Neuronal glutamate transporters control reciprocal inhibition and gain modulation in D1 medium spiny neurons

Understanding the function of glutamate transporters has broad implications for explaining how neurons integrate information and relay it through complex neuronal circuits. Most of what is currently known about glutamate transporters, specifically their ability to maintain glutamate homeostasis and limit glutamate diffusion away from the synaptic cleft, is based on studies of glial glutamate transporters. By contrast, little is known about the functional implications of neuronal glutamate transporters. The neuronal glutamate transporter EAAC1 is widely expressed throughout the brain, particularly in the striatum, the primary input nucleus of the basal ganglia, a region implicated with movement execution and reward. Here, we show that EAAC1 limits synaptic excitation onto a population of striatal medium spiny neurons identified for their expression of D1 dopamine receptors (D1-MSNs). In these cells, EAAC1 also contributes to strengthen lateral inhibition from other D1-MSNs. Together, these effects contribute to reduce the gain of the input-output relationship and increase the offset at increasing levels of synaptic inhibition in D1-MSNs. By reducing the sensitivity and dynamic range of action potential firing in D1-MSNs, EAAC1 limits the propensity of mice to rapidly switch between behaviors associated with different reward probabilities. Together, these findings shed light on some important molecular and cellular mechanisms implicated with behavior flexibility in mice.


Introduction
The neuronal glutamate transporter EAAC1, encoded by the Slc1a1 gene, is distributed broadly throughout the brain (Rothstein et al., 1994;Shashidharan et al., 1997), with plasma membrane surface density values thought to be significantly lower than those of astrocytic glutamate transporters (Holmseth et al., 2012). Glutamate binding to glutamate transporters activates currents that can be easily recorded in heterologous expression systems and astrocytes, where the density of expression of these molecules is high (Wadiche et al., 1995a). In most neurons, recording glutamate transporter mediated currents continues to be elusive (Holmseth et al., 2012). This finding sparked some doubts on the functional relevance of neuronal glutamate transporters like EAAC1 in the brain (Holmseth et al., 2012). Despite this concern, multiple lines of evidence point to the fact that EAAC1 is an important player in the regulation of synaptic function and behavior. For example, in the hippocampus, EAAC1 limits NMDA receptor activation and increases GABA release (Diamond, 2001;Mathews and Diamond, 2003). In the dorsolateral striatum (DLS), EAAC1 limits activation of group I metabotropic glutamate receptors (mGluRI) and is associated with increased execution of stereotyped motor behaviors (Bellini et al., 2018). While most of these works relied on the use of Slc1a1 -/mice (Peghini et al., 1997), other studies that used overexpression models of EAAC1 showed behavioral abnormalities, including increased anxiety-like behaviors (Delgado-Acevedo et al., 2019;Escobar et al., 2021) and reduced responses to amphetamine-induced hyperlocomotion (Zike et al., 2017). The emerging picture is that maintaining optimal levels of EAAC1 expression is key for proper execution of a wide range of complex motor behaviors and habitual actions, many of which are critically dependent on the activity of the DLS (Gremel and Costa, 2013;Yin et al., 2004;Yin et al., 2006).
What continues to puzzle the field is how we can reconcile the obvious behavioral abnormalities of Slc1a1 -/mice with the apparent inability to record EAAC1-mediated currents in neurons. A possible way out of this conundrum comes from considerations of the biophysical properties and sub-cellular distribution of this neuronal transporter. First, EAAC1 has a very low single channel conductance (~0.3 fS; Grewer et al., 2000) and is not evenly distributed along the plasma membrane (Cheng et al., 2002;Conti et al., 1998;Rothstein et al., 1994;He et al., 2000). Second, the fact that currents generated at a distance from the soma are dramatically reduced by electrotonic filtering and attenuation could make them particularly challenging to identify using somatic patch-clamp recordings (Tønnesen et al., 2014;Svoboda et al., 1996;Rall, 1959). Third, tentative estimates of the average surface density of expression EAAC1 obtained from immunoblot experiments do not provide information on the local density of expression of EAAC1 in dendritic spines and axonal boutons, where EAAC1 is thought to be confined. Fourth, the amplitude of local EAAC1-mediated currents does not provide a direct readout of the number of receptors that EAAC1 might protect from glutamate spillover . Together, these findings suggest that the inability to record EAAC1-mediated currents from the soma should not be interpreted as evidence of a lack of any functional role for glutamate uptake via EAAC1.
Interestingly, qualitative pre-and post-embedding ultrastructural works show that EAAC1 has a distinctive punctate peri-synaptic and post-synaptic expression in glutamatergic neurons, attributed to the presence of a specific domain in the C-terminal region that controls the targeting of EAAC1 to dendritic spines and shafts (Cheng et al., 2002;He et al., 2000). EAAC1 is also expressed in a population of GABAergic neurons identified by their expression of glutamic acid decarboxylase, the biosynthetic enzyme for GABA (Conti et al., 1998). In these neurons, EAAC1 shows a clustered expression in a subset of axonal presynaptic terminals (He et al., 2000;Rothstein et al., 1994), although these findings have been brought into question (Holmseth et al., 2012). In this case, the inconsistent results may be due to the limited sensitivity and specificity of antibodies directed against EAAC1 used for immunolabeling studies, which is a historically challenging, if not insurmountable issue. Other labeling strategies based on the use of pharmacological tools have been hampered by the fact that there continues to be no drug that targets EAAC1 without also affecting other glial glutamate transporters (Shimamoto et al., 1998;Tsukada et al., 2005). For these reasons, functional studies combined with genetic approaches currently represent an ideal tool to fill current gaps of knowledge on the functional roles of EAAC1 on synaptic communication in different brain areas.
Multiple lines of evidence suggest the existence of potential association between loss of function of EAAC1, altered synaptic activity in the striatum, and increased compulsive behaviors. First, the striatum is one of the brain regions with the most abundant expression of EAAC1. Second, hyperactivity of striatal circuits is associated with the emergence of compulsive behaviors (Gassó et al., 2015;Gilbert et al., 2008;Menzies et al., 2008b;Abramowitz et al., 2009;Carmin et al., 2002). Third, several genome-wide studies identified loss-of-function variants and single nucleotide polymorphisms of SLC1A1, the homolog gene encoding EAAC1 in humans diagnosed with OCD, autism spectrum disorder (ASD) and attention deficit/hyperactivity disorder (ADHD; Veenstra-VanderWeele et al., 2012;Wendland et al., 2009;Porton et al., 2013;Stewart et al., 2007;Stewart et al., 2013a;Stewart et al., 2013b;Gadow et al., 2010;Brune et al., 2008).
One of the hypotheses that has been brought forward to explain how EAAC1 may limit striatal hyperactivity is that impairing glutamate uptake via EAAC1 might lead to an increased ambient glutamate concentration in the striatum, which in turn would trigger hyperactivity (Porton et al., 2013). This interpretation is difficult to reconcile with evidence that EAAC1 consitutes only ~5% of all glutamate transporters (Holmseth et al., 2012). Even if all EAAC1 molecules lost their ability to bind and transport glutamate, under steady-state conditions, the remaining ~95% of glutamate transporters in glial cells would be able to keep the extracellular glutamate concentration at low nanomolar levels (Herman and Jahr, 2007;Chiu and Jahr, 2017). Accordingly, experimental data show that loss of EAAC1 does not change the extracellular glutamate concentration (Rothstein et al., 1996), a result supported also by our own previous works in the striatum (Bellini et al., 2018) and hippocampus ). An alternative hypothesis comes from computational models of neural dynamics in cortico-striatal-thalamic networks, which suggest that loss of EAAC1 might contribute to striatal hyperactivity by changing the relative strength of synaptic excitation and inhibition (E/I) locally, in some or all medium spiny neurons (MSNs), the main long-range projection neurons in the striatum (Rădulescu et al., 2017). According to this model, local changes in E/I of striatal MSNs can alter the firing rates of neurons not only in the striatum but also in larger neural networks that include the cortex and thalamus, bringing them in a regime of hyperactivity (Rădulescu et al., 2017). Although these theoretical inferences provide potential explanations of how local changes in E/I may lead to changes in the activity of more complex neural networks and behaviors, an experimental investigation of how EAAC1 alters E/I in different populations of striatal MSNs has not been previously performed.
Here, we show that EAAC1 limits synaptic excitation and strengthens synaptic inhibition in the DLS (i.e., reduces E/I), by modulating the strength of excitatory synaptic transmission onto D1 dopamine receptor expressing MSNs (D1-MSNs), and reciprocal inhibition among D1-MSNs. Through these mechanisms, EAAC1 increases the offset of the input-output relationship of D1-MSNs, to levels that differ depending on the rate of incoming inhibition. Together, these findings indicate that impairing the activity of EAAC1 brings D1-MSNs in a hyperexcitable state where these cells fire more action potentials in response to (i) lower frequencies and (ii) smaller changes in the frequency of activation of thalamo-cortical excitatory inputs. The reduced lateral inhibition among D1-MSNs may promote task switching, increasing the likelihood that mice rapidly change from one motor behavior to another. These findings shed new light on the synaptic and circuit mechanisms through which EAAC1 modulates the activity of specific populations of MSNs, with potential implications for understanding the cellular basis of impulsivity and compulsive behaviors (Grassi et al., 2015;Bari and Robbins, 2013;Boisseau et al., 2012;Benatti et al., 2014;Ettelt et al., 2007).

EAAC1 is expressed in DLS MSNs
Although there is evidence that EAAC1 is expressed in the DLS (Holmseth et al., 2012), it is not known whether its cellular distribution is limited to MSNs or differs between D1-and D2-MSNs. To address this, we performed an RNAscope FISH analysis in the DLS using probes for Drd1a, Drd2 and Slc1a1, the genes encoding D1/2 dopamine receptors and EAAC1 in mice, respectively ( Figure 1A and B). We identified a population of cells that did not express EAAC1, which we excluded from further analysis (24% of all cells stained with the nuclear marker DAPI; Figure 1B). The EAAC1-expressing cells could be further classified into three main groups using unsupervised clustering ( Figure 1C), principal component analysis ( Figure 1D) and a dimensionality reduction approach based on t-distributed stochastic neighbor embedding (t-SNE) analysis ( Figure 1F). The graphs in Figure 1C, D and F describe the distribution of the raw data in each cluster, for different types of dimensionality spaces, and the fact that they all identify three data clusters indicate that this result is not dependent on the type of clustering approach being used. The three EAAC1-expressing cohorts of cells were identified as being either D1-MSNs (25%), D2-MSNs (24%; Figure 1E), or cells with a very low, but detectable expression of EAAC1 and D1/2 receptors (51%; Figure 1E). These findings suggest that: (i) in the DLS, EAAC1 mRNA is mostly expressed in MSNs compared to other types of cells, and (ii) at the mRNA level, the expression of EAAC1 is indistinguishable between D1-and D2-MSNs ( Figure 1G and H).

EAAC1 reduces spine number and size in D1-MSNs
Dendritic growth and an increased complexity of the dendritic arbor can be promoted by increased neuronal activity, in vitro and in vivo (Yu and Malenka, 2003;Redmond et al., 2002;Sin et al., 2002). Since changes in dendritic morphology are often concurrent with changes in synaptic structure and function, we asked whether the structural changes in the dendritic arborization of D1-MSNs in the absence of EAAC1 might be associated with changes in the number and functional properties of excitatory synaptic inputs onto these cells (Peng et al., 2009). We addressed this question by analyzing the density and subtype distribution of dendritic spines, the anatomical correlates of excitatory synapses, in biocytin-filled MSNs. Spine size scales with AMPA receptor content and the release probability of the presynaptic terminal with which they are in contact (Schikorski and Stevens, 1997). We found that the spine density along the dendrites of D1-and D2-MSNs was not altered in the absence of and cells with no detectable Slc1a1 expression (which also had a very low expression of Drd1/Drd2; gray). (C) Circular dendrogram representation of RNAscope FISH data from EAAC1-expressing cells in the DLS. The dendrogram was obtained using the Ward's minimum variance method for hierarchical cluster analysis, using three as the number of groups cutting the tree (i.e., the optimal number of clusters obtained using gap statistics). The three main groups represent D1-MSNs (blue), D2-MSNs (salmon) and cells that express low levels of D1, D2 and EAAC1 (green). (D) Principal component analysis obtained using an enhanced k-means clustering with the three populations of cells. (E) Pie chart describing the percentage of the cell types identified with the hierarchical clustering. (F) Annotated clusters in the t-SNE map showing three specific groups of cells identified by the k-means algorithm based on their differential expression of D1, D2 and EAAC1. (G) Analysis of mRNA puncta in D1-MSNs (n=229 puncta; blue), D2-MSNs (n=212 puncta, salmon) and low D1/2, EAAC1 (n=460 puncta, burgundy). (H) Summary graph showing the mean number of puncta and S.E.M. measured in each of the three identified cell populations described in C. EAAC1 (p=0.34 and p=0.12, respectively), but was higher in D2-compared to D1-MSNs in Slc1a1 +/+ and Slc1a1 -/mice (**p=2.4e-3 and **p=2.7e-4, respectively; Figure 2G and H). Given that the total dendritic length is larger in Slc1a1 -/mice, the presence of a similar density of spines in D1-MSNs of Slc1a1 +/+ and Slc1a1 -/mice indicates that D1-MSNs receive a larger number of excitatory synaptic inputs in the absence of EAAC1 (spine number = spine density · dendritic length). In addition, there was an increase in the proportion of mushroom spines and a decrease in that of thin and stubby spines in D1-MSNs of Slc1a1 -/mice (D1-MSN mushroom spines: **p=1.1e-3; thin spines: *p=0.01; stubby spines: *p=0.02; Figure 2I). This effect was not detected when comparing the spine size distribution in D2-MSNs of Slc1a1 +/+ and Slc1a1 -/mice (D2-MSN mushroom spines: p=0.38; thin spines: p=0.28; stubby spines: p=0.42; Figure 2J). Therefore, the size of the dendritic arbor of D1-MSNs is smaller when EAAC1 is expressed, and these cells receive fewer and weaker excitatory synaptic inputs.

EAAC1 limits excitation onto D1-MSNs
An increase in the spine number and size in D1-MSNs of Slc1a1 -/mice might be indicative of functional changes in the quantal parameters (N, p and q), which can manifest themselves as changes in the frequency and/or kinetics of action potential-independent miniature EPSCs (mEPSCs). Therefore, we recorded mEPSCs in the presence of the voltage-gated sodium channel blocker tetrodotoxin (TTX 1 µM) and analyzed their frequency, amplitude, and kinetics. In our experiments, the mEPSC frequency was larger in D1-MSNs of Slc1a1 -/mice compared to Slc1a1 +/+ mice (*p=0.01; Figure 3A and B). This result is consistent with an increase in the number and/or release probability of excitatory synaptic contacts onto D1-MSNs in Slc1a1 -/mice. In Slc1a1 +/+ mice, the mEPSC frequency was larger in D2-MSNs compared to D1-MSNs (*p=0.02), whereas the frequency of these event in Slc1a1 -/mice was similar (p=0.51; Figure 3A and B). To determine whether the increased mEPSC frequency in D1-MSNs of Slc1a1 -/mice could be attributed exclusively to differences in N (the number of functional release sites) or also differences in p (the release probability), we measured the paired-pulse ratio (PPR) of evoked AMPA EPSCs in D1-MSNs, as this parameter is inversely proportional to the release probability, p. The PPR of AMPA EPSCs was similar in D1-MSNs from Slc1a1 -/and Slc1a1 +/+ mice (p=0.43), arguing against differences in p between excitatory synapses onto D1-MSNs in Slc1a1 -/and Slc1a1 +/+ mice. The mEPSC analysis also revealed that the mEPSC amplitude was larger in the absence of EAAC1 (**p=1.8e-3; Figure 3C), with no significant difference in mEPSC kinetics (rise:  Figure 3D). This is consistent with an increased quantal size (q) at excitatory synapses onto D1-MSNs in Slc1a1 -/mice, and with the increased spine size in these cells. Therefore, both N and q are increased in the absence of EAAC1.

EAAC1 strengthens inhibition onto D1-MSNs
One interesting property of EAAC1 is that its expression is not limited to glutamatergic synapses, but can also be detected in axonal boutons of GABAergic neurons, where it may serve to supply these cells with a substrate for GABA synthesis and release (Conti et al., 1998;Rothstein et al., 1994;He et al., 2000). This is because even though GABA can be synthesized de novo in axonal boutons, part of it can also be synthesized from glutamate recycled from the extracellular space via glutamate transporters (Scimemi, 2014). This recycling pathway differs in complexity depending on whether recycling of extracellular glutamate relies on neuronal or glial glutamate transporters ( Figure 4A, left). Whereas glutamate taken up by EAAC1 is converted into GABA via decarboxylation in the presynaptic terminal, glutamate taken up by glial transporters needs first to be converted into glutamine in the astrocyte cytoplasm, which is shuttled to neurons and converted first into glutamate and ultimately into GABA ( Figure 4A, left). There are multiple unknowns about these two recycling pathways. For example, it is not known how much GABA release onto MSNs relies on de novo synthesis versus recycling via neuronal or glial glutamate transporters. It is also unknown whether EAAC1 contributes differently to action potential-dependent and -independent GABA release. This is important, given the growing body of work indicating that vesicles mediating spontaneous and evoked inhibitory synaptic transmission are partially segregated among synapses and may utilize partially different molecular machineries (Horvath et al., 2020;Sara et al., 2011;Wang et al., 2022).

EAAC1 strengthens reciprocal inhibition between D1-MSNs
MSNs form extensive collaterals within the DLS. Accordingly, the dendritic field of each MSN extends over a volume that contains >2800 MSNs, and forms ~1200-1800 synaptic contacts onto these cells (Wilson, 2007). Functional studies suggest that each D1-MSN has a connection rate of 26% and 6% with other D1-and D2-MSNs, respectively (Taverna et al., 2008;Tecuapetla et al., 2009). By contrast, each D2-MSN has a connection rate of 28% and 36% with other D1-and D2-MSNs, respectively (Tecuapetla et al., 2009;Taverna et al., 2008). The presence of such an intricate inhibitory network of connections is thought to be important for regulating the excitability of MSNs and, more generally, the output of the striatum and the execution of coordinated motor behaviors (Burke et al., 2017; mice in control conditions (black) and in the presence of T-TBOA (1 µM; blue), at V hold = 40 mV. Center, Summary of the mIPSC amplitude and kinetics recorded before and after T-TBOA (n=11 neurons). Right, Summary of the relative effect of T-TBOA on the mIPSC amplitude and kinetics. T-TBOA induced a significant reduction in the amplitude of mIPSCs recorded from D1-MSNs. (C-E) As in B, for DLS D1-MSNs of Slc1a1 -/mice (n=10 neurons), and D2-MSNs of Slc1a1 +/+ mice (n=15 neurons), and Slc1a1 -/mice (n=13 neurons), respectively. In panels C, E we also included data collected from Slc1a1 +/+ mice to show that loss of EAAC1 leads to smaller and faster mIPSCs in D1-MSNs, whereas no significant difference is detected between D2-MSNs of Slc1a1 +/+ and Slc1a1 -/mice. (F) Summary of the effect of T-TBOA on the mIPSC frequency in D1-MSNs of Slc1a1 +/+ and Slc1a1 -/mice, with representative traces (left), raw data (center) and normalized values (right). (G) As in F, for D2-MSNs. Data represent mean ± SEM.
The online version of this article includes the following figure supplement(s) for figure 4:    Dobbs et al., 2016). To determine to which extent EAAC1 shapes lateral inhibition among different types of MSNs, we performed bilateral stereotaxic injections of a conditional ChR2-expressing AAV in the DLS of Drd1a Cre/+ :Rosa26 tm9/tm9 or Adora2a Cre/+ :Rosa26 tm9/tm9 mice ( Figure 6A, left). Three weeks later, we prepared acute coronal slices, patched D1-or D2-MSNs, identified for their expression of the tdTomato reporter, and used blue light stimuli for full-field activation of ChR2 expressed in either D1-or D2-MSNs ( Figure 6A, right). We set the intensity of the blue light stimulation to a power of ~250 µW, measured at the sample plane ( Figure 6-figure supplement 1A). Each stimulation lasted 5ms, a duration that did not evoke any adaptation in post-synaptic, ChR2-mediated photocurrents ( Figure 6-figure supplement 1B), and was repeated every 30 s, to allow for full recovery of ChR2 from activation ( Figure 6-figure supplement 1D). The ChR2 photocurrents, isolated in the presence of TTX (1 µM), reversed at ~20 mV ( Figure 6-figure supplement 1C) and their amplitude did not change in the presence of the GABA A receptor antagonist picrotoxin (100 µM). Therefore, we used 20 mV as the holding potential to record optogenetically-evoked IPSCs (oIPSCs), isolated pharmacologically by adding NBQX (10 µM) and APV (50 µM) to the external solution. Under these experimental conditions, the oIPSCs were completely blocked by the GABA A receptor antagonist picrotoxin (100 µM). To confirm the accuracy of the stereotaxic injections in the DLS, at the completion of the recordings, we fixed the slices and imaged them using a confocal microscope ( Figure 6B). Our experiments showed that the oIPSC amplitude was smaller at D1-D1 synapses from Slc1a1 -/compared to Slc1a1 +/+ mice, suggesting that EAAC1 might enhance synaptic inhibition at these synapses (*p=0.04; Figure 6D and H). By contrast, the amplitude of oIPSCs evoked at D2-D1 synapses was similar across the two genotypes (p=0.65; Figure 6E, I). The oIPSC amplitude was also similar at D1-D2 (p=0.79; Figure 6F and J) and D2-D2 synapses of Slc1a1 +/+ and Slc1a1 -/mice (p=0.54; Figure 6G and K).
Together, these results indicate that glial glutamate transporters enhance inhibition at homo-and hetero-synaptic contacts formed between different types of MSNs and, to a lesser extent, at synapses mice (n=9 neurons), respectively. (E-H) Summary of the relative effect of T-TBOA on the IPSC amplitude and kinetics. T-TBOA induced a significant reduction on the IPSC amplitude, but not of the rise time, 50% decay time and PPR. T-TBOA reduced the IPSC amplitude in D1-MSNs of Slc1a1 +/+ mice more than in any other type of MSNs tested in the experiments. Data represent mean ± SEM.
The online version of this article includes the following figure supplement(s) for figure 5:    Figure 6C). By contrast, EAAC1 significantly enhances synaptic inhibition only at homosynaptic GABAergic contacts formed between D1-MSNs.

EAAC1 limits the firing output of D1-MSNs
An essential aspect of information processing is the ability to transform synaptic inputs into action potential outputs, allowing the DLS to control the activity of its target regions. The data obtained so far provide an opportunity to shed light on the functional role of homosynaptic, reciprocal inhibition between D1-MSNs. For this reason, we asked how the changes in synaptic excitation and reciprocal inhibition between D1-MSNs induced by EAAC1 shape the firing output of these cells. To answer this, we used the NEURON platform and performed modeling experiments on reconstructed D1-MSNs from Slc1a1 +/+ and Slc1a1 -/mice, to account for the different morphological features of these cells (Figure 2; Hines and Carnevale, 1997). In these simulations, the passive and active membrane properties of D1-MSNs were consistent with those of D1-MSNs in Slc1a1 +/+ and Slc1a1 -/mice (Figure 7-figure supplement 1), and the weight of each excitatory and inhibitory input was set to be consistent with our experimental data (Figure 7-figure supplements 2-4). We allowed each D1-MSN to receive 100 excitatory and 100 inhibitory inputs, randomly distributed along the dendrites, activated at a range of frequencies consistent with the range of synaptic inputs received by the striatum in vivo. We analyzed the effect of these parameter manipulations on firing rates of D1-MSNs across the theta (∼8 Hz) and beta range (∼20 Hz), also representative of the firing activity detected in the striatum in vivo ( Figure 7A and B; Berke et al., 2004;Pennartz et al., 2009). We then asked whether and how EAAC1 altered the offset and the gain in the input-output relationship of D1-MSNs ( Figure 7C, left and center). Briefly, changing the offset allows D1-MSNs to subtract basal levels of synaptic activity, altering the range of input stimulation frequencies that evoke firing (an additive/subtractive operation). Changing the gain alters the sensitivity of D1-MSNs to varying levels of synaptic input rates, while preserving the range over which these inputs evoke spiking activity (a multiplicative/divisive operation; Mitchell and Silver, 2003). In the absence of synaptic inhibition, EAAC1 induced a modest increase in offset, which became more pronounced as the inhibition rate increased ( Figure 7D). With increasing rates of inhibition onto D1-MSNs, EAAC1 also decreased the gain, and therefore switched from having a purely additive effect to also having a divisive effect ( Figure 7D, right). Conversely, when excitation was low, EAAC1 reduced the basal firing rate of D1-MSNs, with a negligible effect on gain ( Figure 7E, left). As excitation increased, the increase in gain becomes more evident and is associated with a slight decrease in the basal firing rate of D1-MSNs ( Figure 7E, right). The overall effect of EAAC1 across a broader range of activity of E/I is to introduce a frequency-dependent increase in offset and a frequency-independent decrease in gain ( Figure 7C, right). Together, these results suggest that EAAC1 can perform both subtractive and divisive operations, depending on the rate of incoming E/I onto D1-MSNs. The higher is the rate of synaptic E/I onto D1-MSNs, the greater is the reduction of the firing output of D1-MSNs by EAAC1. By modulating E/I, EAAC1 narrows the dynamic range and reduces the sensitivity of D1-MSNs to incoming inputs.   The role of excitation and reciprocal inhibition between D1-MSNs in reward-based behaviors The activity of MSNs is important for the execution of coordinated movements and reward-based behaviors (Freeze et al., 2013;Kravitz et al., 2010;Lobo et al., 2010;Kravitz et al., 2012). Although heterosynaptic inhibition from D2-MSNs has been shown to inhibit action potential firing in D1-MSNs (Dobbs et al., 2016), much less is known about the functional significance of homosynaptic inhibition between D1-MSNs. Computational models suggest that this form of inhibition might be involved in driving coherence or synchronizing functional units of information processing in the striatum, known as ensembles (Humphries and Prescott, 2010a;Humphries et al., 2010b;Humphries et al., 2009;Moyer et al., 2014;Wickens, 2012, Ponzi andWickens, 2010;Ponzi and Wickens, 2013;Yim et al., 2011). If the striatum truly operates as a collection of ensembles driving specific behaviors, one may hypothesize that the reduction in synaptic excitation and reciprocal inhibition among D1-MSNs in Slc1a1 -/mice might increase their propensity to switch between different rewardbased behaviors. We tested this hypothesis using a simple probabilistic reward lever press task, in which we trained mice to receive a water reward at different probability (P rew = 0.5‖0.9; Figure 8A-E).
Although the DLS is recruited during task switching, reward-based flexibility in executive control also relies on neuronal activity in ventral medial regions of the striatum (VMS; Wallis, 2007;Gu et al., 2008). Given the presence of an abundant population of D1-MSNs in the VMS, we asked whether removing EAAC1 from D1-MSNs in the VMS could also allow mice to engage more effectively in slowly switching reward-based behaviors. To test this hypothesis, we repeated the task switching test in Slc1a1 f/f mice that received stereotaxic injections of a Cre-dependent viral construct (AAV-D1Cre) either in the DLS or VMS, respectively (Figure 8-figure supplement 2). Mice lacking EAAC1 in DLS or VMS D1-MSNs displayed task switching behaviors similar to those of Slc1a1 -/mice (Figure 8figure supplement 2) suggesting that EAAC1 expression in D1-MSNs throughout the striatum limits the execution of reward-based behavior flexibility for slowly switching time intervals.
Lastly, we asked whether loss of EAAC1 in D1-MSNs outside the striatum could also recapitulate the behavior of Slc1a1 -/mice. To address this question, we repeated the injections of AAV-D1Cre in the substantia nigra (pars compacta and reticulata; SN) and ventral tegmental area (VTA) of Slc1a1 f/f mice, where D1 receptors are also abundantly expressed (Cadet et al., 2010;Savasta et al., 1986;Boyson et al., 1986;Wamsley et al., 1989; Figure 8-figure supplement 3). The behavior of these mice in the operant task switching test was similar to that of Slc1a1 +/+ mice, pointing to a specific role of EAAC1 in striatal D1-MSNs in the execution of slowly switching reward-based behaviors.
Together, these findings suggest that increased excitation onto D1-MSNs and reciprocal inhibition between striatal, but not nigral/tegmental D1-MSNs, limits execution of reward-based behaviors and compulsivity during slow action switching. E) As in B, C, showing the total number of lever presses performed by the mice. (F) Temporal distribution of the number of rewards collected by Slc1a1 +/+ (left) and Slc1a1 -/mice (right) when the reward probability was switched from P rew = 0.5 to P rew = 0.9. The thick green line represents a binomial smoothing of the mean. (G) As in F, for the number of lever presses. (H) Raster plot showing the temporal distribution of lever presses in a 5 min long trial, in which the reward probability was switched from P rew = 0.5 to P rew = 0.9 every 30 s. The raster plots in this panel were collected at P rew = 0.5. Color-coded tick marks (gray/pink) represent the times when the lever was pressed. Other tick marks (black/red) represent the times when the water rewards were collected. (I) As in H, for P rew = 0.9. (J) Schematic representation of the test session, with water rewards delivered at two different values of P rew , switching at different time intervals (5-75 s). (K-N) As in B-E, for the test session. Data represent mean ± SEM.
The online version of this article includes the following figure supplement(s) for figure 8:

Discussion
The main finding in this work is that the neuronal glutamate transporter EAAC1 exerts a cellpreferential control of E/I in the DLS: it limits excitation onto D1-MSNs and enhances homosynaptic lateral inhibition between D1-MSNs. By doing so, EAAC1 increases the offset and decreases the gain of the firing output of D1-MSNs. These effects are associated with reduced ability to engage in slowly switching reward-based behaviors.

EAAC1 regulates D1-MSN morphology
To the best of our knowledge, the data presented here provide the first indication that the neuronal glutamate transporter EAAC1 is implicated with the control of neuron and spine morphology. Our previous analysis in the hippocampus indicated that the spine density and spine head diameter are not different in CA1 pyramidal cells of Slc1a1 +/+ and Slc1a1 -/mice . However, this work did not include a Sholl analysis or a detailed analysis of spine type distribution, like the one included in this manuscript. It is not known whether the effects we detect here are induced specifically by EAAC1 or whether they could also be induced by a comparable loss of glial glutamate transporters. Although there is evidence that the expression of glial glutamate transporters is altered in neuropsychiatric disorders, and that this is associated with a reduction in spine density and dendritic arborization (O'Donovan et al., 2017), a causal link between these structural effects and disease onset remains to be established. It might be tempting to speculate that the increase in spine size and dendritic complexity in D1-MSNs of Slc1a1 -/mice might be due to increased spillover onto NMDA receptors, since there is evidence that glutamate triggers morphological changes in neurons and promotes the acquisition of a mature spine morphology largely by acting on these receptors (Mattison et al., 2014;Kwon and Sabatini, 2011). If this were true, one could hypothesize that the lack of effect of genetic loss of EAAC1 on D2-MSNs might be due to the inability of EAAC1 to alter synaptic transmission onto these cells. This, however, would not be supported by the results obtained in the hippocampus, where EAAC1 limits spillover onto NMDA receptors, yet has no detectable effect on spine size . It is also possible that the effects on the morphology of D1-MSNs may be a downstream effect of EAAC1 on other cells, although interestingly dopamine depletion induces spine pruning in D2-more than D1-MSNs (Witzig et al., 2020;Gagnon et al., 2017;Day et al., 2006;Suárez et al., 2014;Toy et al., 2014;Fieblinger et al., 2014). While these findings indicate that the signaling pathways implicated in the control of spine and dendrite size are many and complex, they also point out to the fact that EAAC1 might be a critical, unsuspected component of these regulatory mechanisms.

The role of EAAC1 in controlling synaptic integration in D1-MSNs
The DLS, one of the four main subdivisions of the striatum, integrates synaptic inputs from multiple brain regions to control the execution of habitual behaviors (Hunnicutt et al., 2016;Wall et al., 2013). In mice, ~1/3 of excitatory inputs to the DLS originates from the thalamus (Huerta-Ocampo et al., 2014), and the remaining 2/3 from the frontal associative and sensorimotor cortex (Hunnicutt et al., 2016;Guo et al., 2015;Wall et al., 2013;Pan et al., 2010). All these inputs provide excitation onto both D1-and D2-MSNs, but sensory inputs preferentially innervate D1-MSNs, whereas motor inputs preferentially target D2-MSNs (Wall et al., 2013). Our findings do not distinguish whether EAAC1 limits excitation at some or all these glutamatergic projection types to the DLS, but multiple lines of evidence indicate that only excitation onto D1-MSNs is affected by EAAC1. First, there is a reduced spillover-activation of NMDA receptors in these cells (Figure 3). Second, EAAC1 expression is associated with a reduced spine size, quantal size (q) and quantal content (N) in D1-MSNs (Figures 2  and 3). Third, the dendritic arbor of D1-MSNs is smaller when EAAC1 is expressed (Figure 2). Since D1-MSNs preferentially process sensorimotor information, these findings suggest that, in the DLS, EAAC1 is primarily involved with sensorimotor input integration.
How excitation onto D1-MSNs is processed, relayed to other brain regions and ultimately converted into different behavioral outputs depends on multiple factors, including: (i) the spatial distribution and activation time of these thalamo-cortical excitatory inputs; (ii) the local connectivity and pattern of activity of inhibitory connections formed by striatal interneurons and other MSNs; (iii) the intrinsic excitability properties of these cells, which however do not change in the absence of EAAC1 (Figure 7-figure supplement 1; Burke et al., 2017;Preston et al., 1980;Somogyi et al., 1981;Wilson and Groves, 1980;Kawaguchi et al., 1990;Yung et al., 1996;Park et al., 1980). MSNs receive strong peri-somatic feedforward inhibition from fast spiking PV-INs, which is not altered by EAAC1 (Burke et al., 2017). A large population of 1200-1800 inhibitory inputs, however, targets the distal dendrites of these cells, and is formed by axon collaterals of other MSNs. Although these distal inputs are thought to be at a positional disadvantage and weaker than peri-somatic inhibitory inputs, they can change local synaptic integration and the firing output of MSNs (Tunstall et al., 2002;Czubayko and Plenz, 2002;Dobbs et al., 2016;Koós and Tepper, 1999;Wilson, 2007). Accordingly, simultaneous activation of D2-MSNs suppresses D1-MSN firing (Dobbs et al., 2016). Given that the coordinated activity of MSNs accounts for the complementary roles of these cells in the temporal control of movement execution (Kravitz et al., 2010;Cui et al., 2013), an increased spiking of D1-MSNs caused by altered E/I in the absence of EAAC1 could provide an important mechanism to disrupt the coordinated recruitment of MSNs during sensorimotor integration and stereotyped movement execution. This hypothesis is supported by our own previous work, showing that Slc1a1 -/mice display an increased grooming frequency (Bellini et al., 2018).

Physiological implications of offset-and gain-modulation by EAAC1
Understanding the input-output transformations of D1-MSNs is a key step in tying together the effect of EAAC1 on E/I. In the input-output relationships, changes in offset or gain exert different effects on information processing (Abbott et al., 1997;Salinas and Thier, 2000;Schwartz and Simoncelli, 2001;Prescott and De Koninck, 2003;Mehaffey et al., 2005). Changes in offset alter the input detection threshold (Pavlov et al., 2009). Changes in gain alter the sensitivity of a neuron to changes in excitatory inputs, and the range of inputs that can be discriminated (Pavlov et al., 2009). Excitatory afferents to the striatum generate action potentials at different rates. In the absence of inhibitory inputs, EAAC1 increases the offset of D1-MSNs (i.e. it increases their detection threshold), but preserves the gain of their input-output relationship. As the rate of D1-D1 inhibition increases, EAAC1 increases the offset even further, while also reducing the gain (i.e. the sensitivity to changes in excitation rates). That is to say that the input detection threshold of D1-MSNs for thalamo-cortical excitatory inputs is increased and the gain is slightly reduced at increasing levels of reciprocal inhibition among these cells, when EAAC1 is expressed. This flexible offset-and gain-control mechanism, which varies with the frequency of E/I and limits D1-MSNs firing in response to ongoing thalamo-cortical excitation, is lost in Slc1a1 -/mice.
Multiple studies suggest that MSNs in the DLS are organized into ensembles that can control the execution of different behaviors through their coordinated activation (Adler et al., 2012;Carrillo-Reid et al., 2011;Carrillo-Reid et al., 2008;Barbera et al., 2016). One of the conceptual models for the mechanisms of action of these functional units posits that D1-MSNs in units driving a specific behavior and D2-MSNs in units inhibiting competing behaviors are simultaneously active (Mink, 1996). Reciprocal inhibition between D1-MSNs would then limit the execution of competing actions, whereas reciprocal inhibition between D2-MSNs suppresses the inhibition of the desired behavior (Mink, 1996). According to this working model, disrupting reciprocal inhibition between D1-MSNs, as it happens in Slc1a1 -/mice, would promote unsynchronized activity across different functional units (Humphries and Prescott, 2010a;Moyer et al., 2014;Wickens, 2012, Ponzi andWickens, 2010;Ponzi and Wickens, 2013;Yim et al., 2011). Would this cause the DLS to remain in a given state and promote the sustained execution of a given behavior, or would it perhaps allow easier switcing between different behaviors? Our experiments support a model in which reciprocal inhibition among D1-MSNs limits action switching only at long time intervals, and likely promotes coordinated movement execution (Figure 8). Consistent with this hypothesis, loss-of-function mutations of EAAC1 have been implicated with a neuropsychiatric disease characterized by loss of movement coordination, compulsions and impulsivity, like OCD (Porton et al., 2013). Therefore, by modulating E/I onto D1-MSNs, EAAC1 may be a key regulator of the activity of functional units in the striatum also implicated with this disease. These mechanisms are particularly important in the context of neuropsychiatric diseases like OCD, as they might contribute to increased impulsivity and hypersensitivity to triggers, respectively (Grassi et al., 2015;Boisseau et al., 2012;Bari and Robbins, 2013;Benatti et al., 2014;Ettelt et al., 2007). Different circuits support the execution of fast and slow switching tasks An interesting finding highlighted by our experiments is that the perception of time is important for recruiting different neuronal circuits to support cognitive flexibility. In this context, the fact that the task switching behavior of Slc1a1 -/mice differs from that of Slc1a1 +/+ mice only at long time intervals (>30 s) suggests that excitation and lateral inhibition between D1-MSNs are engaged or change their properties slowly compared to other circuits that are preferentially involved in the execution of rapid switching tasks. Multiple open questions remain. For example, we do not know: (i) why is D1-D1 inhibition is recruited slowly? (ii) which other circuits, within or outside the striatum, support rapid task switching? Although our study does not provide a clear answer to these questions, evidence from previous works shows that: (i) the dopamine tone in the striatum ramps up slowly and smoothly, and alters the speed of temporal processing during timing tasks (Rao et al., 2001;Matell et al., 2003;Howe et al., 2013;Westbrook and Braver, 2016); (ii) striatal neurons encode temporal information slowly, over the course of many seconds, via time-dependent ramping of action potential firing (Emmons et al., 2017;Bakhurin et al., 2017;Matell et al., 2003;Mello et al., 2015;Wang et al., 2018;Bruce et al., 2021). Therefore, one could speculate that slowly evolving dopamine ramps might contribute to the recruitment of lateral D1-D1 inhibition in both Slc1a1 +/+ and Slc1a1 -/mice, but they may exert different effects on slow task switching behaviors (i.e. different number of lever presses/ rewards) due to differences in the offset-and gain-control properties of D1-MSNs due to altered excitation and lateral inhibition between D1-MSNs in Slc1a1 +/+ and Slc1a1 -/mice. The identity and location of the circuits responsible for controlling fast task switching programs (e.g. striatal or cortical ones), at present, remains unknown.

Functional implications of EAAC1 in different brain regions and in the context of compulsive behaviors
Increasing evidence implicates EAAC1 in the modulation of synaptic transmission in at least two regions of the brain: the hippocampus and DLS (Bellini et al., 2018;Diamond, 2001;Mathews and Diamond, 2003). In our studies, we noticed commonalities in the mechanisms through which EAAC1 controls synaptic transmission in the hippocampus and DLS, because in both regions EAAC1 acts to limit glutamate escape at excitatory synapses and increases GABA release. However, the functional consequences of these effects can vary, due to existing differences in the molecular landscape of excitatory and inhibitory synapses across brain regions. In CA1 pyramidal cells of the mouse hippocampus, EAAC1 limits glutamate escape onto NMDA receptors and promotes long-term plasticity (Diamond, 2001;. In the DLS, EAAC1 also limits glutamate escape onto NMDA receptors. In addition, by preventing activation of mGluRI receptors, it promotes D1 dopamine receptor expression and long-term plasticity (Bellini et al., 2018). EAAC1 contributes to GABA synthesis and release from stratum oriens interneurons to CA1 pyramidal cells (Mathews and Diamond, 2003), and only at a subset of inhibitory synapses mediating lateral inhibition across D1-MSNs in the striatum. Overall, our data indicate that despite being expressed in D1-and D2-MSNs, EAAC1 exerts a preferential effect on different types of synaptic inputs targeting D1-MSNs (Bellini et al., 2018). This is confirmed also by our behavioral assays, showing that loss of EAAC1 in D2-MSNs does not change the ability of mice to engage in switching reward-based tasks (Figure 8-figure  supplement 1). By contrast, loss of EAAC1 in D1-MSNs recapitulates the behavior of Slc1a1 -/mice ( Figure 8-figure supplement 1). Although our investigation of changes in cell morphology and synaptic function was limited to the DLS, we cannot rule out the possibility that analogous changes in neuron morphology and synaptic transmission may also take place in the VMS, given the similar behavioral phenotype of mice lacking EAAC1 in D1-MSNs of the DLS and VMS (Figure 8-figure  supplement 2). Since the behavioral phenotype of mice lacking EAAC1 in D1-MSNs of the SN/VTA is similar to that of Slc1a1 +/+ mice (Figure 8-figure supplement 3), our findings point out to a cell-type specific and location-specific role of EAAC1 expression in striatal D1-MSNs with respect to compulsive-like behaviors, likely mediated via the regulation of lateral inhibition between these cells.
There are indications that there are structural changes in the hippocampus of patients affected by compulsions (including OCD, ADHD, and ASD patients; Reess et al., 2018;Al-Amin et al., 2018). Although their implications remain unclear, these findings support emerging evidence that limbic structures like the hippocampus might be implicated with compulsive behaviors in a variety of neuropsychiatric disorders (Milad and Rauch, 2012;Menzies et al., 2008a;Wood and Ahmari, 2015;Ullrich et al., 2018).
Obviously, it is important to remember that although animal studies provide strong evidence for dysfunction in glutamatergic transmission at striatal thalamo-cortical synapses in human patients with compulsions, several other genes in the glutamatergic system, in addition to SLC1A1, are associated with different neuropsychiatric diseases (Welch et al., 2007;Shmelkov et al., 2010). For example, other candidate genes for OCD belong to the serotonergic and dopaminergic systems. Furthermore, non-genetic, environmental risk factors are also crucial for the manifestation of symptoms for OCD and other neuropsychiatric diseases associated with compulsions, making them complex polygenic and multifactorial diseases (Pauls et al., 2014;Hoffman and Cano-Ramírez, 2022;Chen et al., 2022). Despite this apparently daunting scenario, there is considerable agreement on the implication of cortico-striatal hyperactivity in compulsivity, suggesting the existence of a possible convergence of different genetic and epigenetic factors in the control of striatal function. Therefore, the findings described here may provide an example of circuit and synaptic dysfunctions shared across different genetic models of neuropsychiatric disease.

Trans-cardial perfusion and RNAscope fluorescence in situ hybridzation (FISH)
For RNAscope FISH, mice of either sex aged P30 were anesthetized with a sodium pentobarbital solution (Euthanasia-III Solution, Med-Pharmex, Pomona, CA; 390 mg/ml, 3.9 g/Kg). The mice were perfused through the ascending aorta with 20 ml PBS and 20 ml 4% PFA/PBS at 4 °C at 8 ml/min. The brains were removed, post-fixed with 4% PFA/PBS overnight at 4 °C, cryo-protected in 30% sucrose PBS at 4 °C for 48 hr and stored in PBS for 24 hr. To prepare coronal sections for RNAscope, we separated the two hemispheres, embedded them in 4% w/v agar/PBS and prepared 40-µm-thick slices using a vibrating blade microtome (VT1200S, Leica Microsystems, Buffalo Grove, IL). The slices were post-fixed in 4% PFA/PBS for 30 min at room temperature (RT) and mounted onto Superfrost plus microscope slides. Mounted slices were used for fluorescence in situ hybridization (FISH) using an RNAscope multiplex fluorescent assay (Advanced Cell Diagnostics, Newark, CA) according to manufacturer instructions, using Mm-Drd1-C1, Mm-Sc1a1-C2 and Mm-Drd2-C3 RNA probes and Opal 520, 570 and 690 dyes (Akoya Biosciences, Menlo Park, CA). DAPI Fluoromount G was used as the mounting medium (Cat# 0100-02; SouthernBiotech, Birmingham, AL). The presence of mRNA transcripts was assessed using a confocal microscope (Zeiss LSM710) equipped with a Plan-Apochromat 63 X/1.4NA oil objective. Image size was set to 1,024 · 1,024 pixels and represented the average of 8 consecutive frames. The image analysis was performed using CellProfiler version 4.0.7 and a modified version of the Speckle Counting Pipeline (Carpenter et al., 2006). In this pipeline, we set a feature size of 20 for the EnhanceOrSuppressFeatures tool. We used Advanced Settings for the IdentifyPri-maryObjects tool. Here, the typical diameter of objects was set to 3-20 pixels, the selected thresholding method was Robust Background, and the Upper Outlier Fraction was set to 0. The size of the smoothing filter for declumping was set to 4. The minimal allowed distance to suppress local maxima was set to 4. We did not use lower-resolution images to find local maxima. The EnhanceOrSuppress-Features, MaskImage, IdentifyPrimaryObjects and RelateObjects tools were applied for each label color (green, red, far red; i.e. three times).

Confocal imaging of biocytin filled MSNs and spine analysis
Biocytin 0.2%-0.4% (w/v) was added to the intracellular solution used to patch MSNs. Each cell was filled for at least 20 min. The slices were then fixed overnight at 4 °C in 4% PFA/PBS, cryo-protected in 30% sucrose PBS, and incubated in 0.1% streptavidin-Alexa Fluor 488 or streptavidin-Alexa Fluor 647 conjugate and 0.1% Triton X-100 for 3 hr at RT. The slices were then mounted onto microscope slides using Fluoromount-G or DAPI Fluoromount-G mounting medium (Cat# 0100-01 and Cat# 0100-02, respectively; SouthernBiotech, Birmingham, AL). Confocal images were acquired using a Zeiss LSM710 inverted microscope equipped with 488 nm Ar or 633 nm HeNe laser. All images were acquired as stitched z-stacks of 4-5 frames averages (1024x1024 pixels; 1 µm z-step) using a 40 X/1.4 NA or 63 X/1.4NA Plan-Apochromat oil-immersion objectives. We generated 2D maximum intensity projections of biocytin-filled MSNs and traced the dendritic processes manually using Simple Neurite Traces in Fiji (https://fiji.sc/). Confocal images for spine analysis were also collected as z-stacks of 8 frame averages (1024x1024 pixels; 0.5 µm z-step; 3-5 digital zoom) using a 63 X/1.4 NA Plan-Apochromat oil-immersion objective. For structural analysis, dendritic spines were classified into four groups according to their neck and head size (i.e. mushroom, thin, stubby, filopodia), using Fiji (Peters and Kaiserman-Abramof, 1970;Jones and Powell, 1969;Harris et al., 1992).
Each conductance was adjusted as follows: sodium channels (2.4 S/cm 2 ); voltage-gated calcium channels (Ca V 1.2: 6.7e-6 S/cm 2 , Ca V 1.3: 1.0e-4 S/cm 2 ) and potassium channels (K AS : 4.0e-5 S/cm 2 , K DR : 5.0e-3 S/cm 2 , K IR : 1.0e-4 S/cm 2 , K RP : 2.0e-4 S/cm 2 ). Voltage-gated sodium channels were placed in the axon initial segment (a 50 µm axonal segment with a diameter of 0.5 µm). All voltage-gated calcium and potassium channels were distributed throughout the plasma membrane of the soma and the dendrites. Passive leak channels were present throughout the plasma membrane. We did not make assumptions on the location of the synaptic inputs active during our electrophysiology recordings, or on the potential existence of synaptic scaling mechanisms that might balance for passive attenuation of distal synaptic inputs, as this has not been documented for D1-MSNs (Nicholson et al., 2006;Magee and Cook, 2000). Instead, we assumed that all inputs onto D1-MSNs are of equal strength, regardless of which portion of the dendrite they target. This assumption was important to set the synaptic weights of E/I inputs to ensure the model reproduced the amplitude of mEPSCs and mIPSCS recorded experimentally from D1-MSNs. To accomplish this, each simulation was repeated 100 times, each time randomizing the location of either one excitatory or inhibitory input (Figure 7figure supplement 2A, Figure 7-figure supplement 4A). We then averaged the somatic mPSCs and compared their amplitude and kinetics with those of mPSCs recorded from D1-MSNs in Slc1a1 +/+ mice. We performed an initial adjustment of the peak conductance of the synaptic inputs in the model (AMPA: 7.2e-4 µS, GABA A : 3.6e-4 µS) to match the mPSC amplitude recorded from experiments.
We modeled mPSCs decay using a mono-exponential function, whereby decay = A 1 e −t/τ1 . We set the amplitude (A 1 ) and decay time constant (τ 1 ) of mPSCs in the model to match that of A 1 and τ 1 measured across experiments (Figure 7- In the case of NMDA EPSCs, we did not make a direct comparison between the model and experimental NMDA mEPSCs, as these events are particularly small in D1-MSNs, and technically challenging to record. Instead, we relied on knowledge of the AMPA mEPSC amplitude at V hold =-70 mV (Figure 3), the reversal potential for glutamatergic currents (0 mV), and the experimentally measured value for the NMDA/AMPA ratio corrected by the driving force of each current (Slc1a1 -/-0.57±0.14; Slc1a1 +/+ 0.68±0.14; p=0.57; Figure 3-figure supplement 1A, B), to predict the NMDA mEPSC amplitude at V hold = 40 mV. The estimated NMDA mEPSC amplitude was calculated as the product of the AMPA mEPSC amplitude at V hold =-70 mV, the NMDA/AMPA ratio, and the ratio of the NMDA and AMPA driving force (Slc1a1 +/+ 15.0 pA · 0.68 · 40 mV/70 mV =5.8 pA; NMDA mEPSC Slc1a1 -/-19.3 pA · 0.57 · 40 mV/70 mV =6.3 pA). This estimated mEPSC amplitude was used for the initial adjustment of the peak conductance of the synaptic NMDA input in the model (NMDA: 6.8e-5 µS). The estimated NMDA mEPSC decay matched that of evoked NMDA EPSCs recorded experimentally (Figure 7-figure supplement 3). We then determined the values for A 1 and τ 1 for synaptic NMDA inputs in our model (Figure 7-figure supplement 3C, D) using the same method previously mentioned for synaptic AMPA and GABA inputs. Together, these parameter-constraints allowed the compartmental model to reproduce the larger AMPA mEPSCs amplitude ( Figure 3A-D), similar NMDA/AMPA ratio (Figure 3-figure supplement 1), and smaller GABA mIPSC amplitude and decay time in D1-MSNs from Slc1a1 -/mice ( Figure 4).

Behavioral analysis
We performed our behavioral analysis on C57BL/6 J wild type and Slc1a1 -/mice of either sex aged P28-37, maintained on as 12 h:12 h L/D cycle. Three days prior to training, and for the whole duration of the training and testing sessions, the drinking water in the home cage was supplemented with 2% citric acid (w/v) (Reinagel, 2018;Whiteway et al., 2021). Each behavioral session lasted 5 min. During each one of the ten training sessions, a 70 µl water reward was delivered in response to a lever press, and the timing of each lever press and reward was monitored using custom made MATLAB code and a B-pod state machine (Sanworks, Rochester, NY). The reward probability was set to P rew = 0.5‖0.9. Each mouse was subject to 10 training sessions with P rew = 0.5 and 10 training sessions with P rew = 0.9. Mice that did not perform lever presses during this training period were excluded from further analysis. During the one final testing session, we changed the reward probability from P rew = 0.5 to P rew = 0.9 at one of the following switching time intervals (5, 10, 15, 30, 50, 75 s). The number and time distribution of lever presses and rewards for each mouse was analyzed in IgorPro 6.37 (Wavemetrics, Lake Oswego, OR).

Ethics
All experimental procedures were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee at the State University of New York (SUNY) Albany and guidelines described in the National Institutes of Health's Guide for the Care and Use of Laboratory Animals.

Supplementary files • MDAR checklist
Data availability All primary data used in this work and complete statistical analyses for each figure have been deposited to the Open Science Framework (https://osf.io/dw5n7/).
The following dataset was generated: