Dynamic regulation of corticostriatal glutamatergic synaptic expression during reversal learning in male mice

Abstract

Behavioral flexibility, one of the core executive functions of the brain, has been shown to be an essential skill for survival across species. Corticostriatal circuits play a critical role in mediating behavioral flexibility. The molecular mechanisms underlying these processes are still unclear. Here, we measured how synaptic glutamatergic α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) and N-methyl-D-aspartic acid receptor (NMDAR) expression dynamically changed during specific stages of learning and reversal. Following training to well-established stages of discrimination and reversal learning on a touchscreen visual task, lateral orbitofrontal cortex (OFC), dorsal striatum (dS) as well as medial prefrontal cortex (mPFC), basolateral amygdala (BLA) and piriform cortex (Pir) were micro dissected from male mouse brain and the expression of glutamatergic receptor subunits in the synaptic fraction were measured via immunoblotting. We found that the GluN2B subunit of NMDAR in the OFC remained stable during initial discrimination learning but significantly increased in the synaptic fraction during mid-reversal stages, the period during which the OFC has been shown to play a critical role in updating outcome expectancies. In contrast, both GluA1 and GluA2 subunits of the AMPAR significantly increased in the dS synaptic fraction as new associations were learned late in reversal. Expression of NMDAR and AMPAR subunits did not significantly differ across learning stages in any other brain region. Together, these findings further support the involvement of OFC-dS circuits in moderating well-learned associations and flexible behavior and suggest that dynamic synaptic expression of NMDAR and AMPAR in these circuits may play a role in mediating efficient learning during discrimination and the ability to update previously learned associations as environmental contingencies change.

Introduction

The ability to efficiently learn associations that lead to positive outcomes and flexibly change those associations in response to new environmental conditions is essential for an organism’s survival. Associative learning and flexible action are sub-served by neural circuitry that is highly conserved across species with topographically-organized loops originating in cortical subregions that project through structures in the basal ganglia before returning to the cortex (Schilman et al., 2008; Voorn et al., 2004). It is well established that circuits connecting the lateral orbitofrontal cortex (lOFC) and dorsal striatum (dS) mediate the acquisition of well-learned associations and the ability to flexibly change these associations when required (Schilman et al., 2008; Voorn et al., 2004). Studies in primates and rodents using an array of associative stimuli and responses have demonstrated that the OFC is not essential for initial choice learning, but is functionally necessary for optimal behavioral flexibility (Chudasama & Robbins, 2003; Dias et al., 1996; Moore et al., 2009; Rudebeck & Murray, 2008). In contrast, the dorsal striatum (dS) mediates choice behavior by balancing bottom up stimulus preference and action-outcome learning (Featherstone & McDonald, 2004; Palencia & Ragozzino, 2005; Yin et al., 2004). Reversal learning tasks recruit both of these systems, by requiring a subject to learn a choice or response pattern that leads to reward and then shift to the previously unrewarded choice when reward contingencies are reversed. We have previously demonstrated that mouse touchscreen reversal recruits these corticostriatal circuits, as dS is selectively recruited during discrimination learning and OFC is activated when flexible behavior is highly taxed during early reversal (Brigman et al., 2013a; Graybeal et al., 2011; Marquardt et al., 2017). However, the specific mechanism by which OFC-dS cortico-striatal circuits balance efficient learning and flexible behavior is not fully understood.

It is well established that both learning and reversal require the induction of synaptic plasticity via the activation of both α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPAR) and N-methyl-_D-aspartate receptors (NMDAR). AMPAR play a critical role in post-synaptic depolarization, and are required to release the voltage-sensitive magnesium block and allow glutamate binding to active NMDAR. NMDAR are heteromeric complexes composed of an obligatory GluN1 subunit together with one or more GluN2 subunits (GluN2A-2D) which confer distinct physiological and molecular properties to the receptor. The predominant subunits in the adult forebrain, GluN2A and GluN2B, have been widely studied in order to understand the role of NMDAR in the synaptic plasticity required for learning and flexible choice. The ratio of GluN2A to GluN2B-containing NMDAR is hypothesized to play a crucial role in the plasticity required to form, and flexibly alter new associations. Multiple learning paradigms increase the GluN2A/GluN2B ratio and increase the threshold for inducing plasticity (Holehonnur et al., 2016; Philpot et al., 2001; Quinlan et al., 2004). GluN2A and GluN2B expression has been directly implicated in efficient reversal learning. Loss of GluN2A has been shown to globally impair associative learning (Brigman et al., 2008), while loss of GluN2B function spares associative learning but impairs behavioral flexibility (Brigman et al., 2013b). In addition, loss of GluN2B also significantly alters spike-firing activity and reduces and delays coherence of OFC pyramidal neurons during early reversal (Marquardt et al., 2019).

The role of AMPAR subunits in various forms of experience-dependent plasticity is less well-studied outside the area of substance use disorders. AMPAR are heterotetramers, composed of 4 subunits GluA1–4 (Lu et al., 2009; Malenka, 2003), with GluA1 and GluA2 predominantly expressed in the hippocampus and the cortex (Lu et al., 2009). The trafficking of the AMPAR in and out of the synapse plays a central role in LTP and LTD respectively (Malenka, 2003). AMPAR subunits, specifically the subunit specific trafficking is critical for spatial reversal learning. Impaired GluA1 endocytosis in MAPK-activated protein kinases 2 and 3 (MK2/3) knock out mice showed deficits in hippocampal dependent spatial reversal task(Eales et al., 2014). As with NDMAR, different subunits confer different electrophysiological property as GluA2 containing AMPAR are calcium impermeable whereas GluA2 lacking are calcium permeable (Keifer & Zheng, 2010; Malenka, 2003).

While there is strong evidence from genetic and pharmacological studies that NMDAR and AMPAR subunits are involved in specific aspects of learning and reversal, how subunit synaptic expression may be dynamically expressed or altered during these behaviors is not well understood. In order to address this issue, we tested cohorts of mice to previously validated learning stages of discrimination and reversal known to recruit activation of corticostriatal circuits specifically. We then examined the expression of AMPA and NMDA receptor subunits in the synaptic fraction in brain regions known to be involved in regulation of learning and reversal: the lateral OFC, dS (both medial and lateral aspects) as well as brain regions known to be involved in behavioral flexibility including the medial prefrontal cortex (infralimbic/prelimbic; mPFC) basolateral amygdala (BLA) and a control region (piriform cortex; Pir). By examining how NMDA and AMPA receptor subunits are expressed in the synapse we sought to understand how synaptic levels of both NMDAR and AMPA subunits are changed across states of learning and reversal behavior during the touchscreen task. Our results suggest that visual discrimination reversal learning is associated with significant alterations in synaptic expression of NMDAR and AMPAR subunits in corticostriatal subregions.

Methods

Male C57BL/6J mice from Jackson laboratories were used for all experiments. 6–7 week-old male mice were housed in pairs in cages in a temperature- and humidity-controlled environment under a reverse 12-hour light/dark cycle (lights off 8:00 hours, on at 20:00 hours). Mice were given one-week period for acclimation to the new environment and were handled on the last two days by the investigator to ensure familiarity. Following acclimation, all mice weighing above 20gms were started on food restriction regimen to enable gradual weight reduction over a period of 1–2 weeks to reach approximately 85% of their free feeding weight. Upon reaching the target weight, mice were started on pretraining. All behavioral experiments were conducted during the dark phase and all experimental procedures were performed in accordance with the National Institutes of Health Guide for Care and Use of Laboratory Animals and were approved by the University of New Mexico Health Sciences Center Institutional Animal Care and Use Committee.

Pretraining and Behavioral testing:

Mice were 7–8 weeks of age at onset of pretraining. All mice were first habituated to receiving reward in the operant chambers by placing them.in the chambers for 30 minutes during which time 30μl of liquid reward (strawberry milk) was delivered 30 secs following each reward retrieval. Mice retrieving 10 rewards in under 15 minutes were moved to the next stage of pretraining, lever press training. Here, mice had to learn to initiate each trial by pressing an ultra-sensitive lever in the operant chamber. Mice, completing 30 trials under 30 minutes moved on to touch training where the lever press was followed by presentation of various white shapes in 1 of the 2 response windows. The appearance of the stimulus in either window during various trials was pseudorandomized and the stimulus remained on the screen until a response was made. Touch of a stimulus was immediately followed by delivery of reward cued by the light in the reward chamber turning on and presentation of tone for 1 second. Touches on the blank response window had no response. Mice initiating, touching, and retrieving 30 rewards within 30 minutes were moved to punish training. Punish training was identical to touch training except that responses at a blank window during stimulus presentation resulted in a 15-second time-out (signaled by illumination of the house light) to discourage indiscriminate screen responding. Errors on this stage were followed by correction trials, in which the same stimulus and left/right position were presented until a correct response was made. Mice making ≥75% (excluding correction trials) of their responses at a stimulus-containing window over a 30-trial session were moved onto discrimination.

Discrimination and Reversal (DR) Learning:

Mice were tested as previously described (Marquardt et al., 2014). Briefly, for discrimination learning, 2 novel approximately equiluminescent stimuli were presented in a spatially pseudorandomized manner over 30-trial sessions (5-second inter-trial interval, 60 minutes maximum per session). Responses at one stimulus resulted in reward while responses at the other stimulus resulted in a 15-second time-out (signaled by illumination of the house light) and were followed by a correction trial. Designation of initially rewarded stimulus was randomized across mice. Stimuli remained on screen until a response was made. As during pretraining, errors on first presentation trials were followed by correction trials which continued until a correct response was made or the session ended. Mice were trained to a criterion of ≥85% correct responding (excluding correction trials) over 2 consecutive sessions.

Reversal training began on the session following discrimination criterion. Here, the stimulus-reward contingencies for each mouse was simply switched wherein the previously correct stimulus resulting in reward is now the incorrect stimulus leading to 15-second time out with illumination of house light and vice versa. Mice were trained on reversal to a criterion of ≥85% correct responding (excluding correction trials) over 2 consecutive sessions.

Discrimination and reversal learning were subdivided into five distinct stages: Disc^early where the mice were performing at ~50% correct responses; Disc^late where mice had learned the correct association to ~85% or more correct performance over two consecutive days; Rev^early, the first day of reversal session following during which mice were highly perseverative; Rev^mid where the mice had ceased perseverating and were performing the new association at ~50% correct responding; Rev^late where the mice had learned the new stimulus reward associations to 85% correct responses over 2 consecutive days. 2 hours following attaining the criterion for respective stages, mice were sacrificed via cervical dislocation and rapid decapitation and brains were extracted and snap frozen using 2% iso-butenol in dry ice. Brains were stored in −80°c until they were ready for further processing (Fig 1A).

Figure 1. Timeline and performance through stages of touchscreen-based discrimination and reversal learning.

A. Experimental timeline showing the sequence of behavioral testing and analysis of protein expression. Following attaining criteria for each behavioural stage (Disc^early, Disc^late, Rev^early, Rev^late), brains were extracted, regions of interest were micropunched. Subcellular fractionization was done to isolate to synaptic fraction, where the expression of NMDAR and AMPAR subunits were quantified using immunoblotting. B. Percent correct was highest during the well learned stages (Disc^late, Rev^late) when compared to all other learning stages and lowest during the Rev^early stage. C. Total errors made was significantly lower during the well learned stages (Disc^late,Rev^late). D. Total correction errors were significantly higher during the early perseverative phase pf reversal (Rev^early). E. Stimulus reaction time was significantly higher during the early stages of learning and reversal. F. No significant differences were seen in reward reaction time across the stages. Data are group mean ± SEM.  *= p<0.05 significantly diff from Rev^early, $= p<0.05 significantly diff from Rev^late, #= p<0.05, significantly diff from Disc^late.

Subcellular fractionation:

In order to obtain sufficient synaptic membrane fraction for immunoblotting, micro punches of the target regions (lOFC, dS, mPFC, BLA and Pir) from 4 brains were pooled resulting in an n=4 per behavioral stage. Subcellular fractionation was performed as previously described (Goebel-Goody, S.M., et al., 2009) to isolate the synaptic membrane fraction bound to post synaptic density. Briefly, tissue was homogenized in homogenization buffer (Tris 20mM, pH 7.4, 1.0M EDTA, 320mM sucrose, 20mM sodium pyrophosphate, 10mMsodium fluoride, 20mM β-glycerophosphate, 0.2mM sodium orthovanadate, and protease inhibitor cocktail (Sigma, St. Louis, MO)) and centrifuged at 1000 × g for 10 minutes at 4°C. The supernatant was isolated and the above process was repeated in order to remove the nuclei and large debris. The isolated supernatant was then centrifuged (15,0000 × g for 30 minutes at 4°C) which resulted in separation of the crude synaptic fraction (P1) which contained both synaptic and extra synaptic membrane. P1 fraction was then hypo-osmotically lysed using protease inhibitor cocktail(sigma) in ice-cold deionized water and sonicated. Subsequent addition of HEPES-NaOH buffer (pH 7.4) and centrifugation 15,0000 × g for 30 minutes at 4°C resulted in supernatant which is the plasma membrane enriched synaptosomal membrane fraction. This fraction was resuspended in Triton X-100 buffer (pH 7.4) containing 10mMTris, 5mM NaF, 1mM EDTA, 0.5mM EGTA, 0.2mM sodium orthovanadate, protease and phosphatase inhibitors (Sigma protease inhibitor cocktail and phosphatase inhibitor cocktails 1 and 2, 10 μM NaF, 1% Triton-X 100, 25 mM Tris, pH 6.8), gently vortexed and incubated at 4°C for 30 minutes. Triton X-100, a non-ionic detergent solubilizes all membrane components except synaptic junctional membrane complexes tightly bound to post synaptic density (PSD) which are resistant to detergent treatment(Cotman et al., 1971). Treatment with nonionic detergent such as Triton X-100 has been shown to reliably separate the synaptic membrane fraction bound to PSD from the extra synaptic membrane (Brady et al., 2013; Cotman et al., 1971; Goebel-Goody et al., 2009). Following 30 minutes incubation in Triton X-100 buffer, samples were centrifuged (15,0000 × g for 30 minutes at 4°C) to yield triton X-100 soluble supernatant, the extra-synaptic fraction (Triton-soluble, TxS) and triton x-100 insoluble pellet, the synaptic fraction bound to PSD(Triton-insoluble particulate, TxP). The resulting supernatant, extra-synaptic fraction (TxS) was then isolated from the insoluble pellet which is the synaptic fraction bound to PSD (TxP). The pellet (TxP) was then re-suspended in homogenization buffer containing sodium dodecyl sulfate to a final concentration of 1%, and stored in −80°C until further processing. The TxP fraction was tested for the presence of PSD associated synaptic proteins such as, HOMER 1a and HOMER 1b/c (Fig. S1) using immunoblotting to confirm that TxP fraction contained PSD associated synaptic membrane components. Total protein levels in each TxP fraction were measured using Qubit 4 Fluorometer (Thermo Fisher Scientific Inc., Waltham, MA).

Quantification of NMDAR Subunits:

To examine changes in NMDAR subunit expression during discrimination learning and reversal, GluN1, GluN2A and GluN2B levels were measured in the synaptic fractions bound to PSD in all five target regions via immunoblotting. First, the appropriate amount of protein to be loaded for each target region was determined by performing optimization blot for each region of interest (data not shown). This was done to improve the accuracy of quantitative protein analysis. Briefly, samples were loaded in increasing concentration, subjected to immunoblotting (detailed description below). Following incubation with primary and secondary fluorescence antibodies against the target protein, the resultant fluorescence intensity elicited was calculated via the Odyssey imaging system for each concentration. The protein concentration where the fluorescence intensity showed linear relationship is said to be in linear range of detection and this concentration was used to calculate the appropriate amount of sample to be loaded for each region.

The calculated sample amounts were treated with reducing agent (NuPage sample reducing agent; #NP0004, Thermo Fisher Scientific) which enabled the denaturing of the proteins, and diluted with loading buffer (NuPage sample buffer; #NP0007, Thermo Fisher Scientific) containing lithium dodecyl sulfate, pH 8.4, enabling the maximum activity of the reducing agent and tracking dyes such as Coomassie G250 and Phenol Red which helps in monitoring the progress of electrophoresis. 20μl of prepared sample mixture along with the Li-Cor molecular weight ladder was loaded in the 15 well 4–12% Bis-Tris polyacrylamide gel (#NP0336, Thermo Fisher Scientific) and electrophoresis was performed at 165 volts for 1 hr and 20 minutes. This resulted in the separation of the proteins in the sample based on their molecular weight. The proteins were transferred to PVDF transfer membrane (Immobilon-FL; #IPFL07810, Sigma Aldrich) via electroblotting at 20volts for 1hr 10 minutes. The membrane was incubated with primary antibodies anti-GluN1 (1:1000; 5704S, Cell Signaling), anti-GluN2A (1:1000, #1500, BD Biosciences) and anti-GluN2B (1:750; #4212S, cell signaling) in blocking buffer (#P/N 927–70001, Li-Cor biosciences) overnight at 4°C. The following day, membrane was probed with fluorescent labelled secondary antibodies goat anti-mouse IRDye 800CW antibody (1:15,0000; #92632210, Li-Cor biosciences) and goat anti-rabbit IRDye 680RD antibody (1:15,0000; #926–68071, Li-Cor biosciences) at room temperature for 45 minutes. The membrane was scanned with Li-Cor odyssey imager to detect immunofluorescence of the target proteins bound to their specific fluorescent antibodies, analyzed using Image-Lite studio software (Li-Cor Biosciences) to determine to expression of NMDAR subunits.

Quantification of AMPAR Subunits:

The expression of AMPAR subunits GluA1 and GluA2 were quantified using immunoblotting similarly as the NMDARs as described above. Following the separation of the proteins via electrophoresis, and electroblotting, resulting in the transfer of proteins, the PVDF membrane with the transferred proteins was incubated with the primary antibodies, Anti GluA1 antibody(1:1000; #4969782, Sigma Aldrich) and anti-GluA2 antibody( 1:500; #13607S, Cell signaling). The following day the membrane was incubated with fluorescent labelled secondary antibodies, goat anti-mouse IRDye 800CW antibody (1:15,0000; #92632210, Li-Cor biosciences) and goat anti-rabbit IRDye 680RD antibody (1:15,0000; #926–68071, Li-Cor biosciences). The membrane was scanned with Li-Cor odyssey imager to detect immunofluorescence of the fluorescent antibodies binding to respective target proteins, analyzed using Image-Lite studio software (Li-Cor Biosciences) to determine to expression of AMPAR subunits.

Statistical Analysis:

All behavioral measures of interest were analyzed using analysis of variance (ANOVA) followed by Tukey’s post hoc test to identify significant differences across behavioral stage. The data from immunoblotting experiments represents the protein expression across behavioral stages and was determined by calculating the near-infrared fluorescence after subtraction of background fluorescence using the Image lite Version 5.2 software. One-way ANOVA was used to determine to statistical difference in protein expression across stages, which was followed by Tukey’s post hoc test to identify the stages across which the protein expression was significantly different. All statistical tests were performed using the GraphPad Prism software; version 9.4.2.

Results:

Analysis of Behavioral Stages:

Training mice to 5 specific stages of learning for discrimination (Disc^early; Disc^Late) and reversal (Rev^arly; Rev^mid; Rev^late) resulted in statistically distinct learning profiles as measured by percent correct responses, errors and correction errors. Analysis of percent correct responses showed a significant main effect of stage (F_{4, 95} = 615.8, P<0.0001) with Disc^early, Rev^arly, Rev^mid significantly differing from both Rev^late and Disc^late stages as shown by Tukey’s post-hoc tests (P<0.0001; Fig 1B). Similarly, analysis of total errors showed a main effect of behavioral stage (F_4,95 = 94.02, P<0.0001) with total errors being significantly lower in the well learned stages, Rev^late and Disc^late when compared to Disc^early, Rev^arly, Rev^mid (P<0.0001; Fig 1C). Correction errors also showed a main effect of behavioral stage (F_4,95 = 213.4, P<0.0001; Fig 1D). Additionally, Tukey’s post hoc analysis revealed significantly lower correction errors were made in the well learned stages, Rev^late and Disc^late, compared to Disc^early,Rev^arly, Rev^mid (P<0.0001). Rev^arly stage had significantly higher number of correction errors compared to all other stages (P<0.0001; Fig 1D). Stimulus reaction time also showed a main effect of stage (F_4,95 = 3.986, P=0.0049), where Rev^late stage had significantly lower stimulus reaction time when compared to Disc^early (P=0.0352) and Rev^arly stages (P=0.0135; Fig 1E). Reward reaction time, which measures the level of motivation of the mice performing the task remained comparable across all behavioral stages (F_4,95 = 0.9254, P=0.4526; Fig 1F).

We were able to reliably detect GluN1, GluN2B (Fig 2A) and the GluN2A subunits (Fig 2B) while Coomassie staining was used to quantify total protein (Fig 2C). Similarly, we were able to reliably detect AMPAR subunits GluA1, GluA2 (Fig 2D&E) and Coomassie staining used to quantify the total protein expression of AMPAR subunits (Fig 2F). No significant main effect of learning stage was found for the mPFC (Fig S2), BLA (Fig S3) or Pir (Fig S4) regions analyzed either for NMDAR subunits (Table 1) or AMPAR (Table 2).

Figure 2. Representative blot images used in the quantification of protein expression in the OFC and dS.

A. Blot showing bands for NMDAR subunits GluN1 at 120kda and GluN2b at 180kda. B. Blot showing bands for NMDAR subunit, GluN2A at 180kda. C. Coomassie staining used for total protein analysis. D. Blot showing bands for AMPAR subunit GluA1 at 100kda. E. Blot showing bands for AMPAR subunit GluA1 at 100kda. F. Coomassie staining used for total protein analysis.

Table 1:

Quantification of NMDAR subunit expression. Data are average immunoreactivity values per behavioral stage for each region in a subunit specific manner and the result of one-way ANOVA.

	GluN1						GluN2B						GluN2A
	Discrimination (Immunoreactivity)		Reversal (Immunoreactivity)			Main effect of stage (p value)	Discrimination (Immunoreactivity)		Reversal (Immunoreactivity)			Main effect of stage (p value)	Discrimination (Immunoreactivity)		Reversal (Immunoreactivity)			Main effect of stage (p value)
	Discearly	Disclate	Revearly	Revmid	Revlate		Discearly	Disclate	Revearly	Revmid	Revlate		Discearly	Disclate	Revearly	Revmid	Revlate
mPFC	0.41 ± 0.08	0.34 ± 0.03	0.31 ± 0.02	0.33 ± 0.04	0.36 ± 0.05	0.72	0.30 ± 0.04	0.26 ± 0.03	0.23 ± 0.01	0.25 ± 0.02	0.33 ± 0.05	0.24	0.04 ± 0.006	0.04 ± 0.005	0.04 ± 0.004	0.04 ± 0.003	0.05 ± 0.001	0.42
BLA	0.17 ± 0.03	0.12 ± 0.006	0.13 ± 0.01	0.15 ± 0.02	0.12 ± 0.01	0.20	0.13 ± 0.03	0.09 ± 0.007	0.09 ± 0.01	0.10 ± 0.02	0.70 ± 0.009	0.36	0.03 ± 0.007	0.02 ± 0.003	0.02 ± 0.004	0.02 ± 0.004	0.02 ± 0.002	0.43
Pir	0.18±0.03	0.24 ± 0.02	0.25 ± 0.06	0.26 ± 0.02	0.33 ± 0.02	0.07	0.13±0.02	0.18 ± 0.03	0.19 ± 0.04	0.18 ± 0.02	0.24 ± 0.05	0.27	0.03 ± 0.006	0.05 ± 0.01	0.04 ± 0.007	0.04 ± 0.004	0.05 ± 0.01	0.29

mPFC = medial prefrontal cortex; BLA= basolateral amygdala; Pir= piriform cortex (Immunoreactivity data are Means ± SEM, Main effect of stage = P value of One-way ANOVA)

Table 2:

Quantification of AMPAR subunit expression. Data are average immunoreactivity values per behavioral stage for each region in a subunit specific manner and the result of one-way ANOVA.

	GluA1						GluA2
	Discrimination		Reversal			Main effect of stage (p value)	Discrimination		Reversal			Main effect of stage (p value)
	Discearly	Disclate	Revearly	Revmid	Revlate		Discearly	Disclate	Revearly	Revmid	Revlate
mPF C	0.06 ± 0.01	0.05 ± 0.004	0.05±0.006	0.08±0.009	0.07 ± 0.008	0.06	0.24 ± 0.03	0.21 ± 0.04	0.21 ± 0.03	0.36±0.05	0.31±0.04	0.06
BLA	0.04 ± 0.007	0.04±0.004	0.04±0.002	0.05 ± 0.003	0.05±0.006	0.32	0.03 ± 0.004	0.03 ± 0.002	0.03 ± 0.001	0.04 ± 0.002	0.04±0.004	0.06
Pir	0.04 ± 0.007	0.07 ± 0.002	0.06±0.007	0.06 ± 0.007	0.06 ± 0.01	0.49	0.04 ± 0.006	0.07 ±0.01	0.07 ± 0.006	0.06 ± 0.006	0.07 ± 0.001	0.25

mPFC = medial prefrontal cortex; BLA= basolateral amygdala; Pir= piriform cortex (Immunoreactivity data are Means ± SEM, Main effect of stage = Pvalu of One-way ANOVA)

Orbitofrontal Expression of Ionotropic Glutamate Receptor Subunits:

To measure changes of NMDAR subunits in learning and reversal, the expression of these subunits was quantified via immunoblotting following reaching criteria in the specific behavioral stages. Quantification of the obligatory NMDAR subunit, GluN1 remained comparable through learning and reversal stages (F_4,15 = .3036, P=0.8711; Fig 3A). Similarly, GluN2A levels did not reveal any significant changes in the expression across the stages (F_4,15 = 1.750, P=0.1915; Fig 3B). GluN2B expression in the OFC showed a significant change across behavioral stages (F_4,15 = 4.018, P=0.0207; Fig 3C). Specifically, Tukey’s post hoc analysis revealed a significant increase in GluN2B expression during Rev^mid when compared to Rev^early stage (P= 0.0137, Fig 3C). The GluN2A/GluN2b ratio remained comparable across stages (F_4,15 = 0.3208, P=0.8596; Fig 3D). Similarly, The ratio of GluN2A (F_4,15 = 0.4487, P=0.7718; Fig 3E) and GluN2b (F_4,15 = 1.294, P=0.3164; Fig 3F) expression to the obligatory NMDAR subunit, GluN1 did not show any significant difference across the various behavioral stages.

Figure 3. OFC showed dynamic alteration in the NMDAR subunit GluN2b expression with significant increase during the early perseverative phase of reversal.

A. GluN1 expression remained stable across behavioral stages. B. There was no significant difference in the GluN2A subunit expression across the behavioral stages. C. GluN2B subunit was significantly increased during the early perseverative stage, from Rev^early through R^mid stage. D. The ratio of GluN2A to the obligatory subunit GluN1 did not change through learning and reversal. E. The ratio of GluN2B to the obligatory subunit GluN1 did not change through learning and reversal. F. The ratio of GluN2A to GluN2B also remained stable throughthe behavioural stages. G-I. Analysis of AMPAR subunits GluA1, GluA2 expression. G. No significant difference in the GluA1 subunit in the OFC across various learning stages. H. GluA2 expression also remained comparable across the stages I. The ratio of the subunits GluA1 to GluA2 did not alter through the behavioral stages.  The expression of all subunits were normalized to the total protein expression determined using Coomassie staining. Data are group mean ± SEM. * = p<0.05 main effect of behavioral stages. # = p<0.05 post hoc difference vs. Disc^early.

Quantification of AMPAR subunits GluA1 (F_4,15 = 0.2904, P=0.8796) and GluA2 (F_4,15 = 0.2346, P=0.9145) in the OFC did not reveal any significant differences across stages (Fig 3G&H). Similarly, GluA1/GluA2 ratio revealed no change in the OFC (F_4,15 = 0.2465, P=0.9073) Fig 3I).

Dorsal Striatal Expression of Ionotropic Glutamate Receptor Subunits:

NMDAR subunit expression was quantified in the dS across the 5 behavioral stages described above. Expression of total synaptic GluN1 showed no significant changes (F_4,15 = 2.194, P=0.1191) across learning stages (Fig 4A). Similarly, GluN2A (F_4,15 = 2.227, P=0.1151) total subunit expression also did not significantly differ (Fig 4B). However, GluN2B (F_4,15 = 3.194, P=0.0438) subunit expression revealed significantly different expression level across learning stages (Fig 4C). In contrast, the ratio of GluN2A/GluN2B (F_4,15 = 1.901, P=0.1627), relative expression of GluN2A (F_4,15 = 1.559, P=0.2362) and GluN2B (F_4,15 = 1.856, P=0.1708) to GluN1 remained comparably stable through learning and reversal stages (Fig 4D–F).

Figure 4. The dS showed increase in both AMPAR subunits during the well learned stage of reversal.

A. GluN1 expression remained stable across behavioral stages. B.There was no significant difference in the GluN2A subunit expression across the behavioral stages. C.GluN2B subunit was significantly different across the behavioral stages. D. The ratio of GluN2A to the obligatory subunit GluN1 did not change through learning and reversal. E. The ratio of GluN2B to the obligatory subunit GluN1 did not change through learning and reversal. F. The ratio of GluN2A to GluN2B also remained stable through the behavioural stages. G-I. Analysis of AMPAR subunits GluA1, GluA2 expression. G. GluA1 subunit expression was significantly increased during the well learned Rev^late stage when compared to the early perseverative Rev^late stage. H. GluA2 expression was also increased during the well learned Rev^late stage when compared to the early perseverative Rev^late stage. I. The ratio of the subunits GluA1 to GluA2 did not alter through the behavioral stages.  The expression of all subunits were normalized to the total protein expression determined using Coomassie staining. Data are group mean ± SEM. * = p<0.05 main effect of behavioral stages. # = p<0.05 post hoc difference vs. Disc^early.

Analysis of AMPAR subunit expression in the dS revealed a significant increase in the GluA1 (F_4,15 = 5.889, P=0.0047) and GluA2 expression (F_4,15 = 9.123, P=0.0006) during well learned reversal (Rev^late) versus early reversal (Rev^early, Fig 4G,H). GluA1/GluA2 ratio revealed no significant change in the dS (F_4,15 = 2.521, P=0.0849; Fig 4I).

Discussion

In the current study, we investigatged the whether learning or reversal would altered the expression of synpatic NMDA and AMPA using a well-validated task of behavioral flexibility in mice (Citri & Malenka, 2008; Malenka, 2003; Yashiro & Philpot, 2008). We found that reversal learning corresponded with significant changes in synaptic glutamatergic receptor expression in a subunit and stage specific manner. Specifically, synaptic NMDAR subunit GluN2B was significantly increased in the OFC as animals transitioned from perseverative to chance learning during revesal. In the dS, synaptic GluN2B showed a step-wise pattern, increasing across discrimination and reversal concominant with correct responding to the rewarded stimulus. Synpatic AMPAR subunits did not alter significantly in the OFC, but showed significant increase across all learning stages. Together, these results suggest that dynamic changes in synpatic NMDAR and AMPAR expression may play a role in the transition from well-learned responses to new choice behaviors as contingencies change.

It is well established that NMDARs are mediators of long-term potentiation at cortical and striatal synapses both in vivo (Partridge et al., 2000) and in vitro (Charpier & Deniau, 1997). The subunit composition of NMDAR confers distinct properties on the channel, with the presence of GluN2B conferring slower channel kinetics, lower open probabilities and higher sensitivity to glutamate versus those containing GluN2A (Cull-Candy et al., 2001). While it was previously proposed that these characteristics implicate specific subunits in the induction of LTP versus LTD, there is strong supporting evidence that the GluN2A/GluN2B ratio controls the threshold required to induce plasticity in order to stabilize associations or make them more labile (Yashiro & Philpot, 2008). Both learning and sensory input have been shown to increase the GluN2A/GluN2B ratio and the threshold of plasticity induction (Quinlan et al., 2004) which is thought to be critical for stabilizing new associations to support learning and memory (Kirkwood et al., 1996). In contrast, decreases in GluN2A/GluN2B ratio allow the induction of plasticity and are thought to be required to make associations more labile and modify existing behavioral responses (Holehonnur et al., 2016).

Previous studies support the role of GluN2B-containing NMDAR in plasticity and learning (Brigman et al., 2010; Duffy et al., 2008; Higgins et al., 2005; von Engelhardt et al., 2008) (Hawasli et al., 2007; Tang et al., 1999) and our current data show that corticostriatal synaptic expression of GluN2B is dynamic, with cortical levels remaining stable during discrimination learning, and signficantly increasing in the synpase during mid-stage reversal. In contrast, we found that NMDAR subunits did not significantly change across learning stages in the mPFC or BLA (Brigman et al., 2013a). Given previous findings that knockdown of GluN2B in the OFC was sufficient to increase the period of perseveration (Brigman et al., 2013a), these results suggest that dynamic trafficking of synaptic OFC GluN2B may be involved in updating choice behaviors during the transition from perseverative to learning phase of reversal. In contrast, synaptic GluN2B in the striatum tracked efficient learning across discrimination and reversal, in as step-wise pattern mirroring activations patterns in the dS seen across the task in IEG studies (Brigman et al., 2013a). These findings compliment previous results showing that cortical loss of GluN2B alters both OFC pyramidal neuron firing patterns, but also striatal firing dynamics and inter-region synchrony during reversal (Marquardt et al., 2019).

Previous studies have shown that forebrain wide knockdown of GluN2A impairs learning across both discrimination and reversal (Brigman et al., 2008) and that sensory experience and learning can alter GluN2A/GluN2B ratios and shortening of NMDAR currents, suggesting that GluN2A would be upregulated at well-learned stages such as criterion performance of discrimination and reversal. We did not detect any significant changes in GluN2A synaptic expression across learningsuggesting that in adulthood, dynamic changes in subunit expression may be primarily driven by changes in GluN2B, at least during visual reversal. However, it should be noted that the current study we were only able to measure subunit expression at 5 discrete points, and may have missed a critical window where GluN2A is dynamically altered.

Although NMDAR-mediated synaptic plasticity changes such as LTP and LTD are triggered by the activation of the NMDARs (Lüscher & Malenka, 2012), their expression requires change in the existing AMPARs and the addition of new AMPARs to the synapse (Citri & Malenka, 2008; Park et al., 2018; Yang et al., 2010). AMPA/NMDA ratios have therefore been used extensively as an indirect measure of plasticity in cortical circuits. However, little is known about how AMPA is dynamically regulated during flexible behavior. Our current results show that both GluA1 and GluA2 synaptic subunit expression was increased in the striatum during late reversal, but was not significantly altered in the OFC. These results are consistent with previous studies examining loss of AMPAR function in other brain regions. For example, impaired AMPAR trafficking via double knockout of MAPK-activated protein kinases 2 and 3 (DKO) in the hippocampus has been shown to impair spatial reversal learning (Eales et al., 2014). The DKO cells showed decreased GluA1 subunit expression at the synapse, which may have contributed to the reversal deficits seen in these mice (Eales et al., 2014). Similarlly, mice with systemic administration of a synthetic peptide Tat-GluA23Y, which blocks the endocytosis of AMPAR in the hippocampus consistently blocks LTD in hippacampal CA1 and can impair spatial reversal learning conducted in the Morris Water Maze (Dong et al., 2013; Ge et al., 2010; Migues et al., 2016).

Our current study found stage specific changes in synaptic glutamatergic subunit receptor expression in the OFC and dS by targeting a very specific time points previously shown to be sensitive to NMDAR manipulation. How extra-synaptic NMDAR and AMPA subunit expression or more specific changes within-learning stages (e.g. across sessions of early reversal) may play a role be involved in learning and flexible behavior is still unknown. In addition, due to the large number of mice required to detect sufficient levels of synaptic NDMAR and AMPAR proteins, only male mice were used in the current study. Although manipulations of GluN2B have not specifically shown sex-specific effects previously, future studies are required to determine if female mice show similar dynamic subunit expression.

Overall, the current results demonstrate that during touchscreen visual reversal, synaptic GluN2B peaks in the OFC during chance reversal while at the same time, postsynaptic striatal GluN2B and AMPAR increases across reversal. Together, these results further underscore the role of corticostriatal glutamatergic subunit expression plays in both learning and reversal. In addition, these findings suggest a potential mechanism for mediating the ability to efficiently learn a choice behavior leading to a positive outcome, and dynamically alter that behavior when contingencies change.

Supplementary Material

1

Figure S1. Representative blot images used in the validation of subcellular fractionation. A. Blot showing bands for Homer 1b/c (upper) and immunoreactivity in the orbitofrontal cortex across stages at 50kda and Homer 1a (lower) at 40kda. B. Blot showing expression of the glutamate transporter GLAST (EAAT1) bands at 60kda (upper) in the untreated crude synaptosomal membrane for NMDAR subunit, GluN2A at 180kda. In the synaptic fraction bound to PSD following Triton-X100 these bands were absent (lower) while GluA1 was robustly expressed.

Figure S2. No significant alterations were seen in NMDAR or AMPAR subunits in the medial prefrontal cortex (mPFC) across learning stages. A. GluN1 expression remained stable across behavioral stages. B-C. There was no significant difference in GluN2A or GluN2B subunit expression across the behavioral stages. D-E. The ratio of GluN2A to the obligatory subunit GluN1 or Glun2B to GluN1 did not change through learning and reversal. F. The ratio of GluN2A to GluN2B also remained stable through the behavioural stages. G-I. Analysis of AMPAR subunits GluA1, GluA2 expression. G-H. No significant difference in the GluA1 or GluA2 subunit in the mPFC across various learning stages. I. The ratio of the subunits GluA1 to GluA2 did not alter through the behavioral stages.  The expression of all subunits were normalized to the total protein expression determined using Coomassie staining. Data are group mean ± SEM.

Figure S3. No significant alterations were seen in NMDAR or AMPAR subunits in the basolateral amygdala (BLA) across learning stages. A. GluN1 expression remained stable across behavioral stages. B-C. There was no significant difference in GluN2A or GluN2B subunit expression across the behavioral stages. D-E. The ratio of GluN2A to the obligatory subunit GluN1 or Glun2B to GluN1 did not change through learning and reversal. F. The ratio of GluN2A to GluN2B also remained stable through the behavioural stages. G-I. Analysis of AMPAR subunits GluA1, GluA2 expression. G-H. No significant difference in the GluA1 or GluA2 subunit in the BLA across various learning stages. I. The ratio of the subunits GluA1 to GluA2 did not alter through the behavioral stages.  The expression of all subunits were normalized to the total protein expression determined using Coomassie staining. Data are group mean ± SEM.

Figure S4. No significant alterations were seen in NMDAR or AMPAR subunits in the piriform cortex (Pir) across learning stages. A. GluN1 expression remained stable across behavioral stages. B-C. There was no significant difference in GluN2A or GluN2B subunit expression across the behavioral stages. D-E. The ratio of GluN2A to the obligatory subunit GluN1 or Glun2B to GluN1 did not change through learning and reversal. F. The ratio of GluN2A to GluN2B also remained stable through the behavioural stages. G-I. Analysis of AMPAR subunits GluA1, GluA2 expression. G-H. No significant difference in the GluA1 or GluA2 subunit in the Pir across various learning stages. I. The ratio of the subunits GluA1 to GluA2 did not alter through the behavioral stages.  The expression of all subunits were normalized to the total protein expression determined using Coomassie staining. Data are group mean ± SEM.

Highlights:

• Glutamatergic NMDA and AMPA receptor subunit expression is dynamically expressed across touchscreen visual discrimination reversal learning in the cortical and striatal synapses.

• NMDAR subunit GluN2B is significantly increased in lateral orbitofrontal cortex synapses in the early perseverative phase, a period where the OFC plays a critical role for efficient reversal.

• Synaptic GluN2B is expressed in the dorsal striatum (medial and lateral aspects) during late stage reversal learning.

• AMPAR subunits GluA1 and GluA2 show increased expression in the synaptic fraction of dorsal striatum when associations are well-learned.

• This study provides evidence that choice learning and reversal is associated with dynamic alteration in the synaptic composition of glutamatergic receptors in the cortex and striatum.