Selective Role of RGS9-2 in Regulating Retrograde Synaptic Signaling of Indirect Pathway Medium Spiny Neurons in Dorsal Striatum

In the striatum, medium spiny neurons (MSNs) are heavily involved in controlling movement and reward. MSNs form two distinct populations expressing either dopamine receptor 1 (D1-MSN) or dopamine receptor 2 (D2-MSN), which differ in their projection targets and neurochemical composition. The activity of both types of MSNs is shaped by multiple neuromodulatory inputs processed by GPCRs that fundamentally impact their synaptic properties biasing behavioral outcomes. How these GPCR signaling cascades are regulated and what downstream targets they recruit in D1-MSN and D2-MSN populations are incompletely understood. In this study, we examined the cellular and molecular mechanisms underlying the action of RGS9–2, a key GPCR regulator in MSNs implicated in both movement control and actions of addictive drugs. Imaging cultured striatal neurons, we found that ablation of RGS9–2 significantly reduced calcium influx through NMDARs. Electrophysiological recordings in slices confirmed inhibition of NMDAR function in MSNs, resulting in enhanced AMPAR/NMDAR ratio. Accordingly, male mice lacking RGS9–2 displayed behavioral hypersensitivity to NMDAR blockade by MK-801 or ketamine. Recordings from genetically identified populations of striatal neurons revealed that these changes were selective to D2-MSNs. Surprisingly, we found that these postsynaptic effects resulted in remodeling of presynaptic inputs to D2-MSNs increasing the frequency of mEPSC and inhibiting paired-pulse ratio. Pharmacological dissection revealed that these adaptations were mediated by the NMDAR-dependent inhibition of retrograde endocannabinoid signaling from D2-MSNs to CB1 receptor on presynaptic terminals. Together, these data demonstrate a novel mechanism for pathway selective regulation of synaptic plasticity in MSNs controlled by GPCR signaling. SIGNIFICANCE STATEMENT This study identifies a role for a major G-protein regulator in controlling synaptic properties of striatal neurons in a pathway selective fashion.


Introduction
The striatum integrates excitatory inputs from the cortex and thalamus to coordinate voluntary movement, reward valuation, motor learning, and decision-making (Corbit and Balleine, 2003;Balleine et al., 2007;Smith et al., 2011). These functions critically rely on medium spiny neurons (MSNs), which account for ϳ95% of neurons in this region (Kemp and Powell, 1971). In the dorsal striatum, two types of MSNs project to different areas and exert opposite effects on behavior. MSNs forming the direct pathway (D1-MSNs) project primarily to substantia nigra pars reticulata and internal globus pallidus, whereas the indirect pathway (D2-MSNs) projects to the external segment of globus pallidus (Smith et al., 1998). In addition, these two types of MSNs differentially express distinct signaling molecules. The most notable example is the selective expression of D1-type dopamine receptors (D1R) in D1-MSNs and D2-type dopamine receptors (D2R) in D2-MSNs. While D1Rs activate G␣s/olf, D2Rs signal via G␣i/o to activate or inhibit adenylate cyclase activity and cAMP production, respectively (Gerfen et al., 1990;Taussig et al., 1994). Behaviorally, activation of D1-MSNs generally stimulates movements, whereas activation of D2-MSNs inhibits them (Ger-fen et al., 1990;Smith et al., 1998;Kravitz et al., 2012). However, recent models propose a more intricate interplay that relies on synergy and synchronization of D1-MSN and D2-MSN in action selection (Jin et al., 2014;Tecuapetla et al., 2016;Nakamura et al., 2017). In addition, striatal MSNs play essential roles in processing reward signals; and their modulation by addictive drugs, such as cocaine, amphetamines, heroin, and cannabinoids, underlies dependence, craving, and relapse (Nestler, 2001;Kosten and George, 2002). In general, the prevalent models posit that the modulation of D1-MSNs and D2-MSNs activity by neuromodulatory inputs plays a key role in reward seeking, motivation, and motor control.
Most neuromodulators and addictive drugs acting on MSNs exert their effects through GPCRs, which are abundantly expressed in this region. In addition to segregated dopamine D1R and D2R receptors, others include opioids that act on -, -, and ␦-opioid receptors (Winzer-Serhan et al., 2003;Le Merrer et al., 2009), and endocannabinoids (eCBs) that engage Type 1 or Type 2 cannabinoid receptors (CB1 or CB2) (Araque et al., 2017). How these various GPCR inputs are regulated to control synaptic properties of MSNs and striatal-mediated behaviors is poorly understood.
A key role in regulation of GPCR signaling belongs to regulators of G-protein signaling (RGS) proteins, which limit G-protein signaling by accelerating the rate of GTP hydrolysis on G␣ subunits (for review, see Siderovski and Willard, 2005). The long splice isoform of the ninth member of RGS family (RGS9 -2) is selectively enriched in striatal MSNs (Rahman et al., 1999;. In the striatum, RGS9 -2 exists in a macromolecular complex with the Type 5 G-protein ␤ subunit and R7 bindingprotein (Xie and Martemyanov, 2011). The RGS9 -2 complex is well documented to have an essential role in motor control and reward responses (Traynor et al., 2009). Mice lacking RGS9 -2 display augmented responses to the rewarding and motorstimulatory effects of morphine and cocaine Psifogeorgou et al., 2007), as well as deficits in motor learning tasks (Blundell et al., 2008). At the molecular level, RGS9 -2 has been shown to control signaling initiated by D2R and -opioid receptor (Rahman et al., 2003;Cabrera-Vera et al., 2004;Psifogeorgou et al., 2011). However, the downstream effector targets of RGS9 -2 complex are not well defined. One of the prominent targets of -opioid receptor signaling is the NMDAR (Rusin and Randić, 1991;Xie and Lewis, 1997;Rodríguez-Muñoz et al., 2012). Accordingly, RGS9 -2 has been implicated in the regulation of NMDAR function Bouhamdan et al., 2006) and synaptic plasticity (Busse et al., 2014). However, the cellular mechanisms of RGS9 -2 mediated effects on synaptic adaptations, and their relevance in behavior and circuit specificity remains unclear.
In this study, we examined the role of RGS9 -2 in controlling synaptic functions of MSN at the molecular level and within the context of the intact striatal circuitry. We report a mechanism whereby RGS9 -2 acts specifically to adjust synaptic strength of inputs to D2-MSN via a retrograde mechanism that involved the action of the postsynaptic NMDARs to control eCB release and modulation of the presynaptic function via CB1R.

Materials and Methods
Experimental models and subject details. All studies were performed in accordance with the National Institutes of Health guidelines and were granted formal approval by the Institutional Animal Care and Use Committee. Generation and characterization of mouse knock-out strains with targeted disruption of RGS9 (Rgs9 Ϫ/Ϫ ) were described previously (Chen et al., 2000). For electrophysiological recording, Rgs9 Ϫ/Ϫ mice were crossed with D2R-EGFP line so that GFP was selectively expressed in D2-MSN neurons. Mice were housed in groups on a 12 h light-dark cycle with food and water available ad libitum. Both male and female mice between 2 and 5 months of age were used.
Primary striatal neuron cultures. Primary cultures of striatal neurons were prepared as developed by Ivkovic and Ehrlich (1999). Briefly, brains were first dissected from wild-type and RGS9 -2 KO mice (all in C57BL/6J background) on postnatal day 1. Striatum were dissected in ice-cold dissection media [Ca 2ϩ /Mg 2ϩ free Hank's buffered salt solution (CellGro) containing 3 mM HEPES, 1 mM kynurenic acid, 12 mM MgSO 4 , and 5.5 mM glucose, pH 7.4]. After dissection, tissues were treated by 250 g/ml papain (Roche) in dissection media, triturated and plated onto 25-mm-diameter, 0.13-mm-thick glass coverslips (Thermo Fisher Scientific) coated with 5 g/ml poly-L-lysine (Sigma-Aldrich) at a density of 2000 cells/mm 2 . Cultures were maintained in Neurobasal-A medium supplemented with 2% B27, 0.5 mM L-glutamine, and 100 U/ml penicillin/streptomycin (Invitrogen). The cultures were incubated at 37°C in a humidified 5% CO 2 incubator. One-half of media was replaced with the fresh media every 72 h.
Ca 2ϩ imaging experiments. All Ca 2ϩ imaging experiments were performed on striatal neurons at 10 -12 DIV at room temperature (22°C-24°C). Neurons were loaded with the cell-permeant 2 M Indo-1/AM (the acetoxymethyl derivative of Indo-1; Invitrogen) for 40 min in aCSF (137 mM NaCl, 5.4 mM KCl, 0.4 mM KH 2 PO 4 , 0.3 mM Na 2 HPO 4 , 3 mM NaHCO 3 , 0.5 mM MgCl 2 , 0.4 mM MgSO 4 , 1.3 mM CaCl 2 , 20 mM HEPES, 5.6 mM glucose, pH 7.4). The Indo-1/AM was then allowed to fully metabolize to Indo-1 by incubating in fresh aCSF for an additional 20 -25 min before imaging acquisition. Coverslips were mounted in a perfusion chamber and positioned on a movable stage of a CTR6000 inverted microscope (Leica Microsystems) and perfused with aCSF by gravity at ϳ2 ml/min. Indo-1 was excited using 380 nm UV light through a dichroic filter cube and a 63ϫ objective (Leica Microsystems). Emitted light was then passed through a DV2 dual-view image splitter (MAG Biosystems) equipped with bandpass filters of 405 nm (30 nm cutoff) and 485 nm (25 nm cutoff). The 405 and 485 nm images were subsequently acquired using an Em CCD camera (Hamamatsu) at 1 s interval. Ratiometric images of Indo-1 fluorescence were analyzed by defining regions of interest (ROIs) outlining the cell soma using MetaFluor Software version 7.5.6.0 (MDS Analytical Technologies).
Fluorescence changes were converted to [Ca 2ϩ ] i by using the formula [Ca 2ϩ ] i ϭ K d ␤ (R Ϫ R min )/(R max Ϫ R), where R is 405/490 nm fluorescence ratio (Grynkiewicz et al., 1985). The dissociation constant (K d ) for indo-1 was 250 nM and ␤ was the ratio of fluorescence emitted at 490 nm and measured in the absence and presence of Ca 2ϩ . R min , R max , and ␤ were determined by bathing intact cells in 2 M ionomycin in Ca 2ϩ -free buffer (1 mM EGTA) and saturating Ca 2ϩ (5 mM Ca 2ϩ ). Values for R min , R max , and ␤ were 0.25, 2.3, and 3.5, respectively. Background light levels were determined at each wavelength and subtracted before calculating ratios.
Patch-clamp recordings. Coronal slices (300 m) containing striatum (0.2-1.3 mm posterior to bregma) were cut in ice-cold aCSF (in mM as follows: 124 NaCl, 2.8 KCl, 1.25 NaH 2 PO 4 , 2 CaCl 2 , 1.25 MgSO 4 , 26 NaHCO 3 , 10 glucose, pH 7.5, bubbled with 95% O 2 /5% CO 2 ) using a vibrating tissue slicer (VT1200, Leica Microsystems). The slice was divided into two hemispheres along the midline, and each hemisphere was placed into an individual well of a custom slice incubation chamber (Moyer and Brown, 2007) where they remained in oxygenated aCSF at 32°C-36°C until use. During recording, striatal slices were transferred to a submerged recording chamber where they were continuously perfused at 2 ml/min with oxygenated aCSF with picrotoxin (100 M) and maintained at 32°C-36°C. Voltage-clamp whole-cell recordings were obtained with borosilicate glass pipettes (2-5 M⍀) filled with the following solution (in mM: 130 CsMeSO 3 , 20 CsCl, 5 NaCl, 10 HEPES, 0.6 EGTA, 20 TEA, 4 MgATP, 0.3 Na 2 GTP, with a pH of 7.3 and osmolarity of 290 mOsmol). Membrane capacitance (C m ) and series resistance (Rs) were determined by application of a depolarizing pulse of 5 mV for 5 ms. After the measurement of C m and Rs, cells were briefly switched to current recording mode to obtain resting membrane potential. Genetic deletion of RGS9 did not affect resting membrane potential or C m in either D2-MSNs or D1-MSNs (Table 1). Cells were recorded only if initial series resistance Յ20 M⍀ and were excluded from analysis if the Rs changed Ͼ20% during recording period. For mEPSC recording, TTX (1 M) was added to the perfusion solution and spontaneous activities were recorded 3 min after whole-cell mode was obtained for at least 3 min. Evoked EPSCs were recorded using a bipolar stimulating electrode located within the striatum and ϳ200 m away dorsal to the soma. AMPAR-and NMDAR-mediated components were identified according to their distinct activation mechanisms and deactivation kinetics (Myme et al., 2003;Chung et al., 2015). AMPAR-mediated EPSCs were recorded at Ϫ70 mV and measured as the peak response following the stimulus. NMDAR-mediated EPSCs were recorded at 40 mV and measured as the mean current over a 5 ms window, 50 ms after the stimulus. Mean EPSCs were an average of 10 -15 sweeps obtained at 0.1 Hz. For paired-pulse ratio (PPR) recordings, a paired-pulse protocol of two stimuli at an interpulse interval of 50 ms was applied. PPR was defined as the second peak amplitude (P2) divided by the first peak amplitude (P1). Mean PPRs were an average of 10 sweeps acquired at 1/15 Hz. Fatty acid amide hydrolase (FAAH) activator PDP-EA (0.3 mM) was added to the internal solution to facilitate hydrolysis of anandamide (AEA). CB1 blocker AM251 (5 M) and calcium chelator BAPTA-AM (10 M) were added in the external solution. TTX (catalog #1069), PDP-EA (catalog #5210), AM251 (catalog #1117), and BAPTA-AM (catalog #2787) were obtained from Tocris Bioscience and made into appropriate stock solutions in DMSO before use. All other chemicals were purchased from Sigma-Aldrich. All data were collected with an EPC10 amplifier system (HEKA Instruments) and transferred to a personal computer using an ITC-16 digital-to-analog converter (HEKA Instruments). The signals were filtered at 2.9 kHz and digitized at 10 kHz using Patchmaster software (HEKA Instruments). Data were analyzed offline using Patchmaster. Voltages were not corrected for the liquid-liquid junction potential.
Behavioral testing in the open field and rotarod. Locomotor performance was evaluated using male mice in an open field arena (140 cm ϫ 140 cm ϫ 140 cm). All animals were allowed to habituate to the behavioral room and apparatus before the start of study. On the test day, mice were injected with MK-801 (0, 0.3, 1.0, 3.0, and 6.0 mg/kg, i.p.) and placed into the center of the arena; activity was recorded with ANY-Maze automated video tracking software (Stoelting) for 90 min. All animals were given several days of recovery between doses and monitored for normal behavior before each session.
Rotarod performance was tested using a five-station rotarod treadmill with a 1.25 inch rod diameter (IITC Life Sciences). Initially, male mice were trained on the accelerating rotarod (8 -20 rpm) to reach a time of 60 s, with 5 min between intertrial intervals. The following day, the mice were retrained on the rotarod with a cutoff time of 60 s to ensure each mouse had the same initial motor performance. Mice were injected with saline or ketamine (3, 10, 20, or 30 mg/kg; i.p.), and their performance on the rotarod was assessed 15, 30, and 60 min following injection. Each trial ended when a mouse fell off the rod or completed one full revolution on the rod or reached 60 s and the time was scored as the latency to fall.
Statistical analysis. Statistical analysis was performed using Prism 6.0 (GraphPad). Student's t test was used to compare means between two groups, and ANOVA followed by Bonferroni post hoc tests was used to determine significant differences among multiple groups. Repeatedmeasures ANOVA followed by Bonferroni post hoc tests was used to determine the effect of ketamine on rotarod test. All statistical tests were performed two-sided. Differences were considered significant if p Ͻ 0.05. Data are mean Ϯ SEM. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those generally used in comparable studies.

Knock-out of RGS9 -2 suppresses calcium influx through NMDARs in MSNs
We began probing the role of RGS9 -2 in controlling signaling in striatal neurons, by evaluating Ca 2ϩ influx in response to glutamate application using ratiometric calcium imaging in neurons grown in primary culture ( Fig. 1 A, B). Bath application of glutamate (30 M) induced a fast increase in the intracellular Ca 2ϩ concentration in both Rgs9 ϩ/ϩ and Rgs9 Ϫ / Ϫ neurons (Fig. 1C). However, the Ca 2ϩ influx was significantly smaller in Rgs9 Ϫ / Ϫ neurons (ϳ35% suppression) compared with Rgs9 ϩ/ϩ neurons (t (83) ϭ 3.29, p ϭ 0.002; two-tailed independent-samples t test), suggesting that RGS9 -2 facilitates glutamate-activated Ca 2ϩ entry ( Fig. 1C,D). In response to glutamate application, Ca 2ϩ may enter the cell through NMDARs or voltage-gated calcium channels (VGCCs). To dissect which route Ca 2ϩ is affected by RGS9 -2 loss, we sequentially blocked Ca 2ϩ entry via NMDAR by bath application of MK-801 (10 M) or via L-type VGCC by nimodipine (10 M) (Fig. 1B). We found that MK-801 significantly suppressed the Ca 2ϩ influx (74% in WT and 61% in Rgs9 Ϫ / Ϫ neurons, respectively) in response to glutamate application (t (78) ϭ 6.80, p Ͻ 0.001 and t (43) ϭ 4.12, p Ͻ 0.001 for Rgs9 ϩ/ϩ and Rgs9 Ϫ / Ϫ neurons, respectively) ( Fig. 1C,D). Addition of nimodipine further suppressed glutamate-induced Ca 2ϩ entry in both genotypes compared with MK-801 alone (t (41) ϭ 4.27, p Ͻ 0.001; and t (24) ϭ 2.82, p ϭ 0.010 for Rgs9 ϩ/ϩ and Rgs9 Ϫ / Ϫ , respectively). Notably, the Ca 2ϩ influx in MSNs in response to glutamate was comparable between the genotypes in the presence of MK-801 alone or in the presence of the mixture of MK-801 and nimodipine ( p Ͼ 0.6 for both experiments), suggesting that lower calcium influx in response to glutamate application seen in Rgs9 Ϫ / Ϫ neurons is mediated through suppression of NMDAR activity, with negligible changes in entry via VGCC.
Because RGS9 -2 is involved in controlling cAMP levels by modulating adenylyl cyclase 5 (AC5) activity in MSN (Xie et al., 2012), we next investigated the role of cAMP in suppression of glutamate-induced Ca 2ϩ influx seen in Rgs9 Ϫ / Ϫ neurons. Application of AC activator forskolin did not significantly change glutamate-induced Ca 2ϩ influx in wild-type ( p ϭ 0.38) but significantly enhanced Ca 2ϩ influx in Rgs9 Ϫ / Ϫ neurons (t (48) ϭ 3.5, p Ͻ 0.01), abolishing the difference between the genotypes ( Fig.  1D; p ϭ 0.97). These data suggest that the inhibition of glutamate-induced Ca 2ϩ influx in the absence of RGS9 -2 involves the reduction in cAMP levels.
To confirm that the observed glutamate-induced Ca 2ϩ influx was mediated by NMDARs, we measured intracellular Ca 2ϩ con- Data are mean Ϯ SE (no. of cells). I AMPA , AMPA receptor-mediated current; I NMDA , NMDA receptor-mediated current; dMSN, direct pathway MSN; iMSN, indirect pathway MSN; RMP, resting membrane potential. iMSNs were identified by their presence of EGFP fluorescence under microscope. dMSNs were identified by their absence of EGFP fluorescence under microscope. RMPs were measured immediately after whole-cell mode was established by switching to current mode. EPSCs were evoked by a stimulating electrode located ϳ200 m away dorsal to the soma within the striatum. I AMPA was measured at the peak current when cells were held at Ϫ70 mV. I NMDA was measured as the average current between 50 and 55 ms following the stimulation when the cells were held at 40 mV. *t (33) ϭ 2.8, p ϭ 0.008, statistically different between iMSN Rgs9 ϩ/ϩ and iMSN Rgs9 Ϫ/Ϫ . centration in response to application of selective NMDAR agonist NMDA (100 M) in the absence of Mg 2ϩ . We found that, similarly to glutamate application, NMDA resulted in a robust increase in Ca 2ϩ influx in MSNs (Fig. 1E). One-way ANOVA revealed a significant difference between the genotypes ( Fig. 1F; F (2,45) ϭ 7.44, p ϭ 0.007). Bonferroni post hoc tests revealed that the Ca 2ϩ influx was also significantly suppressed (40%) in Rgs9 Ϫ/Ϫ neurons compared with Rgs9 ϩ/ϩ neurons ( p Ͻ 0.01). Again, forskolin completely rescued reduction in Ca 2ϩ influx deficits in Rgs9 Ϫ / Ϫ neurons ( p Ͻ 0.01). Collectively, these data suggest that RGS9 -2 positively regulates mobilization of NMDAR-mediated Ca 2ϩ influx in response to glutamate application in a cAMP-dependent manner.

Mice lacking RGS9 -2 are hypersensitive to inhibition of NMDAR by MK-801 and ketamine
We next probed the behavioral relevance of the observed changes in NMDAR-mediated function associated with RGS9 -2 loss. Given the role of the striatum in motor control, we first assessed the effect of NMDAR-selective blocker MK-801 on locomotor activity in an open field ( Fig. 2A), which is commonly used to assess striatum-dependent motor control. Consistent with previ- Figure 1. RGS9 -2 deletion in primary striatal neurons inhibits calcium influx through NMDARs. A, Fluorescent image of a representative striatal neuron loaded with calcium indicator Indo-1 and excited with 380 nm UV light. Red circle represents the ROI (soma) used to measure intracellular calcium concentration. B, Schematic of the assay design to study the role RGS9 -2 on calcium influx through NMDARs. MK-801 was used to block NMDARs, and nimodipine was used to block VGCCs. Forskolin (FSK) was used to activate adenylyl cyclase (AC) to increase cAMP. C, D, Representative traces and summarized bar graphs show that glutamate (30 M) elicited a dramatic increase in intracellular calcium concentration, which was significantly suppressed in Rgs9 Ϫ / Ϫ neurons (t (83) ϭ 3.3, p ϭ 0.002). MK-801 or a mixture of MK-801 ϩ nimodipine significantly reduced the calcium influx evoked by glutamate, but no difference between genotypes was observed in the presence of the blockers. E, F, Representative traces and summarized bar graphs show NMDA (100 M)-induced calcium influx in MSNs. One-way ANOVA (F (2,45) ϭ 7.4, p ϭ 0.007) followed by Bonferroni post hoc tests revealed that NMDA-evoked Ca 2ϩ influx was significantly attenuated in Rgs9 Ϫ / Ϫ neurons ( p Ͻ 0.01), whereas FSK rescued the effect of Rgs9 Ϫ / Ϫ . **p Ͻ 0.01, significantly different between groups.
Once the criteria had been reached, we investigated the effect of ketamine on their motor performance. One day following training, mice were subjected to a test on the rotarod before and 15, 30, and 60 min after either saline or ketamine injection (Fig. 2D). At the first time point following either saline or ketamine injection, we found that Rgs9 Ϫ/Ϫ mice displayed a greater sensitivity to ketamine on the rotarod as indicated by the reduced latency time across several doses (Fig. 2E). Two-way ANOVA revealed a main effect of ketamine (F (4,40) ϭ 46.2, p ϭ 0.010), genotype (F (1,10) ϭ 19.9, p ϭ 0.001), and a significant interaction of dose and genotype (F (4,40) ϭ 3.8, p ϭ 0.010). Further post hoc analysis with Bonferroni correction revealed that the latency to fall from the rotarod was significantly reduced in Rgs9 Ϫ/Ϫ mice at 3 mg/kg ( p Ͻ 0.05) and 10 mg/kg ( p Ͻ 0.001) but were ineffective in Rgs9 ϩ/ϩ mice. There were no significant differences in performance between genotypes in control experiments involving saline (0) or at higher ketamine concentrations (20 and 30 mg/kg) that induce anesthesia (Fig. 2E). At a subthreshold dose of 10 mg/kg, only Rgs9 Ϫ/Ϫ mice displayed reduced motor performance at 15 min following ketamine injection, and this effect lasted for Ͼ60 min ( Fig. 2F; main effect of genotype as revealed by repeated-measures ANOVA followed by Bonferroni post hoc test: F (1,30) ϭ 7.8, p ϭ 0.019). At a higher dose of 20 mg/kg, Rgs9 ϩ/ϩ mice showed a reduced motor performance 15 min after injection but were able to recover (30 min after injection), whereas Rgs9 Ϫ/Ϫ mice were unable recover as evident from the reduced performance at 60 min ( Fig. 2G; main effect of time as revealed by repeated-measures ANOVA followed by Bonferroni post hoc test; F (3,30) ϭ 10.9, p Ͻ 0.001). These data indicate that elimination of RGS9 -2 in mice substantially augments their behavioral sensitivity to NMDAR blockade, consistent with the reduction in the NMDAR function observed in cultured neurons and brain slices.

Deletion of RGS9 -2 suppresses NMDAR current and enhances AMPA/NMDA ratio selectively in D2-MSNs
To test the relevance of the observed NMDAR deficits associated with RGS9 -2 loss in an intact neuronal circuit, we further used whole-cell patch-clamp recordings of MSNs in brain slices directly measuring the NMDAR currents. We reasoned that high variability of Ca 2ϩ responses in cultured neurons ( Fig. 1 D, F ) could be related to differences in the behavior from the two distinct populations of MSNs, D1-MSNs and D2-MSNs, which have distinct electrophysiological properties and regulation (Planert et al., 2013). Therefore, in the rest of our studies, we distinguished between MSN populations by crossing Rgs9 Ϫ / Ϫ mice with the D2R-EGFP line, which expresses GFP selectively in D2-MSN neurons and recording from both fluorescent (D2-MSN) and nonfluorescent (D1-MSNs) neurons containing or lacking RGS9 -2 expression (Fig. 3 A, B). NMDAR current in MSNs was elicited by direct bath application of NMDA (30 M), and the cells were held at Ϫ70 mV in the presence of GABAergic blocker picrotoxin. This manipulation induced a large inward current in both Rgs9 ϩ/ϩ D2-MSNs and D1-MSNs. The amplitude of this current was significantly attenuated in Rgs9 Ϫ/Ϫ D2-MSNs ( Fig.  3C; t (23) ϭ 2.17, p ϭ 0.041). In contrast, we observed no significant differences in the amplitude of NMDA-elicited current between genotypes in D1-MSNs ( Fig. 3D; t (19) ϭ 0.30, p ϭ 0.77). These data reinforce our observations that loss of RGS9 -2 leads to inhibition of NMDAR activity in D2-MSNs.
We next examined the contribution of RGS9 -2 to synaptic events mediated by NMDAR. In these experiments, glutamatergic inputs from the cortex onto MSNs were electrically stimulated by concentric electrode placed ϳ200 m dorsal to the cell body evoking EPSCs while recording from genetically identified D2-MSN and D1-MSN neurons in the presence of picrotoxin (Fig.  4A). Contributions of AMPAR and NMDAR to EPSCs were isolated by holding cells at Ϫ70 mV and 40 mV, respectively (Fig.  4 B, C). Consistent with reduction in chemically evoked NMDAR current in Rgs9 Ϫ / Ϫ mice, we found that the NMDAR component was significantly reduced (t (33) ϭ 2.8, p ϭ 0.008) in Rgs9 Ϫ/Ϫ neurons, whereas the AMPAR component was comparable (t (32) ϭ 1.1, p ϭ 0.278) between genotypes in D2-MSNs (Table 1). Furthermore, deletion of RGS9 -2 significantly increased the AMPA/NMDA ratio in D2-MSNs ( Fig. 4B; t (33) ϭ 2.24, p ϭ 0.032). In contrast, no significant differences were observed between genotypes in any of these measurements in D1-MSNs ( Fig.  4C; Table 1). Together, these data indicate that the loss of RGS9 -2 inhibits NMDAR function selectively in D2-MSN neurons, which results in greater contribution of AMPAR relative to NMDAR to synaptic responses.

Loss of postsynaptic RGS9 -2 selectively facilitates presynaptic glutamate release onto D2-MSNs
To further examine the mechanisms underlying RGS9 -2mediated modulation of synaptic properties, we recorded mEP-SCs in dorsal striatal MSNs in the presence of TTX and picrotoxin. As shown in the representative traces in Figure 5A and summarized in Figure 5C, we found that deletion of RGS9 -2 had no effect on the amplitude of mEPSC (t (18) ϭ 0.64, p ϭ 0.530) in D2-MSNs. Because mEPSCs are predominantly, if not exclusively, contributed by AMPAR currents, this further confirms no influence of RGS9 -2 on the synaptic AMPAR function and demonstrates its exclusivity of NMDAR modulation. Surprisingly, we found that D2-MSN neurons from Rgs9 Ϫ/Ϫ mice showed significantly increased frequency of mEPSC compared with wild-type neurons ( Fig. 5D; t (20) ϭ 3.78, p ϭ 0.0012), suggesting additional presynaptic adaptations. This effect was again specific to the D2-MSN population as neither frequency nor amplitude of mEPSC was significantly different between genotypes in D1-MSNs (Fig.  5 B , E,F; t (20) ϭ 0.23, p ϭ 0.823, and t (20) ϭ 1.39, p ϭ 0.180 for frequency and amplitude, respectively).
To determine the mechanisms by which the loss of RGS9 -2 in postsynaptic D2-MSNs regulates presynaptic release prob-ability, we tested a hypothesis about the involvement of retrograde eCB signaling, which plays a prominent role in retrograde feedback in striatal synapses (Gerdeman and Lovinger, 2001). Our strategy relied on blocking different steps in the retrograde eCB signaling in D2-MSN and D1-MSN neurons from wild-type and Rgs9 Ϫ / Ϫ mice: blocking CB1R with AM251, chelating intracellular calcium with BAPTA-AM, and facilitating the breakdown of AEA with a FAAH activator PDP-EA (Fig. 6A).
These data suggest that the increase in release probability observed in D2-MSN following RGS9 -2 deletion was mediated by reduced activation of the eCB signaling system, which might be associated with the reduction in NMDAR activation.

Discussion
The current study provides the first evidence that RGS9 -2 specifically regulates NMDAR function and synaptic transmission in D2-MSNs in the dorsal striatum. Both calcium imaging from primary culture and electrophysiological recording from brain slices reveal that NMDAR function was significantly attenuated following RGS9 -2 deletion. These observations are consistent with earlier reports that noticed similar dysregulation of NMDAR function . The experiments performed to dissect the consequences of this alteration led to propose a model where NMDAR hypofunction increased AMPAR/ NMDAR ratio and suppressed the eCB signaling system, which led to increased presynaptic glutamate release (Fig. 7). In this model, RGS9 -2 suppresses signaling via Gi/o-coupled GPCRs, which in turn inhibit adenylyl cyclase to control cAMP production. Consequently, deletion of RGS9 -2 would increase inhibitory tone suppressing cAMP production and reducing the activity of its key effector PKA, well known for its stimulatory effects on NMDAR mediated by phosphorylation (Raman et al., 1996;Crump et al., 2001;Scott et al., 2001;Skeberdis et al., 2006). This would reduce Ca 2ϩ permeability of NMDARs suppressing eCB release and engagement of presynaptic CB1R relieving their inhibitory effects on neurotransmitter release. At the behavioral level, suppression of NMDAR function made mice lacking RGS9 -2 hypersensitive to the dissociative effects of ketamine and MK-801 administration. Collectively, these data clarify cellular and molecular mechanisms by which RGS9 -2 provides pathway-selective control of psychomotor functions mediated by the dorsal striatum.

RGS9 -2 specifically modulates synaptic properties of D2R-expressing MSNs
Although highly restricted to striatum, RGS9 -2 is reported to be equally expressed in D1-MSNs and D2-MSNs (Kovoor et al., 2005;Tecuapetla et al., 2016). Interestingly, in the current study, we observed that effects induced by RGS9 -2 deletion were specific to D2-MSNs, including changes in NMDAR current, eCB signaling, AMPA/NMDA ratio, mEPSC frequency, and PPRs. We think that the pathway selectivity of RGS9 -2 effects may be explained by specificity of RGS9 -2 for the G␣i/o protein (Hooks et al., 2003;Xie et al., 2012;Masuho et al., 2013), which mediates the effects of D2R in the D2-MSNs. This explanation is supported by the observations that RGS9 -2 when ectopically coexpressed in transfected cells interacts with D2Rs, but not D1Rs , and that NMDARs interact differently with the two subtypes of dopamine receptors such that activation of D1Rs potentiates, whereas activation of D2Rs attenuates, NMDAR responses (Cepeda et al., 1993(Cepeda et al., , 2008. Furthermore, genetic deletion of RGS9 -2 causes locomotor deficits (e.g., autonomic involuntary movements) that are related to D2R, but not D1R, functions . Conversely, overexpression of RGS9 -2 in rat nucleus accumbens reduced locomotor responses to cocaine and to D2R, but not to D1R, receptor agonists (Rahman et al., 2003). The function of RGS9 -2 in D1-
NMDAR, but not AMPAR, current was attenuated in MSNs in Rgs9 knock-out mice . These data are also consistent with the increase in AMPAR/NMDAR ratio in D2-MSNs lacking RGS9 -2 and suggest that the behavioral deficits observed in Rgs9 knock-out animals might be at least in part related to the hypofunction of NMDAR and ensuing synaptic adaptations.
At the behavioral level, NMDAR function is critical for striatum-dependent motor control and learning. For example, striatum-specific ablation of NMDAR increases locomotor activity (Mohn et al., 1999;Ohtsuka et al., 2008) and significantly impairs behavioral performance in rotarod test (Dang et al., 2006). This is reminiscent of behavioral deficits in the same behavioral test following RGS9 deletion (Blundell et al., 2008). The observation that mice lacking RGS9 were supersensitive to ketamine reported in this study further reinforces the role of RGS9 -2 in NMDAR modulation at the behavioral level.
At the molecular level, the hypofunction of NMDAR in RGS9 -2 deficient mice is likely driven by alterations in NMDARs function. RGS9 -2 deletion is expected to enhance the activity of Gi/o-linked GPCRs prominent in the region, such as D2R or -opioid receptor. The ensuing suppression of PKA activity would reduce its ability to phosphorylate NMDARs (Rusin and Randić, 1991;Martin et al., 1997;Xie and Lewis, 1997;Rodríguez-Muñoz et al., 2012). Phosphorylation of NMDARs by PKA has been shown to increase the surface expression, gating, and Ca 2ϩ permeability of the receptors (Scott et al., 2001;Chen and Roche, 2007;Aman et al., 2014). Conversely, blocking PKA dramatically inhibits NMDAR function (Skeberdis et al., 2006). Thus, loss of RGS9 -2 likely leads to changes in NMDAR gating and/or targeting to plasma membrane. Furthermore, RGS9 -2 may also exert its facilitatory effects on NMDAR activity by suppressing its calcium-dependent inactivation through interacting with ␣-actinin-2 (Bouhamdan et al., 2006). The functional nature of the effects observed is also supported by no significant changes in the expression of NMDAR transcript in the striatum upon RGS9 ablation (Busse et al., 2014).

RGS9 -2 regulates eCB system
In the course of this study, we made a surprising observation that postsynaptic RGS9 -2 expressed in MSN modulates mEPSC frequency and PPR of excitatory inputs to D2-MSNs by a presynaptic mechanism. These results are consistent with a previous study where deletion of RGS9 -2 was noted to increase spontaneous EPSC accompanied by impaired induction of LTD (Busse et al., 2014). Our pharmacological dissection indicates that changes in presynaptic release probability are likely due to an inhibition of eCB signaling caused by RGS9 -2 deletion. Accordingly, blockade of presynaptic CB1R or inhibition of eCB production mimicked the effect of RGS9 -2 ablation. Previous studies showed that eCBs are released from MSNs triggered by calcium influx upon membrane depolarization (Di Marzo et al., 1994;Alger, 2004), and this process is potentiated when cAMP levels rise (Cadas et al., 1996). After being released, eCBs act as retrograde messengers that reg- Figure 7. Proposed model for RGS9 -2 regulation of NMDAR function and synaptic signaling in D2-MSNs. Within MSNs, RGS9 -2 inactivates Gi-coupled GPCR, which inhibits the production of cAMP, which in turn plays a key role in positive modulation of NMDAR function. The change in Ca 2ϩ influx through NMDARs then affects the production of AEA, which retrogradely inhibits presynaptic neurotransmitter release. ulate synaptic transmission through presynaptic CB1Rs and suppress neurotransmitter release (for review, see Heifets and Castillo, 2009). eCB signaling is required for synaptic plasticity in striatum, especially LTD (Gerdeman and Lovinger, 2001;Gerdeman et al., 2002). Genetic deletion of CB1R significantly impaired skill learning on rotarod in adult mice (Bilkei-Gorzo et al., 2005), indicating that the eCB signaling system is also involved in motor control. Together, these observations suggest that RGS9 -2 may ultimately regulate motor learning through modulating GPCR-NMDAR-eCB signaling axis.