Astrocytes gate Hebbian synaptic plasticity in the striatum

Astrocytes, via excitatory amino-acid transporter type-2 (EAAT2), are the major sink for released glutamate and contribute to set the strength and timing of synaptic inputs. The conditions required for the emergence of Hebbian plasticity from distributed neural activity remain elusive. Here, we investigate the role of EAAT2 in the expression of a major physiologically relevant form of Hebbian learning, spike timing-dependent plasticity (STDP). We find that a transient blockade of EAAT2 disrupts the temporal contingency required for Hebbian synaptic plasticity. Indeed, STDP is replaced by aberrant non-timing-dependent plasticity occurring for uncorrelated events. Conversely, EAAT2 overexpression impairs the detection of correlated activity and precludes STDP expression. Our findings demonstrate that EAAT2 sets the appropriate glutamate dynamics for the optimal temporal contingency between pre- and postsynaptic activity required for STDP emergence, and highlight the role of astrocytes as gatekeepers for Hebbian synaptic plasticity.

F ast excitatory transmission at central synapses is dependent on glutamate dynamics. Astrocytes play a major role in the precise regulation of glutamate concentration in the extracellular fluid, via their high-affinity glutamate transporters (excitatory amino acid transporters, EAATs), which determine the extent of receptor stimulation by terminating the neurotransmitter signal [1][2][3][4] . Among the five subtypes of EAATs, the largest proportion of glutamate uptake (95%) in the adult forebrain is mediated by the astrocytic EAAT2 (refs 5-8). Specific deletion of EAAT2 in astrocytes (which express 90% of total EAAT2) revealed that astrocytic EAAT2 contributes to most of the glutamate uptake and that specific EAAT2 deletion in neurons has to this day unidentified consequences 8,9 . Decreased levels of EAAT2 associated with increased ambient glutamate have been observed in neurodegenerative and psychiatric diseases 7,10,11 and in chronic exposure to drugs of abuse 12 .
EAAT2 is of crucial importance in the maintenance of low glutamate concentrations and for ensuring a high signal-to-noise ratio in synaptic and extrasynaptic transmission 4,13 . Astrocytic glutamate uptake via EAAT2 affects both the fast component of the synaptic glutamate transient and slower components by limiting the spill-out to extrasynaptic receptors and the spillover to neighboring synapses [13][14][15] . Although, astrocytic glutamate transporters are not overwhelmed on physiological activity 16 , synaptic isolation is never reached 17 . Thus, fast removal of glutamate by astrocytes contributes to set the strength and timing of synaptic inputs by controlling periand extrasynaptic receptor activation during neuronal activity 18 .
According to Hebbian theory, neural networks refine their connectivity by patterned firing of action potentials in pre-and postsynaptic neurons 19 . Spike timing-dependent plasticity (STDP) is a synaptic Hebbian learning rule that has been the focus of considerable attention in experimental 19,20 and computational 21,22 neuroscience. STDP relies on the precise order and the millisecond timing of the paired activities on either side of the synapse 19,20 . However, the conditions required for the emergence of STDP from distributed neural activity remain unclear.
Temporal coding via STDP may be essential for the role of the striatum in learning of motor sequences in which sensory and motor events are associated in a precise time sequence. Corticostriatal longterm plasticity provides a fundamental mechanism for the function of the basal ganglia in procedural learning 23,24 . MSNs act as detectors of distributed patterns of cortical and thalamic activity. Thus, the physiological or pathological regulation of EAAT2 expression should play a major role in information processing in the basal ganglia, which is based on a precise time-coding process. EAAT2 is highly expressed in the striatum 7 and specific knockout of astrocytic EAAT2 leads to pathological repetitive behaviours due to corticostriatal dysfunction 25 . We have previously shown, by dual astrocyteneuron recordings, that EAAT2 controls corticostriatal transmission and short-term plasticity, and increases the strength of cortical input filtering by the striatum 26 . Here we questioned the role of astrocytes (via EAAT2) in the control of Hebbian plasticity expression, and, more specifically, corticostriatal STDP. We find that under a transient blockade of EAAT2, a non-Hebbian form of plasticity occurring for uncorrelated events replaces STDP. By contrast, EAAT2 overexpression impairs the detection of correlated pre-and postsynaptic activity by MSNs, resulting in the absence of plasticity. We demonstrate here that astrocytes, via EAAT2, set the appropriate glutamate dynamics for the emergence and the establishment of synaptic Hebbian learning rule, such as STDP.

Results
Bidirectional STDP within a narrow temporal window. We investigated the effect of EAAT2 on STDP, using whole-cell recordings from striatal medium-sized spiny neurons (MSNs) in horizontal corticostriatal brain slices from juvenile rats 27 (Fig. 1a). Baseline excitatory postsynaptic currents (EPSCs) were recorded for 10 min in voltage-clamp mode and then recordings were switched to current-clamp mode to pair a single excitatory postsynaptic potential (EPSP) induced by presynaptic stimulation with a single postsynaptic spike induced by a brief depolarization of the MSN (Fig. 1b). The STDP protocol involved pairing pre-and postsynaptic stimulation with a certain fixed timing interval, Dt STDP (Dt STDP o0 indicating that postsynaptic stimulation preceded presynaptic stimulation and Dt STDP 40 indicating that presynaptic stimulation preceded postsynaptic stimulation), repeated 100 times at 1 Hz. After the STDP protocol, recordings were obtained in voltage-clamp mode, and EPSCs were monitored for 1 h.
EAAT2 gates the polarity and temporal window of STDP. Investigation of the role of astrocytic glutamate uptake in corticostriatal STDP required the transient blocking of EAAT2 during the STDP pairings (see Methods section). We considered a pharmacological approach to be most appropriate for this purpose. We previously showed, by dual astrocyte-neuron recordings, that dihydrokainate (DHK; 300 mM), a selective nontransportable inhibitor of EAAT2 (ref. 33), efficiently blocked most of the transporter-mediated currents in striatal astrocytes on corticostriatal stimulation 26 . Brief EAAT2 blockade with DHK for 5 min resulted in a marked depolarization of the recorded MSN in current-clamp mode in the absence of cortical stimulation (22±2 mV, Po0.0001, n ¼ 14; Fig. 2a). This effect was fully reversible after 15 min of DHK washout. These findings suggest that the slice contained sufficiently large amounts of glutamate to induce postsynaptic depolarization during EAAT2 blockade. DHK-induced depolarization involved AMPAR and type-I/II mGluR activation (Fig. 2a). Indeed, during the concomitant inhibition of AMPAR with CNQX (20 mM) and of type-I/II mGluR with MCPG (500 mM) no significant depolarization was observed (1.0 ± 0.3 mV, P ¼ 0.5872, n ¼ 7  We then ensured that brief (5 min) EAAT2 blockade induced no long-term change in synaptic efficacy. A stable baseline was established over a period of 10 min. We then applied DHK for 5 min without STDP pairing. As exemplified in Fig. 2b,c, we observed a transient decrease in EPSC amplitude (65 ± 9%, P ¼ 0.0105, n ¼ 6) due to AMPAR desensitization, as previously reported 26 , and an inward shift of I holding ( À 199 ± 41 pA, P ¼ 0.0022; Ri was not significantly affected, P ¼ 0.8182; Fig. 2c). These effects were fully reversed 15 min after DHK removal (93±9%, P ¼ 0.4749 and 11±15 pA, P ¼ 0.1797, respectively; Fig. 2c). Thus, transient EAAT2 blockade with DHK was compatible with the estimation of long-term changes in synaptic efficacy.
Thus, during the transient blockade of EAAT2 with either DHK or WAY-213,613, any paired activity on either side of the synapse, regardless of Dt STDP , was able to modify synaptic efficacy in the long term (Fig. 2j). This finding contrasts strongly with the STDP observed in control conditions, in which EAAT2 activity was unaffected. In conclusion, the correct functioning of EAAT2 allows the expression of a bidirectional order-dependent STDP during a restricted time window.
EAA2 blockade-induced depolarization and plasticity. We investigated whether the observed plasticity was due to the transient depolarization induced by EAAT2 blockade. For this purpose, we maintained the recorded MSNs at À 80 mV by intracellular current injection (close to MSN resting membrane potential) during STDP pairings, to prevent DHK-induced depolarization (Fig. 3a). In these conditions, pairings for À 70oDt STDP o þ 70 ms and Dt STDP ¼ ±200 ms induced LTD (77 ± 7%, P ¼ 0.0233, n ¼ 5; 5/5 cells displayed LTD; Fig. 3b) and LTP (186 ± 28%, P ¼ 0.0382, n ¼ 5; 5/5 cells displayed LTP; Fig. 3c), respectively. These results are similar to those obtained in presence of DHK when neurons were not maintained at À 80 mV (Fig. 2). Thus, the depolarization of the postsynaptic MSN induced by EAAT2 blockade does not account for the observed plasticity. We then investigated whether postsynaptic depolarization alone (without DHK) during STDP pairings mimicked the effects of transient EAAT2 blockade. When MSNs were held at À 50 mV in the absence of DHK during the STDP protocol (Fig. 3d), pairings for À 70oDt STDP o þ 70 ms and for Dt STDP ¼ ± 200 ms induced exclusively LTD (65 ± 7%, P ¼ 0.0029, n ¼ 7, 7/7 cells displayed LTD and 62 ± 6%, P ¼ 0.0011, n ¼ 7, 7/7 cells displayed LTD, respectively; Fig. 3e,f). This result is in accordance with LTD induced with sustained depolarization in visual cortex 35 , and with hippocampal depolarization-induced LTD 36 . Thus, postsynaptic depolarization in the absence of DHK is not sufficient to reproduce the effects of transient EAAT2 blockade. Glutamate spillover is, therefore, likely to contribute to the observed plasticity.
Given the involvement of VSCCs in the LTD observed under EAAT2 blockade, we investigated the calcium dependence of LTD at the level of the recorded MSN. To do so, we delivered intracellularly a fast calcium buffer, BAPTA, (i-BAPTA, 10 mM) through the patch-clamp pipette in the recorded MSN. Under EAAT2 blockade, i-BAPTA had no effect on LTD (77 ± 9%, P ¼ 0.0482, n ¼ 7; 5/7 cells displayed LTD at À 70oDt STDP o þ 70 ms; Fig. 4b). Thus, LTD observed under EAAT2 blockade is not dependent on postsynaptic MSN calcium. These results indicate that network effects are involved in LTD expression. They also suggest that VSCCs involved are located on neurons other than the recorded MSN and are activated during EAAT2 blockade, due to glutamate spillover-induced depolarization.
We then investigated the involvement of inhibitory networks in LTD. DHK-induced depolarization would also affect GABAergic interneurons resulting in an increased inhibitory tone 38 . Thus, the observed LTD might arguably arise from an increase in GABA release.
Cortical stimulation (of an intensity similar to that used for STDP pairings) evoked action potentials in all recorded FS cells whereas MSNs displayed subthreshold EPSPs (Fig. 4f). Thus, DHK application leads to the recruitment of GABAergic interneurons, resulting in an increase of the inhibitory weight exerted on the recorded MSN. An increase in inhibitory drive may, therefore, promote LTD.
We then bath-applied picrotoxin (50 mM) to investigate the involvement of GABAergic networks in LTD. For pairings at À 70oDt STDP o þ 70 ms under EAAT2 blockade, picrotoxin application prevented LTD, instead promoting LTP (202 ± 20%, P ¼ 0.0075, n ¼ 6; 6/6 cells displayed LTP; Fig. 4g). These findings suggest that LTD was dependent on GABA A R activation. Thus, an increase in inhibitory transmission, probably due to the recruitment of GABAergic interneurons under DHK treatment, is responsible for LTD. Surprisingly, the prevention of this GABAergic inhibition by picrotoxin did not result in the expected lack of plasticity. Instead, it promoted LTP. We analysed the involvement of GABAergic circuits in LTD expression further, by inhibiting GABAergic transmission during transient DHK application. Co-application of gabazine (10 mM; with effects readily reversible by washout) and DHK prevented the expression of plasticity (94 ± 3%, P ¼ 0.0974, n ¼ 5; 1/5 cells displayed LTD; Fig. 4h). Thus, GABAergic transmission during STDP pairings is determinant for LTD induction under transient EAAT2 blockade.
The LTD observed under transient EAAT2 blockade, for pairings at À 70oDt STDP o þ 70 ms, is, thus, dependent on the activation of VSCCs, probably located on striatal GABAergic interneurons. The blockade of GABAergic transmission revealed potent LTP, similar to that observed for uncorrelated pairings ( À 500oDt STDP o À 70 ms and þ 70oDt STDP o þ 500 ms). Thus, an impairment of EAAT2 function leads to LTP over the entire range of Dt STDP , with the exception of a narrow time window ( À 70oDt STDP o þ 70 ms), during which GABAergic microcircuits take over LTP and impose LTD.
The GluN2B subunit is predominantly expressed at extrasynaptic NMDARs but it has also been identified in synaptic NMDARs 41 . We applied memantine (10 mM), a low-affinity uncompetitive NMDAR antagonist that acts as an open-channel blocker with a fast off-rate (see Methods section). Memantine preferentially blocks extrasynaptic NMDARs, without affecting synaptic transmission. Indeed, memantine blocks with a greater extend extrasynaptic NMDARs that are activated due to a low but prolonged elevation of glutamate concentration. By contrast, memantine is relatively inefficient to block NMDARs in the presence of higher synaptic concentrations of glutamate over periods of a few milliseconds, and thus does not interfere with synaptic activity 42 . For STDP during EAAT2 blockade, memantine treatment prevented LTP, as no significant plasticity was observed (99 ± 5%, P ¼ 0.8302, n ¼ 5; 1/5 cells displayed LTP; Fig. 5d). Extrasynaptic GluN2B-containing NMDARs located on the postsynaptic recorded striatal MSN are thus required for LTP induction under EAAT2 blockade.
We previously showed that corticostriatal t-LTP is dependent on postsynaptic NMDARs 31 and, more precisely, that the balance between GluN2A-and GluN2B-containing NMDARs shapes Dt STDP 43 . We further investigated whether extrasynaptic NMDARs were required for t-LTP expression in control conditions, as observed for as for LTP observed under EAAT2 blockade. For this purpose, we performed STDP experiments with post-pre pairings at À 30oDt STDP o0 ms (similar to the experiments in Fig. 1c,e), in presence of memantine (10 mM); LTP was still observed (222 ± 44%, P ¼ 0.0271, n ¼ 8; 7/8 cells displayed LTP; Supplementary Fig. 3 Fig. 2g). This suggests that the induction of plasticity is not dependent on the timing or order of pre-and postsynaptic activity. Timing, order and paired activity are the cardinal features of STDP 11 . We, therefore, investigated whether the plasticity observed under transient EAAT2 blockade nevertheless followed STDP rules. We designed STDP protocols with each of 100 Dt STDP pairings chosen randomly between À 500 and þ 500 ms from a close-to-uniform distribution (see Methods section; Fig. 6). Each of the random pairing protocols (n ¼ 8) was applied both to a MSN recorded in control conditions and to a MSN subjected to transient EAAT2 blockade. An example is shown in Fig. 6a, with two MSNs (one in control conditions and the other under transient EAAT2 blockade) subje-cted to the same random pairing template. A single random Dt STDP pattern (taken from the eight different randomly generated Dt STDP patterns) did not trigger plasticity in the MSN in control conditions (the mean baseline EPSC amplitude, 119 ± 3 pA, was not significantly different from the 120 ± 5 pA   Fig. 6b illustrates that pairings were randomly distributed in a uniform manner. The application of the eight different randomly generated Dt STDP patterns resulted in no significant plasticity in control conditions (99 ± 5%, P ¼ 0.8429, n ¼ 8; 2/8 cells displayed LTP; Fig. 6c), whereas these patterns induced LTP under transient EAAT2 blockade (165±22%, P ¼ 0.0226, n ¼ 8; 7/8 cells displayed LTP; Fig. 6d). Thus, plasticity under transient EAAT2 blockade does not depend on the timing or order of the paired activity on either side of the synapse and does not, therefore, meet the criteria for STDP.
LTP under transient EAAT2 blockade does not require paired activity. The timing and order of pairings are crucial for STDP, but were not critical for the expression of plasticity under EAAT2 blockade. We investigated whether paired activity was required to induce plasticity under EAAT2 blockade, by determining whether unpaired activity consisting in postsynaptic spiking (a single postsynaptic action potential repeated 100 times at 1 Hz) without presynaptic stimulation could trigger long-term plasticity (Fig. 6e). In control conditions, this unpaired activity did not induce plasticity (101 ± 5%, P ¼ 0.9074, n ¼ 6; 1/6 cells displayed LTP; Fig. 6f). By contrast, under transient EAAT2 blockade, this unpaired activity was sufficient to trigger LTP (156 ± 17%, P ¼ 0.0152, n ¼ 7; 6/7 cells displayed LTP; Fig. 6g). This LTP was prevented by D-AP5 (50 mM) and was therefore NMDARmediated (96±10%, P ¼ 0.6693, n ¼ 6; 1/6 cells displayed LTP; Fig. 6g). Finally, we investigated whether postsynaptic suprathreshold activity was required to induce plasticity under transient EAAT2 blockade. To do so, we induced subthreshold depolarization (repeated 100 times at 1 Hz without cortical stimulation) in the recorded MSN (Supplementary Fig. 4a). This subthreshold unpaired postsynaptic stimulation was not sufficient to trigger significant plasticity when the average of all experiments performed in these conditions was considered: 118 ± 10% (P ¼ 0.1213, n ¼ 6; Supplementary Fig. 4b). However, four of the six recorded MSNs displayed significant LTP (see scatter plot in Supplementary  Fig. 4b). The postsynaptic spike therefore seems to be required for the induction of potent NMDAR-mediated LTP under transient EAAT2 blockade.
Correct functioning of EAAT2 is, therefore, required for STDP expression. A cardinal feature for STDP is that it relies on the precise time-correlation between the activities on either side of the synapse. Plasticity under transient EAAT2 blockade therefore does not meet the criteria for STDP.

Discussion
Identifying the conditions required for the expression of Hebbian plasticity, such as STDP, is essential for a better understanding of the mechanisms underlying learning and memory. Our findings demonstrate that astrocytes play a key role in the establishment of STDP, through EAAT2-mediated glutamate uptake. Indeed, EAAT2 allows translating precise pre-and postsynaptic activity into a salient time-coded message. This is a key requirement for STDP, the main characteristic of which is a high degree of sensitivity to timing 19,20 , a feature that was erased by the transient blockade of EAAT2. Under this blockade, STDP was replaced by a non-Hebbian form of plasticity that was not dependent on the timing or order of the activities on either side of the synapse and was even observed in cases of unpaired activity. By contrast, EAAT2 overexpression impaired the detection of correlated pre-and postsynaptic activity by MSNs, resulting in an absence of plasticity. Our results show that astrocytes gate the conversion from non-Hebbian to Hebbian plasticity via EAAT2, leading to the emergence of STDP (Fig. 8).
Astrocytes actively control various synaptic functions and, therefore, play a key role in the modulation of neuronal activity 11,12,45,46 . Control of neuronal computation by astrocytes is via the release and uptake of transmitters, such as glutamate. Glutamate release by astrocytes plays an important role in STDP at L4-L2/3 neocortical synapses, by controlling t-LTD through the activation of astrocytic CB 1 R 47 . By contrast, the involvement of astrocytic glutamate uptake in a time-coding paradigm, such as STDP, has never been investigated. Previous reports indicate that rate-coded plasticity, induced by low-or high-frequency stimulation (LFS and HFS) or theta-burst stimulation (TBS), is sensitive to changes in astrocytic glutamate uptake [48][49][50][51][52][53] . In addition, neuronal EAAT3 regulates the balance between TBS-LTP and LFS-LTD 54 and cerebellar LTD is dependent on the patterned expression of neuronal EAAT4 on Purkinje cells 55 . This study is, to our knowledge, the first to assess the involvement of astrocytic glutamate uptake in the expression of time-coded plasticity. STDP relies on the precise timing and order of inputs on either side of the synapse and thus constitutes a time-coding paradigm for plasticity induction 19,20 by contrast to rate-coding plasticity protocols. The detection of a temporal coincidence between pre-and postsynaptic activities is crucial for STDP expression. Astrocytic glutamate uptake is involved in setting the timing of synaptic inputs. We therefore explored the role of EAAT2 in STDP, by transiently inhibiting (with DHK or WAY-213,613) EAAT2 during STDP pairings. This allows an on-off manipulation compatible with STDP study, whereas genetic approaches (knockout) and long-lasting drug applications have potential long-term effects. DHK and WAY-213,613 have several advantages for studies of this type. In addition to their specificity for EAAT2 and their efficient washout, they are also nontransportable inhibitors of EAAT2, and this property prevents artificial increases in extracellular glutamate concentration due to hetero-exchange 33,34 . We next overexpressed EAAT2 with ceftriaxone, which has been reported to increase EAAT2 expression and activity 44 .
Astrocytic pools of EAAT2 are responsible for 90% of the glutamate uptake 8 . EAAT2 is also found on neurons but at much lower level (B10% of astrocytic EAAT2). The physiological role of neuronal EAAT2 remains uncertain based on their very low level of expression but also on their distribution in most of the axon-terminal membranes and not being concentrated in the synapses 9,56 . Specific deletion of EAAT2 in astrocytes induces dramatic effects, such as excess mortality, lower body weight and spontaneous seizures, whereas no detectable neurological abnormalities are observed with neuronal EAAT2 deletion 8,9 .
The key feature of STDP is its occurrence within a restricted time window. Uncorrelated events (430 ms) therefore fail to trigger plasticity. When EAAT2 activity is transiently impaired, an aberrant form of plasticity occurs during time windows in which plasticity is not normally observed. Uncorrelated events can induce this aberrant plasticity and are considered as pertinent events for an engram. Unlike STDP, the non-Hebbian LTP induced under transient EAAT2 blockade did not depend on the timing or order of pre-and postsynaptic activity. t-LTP has been reported to be mainly dependent on NMDARs 19 , which operate as molecular coincidence detectors 4 . By contrast, non-Hebbian LTP under EAAT2 blockade is dependent on postsynaptic GluN2B-containing NMDARs located extrasynaptically, and these receptors do not act as molecular coincident detectors. Supporting this, we found that even unpaired activity (consisting of a single postsynaptic action potential repeated 100 times at 1 Hz) induced non-Hebbian LTP under EAAT2 blockade (Fig. 6g). Molecular coincidence detectors, such as NMDARs, require concomitant signals to be activated, as in STDP, in which the postsynaptic back-propagating action potential is paired with presynaptic activity 19,20 . In the presence of transient EAAT2 blockade, this feature is lost, because a single signal, the postsynaptic back-propagating action potential removing Mg 2 þ blockade, becomes sufficient to trigger LTP, due to the high ambient glutamate levels present when EAAT2 is blocked. GABAergic microcircuits are involved in plasticity occurring at specific time window ( À 70oDt STDP o þ 70 ms) resulting in LTD (by contrast to the non-timing-dependent LTP). In the presence of DHK, GABAergic inhibition was stronger, due to the recruitment of inhibitory neurons as a result of the increase in glutamate spillover. In the presence of blockers of GABA A Rs or VSCCs, pairings for which À 70oDt STDP o þ 70 ms unmasked NMDAR-mediated LTP.
We previously described the control of STDP polarity by GABA 28 . Here, different mechanisms are involved because concomitant transient blockade of GABAergic transmission and EAAT2 led to an absence of plasticity. GABAergic circuits are efficiently recruited by cortical stimulation in the presence of DHK. We hypothesize that the NMDAR-mediated LTP observed at large Dt STDP is somehow shunted at narrow Dt STDP by an additional pool of GABA, due to the recruitment of GABAergic interneurons by cortical stimulation. Indeed, NMDAR-mediated LTP at larger Dt STDP was exclusively dependent on the postsynaptic spiking (Fig. 6g) and did not require presynaptic stimulation. By contrast, when cortical stimulation (and, thus, the recruitment of GABAergic interneurons) was paired with the postsynaptic spike for narrow Dt STDP , the increased GABAergic transmission prevented LTP expression. Thus, NMDAR-mediated LTP may be expressed only at large Dt STDP , when presynaptic stimulation occurs far from the postsynaptic spike and GABAergic evoked transmission does not interfere with LTP expression. As a result, the blocking of GABA A R transmission revealed LTP. This LTP was similar to the non-timing-dependent LTP (NMDAR-mediated) induced for large Dt STDP . Interestingly, pre-post t-LTD and post-pre t-LTP observed in control conditions are both dependent on VSCC activity 31 , but their induction itself is not dependent on GABAergic transmission 28 . Thus, the t-LTD and t-LTP evoked in control conditions involve signalling mechanisms distinct from those involved in the plasticity observed under EAAT2 blockade.
EAAT2 overexpression by ceftriaxone prevented both t-LTP and t-LTD. We verified that ceftriaxone did not alter the passive and active electrophysiological properties of MSNs, as well as corticostriatal transmission and probability of glutamate release. Ceftriaxone can also mediate the upregulation of system x c -(cystine/glutamate antiporter system) 57 , which, together with EAAT2, is involved in the maintenance of glutamate homeostasis. However, the net effect of up or downregulation and the precise balance between these two systems (glutamate uptake and export) remains to be determined. We have previously shown that the bidirectional corticostriatal STDP relies on two distinct signalling pathways 31,43 . Indeed, t-LTP is NMDAR-dependent, whereas t-LTD is mGluR-mediated. Both receptor subtypes can be localized outside the synaptic cleft 37,41 and thus compete with EAAT2 for the extracellular glutamate. Overexpression of EAAT2 by ceftriaxone is expected to enhance glutamate uptake and reduce spillover. This would reasonably result in a profound alteration of corticostriatal STDP expression. We hypothesize that enhanced glutamate clearance by EAAT2 upregulation may prevent the activation of postsynaptic NMDARs or type-ImGluRs, leading to the lack of t-LTP or t-LTD, respectively. In line with that, increases in glutamate transporter expression have been shown to alter frequency-based plasticity dependent on glutamate spillover, such as mGluR-mediated LFS-LTD and HFS-LTP in the hippocampus 53 . However, ceftriaxone has been mainly used in neurodegeneration and addiction models where extracellular glutamate levels are greatly enhanced 10,12 . Importantly, ceftriaxone abolishes the increase in glutamate spillover (assessed by NMDAR-EPSCs) in heroin-treated animals but not in control yoked saline animals 58 . In agreement with this 58 , we did not detect significant difference between NMDAR-EPSCs decay in saline and ceftriaxone-treated rats ( Supplementary  Fig. 5j-l). One possible explanation is that monitoring NMDAR-EPSCs does not allow differentiating between synaptic and extrasynaptic NMDARs. Therefore, the fraction of ambient versus synaptic glutamate detected by extrasynaptic NMDARs is difficult to assess. We hypothesize that under EAAT2 blockade, a critical number of peri-and/or extrasynaptic NMDARs are recruited leading to non-Hebbian plasticity. On the contrary, EAAT2 overexpression would reduce the pool of activated periand/or extrasynaptic NMDARs and consequently prevents STDP expression.
A few studies have reported effects of changes in EAAT2 expression on behaviour 46 . The pharmacological blockade of EAAT2 with DHK impairs spatial memory and induces depression and anhedonia and ceftriaxone has been reported to display antidepressant effects 46 . EAAT2 downregulation in striatum is also found in a rat model of depression 59 . EAAT2 KO mice exhibit seizures and premature death 6,9 . An inducible astrocytic EAAT2 knockout was recently shown to be associated with pathological repetitive behaviours and an increase in corticostriatal excitatory transmission 25 . Moreover, this phenotype was reversed by memantine treatment, confirming that excessive glutamate spillover due to EAAT2 dysfunction, deregulating the corticostriatal pathway, was responsible for the observed repetitive behaviours. These findings are consistent with our results showing that memantine prevents aberrant LTP in conditions of EAAT2 blockade. Conversely, EAAT2 overexpression has been reported to impair hippocampal learning 60 . This observation is consistent with our results showing a lack of plasticity with ceftriaxone treatment.
EAAT2 dysfunction, associated with higher ambient glutamate levels, has been observed in neurodegenerative and psychiatric diseases including Huntington's, Parkinson's, Alzheimer's and schizophrenia in which cognitive functions are impaired 7,10,11 . Chronic exposure to drugs of abuse has also been shown to induce a downregulation of EAAT2 in the nucleus accumbens 12 . EAAT2 therefore appears to be a major target for the treatment of neurological diseases and addiction (by ceftriaxone), not only to combat glutamatergic neurotoxicity but also to prevent aberrant plasticity, which could be linked to cognitive deficits [10][11][12] . Thus, our results, showing the tight control of STDP by EAAT2, are of importance for linking the expression of timing-dependent plasticity with different physiological or pathological states.
Astrocyte function is not restricted to structural and metabolic support or homeostatic and protective functions. Through glutamate uptake, astrocytes are also involved in higher brain functions, such as learning and memory 11,45,46 . We demonstrate here that EAAT2 operates over a highly controlled range to allow the emergence of bidirectional STDP. If STDP is dependent on the efficiency of glutamate uptake, then we would expect STDP expression to be controlled by the precise location and density of transporter expression, and glial synaptic coverage, which may differ considerably between brain structures and can undergo experience-dependent remodelling 61 (Fig. 8). This work thus identifies astrocytes as key players in the establishment of synaptic Hebbian learning rule, such as STDP.

Methods
Animals. All experiments were performed in accordance with the guidelines of the local animal welfare committee (Center for Interdisciplinary Research in Biology Ethics Committee) and the EU (directive 2010/63/EU). Every precaution was taken to minimize stress and the number of animals used in each series of experiments. OFA rats P18-42 (Charles River, L'Arbresle, France) were used for brain slice electrophysiology. Animals were housed in standard 12-h light/dark cycles and food and water were available ad libitum.
Brain slice preparation. Horizontal brain slices containing the somatosensory cortical area and the corresponding corticostriatal projection field were prepared as previously described 27,28,31 . Corticostriatal connections (between somatosensory cortex layer 5 and the dorsolateral striatum) are preserved in the horizontal plane. Horizontal brain slices (330 mm-thick) were prepared from rats with a vibrating blade microtome (VT1200S, Leica Micosystems, Nussloch, Germany). Brains were sliced in an ice-cold cutting solution (125 mM NaCl, 2.5 mM KCl, 25 mM glucose 25 mM NaHCO 3 , 1.25 mM NaH 2 PO 4 , 2 mM CaCl 2 , 1 mM MgCl 2 , 1 mM pyruvic acid) through which 95% O 2 /5% CO 2 was bubbled. The slices were transferred to the same solution at 34°C for 1 h and then to room temperature.
Spike timing-dependent plasticity protocols and random Dt STDP patterns. Electrical stimulations were performed with a concentric bipolar electrode (Phymep, Paris, France and CBBSE75 FHC, Bowdoin, ME, USA) placed in layer 5 of the somatosensory cortex 27 . Electrical stimulations were monophasic, at constant current (ISO-Flex stimulator, AMPI, Jerusalem, Israel). Currents were adjusted to evoke 100-400 pA EPSCs. Repetitive control stimuli were applied at 0.1 Hz. STDP protocols consisted of pairings of pre-and postsynaptic stimulations (at 1 Hz) separated by a specific time interval (Dt STDP ). Presynaptic stimulations corresponded to cortical stimulations and the postsynaptic stimulation of an action potential evoked by a depolarizing current step (30 ms duration) in MSNs. Dt STDP o0 ms for post-pre pairings, and Dt STDP 40 ms for pre-post pairings. Dt STDP ¼ ±500 ms corresponds to post-pre and pre-post pairings performed around Dt STDP ¼ À 500 ms and Dt STDP ¼ þ 500 ms. Note that for Dt STDP ¼ À 500 ms and Dt STDP ¼ þ 500 ms, the order (post-pre versus pre-post) was determined by the first pairing of the STDP protocol only, because, for the remaining pairings, the pre-and postsynaptic stimulations were separated by 500 ms and could therefore be considered as either post-pre or pre-post pairings at 1 Hz. We therefore pooled the data for Dt STDP ¼ À 500 ms and Dt STDP ¼ þ 500 ms (Dt STDP ¼ ± 500 ms), which are presented as a single average on the figures. Recordings on neurons were made over a period of 10 min at baseline, and for at least 50 min after the SDTP protocols; long-term changes in synaptic efficacy were measured for the last 10 min. We individually measured and averaged 60 successive EPSCs, comparing the last 10 min of the recording with the 10-minute baseline recording. Whole-cell recordings were made in voltage-clamp mode during baseline and for the 60 min of recording after the STDP protocol, and in current-clamp mode during STDP protocol. Experiments were excluded if input resistance (Ri) varied by more than 20%.
For the random Dt STDP patterns, we used the following algorithm (programmed in Igor Pro 6.3 software, WaveMetrics): for each pairing, we first selected a time window with a randomly selected length between 500 and 1,500 ms (with a uniform distribution) and located the presynaptic stimulation time in the middle of this window. The postsynaptic stimulation time was then randomly chosen within this window (with a uniform distribution). The Dt STDP pattern was formed by the concatenation of 100 such windows. This generated both a close-touniform distribution of the Dt STDP and a variable interval between two successive presynaptic stimulations.
were obtained from ceftriaxone-or saline-treated rats 24 h after the final injection, and prepared as described above.
Immunohistochemistry. Rats were treated for eight days with daily i.p. injection of either saline (n ¼ 4 rats) or ceftriaxone (n ¼ 4 rats), as described above. Rats were anesthetized with pentobarbital. Brains were postfixed in 4% paraformaldehyde and cut into 30 mm horizontal sections with a vibratome (Microm HM650V, ThermoScientific). Immunostaining was performed by incubating free-floating sections with a guinea pig anti-EAAT2 antibody (1:5000; AB1783, Merck Millipore) for 48 h at 4°C and then with a secondary Cyanine Cy3-conjugated antibody (1:1,000; Jackson Laboratories) dissolved in PBS 1X for 1 h. Images were acquired with the SP5 confocal system (Leica, Germany).
Data availability. All relevant data are available from the authors.