Male NAc D1-/D2-MSN neurotransmission-blocked mice (D1-/D2-MSN-Blocked) and wildtype (WT) controls, aged between 10 and 16 weeks, were generated using the TRE-TeNT-GFP transgenic mice on a C57BL/6 background, as previously described (Hikida et al., 2010; Macpherson et al., 2016). Tetanus toxin (TeNT) is a bacterial toxin that blocks the release of neurotransmitters from the presynaptic terminal of the neurons in which it is expressed by cleaving the vesicle-associated membrane protein VAMP2 (Schiavo et al., 1992; Wada et al., 2007). In TRE-TeNT-GFP mice, the expression of TeNT and green fluorescent protein (GFP) is under the control of tetracycline responsive element (TRE) and is driven by the interaction of TRE with tetracycline transactivator (tTA) (Yamamoto et al., 2003; Wada et al., 2007).

In both WT and TRE-TeNT-GFP mice, tTA was specifically expressed in either NAc D1-MSNs or D2-MSNs, which is known to coexpress the peptides substance P (SP) or Enkephalin (ENK), respectively (Gerfen et al., 1990; Lu et al., 1997), by bilateral microinjections of a recombinant adeno-associated virus (AAV) construct (AAV2-SP-tTA or AAV2-ENK-tTA) into the NAc (AP: +1.5 mm, ML: ±0.8 mm, DV +3.5 mm; 500 nl infused at 50 nl/min; spread of ±0.5 mm in each area) under anesthesia (90 mg/kg Ketamine and 20 mg/kg Xylazine, i.p. injection). This resulted in persistent blocking of neurotransmitter release from either NAc D1- or D2-MSNs of TeNT mice (D1-MSN-Blocked: n = 8, D2-MSN-Blocked: n = 9), but had no effect on WT (n = 8, per group) mice, an effect that has been previously been electrophysiologically validated (Hikida et al., 2010). Post-surgery, mice were provided with an anti-inflammatory drug (10 mg/kg Rimadyl, Zoetis, Florham Park, NJ, United States) in their drinking water for 1 week and left in their home cages for 3–4 weeks for adequate viral expression and surgical recovery.

Mice were housed in groups of 2–4 and were maintained on a 12-h light/dark schedule (lights on at 8 a.m.) at a temperature of 24 ± 2°C and humidity of 50 ± 5% controlled room. Beginning 3 days before the commencement of experiments, mice were food restricted to 85% of their free-feeding weight on standard lab chow, with water available ad libitum. Behavioral experiments were performed between the hours of 10 a.m. to 6 p.m. Following the completion of experiments, virus infusion locations were histologically verified by immunohistochemical investigation of GFP expression (Figure 1C), and two mice were excluded due to misaligned injection sites. All animal handling procedures and use of viruses were approved by the animal research committees of the Kyoto University Graduate School of Medicine (approval ID: MedKyo17071) and the Institute for Protein Research, Osaka University (approval ID: 29-02-1).

The ASST was performed in a sound-attenuating experimental room using a square opaque acrylic box [45 cm (L) × 45 cm (W) × 30 cm (D)] divided halfway down the front center line by an opaque barrier to create two equal-sized chambers and a rectangular staging area that had a removable acrylic divider placed horizontally 15 cm from the back wall (Figure 1A). In each of the two chamber areas, a shallow square polyethylene platform [15 cm (L) × 15 cm (W) × 5 cm (D)] was added, and on top of each platform was placed a circular plastic weighing dish (7 cm diameter) acting as a digging bowl. The platforms were either left as they were or wrapped in one of the five materials (styrofoam, corrugated cardboard, metal wire, sandpaper, and bubble wrap) to provide six different tactile cues. The digging bowls were filled with woodchips bedding that had been infused with one of the six different odors (coffee, cinnamon, rosemary, garlic, ground ginger, and nutmeg). Odorless sucrose pellets (Dustless Precision Pellets,^® Sucrose, Unflavored, 20 mg, Bio-Serv, Flemington, NJ, United States) were used for rewards, and it was verified before the experiment commencement that mice were unable to detect the location of a baited bowl at a greater chance level (50 ± 10% accuracy across a total of 50 trials using two woodchips filled bowls; one baited and one not) when no location cues (odor/texture) were provided.

Trials to criterion and total errors were collected for each stage; however, as these two measures are correlated and analysis of either produced the same results, only trials to criterion are reported [as previously described (Birrell and Brown, 2000; Young et al., 2010)]. Response latencies (seconds) were separated into the mean latencies to perform a correct or an incorrect response (mean correct/incorrect latency). The total amount of omissions per session was also recorded.

The data were found to be normally distributed (Shapiro–Wilk tests; p > 0.05) and the assumption of homogeneity of variance was not violated (Levene’s tests; p > 0.05). Data were analyzed separately for D1- and D2-MSN-Blocked groups (including their respective WTs) initially using repeated measures of three-way ANOVAs with stage (SD, CD, CDR, IDS, IDR, EDS, and EDR) as a within-subjects variable and group (D1-/D2-MSN-Blocked or WT) and dimension change (odor-to-platform and platform-to-odor) as between-subject variables. The influence of dimension change was also checked separately for D1- and D2-MSN-Blocked groups using univariate three-way ANOVAs with the EDS stage as the dependent variable and group (D1-/D2-MSN-Blocked or WT) and dimension change (odor-to-platform and platform-to-odor) as independent variables. As no significant main effect or interaction of dimension change was found in all the analyses (see Supplementary Table 1), dimensions were grouped together for all the subsequent analyses as well as in the presented graphs. D1- and D2-MSN-Blocked groups were then reanalyzed separately using repeated measures of two-way ANOVAs with stage (SD, CD, CDR, IDS, IDR, EDS, and EDR) as the within-subjects variable and group (D1-/D2-MSN-Blocked or WT) as the between-subject variable. Additionally, to validate the formation of attentional sets, trials to criterion in IDS vs. EDS stages were analyzed separately in D1- and D2-MSN-Blocked groups using repeated measures of two-way ANOVAs with stage (IDS and EDS) as the within-subjects variable, and group (D1-/D2-MSN-Blocked or WT) as the between-subject variable. Post-hoc analyses of significant effects were performed using the Bonferroni test. Correlations between trials to criterion and mean incorrect latency were analyzed using the Pearson’s correlation coefficients. Statistical significance was considered to be p < 0.05. All statistical analyses are presented in Supplementary Table 1.

Following the completion of experiments, mice were anesthetized (90 mg/kg Ketamine and 20 mg/kg Xylazine, i.p. injection) and then transcardially perfused with cold 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) (Nacalai Tesque, Kyoto, Japan). Brains were removed from the skull and soaked in 30% sucrose in phosphate-buffered saline (PBS; pH 7.4) for 3 days until completely submerged, frozen at −20°C with compound medium (Tissue-Tek O.C.T. compound, Sakura Finetech, Tokyo, Japan), and then sliced into 30 μm coronal sections using a cryostat (Leica CM1860, Leica Biosystems, Wetzlar, Germany). Free-floating sections in PBS were subjected to immunohistochemistry [as previously described (Ohishi et al., 1994)] using a rabbit polyclonal anti-GFP primary antibody (A-11122, ThermoFisher Scientific, Waltham, MA, United States) diluted (1:500) in PBS and a fluorescent secondary antibody conjugated to Alexa 488 (Life Technologies, Newark, CA, United States) also diluted (1:200) in PBS. Sections were mounted with Vectashield containing DAPI (Vector Laboratories, CA, United States) and images were captured using a Keyence BZ-X810 fluorescence microscope (Keyence, Osaka, Japan).

Immunohistochemical staining of GFP found expression of the viral vector to be largely restricted to the NAc Core, with minimal spillover to NAc Shell or dorsal striatal regions (Figure 1C).

The ability of WT and D1-/D2-MSN-Blocked mice to perform the ASST was measured. A significant main effect of the stage on trials to criterion was observed in both D1-MSN-Blocked [Figure 2A; F_(6,84) = 10.78, p < 0.001] and D2-MSN-Blocked [Figure 2B; F_(6,90) = 5.34, p < 0.001] groups, indicating that mice’s performance varied across the different stages. Additionally, a comparison of performance on the IDS vs. the EDS stage for internal validation of attentional set formation (Young et al., 2010) revealed a significant main effect of the stage on trials to criterion for both D1-MSN-Blocked [Supplementary Figure 1A; F_(1,14) = 24.06, p < 0.001] and D2-MSN-Blocked [Supplementary Figure 1B; F_(1,15) = 27.85, p < 0.001] groups. Poorer performance on the ED stage by both groups indicated that all animals were able to successfully form an attentional set to the internal stimulus dimension. Analysis of omissions found no significant main effect of stage or genotype, and no stage × genotype interaction for both D1- and D2-MSN-Blocked groups, indicating that inhibition of neurotransmitter release from NAc Core D1- or D2-MSNs likely did not alter task engagement.

The D2-MSN-Blocked group, but not the D1-MSN-Blocked group, demonstrated a significant main effect of genotype [Figure 2B; F_(1,15) = 37.20, p < 0.001], as well as a significant stage × genotype interaction [Figure 2B; F_(6,90) = 5.34, p < 0.001], on trials to criterion. Subsequent post-hoc analyses using Bonferroni’s multiple comparison test revealed that the D2-MSN-Blocked mice took a significantly greater amount of trials to reach criterion on all reversal stages (CDR, IDR, and EDR), but not discrimination (SD and CD) or set-shift stages (IDS and EDS), than WT controls (Figure 2B). These findings suggest that blockade of neurotransmitter release from NAc Core D2-MSNs was able to impair the ability for reversal learning.

Investigation of the mean correct latency revealed a significant main effect of the stage in both D1-MSN-Blocked [Supplementary Figure 2A; F_(6,84) = 4.34, p < 0.01] and D2-MSN-Blocked groups [Supplementary Figure 2B, F_(6,90) = 3.96, p < 0.01]. However, no significant main effect of genotype or interaction between stage and genotype were found (Supplementary Table 1). Thus, in all mice, correct response times appeared to vary across different stages.

Finally, for the mean incorrect latency, a significant main effect of the stage was found for both D1-MSN-Blocked [Figure 3A; F_(6,84) = 2.51, p < 0.05] and D2-MSN-Blocked [Figure 3B; F_(6,90) = 2.35, p < 0.05] groups, suggesting that, with the mean correct latency, incorrect response times varied across different stages. Additionally, in the D2-MSN-Blocked, but not D1-MSN-Blocked, group a significant main effect of genotype [Figure 3B; F_(1,15) = 11.43, p < 0.001], as well as a significant stage × genotype interaction [Figure 3B; F_(6,90) = 4.67, p < 0.001] was found. Post-hoc Bonferroni’s multiple comparison tests revealed that incorrect response times were shorter in all reversal stages (CDR, IDR, and EDR), but not in discrimination (SD and CD) or set-shift stages (IDS and EDS), than in WT controls (Figure 3B). Subsequent Pearson’s correlation coefficient analysis of reversal stages (CDR, IDR, and EDR) demonstrated that the incorrect latency and, for the most part, the mean correct latency were significantly negatively correlated with trials to criterion in D2-MSN-Blocked and WT mice (Supplementary Figures 4A–F). These findings suggest that as response time slows down, accuracy in reversal stages of the ASST increases. Moreover, it is plausible that the reduced response time in error trials in reversal stages may underlie the impaired performance in these stages, with less time for cognitive deliberation resulting in a reduction in response accuracy.

Our study found that neurotransmission blocking in NAc Core D1- and D2-MSNs did not alter the ability for discrimination learning in the initial acquisition stage of the ASST. While these findings are consistent with a previous study from our group indicating that signaling in NAc D1- and D2-MSN is not necessary for the acquisition of a spatial discrimination task (Macpherson et al., 2016), they contrast with other studies reporting NAc D1-MSN activity to be necessary for spatial and visual discrimination tasks (Hikida et al., 2010; Nishioka et al., 2021). These findings may be explained by differences in the complexity of the tasks used. It has been suggested that NAc signaling becomes necessary when task requirements are ambiguous or require considerable cognitive or physical effort (Floresco, 2015; Macpherson et al., 2021). Indeed, in visual discrimination tasks disrupted by NAc inactivation, animals were required to inhibit responses to known cues and respond only at random cues (Nishioka et al., 2021), or to respond correctly at one of the five possible response windows (five-choice serial reaction time test) (Christakou, 2004; Pezze et al., 2007). Similarly, in NAc inactivation-impaired spatial learning tasks, animals were required to navigate through up to eight possible locations in a radial arm maze (Floresco et al., 1997; Gal et al., 1997). Despite the distracting influence of task-irrelevant cues in the current ASST study, the discrimination learning stages are considerably simpler than the above-described visual and spatial discrimination studies, requiring mice to choose between only two possible options. As such, they match with the previous evidence demonstrating NAc inactivation to have no effect on the performance of discrimination tasks with only two-to-four locations (Castañé et al., 2010; Macpherson et al., 2016) or visual cues (Floresco et al., 2006).

The role of the NAc in reversal learning is complicated, with NAc inactivation studies often reporting apparently conflicting results depending on differences in the method of NAc manipulation used, the region of the NAc targeted, the species tested, and the type of reversal learning measured. In rats, NMDA receptor (NMDAR) blockade of either the NAc Core or Shell with the NMDAR antagonist AP5, an effect likely to inhibit the neural activity of these regions, was reported to impair reversal learning in a spatial operant task (Ding et al., 2014), while performance in a similar task was found to be unaltered following quinolinic acid lesions of either the NAc Core or Shell (Castañé et al., 2010). In a spatial T-maze task, dopamine depletion of the NAc using 6-hydroxydopamine lesions has been reported to disrupt reversal learning (Taghzouti et al., 1985); however, reversal learning in an operant probabilistic task and a deterministic task was found to be disrupted by pharmacological inactivation of the NAc Shell, but not Core, using the GABA receptor agonists baclofen and muscimol (Dalton et al., 2014). In non-human primates, ibotenic acid lesions of the NAc were reported to impair visual reversal learning, but, in contrast to inactivation studies in rodents, did not alter spatial reversal learning (Stern and Passingham, 1995). Finally, in humans, fMRI analysis of subjects performing visual reversal learning tasks has similarly identified NAc activation during reversal error responses (Cools et al., 2002).

In set-shifting studies, baclofen and muscimol inactivation of the NAc Core of rats has been revealed to disrupt switching from spatial to visual cue-based strategies in a radial arm-based task, with NAc Shell inactivation oppositely facilitating set-shifting (Floresco et al., 2006). Similarly, infusion of NMDAR antagonist AP5 into the NAc Core, but not Shell, impaired set-shifting from visual to spatial cue-based strategies in an operant task (Ding et al., 2014).

Despite their dissimilar findings, the above-described studies, as well as this and previous studies from our group (Yawata et al., 2012; Macpherson et al., 2016), support an important role for NAc neurons in mediating behavioral flexibility.

In the ASST, neurotransmission blocking in NAc D2-MSNs, but not D1-MSNs, was demonstrated to impair performance in reversal learning stages, where two learned S–R–O associations were switched. This finding supports those of previous studies from our group and others indicating signaling in NAc D2-MSNs to be critical for reversal learning in both visual and spatial discrimination tasks (Yawata et al., 2012; Macpherson et al., 2016; Cui et al., 2018). In general, these findings are also supported by previous studies investigating the effect of pharmacological manipulation of dopamine receptors on reversal learning. Intra-NAc Core infusion of a D2R agonist, but not a D1R agonist or a D1R or D2R antagonist, was revealed to disrupt reversal learning in a visual discrimination task in rats (Haluk and Floresco, 2009). Whereas, a more recent study reported that intra-NAc Core infusion of a D2R antagonist, but not a D1R antagonist, was able to improve reversal learning in a visual discrimination task by reducing perseverative errors (Sala-Bayo et al., 2020). Given that D2Rs are Gi protein-coupled receptors that act to inhibit the D2-MSNs in which they are expressed (Shen et al., 2008), these findings suggest that the activity of NAc Core D2-MSNs contributes to the ability for reversal learning. Interestingly, constitutive deletion of D2Rs has also been shown to reduce reversal learning ability in ASST (DeSteno and Schmauss, 2009), odor discrimination (Kruzich and Grandy, 2004), and visual discrimination (Morita et al., 2016) tasks. Thus, it is possible that disturbance of normal NAc D2-MSN signaling, either by increased or reduced activity, may be sufficient to disrupt reversal learning ability. However, the possible influence of D2R deletion in areas outside of the NAc on reversal learning in these tasks cannot be discounted.

Finally, a potential limitation of the current study is that it included only a single rather than serial reversal stages. Previous studies have revealed that disruption of the OFC is able to impair the initial, but not serial, reversal stages in serial reversal odor discrimination tasks (Schoenbaum et al., 2002, 2003), suggesting that reversal learning deficits may not persist following repeated training. While it is not clear how neurotransmitter release inhibition in NAc Core D2-MSNs may affect performance on serial reversal learning stages in an ASST, previous work from our group has revealed that, in NAc Core D2-MSN-Blocked mice, impaired reversal learning in an initial reversal stage of a serial reversal place discrimination task was gradually restored to the level of controls across repeated reversals (Macpherson et al., 2016). In this study, we speculated that other brain regions, potentially the dorsal striatum, may be able to compensate for impaired reversal learning ability following repeated training across serial reversal stages.

Reversal learning deficits following neurotransmission blocking in NAc Core D2-MSNs were found to be associated with a reduced average latency to make an incorrect response, but no change in the average latency to make a correct response when compared with WT controls. These data suggest that signaling from NAc D2-MSNs plays an important role in increasing the decision-making time concerning whether to respond to previously correct and now outdated learned S–R–O associations, potentially helping to reduce the likelihood of incorrect responding. In support of the importance of NAc Core D2-MSNs in response inhibition, a recent study revealed that intra-NAc Core infusion of the D2R antagonist raclopride, but not the D1 antagonist SCH-23390, selectively reduced early perseverative errors in a visual cue-based serial reversal-learning task (Sala-Bayo et al., 2020). Oppositely, intra-NAc Core infusion of the D2 agonist quinpirole has been found to increase perseverative error responses in five-choice serial reaction time tasks (Pezze et al., 2007). These findings suggest that signaling from NAc Core D2-MSNs may be able to bidirectional control perseveration. In the current study, it was not possible to directly assess perseverative errors due to the choice of only two response options; however, in a previous study by our group, neurotransmitter release inhibition from NAc Core D2-MSNs resulted in an increase in perseverative, but not general, errors in serial reversal learning stages of a four-choice spatial discrimination task (Macpherson et al., 2016).

Inhibition of perseverative responding in reversal learning tasks is suggested to require information feedback concerning response errors (Klanker et al., 2013). Both animal and computational data have revealed that D2Rs play a critical role in such signaling of response errors during visual and probabilistic reversal learning tasks, allowing learning from losses (Alsiö et al., 2019). These findings are also congruent with recent evidence from our group demonstrating NAc D2-MSNs to be critical for the future avoidance of non-rewarded cues following response errors (Nishioka et al., 2021). Studies in humans also appear to support the role of the NAc in response inhibition, with fMRI analysis identifying robust activation of the NAc during the final reversal error of a visual reversal learning task, immediately before a switch in responding toward the correct visual cue (Cools et al., 2002).

Overall, our data add to a growing literature demonstrating the importance of NAc D2-MSNs in inhibiting incorrect behavioral responses, likely by providing necessary feedback or response errors.

A major finding of our study was that the neurotransmission blocking of NAc Core D1- and D2-MSNs did not alter the ability for set-shifting in the ASST. These findings contrast with the previous studies that demonstrate bilateral inactivation of the NAc Core, as well as disconnection of prefrontal and thalamic inputs to the NAc Core, to impair the set-shifting ability (Floresco et al., 2006; Block et al., 2007). Similarly, the same group revealed that intra-NAc administration of pharmacological agents acting at D1R and D2R modulate the set-shifting ability. Haluk and Floresco (2009) reported that a D1R, but not D2R, antagonism as well as D2R, but not D1R, agonism disrupted the set-shifting ability from a visual to a spatial cue-based strategy. However, another group investigating the role of D2Rs in behavioral flexibility found no effect of constitutive D2R deletion on set-shifting from odor to texture cue-based strategies in the ASST (DeSteno and Schmauss, 2009).

It is unclear why impairments in the set-shifting ability reported in the above-described studies of NAc inactivation were not observed in the current study. However, it should be noted that our study differs from these studies in several methodological factors. First, while previous studies of the NAc and set-shifting have tended to use rats, our study used mice. Second, in contrast to previous studies that used visual or spatial cues to guide responding, in our task, mice were required to utilize odor and tactile cues. Thus, it is possible that while activity in NAc Core D2-MSNs is necessary for switching to or from visual or spatial cue-based strategies, these neurons may not be necessary for strategy switching based on odor and tactile cues. As described in the previous section (see Section Evidence for Nucleus Accumbens Control of Behavioral Flexibility), similar outcome differences in studies utilizing different species or cue modalities are not uncommon, and further investigation is likely necessary to identify how behavioral flexibility based on information from various modalities may be differentially controlled within the NAc of various species. Finally, in previous studies, NAc subregions and cell types were inactivated acutely by intracranial infusions of dopamine or GABA receptor ligands, or anesthetics. In contrast, our study utilized a chronic NAc Core D1-/D2-MSN inactivation method. It is possible that such chronic inactivation may result in neuroplastic compensatory mechanisms, such as a different brain region being recruited, that allow the animal to regain the ability for set-shifting. However, if this is the case, it is unclear why reversal learning remained impaired. Future studies utilizing transient-cell-type-specific inactivation methods, such as Cre-dependent inhibitory opsins or artificial receptors in transgenic D1-/D2-Cre lines, may help to elucidate this question.

Studies on humans have indicated that the dorsal striatum and its inputs from the mPFC may contribute significantly to the control of set-shifting. The fMRI analysis of healthy controls found significant activation of dorsal frontal-striatal regions during set-shifting stages of a visual discrimination task, whereas OCD patients demonstrating the dysfunctional set-shifting ability showed no such activation (Gu et al., 2008). Conversely, carriers of a DRD2/ANNK1-Taqla polymorphism that results in reduced D2 receptors, particularly in the dorsal striatum, demonstrated impaired set-shifting performance in a visual cue-guided reward learning task and decreased functional connectivity between mPFC and dorsal striatal regions (Noble et al., 1997; Stelzel et al., 2010). In positron emission tomography (PET) studies, increased dopamine release has been observed in the dorsal striatum during set-shifting (Monchi et al., 2006), while reduced dopamine in the dorsal striatum during the early stages of Parkinson’s disease is associated with the impaired set-shifting ability (Lewis et al., 2005; Cools, 2006; Kehagia et al., 2010). Interestingly, while treatment of early-stage Parkinson’s disease patients with levodopa reverses set-shifting dysfunction, it has been found to impair performance reversal learning tasks, potentially by excessive stimulation of DA receptors in the ventral striatum which is generally less prone to dopamine depletion during the early stage of the disease (Swainson et al., 2000; Cools, 2006; Kehagia et al., 2010). These studies, alongside our finding that NAc Core D2-MSN neurotransmission blocking impairs reversal learning but not set-shifting, suggest a functional dissociation in the striatal regions controlling different aspects of behavioral flexibility, with an OFC-NAc D2-MSN pathway potentially mediating reversal learning and an mPFC-dorsal striatum potentially mediating the set-shifting ability. Future studies using D1- and D2-MSN-specific neurotransmission blocking in various subregions of the dorsal striatum will likely help to identify the precise striatal circuits responsible for set-shifting.