Abstract
Dynamic regulation of synaptic transmission at cortical inputs to the dorsal striatum is considered critical for flexible and efficient action learning and control. Presynaptic mechanisms governing the properties and plasticity of glutamate release from these inputs are not fully understood, and the corticostriatal synaptic processes that support normal action learning and control remain unclear. Here we show in male and female mice that conditional deletion of presynaptic proteins RIM1αβ (RIM1) from excitatory cortical neurons impairs corticostriatal synaptic transmission in the dorsolateral striatum. Key forms of presynaptic G-protein-coupled receptor-mediated short- and long-term striatal plasticity are spared following RIM1 deletion. Conditional RIM1 KO mice show heightened novelty-induced locomotion and impaired motor learning on the accelerating rotarod. They further show heightened self-paced instrumental responding for food and impaired learning of a habitual instrumental response strategy. Together, these findings reveal a selective role for presynaptic RIM1 in neurotransmitter release at prominent basal ganglia synapses, and provide evidence that RIM1-dependent processes help to promote the refinement of skilled actions, constrain goal-directed behaviors, and support the learning and use of habits.
SIGNIFICANCE STATEMENT Our daily functioning hinges on the ability to flexibly and efficiently learn and control our actions. How the brain encodes these capacities is unclear. Here we identified a selective role for presynaptic proteins RIM1αβ in controlling glutamate release from cortical inputs to the dorsolateral striatum, a brain structure critical for action learning and control. Behavioral analysis of mice with restricted genetic deletion of RIM1αβ further revealed roles for RIM1αβ-dependent processes in the learning and refinement of motor skills and the balanced expression of goal-directed and habitual actions.
Introduction
The capacities to select, refine, and automatize actions arise from concerted activity and plasticity of multiple, interconnected neural networks. Central to these networks is the dorsal striatum, the primary input nucleus of the basal ganglia (Balleine et al., 2007; Heilbronner et al., 2016). Excitatory projections emanating broadly from cerebral cortex, thalamus, and amygdala innervate the dorsal striatum in complex patterns (Kelley et al., 1982; McGeorge and Faull, 1989; Smith et al., 2004; Hintiryan et al., 2016; Hunnicutt et al., 2016), contributing to functional heterogeneity across its various subregions. For example, the learning and use of goal-directed action strategies preferentially recruit and rely on dorsomedial striatal subregions (Barnes et al., 2005; Yin et al., 2005a,b; Thorn et al., 2010; Corbit et al., 2013; Gremel and Costa, 2013; Hart et al., 2018), whereas habitual strategies and skilled action patterns preferentially engage and depend on intact dorsolateral subregions (Yin et al., 2004, 2006, 2009; Dias-Ferreira et al., 2009; Thorn et al., 2010; Kupferschmidt et al., 2017). By gating and integrating diverse inputs across these subregions, the dorsal striatum functions to facilitate adaptive behavioral control (Graybiel and Grafton, 2015; Rueda-Orozco and Robbe, 2015; Peak et al., 2018).
Cortical inputs provide a primary source of excitatory drive of dorsal striatal neurons (Wilson, 1995; Huerta-Ocampo et al., 2014); thus, dynamic changes in corticostriatal transmission are considered key determinants of dorsal striatum function and adaptive actions. Postsynaptic plasticity at putative corticostriatal synapses has been well established to contribute to action learning and control (Costa et al., 2004; Barnes et al., 2005; Yin et al., 2005a, 2006, 2009; Dang et al., 2006; Pittenger et al., 2006; Hawes et al., 2015; Santos et al., 2015). Glutamate release from corticostriatal terminals is also potently modulated by various presynaptic signaling and plasticity processes (Calabresi et al., 1990; Lovinger, 1991; Atwood et al., 2014; Park et al., 2014; Kupferschmidt and Lovinger, 2015). Ex vivo electrophysiology and gene deletion studies suggest roles for presynaptic plasticity at corticostriatal terminals in skill learning (Barnes et al., 2005; Kheirbek et al., 2009; Yin et al., 2009; Hawes et al., 2015), and manipulations of presynaptic corticostriatal receptor function bias the learning and use of goal-directed and habitual action strategies (Nazzaro et al., 2012; Gremel et al., 2016). Notably, loss of presynaptic cannabinoid-1 receptor (CB1R)-dependent long-term depression (LTD) of excitatory transmission in the dorsolateral striatum (DLS) has been proposed to promote habit formation (Nazzaro et al., 2012), whereas endocannabinoid signaling in dorsomedial striatum fosters habit formation (Gremel et al., 2016). Furthermore, in vivo monitoring of corticostriatal input function has revealed presynaptic activity dynamics that encode and predict action learning (Kupferschmidt et al., 2017). Despite these advances, little is known about the synaptic mechanisms through which corticostriatal transmission supports action learning and control.
Rab3-interacting molecules (RIMs) are presynaptic scaffolding proteins that coordinate key active zone processes involved in synaptic transmission (Südhof, 2013). RIMs form the core of a macromolecular complex that recruits voltage-gated calcium channels to active zones, docks synaptic vesicles to the exocytotic machinery, and primes them for release (Wang et al., 1997; Coppola et al., 2001; Hibino et al., 2002; Sun et al., 2003; Deng et al., 2011; Han et al., 2011; Kaeser et al., 2011). Seven different RIM isoforms are encoded by four genes (Rims1–4), the first of which gives rise to isoforms RIM1α and RIM1β (collectively, RIM1) (Kaeser et al., 2008a). These isoforms are expressed widely in brain, with particularly high mRNA and protein levels in forebrain and cerebellar structures (Schoch et al., 2006; Kaeser et al., 2008a). Global and conditional knockout (KO) of RIM1 isoforms has revealed synapse-specific roles for the proteins in synaptic transmission and presynaptically expressed plasticity (Castillo et al., 2002; Schoch et al., 2002; Calakos et al., 2004; Chevaleyre et al., 2007; Fourcaudot et al., 2008; Kaeser et al., 2008b; Grueter et al., 2010). For example, RIM1α deletion selectively impairs synaptic transmission and short-term plasticity at hippocampal Schaffer collateral-CA1 synapses (Schoch et al., 2002), whereas mossy fiber synapses show selective deficits in long-term plasticity after such deletion (Castillo et al., 2002; Yang and Calakos, 2010). These and other synaptic phenotypes have been linked to behavioral abnormalities following RIM1 deletion, including altered maternal behavior, fear and spatial learning, locomotion, and prepulse inhibition (Schoch et al., 2002; Powell et al., 2004; Blundell et al., 2010; Haws et al., 2012). Despite the importance of corticostriatal transmission and its presynaptic regulation to dorsal striatal function, the contributions of RIM1 to these synaptic processes have yet to be tested. Moreover, the discrete presynaptic proteins and processes within corticostriatal inputs that support action learning and control remain poorly understood.
Here we show that conditional deletion of RIM1 from cortical pyramidal neurons, including corticostriatal projection neurons, impairs excitatory transmission in the DLS, but spares forms of GPCR-mediated short- and long-term striatal plasticity. Conditional RIM1 KO mice show heightened novelty-induced locomotion and impaired motor learning on the accelerating rotarod. They further show heightened instrumental responding for food under higher-effort conditions and impaired expression of habitual food responding. These findings implicate RIM1-dependent processes in corticostriatal transmission and in the learning and control of motivated behavior.
Materials and Methods
Mice
All mice used in these experiments were obtained from The Jackson Laboratory and backcrossed onto a C57BL/6J background (stock #000664) for at least 5 generations. Floxed RIM1 mice (RIM1fl/fl, stock #015832) (Kaeser et al., 2008a) that carry loxP sites flanking exon 6 of the RIM1 gene were bred with Emx1::Cre mice (stock #005628) (Gorski et al., 2002) that express Cre recombinase in the majority of excitatory cells in the neocortex and hippocampus (and sparser populations in some ventral pallial structures) using a 3-stage breeding strategy: (1) Emx1::Cre mice were crossed with RIM1fl/fl mice; (2) Emx1::Cre;RIM1−/+ offspring were crossed with RIM1fl/fl mice; and (3) Emx1::Cre;RIM1−/− offspring were crossed again with RIM1fl/fl mice. This strategy generated age- and sex-matched littermates with RIM1 either present in (RIM1fl/fl; Control) or deleted from Emx1Cre-expressing cells (Emx1::Cre;RIM1−/−). Mice were housed in groups of 2–4 in a humidity- and temperature-controlled colony room on a 12 h light/dark cycle with ad libitum access to water and food (unless otherwise specified). Mice that underwent testing for novelty-induced locomotion were later tested for motor learning on the accelerating rotarod and prepulse inhibition. Mice tested for homecage locomotion, fixed ratio (FR)/progressive ratio (PR) instrumental training, and fixed/random interval (RI) instrumental training comprised 3 separate cohorts. All procedures conformed to the National Institutes of Health Guide to the care and use of laboratory animals and were approved by the National Institute on Alcohol Abuse and Alcoholism Animal Care and Use Committee.
qRT-PCR
Mice were anesthetized with isoflurane and decapitated, and brains were rapidly removed and rinsed in cold 0.9% NaCl. Brains were cut into 1 mm coronal sections using a stainless-steel brain matrix, and samples from M1+M2 motor cortices, dorsal striatum, or thalamus were placed in RNAlater (Invitrogen), stored at 4°C for 48 h, then removed from RNAlater and frozen at −20°C before RNA extraction. RNA was extracted using the RNeasy Lipid Tissue Mini Kit (QIAGEN), and RNA concentration was measured using a NanoDrop (Thermo Fisher Scientific). RNA (100 ng) was reverse transcribed using the Quantitect Reverse Transcription Kit (QIAGEN). qPCRs contained 2 μl cDNA template, 1 μl predesigned TaqMan Gene Expression Assay (for Rim1, Mm01225745_m1), 10 μl TaqMan Fast Advanced Master Mix (both from Thermo Fisher Scientific), and 7 μl Ultrapure water (Invitrogen). qPCR was run on StepOnePlus system (Thermo Fisher Scientific) and was performed according to the manufacturer's recommended settings: 50°C for 2 min, 95°C for 2 min, and then 40 cycles of melting at 95°C for 1 s and annealing/extension at 60°C for 20 s. Reactions were run in triplicate for 4 or 5 animals per group. Analysis was performed using the 2−ΔCT method (Schmittgen and Livak, 2008). The CT value was determined by identifying the cycle number at which the amplification curve reached a ΔRn threshold set at 0.1 (ΔRn = Rn − baseline, where Rn is the fluorescence of the reporter dye divided by the fluorescence of a passive reference dye). Relative Rim1 mRNA expression was quantified using Actb as an internal control (Applied Biosystems #4351315). Thus, ΔCT = CT (Rim1) − CT (Actb).
Immunohistochemistry
RIM1fl/fl and Emx1::Cre;RIM1−/− mice were transcardially perfused under deep isoflurane anesthesia with 0.1 m PBS and 4% formaldehyde in PBS, pH 7.4. Brains were collected and postfixed in 4% formaldehyde at 4°C for 24 h. The 50 μm coronal sections were made using a vibratome. On a revolving platform, sections were incubated in PBST (PBS with 0.2% Triton X-100) for 2 h, and blocked in 5% BSA in PBST for 4 h. Slices were then incubated in chicken anti-Neurofilament (Encor Biotechnology, catalog #CPCA-NF-H, 1:40,000) with rabbit anti-mu opioid receptor (Immunostar, 24216, 1:4000), rabbit anti-parvalbumin (Swant, PV27, 1:1000), or rabbit anti-cannabinoid-1 receptor (CB1R; Synaptic Systems, 258003, 1:1000) antibodies in PBST for 12 h at 4°C. Following three 1 h washes in PBST, slices were incubated in AlexaFluor-488 goat anti-chicken (Invitrogen, A-11039, 1:2000), AlexaFluor-488 goat anti-rabbit (Invitrogen, A-11034, 1:1000), and/or AlexaFluor-568 goat anti-rabbit (Invitrogen, A-11011, 1:1000) antibodies in PBST for 12 h at 4°C. Following a 1 h wash in PBST containing DAPI (Invitrogen, D3571, 1:20,000) and three 1 h washes in PBS, sections were mounted on Superfrost Plus slides (Daigger, EF15978Z) using Fluoromount Aqueous Mounting Medium (Sigma-Aldrich, F4680), coverslipped, and imaged using Carl Zeiss AxioVision LE 4.3 software with a Carl Zeiss Axiocam on a Carl Zeiss SteREO Lumar microscope. Brightness was uniformly adjusted in each image for presentation.
Electrophysiology
Brain slice preparation for electrophysiology.
Following isoflurane anesthesia, RIM1fl/fl and Emx1::Cre;RIM1−/− mice (P18–P50) were decapitated, and brains were rapidly removed and submerged in ice-cold cutting solution containing the following (in mm): 30 NaCl, 4.5 KCl, 1 MgCl2, 26 NaHCO3, 1.2 NaH2PO4, 10 glucose, and 194 sucrose, continuously bubbled with 95% O2/5% CO2. Slices were submerged in aCSF containing the following (in mm): 124 NaCl, 4.5 KCl, 2 CaCl2, 1 MgCl2, 26 NaHCO3, 1.2 NaH2PO4, and 10 glucose, in a holding chamber and continuously bubbled with 95% O2/5% CO2. Slices were incubated for 30 min at 32°C and then held at room temperature for 1–5 h before whole-cell recordings.
Whole-cell voltage-clamp recordings (excluding depolarization-induced suppression of excitation [DSE] and DHPG experiments).
Individual hemisected slices were moved to a custom recording chamber where they were submerged in and continuously perfused with 30°C–32°C aCSF containing picrotoxin (50 μm) at a rate of ∼1.5 ml/min. Borosilicate glass pipettes (2.5–4.0 mΩ resistance in bath) were filled with internal solution (295–310 mOsm, pH 7.3) containing the following (in mm): 120 CsMeSO3, 5 NaCl, 10 tetraethylammonium-Cl, 10 HEPES, 5 QX-314, 1.1 EGTA, 0.3 Na-GTP, and 4 Mg-ATP. Slices were visualized on a Olympus BX51WI microscope with a 40×/0.8 NA water-immersion objective. EPSCs were recorded from medium-sized spiny projection neurons (MSNs) held at −60 mV using a Multiclamp 700A amplifier and Digidata 1332A (Molecular Devices). Recordings were filtered at 2 kHz and digitized at 10 kHz using Clampex 10.3 software (Molecular Devices). A concentric bipolar stimulating electrode (FHC, CBAEC75), controlled by an isolation unit (Digitimer, DS3) and Clampex 10.3, was placed at the border of the DLS and the overlying white matter. Electrical pulses were delivered every 20 s. Stimulation duration (40–80 μs) and intensity (40–100 μA) were adjusted to yield EPSC amplitudes of 200–500 pA. Experiments were discarded if series resistance varied by >20% or increased >25 mΩ. Data were analyzed using Clampfit 10.3.
Whole-cell voltage-clamp recordings (DSE experiments).
Recordings were performed as described above, with the following exceptions. Recording pipettes were filled with a potassium-based internal solution containing the following (in mm): 120 K-gluconate, 4 NaCl, 20 KCl, 10 HEPES, 0.4 Na-GTP, and 4 Mg-ATP, pH adjusted to 7.25 using KOH. Neurons were held at −70 mV, and electrical pulses were delivered every 5 s. EPSC amplitudes were monitored for 50 s before and 100 s after DSE induction. DSE was induced by depolarizing the neuron to 30 mV for 10 s. The data were averaged from 3 trials per cell. For each trial, EPSC amplitudes were normalized to the average baseline amplitude, and between-group comparisons were made on the first EPSC following DSE induction.
Whole-cell voltage-clamp recordings (DHPG experiments).
EPSCs were recorded through pipettes of 3–5 mΩ in MSNs held at −50 mV. EPSCs were evoked every 20 s by a tungsten bipolar stimulating electrode (FHC) with a tip separation of 505 μm placed in the DLS near the border of the external capsule adjacent to the recorded neuron. Stimulation intensity was adjusted to yield EPSC amplitudes of 200–400 pA. Experiments were discarded if series resistance varied by >15% or increased >20 mΩ.
Homecage locomotion
RIM1fl/fl and Emx1::Cre;RIM1−/− mice were individually housed in standard homecages. Mice were then habituated to the locomotor testing room for 1 h/d for 3 d immediately following onset of the dark cycle. On the fourth day, mice were moved to the locomotor testing room for 1 h and then placed into a clean homecage. Horizontal activity was monitored using a photobeam-based system (San Diego Instruments) for 72 consecutive hours, across normal light-dark cycles.
Novel cage locomotion
RIM1fl/fl and Emx1::Cre;RIM1−/− mice were habituated to the locomotor testing room for 1 h/d for 3 d 1–2 h following onset of the light cycle. On the fourth day, mice were individually placed in clear Plexiglas rectangular chambers with minimal bedding, and horizontal activity was monitored using the photobeam-based tracking system for 2 h/d for 6 d. For one cohort of mice, used chambers were exchanged daily with clean chambers (“novel” cage each day); for another cohort, chambers remained unchanged across the 6 d of locomotor testing.
Prepulse inhibition
Prepulse inhibition was assessed in RIM1fl/fl and Emx1::Cre;RIM1−/− mice using procedures similar to those used previously (Blundell et al., 2010) in RIM1α KO mice. A startle chamber (San Diego Instruments) was fixed above a piezoelectric accelerometer and enclosed in a sound-attenuating box (Med Associates). Acoustic stimuli were delivered by a speaker mounted inside the box, 20 cm above the startle chamber. Chamber sensitivity and sound levels were calibrated before the experiment. Each startle response was recorded over a 65 ms period and quantified in arbitrary units. Mice underwent 62 trials of five trial types across a 22 min session: pulse alone (40 ms pulse of 120 dB white noise), three different prepulse/pulse trials (20 ms prepulse of 4, 8, or 16 dB above background noise level of 70 dB preceding 100 ms, 120 dB pulse; onset to onset), and no stimulus. Trials were presented pseudo-randomly with an average intertrial interval of 15 s (7–23 s). A 5 min acclimation period preceded testing in four blocks of trials. Blocks 1 and 4 consisted of six pulse-alone trials; blocks 2 and 3 contained six pulse-alone trials, five of each level of prepulse/pulse, and five no-stimulus trials. Prepulse inhibition was assessed as the percentage of decrease in startle amplitude in prepulse/pulse trials compared with pulse-alone trials.
Accelerating rotarod
RIM1fl/fl and Emx1::Cre;RIM1−/− mice were trained on a rotarod (ENV-575M, Med Associates) that accelerated from 4 to 40 rpm over 300 s. Latency to fall from the rotarod was recorded during 4 trials per day for 3 d, with 15 min between trials. Individual trials were stopped and latency was recorded if the mouse held onto the rod for two consecutive rotations or reached 40 rpm.
Self-paced instrumental training
Eight identical operant chambers (Med Associates) enclosed in sound- and light-attenuating cases were used for instrumental training. Each chamber was equipped with a pellet dispenser that delivered food pellets (F0071, Bioserv) into a recessed food magazine, an infrared photobeam crossing the magazine to allow detection of magazine head entries, two retractable levers to either side of the magazine, a houselight, and a steel rod floor. A computer running Med-PC software (Med Associates) controlled operant chamber events and recorded lever presses and magazine entries.
RIM1fl/fl and Emx1::Cre;RIM1−/− mice were food restricted to 90%–95% of their free-feeding body weight, and handled daily for 2 d before training on one of two instrumental paradigms: [1] FR training and PR testing, and [2] RI training and valuation testing.
FR training and PR testing
Following 2 d of food restriction and handling, RIM1fl/fl and Emx1::Cre;RIM1−/− mice were given one 15 min session of magazine training, during which food pellets were delivered at random 60 s intervals for 15 outcomes. Mice were then trained to press one of two available levers for food pellets in daily 90 min sessions. Responses on the “active” lever resulted in pellet delivery, whereas responses on the “inactive” lever did not. Mice were trained on an FR1 schedule for 5 d with the potential to earn up to 10 and 20 rewards on training days 1 and 2, respectively, and 30 rewards on days 3–5. Mice were then trained on an FR5 schedule for 2 d, and an FR10 schedule for 2 d, each with the potential to earn 30 rewards/d. Mice were then tested in one session on a PR schedule wherein mice had to press the active lever a progressively increasing number of times (10, 20, 25, 32, 40, 50, etc.) (adapted from Richardson and Roberts, 1996) to obtain successive food pellets. The total number of lever presses and the breakpoint achieved (defined as the last series in which a pellet was obtained) were recorded.
RI training and valuation testing
A separate cohort of RIM1fl/fl and Emx1::Cre;RIM1−/− mice were food restricted to 90%–95% of their baseline body weight. Following 2 d of food restriction and handling, mice were given one 15 min session of magazine training, during which food pellets were delivered at random 60 s intervals for 15 outcomes. Mice were then trained to press a single lever for food pellets in daily 90 min sessions on FR1 for 4–5 d, RI30 for 2 d, and RI60 for 4 d. One to 3 h following each training session, mice were given 1 h access to a 20% sucrose solution in their homecages.
Valuation testing was conducted on the 2 d following training. On the valued day, mice were given ad libitum access to the sucrose solution for 1 h before a 5 min unreinforced session in the operant boxes. On the devalued day, mice were given ad libitum access to the food pellets previously earned by lever pressing for 1 h before a 5 min unreinforced session in the operant boxes.
Drugs
Picrotoxin (Sigma-Aldrich) was dissolved directly in aCSF. LY379268 and (RS)-3,5-DHPG (Tocris Bioscience) were prepared in distilled H2O and aCSF, respectively, and diluted to their final concentrations in aCSF. AM251 (Tocris Bioscience) was prepared in DMSO (Sigma-Aldrich) and diluted to its final concentration in aCSF containing β-cyclodextrin (6% w/v; TCI America) as a carrier. The final concentration of DMSO was <0.1%.
Statistics
Data are mean ± SEM. One-sample Student's t tests were used to compare a single genotype (RIM1fl/fl or Emx1::Cre;RIM1−/−) to a theoretical mean (e.g., 100% EPSC amplitude baseline). Two-tailed paired or unpaired Student's t tests were used to compare RIM1fl/fl or Emx1::Cre;RIM1−/− mice on a single measure (e.g., sEPSC frequency). A nonparametric Mann–Whitney test was used to compare RIM1fl/fl or Emx1::Cre;RIM1−/− mice breakpoints in PR schedule responding. Two-way mixed ANOVAs were used to assess main effects and interactions of genotype with repeated-measures variables (e.g., paired pulse ratios at multiple interpulse intervals in RIM1fl/fl or Emx1::Cre;RIM1−/− mice). Post hoc significance was determined using Sidak's multiple-comparisons test. A hierarchical linear mixed model was used to assess main effects and interactions of genotype with repeated-measures variables that contain nested variables (e.g., rotarod latencies across trials nested within days in RIM1fl/fl and Emx1::Cre;RIM1−/− mice). All pairwise comparisons of the model-derived estimated marginal means underwent Bonferroni corrections, and all reported p values reflect these corrections. Sex differences were analyzed for each category of behavior using one-sample Student's t tests and two-way ANOVAs, where appropriate. Effects were considered statistically significant at p < 0.05. Data were analyzed using MiniAnalysis, Clampfit, Microsoft Excel, GraphPad Prism 6, and PASW Statistics 18.
Results
Verification of conditional RIM1 KO
To assess the efficacy of our conditional RIM1 deletion strategy, we performed qRT-PCR on brain samples from RIM1fl/fl and Emx1::Cre;RIM1−/− mice. Consistent with the known expression patterns of Emx1::Cre mice (Gorski et al., 2002), RIM1 mRNA was reduced in motor cortex (M1 and M2) of Emx1::Cre;RIM1−/− mice relative to RIM1fl/fl littermates (Fig. 1A; n = 4, n = 4; t(6) = 8.421, p < 0.0005) but was indistinguishable from controls in dorsal striatum (n = 5, n = 5; t(8) = 1.327, p = 0.221) and thalamus (n = 4, n = 5; t(7) = 0.986, p = 0.357).
Conditional RIM1 KO reduced cortical gene expression and was without effect on viability, body weight, and brain weight. A, qRT-PCR measurement of RIM1 mRNA in cortex, dorsal striatum, and thalamus of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. *p < 0.05, different from RIM1fl/fl group. B, Offspring ratios of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. C, Average body weights of adult (P90–P100) RIM1fl/fl and Emx1::Cre;RIM1−/− mice. D, Average brain weights of adult (P90–P100) RIM1fl/fl and Emx1::Cre;RIM1−/− mice.
Normal viability, brain and body weight, and gross brain morphology in conditional RIM1 KO mice
Emx1::Cre;RIM1−/− mice were viable and fertile. Offspring of Emx1::Cre;RIM1−/− × RIM1fl/fl were born at the expected Mendelian ratios (Fig. 1B), and survived normally into adulthood. Body and brain weights of Emx1::Cre;RIM1−/− mice did not differ from those of littermate RIM1fl/fl controls (Fig. 1C: n = 14, n = 16; t(28) = 0.766, p = 0.45; Fig. 1D: n = 7, n = 8; t(13) = 1.17, p = 0.263). Morphological analyses revealed no gross structural abnormalities in brain architecture in the conditional RIM1 KO mice. Notably, unlike mice with Emx1 promoter-driven conditional KO of the presynaptic CB1R (Davis et al., 2018), Emx1::Cre;RIM1−/− mice showed normal axonal fasciculation of cortical fibers into and through striatum as revealed by immunostaining for neurofilament (Fig. 2A). Striatal CB1R and mu opioid receptor expression patterns were also normal in Emx1::Cre;RIM1−/− mice (Fig. 2B,C), indicating preserved striosome-matrix striatal compartments in mice lacking cortical RIM1. Mesoscale cortical and hippocampal structure was similarly preserved in these mice, as indicated by labeling with DAPI and immunostaining for neurofilament, parvalbumin, and CB1R (Fig. 2D–I).
Gross brain anatomy is unaltered in conditional RIM1 KO mice. A–I, Neurofilament, CB1R, mu opioid receptor (MOPR), parvalbumin (PV), and/or DAPI staining of striatum (A–D), motor cortex (E,F) and hippocampus (G–I) of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. Scale bars: A–D, G–I, 400 μm; E, F, 200 μm.
Impaired excitatory transmission at corticostriatal synapses in conditional RIM1 KO mice
Given our interest in probing the role of corticostriatal RIM1 in action learning and control, we examined synaptic transmission and plasticity in the conditional RIM1 KO mice at cortical inputs to the DLS, a brain region widely implicated in the development and expression of habitual behaviors (Yin et al., 2004; Balleine et al., 2007). Furthermore, despite evidence of diverse, synapse-specific roles for RIM1α in transmission and plasticity at several central projections (Castillo et al., 2002; Schoch et al., 2002; Chevaleyre et al., 2007; Kiyonaka et al., 2007; Fourcaudot et al., 2008; Kintscher et al., 2013), and detailed characterization of presynaptic forms of plasticity at cortical inputs to the DLS (Gerdeman et al., 2002; Mathur et al., 2011; Atwood et al., 2014), no studies to date have determined physiological roles for RIM1 at corticostriatal synapses in the DLS, or dorsal striatum more generally.
To assess the contribution of RIM1 to excitatory transmission in the DLS, we first examined spontaneous excitatory postsynaptic current (sEPSC) frequency, amplitude, and kinetics in DLS MSNs of Emx1::Cre;RIM1−/− and RIM1fl/fl mice. sEPSC frequency was significantly reduced in Emx1::Cre;RIM1−/− relative to RIM1fl/fl mice (Fig. 3A; n = 15, n = 15, t(28) = 2.207, p < 0.05). However, conditional deletion of RIM1 had no effect on sEPSC amplitude (Fig. 3B; t(28) = 1.624, p = 0.116). sEPSC rise time, decay time, and total charge transfer were indistinguishable between the two groups (data not shown). These findings suggest that loss of RIM1 in cortical excitatory cells causes impaired glutamate release in the DLS.
Impaired spontaneous excitatory synaptic transmission in DLS of conditional RIM1 KO mice. A, Average and cumulative probability plots of frequency and interevent interval of sEPSCs in DLS of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. B, Average and cumulative probability plots of amplitude of sEPSCs in DLS of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. *p < 0.05, different from RIM1fl/fl group. Calibration: 2 s, 20 pA.
To assess whether conditional RIM1 KO altered evoked neurotransmission, we measured EPSCs in DLS MSNs in response to electrical stimulation at the callosal-DLS border (Fig. 4A). Stimulation across a range of intensities revealed a significant reduction in evoked EPSC amplitude in Emx1::Cre;RIM1−/− mice relative to RIM1fl/fl littermates (Fig. 4B; n = 9, n = 11; two-way ANOVA: current × genotype interaction: F(10,180) = 2.557, p < 0.01), indicative of an impairment in evoked excitatory transmission.
Impaired evoked excitatory synaptic transmission in DLS of conditional RIM1 KO mice. A, Schematic diagram of stimulating and recording electrode placement in DLS. B, Input–output curves of electrically evoked excitatory currents in DLS of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. C, Paired pulse ratios of electrically evoked excitatory transmission in DLS of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. D, Activity-dependent depression during 10 Hz train of electrical stimulation at border of DLS in RIM1fl/fl and Emx1::Cre;RIM1−/− mice. *p < 0.05, different from RIM1fl/fl group. Calibration: C, 100 ms, 100 pA; D, 200 ms, 100 pA.
To test whether the changes in corticostriatal neurotransmission following RIM1 KO could be attributed to changes in presynaptic release probability, we assessed the ratio of excitatory synaptic responses to pairs of closely timed electrical pulses (paired pulse ratio) in DLS MSNs of Emx1::Cre;RIM1−/− and RIM1fl/fl mice. Paired pulse facilitation was enhanced in Emx1::Cre;RIM1−/− mice relative to RIM1fl/fl littermates (Fig. 4C; n = 15, n = 11; genotype: F(1,24) = 7.825, p < 0.05; IPI × genotype interaction: F(6,144) = 2.997, p < 0.01), particularly at short interpulse intervals (20 ms, p < 0.001; 40 ms, p < 0.05). Because paired pulse facilitation is normally inversely correlated with neurotransmitter release probability (Thomson, 2000), this finding suggests that transmitter release probability is lower at corticostriatal synapses lacking RIM1.
We also measured the adaptation of synaptic responses during short trains of moderate frequency stimulation (5–10 Hz) to further assess how transmission and use-dependent short-term plasticity are impaired following conditional deletion of RIM1. RIM1fl/fl and Emx1::Cre;RIM1−/− mice showed no difference in the depression of EPSC amplitude observed across a 2 s, 5 Hz train (data not shown); however, relative to RIM1fl/fl mice, Emx1::Cre;RIM1−/− mice showed a resistance to the depression observed across a 2 s, 10 Hz train (Fig. 4D; n = 13, n = 12; main effect of stimulus: F(19,437) = 34.50, p < 0.001; main effect of genotype: F(1,23) = 7.143, p < 0.05; stimulus × genotype interaction: F(19,437) = 1.832, p < 0.05). These findings are further consistent with a use-dependent impairment in synaptic transmission at corticostriatal synapses.
Spared forms of short- and long-term plasticity at corticostriatal synapses in conditional RIM1 KO mice
To assess whether corticostriatal RIM1 was necessary for presynaptic GPCR-mediated forms plasticity in the DLS (Lovinger, 2010), we first tested RIM1fl/fl and Emx1::Cre;RIM1−/− mice for DSE, an effect mediated by short-term endocannabinoid signaling at CB1Rs on striatal glutamatergic terminals (Shonesy et al., 2013). DSE was intact and of similar magnitude and duration in slices from conditional RIM1 KO and control mice (Fig. 5A; n = 10, n = 10; unpaired t test [on amplitude of first EPSC following depolarization]; t(18) = 0.154, p = 0.880). We then assessed whether endocannabinoid/CB1R-mediated LTD of excitatory transmission in the DLS, induced by treatment with DHPG (Kreitzer and Malenka, 2005), was intact in conditional RIM1 KO mice. RIM1fl/fl and Emx1::Cre;RIM1−/− mice showed comparable DHPG-induced LTD (Fig. 5B; n = 7, n = 8; unpaired t test [on average EPSC amplitude during final 10 min]; t(13) = 1.212, p = 0.247). Pretreatment with the CB1R antagonist, AM251, confirmed that DHPG-induced LTD in both groups of mice required activation of this presynaptic receptor (Fig. 5C; DHPG vs AM251+DHPG in Emx1::Cre;RIM1−/− mice; n = 6, n = 8; unpaired t test [on average EPSC amplitude during final 10 min]; t(12) = 2.445, p < 0.05; DHPG vs AM251+DHPG in RIM1fl/fl mice; n = 7, n = 7; unpaired t test; t(12) = 1.278, p = 0.128; DHPG alone in RIM1fl/fl mice; n = 7; one-sample t test, against baseline of 100; t(6) = 4.148, p < 0.001; AM251+DHPG in RIM1fl/fl mice; n = 7; one-sample t test; t(6) = 1.217, p = 0.269). Demonstrating that the lack of effect of conditional RIM1 deletion on long-term forms of plasticity extended beyond LTD mediated through CB1Rs, we found that LTD induced by the mGlu2/3 receptor agonist LY379268 was indistinguishable in RIM1fl/fl and Emx1::Cre;RIM1−/− mice (Fig. 5D; n = 4, n = 7; unpaired t test [on average EPSC amplitude during final 10 min]; t(9) = 1.181, p = 0.268). Together, these findings indicate that RIM1 within cortical inputs to the DLS is necessary for normal corticostriatal transmission, but not for several forms of short- and long-term presynaptic plasticity.
Intact short- and long-term depression of excitatory synaptic transmission in conditional RIM1 KO mice. A, DSE in DLS of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. Arrow indicates 10 s period of postsynaptic depolarization to 0 mV. B, DHPG-induced LTD of EPSCs in DLS of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. C, Effects of CB1R antagonist AM251 preincubation on DHPG-mediated LTD of EPSCs in DLS of RIM1fl/fl and Emx1::Cre;RIM1−/− mice. D, LTD of EPSCs in DLS of RIM1fl/fl and Emx1::Cre;RIM1−/− mice induced by bath application of mGlu2/3 receptor agonist, LY379268. Calibration: 10 ms, 100 pA.
Transiently enhanced homecage locomotion in conditional RIM1 KO mice
Given this synaptic phenotype in a brain region critical for action learning and control, we next evaluated conditional RIM1 KO mice for associated behavioral deficits. RIM1fl/fl and Emx1::Cre;RIM1−/− mice were first assessed for differences in basal locomotion in their homecages for 24 h/d for 3 d. Mice were placed into a homecage with fresh bedding 1 h into the dark period. Emx1::Cre;RIM1−/− mice showed enhanced locomotion during the first dark period relative to their RIM1fl/fl littermates, but this diminished to control levels by the first light period (Fig. 6A–C; n = 8, n = 13; hierarchical linear mixed model; main effect of day: F(2,95) = 29.153, p < 0.001; main effect of genotype: F(1,19) = 1.302, p = 0.268; light × day interaction: F(3,95) = 105.721, p < 0.0001; day × genotype interaction: F(2,95) = 1.306, p = 0.276; light × day × genotype interaction: F(3,95) = 3.701, p = 0.014; day 1, dark, RIM1fl/fl vs Emx1::Cre;RIM1−/−, p < 0.01).
Enhanced novelty-induced locomotion in conditional RIM1 KO mice. A–C, Homecage locomotion in RIM1fl/fl and Emx1::Cre;RIM1−/− across 3 d. Novelty-induced locomotion was observed in the first 10 h following exposure to the novel homecage (B). D, E, Novelty-induced locomotion in RIM1fl/fl and Emx1::Cre;RIM1−/− mice in 10 min (D) or 120 min bins (E) when fresh bedding and a new locomotor chamber were introduced daily. F, No novelty-induced hyperactivity was observed in Emx1::Cre mice. G, H, Novelty-induced locomotion in RIM1fl/fl and Emx1::Cre;RIM1−/− mice in 10 min (G) or 120 min bins (H) when cages were unchanged across 6 d of locomotor testing. *p < 0.05, different from RIM1fl/fl group. I, Prepulse inhibition in RIM1fl/fl and Emx1::Cre;RIM1−/− mice.
Novelty-induced hyperactivity in conditional RIM1 KO mice
To further characterize this hyperactivity phenotype, Emx1::Cre;RIM1−/− and RIM1fl/fl mice were tested for locomotion during the light period in a novel cage for 2 h/d for 6 d. One cohort of mice had their testing cages replaced with a clean cage daily, and a different cohort had their cages left unchanged for the 6 d period. Emx1::Cre;RIM1−/− mice whose cages were changed daily showed enhanced locomotion on day 1 that gradually converged with levels seen in RIM1fl/fl littermate controls by day 4 (Fig. 6D,E; n = 11, n = 12; two-way ANOVA; main effect of genotype: F(1,21) = 6.976, p < 0.05; day × genotype interaction: F(5,105) = 8.130, p < 0.001; RIM1fl/fl vs Emx1::Cre;RIM1−/−, day 1: p < 0.001; day 2: p < 0.05; day 3: p < 0.05). Importantly, these locomotor effects were not driven by Cre expression alone because Emx1::Cre mice showed normal locomotion in a novel cage (Fig. 6F; n = 9, n = 9; two-way ANOVA; main effect of genotype: F(1,16) = 0.011, p = 0.919). Emx1::Cre;RIM1−/− mice whose cages were left unchanged showed initially enhanced locomotion that reduced to control levels by day 2 (Fig. 6G,H; n = 10, n = 11; two-way ANOVA; main effect of genotype: F(1,19) = 0.863, p = 0.365; day × genotype interaction: F(5,95) = 9.086, p < 0.001; RIM1fl/fl vs Emx1::Cre;RIM1−/−, day 1: p < 0.05). Together, these findings suggest that Emx1::Cre;RIM1−/− mice show novelty-induced hyperactivity.
To assess sex differences in this hyperactivity, we combined data collected on locomotor test day 1 from both experimental cohorts (those whose cages were subsequently changed or unchanged). Male and female Emx1::Cre;RIM1−/− mice showed comparable novelty-induced hyperactivity relative to their RIM1fl/fl littermates (males, n = 11, n = 10; females, n = 10, n = 13; two-way ANOVA; main effect of sex: F(1,40) = 1.21, p = 0.278; main effect of genotype: F(1,40) = 20.57, p < 0.0001; sex × genotype interaction: F(1,40) = 2.841, p = 0.10). The novelty-induced hyperactivity seen here parallels that seen in global RIM1α KO mice (Blundell et al., 2010) and mice with conditional RIM1 KO driven by the kainate receptor subunit 1 promoter (Haws et al., 2012). Notably, Emx1::Cre;RIM1−/− mice did not recapitulate every phenotype seen in global RIM1α KO mice, including impairments in prepulse inhibition (Fig. 6I; n = 10, n = 10; two-way ANOVA; main effect of genotype: F(1,18) = 0.079, p = 0.781) (Blundell et al., 2010).
Impaired motor skill learning in conditional RIM1 KO mice
In a further phenotypic departure from global RIM1α mice (Powell et al., 2004), Emx1::Cre;RIM1−/− mice showed impaired performance on the accelerating rotarod relative to their RIM1fl/fl littermates (Fig. 7; n = 16, n = 15; hierarchical linear mixed model; main effect of day: F(3,319) = 39.759, p < 0.0001; main effect of genotype: F(1,29) = 4.411, p < 0.05; trial × day interaction, F(8,319) = 12.067, p < 0.0001). Male and female Emx1::Cre;RIM1−/− mice showed comparable deficits in overall rotarod performance relative to their RIM1fl/fl littermates (males, n = 10, n = 10; females, n = 6, n = 5; two-way ANOVA; main effect of sex: F(1,27) = 2.616, p = 0.117; main effect of genotype: F(1,27) = 4.617, p < 0.05; sex × genotype interaction: F(1,27) = 0.3521, p = 0.558). These findings suggest that cortical RIM1 may be particularly important for aspects of action learning, including the refinement of actions into motor skills.
Impaired motor learning in conditional RIM1 KO mice. Latency to fall from the accelerating rotarod in RIM1fl/fl and Emx1::Cre;RIM1−/− mice. *p < 0.05, different from RIM1fl/fl group.
Heightened responding on FR and PR schedules in conditional RIM1 KO mice
We next assessed whether RIM1-dependent processes in forebrain excitatory neurons were required for normal instrumental learning. RIM1fl/fl and Emx1::Cre;RIM1−/− mice were food restricted and trained to lever press for food pellets on escalating FR schedules of reinforcement. Both groups pressed the “active” lever sufficient numbers of times to receive all allotted food pellets each day (Fig. 8A). However, Emx1::Cre;RIM1−/− mice showed heightened rates of responding for food during the higher FR sessions (Fig. 8B; n = 14, n = 12; two-way ANOVA; main effect of genotype: F(1,24) = 6.063, p < 0.05; day × genotype interaction: F(8,192) = 4.519, p < 0.001; RIM1fl/fl vs Emx1::Cre;RIM1−/−, day 8 and 9: p < 0.05). Emx1::Cre;RIM1−/− mice also showed temporarily heightened numbers of head entries into the food dispenser early in instrumental training (Fig. 8C; n = 11, n = 11; two-way ANOVA; day × genotype interaction: F(8,160) = 2.076, p < 0.05; RIM1fl/fl vs Emx1::Cre;RIM1−/−, day 2: p < 0.05). These behavioral differences were not attributable to differences in body weight responses to food restriction (Fig. 8D) or to general differences in propensity for consumption, as RIM1fl/fl and Emx1::Cre;RIM1−/− mice ate comparable amounts of food during subsequent post testing ad libitum access to food pellets in their homecages (RIM1fl/fl, 6.27 ± 0.34 g vs Emx1::Cre;RIM1−/−, 6.66 ± 0.23 g). Heightened responding also did not simply reflect general hyperactivity as average daily inactive lever pressing rates were statistically indistinguishable across groups (two-way ANOVA; main effect of genotype: F(1,24) = 0.882, p = 0.357) and press counts were consistently low across all training sessions (RIM1fl/fl, 5.18 ± 1.50 presses vs Emx1::Cre;RIM1−/−, 7.31 ± 1.83 presses).
Enhanced instrumental responding for food in conditional RIM1 KO mice. A, B, Active lever presses (A), active lever presses/min (B), and head entries into the food receptacle (C) during FR schedule training in RIM1fl/fl and Emx1::Cre;RIM1−/− mice. D, Body weight response to food restriction across training. E, F, Active and inactive lever presses (E) and breakpoints achieved (F) during PR testing in RIM1fl/fl and Emx1::Cre;RIM1−/− mice. *p < 0.05, different from RIM1fl/fl group.
Mice subsequently underwent a PR test in which they were required to press the active lever a progressively increasing number of times to receive a food pellet. Relative to littermate RIM1fl/fl controls, Emx1::Cre;RIM1−/− mice showed heightened active lever pressing for food during this test (Fig. 8E; unpaired t test; t(24) = 3.295, p < 0.005), and obtained higher breakpoints in their responding (Fig. 8F; Mann–Whitney test; U = 26, p < 0.005). Male and female Emx1::Cre;RIM1−/− mice showed comparably enhanced active lever pressing during the PR test relative to their RIM1fl/fl littermates (males, n = 7, n = 8; females, n = 5, n = 6; two-way ANOVA; main effect of sex: F(1,22) = 1.13, p = 0.299; main effect of genotype: F(1,22) = 6.505, p < 0.05; sex × genotype interaction: F(1,22) = 0.020, p = 0.888). Responding on the inactive lever during the PR test was also elevated in Emx1::Cre;RIM1−/− relative to RIM1fl/fl mice (t(24) = 2.703, p < 0.05), but inactive lever responding was consistently low relative to active lever responding in both groups (8.06 ± 3.53% and 8.67 ± 1.38% in RIM1fl/fl and Emx1::Cre;RIM1−/− mice, respectively). These results indicate that conditional RIM1 KO mice show heightened instrumental responding for food, most prominently under higher-effort conditions.
Enhanced responding on RI schedules and resistance to developing habitual actions in conditional RIM1 KO mice
Action strategies used during instrumental learning and responding have been shown to exist on a gradient between outcome-dependent/goal-directed and outcome-independent/habitual. To delineate the action strategy used by Emx1::Cre;RIM1−/− mice during food responding, we trained a cohort of Emx1::Cre;RIM1−/− and RIM1fl/fl mice on escalating RI schedules of reinforcement known to bias learning toward the use of habitual response strategies (Dickinson, 1985; Gremel and Costa, 2013). RI schedules promote habitual responding by weakening the contingency and contiguity between the action (lever press) and outcome (food pellet) (Dickinson, 1985; Derusso et al., 2010). To assess their propensity to use a goal-directed or habitual action strategy, mice underwent sensory-specific satiety outcome devaluation testing over two consecutive days, termed valued and devalued days. Mice were permitted 1 h ad libitum consumption of either a familiar sucrose solution on the “valued” day, and food pellets previously earned by lever pressing on the “devalued” day, each followed by a 5 min period of nonrewarded lever pressing. A larger reduction in lever pressing in the devalued versus the valued state is indicative of greater goal-directed action control; indistinguishable lever pressing in the devalued and valued states is indicative of habitual control (Dickinson, 1985; Gremel et al., 2016).
Relative to RIM1fl/fl mice, Emx1::Cre;RIM1−/− mice showed heightened levels of responding (Fig. 9A; n = 16, n = 14; two-way ANOVA; day × genotype interaction: F(9,252) = 2.726, p < 0.005; RIM1fl/fl vs Emx1::Cre;RIM1−/−, day 9 and 10, p < 0.05) and rates of responding for food during high RI training (Fig. 9B; day × genotype interaction: F(9,252) = 2.787, p < 0.005; RIM1fl/fl vs Emx1::Cre;RIM1−/−, day 10, p < 0.05). As seen in the cohort of mice trained on escalating FR schedules, Emx1::Cre;RIM1−/− mice showed transiently elevated numbers of head entries early in instrumental training (Fig. 9C; n = 15, n = 10; two-way ANOVA; day × genotype interaction: F(9,207) = 2.016, p < 0.05; RIM1fl/fl vs Emx1::Cre;RIM1−/−, day 1: p < 0.05). Again, this heightened responding was not attributable to genotype differences in body weight responses to food restriction (data not shown).
Impaired habitual responding for food in conditional RIM1 KO mice. A, B, Lever presses (A), lever presses/min (B), and head entries into the food receptacle (C) during FR and RI schedule training in RIM1fl/fl and Emx1::Cre;RIM1−/− mice. D, Lever presses during valuation testing in RIM1fl/fl and Emx1::Cre;RIM1−/− mice. *p < 0.05, different from valued condition. E, Normalized lever presses during valuation testing in RIM1fl/fl and Emx1::Cre;RIM1−/− mice. *p < 0.05, different from 0.50. F, Revaluation index in response to valuation testing in RIM1fl/fl and Emx1::Cre;RIM1−/− mice. *p < 0.05, different from RIM1fl/fl group. G, Correlation between pellet consumption during ad libitum feeding and revaluation index in RIM1fl/fl mice (gray line), Emx1::Cre;RIM1−/− mice (red line), and both combined (black line). All correlations were nonsignificant.
Devaluation testing revealed that RIM1fl/fl mice responded comparably on the valued and devalued days (Fig. 9D; two-way ANOVA; valuation × genotype interaction: F(1,28) = 5.862, p < 0.05; valued vs devalued, RIM1f/f: p = 0.428), consistent with the expected use of a habitual action strategy following RI schedule training. In contrast, Emx1::Cre;RIM1−/− mice showed reduced lever pressing on the devalued day relative to the valued day (Fig. 9D; ANOVA [as above]; valued vs devalued, Emx1::Cre;RIM1−/−: p < 0.05). Analysis of normalized lever pressing during valued and devalued conditions revealed that Emx1::Cre;RIM1−/−, but not RIM1fl/fl, mice were significantly sensitive to devaluation (Fig. 9E; one-sample t test, against “no devaluation” point of 0.5; Emx1::Cre;RIM1−/−: t(15) = 7.293, p < 0.0001; RIM1fl/fl: t(13) = 2.119, p = 0.054). Notably, devaluation in RIM1fl/fl mice narrowly avoided significance, indicative of modest expression of a habitual action strategy in these mice. Significant sensitivity to devaluation was seen in male and female Emx1::Cre;RIM1−/− mice (males, n = 7: t(6) = 7.901, p < 0.0005; females, n = 9: t(8) = 4.136, p < 0.005), but not male and female RIM1fl/fl mice (males, n = 6: t(5) = 2.175, p = 0.082; females, n = 8: t(7) = 1.077, p = 0.317). Furthermore, using a “revaluation index” ([Valued − Devalued Lever Presses]/Total Lever Presses) as a measure of goal-directness (Gremel and Costa, 2013; Gremel et al., 2016), we found that Emx1::Cre;RIM1−/− mice showed greater goal-directedness than their RIM1fl/fl littermates (Fig. 9F; t(28) = 2.774, p < 0.01). Notably, despite consuming comparable amounts of sucrose as RIM1fl/fl mice during the 1 h ad libitum feeding session preceding valuation testing (0.92 ± 0.16 vs 0.79 ± 0.13 g, t(28) = 0.612, p = 0.546), Emx1::Cre;RIM1−/− mice consumed more food pellets than RIM1fl/fl mice during the corresponding prefeeding session (0.69 ± 0.10 vs 0.40 ± 0.06 g, t(28) = 2.459, p = 0.020). However, no correlation was observed between pellet consumption and revaluation index in either group of mice (Fig. 9G), suggesting that variation in pellet consumption did not meaningfully impact devaluation of responding for pellets. Together, these data indicate that RIM1-dependent processes in cortical excitatory neurons may help to constrain goal-directed action and contribute to the learning and/or use of habits.
Discussion
Here we combine genetic, electrophysiological, and behavioral approaches to assess presynaptic processes within corticostriatal projections that subserve normal action learning and control. We show that conditional deletion of RIM1 presynaptic scaffold proteins from corticostriatal projection neurons (and other forebrain excitatory neurons) impairs excitatory transmission in the DLS but spares many forms of presynaptic GPCR-mediated short- and long-term striatal plasticity. Conditional RIM1 KO mice show elevated locomotor responses to novelty and impaired motor learning on the accelerating rotarod. They further show abnormally robust goal-directed responding for food. These findings implicate RIM1-dependent processes in corticostriatal transmission and in the learning and control of motivated behavior.
The selective effect of RIM1 deletion on corticostriatal transmission (not CB1R- or mGlu2/3 receptor-mediated plasticity), although unexpected, is consistent with selective roles for RIM1-dependent processes in transmission or plasticity at other central synapses (Castillo et al., 2002; Schoch et al., 2002; Lonart et al., 2003; Abrahamsson et al., 2017). Global deletion of RIM1α has been shown to attenuate DHPG-induced LTD of excitatory transmission in the nucleus accumbens (Grueter et al., 2010); however, unlike in the present study, the terminal population driving the effect of this single isoform deletion is unclear. Furthermore, plasticity mechanisms are well documented to be differentiable in dorsal and ventral striatum subregions (Lovinger, 2010; Russo et al., 2010). Although it is possible that the lack of effect of RIM1 deletion on GPCR-mediated forms of plasticity resulted from recruitment of DLS inputs derived from populations of non-Emx1::Cre (and therefore RIM1-expressing) neurons, we believe this interpretation to be unlikely. First, Cre expression in Emx1::Cre mice drives recombination in the vast majority (∼88%) of cortical pyramidal neurons (Gorski et al., 2002). Indeed, we observed robust Cre-mediated knockdown of RIM1 mRNA in M1 and M2 motor cortices that densely innervate the DLS (Fig. 1A). Second, the other principle source of excitatory inputs to the DLS is the thalamus (Hunnicutt et al., 2014, 2016), but the synaptic properties (e.g., paired pulse ratios) of the mixed striatal inputs stimulated in the present study more closely mimicked those of cortical rather than thalamic origin (Ding et al., 2008). Moreover, thalamostriatal inputs exhibit minimal CB1R expression and CB1R-mediated forms of plasticity (Wu et al., 2015; for mGlu2 receptor-mediated thalamostriatal plasticity, see Johnson et al., 2017). Therefore, effects of RIM1 deletion on presynaptic GPCR-mediated forms of corticostriatal plasticity are unlikely to be blunted by recruitment of RIM1-expressing DLS inputs.
Conditional RIM1 deletion from cortical pyramidal neurons favored, and intensified, the use of goal-directed action strategies during instrumental responding for food. These findings, in conjunction with the observed impairment in excitatory transmission in the DLS, support the view that the normal learning and use of habitual action strategies depend on DLS engagement (Yin et al., 2004; Dias-Ferreira et al., 2009; Thorn et al., 2010; Smith and Graybiel, 2013). In this way, weakened excitatory drive of the DLS may bias animals toward the adoption of goal-directed strategies that are more reliant on dorsomedial striatum-based circuitry (Yin et al., 2005b; Gremel and Costa, 2013; Hart et al., 2018). Of course, in the present study, RIM1 was deleted from widespread cortical structures, including from associative and limbic cortical regions that innervate the dorsomedial striatum and ventral striatum (McGeorge and Faull, 1989; Hintiryan et al., 2016). Deletion-induced impairments in excitatory transmission may also occur in these and other regions and contribute to the array of behavioral phenotypes observed in these animals. It remains notable, however, that widespread cortical RIM1 deletion led to a net bias toward goal-directed actions. Future work mapping the synaptic phenotypes seen across striatal subregions following cortical RIM1 deletion would help to explain this net bias.
A major impetus for targeting electrophysiology recordings to the DLS was to explore the relationship between prominent forms of DLS presynaptic plasticity and action control. Indeed, prior work has proposed that CB1R-mediated LTD in the DLS plays an important role in habit expression because the two phenomena are impaired following the development of tolerance to a CB1R agonist (Δ9-THC) and are rescued upon restoration of endocannabinoid signaling in the DLS (Nazzaro et al., 2012). The sparing of CB1R-mediated LTD in the DLS following RIM1 deletion did not permit direct testing of this hypothesis. However, it is probable that impairment of synaptic mechanisms other than LTD can interfere with habit learning even when LTD is normal. Indeed, multiple mechanisms are likely to contribute to over-reliance or under-reliance on DLS-based circuits that can promote or impair habit learning and use (Corbit et al., 2014; Barker et al., 2015; Gremel et al., 2016; O'Hare et al., 2017). Variation in CB1R-mediated plasticity in the DLS is only one such mechanism.
Motor skill learning was also found to be impaired following cortex-wide conditional RIM1 deletion. Global RIM1α KO mice showed no such impairment across a single-day rotarod training paradigm (Powell et al., 2004), but distinct methodologies (e.g., single isoform deletion, developmental compensation, extent of training, etc.) may account for these distinct behavioral phenotypes. Our evidence of impaired motor learning, paired with a deficit in DLS excitatory transmission, supports the view that like habitual responding the learning of stereotyped motor actions depends on DLS engagement (Miyachi et al., 2002; Floyer-Lea and Matthews, 2005; Yin et al., 2009). Indeed, recent work from our group showed that sensorimotor cortical inputs to DLS are robustly engaged from the outset of action learning, and incrementally disengage as actions become refined and stereotyped (Kupferschmidt et al., 2017). Findings from our group further show that conditional deletion of CB1R from cortical pyramidal neurons (in Emx1::Cre;CB1R−/− mice) has no effect on rotarod learning (D.A.K. and D.M.L., unpublished observations). Together, these findings suggest that basal glutamate release at corticostriatal synapses (and associated postsynaptic plasticity), and not presynaptic GPCR modulation of this release, is particularly critical for motor skill learning. As above, synaptic impairments following cortical RIM1 deletion undoubtedly extend beyond those presently observed in the DLS and may contribute to the observed motor skill learning deficit.
Last, mice with the conditional RIM1 deletion showed a pronounced enhancement of the locomotor response to a novel environment. A similar enhancement was seen in global RIM1α mice (Powell et al., 2004; Blundell et al., 2010), suggesting that RIM1α-dependent processes in forebrain excitatory neurons contribute to this behavioral phenotype. Haws et al. (2012) showed a modest increase in novelty-induced locomotion in mice with RIM1 conditionally deleted from nearly all hippocampal CA3 pyramidal neurons plus sparse populations in dentate gyrus, cerebellum, cortex, and thalamus. Therefore, the RIM1-lacking neurons driving the novelty response in the present experiments may be primarily localized to the hippocampus and likely differ from those driving the impaired skill learning and biased instrumental action control. Notably, however, the neural substrates underlying novelty responses and appetitive motivational processes, like those that support instrumental responding under high-effort conditions, may show greater convergence, including a shared reliance on ventral striatal brain systems (Bardo et al., 1996; Legault and Wise, 2001; Salamone et al., 2003, 2016; Lisman and Grace, 2005). Therefore, neuroadaptations in such systems may underlie the heightened novelty-induced locomotion and PR food responding seen here in conditional RIM1 KO mice.
For reasons both technical and historical, study of the neural basis of action control has focused heavily on learning-related postsynaptic adaptations in basal ganglia circuits. However, the molecular players and physiological processes at work within targeted populations of presynaptic terminals have become accessible to modern neuroscientists, and their contributions to normal and disordered action control are becoming increasingly recognized. Here we advance this effort by revealing a selective role for presynaptic RIM1 in neurotransmitter release at a key input structure of the basal ganglia and provide evidence that RIM1-dependent processes help to promote the refinement of skilled actions, constrain goal-directed behaviors, and contribute to the learning and use of habits.
Footnotes
This work was supported by National Institute on Alcohol Abuse and Alcoholism Division of Intramural Clinical and Biological Research ZIA AA000416, Natural Sciences and Engineering Research Council postdoctoral fellowship 438487-13 to D.A.K., and National Institute of General Medical Sciences PRAT Fellowship and K99 AA025403 to K.A.J. We thank all members of the D.M.L. laboratory for helpful discussions about the present experiments and manuscript; Guoxiang Luo for help with mouse genotyping; and the Fishers Lane Animal Care staff for excellent animal husbandry.
The authors declare no competing financial interests.
- Correspondence should be addressed to David M. Lovinger at lovindav{at}mail.nih.gov
