|Home | About | Journals | Submit | Contact Us | Français|
Dopamine (DA)-dependent corticostriatal plasticity is thought to underlie incremental procedural learning. A primary effector of striatal DA signaling is cAMP, yet its role in corticostriatal plasticity and striatum-dependent learning remains unclear. Here, we show that genetic deletion of a striatum-enriched isoform of adenylyl cyclase, AC5 (AC5KO), impairs two forms of striatum-dependent learning and corticostriatal synaptic plasticity. AC5KO mice were severely impaired in acquisition of a response strategy in the cross maze, a striatum dependent task requiring a correct body turn to find a goal arm. In addition, AC5KO mice were impaired in acquisition of a motor skill, as assessed by the accelerated rotarod. Slice electrophysiology revealed a deficit in corticostriatal LTD following high frequency stimulation of tissue from AC5KO mice. LTD was rescued by activation of either presynaptic cannabinoid type 1 (CB1) receptors, or postsynaptic metabotropic glutamate receptors (mGluRs), suggesting a postsynaptic role of AC5-cAMP, upstream of endocannabinoid release. In striatopallidal projecting medium spiny neurons (MSNs), DA D2 receptors are negatively coupled to cAMP production and activation of these receptors is required for endocannabinoid release and corticostriatal LTD. Recordings from striatopallidal neurons indicated that this is mediated by AC5, as co-activation of D2 and mGluR receptors could induce LTD in WT, but not in AC5KO neurons. To further examine the role of cAMP in corticostriatal plasticity, we elevated cAMP in striatal neurons of wild-type mice via the recording electrode. Under these conditions corticostriatal LTD was eliminated. Together, these data suggest an AC5-cAMP-endocannabinoid-CB1 signaling pathway in corticostriatal plasticity and striatum-dependent learning.
In humans, the dorsal striatum plays a critical role in procedural learning, the memory for motor skills and stimulus-response habits, that is dissociable from hippocampus dependent declarative learning, the memory for facts and events (Packard et al., 1989; Packard and White, 1991; Packard and McGaugh, 1996; Packard and Knowlton, 2002; Schroeder et al., 2002; Squire, 2004; Yin et al., 2004; Yin et al., 2005). In rodents, lesion studies have indicated that striatum-dependent learning of stimulus-response habits is dissociable from hippocampus dependent learning of spatial information (Packard and McGaugh, 1996; Packard and Knowlton, 2002; Yin and Knowlton, 2006). This dissociation has been documented through use of a cross maze, where inactivation of the dorsal striatum impairs the animal’s ability to learn to make the correct body turn at a choice point to find the goal arm (Packard and McGaugh, 1996). The dorsal striatum receives dense glutamatergic innervation from cortical areas, as well a dopaminergic input from midbrain nuclei. The interplay between dopamine (DA) and glutamate input to medium spiny neurons (MSNs) of the dorsal striatum has provided the basis for models of reinforcement learning that suggest DA dependent plasticity at corticostriatal synapses is necessary for striatum-dependent learning (Calabresi et al., 1992; Schultz et al., 1997; Reynolds et al., 2001; Reynolds and Wickens, 2002). Yet, few studies have addressed the role of DA-dependent intracellular signaling pathways that underlie this form of striatum dependent-learning, and how they contribute to DA dependent corticostriatal plasticity.
In brain slice recordings, the predominant form of synaptic plasticity seen in the dorsolateral striatum is long term depression (LTD), which requires DA D2 receptor activation, increases in intracellular calcium, and retrograde release of endocannabinoids that inhibit glutamate release via presynaptic CB1 receptors (Calabresi et al., 1994; Gerdeman et al., 2002; Calabresi et al., 2007; Kreitzer and Malenka, 2007). Although modulation of intracellular cAMP content is a direct downstream effect of DA receptor activation (Neve et al., 2004), the role of cAMP in corticostriatal plasticity has remained largely uninvestigated. Two downstream targets of cAMP dependent protein kinase (PKA), cAMP response element binding protein (CREB) and the dopamine and cAMP-regulated phosphoprotein 32 kDa (DARPP-32) have been implicated in corticostriatal LTD (Calabresi et al., 2000; Pittenger et al., 2006), however, these proteins are also regulated by non-cAMP dependent signaling pathways (Kang et al., 2001; Svenningsson et al., 2004), leaving no direct evidence for a contribution of cAMP to corticostriatal LTD.
The highly conserved cAMP second messenger system has been implicated in synaptic plasticity and learning in many different organisms (Kandel, 2001). The enzyme that catalyzes cAMP production, adenylyl cyclase (AC), has at least nine membrane-bound isoforms that have different expression patterns and regulatory properties (Cooper et al., 1998; Hanoune and Defer, 2001; Wang and Storm, 2003). CaCaM-dependent AC isoforms have been extensively studied and are essential for induction of long term potentiation (LTP) in the hippocampus, as they couple glutamate mediated increases in intracellular calcium to cAMP accumulation (Xia et al., 1993; Wu et al., 1995; Matsuoka et al., 1997; Wong et al., 1999; Wang and Storm, 2003; Nicol et al., 2005). The dorsal striatum expresses high levels of AC5 (Matsuoka et al., 1997), a primary downstream effector of DA receptor signaling in the striatum. Mice that lack AC5 expression (AC5KO) show severely impaired striatal D1 and D2 receptor modulation of cAMP signaling (Iwamoto et al., 2003). In addition, recent studies have indicated that AC5KO mice are unable to predict reward delivery in a DA-dependent Pavlovian conditioning task (Kheirbek et al., 2008). Thus, the AC5KO mouse provides an opportunity to test the contribution of cAMP to corticostriatal plasticity and dorsal striatum-dependent learning. Here, we demonstrate a critical role for AC5 in a striatum-dependent learning task, response learning in the cross maze. This deficit in striatum dependent learning was confirmed in the accelerated rotarod, a motor skill learning task previously correlated with plasticity in the dorsal striatum. In addition, we reveal a role for AC5 and cAMP signaling in corticostriatal LTD by regulating D2 mediated retrograde release of endocannabinoids.
AC5KO mice were generated as described (Iwamoto et al., 2003) and backcrossed to c57BL/6 for nine generations. All mice tested were 8–12 weeks of age and behavioral experiments were conducted during the light phase. WT littermates were used as controls and independent groups were used for each experiment. Drd2-EGFP BAC transgenic mice were obtained from the Mutant Mouse Regional Resource Center. Drd2-EGFP mice were backcrossed to AC5KO background to generate AC5KO/Drd2-EGFP and WT/Drd2-EGFP littermate controls for electrophysiological experiments. All experiments were approved by the Institutional Animal Care and Use Committee of the University of Chicago.
Mice were decapitated, and brains were fresh frozen on dry ice. 20 μm coronal sections were cut and thaw mounted on slides, and stored at −80 degrees until hybridization. Sections were fixed in 4% paraformaldehyde for 10 min, digested with protinase K, followed by acetylation for 10 min and pre-hybridization for 2 hours at 65 °C in a buffer containing 50% formamide, 20x SSC, 5X Denhardt’s solution, 250 μg/ml tRNA and 500 μg/ml fish sperm DNA. For radiolabeled probe, 106 CPM 33P-rUTP probe per 50 μl of hybridization buffer was used for overnight incubation at 65 °C. Signal was visualized after 48-hour development on Kodak BioMax MR film. For DIG labeled probes, sections were incubated with 400 ng/ml probes overnight at 65 °C. Slides were then washed, and incubated in 1:5000 dilution of anti-DIG-AP antibody (Roche) for two hours, then visualized using NBT/BCIP staining for alkaline phosphatase. For double labeling in situ, sections were hybridized with DIG and biotin labeled mRNA probes simultaneously. After washing, the sections were incubated in 1:5000 dilution anti-biotin-AP antibody, followed by visualization with FastRed (Sigma). Slides were photographed, then washed in 0.1M glycine (pH 2.2) and incubated in anti-DIG-AP, followed by development in NBT/BCIP solution.
AC1, AC5, and ENK probes were isolated from a mouse brain cDNA library using the following primer pairs. AC5: 5′CTGGAGGAGCGTCGAGGAAA3′ and 5′GGTCCTGGGAGTTGGTGTGC3′; ENK: 5′GGATGCAGATGAGGGAGACA3′ and 5′AAACCCGTGACCCCAACTGC3′. The cDNA fragment was cloned into PCRII-TOPO vector (Invitrogen), linearized and antisense probe was transcribed using the appropriate RNA polymerase.
Experiments were conducted in a water tank (white walls, 100cm in diameter), with white plexiglass walls inserted in the maze to make alleyways. Water was at room temperature, and white non-toxic paint (FunStuff, Toronto, CA) was added to obscure the hidden platform. A black curtain surrounded the maze to reduce spatial cues. The walls of the maze were rotated every ten trials to further reduce the contribution of spatial cues. A trial was started by placing mouse in the start arm facing the wall of the tank. Once the mouse made a right or left choice, a divider closed off the arm and the mouse was either allowed to stand on the platform for 15 s (correct choice) or explore the arm of the maze without the platform for 15 s (incorrect choice). Mice were then removed from the maze and placed under a heat lamp for 30 seconds before beginning the next trial. Trials were started randomly from either the north or the south end of the maze with the goal arm always to the right of the start arm during initial acquisition and to the left of the start arm during reversal learning. Correct or incorrect choices were recorded.
Experiments were conducted as previously described (Vorhees and Williams, 2006). Experiments were conducted in a water tank (white walls, 100 cm in diameter), with white plexiglass walls inserted in the maze to make alleyways. Water was at room temperature, and white non-toxic paint (FunStuff, Toronto, CA) was added to obscure the hidden platform. Around the tank, white paper cutouts of stars and arrows were placed on the black surrounding curtain to serve as spatial cues. Mice were first trained for two days, three trials a day with a visible platform (8 cm diameter) for habituation purposes and to determine any swimming deficits. Following habituation, training trials began. A 1 cm submerged platform (8 cm diameter) was placed in one quadrant. During training, mice were started from random start positions, and latency to reach hidden platform was measured. Mice were given 3 training trials a day for seven consecutive days. All measurements were made using a live camera and EthoVision (Noldus) software. Twenty-four hours after the last days of training mice were given a one-minute probe test (Vorhees and Williams, 2006) where the platform was removed, and time spent in the target quadrant was measured by the Ethovision software.
Mice were placed on a rotarod apparatus (Columbus Instruments) accelerating from 4–40 rpm in 5 min. Trials began by placing mouse on the rod and beginning rotation. Each trial ended when the mouse fell off the rod, and latency was recorded. Mice were tested for 4 trials a day (1 min inter-trial interval) for 3 consecutive days.
Each mouse was placed in an acrylic open field chamber 40 cm long × 40 cm wide × 37 cm high (Med Associates, St. Albans, VT). The floor of the chamber is white, the sides transparent. Each chamber was surrounded by black drop cloth obscuring views beyond the chamber. Illumination of open field was set to 16 lux. No background noise was provided. Infrared beams recorded the animal’s location and path (locomotor activity) and recorded distance traveled. Data were collected in 5-min bins during 60-min trials.
Coronal brain slices (250 μM thick) were obtained from 6–10 week old mice. The mice were anesthetized with isoflurane (Abbott Laboratories, North Chicago, IL) and rapidly decapitated. Their brains were transferred to cold, sucrose-artificial cerebrospinal fluid (ACSF) containing the following (in mM): 200 sucrose, 25 NaHCO3, 20 glucose, 10 ascorbic acid, 2.5 KCl, 2.5 CaCl2, 1 MgCl2 and 1 NaH2PO4, pH 7.4, saturated with 95% O2 and 5% CO2 (Fagen et al., 2007). Striatal brain slices were made on a vibratome (VT100S, Leica Instruments, Germany). Slices were incubated for at least 1 hour in a bath circulated at 20 ml/min with normal, 32°C ACSF containing (in mM): 125 NaCl, 25 NaHCO3, 20 glucose, 2.5 KCl, 2.5 CaCl2, 1 MgCl2, 1 NaHCO3, and 1 ascorbic acid, pH 7.4, saturated with 95% O2 and 5% CO2 (Fagen et al., 2007). During recording, slices were superfused (2 ml min−1) with this same ACSF at ~34°C, but without the ascorbic acid and with 10 μM (−) bicuculline methiodide to block GABAA receptor-mediated synaptic responses. 10 μM DNQX (6,7-Dinitroquinoxaline-2,3-dione) was added to the ACSF for experiments testing intrinsic membrane excitability. 1 μM tetrodotoxin was included in the ACSF for the recording of mEPSCs.
Whole-cell voltage-clamp recordings from MSNs located in the dorsal lateral striatum were obtained under visual control on a differential interference contrast, upright microscope with infrared illumination (Axioskop, Carl Zeiss, Thornwood, NY). Recordings were obtained using 2.5 to 6 MΩ resistance pipettes made from borosilicate glass capillary tubing (G150-4, Warner Instruments, Inc, Hamden, CT) that was pulled on a Flaming/Brown micropipette puller (Model P-97, Sutter Instrument Company, Novato, CA). To test membrane excitability and mEPSCs, recording pipettes were filled with an internal solution containing the following (in mM): 154 K-gluconate, 1 KCl, 1 EGTA, 10 HEPES, 10 glucose, 5 ATP, and 0.1 GTP and brought to a pH of 7.4 with KOH (Mansvelder and McGehee, 2000). To test stimulated synaptic currents, recording pipettes were filled with an internal solution containing the following (in mM): 120 CsMeSO3, 10 TEA-Cl, 10 HEPES, 10 sucrose, 5 NaCl, 5 QX-314, 4 ATP-Mg2, 1.1 EGTA, 0.3 GTP-Na2 and brought to a pH of 7.25 with CsOH (Wang et al., 2006). Currents were measured using a Multiclamp 700A amplifier (2 kHz low-pass Bessel filter) with a DigiData 1322A interface (10 kHz digitization) and pCLAMP 9.2 software (Molecular Devices Corp.). MSNs in the dorsal lateral striatum, identified by their morphology and hyperpolarized resting membrane potential, were voltage clamped at −80 mV, unless noted otherwise. A bipolar tungsten electrode with a 500 μM tip separation (FHC, Bowdoinham, ME) was placed in the white matter 200 to 300 μm dorsal to the recording electrode. Test stimuli (2 pulses 25 ms apart) were delivered through a Master-8 stimulator (AMPI, Jerusalem, Israel) every 20 s. Stimulus intensity ranged from 0.1 to 1 mA and pulse duration ranged from 50 to 400 μs. Series resistance in the recorded cells was monitored online from a 5 mV, 20 ms depolarizing pulse delivered 60 ms after the second of the paired stimuli. The series resistance was 60% compensated and cells were discarded if this varied >20% or increased to >30 MΩ during the course of the experiment. After evoked EPSC amplitudes remained relatively stable for 10 min, high frequency stimulation (HFS) was delivered in some experiments, which consisted of four 1 s 100 Hz trains delivered every 10 s, during which the postsynaptic cell was depolarized to 0 mV. EPSC amplitudes are expressed as a percentage of the baseline values, mean ± SEM. For prevalence data, plasticity was determined by paired t-tests comparing the 10 min baseline period with the last 10 min of recording.
Carbon fiber microelectrodes were fabricated by threading a 7 μm diameter carbon fiber (Fortafil Fibers, Knoxville, TN) through borosilicate glass capillary tubing (G150-4, Warner Instruments, Inc, Hamden, CT) that was then pulled on a Flaming/Brown micropipette puller (Model P-97, Sutter Instrument Company, Novato, CA). The carbon fiber was subsequently cut to restrict the length of the exposed fiber in the slice to ~50 μm. Recording electrodes were placed in the dorsal striatum ~250 μm from a bipolar tungsten electrode with a 500 μM tip separation (FHC, Bowdoinham, ME). Dopamine release was evoked by electrical stimulation (400 μA, 1 msec) delivered every 2 minutes with an Iso-Flex stimulus isolator triggered by a Master-8 pulse generator (A.M.P.I., Jerusalem, Israel). Currents were measured using a Multiclamp 700A amplifier (10 kHz low-pass Bessel filter) with a DigiData 1322A interface (50 kHz digitization) and pCLAMP 9.2 software (Molecular Devices Corp.). For fast-scan cyclic voltammetry (FSCV), the electrode voltage was ramped from the baseline voltage of −400 mV to +1000 mV and then back again at 300 V/sec versus Ag/AgCl at 100 msec intervals. Background-subtracted FSCV was used to visualize oxidation and reduction currents, allowing for identification of the oxidized substance and for calibration with 5 μM dopamine at the end of each experiment.
We first confirmed the enrichment of AC5 mRNA in the adult mouse striatum using in situ hybridization with 33P labeled probes (Fig. 1A–B). AC5 mRNA expression was very high in the dorsal striatum, nucleus accumbens and olfactory tubercle (Fig. 1A). AC5 message was detected at low levels in hippocampus and cortex (Fig. 1B). AC5 levels were low in other regions of the basal ganglia, including the globus pallidus and substantia nigra (data not shown). To examine AC5 expression in the dorsal striatum in more detail, we conducted double labeling in situ hybridization for AC5 and preproenkephalin (ENK) to ask whether AC5 was differentially expressed in subsets of medium spiny neurons (MSN) in the striatum (Figs.1C–D). ENK is expressed in neurons projecting via the striatopallidal indirect pathway to the substantia nigra and is absent in those neurons projecting via the striatonigral direct pathway (Beckstead and Kersey, 1985; Gerfen and Young, 1988). AC5 mRNA expression was seen in both ENK positive and ENK negative cells (Figs. 1C–D), suggesting that AC5 was expressed in both subtypes of striatal projection neurons.
AC5 deficient mice (AC5KO) have impaired DA receptor mediated modulation of cAMP levels in the striatum (Iwamoto et al., 2003). In addition, recent studies have indicated AC5 plays a role in appetitive Pavlovian learning (Kheirbek et al., 2008), a task attributed to DA in the ventral striatum (Parkinson et al., 2002). To examine whether AC5 also plays a role in dorsal striatum-dependent learning, we tested mice in a well-established dorsal striatum-dependent response learning task (Tolman and Gleitman, 1949; Restle, 1957; Packard and McGaugh, 1996; Packard, 1999; Schroeder et al., 2002). In a water-filled cross maze, mice were trained to swim to find a submerged hidden platform using a response (egocentric) strategy, i.e. making the same body turn on each trial to find the goal arm. In the first phase of the experiment, mice were randomly started from either the north or south arm of the maze, and trained to find the hidden platform in the arm to the right of the start arm. Mice were given ten trials a day until they reached a learning criterion of nine out of ten correct choices for three consecutive days. Wild-type (WT) mice learned the task in approximately 3 sessions, while AC5KO mice required 2–3 times as many sessions to reach learning criterion, reflecting severe impairment in response learning (Fig 2A,C, n=8 AC5KO, 8 WT, number of trials to reach criterion WT, 2.9 ± 0.5 SEM, AC5KO 8.6 ± 2.4 SEM, t-test, t14=2.31, genotype effect p=0.035). After learning the task, mice were trained to find the hidden platform on the opposite side, i.e. the arm to the left of the start arm, to test their ability to learn a new response strategy. AC5KO mice required significantly more trials to reach learning criterion, again demonstrating severely impaired learning (Fig 2A, D, number of trials to reach criterion WT 10.38 ± 2.1 SEM, AC5KO 24.75 ± 2.4 SEM, t-test, genotype effect t14= 4.49, p=0.0005). During reversal learning, AC5KO mice required significantly more trials to extinguish previous choice and reach chance performance (Fig 2B, number of number of trials to reach chance, WT 6 ± 1.0 SEM, AC5KO 14.8 ± 3.8 SEM, t test, genotype effect t14 = 2.24, p=0.04) and once chance performance was reached, significantly more trials to reach learning criterion (Fig 2B, number of trials to reach criterion after chance WT 3.75 ± 1.1 SEM, AC5KO 10.3 ± 2.3 SEM, t test, t14=2.59, p=0.021). Interestingly, AC5KO mice do eventually learn the task, suggesting that either non-AC5 mechanisms in the striatum or non-striatal dependent mechanisms can compensate for loss of AC5. Similar phenomena have been well documented in other systems (Wiltgen et al., 2006).
To test if the above behavioral deficits were due to a global deficit in learning, mice were subjected to a spatial learning task in the Morris water maze. Mice were trained for seven days, three trials a day to find a hidden platform using spatial cues situated around the room. Both genotypes reduced latency to find the hidden platform across training days (Fig 2G, n=6 AC5KO, 8 WT, ANOVA, genotype x day interaction, F(6,72)=0.572, p=0.75). To test spatial memory, 24 hrs after the last training day, mice were given a one-minute probe trial to measure percent of total time spent in each quadrant. Both AC5KO and WT mice spent significantly more than 25% of their time in the target quadrant (Fig 2H, single sample t test for WT p=0.012, AC5KO p=0.008). In addition, there was no difference between genotypes in time spent in the target quadrant (genotype main effect, F(1,12)=0.13, p=0.7). This suggests spatial memory in the Morris water maze is intact in AC5KO mice. There was an insignificant genotype effect in latency to find the hidden platform during training sessions (Fig 2G, ANOVA genotype main effect, F(1,12)=4.62, p=0.053), indicating a performance deficit and not a learning deficit. This could be attributed to a slight motor skill deficit associated with swim speed, as a similar deficit was seen during habituation sessions when mice were tested for latency to find a visible platform (Supplemental figure S1, ANOVA, genotype main effect, F(1,12)=5.08, p=0.04, genotype X trial interaction F(1,12)=0.103, p=0.75).
Plasticity in corticostriatal circuits has also been implicated in motor learning in the accelerating rotarod task (Costa et al., 2004; Dang et al., 2006; Yin et al., 2009). In this task, mice were placed on an accelerating rotarod (4–40 RPM in 5 min) and staying on the rod required learning a sequence of coordinated movements. To confirm and extend previous results reporting rotorod deficits in AC5KO mice (Iwamoto et al., 2003), Mice were trained for four trials a day for 3 consecutive days and learning was assessed by an increase in latency to fall off the rod across trials. AC5KO mice exhibited a severe deficit in performing this task (Fig 3A, n= 8 AC5KO, 8 WT, ANOVA genotype main effect, F(1,14)=17.31, p=0.001, genotype X day interaction F(2,28) =0.07, genotype X trial interaction F(3,42)=2.49, p=0.07, genotype X day X trial interaction F(6,84) =1.4, p=0.22). Although both genotypes performed similarly on the first trial, WT mice greatly improved their performance across the four trials on the first day, while AC5KO mice did not (first day ANOVA, genotype main effect F(1,14)= 5.12, p=0.04, genotype X trial interaction effect, F(3,42)=3.65 p=0.02). In subsequent days, performance in AC5KO was significantly impaired as compared to WT mice as seen by comparison of the average time on the rod across days (Fig 3B, ANOVA genotype main effect, F(1,14)=17.31 p=0.001, genotype X day interaction F(2,28)= 2.9, p=0.07) as well as a failure to increase their performance on first trial each day (Fig 3C, ANOVA genotype main effect, F(1,14)=8.81, p=0.01, day X genotype interaction F(2,28)= 5.27, p=0.01). This deficit was not due to a general locomotor deficit, as activity in AC5KO mice did not differ significantly from their WT littermates when tested in one-hour sessions for three consecutive days for distance traveled in the open field (Fig 3D, ANOVA genotype main effect, F(1,14)=1.9, p=0.2). In addition, AC5KO and WT mice habituated similarly to the novel environment of the locomotor chamber during the one-hour session, on the first day of testing (Fig 3E, ANOVA genotype X blocks interaction effect, F(1,14)=3.14, p=0.1).
To assess potential differences in striatal dopaminergic transmission that may account for the observed behavioral deficits in AC5KO mice, we used fast-scan cyclic voltammetry to compare electrical stimulus evoked DA overflow in tissue slices from each genotype (Fig. 4A). The amount of DA overflow observed with one, two, and four local striatal stimulations given at 25 Hz did not significantly differ between WT and AC5KO mice (Fig. 4B, ANOVA genotype effect F1,38 = 0.02, P = 0.90; stimulus number effect F1,38 = 4.31, P = 0.04; interaction effect F1,38 = 0.04, P = 0.85). We also examined the DA overflow elicited by paired pulses with interstimulus intervals ranging between 2 and 60 seconds and found no differences between the genotypes (Fig. 4C–D, ANOVA genotype effect F1,36 = 0.07, P = 0.79; interstimulus interval effect F1,36 = 77.90, P < 0.001; interaction effect F1,36 = 0.01, P = 0.94). The lack of effect of AC5 deletion on DA release, suggests the D2 autoreceptor function is intact in these mice, signaling through an AC5-independent mechanism, likely via other isoforms of AC. This is consistent with our previous report indicating that injection of a low dose of quinpirole to preferentially activate pre-synaptic D2 autoreceptors significantly reduces locomotor activity in AC5KO mice (Kheirbek et al., 2008).
Using striatal tissue slices, we investigated possible differences in the physiological properties of striatal MSNs between WT and AC5KO mice. Whole-cell voltage clamp recordings from MSNs in the dorsolateral striatum revealed that genetic deletion of AC5 did not significantly alter MSN resting membrane potential (WT = −76 ± 1 mV, KO = −77 ± 1 mV, t62 = 0.51, P = 0.61), cell capacitance (WT = 35 ± 2 pF, KO = 32 ± 2 pF, t60 = 1.00, P = 0.32), or cell input resistance (WT = 66 ± 7 MΩ, KO = 77 ± 6 MΩ, t62 = 1.11, P = 0.27). To test the intrinsic membrane excitability of striatal MSNs, we measured action potential frequency during depolarizing current injections (Supplemental Figure 2A). The average firing rate during 500 ms current injections did not differ between the genotypes (Supplemental Figure 2B, ANOVA genotype effect F1,237 = 0.11, P = 0.74; input current effect F1,237 = 53063, P < 0.001; interaction effect F1,237 = 4.31, P = 0.65). MSNs from AC5KO mice did show an exaggerated adaptation in spike frequency (Supplemental Figure 2C, ANOVA genotype effect F1,158 = 0.64, P = 0.43; spike number effect F1,158 = 85.23, P < 0.001; interaction effect F1,158 = 24.02, P < 0.001), which suggests possible differences in expression of small conductance calcium-activated potassium current (SK) (Stocker et al., 1999). Although the AC5KO mice showed larger mEPSC amplitudes and less spike-frequency adaptation, it is unlikely that these differences in baseline excitability contributed to the observed differences in synaptic plasticity.
To examine the synaptic properties of the glutamatergic afferents to MSNs, we recorded miniature excitatory postsynaptic currents (mEPSCs) in MSNs from both WT and AC5KO mice (Supplemental Figure 3A). There was no significant difference in mEPSC frequency between the genotypes (Supplemental Figure 3B t30 = 1.08, P = 0.29), but AC5KO mice did have larger amplitude mEPSCs (Fig. 5C, t30 = 2.96, P = 0.006). This suggests that the AC5KO mice may have more synaptic AMPA receptors on MSNs than the WT mice. To assess possible differences in the status of synaptic plasticity and presynaptic function, we measured the AMPAR/NMDAR ratio and paired pulse ratio (PPR) between the two genotypes. For these studies, we electrically stimulated the white matter adjacent to the dorsolateral striatum to preferentially activate cortical afferents to MSNs (Fig. 5A). The AMPAR/NMDAR ratio was not different between the WT and AC5KO mice (Fig. 5B, t19 = 0.44, P = 0.66). The fact that AC5KO mice exhibited larger mEPSC amplitudes without any difference in AMPAR/NMDAR ratio suggests that the MSNs in the AC5KO mice had an increased density of both AMPA and NMDA receptors. This may be a consequence of homeostatic plasticity due to differences in network activity in the absence of AC5 (Turrigiano, 2007). Consistent with the similarity in mEPSC frequency, the paired pulse ratio did not differ between the genotypes (Fig. 5C, ANOVA genotype effect F1,110 = 0.45, P = 0.50; ISI effect F3,110 = 0.92, P = 0.43; interaction effect F3,110 = 0.19, P = 0.91), suggesting that presynaptic release probability is similar between the genotypes.
The best characterized form of plasticity in the dorsal striatum is long-term depression (LTD) of corticostriatal synapses (Calabresi et al., 2007). This form of plasticity is induced by pairing high frequency stimulation of glutamatergic afferent fibers with postsynaptic depolarization (Choi and Lovinger, 1997). To assess possible differences in corticostriatal plasticity in AC5KO mice, we compared LTD induction in brain slices from AC5KO and WT mice. Cortical glutamatergic inputs to MSNs in the dorsolateral striatum were stimulated by an extracellular electrode placed in the adjacent white matter (Fig. 5D). High frequency stimulation (four 1 s 100 Hz trains delivered every 10 s) paired with postsynaptic depolarization (to 0 mV) produced LTD on average in recordings from WT mice (Fig. 5E, 73 ± 8% of baseline at 30–40 min, n = 11 cells/slices from 8 animals t10 = 2.83, P = 0.02). This LTD accompanied an increase in the paired pulse ratio (109 ± 2% of baseline at 30 to 40 min, t10 = 2.53, P = 0.03), which is consistent with previous observations that corticostriatal LTD is mediated presynaptically by a decreased release probability (Choi and Lovinger, 1997). In tissue slices from AC5KO mice, the same HFS protocol did not induce long term plasticity on average (Fig 5E, 105 ± 8% of baseline at 30 to 40 min, n = 9 cells/slices from 7 animals, t8 = 1.79, P = 0.86). Individual recordings showed some heterogeneity in the type and magnitude of plasticity observed following the LTD induction protocol. In WT recordings, where 7/11 cells showed significant LTD (t test, p < 0.05) that averaged to 59 ± 7% of baseline at 30 to 40 min, and 4/11 cells showed no significant change in synaptic strength (t test, p > 0.05). Such heterogeneity is in agreement with previous studies (Kreitzer and Malenka, 2007). In AC5KO slices, only 2 cells exhibited a low magnitude LTD (85% for both), 4 cells showed no significant long-term changes, and 3 cells exhibited low magnitude LTP (131 ± 14%). Thus, the magnitude and prevalence of striatal LTD were both significantly lower in AC5KO mice relative to WT. Although the AC5KO mice showed larger mEPSC amplitudes and less spike-frequency adaptation (Supplementary Figs 2–3), it is unlikely that these differences in baseline excitability contributed to the observed differences in synaptic plasticity. The firing properties to prolonged depolarization would not effect the presynaptic stimulation, as each stimulus was ≤ 0.4 msec, even during HFS. While stronger quantal responses could affect plasticity, stronger postsynaptic excitation and calcium entry should have enhanced plasticity, but we saw the opposite effect in the slices from AC5KO mice.
To determine if the impairment in corticostriatal plasticity was a result of presynaptic deficits, we bath applied the cannabinoid receptor type 1 (CB1) agonist Win 55,212-2 (1 μM) while monitoring evoked EPSCs. Both WT and AC5KO mice exhibited robust, persistent suppression of EPSC amplitude following Win 55,212-2 application, demonstrating that the presynaptic glutamatergic terminals were capable of depression in response to CB1 stimulation in AC5KO mice (Fig 6A; WT: 52 ± 1% of baseline at 35 to 45 min, n = 5 cells/slices from 3 animals, t4 = 4.61, P = 0.01; KO: 59 ± 1% of baseline at 35 to 45 min, n = 8 cells/slices from 3 animals, t7 = 7.73, P < 0.001). The lack of HFS-induced LTD in AC5KO mice could either be due to a lack of AC5-mediated modulation in endocannabinoid synthesis, or a deficit in endocannabinoid release. To address this issue, we tested DA-independent endocannabinoid-mediated LTD in MSNs using the mGluR agonist DHPG. Bath application of DHPG (100 μM) elicited LTD in both genotypes, while the cells were voltage clamped at −50 mV (Fig 6B; WT: 78 ± 1% of baseline at 30 to 40 min, n = 8 cells/slices from 4 animals, t7 = 2.45, P = 0.04; KO: 80 ± 1% of baseline at 30 to 40 min, n = 11 cells/slices from 5 animals, t10 = 2.46, P = 0.03). This finding demonstrates that endocannabinoid-mediated synaptic plasticity can be induced in MSNs from AC5KO mice.
To address whether D2 receptors functionally couple to AC5 to mediated endocannabinoid release and LTD, we generated AC5KO/D2-EGFP and WT/D2-EGFP littermate control mice (see Materials and Methods) to selectively record from D2-expressing indirect pathway neurons in both genotypes. In WT, D2-positive MSNs a brief, one minute application of DHPG (100 μM) has been demonstrated to induce a lasting depression of EPSCs if applied in the presence of the D2 receptor agonist quinpirole (10 μM; Kreitzer & Malenka, 2007). This manipulation effectively depressed EPSCs in GFP-labeled D2 receptor-containing neurons in WT but not AC5KO mice (Fig 6C; WT: 68 ± 4% of baseline at 30 to 40 min, n = 7 cells/slices from 3 animals, t6 = 4.24, P = 0.01; KO: 97 ± 5% of baseline at 30 to 40 min, n = 6 cells/slices from 3 animals, t5 = 0.92, P = 0.40), suggesting that D2 receptors couple to AC5 to induce endocannabinoid release and LTD.
Considerable evidence exists that corticostriatal LTD and retrograde signaling via endocannabinoid release is dependent on D2 receptor activation, yet the signaling cascades downstream of D2 receptor activation that are necessary for LTD induction are unknown (Piomelli, 2003; Yin and Lovinger, 2006; Calabresi et al., 2007). As D2 receptors are negatively coupled to cAMP production by Gi/o inhibition of AC5 (Iwamoto et al., 2003), we hypothesized that D2 receptors reduce AC5 activity and cAMP levels during LTD induction. To test this, we attempted to induce LTD in WT neurons in which cAMP levels were clamped to 1 mM by including cAMP in the internal solution. High frequency stimulation paired with postsynaptic depolarization in WT cells that were dialyzed with 1 mM cAMP in the internal solution induced LTP on average (Fig. 6D, 123 ± 1% of baseline at 35 to 45 min, n = 8 cells/slice from 5 animals, t7 = 2.45, P = 0.04). Tabulation of individual responses revealed that half the cells exhibited LTP that averaged to 140 ± 6% of baseline and half the cells showed no significant changes in synaptic strength and averaged to 100 ± 2% baseline at 30 to 40 min. In slices from AC5KO mice, the same induction protocol in combination with 1 mM cAMP in the internal solution did not, on average, induce any long-term changes in synaptic strength (Fig. 6D, 97 ± 1% of baseline at 35 to 45 min, n = 7 cells/slices from 3 animals, t7 = 0.24, P = 0.82). Of this population, 2 cells exhibited significant LTP (131 ± 9%), 2 cells exhibited significant LTD (82 ± 7%), and 3 cells did not show any significant long-term changes (102 ± 2%). Thus, cells from AC5KO mice after cAMP loading behaved similarly to cells without cAMP loading.
Here we demonstrate that striatum-dependent response and skill learning, as well as LTD of the cortical inputs to dorsolateral striatum are dependent upon AC5 and cAMP signaling. Our data show that loss of AC5, an isoform of AC highly expressed in the striatum, severely impairs response and motor skill learning, but not spatial learning. These behavioral deficits are correlated with a loss of LTD in the dorsal striatum of AC5KO mice. This deficit in LTD could be rescued by agonist activation of presynaptic cannabinoid receptors, or by DA-independent stimulation of postsynaptic endocannabinoid release. Based upon the results presented here, we hypothesize that AC5 regulation of cAMP production during LTD induction facilitates endocannabinoid release and activation of presynaptic CB1 signaling to suppress glutamatergic transmission. In addition, we hypothesize that impairment in these plasticity mechanisms underlie the deficits in striatum-dependent memory exhibited by AC5KO mice.
Deficits in procedural learning in human patients or experimental animals with either damage or lesions to the dorsal striatum supports the critical role of this brain area (Packard and Knowlton, 2002). In double dissociation studies, inhibition of the dorsal striatum, but not hippocampus, resulted in impaired motor response learning, while hippocampal inhibition interfered with spatial learning (Packard and White, 1991; Packard and McGaugh, 1996). In addition, recent studies have implicated modulation of dorsal striatal circuits during learning a motor skill on the accelerated rotorod (Costa et al., 2004; Yin et al., 2009). Our data indicates that AC5 plays a critical role in learning both these tasks by modulating corticostriatal plasticity. It is important to note, as with many studies on the molecular basis of behavior, the data presented here provides a correlation between the physiology and behavior. To further strengthen the link between physiology and behavior, future studies using more elegant genetic designs to manipulate other components of this signaling pathway will be required. One possibility would be to conditionally delete AC5 in D2 expressing MSNs to observe if this manipulation would replicate the behavioral deficits seen here. An alternative strategy could be conditional deletion of CB1 receptors on presynaptic corticostriatal terminals. In addition, care must be taken in interpretation of results with constitutive knockouts, like AC5KO, as compensatory developmental changes in striatal signaling may contribute to the deficits seen here. As AC5 is a direct downstream target of DA signaling (Iwamoto et al., 2003), and the role of DA in regulating corticostriatal plasticity and striatum dependent behaviors is well established (Packard and Knowlton, 2002; Reynolds and Wickens, 2002), it is likely that the behavioral deficits seen in AC5KO mice is directly due to loss of AC5. Yet, future studies using specific pharmacological inhibition of dorsal striatal AC5, or temporal control of AC5 gene expression in specific cell types will be required to rule out developmental confounds.
In brain slices, LTD is a predominant form of synaptic plasticity at corticostriatal synapses (Calabresi et al., 1992; Lovinger et al., 1993; Calabresi et al., 1994; Kreitzer and Malenka, 2007). There is general agreement that corticostriatal LTD in the dorsal striatum requires D2 receptor activation, increases in intracellular calcium postsynaptically, and retrograde signaling via endocannabinoids that depress glutamate release via presynaptic CB1 receptors (Lovinger et al., 1993; Calabresi et al., 1994; Calabresi et al., 1997; Gerdeman et al., 2002; Wang et al., 2006; Kreitzer and Malenka, 2007). A direct downstream target of DA receptor stimulation is AC, and signaling mechanisms downstream of cAMP, such as DARPP-32 and CREB have been implicated in corticostriatal LTD (Calabresi et al., 2000; Pittenger et al., 2006). However, as these proteins are modulated by calcium-dependent signaling pathways (Kang et al., 2001; Nishi et al., 2005), the contribution of cAMP upstream of these signals is unknown. The data presented here demonstrates AC5 plays a critical role in this pathway.
A provocative finding in the current study is that LTD could be rescued in AC5KO mice by bath application of the CB1 agonist WIN 55,212-2. CB1 receptors are localized on glutamatergic corticostriatal presynaptic terminals and mediate the retrograde endocannabinoid signal from striatal neurons (Gerdeman et al., 2002). In addition, it has been documented that D2 receptor stimulation produces an increase in anandamide overflow in the striatum (Giuffrida et al., 1999), yet the underlying signaling mechanisms remain unclear (Piomelli, 2003). As CB1 receptor activation could rescue LTD, our data suggest that AC5-cAMP in corticostriatal LTD is in postsynaptic striatal neurons, upstream of endocannabinoids and CB1 activation. In addition, this result indicates that although very low levels of AC5 could be detected in cortical areas, CB1 receptors on cortical terminals do not require AC5 for regulating LTD induction. This ability to rescue LTD with CB1 receptor agonist suggested AC5 may contribute to either the release or biosynthesis of endocannabinoids in the MSN. We examined these possibilities by testing whether the mGluR agonist DHPG could induce LTD in AC5KO mice. In the dorsal striatum, DHPG can trigger LTD in a D2-independent, calcium-dependent manner by promoting release of postsynaptic endocannabinoids from MSNs (Kreitzer and Malenka, 2005). AC5KO mice exhibited a DHPG-LTD comparable to WT mice. This suggests the impaired HFS-induced LTD in AC5KO mice is not due to a deficit in endocannabinoid signaling, but rather to impaired DA modulation of endocannabinoid release during the induction protocol. Finally, as D2 receptor stimulation promotes endocannabinoid release in vivo (Giuffrida et al., 1999) and D2 receptor activation is required for LTD induction (Calabresi et al., 1994; Calabresi et al., 1997), we directly tested whether D2 receptor coupling to AC5 is required for endocannabinoid production and LTD. Using mice with genetically labeled D2-expressing striatopallidal neurons, we showed that while a short application of DHPG in the presence of quinpirole could induce LTD in WT cells, it could not in AC5KO cells. This suggests that the D2 receptor directly couples to AC5 to regulate endocannabinoid mediated LTD.
Our data further suggest that regulation of cAMP levels by AC5 may be required for corticostriatal LTD, as corticostriatal LTD can be effectively eliminated by elevated postsynaptic cAMP levels in WT striatal neurons. Such a manipulation would block D2-mediated inhibition of cAMP levels, which is similar to that seen following genetic deletion of D2 receptors or chronic haloperidol treatment that also eliminate corticostriatal LTD (Centonze et al., 2004). The data suggest that D2 activation inhibits tonic AC5 activity during LTD induction. The mechanism of tonic AC5 activation is unknown, however one possibility is the adenosine A2a receptors that co-localize with D2 receptors in the striatum and couple to AC5 to increase cAMP production (Fink et al., 1992; Lee et al., 2002). Future studies examining the consequence of brief application of A2a antagonist on this form of LTD may provide insight into the mechanism of tonic AC5 activation in D2-expressing neurons.
In addition, elevated postsynaptic cAMP levels in WT cells resulted in a net LTP of corticostriatal synapses, which was also seen with D2 receptor knockout or chronic haloperidol treatment (Centonze et al., 2004). In our experiments, the elevation of cAMP in the MSN may also contribute to LTP induction, as two receptors that are positively coupled to cAMP signaling, A2a and D1 receptors, contribute to LTP induction at corticostriatal synapses (Shen et al., 2008) (Reynolds et al., 2001; Reynolds and Wickens, 2002). Elevation of cAMP levels in AC5KO mice did not, on average, alter synaptic strength, suggesting impairment in the ability to strengthen synapses in AC5KO. This is not surprising, as cAMP signaling contributes to LTP induction at these inputs, as mentioned.
Independent of the LTP effects, our data shows that D2-mediated changes in cAMP are necessary for LTD at the glutamatergic inputs to striatal MSNs. A major goal for future investigations will be the identification of downstream cAMP-responsive proteins that couple DA receptors and AC5 activity to the induction of plasticity at corticostriatal synapses. The control of endocannabinoid release is poorly understood, which is a major challenge in the field, as recent studies illustrate (Adermark et al, 2007). Focusing upon cAMP-dependent pathways may help identify novel molecular mechanisms that contribute to this important mediator of striatal plasticity.
In conclusion, the current study elucidates the critical role of AC5, an isoform of AC highly expressed in the striatum, in striatum-dependent response learning. The data provide insight into the molecular mechanisms underlying LTD induction in the dorsal striatum, and how DA regulates this form of plasticity via an AC5-cAMP-endocannabinoid-CB1 signaling pathway. It establishes a framework for analysis of the contribution of corticostriatal plasticity to dorsal striatum dependent behaviors.
We thank Wei-Jen Tang for helpful discussions in preparing the manuscript. This work was supported in part by NIH F31MH076422 (MAK), F31DA023340 (JPB), F32DA020427 and RO1DA25875-2 (JAB), HL059139 and GM067773 (YI), DA015918 (DSM), DA022269, MH66216 and Edward Mallinckrodt, Jr. Foundation (XZ).