Materials and Methods

Animals

Our study was conducted in accordance with the Guide for the Care and Use of Laboratory Animals and was approved by the ethics committee of the Hangzhou City University (registration number: 22061). C57BL/6 male mice (2-4 months) were purchased from Hangzhou Ziyuan Laboratory Animal Corporation and housed in groups of three to four per cage. The mice were maintained on a 12-hour light/dark cycle and were provided with ad libitum access to food and water.

Stereotactic virus injection

Stereotactic virus injection was conducted as described previously (1, 2). Briefly, adult mice were deeply anesthetized with sodium pentobarbital (50 mg/kg) and secured in a stereotaxic device with ear bars (RWD, 68930) while their body temperature was maintained at approximately 37 °C using a heating blanket. Their hair was removed using a razor and the skin was sterilized with iodophor. A 1-cm incision in the midline using sterile scissors. Small burr holes were drilled bilaterally using an electric hand drill at the following coordinates: anteroposterior (AP), 2.3 mm from Bregma; mediolateral (ML), ±1.4 mm. Virus particles were then injected bilaterally into the stratum radiatum (1.2 mm from the pial surface) using glass pipettes connected to an injection pump (RWD, R480). The injection rate was controlled at 1 nl/s using the pump. To allow the virus disseminate into the tissue, glass pipette was left in place for 10 minutes after each injection. After the injection, the pipettes were gradually removed and the wound was sutured up. For hippocampal interneuron physiological recording, 200 nl AAV2/9 mDLx EGFP (6.4 × 10¹¹ gc/ml) were injected. For hippocampal slice Ca²⁺ imaging, 500 nl AAV2/5 GfaABC₁D GCaMP6f (1.2 × 10¹² gc/ml) with injected alone or mixed with 500 nl AAV2/5 GfaABC₁D hM3D (Gq) mCherry (5.3 × 10¹² gc/ml). To knockdown γCaMKII in hippocampal interneuron, an shRNA sequence (5’-GCAGCTTGCATCGCCTATATC-3’) was used. A total of 200 nl of AAV2/9 mDLx γCaMKII shRNA (6.4 × 10¹² gc/ml) were injected. All virus were generated by Brainvta and Sunbio Medical Biotechnology (Wuhan, https://www.brainvta.tech/ and Shanghai, http://www.sbo-bio.com.cn/).

Electrophysiology

Mice were anesthetized with isoflurane, and their brains were quickly extracted and immersed in an ice-cold solution which containing (in mM) 235 Sucrose, 1.25 NaH₂PO₄, 2.5 KCl, 0.5 CaCl₂, 7 MgCl₂, 20 Glucose, 26 NaHCO₃, and 5 Pyruvate (pH 7.3, 310 mOsm, saturated with 95% O₂ and 5% CO₂). Coronal hippocampus slices (300-350 μm) were prepared with a vibrating slicer (Leica, V T1200) and incubated for 30 - 40 min in artificial cerebrospinal fluid (ACSF) containing (in mM) 26 NaHCO₃, 2.5 KCl, 126 NaCl, 20 D-glucose, 1 sodium pyruvate, 1.25 NaH₂PO₄, 2 CaCl₂ and 1 MgCl₂ (pH 7.4, 310 mOsm, saturated with 95% O₂ and 5% CO₂).

The slices were transferred to an immersed chamber and continuously perfused with oxygen-saturated ACSF and GABA receptor blockers, picrotoxin (100 μM) and CGP55845 (5 µM), at a rate of 3 ml/min. Interneurons in the stratum radiatum or stratum oriens were visualized using infrared differential interference contrast and epifluorescence imaging. Perforated-path recordings were conducted as previously described (3). Briefly, perforated whole-cell recording from stratum radiatum or stratum oriens interneurons were made with pipettes filled with solution containing (in mM) 136 K-gluconate, 9 NaCl, 17.5 KCl, 1 MgCl₂, 10 HEPES, 0.2 EGTA, 25 μM Alexa 488, amphotericin B (0.5 mg ml⁻¹) and small amounts of glass beads (5–15 μm in diameter; Polysciences, Inc., Warminster, PA, USA) (pH 7.3, 290 mOsm). The pathched neuron was intermittently imaged with epifluorescence to monitor dye penetration. If the patch ruptured spontaneously, the experiment was discontinued.

Whole-cell voltage-clamp recordings were made from either stratum radiatum or stratum oriens interneurons using pipettes with resistance of 3-4 MΩ and filled with a solution containing (in mM) 4 ATP-Na₂, 0.4 GTP-Na, 125 CsMeSO₃, 10 EGTA, 10 HEPES, 5 4-AP, 8 TEA-Cl, 1 MgCl₂, 1 CaCl₂ (pH 7.3–7.4, 280–290 mOsm).

Whole-cell current-clamp recording were made from stratum interneurons using pipettes with resistance of 4 - 6 MΩ and filled with a solution containing (in mM) 125 K-Gluconate, 2 MgCl₂, 10 HEPES, 0.4 Na⁺-GTP, 4 ATP-Na₂, 10 Phosphocreatine disodium salt, 10 KCl, 0.5 EGTA (pH 7.3–7.4, 280–290 mOsm).

Whole-cell recording were performed on stratum radiatum astrocytes using pipettes with resistance of 8 - 10 MΩ and filled with an intracellular solution containing (in mM) 130 K-Gluconate, 20 HEPES, 3 ATP-Na₂, 10 D-Glucose, 1 MgCl₂, 0.2 EGTA (pH 7.3-7.4, 280-290 mOsm). In a subset of experiments, 0.14 mM CaCl₂ and 0.45 mM EGTA were included in the upper intracellular solution to maintain a stable level of astrocytic concentration (calculation by Web-MaxChelator) (2, 4).

Electrical stimuli was delivered via theta glass pipettes in the Schaffer Collateral of stratum radiatum, with a 30 s inter-trial interval during baseline and after LTP induction. Evoked EPSPs were recorded in current-clamp model at the resting membrane potential of the stratum radiatum interneuron. After a 10-min stable baseline period, LTP was induced by applying theta-burst stimulation [TBS; five bursts at 200- ms intervals (5 Hz), each burst consisting of five pulses at 100 Hz], paired with five 60- ms long depolarizing steps to −10 mV. Six episodes of TBS paired with depolarization were given at 20-s intervals. In stratum oriens, LTP was induced by applying TBS paired with five 60-ms long hyperpolarizing steps to −90 mV.

sEPSCs were recorded in whole-cell model at −70 mV in the presence of picrotoxin (100 μM) and D-AP5 (50 μM). Paired pulses were delivered at an inter-pulse intervals of 50 ms, and the paired-pulse ratio was calculated by dividing the peak amplitude of the second EPSC by the peak amplitude of first EPSC. To analyze action potential properties, interneurons were recorded at rest and depolarized with 500-ms current injection pulses at 10-pA increments.

To avoid NMDAR-mediated signaling rapidly washing out when recording interneuron in whole-cell mode, electrical stimuli was delivered with a 5 s inter-trial interval to measure the I-V relationship for NMDAR-mediated EPSPs within 5 minutes of breaking in. The NMDA/AMPA ratio in the stratum radiatum or stratum oriens interneurons was calculated by measuring the amplitude of NMDAR-mediated EPSCs at + 60 mV (50 ms after stimulation) and the peak amplitude of AMPAR-mediated EPSC recorded at −60 mV. To measure pure NMDAR-mediated EPSCs in interneurons at resting membrane potential in perforated whole-cell recording model, Mg²⁺ free ACSF was used, and D-AP5 (50 μM) and picrotoxin (100 μM) were present in ACSF.

Axopatch 700B amplifiers were utilized for patch-clamp recordings (Molecular Devices). The data was filtered at 6 kHz and sampled at 20 kHz before being processed off-line using the pClampfit 10.6 program (Molecular Devices). Negative pules (−10 mV) were employed to monitor series resistances and membrane resistances. The data was included to analysis if the series resistances varied by less than 20% over the course of the trial. All experiments were carried out at 32°C.

Behavioral assays

For the contextual and cued fear conditioning test, the mouse was habituated to the environment and handled for three consecutive days. On the fourth day, the mice were allowed to explore the conditioning cage for 2 min, after which they received three moderate tone-shock pairs [30 stone (80 dB, 4 kHz) co-terminating with a foot shock (0.4 mA, 2 s)]. Following conditioning, the mice were returned to their home cages. The next day, they were place in the same conditioning cage but without receiving any foot shocks, and their freezing behavior was analyzed for the first 5 min. Four hours after the contextual fear conditioning test, the mice were placed in a novel cage with a different shape and texture of the floors compared to the conditioning cage). They were allowed to freely explore the new environment for 2 min and then subjected to 3 tones stimulations [(80 dB, 4 kHz) lasting for 30 s] separated by intervals of 90s, but without receiving any foot-shocks. Once the final tone had ended, all of the mice were allowed to freely explore the chamber for an additional 90 s. Fear responses were measured by calculating the freezing values of the mice using Packwin software (Panlab, Harvard Apparatus, USA). All apparatus were carefully cleaned with 30% ethanol in between tests.

Immunohistochemistry

Immunohistochemistry was conducted as described previously (1, 5, 6). Briefly, mice were administered a single intraperitoneal injection of 50 mg/kg sodium pentobarbital for anesthesia and were transcardially perfused with phosphate-buffered saline (PBS). After the liver and lungs had become bloodless, the mice were perfused with 4% paraformaldehyde (PFA) in 0.1 M PBS. The brains were quickly removed and placed in 4% PFA at 4℃ overnight. Following this, the tissues were cryoprotected in successive concentrations of 10%, 20%, and 30% sucrose before being sliced into 20 μm sections using a freezing microtome. After being washed multiple times with PBS, the sections were blocked for 1.5 hours at room temperature (22-24℃) in a blocking solution consisting of 5% Bovine Serum Albumin (BSA) and 1% Triton X-100. Following the blocking step, the sections were incubated with primary antibodies overnight at 4 ℃. Antibodies used were as follows: mouse monoclonal anti-GFAP (1/1000, Cell Signaling Technology Cat #3670, RRID:AB_561049), mouse monoclonal anti-NeuN (1/500, Millipore Cat# MAB377, RRID: AB_2298772), mouse monoclonal GAD67 (1/500, Synaptic Systems Cat# 198 006, RRID: AB_2713980). After wash several times in PBS, sections were then incubated with the following secondary antibody: Alexa Fluor 594 goat anti-mouse (1/1000, Cell Signaling Technology Cat# 8890, RRID: AB_2714182) at room temperature for 2 h. Afterwards, the sections were rinsed several times in PBS and incubated with DAPI for 5 minutes at room temperature. Following this, the sections were rinsed again and mounted with Vectashield mounting medium. Images were examined using a confocal laser scanning microscope (Olympus, VT1000) and analyzed using ImageJ (NIH, RRID: SCR_003070).

Ca²⁺ imaging

Ca²⁺ signals in hippocampal astrocytes were observed under a confocal microscopy (Fluoview 1000; Olympus) with a 40x water immersion objective lens (NA = 0.8). GCaMP6f was excited at 470 nm. Astrocytes located in the hippocampal CA1 region and at least 40 µm away from the slice surface were selected for imaging. Images were acquired at 1 frame per 629 ms. Hippocampal slices were maintained in ACSF containing picrotoxin (100 μM), and CGP55845 (5 µM) using a perfusion system. A single theta burst stimulation (TBS) was used to stimulate the neuron. CNO (5 μM) was bath-applied to activate hM3D(Gq)-mediated Ca²⁺ signals. The Ca²⁺ signal analysis was described previously (1, 2, 5, 6). In some experiments, image stacks comprising 10-15 optical sections with 1 µM z-spacing were acquired to facilitate the identification of GCaMP6f and hM3Dq (mCherry) co-expression in astrocytes. Briefly, movies were registered using the StackReg plugin of ImageJ to eliminate any x-y drift. Ca²⁺ signals were then analyzed in selected ROIs using the Time Series Analyzer V3 plugin of ImageJ. GCaMP6f fluorescence was calculated as ΔF/F = (F-F₀)/F₀. Mean ΔF/F of Ca²⁺signals were analyzed using Clampfit 10.6.

Statistics

All data processing, figure generation, layout, and statistical analysis were performed using Clampfit 10.6, Prism, MATALAB and Coreldraw. To assess the normality of the data, the Shapiro-Wilk test was used. If the results of the test were not statistically significant (p> 0.05), the data were assumed to follow a normal distribution, and a paired t-test was employed. On the other hand, if the test was statistically significant, a Wilcoxon Signed Rank Test or Mann-Whitney Ran Sum Test was utilized. To statistically analyze cumulative frequency distributions, the Kolmogorov-Smirnov test was employed. When comparing two groups, either a Wilcoxon signed-rank test or a Student’s t-test (paired or unpaired) was used. When comparing three groups, One-way repeated measures (RM) ANOVA followed by Dunnett’s post-hoc test was used.

All values were presented as mean ± standard error of the mean (SEM). Data were considered significantly different when the p value was less than 0.05 (*p < 0.05, **p < 0.01, ***p < 0.001). The numbers of cells or mice used in each analysis were indicated in figure legends.

Introduction

Long-term potentiation (LTP) was originally discovered as a long-term increase in synaptic efficacy at glutamatergic synapses onto granule cells of the hippocampus (7). This form of plasticity has been well documented at CA1 pyramidal cells, where synaptic plasticity can be induced by the activation of postsynaptic (N-methyl-D-aspartate) NMDA receptors (NMDARs), voltage-dependent Ca²⁺ channels, and metabotropic glutamate receptors (mGluRs) (8, 9) . Numerous studies have provided compelling evidence that LTP in hippocampal network occurs not only at excitatory glutamatergic synapse onto excitatory pyramidal and granule cells (LTP_E→E), but also at excitatory glutamatergic synapses onto inhibitory interneurons (LTP_E→I) (10–14). However, the mechanism behind LTP_E→I remains controversial due to the heterogeneous expression of proteins in various types of hippocampal interneurons and the absence of synaptic spines between excitatory inputs on interneurons (10, 15). Two distinct forms of LTP_E→I, namely NMDARs-dependent LTP and NMDARs- independent LTP_E→I, have been observed in hippocampal interneurons.

Astrocytes are the most prevalent glial cell type in the central nervous system (CNS), and play a critical role in regulating the development and function of the nervous system (16, 17). Astrocyte display multipolar branches with numerous micro- processes that allow them to closely associate with blood vessels, neuronal cell bodies and axons, other glial cells and as well as synapses (18–20). They also express varies ion channels, transporters and neurotransmitter receptors (21, 22). With these membrane proteins, they can sense neuronal activity and exhibit increases in intracellular Ca²⁺ in reaction to neurotransmitters and, in turn, they release neuroactive chemicals called gliotransmitters that regulate synaptic transmission and plasticity (17, 23–25). Numerous studies have been shown that the release of D-serine, a co-agonist of NMDAR, from astrocytes is capable of enabling LTP in cultures, in slice and as well as in vivo (4, 26–29). Given the growing number of studies demonstrating the direct roles that astrocytes play in regulating LTP_E→E, it is of particular interest to understand whether and how astrocytes in modulation LTP of excitatory postsynaptic currents in interneuron.

In this particular study, our focus was on the interneurons distributed in the stratum radiatum layer of the CA1 region of the hippocampus. Approximately 80% of these interneurons exhibit NMDAR-dependent LTP_E→I when presynaptic stimulation is paired with postsynaptic depolarization. We found that blocking astrocyte metabolism and clamping of astrocyte Ca²⁺ signaling can prevent the LTP_E→I in large NMDAR- containing interneurons, which could be rescued by bath application of D-serine. Furthermore, pharmacological and Ca²⁺ imaging studies have shown that astrocytes respond to Schaffer Collateral stimulation with Ca²⁺ increases through activation of type-1 cannabinoid receptors (CB1R), which stimulates the release of D-serine, and further regulates LTP_E→I through binding to the glycine site of NMDARs. We also found that activating astrocytes with Gq designer receptors exclusively activated by designer drugs (DREADDs) induced a potentiation of excitatory to inhibitory synapses in stratum radiatum. Additionally, knockdown γCaMKII hampered LTP_E→I in stratum radiatum of hippocampus in brain slice and disrupted contextual fear conditioning memory in vivo. Taken together, these results are the first to indicate that astrocytes are an integral component of a form of long-term synaptic plasticity between glutamatergic neurons and GABAergic interneurons, and that memory is also regulated by LTP_E→I.

Results

Astrocyte play a role in the formation of NMDAR-dependent LTP_E→I in CA1 stratum radiatum

To visualize CA1 stratum radiatum interneurons, we delivered an adeno-associated virus serotype 2/9 (AAV2/9) vector encoding EGFP under the control of the interneuronal mDLx promoter (AAV2/9-mDLx-EGFP) to this region (30). Within the virally transduced region, EGFP expression was limited to the interneuron, with high penetrance (>98% of the GAD67 cells expressed EGFP) (Fig. S1 A-C) and almost complete specificity (>98% EGFP positive cells were also GAD67 positive) (Fig. S1 D). These immunostaining results indicate that EGFP expression was limited to the interneuron in the CA1 region of stratum radiatum.

We next examined whether expressing EGFP in interneurons affects their membrane properties and synaptic transmission. To examine these, we performed whole-cell patch recordings of EGFP⁺ interneurons and putative interneurons in the CA1 stratum radiatum of the hippocampus in the presence of GABAA receptor blocker picrotoxin. We found that the excitability and resting membrane potential are no difference between these EGFP⁺ interneurons and putative interneurons (Fig. S2 A-C). Moreover, we also found that no difference in spontaneous excitatory postsynaptic currents (sEPSC) and paired-pulse ratio (PPR) at 50 ms interpulse intervals between EGFP⁺ interneurons and putative interneurons (Fig. S2 D-H). These results indicate that AAV injection and expression exogenous protein in interneurons has no effect on the membrane properties and baseline synaptic transmission of interneuron.

To avoid some necessary ingredient for LTP_E→I induction be diluted from the cytoplasm, perforate patch-clamp were used to record EPSPs from CA1 stratum radiatum interneurons. After a 10 min baseline recording, we delivered theta burst stimulation (TBS) consisting of 100 Hz stimulation with 25 pulses, delivered in six trains separated by 20-second intervals. The TBS was applied to induce LTP_E→I of the excitatory inputs to CA1 interneurons. The interneurons were depolarized to −10 mV using a voltage-clamp model during the TBS delivery. Under this condition, we found that in 8 out of 10 cells were induced LTP_E→I lasting at least 45 min (Fig. 1 A-C). We repatched 6 of these successful to induction LTP_E→I cells and randomly patched 2 EGFP⁺ in stratum radiatum in the whole-cell voltage-clamp model. The results indicated that all cells showed a linear current-voltage (I-V) curve for AMPAR, as well as a significant component of NMDAR-mediated currents (Fig. S3). In addition, we found that the induction of LTP_E→I was completely blocked by NMDAR blocker D- AP5 (Fig. 1 A-C). This result confirmed that LTP_E→I in CA1 stratum radiatum interneurons is NMDAR dependent (31, 32).

Fig. 1.

To investigate whether astrocytes were involved in NMDAR-dependent LTP_E→I in CA1 stratum radiatum interneurons, we treated slice with fluoroacetate (FAC, 5 mM) to block glial metabolism (4, 33). The results showed that the induction of LTP_E→I was blocked by FAC but not in control slice (Fig. 1 D-F). Next, we tested whether glial metabolism is involved in LTP_E→I in CA1 stratum orient, which has been shown to depend on calcium-permeable AMPARs (CP-AMPARs) and metabotropic glutamate receptors (mGluRs) (13). We found that the induction of LTP_E→I is not blocked by FAC in CA1 stratum oriens (Fig. S4 A-B). We repatched 6 of these cells (4 cells from control group and 2 cells from FAC treated group) in whole-cell voltage-clamp mode and observed that they exhibited high rectification AMPARs and a small component of NMDAR-mediated current (Fig. S5). This observation confirms previous study findings (32, 34). Moreover, we found that the induction of LTP_E→I is not blocked by D-AP5 in CA1 stratum oriens (Fig. S4). Overall, the results indicate that the formation of LTP_E→I in CA1 stratum radiatum is tightly regulated by astrocyte function. The results in stratum oriens also exclude the possibility that FAC directly affects the metabolism of interneuron which inhibit the formation of LTP_E→I.

It is commonly accepted that astrocytic calcium signaling plays a pivotal role in triggering the release of gliotransmitters and modulating synaptic transmission (24, 35–38). In addition, it is well documented that astrocytic calcium signaling is necessary to the formation of LTP_E→E in CA1 stratum radiatum (4, 26). Thus, we tested whether intracellular Ca²⁺ signals are required for the induction of LTP_E→I in CA1 stratum radiatum. We found that clamping of astrocyte Ca²⁺ concentration significantly suppressed LTP_E→I (Fig. 2 A-D). Consistent with the previous study in LTP_E→E, supply of D-serine (50 μM) fully rescued NMDAR-dependent LTP_E→I (Fig. 2D). For the control, when the intracellular Ca²⁺ concentration of astrocytes was not clamped, but the astrocyte was recorded with a glass pipette, LTP_E→I was indistinguishable from that induced without patching an astrocyte (Fig. 2D). Overall, these results indicate that functional preservation of astrocytic metabolism and Ca²⁺ mobilization is critical for maintaining the induction of LTP_E→I.

Fig. 2.

Astrocytic Ca²⁺ transients, induced by activation of astroglial cannabinoid type 1 receptors (CB1Rs), are involved in the regulation of LTP_E→I formation

Accumulating evidence indicates that neuronal depolarization in the hippocampus induced astrocytic Ca²⁺ transients, which is mediated by the activation of astroglial CB1Rs (39–43). Moreover, it has been shown that astroglial CB1R-mediated Ca²⁺ elevation is necessary for LTP_E→E in the CA1 region of hippocampus (26). Therefore, we asked whether astroglial CB1Rs-mediated Ca²⁺ elevations are required for hippocampal LTP_E→I. We first analyzed whether TBS could evoke astrocytic Ca²⁺ transient via CB1Rs in the stratum radiatum of hippocampus. In this study, GCaMP6f was used to analyze astrocytic Ca²⁺ signal, which were specifically expressed in astrocytes by using adeno-associated viruses of the 2/5 serotype (AAV 2/5) with the astrocyte-specific gfaABC₁D promoter (Fig. S6 A). Furthermore, the expression was confirmed by immunohistochemistry. Within the virally transduced region, GCaMP6f expression positive cells in CA1 stratum radiatum were also positive to the astrocyte specific marker GFAP (Fig. S6 A, C-D). Co-staining with the neuron marker NeuN showed no overlap with GCaMP6f expression (Fig. S6 B, E). Consistent with previously studies (26, 44), we found that TBS significantly increased the Ca²⁺ signaling in astrocytes of hippocampal slices in the presence of picrotoxin and CGP55845 (Fig. 3 A-B, F). As predicted, the increase in Ca²⁺ signals after TBS were inhibited by the CB1 receptor inhibitor AM251 (Fig. 3 C-D, F). Previous studies have demonstrated that activation of ɑ1-adrenoceptors increases calcium signals (1, 45–49) and triggers the release of D-serine release in the neocortex (50). However, our results show that ɑ1-adrenoceptors receptors are not involved in the Ca²⁺ signal increase observed after TBS (Fig. 3F).

Fig. 3.

Next, we explore whether CB1Rs-mediated Ca²⁺ elevation was accompanied by the formation of LTP_E→I. We found that LTP_E→I was significantly reduced in AM251 treated slice relative to controls (Fig. 3 G-I). Consistent with the above results shown in Fig. 2D, LTP_E→I was rescued by the addition of D-serine in AM251-treated slices

D-serine release from astrocyte potentiates NMDAR-mediated synaptic response

Next, we explored the underlying mechanisms by which astrocytes control the formation of LTP_E→I in the CA1 stratum radiatum. Our above results indicate that D-serine is a downstream signal pathway of astrocyte Ca²⁺ signal. D-serine, a co-agonist of NMDAR, could be released by astrocytes through Ca²⁺-dependent exocytosis and regulate the function of NMDAR. It has been shown that the occupancy of synaptic NMDAR co-agonist sites by D-serine is not saturated in CA1 pyramidal cells (26, 51). However, the level of occupancy of synaptic NMDAR co-agonist sites by D-serine in interneurons of CA1 stratum radiatum remains unclear. Furthermore, previous studies have demonstrated that NMDAR-dependent synaptic responses of interneurons in the stratum radiatum display a strong rundown effect under whole-cell mode (31). Thus, we recorded the NMDAR-mediated EPSPs from stratum radiatum interneurons in the perforated-patch configuration in Mg²⁺ free ACSF, which showed no rundown effect in this configuration (Fig. 4 A-B). Bath application of 50 μM D-serine enhanced the NMDAR-mediated EPSPs (Fig. 4 A-C), indicating that the level of D-serine in the excitatory to inhibitory synapse cleft was not saturated to occupy the co-agonist site. Subsequently, we tested whether TBS could increase extracellular level of D-serine and promote NMDAR-mediated synaptic responses. Our results indicated that NDMAR-mediated responses were transiently enhanced after one TBS, and the percentage of potentiation was reduced by pre-treatment hippocampal slice with 50 μM D-serine (Fig. 4 D-F). However, disrupting glial metabolism with FAC rendered the percentage of TBS-induced potentiation insensitive to bath application of D-serine (Fig. 4F). Furthermore, the percentage of potentiation in AMPAR-mediated responses after TBS is insensitive to bath application of D-serine, indicating that the enhancement of NMDAR-mediated response by TBS is not due the change of release probability and cell excitability (Fig. 4F). Taken together, our results indicate that D-serine release from astrocyte potentiates NMDAR-mediated responses by binding the co-agonist site and regulates the formation of LTP_E→I.

Fig. 4.

Chemogenetic activation of astrocytes induced E→I synaptic potentiation

The crucial role of astrocytes in LTP_E→I has been extensively demonstrated in brain slices (Henneberger et al., 2010; Min and Nevian, 2012; Pascual et al., 2005; Perea and Araque, 2007; Suzuki et al., 2011) and in vivo (26). In this study, we aimed to investigate if the activation of astrocytic G protein-coupled receptor (GPCR) pathway could trigger potentiation of E→I synaptic transmission. We utilized an adeno-associated virus serotype 5 (AAV2/5) vector encoding hM3Dq fused to mCherry for the specific activation of astrocytes via clozapine-N-oxide (CNO). To ensure specific expression in astrocytes, the vector was also under the control of the astrocyte-specific gfaABC1D promoter (Fig. S7).

To verify whether CNO application could evoke Ca²⁺ transients in astrocytes, we delivered AVV of hM3Dq along with AAV of GCaMP6f, and conducted confocal Ca²⁺ imaging in brain slices. Our results showed that CNO application indeed induced an increase in intracellular Ca²⁺ levels in cells co-expressing hM3Dq and GCaMP6f (Fig. 5 A-D). These results suggest that the expression of hM3Dq is selective to astrocytes and can elicit a rise in intracellular Ca²⁺ levels upon administration of CNO.

Fig. 5.

Subsequently, we investigated the impact of astrocytic Gq activation on evoked synaptic events in interneurons of CA1 stratum radiatum induced by Schaffer Collaterals stimulation, both before and after the administration of CNO. Interestingly, we observe that the EPSC amplitude was potentiated by 60% in response to the exact same stimulus in gfaABC1D::hM3Dq slices treated with CNO (Fig. 5 E-G), while no such potentiation was detected in slices obtained from the mice that were injected a control virus (AAV2/5-gfaABC1D::mCherry) (Fig. 5 E-G).

Previous studies have indicated that the synaptic potentiation triggered by chemogenetic activation of astrocyte is mediated through the release of D-serine by astrocytes, resulting in the activation of NMDARs (52). To verify whether the astrocytic-induced synaptic potentiation between excitatory and inhibitory neurons is mediated by D-serine release from astrocytes and the subsequent activation of NMDARs, we conducted an experiment in which we administered CNO after blocking the NMDARs with D-AP5 or saturating the glycine site of NMDARs with 50 μM D- serine. Our results showed that both the NMDAR blocker D-AP5 and 50 μM D-serine completely inhibited the potentiation in EPSP amplitude observed in response to CNO- induced astrocytic activation (Fig. 5 H-J). Our findings demonstrate, for the first time, that astrocytic activation alone can trigger de novo potentiation of synapses between excitatory and inhibitory neurons, and that this potentiation is indeed mediated by the release of D-serine from astrocytes and subsequent activation of NMDARs.

LTP_E→I in CA1 of stratum radiatum is necessary for long-term memory formation

We have previously shown that GABAergic interneurons in hippocampus express high levels of γCaMKII, while αCaMKⅡ, βCaMKⅡ and δCaMKⅡ are expressed at a lower frequency (53). Additionally, studies have demonstrated that γCaMKII expressed in hippocampal parvalbumin positive (PV⁺) interneurons and cultured hippocampal inhibitory interneurons is essential for the induction of LTP_E→I (53, 54). Specifically, in hippocampal PV⁺ interneurons, this protein is also vital for the formation of hippocampal-dependent long-term memory in vivo (53). Therefore, we asked whether astrocyte gated LTP_E→I in CA1 stratum radiatum involved in the hippocampal- dependent long-term memory formation. Several independent groups provide compelling evidence that astroglial CB1R-mediated signaling pathway regulates excitatory synapse formed by excitatory neurons, rather than excitatory synapses between excitatory neurons and interneurons (39–42, 55–57). Thus, to manipulate this pathway may also affect excitatory synapse formed by excitatory neurons. To avoid this side effect, we specifically knockdown of γCaMKII expression in inhibitory neurons of CA1 stratum radiatum by delivering an adeno-associated virus serotype 2/9 (AAV2/9) vector encoding shRNAs against γCaMKII and EGFP under the control of an inhibitory neuron-specific promoter (AAV2/9-mDLx-EGFP-γCaMKII shRNA) through bilateral stereotactic injection. We observed that interneurons identified by the specific markers GAD67 were also positive for EYFP and therefore likely expressed shRNAs, which led to knockdown γCaMKII in these cells (Fig. S8). To confirm that knockdown γCaMKII could hamper the induction of LTP_E→I, the EYFP positive interneurons in CA1 stratum radiatum were recorded in perforated-patch mode. We found that knocking down γCaMKII in CA1 stratum radiatum interneurons impaired LTP_E→I, but LTP_E→I in the putative interneurons in stratum radiatum of WT mice is unimpaired (Fig. 6 A-C). In addition, robust LTP_E→I induced by TBS in EYFP positive interneurons injected with AAV-mDLx-scramble shRNA into the CA1 stratum radiatum of hippocampus (Fig. 6C). Moreover, it is worth noting that the resting membrane potential, frequency and amplitude of sEPSC, excitability and PPR recorded from γCaMKII knockdown interneurons did not differ from those recorded from putative interneurons and EYFP positive interneurons (infected with scramble shRNA) (Fig. S9). Taken together, these results indicate that knocking down γCaMKII from interneuron has no effect on their synaptic transmission at baseline, but impaired the induction of LTP_E→I in these interneurons.

Fig. 6.

Next, we examined the behavioral consequences of destroying astrocyte gated LTP_E→I in CA1 stratum radiatum. We analyzed contextual fear conditioning memory, which is associated with activation of the hippocampus. We found that no effect of γCaMKII knockdown on exploration of the context before conditioning was observed (Fig. 6 D-E). 24 hours after training, mice were returned to the training box, and freezing was measured at the first 2 min. We found that γCaMKII knockdown mice showed a significantly reduction in freezing during contextual conditioning (Fig. 6F). Consistent with our previous study (53), γCaMKII knockdown in interneurons in CA1 stratum radiatum did not produce a significant effect on freezing during tone conditioning at 28 h after training (Fig. 6G). Taken together, our results strongly suggest that astroglial CB1R signaling pathway gated LTP_E→I in CA1 stratum radiatum plays a vital role in hippocampus-dependent long-term memory.

Discussion

In the present study, we identify LTP_E→I in CA1 stratum radium is tightly control by D-serine release from astrocyte via the astroglial CB1Rs-mediated Ca²⁺ elevation. In addition, knockdown of γCaMKII by specific shRNA in interneurons in CA1 stratum radium causes cognitive function deficit. Taken together, out data support that astrocyte gated LTP_E→I in CA1 stratum radium plays a critical role in preserve normal cognitive function.

CB1Rs are widely expressed in various brain regions including hippocampus and detectable in presynaptic terminals (56, 58), postsynaptic terminals (59, 60), intracellular organelles (61, 62) and also on astrocytes (26, 39–42, 55, 57, 63). However, the effects of astroglial CB1R-mediated signaling on synaptic transmission and plasticity is highly debated. It has been shown that exogenous administration of Δ⁹- tetrahydro-cannabinol (THC) lead to a temporally prolonged and spatially widespread activation of astroglial CB1 receptors and trigger glutamate release, which activates postsynaptic NMDARs and induces LTD in the CA3–CA1 hippocampal synapses, resulting working memory deficit (57). Araque and colleagues reported that eCB released from depolarized CA1 pyramidal cell activates astroglial CB1Rs with a shorter and localized way and induces glutamate release, which activates lateral presynaptic mGluRs and induces LTP (41, 42). In another study, Robin et al found that high frequency stimulation (HFS) of Schaffer Collateral induces LTP, which is gated by the activation of astroglial CB1Rs and the release of D-serine from astrocytes (26). These diverse consequences by activation of CB1 receptor may be due to different neuronal activity pattern that induce eCB release and different nature of the agonists. Interestingly, it has been discovered that individual hippocampal astrocytes are capable of releasing both ATP/adenosine and glutamate, and that this release occurs in a time-dependent and activity-sensitive manner in response to neuronal interneuron activity (64). These findings suggest that the specific type and intensity of astrocyte stimulation plays a critical role in determining the downstream signaling pathways that are triggered by CB1R activation in astrocytes. Consistent with previous study, we found that the activation of astroglial CB1Rs induces Ca²⁺ elevation and triggers the release of D-serine that binds to postsynaptic NMDAR and induces LTP formation (26). Our results provide evidence that astroglial CB1R-mediated singling not only modulates the E→E synapses, but also regulates E→I synapses.

The precise mechanisms through which neurons and astrocytes differentially regulate D-serine levels are yet to be fully elucidated (65–67). However, it is evident that astrocytes play a significant role in regulating the availability of D-serine. The enzyme serine racemase catalyzes the conversion of L-serine into D-serine, which was initially found in astrocytes and microglia in the mammalian brain (29, 68, 69). It should be highlighted that serine racemase has also been detected in neurons (70–72). A study showed that, despite a significant reduction in SR protein levels in the brains of neuronal SR knockout mouse brains, the reduction in D-serine levels was minimal, suggesting that neurons are not the exclusive source of D-serine (70) and neurons may produce and release D-serine under certain conditions. Notably, activation of G protein-coupled receptors in astrocytes through chemogenetic means results in increased synaptic transmission, which is dependent on the availability of D-serine (52, 73). Specifically, the release of D-serine, which is regulated by CB1R in astrocytes, is required for Ca²⁺-dependent modulation of LTP in vivo (26), as well as the threshold and amplitude of dendritic spikes (74). Moreover, recent studies have shown that conditional connexin double knockout (75) or knockdown of α4nAChR (76) in astrocytes can decrease the extracellular concentration of D-serine, which in turn reduces NMDAR-dependent synaptic potentiation. These findings suggest that astrocytes are the primary source of D-serine, which plays a crucial role in modulating the function of NMDARs.

It is well established that LTP_E→E observed in the CA1 region of the hippocampus is triggered by activation of NMDARs. However, LTP_E→I is less studied compared to LTP_E→E, but recent evidence suggests that it also involves the activation of NMDARs (10, 11, 31, 32, 53, 77). There have been reports of NMDAR-dependent LTP_E→I in various regions of the brain, including the hippocampus and cortex (10, 11). Our results suggest that different synaptic mechanisms are involved in the induction of LTP_E→I in different subregions of the hippocampus. The stratum radiatum, where interneuorns contain NMDARs, is known to be sensitive to NMDAR-dependent LTP_E→I. In contrast, the stratum oriens, where interneuorns contain CP-AMPARs, appears to rely on the activation of CP-AMPA receptors for LTP_E→I induction. Notably, our findings suggest that astrocytes contribute to NMDAR signaling in the induction of LTP_E→I in the stratum radiatum through the modulation of extracellular levels of the co-agonist D-serine. It is worth noting that our study found that prolonged activation of astrocytes via the Gq-DREADD pathway resulted in a substantial and persistent increase in Ca²⁺ events and significantly potentiated EPSP responses. This is in consistent with earlier observations made by other groups regarding LTP_E→E (52, 73). Above all, the mechanism of LTP_E→I in stratum radiatum appears to be share by LTP_E→E observed in the CA1 region.

A previous study has demonstrated that knocking down γCaMKII from interneurons can disrupt LTP_E→I and cognitive function (53, 54). Our results confirmed that knocking down γCaMKII in interneurons of the stratum radiatum also leads to disruption of LTP_E→I and cognitive function. Ma and colleagues showed that hippocampal network oscillations in the gamma and theta bands were significantly weaker in γCaMKII knockout mice compared to wild-type mice, following learning (53). This suggests that impaired experience-dependent oscillations in the hippocampus of γCaMKII PV-KO mice may lead to cognitive dysfunction. In this respect, it will be intriguing to investigate the network oscillation after learning in our condition in the future study.

In the hippocampus, GABAergic local circuit inhibitory interneurons make up approximately 10-15% of the total neuronal cell population (78). However, these interneurons are diverse in their subtypes, morphology, distribution, and functions (15, 79). In our study, we mainly focused on a subpopulation of interneurons in the stratum radiatum of hippocampus. Although it is unclear which type of interneuron recorded in our study, but our study indicated that most of interneurons in the stratum radiatum do not express CP-AMPARs but express abundant of NMDARs. These findings are in line with a previous study conducted by Lasmsa et al. (32). In our study, we observed that approximately 80% of interneurons in stratum radium were able to induce LTP successfully. This finding contrasts with the observation made by Lamsa et al., who reported a figure of approximately 52% interneurons capable of inducing LTP. The reason for this discrepancy could be attributed to differences in the induction protocol used in the respective studies.

Our study corroborates earlier research that suggests distinct synaptic mechanisms are involved in LTP induction in the CA1 region of the hippocampus across different subregions (11, 13, 32). However, the major breakthrough of our study is the demonstration that astrocytic function serves as the gating mechanism for LTP _E→I induction in the stratum radiatum. Additionally, our data reveals that the activation of astrocytes via the Gq-DREADD pathway produces de novo long-lasting potentiation of EPSP in stratum radiatum interneurons and the knockdown of γCaMKII disrupts cognitive function. These results shed light on the complex mechanisms underlying learning and memory in the hippocampus, and may have implications for developing new therapies targeted at modulating astrocytic function for the treatment of memory disorders.

Acknowledgements

This work was supported by grants from key R&D Program Project of Zhejiang (2022C03034), Zhejiang Provincial Natural Science Foundation of China (LQ23C090001), Scientific Research Foundation of Zhejiang University City College, (X-202103), Scientific Research Foundation of Hangzhou City University (J-202325), Science and Technology Department of Zhejiang Province (2021RC051).

Conflict of interest

The authors declare no competing interests.

Data availability statement

Data are available from the corresponding author on reasonable request.

Author contributions

W.S. and L.Z. designed the study; W.S. Y.T. and L.Z. analyzed data, wrote the manuscript; W.S., Y.T. J.Y. and L.Z. carried out most experiments; F.Z. and W.Z. helped with viral injection; Y.X. and J.D. helped with immunohistochemistry; J.D. and F.Z. helped with slice electrophysiology; W.Z. and L.X. helped with data analysis. All authors discussed the results and comments on the manuscript.

Supporting Figures

Fig. S1.

Fig. S2.

Fig. S3.

Fig. S4.

Fig. S5.

Fig. S6.

Fig. S7.

Fig. S8.

Fig. S9.