in a circle (diameter, 895 nm) positioned on the spot, and in a concentric annulus (inner diameter, 895 nm; outer diameter, 1195 nm). Only when the fluorescence increased, spread, and then declined was the event counted as a flash (67). Temporal dynamics in X-rhod-1 fluorescence were expressed as background-subtracted ⌬F/F 0 (%), where F 0 represents the fluorescence level of the cells before stimulation, and ⌬F represents the change in fluorescence occurring during stimulation of the cell. Statistical differences were established using the Student's t test at p Ͻ 0.01. Data are expressed as mean Ϯ S.E.
Monitoring of TNF␣ Release and PGE 2 Formation-Cell-free supernatants from astrocyte cultures were collected after 3-or 5-min incubations with stimulants. Release of TNF␣ was monitored using a highly sensitive enzyme-linked immunosorbent assay kit (R&D Systems Europe, Oxon, UK). PGE 2 generation was detected using the sensitive prostaglandin E 2 EIA kit containing monoclonal antibody (Cayman Chemical, Ann Arbor, MI).
Western blot analysis of P2Y1R expression was performed in hippocampi from TNF␣ ϩ/ϩ and TNF␣ Ϫ/Ϫ mice by conventional SDS-PAGE. Proteins were transferred to polyvinylidene fluoride membranes by means of a semidry electroblotter (Bio-Rad). Nonspecific binding was blocked for 1 h at room temperature with 5% nonfat milk in 50 mM Tris-HCl, 200 mM NaCl, 0.1% Tween 20 (pH 7.4). Membranes were subsequently incubated overnight at 4°C with antibodies to P2Y1R (5 g/ml). Immunoreactive proteins were visualized with an enhanced chemiluminescence substrate (Amersham Pharmacia). Densitometric analysis was performed with the NIH Image program.
P2Y1R-evoked Glutamate Release Occurs via Ca 2ϩ -dependent Exocytosis-To distinguish whether P2Y1R stimulation induces glutamate release via a Ca 2ϩ -dependent or a Ca 2ϩindependent mechanism, we applied 2MeSADP (10 M) to astrocyte cultures preincubated with the intracellular Ca 2ϩ chelator BAPTA/AM (50 M, 30 min); in this situation the glutamate release response to the purinergic agonist was abolished (Fig. 2a). In view of the recent demonstration that cultured astrocytes secrete glutamate by regulated exocytosis of synaptic-like microvesicles in response to stimulation of group I metabotropic glutamate receptors (37), we tested whether P2Y1R stimulation would trigger glutamate release by the same process. At first, we tested whether drugs interfering with exocytosis affected P2Y1R-evoked glutamate release. Indeed, the release response to 2MeSADP (10 M) was suppressed (  , and the suppression of ATP-induced glutamate release in cultures pre-exposed to the P2Y1R-selective antagonist A3P5PS (100 M, n ϭ 4). b, dose-response curve of ATP-induced glutamate release. Each point is the mean of 2-4 determinations in triplicate. c, pharmacological identification of the purinergic receptors that mediate the ATP effect. Histograms represent glutamate release evoked by the following: (i) ATP in the absence or presence of P2 purinergic receptor antagonists, including the P2Y1R-selective agent A3P5PS; (ii) P2 receptor agonists, i.e. 2MeSADP, a P2Y1R-selective agent, UTP, which acts on many P2YR but not P2Y1R, BzATP, and ␣,␤-meATP, P2XR agonists active mainly on P2X7 and P2X1 and -3, respectively. All agents were added at 100 M, except 2MeSADP that was added at 10 M. Results (mean Ϯ S.E.) are expressed as percentage of the glutamate release response evoked by 100 M ATP (0.78 Ϯ 0.12 nmol/mg protein; for each condition n ϭ 3-6 in triplicate). d, immunocytochemical detection of P2Y1R expression (green) in GFAP-positive (red) astrocytes. Scale bar indicates 25 m. tetanus neurotoxin (TeNT; 20 h, 2 g/ml), an endopeptidase that cleaves specifically and inactivates VAMP, the v-SNARE protein essential for the formation of the vesicle fusion complex (71).
To directly demonstrate that the P2Y1R-evoked glutamate release occurs by regulated exocytosis, we utilized TIRF microscopy (72) and an experimental approach recently developed in our laboratory (4, 37, 65). Astrocytes in high density cultures were transfected with a chimeric vesicular glutamate transporter construct (VGLUT2-EGFP) and double-stained with acridine orange (AO), a cargo dye that accumulates in a self-quenched state inside acidic compartments (73), including VGLUT-EGFP-positive (VGLUT ϩ ) vesicles (37). Therefore, EGFP fluorescence allows the identification of VGLUT ϩ vesicles, whereas abrupt changes in AO fluorescence signal their exocytic fusion. Thus, AO release in the extracellular space upon vesicle fusion with the plasma membrane leads to dequenching of the dye with a transient increase of fluorescent light (a "flash") followed by diffusion and, eventually, disappearance of the fluorescence (37,73,74). Therefore, each fusion event can be detected and counted as a flash of AO fluorescent light (AO flash) (37,74). A brief (2-s) localized pulse of 2MeSADP (10 M) evoked numerous AO flashes from perfused astrocytes. Such flashes were counted in individual cells containing VGLUT ϩ vesicles. Histograms in Fig. 2b show the average time distribution of VGLUT ϩ vesicle fusions counted in 50-ms time frames (n ϭ 5 different experiments). The number of fusion events rapidly increased to reach a peak within 350 ms from the start of 2MeSADP application to then gradually decrease and slowly return to background values after a total of about 5 s. Overall, 55.2 Ϯ 9.2% of the VGLUT ϩ vesicles present in the TIRF field underwent exocytosis. When saline was applied instead of 2MeSADP, no increase above background in the number of fusions was observed demonstrating the specificity of the 2MeSADP-dependent effect. In a second series of experiments INS-1 cells were co-cultured with astrocytes and used as biosensors of astrocytic glutamate release events in response to P2Y1R stimulation. INS-1 cells are well suited for this purpose because, when cultured in the absence of astrocytes, (a) they respond to glutamate with the NMDA receptor (NMDAR)-dependent [Ca 2ϩ ] i elevations, blocked by D-AP5, duplicated by NMDA (37) (Fig. 2c), and (b) they do not show any detectable Ca 2ϩ response to the P2Y1R agonist 2MeSADP (10 M, see Fig. 2b). Experiments in low density astrocyte-INS-1 cell co-cultures were performed on isolated pairs of the two cell types, generally arranged with the INS-1 cell and the astrocyte in direct lateral contact (37). Cells were stimulated by two subsequent pulses (2 s) of 2MeSADP (10 M) separated by an interval (washing period) of about 3 min (Fig. 2e). In nine experiments we always noticed the same sequence of events as follows: (a) at first 2MeSADP induced AO flashes from VGLUT ϩ vesicles (33.4 Ϯ 4.5% of those present in the TIRF field) in the astrocyte; such flashes occurred in a burst that lasted about 1s and peaked 350 ms from the start of drug application; (b) the AO flashes were then rapidly followed by [Ca 2ϩ ] i elevations (at peak; 35.4 Ϯ 4.8 ⌬F/F 0 %) in the INS-1 cell (average peak AO-to-peak [Ca 2ϩ ] i delay, 106 Ϯ 52 ms). As shown in  (Fig. 2d). Because D-serine is a glutamate co-agonist at NMDAR and could be released by astrocytes via Ca 2ϩ -dependent exocytosis (75), we verified whether NMDARdependent [Ca 2ϩ ] i elevation in INS-1 cells contained a D-serinedependent component. To this end, 2MeSADP was applied in the presence of the D-serine catabolic enzyme, D-amino acid oxidase (0.5 units/mg protein, see Ref. 31). The NMDAR-dependent [Ca 2ϩ ] i elevation was in this case 86.7 Ϯ 12.4% of the elevation observed in the absence of D-amino acid oxidase (n ϭ 7, t test p Ͼ 0.85; not significant), suggesting that D-serine does not contribute significantly to the response. Altogether these observations demonstrate that activation of astrocytic P2Y1 receptors induces Ca 2ϩ -dependent glutamate exocytosis and that this requires Ca 2ϩ release from the intracellular stores.
TNF␣ Signaling via TNFR1 Controls P2Y1R-dependent Glutamate Release-The next intriguing observation was that purinergic P2Y1R signaling in astrocytes induces release of TNF␣. Stimulation of astrocyte cultures with ATP (100 M, 5 min) resulted in a significant enhancement of the extracellular levels of this cytokine, as measured by a sensitive enzyme-linked immunosorbent assay (Fig. 3a). The effect of the nucleotide was blocked by A3P5PS (100 M) and reproduced by 2MeSADP (10 M). In contrast, agonists of other P2 receptor subtypes, UTP (100 M) and BzATP (100 M), were devoid of any effect (Fig.  3a). Given that BB3103 (3 M), an inhibitor of the TNF␣-converting enzyme (TACE), abolished P2Y1R-dependent TNF␣ release, the cytokine must be shed from its transmembrane precursor via the metalloproteinase activity of this enzyme.
TNF␣ Mediates a Component of P2Y1R-dependent [Ca 2ϩ ] i Elevation-We next investigated by which mechanism(s) TNF␣ could influence P2Y1R-dependent glutamate release. Ca 2ϩ signaling represents a converging pathway through which different stimuli induce glutamate release from astrocytes (14,33). Therefore, we verified whether TNF␣-dependent signaling has a role in the Ca 2ϩ response evoked by P2Y1R activation.

P2Y1-evoked Glutamate Exocytosis from Astrocytes
pharmacological experiments in cultures of normal rat astrocytes exposed to the TNF␣ scavenger, sTNFR1 (200 ng/ml; Fig.  3f, right). In control experiments without sTNFR1, astrocytes stimulated by two sequential applications of 2MeSADP, separated by a 20-min interval, consistently responded with highly reproducible [Ca 2ϩ ] i elevations (data not shown). When the first challenge with the P2Y1R agonist was performed in the presence of sTNFR1, the amplitude of the Ca 2ϩ response was lower (60.6 Ϯ 7.8%; n ϭ 7; t test p Ͻ 0.01) than the amplitude observed at the second challenge, performed after washing out sTNFR1.
When analyzing the kinetics of the 2MeSADP-evoked [Ca 2ϩ ] i elevations, we noticed that sTNFR1 caused a significant delay in the time-to-peak (mean value, 1.34 Ϯ 0.64 s in control condition versus 2.24 Ϯ 0.98 s in the presence of sTNFR1; n ϭ 7; t test p Ͻ 0.01; Fig. 3f, right, inset). Overall, these data suggest that TNF␣ signaling contributes to the amplitude and kinetics of the astrocyte [Ca 2ϩ ] i elevations in response to P2Y1R stimulation.
Prostaglandins Are Involved Downstream to the TNF␣-Prostaglandins can be rapidly formed and released in association with astrocytic [Ca 2ϩ ] i rises (9,14,76). In addition to their effects on the vasculature (9), they play a key role in the control of glutamate release in response to glutamatergic stimulation (14,43). So we checked whether prostaglandins are involved in the P2Y1R-dependent signaling pathway. Indeed, by measuring PGE 2 with a sensitive EIA, we found that both ATP (100 M) and 2MeSADP (10 M) increased the extracellular levels of this prostaglandin (Fig. 4a). The TACE inhibitor BB3103 (3 M) strongly inhibited the 2MeSADP-dependent effect (Fig. 4a), indicating that PGE 2 formation largely depends on TNF␣ shedding. Prostaglandins contribute to P2Y1R-evoked glutamate release, because the phenomenon is strongly inhibited in the presence of cyclooxygenase blockers, either indomethacin (INDO, 1 M) or aspirin (ASA, 10 M) (Fig. 4, b and c).
P2Y1R Coupling to Glutamate Release Is Restricted to Astrocytes-We next conducted studies in mixed hippocampal cultures that stain positively for neuronal, astrocytic, and microglial markers (33). Administration of 2MeSADP (10 M) to these mixed cultures induced a rapid glutamate release response (Table 1). Interestingly, the extent of such a response was identical to that of the response evoked in astrocyte cultures plated at the same density (see Fig. 1b). This observation suggests that the neuronal cells do not contribute significantly to P2Y1R-evoked glutamate release. In agreement, no detectable glutamate release was observed in neuron-enriched cultures stimulated with 2MeSADP; moreover, in mixed cultures depleted of their neuronal component (61), P2Y1R-evoked glutamate release was not reduced (Table 1). A possible contribution of microglia was also excluded because the glutamate release response evoked by 2MeSADP in astrocyte cultures was not amplified by addition of microglial cells (10 -12%) to the cultures, even when the microglia had been preactivated by exposure to lipopolysaccharide (24 h). In agreement, lipopolysaccharide-activated microglia-pure cultures did not show any detectable glutamate release in response to 2MeSADP (Table 1). Therefore, in our hippocampal cultures, P2Y1R-dependent glutamate release appears to be astrocyte-specific.

P2Y1R Stimulation Induces Ca 2ϩ -dependent Glutamate Release from Hippocampal Slices; Roles for TNF␣ and
Prostaglandins-Next, we addressed whether the P2Y1R-dependent signaling system identified in cultured astrocytes is operant also in tissue preparations. At first, we tested whether purinergic stimulation induces glutamate release from acute hippocampal slices prepared from adult rats. Indeed, ATP (100 M) evoked rapid glutamate release responses from the slices (n ϭ 7; Fig. 5a). Like in astrocyte cultures, 10 M 2MeSADP mimicked the effect of 100 M ATP (n ϭ 5), and the responses to both the P2Y1R agonist and the endogenous nucleotide were blocked by the P2Y1R antagonist, A3P5PS (100 M, n ϭ 5), indicating that the observed glutamate release in situ is mediated by the P2Y1 subtype of purinergic receptors. Preincuba-  tion of the slices with the Ca 2ϩ chelator, BAPTA/AM (50 M, n ϭ 4) (Fig. 5a), abolished 2MeSADP-evoked release, revealing the Ca 2ϩ dependence of the process. In the hippocampus, two distinct Ca 2ϩ -dependent processes that may be responsible for glutamate release have been described to date, i.e. classical exocytosis from neuronal terminals and release from astrocytes (14,33,37). To distinguish between them, we compared the characteristics of 2MeSADP-evoked glutamate release with those of synaptic glutamate exocytosis evoked by high K ϩ stimulation (14,33). The two processes were found to differ fundamentally in several aspects. First of all, when utilizing hippocampal slices from TNF␣ Ϫ/Ϫ mice (Fig. 5b), we could not see appreciable glutamate release responses to 2MeSADP (10 M) stimulation, whereas responses to high K ϩ (20 mM) were similar to those observed in slices of TNF␣ ϩ/ϩ mice (TNF␣ ϩ/ϩ , 3.32 Ϯ 0.10 nmol/mg protein, n ϭ 7; TNF␣ Ϫ/Ϫ , 3.17 Ϯ 0.07 nmol/mg protein, n ϭ 6). The same dichotomy was observed using slices from TNFR1 Ϫ/Ϫ animals (Fig. 5b). Likewise, blocking cyclooxygenases with INDO (1 M) reduced by 6-fold the 2-MeSADP-induced glutamate release from rat hippocampal slices (2MeSADP, 1.25 Ϯ 0.1 nmol/mg protein; 2MeSAADP ϩ indomethacin, 0.2 Ϯ 0.05 nmol/mg protein; n ϭ 4) but left intact high K ϩ -induced glutamate release (high K ϩ , 1.93 Ϯ 0.1 nmol/mg protein; high K ϩ ϩ indomethacin, 1.95 Ϯ 0.2 nmol/mg protein; n ϭ 4). Finally, although both processes were found to be sensitive to the vesicular exocytosis blocker Baf A1, the time course of inhibition was different in the two cases. High K ϩ -evoked glutamate release was blocked within 2 h of exposure of the slices to Baf A1 (1 M), whereas 2MeSADPevoked release was blocked by about 50% within 6 h (n ϭ 3; *, t test p Ͻ 0.05; Fig. 5c). However, by changing the experimental protocol, we accelerated the inhibitory effect of Baf A1 and also strongly blocked the 2MeSADP-evoked process within 2 h. Because Baf A1 prevents refilling of transmitter-empty vesicles, although it is without effect on transmitter-filled vesicles (77), we reasoned that an initial stimulation of the release leading to discharge of the vesicles contents would result in an enhanced inhibitory efficacy of the drug (43). So, we decided to challenge the slices with 2MeSADP (10 M) at the beginning of the incubation period with Baf A1 (1 M); this procedure was sufficient to suppress the release in response to a second challenge with the P2Y1R agonist after 2 h of exposure to Baf A1 (n ϭ 7; p Ͻ 0.05; see Fig. 5d). Inhibition was because of the specific action of Baf A1, because repeated challenges with 2MeSADP at a 2-h For each condition n ϭ 4 -7. b, left, representative fluorescence traces of the glutamate release responses to sequential application of 2MeSADP (10 M) and high K ϩ (KCl, 20 mM) in hippocampal slices from TNF␣ ϩ/ϩ and TNF␣ Ϫ/Ϫ mice. Notice in TNF␣ Ϫ/Ϫ mice a selective defect in the response to the P2Y1R agonist, but not in those induced by high K ϩ . Right, histograms compare the glutamate release response to 2MeSADP (10 M) and high K ϩ (KCl, 20 mM) in hippocampal slices from TNF␣ ϩ/ϩ and TNF␣ Ϫ/Ϫ mice as well as from TNFR1 ϩ/ϩ and TNFR1 Ϫ/Ϫ mice (n ϭ 6 -7 for each animal group). Responses to 2MeSADP are selectively defective in TNF␣ Ϫ/Ϫ and TNFR1 Ϫ/Ϫ mice, whereas responses to high K ϩ are identical in the four groups of mice. c, glutamate release responses to 2MeSADP (10 M, black bars) and high K ϩ (20 mM, white bars) in rat hippocampal slices preincubated with Baf A1 (1 M) for 2 or 6 h (n ϭ 3 for each time). Data are expressed as percentage of the effect observed at identical times in sister slices not exposed to Baf A1. Baf A1 maximally inhibits high K ϩ -induced release within 2 h, whereas it partially inhibits 2MeSADP-evoked release after 6 h (*, p Ͻ 0.05 versus control; t test). d, prestimulation of the slices with 2MeSADP speeds Baf A1 inhibition. Histograms represent glutamate release responses to two sequential stimulations with 2MeSADP, the first performed at the beginning of the experiment (1st: 0 h) and the second 2 h later (2nd: 2 h). Responses in naive slices (Ctrl, n ϭ 6) and in slices exposed to Baf A1 (Baf A1, n ϭ 7) are compared; in the latter, the 2nd response to 2MeSADP is abolished, whereas the former is identical to the 1st one. Therefore, the inhibitory potency of Baf A1 is greatly enhanced (compare with c) by the initial stimulation with 2MeSADP. e, glutamate release responses in rat hippocampal synaptosomes; the representative fluorescence traces show that the same synaptosomal preparation responds with massive glutamate release to high K ϩ (KCl, 20 mM, n ϭ 12) but does not show any release in response to 2MeSADP (10 M, n ϭ 6) or to TNF␣ (30 ng/ml, n ϭ 6).
interval in slices not incubated with Baf A1 always gave identical glutamate release responses (n ϭ 6; see Fig. 5d). This result supports the exocytotic nature of the glutamate release process induced by P2Y1R stimulation in situ. 4 Overall, the 2MeSADPinduced glutamate release from hippocampal slices has features corresponding to those of the P2Y1R-dependent effect in cultured astrocytes and distinct from exocytosis from neuronal terminals. P2Y1R Stimulation Does Not Induce Glutamate Release from Hippocampal Synaptosomes-To more directly exclude involvement of synaptic exocytosis, we tested whether hippocampal synaptosomal preparations, enriched in isolated neuronal terminals, released glutamate in response to activation of the P2Y1R-dependent cascade (Fig. 5e). Stimulation of nerve terminals with high concentrations of either 2MeSADP (10 -100 M) or TNF␣ (30 -100 ng/ml) did not induce detectable glutamate release. The same nerve terminal preparations, however, promptly responded with massive glutamate release to depolarization with high K ϩ (20 mM, 3.17 Ϯ 0.2 nmol/mg protein, n ϭ 6).

DISCUSSION
This study identifies the coupling of purinergic P2Y1 receptors to glutamate release in astrocytes. The stimulus-secretion coupling is Ca 2ϩ -dependent, mediated by Ca 2ϩ release from internal stores, and occurs via regulated exocytosis, as demonstrated by real time imaging of VGLUT-expressing vesicle fusions in response to P2Y1R stimulation. Both [Ca 2ϩ ] i elevation and glutamate release appear to be under the control of TNF␣ signaling, acting via its p55 receptor, TNFR1, and downstream prostaglandins. Finally, we provide strong evidence that the P2Y1R-glutamate release coupling is restricted to astrocytes, both in hippocampal cell cultures and in acute hippocampal slices.
Glutamate release from astrocytes in response to ATP has been reported in several studies (4). The underlying mechanisms remain controversial, however, and may actually be multiple, mediated by distinct purinergic receptor subtypes. Astrocytes express several pharmacologically distinguishable P2YR (59,79) as well as several ionotropic P2XR subunits, including P2X7 (49). The glutamate release process identified here in response to the nucleotide appears to be specifically mediated by P2Y1R. It is faithfully reproduced by 2-MeSADP, a rather selective P2Y1R agonist (59), and it is blocked by A3P5PS, a specific P2Y1R antagonist (80). Roles for other P2YR subtypes, such as P2Y2, -4 and -6, are excluded because UTP, a potent activator of these receptors, was unable to mimic the effect of 2MeSADP. Moreover, the ATP-evoked glutamate release is sensitive to pyridoxal phosphate-6-azo(benzene-2,4-disulfonic acid) tetrasodium salt, whereas P2Y4R and P2Y6R are largely insensitive to this purinergic antagonist (69). A previous report identified ionotropic P2X7R as source of glutamate release from astrocytes (49). In many cell types, P2X7R form homomeric complexes that, after prolonged exposure to high concentrations of ATP, open large ion channels permeable to mol-ecules up to 900 Da in size, including glutamate and aspartate (81). The characteristics of P2X7R-dependent glutamate release in astrocytes are fully distinct from those of the P2Y1Rdependent process identified here: P2X7R-induced release is Ca 2ϩ -independent, nonvesicular, and catalyzed by reducing the Ca 2ϩ and Mg 2ϩ ion concentration in the cell-bathing medium, a procedure that increases the affinity of P2X7R for their ligands (58). However, in normal extracellular medium, as used in our experiments, we could not observe significant glutamate release in response to the P2X7R agonist BzATP (100 M). This is perhaps not surprising given that astrocyte P2X7R are only weakly activated in physiological media (49). Therefore, in view of the peculiar gating properties of P2X7R channels and the weak agonist potency of ATP at these receptors (81), it remains to be demonstrated that P2X7R-dependent glutamate release occurs under physiological conditions. However, this pathway could be an important player in pathological states such as spinal cord injury and could contribute to excitotoxic neuronal death (82). ATP-evoked glutamate release through a Ca 2ϩ -dependent pathway has been ascribed to metabotropic P2YR acting via Ca 2ϩ release from internal stores. The downstream events remain controversial, however. Kang et al. (35) have shown that inositol 1,4,5-trisphosphate formation induces exocytotic release of glutamate from hippocampal astrocytes, corroborating previous findings in cultured cells (37,39). However, another study by the same authors (36) reported that ATP or UTP evokes a glutamate release process in cultured astrocytes that is, at the same time, dependent on Ca 2ϩ release from internal stores and insensitive to exocytosis inhibitors. The authors propose that the release occurs via volume-regulated anion channels, as they find that these channels can permeate high millimolar glutamate concentrations (36). Whether the channels would drive significant glutamate release at the physiological cytosolic concentrations of the amino acid, 50-fold lower than those used in the study (83), 5 is not yet known. The P2Y1R-mediated release identified here is also clearly distinct from the above process, not just because it occurs by vesicular exocytosis but also because it displays distinct purinergic pharmacology (i.e. preferential activation by nucleoside diphosphates and unresponsiveness to UTP), and because it is inhibited by cyclooxygenase blockers, although the process reported by Takano et al. (36) is not. We cannot easily explain the different properties of ATP-evoked Ca 2ϩdependent glutamate release in the two studies. One possibility is that different culture conditions lead to the expression of different P2YR subtypes in the cultures and that these subtypes activate distinct signaling pathways. Importantly, all the properties of the P2Y1R-dependent pathway observed in cell cultures were seen also in hippocampal slices, indicating that this pathway cannot be an artifact of the culture conditions.
The characteristics of P2Y1R-induced glutamate release, i.e. its activation by low ATP concentrations (EC 50 , 6 M) and probably much lower ADP concentrations, in view of the preferential affinity of P2Y1R for nucleoside diphosphates (56), its sophisticated Ca 2ϩ regulation involving TNF␣ and prostag-landin signaling (see below), and the quantal nature of the release indicate that this pathway is potentially implicated in physiological processes. Indeed, we now have direct evidence that astrocyte P2Y1R receptors are activated during normal synaptic transmission in the hippocampal dentate gyrus and control synaptic strength via glutamate release. 4 The use of TIRF microscopy allowed us to directly document P2Y1R-dependent exocytosis of glutamatergic vesicles and to define several properties of this process. The limited penetration of the evanescent wave of excitation beyond the astrocyte plasma membrane (about 80 nm in our experiments) selectively visualized a population of double-fluorescent vesicles loaded with the dye AO and expressing fluorescent vesicular glutamate transporters (VGLUT2-EGFP). Such vesicles were therefore docked or immediately adjacent to the astrocyte plasma membrane and most likely represented the release-ready pool. The fluorescence spots had a diameter of 325 Ϯ 96 nm, comparable with that of classical synaptic microvesicles seen under TIRF illumination (67). This observation is in line with previous colocalization studies (38) that concluded that the spots represent the population of VGLUT2-and cellubrevin-bearing clear synaptic-like microvesicles identified at the electron microscopic level (37,38).
Single fusion events were monitored as "AO flashes," a sequence of stereotyped fluorescence changes (brightening, spreading, and eventually disappearing) that signal discharge and then diffusion of the AO cargo in the extracellular medium upon fusion of the vesicle membrane with the plasma membrane (4,37,73,74). By this approach we directly counted in individual astrocytes the number of VGLUT-expressing vesicles undergoing exocytosis in response to application of 2MeSADP. About 55% of the vesicles present in the TIRF field (a total of 550 vesicles) were exocytosed within 5 s. Interestingly, in experiments on isolated astrocyte-INS-1 cell pairs, a lower proportion (30%) of vesicles was exocytosed and on a shorter time scale (about 1 s). In both high density astrocyte cultures and in isolated astrocyte-INS-1 cell pairs the maximal fusion rate was achieved within 350 ms from the start of 2MeSADP application. Therefore, the density of the cultures seems to influence the entity and duration of the exocytotic process, particularly its slower phase. One possible explanation is that late exocytic events in high density cultures are caused by intercellular cross-talk and paracrine amplification of the primary stimulus.
The secretion of VGLUT-expressing vesicles was accompanied by glutamate release as detected by temporally correlated NMDAR-dependent [Ca 2ϩ ] i elevations in co-cultured INS-1 cells (37). We verified the correspondence between the amount of glutamate released in response to 2MeSADP or ATP as measured by the GDH assay ( Fig. 1b; 0.2 fmol/cell in our high density cultures) and the amount of glutamate released by vesicular exocytosis. Because the region of an astrocyte monitored by TIRF illumination represents about 1/30 of its total surface (calculated with Imaris software), and assuming that the fusion events in this region are representative of the events occurring in the rest of the cell, we estimate that each astrocyte secretes about 9000 vesicles in response to 2MeSADP. The reported diameter of VGLUT2-expressing vesicles in cultured astrocytes is varied, ranging between 30 and 80 nm (37,38). Taking an average diameter of 55 nm and assuming that the glutamate concentration in astrocytic vesicles is 100 -400 mM as in synaptic vesicles (84,85), the amino acid content of each vesicle will be 8.7-34.8 ϫ 10 Ϫ6 fmol, corresponding to a P2Y1R-evoked release of 0.078 -0.312 fmol/cell. This estimation is in good agreement with the value obtained in GDH assays. It differs, however, from the value calculated by Takano et al. (36). One of the problems with those calculations is that the authors did not take into account that the region monitored under TIRF illumination represents a small part of the astrocyte surface in three dimensions.
AO flashes from VGLUT-expressing vesicles as well as NMDAR-dependent responses in INS-1 cells were abolished in thepresenceofTeNT,confirmingthatbothdependonaSNAREdependent process. They were similarly abolished in the presence of CPA, a blocker of the endoplasmic reticulum Ca 2ϩ -ATPase, indicating that the secretory process is triggered by Ca 2ϩ release from the internal stores, consistent with the reported coupling of P2Y1R with inositol 1,4,5-trisphosphatedependent [Ca 2ϩ ] i elevations (86). Additional transductive events, notably those depending on TNF␣ and PGs, exert a control on P2Y1R-dependent glutamate release. At least three lines of evidence support the implication of these mediators as follows: 1) both TNF␣ and PGE 2 are released from astrocytes in response to ATP-dependent stimulation of P2Y1R (PGE 2 release is TACE-dependent and therefore occurs downstream to the TNF␣ release (see also Ref. 33); 2) P2Y1R-evoked glutamate release is strongly reduced in astrocyte cultures and in acute hippocampal slices treated with pharmacological blockers of either TNF␣ or PG production; and 3) the release is similarly reduced in preparations from TNF Ϫ/Ϫ and TNFR1 Ϫ/Ϫ mice.
Although the above observations highlight a critical control of TNF␣ and PGs on P2Y1R-dependent glutamate release, we cannot conclude that these mediators are indispensable for glutamate release because conditions abolishing TNF␣ or PG signaling strongly reduced but did not completely block glutamate release. The residual release accounts for 10 -25% of the normal release in astrocyte cultures and for an even smaller proportion in acute slices. Incomplete penetration of GDH in the slices and competing uptake could, however, explain the latter result.
The specific roles of TNF␣ and PGs in the stimulus-secretion coupling mechanism remain to be established. Signaling by these mediators could contribute to one or several of the following processes: (a) Ca 2ϩ release from the internal stores triggering vesicle fusion; (b) priming of the secretory process itself; and (c) amplification of the whole cascade by autocrine/paracrine loops. For instance, PGE 2 is known to induce [Ca 2ϩ ] i elevation in astrocytes, possibly by binding to surface E-series prostaglandin (EP) receptors (14,32,76).
By comparing 2MeSADP-evoked [Ca 2ϩ ] i elevations in normal astrocytes and in astrocytes from TNF Ϫ/Ϫ and TNFR1 Ϫ/Ϫ mice or in astrocytes where sTNFR1 neutralized released TNF␣, we observed specific changes in the initial phase of the Ca 2ϩ transients, notably a 30 -40% reduction of the Ca 2ϩ peak accompanied by a slowing down of the time-to-peak. These observations suggest that TNF␣ signaling, although not indispensable for the generation of P2Y1R-dependent [Ca 2ϩ ] i eleva-

P2Y1-evoked Glutamate Exocytosis from Astrocytes
tions, contributes to their shaping. The impact of this contribution to the overall glutamate release process remains to be established with approaches offering higher temporal-spatial resolution.
Another aspect that needs future clarification is the time course of P2Y1R-evoked TNF␣ and PG release. In order for these agents to control glutamate secretion, their release must occur in the time scale of milliseconds. Unfortunately, the assays used in this study lack adequate temporal resolution and cannot confirm this possibility. A recent study using appropriate sniffer cells, detected PGE 2 release from astrocytes within 2 s from stimulation (76). As for TNF␣, the soluble ectodomain of the protein can be rapidly released from its transmembrane precursor by the action of TACE, a specific metalloproteinase constitutively expressed at the surface of the astrocytes (87). Based on the present evidence, however, we cannot exclude the possibility that tonic basal release of TNF␣ and PGs sets the appropriate conditions for efficient [Ca 2ϩ ] i elevations and glutamate release in response to 2MeSADP stimulation. Indeed, evidence for tonic TNF␣ production in astrocyte cultures exists (88,89). In this perspective, P2Y1R-stimulated release of the cytokine could act to reinforce the tonic effect.
Another relevant observation of this study is that P2Y1R are coupled to glutamate release selectively in astrocytes. By applying 2MeSADP to different types of hippocampal cultures, enriched in neurons, microglia, or astrocytes, we could detect glutamate release only from the astrocyte cultures. Moreover, by applying the P2Y1R agonist to mixed neuron-glia cultures or to co-cultures of astrocytes and microglia, we never observed higher glutamate release than in cultures containing only astrocytes. In the in situ experiments we found that P2Y1R-evoked glutamate release had properties analogous to the process in cultured astrocytes and different from synaptic glutamate release evoked by high K ϩ stimulation, despite that both processes were Ca 2ϩ -dependent. Indeed, 1) high K ϩ but not 2MeSADP induced glutamate release from hippocampal synaptosomes, despite that this preparation apparently contains P2Y1R (90); 2) P2Y1R-evoked glutamate release from hippocampal slices, but not high K ϩ -evoked release, was sensitive to inhibition of TNF␣ and PG signaling; and 3) both processes were inhibited by Baf A1 but with a different time course. The sensitivity to Baf A1 of the P2Y1R-evoked release in situ is a relevant finding as it corroborates the evidence that the process is vesicular in nature. The slower rate of inhibition with respect to the high K ϩ -evoked process highlights functional differences with synaptic release, and possibly the fact that spontaneous exocytosis of astrocytic vesicles occurs less frequently than spontaneous exocytic events at the synapses.
In the hippocampus, P2Y1R are expressed in both neurons and astrocytes, although not homogeneously. In the area CA1, they are present in astrocytes of the stratum radiatum, pyramidal neurons, and GABAergic interneurons (8,68,91), whereas in the dentate gyrus they are expressed mostly, if not exclusively, in astrocytes. At the ultrastructural level, P2Y1R in the dentate molecular layer are observed in astrocytic processes surrounding asymmetric synapses. 4 Such localization strongly suggests that they can be activated by ATP released during synaptic transmission. There is increasing evidence that ATP par-ticipates in the excitatory transmission in the hippocampus; for example, a P2XR-dependent component of excitatory postsynaptic currents (EPSCs) was identified at various hippocampal synapses (92,93). Moreover, work in hippocampal slices shows that ATP released as result of Schaffer collateral stimulation induces [Ca 2ϩ ] i increases in astrocytes through P2Y1R activation (8). As we demonstrate here that P2Y1R trigger glutamate exocytosis from astrocytes, activation of this astrocyte receptor may serve important physiological regulatory functions. Indeed, we now have evidence that this pathway is involved in a positive control of perforant path-to-granule cell synapses. 4 In addition, astrocyte P2Y1R can be activated by ATP released from the astrocytes during propagation of intercellular Ca 2ϩ waves (6,7,59). The resulting glutamate release may be part of the mechanism sustaining signal propagation in the glial network (78) while at the same time providing an interface with the neuronal circuitry.