The α2δ-1–NMDA receptor coupling is essential for corticostriatal long-term potentiation and is involved in learning and memory

The striatum receives extensive cortical input and plays a prominent role in motor learning and habit formation. Glutamate N-methyl-d-aspartate (NMDA) receptor (NMDAR)-mediated long-term potentiation (LTP) is a major synaptic plasticity involved in learning and memory. However, the molecular mechanism underlying NMDAR plasticity in corticostriatal LTP is unclear. Here, we show that theta-burst stimulation (TBS) consistently induced corticostriatal LTP and increased the coincident presynaptic and postsynaptic NMDAR activity of medium spiny neurons. We also found that α2δ-1 (previously known as a subunit of voltage-gated calcium channels; encoded by the Cacna2d1 gene) physically interacted with NMDARs in the striatum of mice and humans, indicating that this cross-talk is conserved across species. Strikingly, inhibiting α2δ-1 trafficking with gabapentin or disrupting the α2δ-1–NMDAR interaction with an α2δ-1 C terminus–interfering peptide abolished TBS-induced LTP. In Cacna2d1-knockout mice, TBS failed to induce corticostriatal LTP and the associated increases in presynaptic and postsynaptic NMDAR activities. Moreover, systemic gabapentin treatment, microinjection of α2δ-1 C terminus–interfering peptide into the dorsomedial striatum, or Cacna2d1 ablation impaired the alternation T-maze task and rotarod performance in mice. Our findings indicate that the interaction between α2δ-1 and NMDARs is of high physiological relevance and that a TBS-induced switch from α2δ-1–free to α2δ-1–bound NMDARs is critically involved in corticostriatal LTP and LTP-associated learning and memory. Gabapentinoids at high doses may adversely affect cognitive function by targeting α2δ-1–NMDAR complexes.

synaptic NMDAR activity of MSNs in the dorsomedial striatum. We also show the important role of ␣2␦-1-NMDAR complexes in the physiological process such as striatal learning and memory. This information advances our understanding of the molecular mechanisms that govern the synaptic plasticity engaged in learning and memory.

Theta-burst stimulation induces corticostriatal LTP through synaptic NMDARs
Theta-burst stimulation (TBS) can mimic the theta rhythm firing of dorsomedial neurons during various cognitive tasks (24,25). In extracellular recording of population spikes, TBS is much more effective than is single tetanic stimulation at evoking LTP in the dorsomedial striatum (8). MSNs constitute the major cell type, comprising ϳ95% of striatal neurons in rodents (26). We performed whole-cell recording of excitatory postsynaptic potentials (EPSPs) in individual MSNs to determine the TBS-induced LTP in the dorsomedial striatum of mice. The MSNs were visually identified in brain slices and were characterized by inward rectification and delayed firing in response to depolarization (Fig. 1A), as reported previously (27,28). As expected, TBS (10 trains of stimuli spaced at 10-s intervals, with each train containing bursts of 4 spikes at 100 Hz, repeated 10 times at 5 Hz) (8) in cortical layer VI reliably induced a profound increase in the amplitude of EPSPs in all MSNs after a latency of ϳ10 min; this increase lasted for at least 80 min (Fig.  1). The time course of this electrical TBS-induced LTP expression (via whole-cell current-clamp recording in parasagittal striatal slices) was similar to what was induced using an optical TBS approach (7).
Presynaptic NMDARs at cortical axon terminals are critically involved in TBS-induced corticostriatal LTP (7). We next determined whether postsynaptic NMDARs are similarly involved in corticostriatal LTP induced by TBS. We blocked the postsynaptic NMDARs of MSNs through intracellular dialysis of MK-801 (50 M), an NMDAR open-channel blocker, included in the pipette recording solution (29). TBS failed to induce LTP in all MSNs when an MK-801-containing intracellular solution was used for whole-cell recording (Fig. 1B). Thus, both presynaptic and postsynaptic NMDARs are equally important to corticostriatal LTP induction.
In addition, we used Cacna2d1-knockout mice (17,33) to validate the critical role of ␣2␦-1 in TBS-induced LTP. TBS reliably induced LTP in all MSNs recorded in brain slices obtained from WT mice. In contrast, TBS did not significantly change the amplitude of EPSPs in any of the MSNs examined in Cacna2d1-knockout mice (Fig. 3B). Taken together, these findings indicate that ␣2␦-1 plays a pivotal role in TBS-induced corticostriatal LTP.

␣2␦-1 is essential for increased presynaptic and postsynaptic NMDAR activity of MSNs by TBS
Because both ␣2␦-1 and NMDARs are required for TBSinduced LTP in the dorsomedial stratum, we determined to , co-IP analysis shows the protein-protein interaction between ␣2␦-1 and GluN1 in the membrane extracts from striatal tissues of mice (A) and humans (B, two separate subjects). The proteins were immunoprecipitated initially with a mouse anti-␣2␦-1 antibody or IgG. Western immunoblotting was performed using a rabbit anti-GluN1 or a rabbit anti-␣2␦-1 antibody. IgG and input (tissue lysates only, without immunoprecipitation) were used as negative and positive controls. Similar data were obtained from striatal tissues from each of the four human donors. C, a schematic shows the determination of relative amounts of GluN1 proteins in the membrane fraction of the mouse striatum using a two-step co-IP protocol. D and E, representative gel images (two pairs of samples) and quantification show the relative amount of GluN1 proteins present in the input (total membrane protein lysates) and flow-through (FT) samples from mouse striatal tissues (n ϭ 6 samples/group). F, original gel images show that both GluN1 and ␣2␦-1 were present in samples pulled down by ␣2␦-1 antibody. However, GluN1, but not ␣2␦-1, was detected in FT samples subjected to immunoprecipitation using GluN1 antibody (shown in C). M, molecular marker. *, p Ͻ 0.05 (versus input). Paired Student's t test was used.

␣2␦-1bound NMDA receptors in learning
what extent ␣2␦-1 is involved in the increased synaptic NMDAR activity of MSNs during LTP induction. To assess the presynaptic NMDAR activity of MSNs associated with TBSinduced LTP, we recorded miniature excitatory postsynaptic currents (mEPSCs), which reflect quantal glutamate release from presynaptic terminals (29,36). The mEPSCs of MSNs were recorded at 20 -30 min (at which time LTP was stably induced by TBS; Figs. 1 and 3) after TBS. In brain slices obtained from WT control mice, the frequency of mEPSCs of MSNs was low in the absence of TBS and was not significantly altered by bath application of 50 M AP5 (Fig. 4, A-C). In contrast, TBS caused a large increase in the frequency, but not the amplitude, of mEPSCs in all MSNs of WT mice. In addition, this TBS-induced increase in mEPSC frequency was completely restored to the level of MSNs without TBS within 5 min after bath application of AP5 (Fig. 4, A-C). The baseline frequency and amplitude of mEPSCs in MSNs were comparable in WT and Cacna2d1-knockout mice. In brain slices from Cacna2d1knockout mice, however, TBS failed to alter the frequency or amplitude of MSNs. Furthermore, bath application of AP5 had no effect on the frequency and amplitude of MSNs in brain slices from Cacna2d1-knockout mice, with or without TBS (Fig. 4, A-C).
To determine whether ␣2␦-1 is involved in the postsynaptic NMDAR activity augmented by TBS, we measured the current elicited by puff application of NMDA (37) directly onto the MSN being recorded. In MSNs from WT mice, puff application of 100 M NMDA caused a much larger current amplitude 20 -30 min after TBS than that without TBS (Fig. 4D). The baseline amplitude of currents elicited by the puff application of NMDA was similar in WT and Cacna2d1-knockout mice. However, in MSNs from Cacna2d1-knockout mice, TBS had no significant effect on the amplitude of currents elicited by puff application of NMDA compared with that without TBS (Fig. 4D). These data indicate that ␣2␦-1 is critically involved in the potentiated coincident presynaptic and postsynaptic NMDAR activity of MSNs during TBS-induced corticostriatal LTP.

␣2␦-1 contributes to spatial learning and working memory
Adult WT and Cacna2d1-knockout mice were indistinguishable from one another on the basis of a visual inspection for overt abnormalities. We performed an open-field test to assess the exploration and locomotor activity of WT and Cacna2d1-knockout mice. WT and Cacna2d1-knockout mice did not show any significant difference in the total travel distance or the total time spent in the inner zone and outer zone (Fig. 5).
The electrophysiological studies described above demonstrate that TBS-induced corticostriatal LTP depends on ␣2␦-1-bound NMDARs. The NMDARs in the dorsomedial striatum play a crucial role in spatial learning and working memory (11,38,39). However, it is unclear whether ␣2␦-1bound NMDARs are involved in this process. The natural tendency of rodents in a T-maze is to alternate their choice of goal arms, and spatial learning and working memory can be assessed using an alternation T-maze test (13,38,40). In our study, the mice were tested with a rewarded alternation T-maze task (five trials per day, 20-min intervals between trials) for five consecutive days. The percentage of correct alternations was recorded for each trial daily. Gabapentin (100 g/kg, intraperitoneally) or vehicle (normal saline) was injected 20 min before the first trial each day. The mice injected with saline showed a gradual increase in the percentage of correct alternations in the second run of the T-maze task over 5 days (Fig. 6A). Compared with salinetreated mice, treatment with gabapentin significantly reduced the percentage of correct alternation in the T-maze task starting from day 3 (Fig. 6A).

␣2␦-1bound NMDA receptors in learning
We also determined directly whether the ␣2␦-1-NMDAR complex in the dorsomedial striatum is involved in corticostriatal learning. The ␣2␦-1Tat peptide or control peptide (200 pmol, 200 nl) was microinjected bilaterally into the dorsomedial striatum through the implanted cannula 20 min before initiating each T-maze task training. Mice injected with the control peptide exhibited a steady increase in the percentage of correct alternations over 5 days of training. However, mice injected with ␣2␦-1Tat peptide showed a significant reduction in the percentage of correct alternation starting from day 3, compared with mice receiving the control peptide microinjection (Fig. 6B).
We next determined the potential difference between WT and Cacna2d1-knockout mice in the performance of the T-maze test. Both groups of mice were subjected to the alternation T-maze test for 5 consecutive days. The WT mice showed a gradual improvement in the correct turning response on the second run over 5 days (Fig. 6C). By comparison, the correct alternation of the T-maze task was significantly lower in Cacna2d1-knockout mice than in WT mice throughout the 5 days (Fig. 6C). These data collectively suggest that the ␣2␦-1-NMDAR complex in the dorsomedial striatum is critically involved in learning and memory.

␣2␦-1 is involved in motor skill learning
The NMDARs in the dorsal striatum also contribute to motor skill learning (11,41). Thus, we used the accelerating rotarod test (three trials per day, with a 20-min interval between trials, for 2 consecutive days) to determine the role of ␣2␦-1 in motor coordination learning. Gabapentin (100 g/kg, intraperitoneally) or vehicle saline was injected 20 min before the first trial each day. Mice receiving saline injection showed a gradual increase in the latency to fall from the rotarod over the six trials (Fig. 7A). Although the two groups of mice exhibited increased durations on the rotating cylinder with successive trials, the falling latency was significantly shorter in mice receiving gabapentin treatment than in those injected with vehicle at trials 3-6 ( Fig. 7A).
We then determined whether the ␣2␦-1-NMDAR complex in the dorsomedial striatum is involved in the motor learning. The control peptide or ␣2␦-1Tat peptide (200 pmol, 200 nl) was microinjected bilaterally into the dorsomedial striatum 20 min before each training. Mice injected with the control peptide displayed a continuing increase in the falling latency with successive trials (Fig. 7B). By comparison, mice injected with ␣2␦-1Tat peptide showed a significantly lower improvement in the falling latency (Fig. 7B).
In addition, we used the accelerating rotarod test to evaluate motor learning in WT and Cacna2d1-knockout mice. Both groups of mice showed an incremental increase in the latency to fall from the rotarod over the six successive trials (Fig. 7C). However, Cacna2d1-knockout mice had significantly less time on the rotarod than did WT mice throughout the six trials (Fig.  7C). Together, these results strongly suggest that ␣2␦-1bound NMDARs in the striatum contribute to working memory and motor learning.

Discussion
The striatum is well-known for its role in developing habits and acquiring motor skills (42,43). Striatum learning requires changes in the strength of synaptic connections during the memorization of a complex task. The LTP of excitatory cortical afferents to the dorsal striatum likely occurs when learning to encode new skills and habits (44,45). Activation of cortical afferents with the theta-ranged frequency can induce corticostriatal LTP in striatal MSNs in vivo (45) and in brain slices in the physiological Mg 2ϩ concentration (7,8). Blocking or genetically deleting NMDARs can prevent corticostriatal LTP induced by TBS (7, 8). The theta-burst rhythm is considered a naturally occurring cell activity pattern, and TBS exploits the

␣2␦-1bound NMDA receptors in learning
endogenous circuit properties to maximize NMDAR activation with the least amount of afferent stimulation (46). In corticostriatal LTP induction, the enhanced presynaptic NMDAR activity augments glutamate release from cortical afferent terminals through Ca 2ϩ influx (7). We demonstrated that blocking postsynaptic NMDARs also completely blocked corticostriatal LTP in MSNs. Furthermore, we showed that TBS caused a large increase in the activity of presynaptic and postsynaptic NMDARs in corticostriatal synapses. These findings strongly suggest that TBS-induced corticostriatal LTP is associated with the potentiated activity of both presynaptic and postsynaptic NMDARs.
␣2␦-1 is expressed in distinct brain regions involved in learning and memory, including the cortex, hippocampus, and striatum (21,22). ␣2␦-1 may interact with thrombospondin, an astrocyte-secreted protein, to affect synaptogenesis (but not already-formed synapses) (47). However, the results of a recent study suggest that the association between ␣2␦-1 and thrombospondin is rather weak and that there is no ␣2␦-1-thrombospondin interaction on the cell surface (48). ␣2␦-1 also interacts with BK channels to indirectly modulate VGCC activity (49), although this interaction may not be relevant to TBS-induced LTP. In this study, we showed that ␣2␦-1 physically interacted with NMDARs in the striatum in both mice and humans. This finding is consistent with the results of our recent studies showing that ␣2␦-1 interacts directly with NMDARs to promote their synaptic trafficking in the spinal cord and brain (17,35,50,51). We showed that ␣2␦-1 protein was not detected in flowthrough membrane protein samples after further immunoprecipitation with the GluN1 antibody. These data suggest that most, if not all, ␣2␦-1 membrane proteins in the striatum may be associated with NMDARs. Importantly, we found that targeting ␣2␦-1bound NMDARs with gabapentin or ␣2␦-1Tat peptide abolished TBS-induced corticostriatal LTP. In addi-tion, TBS was unable to induce LTP in the corticostriatal synapses of Cacna2d1-knockout mice. Therefore, the switch from ␣2␦-1-free to ␣2␦-1bound NMDARs plays an obligatory role in corticostriatal LTP induced by TBS.
The most striking finding of our study is that TBS-induced corticostriatal LTP depends on ␣2␦-1bound NMDARs at both presynaptic and postsynaptic sites. Activation of NMDARs is a critical requirement for the induction of LTP at corticostriatal synapses (7,8). Although postsynaptic NMDARs of MSNs are tonically active, blocking NMDARs had no effect on the baseline frequency of mEPSCs in MSNs in the absence of TBS, suggesting that presynaptic NMDARs are normally latent and not functionally active under basal conditions. We demonstrated that TBS increased the activity of both presynaptic and postsynaptic NMDARs in corticostriatal synapses. Thus, TBSinduced corticostriatal LTP may depend on coincident activity in presynaptic and postsynaptic neurons, resulting in calcium influx through synaptic NMDARs. Sustained synaptic glutamate release via presynaptic NMDAR activation can lead to long-lasting stimulation of postsynaptic NMDARs, which can activate calcium-dependent signaling, including AMPA receptor insertion and lasting structural changes at dendritic spines (52)(53)(54). We showed that inhibiting ␣2␦-1 activity with gabapentin, interfering with the ␣2␦-1-NMDAR interaction using ␣2␦-1Tat peptide, and genetic ablation of ␣2␦-1 completely blocked the increase in presynaptic and postsynaptic NMDAR activity during LTP induction. These findings clearly indicate that ␣2␦-1bound NMDARs account for most, if not all, increased NMDAR activity by TBS at the corticostriatal synapses. Therefore, ␣2␦-1bound NMDARs are required for the coordination of functional changes in synaptic efficacy in TBSinduced LTP.
Another major finding of our study is that targeting the ␣2␦-1-NMDAR complex in the dorsomedial striatum or the aboli-

␣2␦-1bound NMDA receptors in learning
tion of ␣2␦-1 attenuates performance on T-maze and rotarod tests, both of which are associated with striatum-based learning and memory. These findings support the notion that ␣2␦-1bound NMDARs contribute to learning and memory. The dorsomedial striatum lesions impair the acquisition of spatial alternation behavior by disrupting the signal necessary to link a goal with a specific spatial sequence (13,38,55). The NMDARs in the dorsal striatum are involved in learning and memory (11,39,56). It should be noted that learning and memory involve neural circuits among the cortex, hippocampus, and striatum that coordinate information storage, retrieval, and manipulation to correlate task performance (1,55,57). Because ␣2␦-1 is extensively expressed in the cortex, amygdala, and hippocampus, ␣2␦-1bound NMDARs in these brain regions are also likely involved in various learning, association, motivational, and cognitive functions.
In summary, we provide new evidence that ␣2␦-1 is essential for corticostriatal LTP induction through its interaction with

␣2␦-1bound NMDA receptors in learning
NMDARs. ␣2␦-1 may mobilize and recruit NMDARs at both presynaptic and postsynaptic sites during TBS-induced corticostriatal LTP. Our findings reveal a critical role for ␣2␦-1 in dynamic NMDAR regulation in LTP induction and in learning and memory. This information provides new insight into the molecular mechanism that is fundamentally important for synaptic plasticity and the associated learning and memory.

Animals
All surgical procedures and protocols were approved by the Animal Care and Use Committee of the University of Texas M. D. Anderson Cancer Center. Adult male and female mice (8 -10 weeks old), housed three to four per cage, were used for final experiments. WT and Cacna2d1-knockout mice (C57BL/6 genetic background) were generated as described previously (33). Two breeding pairs of heterozygous (Cacna2d1 ϩ/Ϫ ) mice were purchased from Medical Research Council (Harwell Didcot, Oxfordshire, UK), and Cacna2d1 Ϫ/Ϫ mice (knockout) and Cacna2d1 ϩ/ϩ (WT) littermates were obtained by breeding the heterozygous mice. The animals were ear-marked at the time of weaning (21-23 days after birth), and tail biopsies were used for genotyping. The housing room was maintained at 23°C on a 12-h light/dark cycle.

Co-immunoprecipitation of GluN1 and ␣2␦-1 in the striatum tissue
We obtained frozen human striatum tissues (two males and two females; age range, 66 -77 years) from the Harvard Brain Tissue Resources Center, a NeuroBioBank Repository funded by the National Institutes of Health. The striatum tissues from humans and mice were dissected and homogenized in ice-cold hypotonic buffer (20 mM Tris, 1 mM CaCl 2 , 1 mM MgCl 2 , and protease inhibitors, pH 7.4). The nuclei and large debris were removed by centrifugation at 300 ϫ g for 5 min. The supernatant was centrifuged for 20 min at 21,000 ϫ g. The pellets were resuspended and solubilized in immunoprecipitation buffer (50 mM Tris, 250 mM NaCl, 10% glycerol, 0.5% Nonidet P-40, 20 mM NaF, 1 mM Na 3 VO 4 , 10 mM N-ethylmaleimide, 1 mM phenylmethylsulfonyl fluoride, 2 mM benzamide, and protease inhibitors, pH 7.4). The membrane protein was incubated at 4°C overnight with protein A/G beads (catalog no. 16-266, Millipore) prebound to the mouse anti-␣2␦-1 (catalog no. sc-271697, Santa Cruz Biotechnology, Dallas, TX). Protein A/G beads prebound to mouse IgG were used as controls. The samples were washed five times with immunoprecipitation buffer and then immunoblotted. Furthermore, flow-through samples containing the same amount of proteins were incubated at 4°C overnight with protein A/G beads prebound to the mouse anti-GluN1 antibody (catalog no. 75-272, NeuroMab, Davis, CA). GluN1 and ␣2␦-1 protein bands were detected using Western immunoblotting. The following antibodies were selected for immunoblotting and validated in our previous study (17): rabbit anti-␣2␦-1 (catalog no. C5105, 1:1,000, Sigma-Aldrich) and rabbit anti-GluN1 (catalog no. G8913, 1:1,000, Sigma-Aldrich). The protein bands were visualized with an ECL kit (Thermo Fisher Scientific), and protein-band intensity was visualized and quantified using the Odyssey Fc Imager (LI-COR Biosciences, Lincoln, NE).

Brain slice preparation
Under deep anesthesia with 5% isoflurane, the mice were decapitated; their brains were quickly removed and placed into ice-cold artificial cerebrospinal fluid composed of 124.0 mM NaCl, 3.0 mM KCl, 1.3 mM MgSO 4 , 2.4 mM CaCl 2 , 1.4 mM NaH 2 PO 4 , 10.0 mM glucose, and 26.0 mM NaHCO 3 , which was saturated with a mixture of 95% O 2 and 5% CO 2 . The tissue block was fixed on the stage of vibratome (Leica) and sagittally sectioned into 300-m-thick slices. The striatum slices were then transferred into 32°C artificial cerebral spinal fluid that was continuously oxygenated with a mixture of 95% O 2 and 5% CO 2 and incubated for at least 1 h before the electrophysiological recording was obtained. The slices were placed in the recording chamber, fixed, and perfused at 3 ml/min with oxygenated artificial cerebral spinal fluid at a temperature of 34°C, maintained using an inline solution heater. The recording electrodes were pulled from borosilicate glasses with a final resistance of 3-5 M⍀ when filled with the following intracellular solution: 140.0 mM potassium gluconate, 2.0 mM MgCl 2 , 0.1 mM CaCl 2 , 10.0 mM HEPES, 1.1 mM EGTA, 0.3 mM Na 2 -GTP, and 2.0 mM Na 2 -ATP adjusted to pH 7.25 with 1 M KOH, 270 -290 mOsm. The whole-cell recordings were performed using a Multiclamp 700B amplifier (Molecular Devices), and all signals were filtered at 2 kHz, digitized at 20 kHz using Digidata 1320A, and stored on the computer for off-line analysis. The medium spiny neurons (MSNs) in the dorsomedial striatum were visualized by IR differential interference optics with a water-immersion objective. MSNs were also identified by their intrinsic membrane properties: resting membrane potential more negative than Ϫ80 mV, inward and outward rectification in response to somatic current injection, and a long depolarization ramp to the action potential threshold leading to a delayed spike discharge (27,28).

LTP induction
Electrical stimulation was applied to elicit LTP using a tungsten bipolar electrode placed on cortical layer VI close to the white matter (8). After a whole-cell recording was made from MSNs in the dorsomedial striatum, a "test" stimulus was applied at 0.017 Hz to elicit EPSPs. The duration of all stimulation pulses was 0.1 ms, and the stimulation intensity was adjusted to a level at which the baseline EPSP amplitude was ϳ5 mV. Recordings were rejected if the peak amplitude of baseline EPSPs was less than 2 mV, if the resting membrane potential changed by more than 10%, or if the input resistance changed more than 30% during recordings. The stable responses were recorded for at least 10 min before applying TBS, which consisted of 10 trains of stimuli (each train contained bursts of 4 spikes at 100 Hz, and the bursts repeated 10 times at 5 Hz) at 10-s intervals (8). The intensity and pulse duration of stimulation for baseline and LTP induction were kept the same for the same neurons. After TBS, EPSP were continuously recorded for at least another 80 min with a single stimulus pulse at 0.017 Hz, and the EPSP amplitude was analyzed using MiniAnalysis software (Synaptosoft, Inc., Decatur, GA). All experiments ␣2␦-1bound NMDA receptors in learning were performed in artificial cerebrospinal fluid containing normal Mg 2ϩ and in the presence of picrotoxin (100 M) to eliminate GABAergic input (7,8). Gabapentin was purchased from Tocris Bioscience (Bristol, UK), and ␣2␦-1Tat peptide and control peptide were synthesized by Bio Basic Inc. (Marham, Ontario, Canada) and validated using LC and MS.

Recording presynaptic and postsynaptic NMDAR activity
The mEPSCs were recorded from MSNs in the dorsomedial striatum at a holding potential Ϫ60 mV in the presence of 1 M tetrodotoxin (37,58). The mEPSCs were analyzed, and cumulative responses of frequency and amplitude were calculated using MiniAnalysis software (Synaptosoft, Inc.). To measure postsynaptic NMDAR activity, puff NMDA-elicited currents were recorded in MSNs. In brief, NMDAR currents were recorded at a holding potential of Ϫ60 mV and elicited by puff application of NMDA through a Pressure System IIe (Toohey Company, Fairfield, NJ). The puff pipette (15-m tip diameter) was placed ϳ150 m away from the recorded cells. Positive pressure (3 p.s.i.) was applied (durations of 200 ms) to eject NMDA (100 M) onto the recorded cell to elicit currents with the intracellular solution containing QX314, a sodium channel blocker. Because the NMDAR channel is voltage-dependently blocked by Mg 2ϩ at a negative holding potential and co-activated by glycine, puff NMDA-elicited currents were recorded in Mg 2ϩ -free solution and in the presence of 10 M glycine (37). In all electrophysiological experiments, only one neuron was recorded from each hemislice, and at least three mice were included for recording in each group.

Cannula implantation and microinjection
The mice were anesthetized via inhaling 1-2% isoflurane, and their heads were stereotaxically fixed using the adapter and ear bars. An incision was made along the middle line on the head, and two small holes were drilled at the coordinates: anteroposterior ϩ0.5 mm and mediolateral 1.5 mm. The guide cannulas were bilaterally implanted dorsally and fixed using the dental acrylic cement. The guide cannula was made of stainless steel (2.3 mm long, catalog no. 62004, RWD Life Science Inc., San Diego, CA). A dummy cannula was kept in the guide cannula for 10 days before conducting behavioral tests. For the dorsomedial microinjection, the mice were briefly anesthetized with 1% isoflurane. The injection cannula (diameter, 0.11 mm) was 0.5 mm longer than the guide cannula. The dorsomedial striatum was targeted at anteroposterior ϩ0.5 mm, mediolateral 1.5 mm, and dorsoventral Ϫ2.8 mm (59). Each microinjection took ϳ3 min. At the end of the behavioral tests, red Fluo-Spheres (0.4 m, 100 nl) were bilaterally microinjected into the dorsomedial striatum in each mouse. The brain was fixed with 4% paraformaldehyde for 72 h and coronally cut to 50-mthick slices to confirm the injection site. We excluded the respective behavioral data if the microinjection missed the dorsomedial striatum in mice.

Open-field test
The open-field test was performed in a quiet behavioral testing room to examine the exploration and locomotor activity of mice (60). The mice were allowed to acclimate in the room for 30 min before starting the test. The size of the plastic open-field box is 40 cm ϫ 40 cm ϫ 35 cm (Noldus, Boston, MA). The central inner zone was defined as a square of 24 cm ϫ 24 cm within the open-field testing area, whereas the remaining outer space was marked as the outer zone. The open-field testing area was wiped with 95% ethanol before each test. The mice were placed in the middle of the testing area and allowed to move freely for 10 min. The travel path of each mice was video-recorded using a GigE color camera (Noldus). The total travel distance and total time spent in the inner zone and outer zone for each mice were analyzed using Autotyping software (61).

Rewarded alternation T-maze test
The mice were initially food-restricted for 5 days to achieve 85-90% of their free body weight (40). The mice were trained to habituate in a T-maze apparatus with two open-side (goal) arms for 3 min, twice a day with a 30-min interval, for 5 days before starting the appetite-motivated T-maze task. The test contained five trials at 20-min intervals for 5 consecutive days. Each trial was composed of two runs in order. On the first run, the correct arm (closed) was randomly defined to avoid adaption. The mouse was placed in the start area at the bottom of T-maze stem, and the door in the start area was opened to allow the mouse to explore the chocolate reward at the end of the opened arm. After the reward had been completely consumed, the mice were guided back to the start area for the second run. On the second run, the correct arm was also opened, and the mouse had to choose between the previously entered arm and the correct arm. If it entered the correct arm (correct alternation), it was given time to completely consume the chocolate reward; if not, the mouse received no reward. No more than 2 min was spent in each trial. For five trials conducted each day, the percentage of correct alternation was calculated for each mouse (40). During the T-maze test, the mice were weighed every day to verify maintenance of 85-90% of their free body weight.

Rotarod performance test
The rotarod test was conducted in mice using an accelerating rotarod (IITC Life Science, Woodland Hills, CA). The mice were initially placed on a rotarod opposite to the rotating direction. The initial rotating speed was 4 rpm/min, which progressively increased to a maximum of 40 rpm/min over 300 s for each trial (41,62). Each animal was given three trials at 20-min intervals per day for 2 consecutive days. Overall performance was expressed as the latency to fall from the accelerating rotating rod. These latencies were automatically detected and measured using a timer built into the apparatus. The mouse was given a new trial if it immediately fell down at the beginning of each trial.

Study design and statistical analysis
All data were presented as means Ϯ S.E. No statistical methods were used to predetermine sample sizes for biochemical ␣2␦-1bound NMDA receptors in learning studies, but our sample sizes were similar to those generally employed in the field. For proper exclusion of data points, the criteria were established before data collection. In electrophysiological recording experiments, we monitored cell capacitance, input resistance, series resistance, resting membrane potential, and baseline holding current. We excluded cells if the recording indicated a rundown condition. The peak amplitude of EPSPs and puff NMDAR currents was determined and analyzed using pClamp 10 (Molecular Devices). The amplitude and frequency of mEPSCs were analyzed with a peak detection program (MiniAnalysis, Synaptosoft, Leonia, NJ). Detection of events was accomplished by setting a threshold above the noise level. The distribution cumulative probability of the amplitude and interevent interval of mEPSCs was compared using the Komogorov-Smirnov test, which estimates the probability that two cumulative distributions are similar. The behavioral test and electrophysiological data were analyzed using one-way ANOVA followed by Dunnett's and Tukey's post hoc tests or two-way ANOVA followed by Tukey's post hoc test. p Ͻ 0.05 was considered statistically significant.