Presynaptic Spike Timing-Dependent Long-Term Depression in the Mouse Hippocampus

1One of the most interesting properties of the mammalian brain is its ability to change in response to experience. This property was termed plasticity by the Spanish neuroscientist Santiago Ramón y Cajal more than a century ago. Plasticity is involved in the organization of cortical maps during development, and in learning and memory processes in the adult (for review, see ). The most extensively studied forms of plasticity are long-term potentiation (LTP) and long-term depression (LTD) of synaptic transmission. Spike timing-dependent plasticity (STDP) is a Hebbian form of long-term synaptic plasticity found in all species studied, from insects to humans, and is a strong candidate for a synaptic mechanism underlying circuit remodeling during development as well as learning and memory. In STDP, the order and millisecond-precision relative timing of pre- and postsynaptic action potentials (spikes) determine the direction and magnitude of synaptic change. Thus, timing-dependent LTP (t-LTP) occurs when a presynaptic spike is followed by a postsynaptic spike within 10–15 ms, whereas timing-dependent LTD (t-LTD) is induced when this order is reversed (;;; for detailed reviews of STDP, see; ). t-LTP and t-LTD have been observed in neocortical slices using different stimulation frequencies from 0.1 to 20 Hz, indicating that this form of plasticity can be elicited at low frequencies of stimulation. STDP has been described in hippocampal primary dissociated cultures, hippocampal organotypic slice cultures, and acute hippocampal slices from young animals. However, whereas the mechanisms of STDP have been extensively studied at neocortical synapses, comparatively less is known about STDP at hippocampal synapses.

2Conventional single-spike STDP in the hippocampus has been reported in juvenile animals; with maturation, induction of t-LTP appears to require the pairing of Schaffer collateral stimulation with bursts of postsynaptic action potentials. Whereas hippocampal LTP and LTD induced by high-frequency stimulation (HFS) and low-frequency stimulation (LFS), respectively, have been studied in detail (see;; and references therein), surprisingly few studies have investigated the mechanisms involved in hippocampal t-LTP and t-LTD.

3To better understand the mechanisms of plasticity during development, in the present work, we studied the properties and mechanisms of t-LTP and t-LTD induction at Schaffer collateral-CA1 synapses of young (P12–P18) mouse hippocampus using whole-cell patch-clamp recordings. P12–P18 is a critical period of brain development during which much of refinement of synaptic connections occurs. We found that NMDA receptor-dependent t-LTP and t-LTD can be reliably induced by 100 pairings of presynaptic activity with single postsynaptic spikes at 0.2 Hz. However, while t-LTP requires postsynaptic GluN2A and GluN2B subunit-containing NMDA receptors, t-LTD does not, but instead requires mGlu5 receptors, CB1 receptors, and nonpostsynaptic NMDA receptors. Furthermore, we found that presynaptic calcineurin is involved in t-LTD induction and that t-LTD is presynaptically expressed as indicated by fluctuation analysis, paired-pulse ratios (PPRs) and rate of use-dependent block of NMDA receptor currents by MK-801. Whereas 2 distinct forms of LFS-induced LTD have been described, requiring either mGlu receptor (mGluR) or postsynaptic NMDA receptor signaling, our results demonstrate a new form of presynaptically expressed hippocampal t-LTD requiring postsynaptic Ca2+, astrocytic signaling, and nonpostsynaptic NMDA receptors for its induction, properties shared with t-LTD at neocortical synapses during development.

4All animal procedures were in accordance with the European Union Directive 2010/63/EU on the protection of animals used for scientific purposes and were approved by the local Ethical Committees. For the experiments performed in the United Kingdom, all animal procedures were in accordance with the UK Animals (Scientific Procedures) Act 1986. C57BL/6 mice were obtained from Harlan Laboratories (Spain) and from Harlan Laboratories UK Ltd (Bicester, UK). For most experiments, C57BL/6 mouse pups of either sex at postnatal day (P) 12–18 were used, avoiding the first 7–10 postnatal days when LTP depends on cAMP-dependent protein kinase rather than Ca2+/calmodulin-dependent protein kinase II. A total of 133 mouse pups were used. For the study of the developmental profile of t-LTD, an additional 12 male mice between P18 and P28 were used.

5Mice were anesthetized with isoflurane (2%) and decapitated for slice preparation. Hippocampal slices were prepared as described previously. Briefly, after decapitation, the whole brain containing the 2 hippocampi was removed into ice-cold solution containing (in mM): NaCl, 126; KCl, 3; NaH2PO4, 1.25; MgSO4, 2; CaCl2, 2; NaHCO3, 26 and glucose, 10 (pH 7.2, 300 mOsm L−1), positioned on the stage of a vibrating blade microtome (Leica VT1000S) and cut coronally to obtain transverse hippocampal slices (350 µm thick). The slices were maintained continuously oxygenated (95% O2/5% CO2) in this solution for at least 1 h before use. All experiments were carried out at room temperature (22–25 °C). During experiments, slices were continuously superfused with the solution described above.

6Whole-cell patch-clamp recordings were made from pyramidal cells located in the CA1 field of the hippocampus. CA1 pyramidal cells were patched under visual guidance by infrared differential interference contrast microscopy and verified to be pyramidal neurons by their characteristic voltage response to a current step protocol. Neurons were recorded in either voltage- or current-clamp configuration with a patch-clamp amplifier (Multiclamp 700B), and data were acquired using pCLAMP 10.2 software (Molecular Devices) or custom-made procedures in Igor Pro (WaveMetrics). Patch electrodes were pulled from borosilicate glass tubing, and had resistances of 4–7 MΩ when filled with (in mM): potassium gluconate, 110; HEPES, 40; NaCl, 4; ATP-Mg, 4; and GTP, 0.3 (pH 7.2–7.3, 290 mOsm L−1). Only cells with a stable resting membrane potential negative to −60 mV were included, and cells were excluded from analysis if the series resistance changed by more than 15% during the recording. All recordings were low-pass filtered at 3 kHz and acquired at 10 kHz. For plasticity experiments, EPSPs were evoked alternately in 2 input pathways, test and control, each at 0.2 Hz, by 2 monopolar stimulation electrodes placed in the “stratum radiatum” (see Fig. 1) using brief current pulses (200 µs, 0.1–0.2 mA). Stimulation was adjusted to obtain an EPSP peak amplitude of approximately 5 mV during control conditions. Pathway independence was assured by lack of cross-facilitation when the pathways were stimulated alternately with a 50 ms interval. Plasticity was assessed as changes in the slope of the EPSP, measured on its rising phase as a linear fit between time points corresponding to 25–30% and 70–75% of the peak amplitude during control conditions.

7After a stable EPSP baseline period of 10 min, the test input was paired 100 times with a single postsynaptic spike. The single postsynaptic spike was evoked by a brief somatic current pulse (5 ms, 0.1–0.5 pA). The control pathway was not stimulated during the pairing period. To induce t-LTD, the postsynaptic action potential was evoked within 18 ms before the onset of the EPSP, whereas, to induce t-LTP, the postsynaptic action potential was evoked within 10 ms after the onset of the EPSP. Both EPSP slopes and peak amplitudes were monitored for at least 30 min after each pairing episode. Presynaptic stimulation frequency remained constant throughout the experiment. Interleaved control t-LTP and t-LTD experiments were performed for each pharmacological compound tested.

8Pharmacological agents were purchased from: Sigma-Aldrich: Fluoroacetate, thapsigargin, BAPTA, bicuculline methbromide, Tricine, Zinc chloride, d-serine, and GDPβS; Tocris Bioscience: nimodipine, (+)-MK-801 maleate, d-AP5, NBQX, TTX citrate, PPDA, Ro 25-6981 maleate, MCPG, MPEP, LY341495, AM251, 2-AG, and FK506; and Abcam: UBP-141. Salts used for internal and external solutions were purchased from Sigma-Aldrich. Compounds were dissolved in H2O or Ringer solution with the exception of thapsigargin, nimodipine, PPDA, FK506, THL, AM251, and 2-AG, which were dissolved in DMSO. Vehicle (DMSO) at the concentrations used did not affect baseline EPSP amplitudes and had no other detectable effects on the neurons. When investigating the effect of pharmacological agents on plasticity, all drugs were included in the superfusion fluid or patch pipette from the start of the experiment until completion (from 0 to 50 min in a standard plasticity experiment), except for fluoroacetate, which was applied from 60 min before the start of recording. When determining the effect of a pharmacological agent on baseline condition, a stable baseline of at least 10 min was first recorded and then the drug was bath applied by switching to a different perfusion line.

9Data were analyzed using the Clampfit 10.2 software (Molecular Devices) and custom-made procedures in Igor Pro. The last 5 min of recording were used to estimate changes in synaptic efficacy compared with baseline. For PPR experiments, 2 EPSPs were evoked 40 ms apart for 30 s at baseline frequency at the beginning of the baseline recording and again 30 min after the end of the pairing protocol. The PPR was expressed as the slope of the second EPSP divided by the slope of the first EPSP. In experiments in which the effect of MK-801 on NMDA-EPSCs was evaluated, the half-life was estimated from a single exponential fit for each individual cell as the number of pulses required until the NMDA-EPSC peak amplitude was reduced to 50% of the baseline amplitude. Statistical comparisons were made using paired or unpaired Student's t-test as appropriate. P-values <0.05 were considered significant. Data are presented as mean ± SEM. Coefficient of variation (CV) analysis was done on EPSP slopes as previously described.

10First, we wanted to confirm that pairing presynaptic stimulation with single postsynaptic spikes at low frequency (0.2 Hz) is sufficient to induce both t-LTP and t-LTD at CA3-CA1 synapses. We monitored excitatory postsynaptic potentials (EPSPs) evoked by extracellular stimulation in the stratum radiatum during whole-cell recording of CA1 pyramidal cells in slices prepared from the mouse hippocampus (postnatal days 12–18) as previously described. t-LTP and t-LTD were induced in current clamp using 100 pairings of single EPSPs and single postsynaptic spikes at 0.2 Hz. A pre-before-post pairing protocol (with a postsynaptic spike occurring within 10 ms of EPSP onset) elicited robust t-LTP (147 ± 6%, n = 15), while an unpaired control pathway was unchanged (101 ± 6%, n = 15; Fig. 2A,C). Conversely, a post-before-pre pairing protocol (with a postsynaptic spike occurring ∼18 ms before presynaptic stimulation) induced robust t-LTD (73 ± 4%, n = 21), while an unpaired control pathway remained unchanged (99 ± 6%, n = 21; Fig. 2B,C).

11In slices treated with the NMDA receptor antagonist d-2-amino-5-phosphonopentanoic acid (d-AP5), a pre-before-post pairing protocol failed to induce t-LTP (102 ± 7%, n = 5; vs. interleaved controls, 143 ± 7%, n = 5; Fig. 2D,F). d-AP5 also blocked t-LTD; in d-AP5-treated slices, a post-before-pre pairing protocol did not induce t-LTD (110 ± 7%, n = 7 vs. interleaved controls, 71 ± 8%, n = 5; Fig. 2E,F). These results indicate that both t-LTP and t-LTD require NMDA receptors.

12To investigate whether the NMDA receptors that are required for t-LTP and t-LTD are located postsynaptically, we repeated the pairing experiments following loading of the use-dependent NMDA receptor channel blocker MK-801 into the postsynaptic neuron via the recording patch pipette. Consistent with previous reports at neocortical synapses, inhibiting postsynaptic NMDA receptors by including MK-801 (1 mM) in the recording pipette blocked the induction of t-LTP (101 ± 6%, n = 7; vs. interleaved controls, 150 ± 6%, n = 7; Fig. 3A,C). In contrast, t-LTD was unaffected (74 ± 6%, n = 8; vs. interleaved controls, 70 ± 6%, n = 9; Fig. 3B,C). To rule out any lack of effect due to insufficient MK-801 concentration, we repeated the experiment with 4 mM MK-801 in the recording pipette; at this concentration, as with 1 mM MK-801, t-LTP was completely prevented but t-LTD was unaffected (76 ± 6%, n = 5, vs. control t-LTD in interleaved slices 71 ± 7%, n = 5), supporting the suggestion that postsynaptic ionotropic NMDA receptors are required for t-LTP but not for t-LTD induction. To further support this conclusion, we did both pre-before-post and post-before-pre, single-spike pairing in the same cells treated with MK-801 (1 mM). Potentiation was not observed after pre-before-post pairing (104 ± 7%, n = 6 with an unpaired pathway unchanged, 101 ± 7%, n = 6; Fig. 3D,E), but subsequent post-before-pre pairing in the same pathway induced robust t-LTD (75 ± 7%, n = 6), while the unpaired pathway remained unchanged (102 ± 5%, n = 6; Fig. 3D,E). Thus, during inhibition of postsynaptic ionotropic NMDA receptors sufficient to completely block the induction of t-LTP, t-LTD could still be successfully induced.

13LFS-induced NMDA receptor-dependent hippocampal LTD has been suggested to be independent of ion flow through NMDA receptors (; but see ). To exclude the possibility that the lack of effect of postsynaptic MK-801 on t-LTD is due to postsynaptic metabotropic NMDA receptor function, we investigated the effect of extracellular application of MK-801 on the induction of t-LTD. We found that 100 µM MK-801 in the superfusate completely blocked the induction of t-LTD (96 ± 8%, n = 6), indicating that nonpostsynaptic ionotropic NMDA receptor function is required for the induction of t-LTD.

14After confirming that both t-LTP and t-LTD require ionotropic NMDA receptor function, but at different locations, we wanted to determine whether this was reflected in different NMDA receptor subunit involvement.

15To test whether t-LTP and t-LTD are dependent upon GluN2A subunit-containing receptors, we used the GluN2A subunit-preferring antagonists Zn2+ and NVP-AAM077. Both Zn2+ (300 nM) and NVP-AAM077 (100 nM) completely blocked the induction of t-LTP in P12–P18 mice (slope, 86 ± 12%, n = 9 and 103 ± 7%, n = 6, for Zn2+ and NVP-AAM077, respectively, vs. control slices, pooled, 177 ± 18%, n = 10; Fig. 4A,C), whereas t-LTD was unaffected by bath application of Zn2+ (76 ± 5%, n = 5) or NVP-AAM077 (73 ± 6%, n = 6) compared with interleaved control slices (75 ± 7%, n = 9; Fig. 4B,C). Thus, Zn2+ and NVP-AAM077 dissociated the NMDA receptor subunit requirement of plasticity at CA3-CA1 synapses. To further characterize the subunit composition of the NMDA receptors involved in t-LTP and t-LTD, we next investigated whether GluN2B subunit-containing NMDA receptors are necessary for the induction of t-LTP and t-LTD using the GluN2B subunit-selective noncompetitive antagonist Ro 25-6981 (;, see ). Ro 25-6981 (0.5 µM) almost completely blocked t-LTP induction (110 ± 10%, n = 9) versus interleaved control slices (139 ± 8%, n = 6; Fig. 4D,F), but did not significantly affect t-LTD induction (80 ± 7%, n = 11) versus interleaved control slices (75 ± 8%, n = 6; Fig. 4E,F), indicating that GluN2B subunit-containing NMDA receptors are required for t-LTP but are not obligatory for t-LTD.

16We next investigated the possible involvement of GluN2C/2D subunits, which are expressed during development. The GluN2C/2D subunits are expressed postnatally in the hippocampus, and this expression peaks around the first week of postnatal development and then decays. To test whether GluN2C/2D subunits are involved in timing-dependent plasticity at CA3-CA1 synapses in the mouse hippocampus, we used PPDA, a moderately selective, competitive antagonist at GluN2C/2D subunit-containing NMDA receptors. Bath application of PPDA (10 µM) did not affect t-LTP induction (161 ± 23%, n = 6 vs. 162 ± 11%, n = 10 in interleaved control slices; Fig. 4G,I), but completely blocked t-LTD (108 ± 8%, n = 6 vs. interleaved control slices 76 ± 6%, n = 10; Fig. 4G,H). A less potent but more selective antagonist, UBP-141, also selectively blocked t-LTD (106 ± 6%, n = 6 vs. interleaved control slices 76 ± 6%, n = 10; Fig. 4H,I) with no effect on t-LTP (175 ± 21%, n = 7, vs. interleaved control slices, 162 ± 11%, n = 10; Fig. 4G,I). These results indicate that t-LTP requires GluN2A and GluN2B, but not GluN2C/2D subunit-containing NMDA receptors, whereas t-LTD requires NMDA receptors that contain GluN2C and/or GluN2D subunits.

17While t-LTD seems not to require postsynaptic ionotropic NMDA receptors, both t-LTP and t-LTD have been shown to require postsynaptic Ca2+ at neocortical synapses. We therefore investigated the postsynaptic Ca2+ requirement of t-LTD at hippocampal CA3-CA1 synapses by loading the Ca2+ chelator BAPTA into the postsynaptic cell via the patch pipette. t-LTD was prevented when BAPTA (20 mM) was included in the recording pipette (97 ± 9%, n = 5, vs. interleaved controls, 67 ± 5%, n = 6) as was t-LTP (104 ± 8%, n = 6, vs. interleaved controls, 155 ± 7%, n = 5; Fig. 5A,B), indicating that both t-LTP and t-LTD require postsynaptic Ca2+. If t-LTD does require postsynaptic Ca2+ but NMDA receptors are not the source of this postsynaptic Ca2+, what is the source and function of this postsynaptic Ca2+ required for the induction of t-LTD? It has previously been reported that Ca2+ channels and release of Ca2+ from intracellular stores are required for t-LTD induction at neocortical synapses. To investigate this possibility at the CA3-CA1 synapses of the hippocampus, we first repeated the pairing protocols following bath application of an L-type Ca2+ channel blocker, nimodipine; in the presence of nimodipine (10 µM), t-LTD induction was completely prevented (105 ± 7%, n = 6, vs. interleaved controls, 75 ± 9%, n = 5; Fig. 5C,D). Next, we performed the t-LTD pairing protocol in the presence of thapsigargin, which depletes intracellular Ca2+ stores; in the presence of bath-applied thapsigargin (10 µM), t-LTD was also prevented (98 ± 6%, n = 6 vs. interleaved controls, 65 ± 6%, n = 5; Fig. 5C,D). The presence of heparin (400 U/mL), a blocker of IP3R-mediated Ca2+ release, in the recording pipette also completely prevented t-LTD induction (106 ± 8%, n = 6 vs. interleaved control slices, 73 ± 8%, n = 5; Fig. 5D), suggesting that postsynaptic IP3R-mediated Ca2+ release is required for t-LTD. In contrast, inclusion in the patch pipette of ryanodine (100 µM), a blocker of ryanodine receptors (RyRs) and Ca2+-induced Ca2+ release from internal stores, did not prevent induction of t-LTD (80 ± 8%, n = 6 vs. 72 ± 8% in interleaved control slices, n = 5, Fig. 5D), suggesting that IP3R-mediated Ca2+ release, but not RyR-sensitive Ca2+ stores, is required for t-LTD..

18The source of IP3 responsible for Ca2+ release from intracellular stores during induction of t-LTD might be through activation of phospholipase C (PLC). We tested this possibility by using the PLC inhibitor U73122. t-LTD was completely prevented by bath application of U733122 (10 µM, 113 ± 8%, n = 6; Fig. 6B), confirming the involvement of the PLC pathway in t-LTD. The activation of PLC during the induction protocol could occur through the activation of mGluRs; in fact, in neocortical as well as in hippocampal neurons the induction of some types of t-LTD has been reported to require the stimulation of mGluRs, which activates PLC to produce IP3. We therefore tested whether t-LTD at CA3-CA1 synapses requires mGluRs. In the presence of the broad-spectrum mGluR antagonists MCPG (500 µM) or LY367385 (100 µM), t-LTD was completely prevented (MCPG, 106 ± 8%, n = 5; LY341495, 104 ± 7%, n = 7; Fig. 6A,B). Neither MCPG nor LY367385 affected baseline EPSP slope (not shown). t-LTD was also blocked by the specific mGlu5 receptor antagonist MPEP (20–40 µM, 97 ± 6%, n = 7, vs. interleaved control slices for the 3 experimental conditions, pooled together, 70 ± 8%, n = 19; Fig. 6A,B). Neither MCPG nor LY367385 prevented t-LTP induction (MCPG, 153 ± 8%, n = 5; LY367385, 155 ± 6%, n = 5). These results suggest that t-LTD requires an mGlu5 receptor-mediated increase of intracellular Ca2+ from intracellular stores. To test the possible postsynaptic location of the metabotropic receptors involved in t-LTD we repeated the experiments with the postsynaptic neuron loaded with GDPβS to prevent G-protein-mediated signaling. In this condition, t-LTD was completely prevented (99 ± 5%, n = 5 vs. interleaved control slices with no GDPβS loaded into postsynaptic cells 69 ± 4%, n = 5, Fig. 6C,D). These results indicate that postsynaptic G protein-coupled receptors, possibly mGluRs, are involved in t-LTD..

19Endocannabinoids (eCBs) are synthetized and released by postsynaptic cells in response to depolarization, Ca2+ elevation and/or mGluR signaling, and some synapses require signaling by eCBs for plasticity. To investigate whether t-LTD at CA3-CA1 synapses of the mouse hippocampus requires postsynaptic eCB synthesis, we performed t-LTD experiments with the postsynaptic neuron loaded via the patch pipette with tetrahydrolipstatin (THL, 5 µM), an inhibitor of the eCB synthesizing enzyme diacylglycerol lipase. In this experimental condition, t-LTD induction was completely prevented (107 ± 5%, n = 6, vs. interleaved control slices, 66 ± 9%, n = 5; Fig. 7A,B), thus indicating that postsynaptic eCB synthesis is necessary for t-LTD. eCBs diffuse and activate CB1 receptors on presynaptic neurons and/or glial cells. To investigate the involvement of CB1 receptors in hippocampal t-LTD, we repeated the t-LTD induction protocol in the presence of the CB1 receptor antagonist AM251 (3 µM). In the presence of AM251, t-LTD was completely prevented (96 ± 6%, n = 9, vs. interleaved slices, 74 ± 5%, n = 5; Fig. 7A,B). To investigate whether CB1 receptors are necessary only during t-LTD induction or also during expression, we applied AM251 after the t-LTD induction protocol. AM251 applied 20 min after the t-LTD induction protocol did not affect the magnitude of t-LTD (49 ± 17%, n = 4 vs. 57 ± 16% in interleaved control slices, Supplementary Fig. 1). Thus, t-LTD induction requires activation of postsynaptic eCB synthesis and activation of CB1 receptors. THL prevents the formation of the eCB 2-arachidonylglycerol (2-AG). To more directly investigate the possible involvement of 2-AG during t-LTD induction, we bath-applied 2-AG. A small reduction in the slope of the EPSP was observed (87 ± 5%, n = 8, Fig. 7C,D). Importantly, following 2-AG application, a t-LTD pairing protocol was no longer able to induce t-LTD (97 ± 6, n = 8, Fig. 7C,D), suggesting that 2-AG occludes the induction of t-LTD and that 2-AG or a structurally similar eCB is a component of the signaling cascade required for the induction of hippocampal t-LTD..

20CB1 receptors have been localized to presynaptic boutons and astrocytes. In astrocytes, CB1 receptor activation has been suggested to stimulate the release of glutamate and other gliotransmitters, including d-serine. To investigate a possible involvement of astrocytes in the induction of hippocampal t-LTD, we preincubated the slices for 1 h with the gliotoxin fluoroacetate (10 mM); this completely abolished t-LTD (111 ± 18%, n = 5 vs. interleaved control slices, 64 ± 7%, n = 6; Fig. 8A,B). In contrast, t-LTP was resistant to fluoroacetate treatment, as a pre-before-post pairing protocol still induced t-LTP in fluoroacetate-treated slices (136 ± 6%, n = 6 vs. interleaved control slices, 141 ± 11%, n = 6, Supplementary Fig. 2). Next, using a patch pipette, we loaded individual astrocytes with the Ca2+ chelator BAPTA (20 mM) for 1–4 h before patching a pyramidal neuron. Ca2+-dependent release of gliotransmitters is prevented by this treatment (see and references therein). We recorded CA1 pyramidal neurons in the proximity (50–100 µm) of the BAPTA-loaded astrocyte and found that BAPTA loading of the astrocyte (a-BAPTA) prevented the induction of t-LTD at CA3-CA1 synapses (121 ± 8%, n = 5 vs. interleaved control slices, 61 ± 6%, n = 5; Fig. 8C,D). t-LTP was not affected by BAPTA loading into nearby astrocytes as a pre-before-post pairing protocol still induced t-LTP in this experimental condition (147 ± 4%, n = 5, vs. interleaved control experiments, 151 ± 8%, n = 6; Supplementary Fig. 2). These results suggest that astrocytic Ca2+-dependent gliotransmitter release is necessary for t-LTD induction. d-serine, a co-agonist at NMDA receptors, is a candidate gliotransmitter, and astrocytes in CA1 have been reported to release d-serine. To test whether d-serine might be responsible for the requirement of astrocytes during induction of t-LTD, we repeated the BAPTA-loading experiments in the presence of 100 µM d-serine added to the superfusion fluid. In this experimental condition, t-LTD was completely recovered (60 ± 8%, n = 7, vs. interleaved slices with a-BAPTA, 108 ± 4%, n = 5; Fig. 9A,B). d-serine without the pairing protocol did not affect baseline transmission, as the EPSP slope was not affected by application of d-serine to the superfusion fluid for 30 min without a pairing protocol (103 ± 4%, n = 6; Fig. 9B, Supplementary Fig. 3). This result supports the conclusion that the contribution of astrocytes during induction of t-LTD involves the release of a co-agonist at NMDA receptors, most likely d-serine, as t-LTD in the a-BAPTA condition but in the presence of d-serine was restored to a level similar to that of control t-LTD experiments. To investigate a possible mechanism involved in d-serine release, we repeated the experiments loading the astrocytes with the G-protein signaling blocker GDPβS and recording neurons in the proximity. In this experimental condition, t-LTD was also completely prevented (103 ± 5%, n = 6). In the same cells, the subsequent application of d-serine recovered t-LTD (73 ± 5%, n = 6, Fig. 9C,D). This result suggests that a G protein-dependent mechanism is involved in the astrocytic signaling required for t-LTD. Consistent with a role for CB1 receptors in activating astrocytes, application of 100 µM d-serine also completely rescued t-LTD in the presence of the CB1 receptor antagonist AM251 (65 ± 5%, n = 5, vs. 101 ± 3%, n = 5 in the presence of 3 µM AM251 only; Fig. 9E,F). These results suggest that CB1 receptors, possibly located on astrocytes, are controlling astrocytic release of a co-agonist at nonpostsynaptic NMDA receptors. Interestingly, and consistent with the results of in the rat barrel cortex, direct stimulation of astrocytes by depolarizing current pulses paired with low-frequency presynaptic activity without postsynaptic action potentials was sufficient to induce LTD (64 ± 14%, n = 5, Supplementary Fig. 4), suggesting that, in this experimental condition, astrocytes may contribute to synaptic depression by mechanisms additional to that of release of a co-agonist at NMDA receptors..

21To gain mechanistic insight into how activation of nonpostsynaptic NMDA receptors could lead to t-LTD, we conjectured that a Ca2+-dependent enzyme might be involved. Since the Ca2+-dependent protein phosphatase calcineurin has earlier been implicated in other forms of LTD, both in the hippocampus and neocortex, we tested the effect of the calcineurin inhibitor FK506 on hippocampal t-LTD. We found that while t-LTD was not affected by loading the postsynaptic neuron or an astrocyte with FK506 (10 µM) via a patch pipette (69 ± 5%, n = 5 and 67 ± 5%, n = 6, respectively; Fig. 10A–C), it was completely blocked by bath application of FK506 (10 µM) (106 ± 7%, n = 7 vs. 68 ± 6%, n = 13 in interleaved control slices; Fig. 10A,C). FK506 was active when loaded into the postsynaptic neurons as loading the same concentration of FK506 into neurons of layer 2 of the medial entorhinal cortex completely prevented LTD in synaptic input from layer 1 (not shown), as previously reported, indicating that presynaptic calcineurin is involved in the induction of hippocampal t-LTD..

22The fact that t-LTD was blocked by extracellular application, but not by postsynaptic intracellular application, of an NMDA receptor channel blocker raises the possibility that nonpostsynaptic, possibly presynaptic, NMDA receptors are necessary for t-LTD, as demonstrated at neocortical synapses.

23To determine the site of expression of hippocampal t-LTD, we combined several approaches. First, we estimated the noise-subtracted CV of the synaptic responses before and after t-LTD induction. A plot of CV−2 versus the change in the mean evoked EPSP slope (M) before and after t-LTD yielded points along the diagonal line indicating a presynaptic modification of release parameters (; Fig. 11A). Second, in several experiments we observed failures in synaptic transmission, and thus we analyzed whether a change in the number of failures occurred after t-LTD. A consistent increase in the number of failures after t-LTD was observed in our experiments (24 ± 4%, n = 10 after t-LTD vs. 9 ± 2% in baseline, n = 10) suggesting also a presynaptic locus of expression of this form of t-LTD (Fig. 11B). Third, we analyzed the PPRs during baseline and 30 min after a t-LTD pairing protocol was applied. The analysis of PPRs before and after t-LTD showed a significant increase of PPR after t-LTD (2.4 ± 0.3, n = 6 vs. 1.4 ± 0.1 during baseline, n = 6; P < 0.01 Student's t-test), which is also indicative of a presynaptic change (Fig. 11C). Finally, to further corroborate the presynaptic expression of t-LTD, we analyzed the progressive block of NMDA receptor-mediated currents by MK-801 after induction of t-LTD compared with a control pathway. As MK-801 is an irreversible NMDA receptor open-channel blocker, NMDA receptors are blocked only at synapses that release transmitter, so the trial-to-trial progressive rate of block provides a measure of the release probability. We induced t-LTD in 1 pathway by 100 pairings of a postsynaptic action potential followed after 18 ms by an EPSP. This post-before-pre pairing protocol induced robust t-LTD (55 ± 11%, n = 5; Fig. 11D). An unpaired naïve pathway served as a control, and no change in EPSPs slope was observed in that pathway (105 ± 5%, n = 5; Fig. 11D). Thirty minutes after the pairing protocol, we switched to voltage-clamp mode and recorded NMDAR-EPSCs at −30 mV in the same cells. NMDA receptor currents were isolated by the addition of gabazine (2 µM) and NBQX (10 µM) to the superfusion solution to block GABAA and AMPA/kainate receptors, respectively. Following bath application of 100 µM MK-801, a gradual decrease of the amplitude of the NMDA receptor currents, and eventually an almost complete block, was observed in the paired as well as the unpaired pathway. However, a slower rate of decay (measured as the number of stimuli necessary to reduce NMDAR-EPSC amplitude) was observed in the paired compared with the unpaired pathway (Fig. 9E). The half-life of the NMDAR-EPSC was 40 ± 5 stimuli for the unpaired pathway (n = 5) and 68 ± 4 stimuli for the paired pathway (n = 5; Fig. 11F). These results are consistent with a reduction in the neurotransmitter release probability in the paired pathway. Together, these results are all indicative of a presynaptic locus of expression of this form of t-LTD..

24In summary, we have found here that both t-LTP and t-LTD can be induced at hippocampal CA3-CA1 synapses in young mice by pairing presynaptic activity with single postsynaptic action potentials at low stimulation frequency (0.2 Hz). These 2 forms of plasticity both require NMDA receptors for their induction, but the NMDA receptors required for t-LTP and t-LTD are at different locations; while t-LTP requires postsynaptic ionotropic NMDA receptors, t-LTD does not. We have further characterized the subunit composition of these 2 different populations of NMDA receptors and found that the postsynaptic NMDA receptors required for t-LTP contain GluN2A and GluN2B, but not GluN2C or GluN2D subunits, whereas NMDA receptors mediating t-LTD contain GluN2C and/or GluN2D subunits. Both t-LTP and t-LTD require postsynaptic Ca2+ for their induction, and t-LTD also requires L-type voltage-dependent Ca2+ channels, activation of mGlu5 receptors, PLC and postsynaptic IP3 receptor-mediated Ca2+ release from internal stores, postsynaptic eCB synthesis, activation of CB1 receptors, and astroglial signaling. We have furthermore found that application of the NMDA receptor co-agonist d-serine bypasses the requirement of astrocytes, suggesting that astrocytes deliver a co-agonist at NMDA receptors during induction of t-LTD. Finally, we demonstrated that t-LTD is presynaptically expressed as indicated by analysis of trial-to-trial EPSP fluctuations and failure rates, as well as PPRs and the rate of use-dependent depression of NMDA receptor currents by MK-801. This new form of hippocampal t-LTD is in many aspects similar to t-LTD at layer 4-to-layer 2/3 synapses in neocortex. However, one notable difference is the nature of the putative gliotransmitter released by astrocytes, which in the case of layer 4 to layer 2/3 synapses is glutamate, while our data at hippocampal CA3-CA1 synapses suggest the involvement of d-serine or another co-agonist acting on nonpostsynaptic ionotropic NMDA receptors.

25Both t-LTP and t-LTD require NMDA receptors as both were prevented by the inclusion of the NMDA receptor antagonist d-AP5 in the superfusion fluid. Similar to neocortical layer 4-to-layer 2/3 synapses we found that while postsynaptic ionotropic NMDA receptors are required for hippocampal t-LTP, they are not required for t-LTD as the inclusion of the NMDA receptor channel blocker MK-801 (1 or 4 mM) in the postsynaptic cell prevented induction of t-LTP but not t-LTD, indicating that NMDA receptors required for t-LTP and t-LTD are at different locations. Importantly, extracellular application of MK-801 blocked the induction of t-LTD, demonstrating that nonpostsynaptic ionotropic NMDA receptor function is required for t-LTD, but an additional role for a postsynaptic NMDA receptor-mediated metabotropic effect cannot be completely ruled out.

26The presence of different subpopulations of NMDA receptors in different brain regions suggests that different subtypes play different roles in brain function. By using different subunit-preferring pharmacological agents, it was reported that LTP and LTD induced by high-frequency and low-frequency afferent stimulation, respectively, could be dissociated, with LTP being dependent of GluN2A, but not GluN2B subunit-containing receptors, and LTD requiring GluN2B, but not GluN2A subunit-containing receptors in the hippocampus as well as in the perirhinal cortex. From these first studies the situation has been shown to be more complex. Most of the available antagonists have limitations in selectivity and caution should be exercised when interpreting these data. Nevertheless, in our experiments, the GluN2A antagonists Zn2+ and NVP-AAM077 (at 100 nM) both completely abolished t-LTP at CA3-CA1 synapses in the hippocampus without affecting t-LTD, whereas the GluN2C and 2D subunit-preferring antagonists PPDA and UBP141 both completely blocked t-LTD without affecting t-LTP. The GluN2B subunit-selective antagonist Ro 25-6981 significantly reduced t-LTP without significantly affecting t-LTD. These results are similar to the double dissociation found at neocortical layer 4-to-layer 2/3 synapses in the sense that t-LTP was blocked by GluN2A subunit-preferring NMDA receptor antagonists but not GluR2C/2D subunit-preferring NMDA receptor antagonists and vice versa, but, whereas a GluN2B subunit-selective NMDA receptor antagonist reduced t-LTP in the hippocampus, t-LTP was not affected in the somatosensory neocortex. Nevertheless, both in the hippocampus and neocortex, the differential NMDA receptor requirement for the induction of t-LTP and t-LTD might reflect compartment-specific expression of different NMDA receptor subunits.

27We found that both t-LTP and t-LTD require a rise in postsynaptic Ca2+, as both were blocked by the presence of BAPTA in the postsynaptic cell, similar to what was reported in the somatosensory cortex. t-LTD requires Ca2+ from intracellular stores as Ca2+ store depletion (by thapsigargin) and postsynaptic IP3 receptor blockade (using heparin) blocked t-LTD induction. This form of t-LTD thus resembles previously described postsynaptic NMDA receptor-independent forms of LTD, which often involves voltage-dependent Ca2+ channels. This was also observed in our results, since t-LTD was blocked by the L-type Ca2+ channel blocker nimodipine. These requirements of hippocampal t-LTD are also shared with presynaptic t-LTD at neocortical synapses. The rise in intracellular Ca2+ may be due to the activation of mGlu5 receptors and subsequent activation of PLC as t-LTD was prevented by the general metabotropic glutamate receptor antagonists MCPG and LY341495 as well as the selective mGlu5 receptor antagonist MPEP and intracellular loading of the PLC inhibitor U73122. Moreover, t-LTD induction was prevented when the postsynaptic neuron was loaded with GBPβS, which blocks G-protein-mediated signaling. However, presynaptic mGluRs have also been described at these synapses, modulating glutamate release and participating in plasticity. The existence of glial mGluRs has also been reported and they might also have a role in this form of t-LTD. Loading GDPβS into astrocytes prevented t-LTD induction, but not the synaptic depression induced by mGluR activation (Supplementary Fig. 5); thus, this result does not allow us to distinguish between the possibilities that GDPβS blocks t-LTD due to inhibition of mechanisms mediated by mGluRs or by other types of metabotropic receptor, such as CB1 receptors. More work will be necessary to clarify this point.

28Postsynaptic loading of THL, a selective inhibitor of the eCB synthesizing enzyme diacylglycerol lipase, blocked, and addition of the eCB 2-AG occluded, the induction of t-LTD, suggesting that eCBs are involved in the induction of t-LTD. Whether postsynaptic Ca2+ or another signaling cascade is required for the release of eCBs is not known. eCBs released from pyramidal neurons have also been reported to induce long-term enhancement of evoked glutamate release though CB1 receptor activation and stimulation of astrocytes. In our experiments, AM-251, a CB1 receptor antagonist, prevented the induction of t-LTD, suggesting that eCB binding to CB1 receptors is required for induction of t-LTD. CB1 receptors involved in synaptic plasticity are located in the presynaptic neuron and/or in astrocytes. In the present work, we did not elucidate the site of CB1 receptors mediating t-LTD and further work will be required to address this question, although the demonstration that astrocytes are required for t-LTD is suggestive that astrocytic CB1 receptors might be involved.

29We used 2 methods to investigate the involvement of astrocytic activity in t-LTD. First, the gliotoxin fluoroacetate prevented the induction of t-LTD. Second, loading individual astrocytes with the Ca2+ chelator BAPTA completely abolished t-LTD, suggesting that Ca2+-regulated release of a gliotransmitter is necessary for t-LTD induction at hippocampal CA3-CA1 synapses.

30A role for astrocytes in synaptic depression is well established. For instance, ATP released from astrocytes as a result of neuronal activity can modulate synaptic transmission in cultured hippocampal neurons. ATP tonically suppresses glutamatergic transmission via P2Y receptors. This effect depends on the presence of co-cultured astrocytes. Glutamate activates non-NMDA receptors on astrocytes and triggers ATP release which causes homo- and heterosynaptic depression. GABAergic network activation of glial cells can also induce hippocampal heterosynaptic depression. In addition, astrocytes have been implicated in transient heterosynaptic depression in the CA1 region of acute hippocampal slices.

31The NMDA receptor co-agonist d-serine is an interesting candidate for a gliotransmitter involved in t-LTD induction. Indeed, addition of d-serine to the superfusion fluid completely restored t-LTD in experiments in which astrocytes were loaded with BAPTA, suggesting that astrocytes release a gliotransmitter required for t-LTD, acting as co-agonist on nonpostsynaptic NMDA receptors.

32Additionally, d-serine recovered t-LTD when astrocytes were loaded with GDPβS and when AM251 was added in the superfusion fluid. These results are consistent with the possibility that CB1 receptors are located on the astrocytes and that d-serine acts on nonpostsynaptic NMDA receptors involved in mediating t-LTD. GluN3A subunit-containing presynaptic NMDA receptors are required for t-LTD in the mouse visual cortex during development, and di-heteromeric GluN1/GluN3A NMDA receptors without glutamate-binding sites have been suggested to exist. In support of this possibility, pairing of astrocytic activity with presynaptic activity without postsynaptic action potentials was sufficient to induce LTD. However, the finding that in the presence of d-serine post-before-pre pairing, rather than only presynaptic activation, was required to induce LTD is consistent with the involvement of NMDA receptors that also contain glutamate-binding GluN2 subunits. It remains possible that astrocytes might release both glutamate and d-serine.

33The results obtained are suggestive that hippocampal t-LTD is mediated by presynaptic NMDA receptors, as ionotropic NMDA receptors are involved but postsynaptic ionotropic NMDA receptors are not required, as demonstrated by the failure of postsynaptic MK-801 to block t-LTD. In the somatosensory cortex, a similar form of t-LTD has been described that requires presynaptic NMDA receptors (; see also;;, ). While the focus of the present work was the characterization of hippocampal t-LTP and t-LTD and early insights into the underlying mechanisms, there is no definitive demonstration of the location of the NMDA receptors involved in hippocampal t-LTD. Future work using paired recordings at CA3-CA1 synapses and including MK-801 in the presynaptic neuron will determine whether the nonpostsynaptic NMDA receptors involved in hippocampal t-LTD have a presynaptic location. The intracellular uncaging of a caged form of MK-801 could more precisely identify the location of these receptors in the cells, as has been done at layer 4-to-layer 2/3 synapses in somatosensory cortex. The presence of presynaptic NMDA receptors in the hippocampus has been suggested previously from experiments monitoring noradrenaline release in synaptosomes, and immuno-electron microscopy has observed NMDA receptor immunolabeling at presynaptic elements of the hippocampus. Physiological roles for presynaptic NMDA receptors have also been suggested, including modulating transmitter release by acting as autoreceptors (;; see for review). Presynaptic NMDA receptors have been suggested to be involved in STDP in visual cortex and somatosensory cortex with direct evidence obtained at layer 4-to-layer 2/3 neurons of somatosensory cortex. In the hippocampus, presynaptic NMDA receptors have been suggested to participate in the induction of LTP. Future experiments should elucidate their precise location and physiological roles at hippocampal CA3-CA1 synapses.

34Protein phosphatases, including calcineurin, have been reported to be required for several different forms of LTD, both in the hippocampus and neocortex. Using FK506 to block calcineurin activity, our results indicate the involvement of presynaptic calcineurin in t-LTD induction, in similarity to the involvement of presynaptic calcineurin in neocortical p-LTD.

35While the exact mechanisms of how calcineurin mediates LTD of evoked transmitter release are unknown, several presynaptic proteins involved in transmitter release processes are targets of phosphorylation/dephosphorylation cascades and therefore candidates to mediate such changes. These include proteins involved in exocytosis, endocytosis, and the regulation of the size of the releasable, recycling and reserve pools of synaptic vesicles, as well as presynaptic calcium channels and their association to the release machinery. Future experiments should determine the exact mechanism by which calcineurin mediates LTD of evoked glutamate release.

36We used 3 different approaches to determine the locus of expression of this form of t-LTD, and all 3, fluctuation analysis (including failure rate), PPR and MK801-induced progressive decay of NMDA receptor currents were consistent with presynaptic changes, strongly suggesting that the locus of expression of this form of hippocampal t-LTD is presynaptic.

37The exact role of STDP in the hippocampus is not known, and more work is necessary to determine specific functions for t-LTP and t-LTD. Temporally asymmetric synapse strengthening has been predicted to drive learning of sequences. Computational models have indicated that temporally asymmetric LTP in the hippocampus could store sequences of spatial positions and that place fields shift backward along repeated paths due to LTP at synapses from earlier- to later-activated place cells. Experimental support for this model was found by, consistent both with simple Hebbian STDP and with a unified model of rate- and timing-dependent plasticity. showed that a rate- and timing-dependent plasticity model could explain both learning of spatial sequences and increased functional connectivity between neurons with overlapping place fields. Thus, STDP seems a good candidate to mediate spatial learning and the possible role of t-LTP and t-LTD in forms of learning involving the hippocampus will be addressed in future studies.

38Our studies were done in developing P12–P18 animals, and the form of t-LTD studied here was absent after P21 (Supplementary Fig. 6). The functions of t-LTP and t-LTD during development are most probably related to the refinement of synaptic connections and remodeling of neuronal circuits. As a Hebbian learning rule, t-LTP should occur when the spike order is pre-before-post, strengthening those connections in which the presynaptic neuron takes part in firing the postsynaptic cell, as predicted by Hebb, whereas t-LTD occurs when the spike order is reversed, so that noncausal spiking weakens the connections involved, possibly as a first step in the elimination of those connections during development in a similar way to that suggested in the neocortex (see ). The form of t-LTD described here may be specifically related to development as it was only observed until the third week of development (in fact, after P21 the same protocol induced t-LTP; Supplementary Fig. 6). Whether or not the specific form of t-LTP studied here is present in adult animals remains to be confirmed (see also; ).

39In summary, these results indicate that at CA3-CA1 synapses of mouse hippocampus, both t-LTP and t-LTD can be induced at low frequency by temporal pairing of presynaptic activity with single postsynaptic spikes. Both t-LTP and t-LTD require NMDA receptors but these NMDA receptors are different and with different location; whereas t-LTP requires postsynaptic NMDA receptors located at CA1 neurons, t-LTD does not require postsynaptic ionotropic NMDA receptors but requires nonpostsynaptic, likely presynaptic NMDA receptors. We also studied the mechanisms involved in t-LTD and found that its induction requires postsynaptic Ca2+ through L-type Ca2+ channels as well as mGlu5 receptor activation and release of Ca2+ from internal stores, which mediates the synthesis and release of eCBs activating CB1 receptors located on astrocytes and/or on presynaptic neurons. Astrocytes were shown to be crucial for the induction of t-LTD most likely by releasing the gliotransmitter d-serine, which, together with glutamate released by presynaptic neurons, activates presynaptic NMDA receptors to induce hippocampal t-LTD (Fig. 12).

40Supplementary material can be found at: http://www.cercor.oxfordjournals.org/.

41This work received support from the Ministerio de Ciencia e Innovación/European Regional Development Fund (grant BFU2009-10034), Ministerio de Economía y Competitividad/European Regional Development Fund (grant BFU2012-38208) and the Junta de Andalucía/European Regional Development Fund (grant P11-CVI-7290 to A.R.-M.), from the Biotechnology and Biological Sciences Research Council, UK (grant BB/H002383/2 to O.P.), and from a Royal Society International Exchanges grant to A.R.-M. and O.P. Funding to pay the Open Access publication charges for this article was provided by Research Councils UK.