Tsc1-mTOR signaling controls the structure and function of midbrain dopamine neurons

mTOR complex 1 (mTORC1) is a central coordinator of cell growth and metabolism. Mutations in regulators of mTORC1 cause syndromic disorders with a high prevalence of cognitive and psychiatric conditions. To elucidate the cellular origins of these manifestations, we conditionally deleted the gene encoding the mTORC1 negative regulator Tsc1 from mouse midbrain dopamine neurons, which modulate motor, affective, and cognitive behaviors that are frequently affected in psychiatric disorders. Loss of Tsc1 and constitutive activation of mTORC1 strongly impacted the properties of dopamine neurons, causing somatodendritic hypertrophy, reduced intrinsic excitability, altered axon terminal ultrastructure, and severely impaired dopamine release. These perturbations were associated with selective deficits in cognitive flexibility, which could be prevented by genetic reduction of the obligatory mTORC1 protein Raptor. Our results establish a critical role for mTORC1 in setting the functional properties of midbrain dopamine neurons, and indicate that dopaminergic dysfunction may underlie cognitive inflexibility in mTOR-related syndromes.


Introduction
The mechanistic target of rapamycin (mTOR) pathway is a highly conserved, fundamental signaling cascade that integrates intra-and extracellular signals to regulate a variety of cellular metabolic processes. Deregulation of mTOR signaling is linked to numerous diseases, with particularly detrimental outcomes for nervous system development and function (Costa-Mattioli and Monteggia, 2013;Lipton and Sahin, 2014). Precisely how perturbed mTOR signaling affects the activity of neurons and neural circuits is a key open question. mTOR functions as a kinase within two complexes, mTORC1 and mTORC2, that have distinct protein components, upstream activators, and downstream targets (Hoeffer and Klann, 2010).
Activation of mTORC1 via PI3K/Akt signaling promotes anabolic functions including protein and lipid synthesis, and suppresses catabolic processes such as autophagy, leading to cell growth and division (Huang and Manning, 2008). mTORC1 activity is negatively regulated by the heterodimeric Tsc1/2 protein complex that exerts GTP-ase activating (GAP) activity on the small GTP-ase Rheb, a direct activator of mTORC1 (Laplante and Sabatini, 2012). Disruption of the Tsc1/2 complex results in constitutively active mTORC1 leading to altered cell growth, metabolism, and proliferation (Huang and Manning, 2008).
In post-mitotic neurons, Tsc1/2-mTORC1 signaling controls neuronal communication via regulation of membrane excitability, synaptic transmission, and synaptic plasticity (Bateup et al., 2011(Bateup et al., , 2013Benthall et al., 2018;Ehninger et al., 2008;Normand et al., 2013;Tavazoie et al., 2005;Tsai et al., 2012;Weston et al., 2014;Yang et al., 2012). Loss-of-function mutations in negative regulators of mTORC1 cause neurodevelopmental syndromes associated with benign tumors in multiple organs and significant neurological and psychiatric impairments. In particular, patients with Tuberous Sclerosis Complex (TSC), caused by mutations in TSC1 or TSC2, frequently present with neuropsychiatric conditions including autism spectrum disorder (ASD), attention deficit hyperactivity disorder (ADHD), anxiety, and aggression, which are collectively termed TAND (TSC-Associated-Neuropsychiatric-Disorders) (de Vries et al., 2015). In contrast to the well-studied mechanisms of tumor formation in TSC, less is known about how mutations in TSC1/2 and deregulation of mTORC1 signaling cause neuropsychiatric and behavioral abnormalities. In particular, the specific neuronal populations responsible for TAND and the functional consequences of mTOR-pathway mutations on these neurons remain to be defined.
Given the involvement of DA signaling in many of the psychiatric conditions associated with TSC and other mTOR-related disorders, we hypothesized that changes in DA signaling may be central to TAND. To test this, we selectively deleted Tsc1 from mouse DA neurons and investigated how developmental deregulation of mTORC1 affects midbrain DA neuron function.
This approach also allowed us to test whether activation of mTORC1 signaling in DA neurons alone was sufficient to drive TAND-related behavioral phenotypes.
We found that deletion of Tsc1 from DA neurons profoundly affected their structural and functional properties resulting in significant impairments in DA release. This dopaminergic deficit led to reduced cognitive flexibility in the absence of changes to motor, social, or affective behaviors. Our findings identify mTORC1 as a critical regulator of midbrain DA neuron output and suggest that cognitive flexibility deficits in mTOR-related disorders may be driven by changes in DA signaling.

Loss of Tsc1 from DA neurons causes somatic and dendritic hypertrophy
To selectively activate mTORC1 signaling in DA neurons, we conditionally deleted Tsc1, which results in loss of function of the Tsc1/2 complex (Kwiatkowski et al., 2002) (Figure S1A), from DA neurons using DAT IRES Cre mice (Bäckman et al., 2006). For all experiments, mice were heterozygous for DAT IRES Cre and had either wild-type ("DA-Tsc1 WT") or homozygous floxed alleles of Tsc1 ("DA-Tsc1 KO") ( Figure S1B). To confirm that loss of Tsc1 resulted in functional activation of mTORC1 signaling in midbrain DA neurons, we quantified phosphorylation of the mTORC1 pathway target ribosomal protein S6 (p-S6) in tyrosine hydroxylase ( In addition to somatic hypertrophy, altered dendrite branching has been observed in neurons with mutations in mTORC1 regulators (Benthall et al., 2018;Goto et al., 2011;Kwon et al., 2006;Weston et al., 2014). To investigate how loss of Tsc1 affects the dendrites of DA neurons, we performed Sholl analysis on three-dimensional reconstructions of neurobiotin-filled neurons ( Figure S1I). We found that SNc and VTA DA neurons in DA-Tsc1 KO mice had more complex dendrites and increased total dendritic length ( Figure 1G-I and S1F-H). Compared to WT neurons, the dendrites of SNc DA-Tsc1 KO neurons extended about the same radial distance from the soma ( Figure 1H), but were longer in VTA DA-Tsc1 KO neurons ( Figure   S1G). Together these data demonstrate that loss of Tsc1, and constitutive activation of mTORC1 signaling, strongly impact the somatodendritic architecture of midbrain DA neurons.

DA neurons exhibit reduced intrinsic excitability and altered action potential shape following deletion of Tsc1
The structural properties of neurons dictate their function. Therefore, we tested whether the extensive somatodendritic remodeling we observed in DA-Tsc1 KO neurons resulted in a change to their intrinsic excitability. To visualize DA neurons in acute slices for electrophysiology recordings, we bred DA-Tsc1 KO mice to 'reporter mice' expressing Credependent tdTomato (Ai9 line; Madisen et al., 2010) ( Figure S1B). Consistent with their somatodendritic hypertrophy, SNc and VTA DA-Tsc1 KO neurons had significantly increased membrane capacitance and decreased membrane resistance compared to WT neurons (Figures 1J,K and S1J,K and Tables S1 and S2). We injected steps of depolarizing current to generate excitability curves and found that, consistent with the changes in passive membrane properties, SNc and VTA DA-Tsc1 KO cells were hypo-excitable compared to DA-Tsc1 WT (Figures 1L,M and S1 L,M and Tables S1 and S2). In particular, significantly larger current step amplitudes were required to evoke action potential (AP) firing in DA-Tsc1 KO neurons ( Figure   1M and S1M). At current amplitudes up to 400 pA, DA-Tsc1 KO neurons fired fewer APs than DA-Tsc1 WT neurons; however, at the 500 and 600 pA current steps, DA-Tsc1 KO neurons had similar firing rates as DA-Tsc1 WT cells ( Figures 1M and S1M). This suggests that higher currents are required to drive increased spiking in DA-Tsc1 KO neurons but that these neurons may still be able to fire a 'burst' of APs given a sufficiently large excitatory input.
DA neurons have been classically defined by their relatively long AP duration and prominent after-hyperpolarization (Bean, 2007). A variety of sodium, calcium and potassium channels control DA neuron firing properties (Gantz et al., 2018). Since mTORC1 signaling can regulate ion channel expression (Raab-Graham et al., 2006;Yang et al., 2012), we examined the properties of single APs evoked by positive current injection in DA-Tsc1 KO neurons. Tsc1 deletion caused a substantial change in the AP waveform of DA neurons (Figures 1N-Q and S1N-Q). Specifically, the AP duration of SNc and VTA DA-Tsc1 KO neurons was significantly shorter and threshold was reduced in SNc neurons compared to WT (Figures 1N,O and S1N,O and Tables S1 and S2). These changes are indicative of altered potassium channel conductances that control repolarization (Kimm et al., 2015). Notably, these results are similar to those of a prior study that reported shorter duration APs in Tsc1 knock-out thalamic neurons (Normand et al., 2013).

Loss of Tsc1 causes severe impairments in striatal DA release
The AP firing of DA neurons does not linearly translate into axonal DA release events (Rice et al., 2011;Sulzer et al., 2016). Therefore, a direct interrogation of DA release is necessary to determine how mTORC1 hyperactivity impacts dopaminergic neurotransmission.
To this end, we used fast-scan cyclic voltammetry (FCV) at carbon fiber microelectrodes (CFM) to monitor electrically-evoked DA release in the striatum, the brain region with the most dense dopaminergic innervation. To directly compare DA transients between the two genotypes, we performed experiments with paired striatal slices from DA-Tsc1 WT and DA-Tsc1 KO mice with interleaved recordings using the same CFM.
Sampling of extracellular dopamine release ([DA] o ) evoked by electrical stimulation at different striatal locations revealed significant reductions in peak [DA] o amplitudes throughout the dorsal striatum, but not in the nucleus accumbens core (p = 0.1650, paired t-test), in DA-Tsc1 KO mice compared to DA-Tsc1 WT (Figures 2A,B). Decreased DA release was found with both single pulse (Figure 2A,B) and high-frequency burst stimulation (4 pulses at 100 Hz, Figure   S2A,B). Because peak-evoked [DA] o in the nucleus accumbens was not significantly altered by Tsc1 deletion, we focused subsequent analyses on the dorsal striatum. Combining all dorsal striatal sites per mouse, we found an average 60% reduction in single pulse ( Figure 2C,D) and 45% reduction in high-frequency burst-evoked [DA] o ( Figure S2C,D) following Tsc1 deletion. We tested multiple stimulation intensities and found that DA release deficits in DA-Tsc1 KO slices were exacerbated at low stimulation intensities, consistent with the intrinsic hypoexcitability observed at the cell body, while peak-evoked [DA] o plateaued at similar stimulation intensities as in WT ( Figure S2E).
Striatal DA release is tightly regulated by striatal cholinergic interneurons (ChIs), which control release probability and can drive axonal DA release directly (Rice et al., 2011;Sulzer et al., 2016). Therefore, we repeated the FCV experiments with cholinergic transmission blockers to test whether the DA release deficits in DA-Tsc1 KO mice were cell autonomous. We used dihydro-ß-erythroidine (DHßE) to block ß2-contaning nicotinic acetylcholine receptors (nAChRs) on DA axon terminals or oxotremorine-M (oxo-M) to activate muscarinic autoreceptors on ChIs, which inhibit acetylcholine release ( Figure S2F). We found that the single pulse-evoked release deficits in DA-Tsc1 KO mice persisted after blockade of cholinergic transmission ( Figure 2C,D).
The reduction in high-frequency burst-evoked release in DA-Tsc1 KO mice was more variable in cholinergic blockers ( Figure S2C,D). However, this is expected for pulse train stimulation paradigms in which DA release is subject to moment-by-moment variations in release probability due to short-term plasticity (Cragg, 2003;Cragg and Rice, 2004;Montague et al., 2004). Overall, we conclude that cell-autonomous changes are responsible for the profound reduction in evoked [DA] o in the dorsal striatum of DA-Tsc1 KO mice.
Changes in D2 receptor occupancy or calcium sensitivity do not account for DA release deficits in DA-Tsc1 KO mice One possible mechanism for a cell autonomous reduction in striatal [DA] o is increased activity or expression of type 2 dopamine (D2) autoreceptors located on DA axons, which inhibit DA release (Sulzer et al., 2016;Zhang and Sulzer, 2012) (Figure S2G). To test this we recorded single pulse-evoked DA release in dorsolateral striatum and applied increasing concentrations of the D2 receptor agonist quinpirole to construct a dose-response curve. Application of quinpirole decreased peak-evoked [DA] o to a similar extent in DA-Tsc1 WT and DA-Tsc1 KO slices (WT IC50=102.7 ± 0.7933 nM, KO IC50=115.9 ± 0.8080 nM, Figure 2E,F). Therefore, it is unlikely that changes in D2 autoreceptor occupancy or expression account for the reduction in evoked DA release upon loss of Tsc1.
Striatal DA release is heavily dependent on voltage-gated calcium channels and is thus sensitive to changes in extracellular calcium levels (Rice et al., 2011;Sulzer et al., 2016). To test whether decreased DA release in DA-Tsc1 KO mice resulted from changes in the calcium sensitivity or calcium coupling of release sites, we tested if increasing extracellular calcium to 5 mM could rescue release deficits. As expected, increasing extracellular calcium led to increased peak-evoked [DA] o in both genotypes compared to normal aCSF containing 2.4 mM Ca 2+ (see Figure 2C,G). However, DA release in DA-Tsc1 KO slices remained ~60% lower compared to DA-Tsc1 WT slices ( Figure 2G,H). Given the inability of increased extracellular calcium to normalize DA release deficits in DA-Tsc1 KO slices, we conclude that calcium sensitivity or coupling are unlikely to be major contributors to the release deficits observed.

DA release probability (P r ) is significantly reduced following Tsc1 deletion
One of the cell-intrinsic features that controls peak-evoked [DA] o is release probability (P r ) (Cragg, 2003;Rice and Cragg, 2004), and a significant reduction in evoked [DA] o may be indicative of compromised P r . To address this possibility, we examined DA P r in paired DA-Tsc1 WT and DA-Tsc1 KO slices using two approaches. First, we compared the ratio of peak [DA] o evoked by 4 stimuli at 100 Hz to that evoked by a single pulse (4p/1p ratio, Figure 2I). The 4p/1p ratio was significantly higher in DA-Tsc1 KO mice compared to WT across all conditions, including drug-free aCSF, nAChR blockade (DHßE), and 5 mM extracellular Ca 2+ ( Figure 2J).
These results indicated reduced initial P r , a conclusion that was supported by the increased frequency-response sensitivity observed in DA-Tsc1 KO mice for short trains of four pulses delivered at 5-100 Hz ( Figure S2H). Enhanced frequency sensitivity and 4p/1p ratio are reflective of the changes in somatic excitability we observed (see Figure 1), where sufficiently large depolarization could drive high firing rates in DA-Tsc1 KO neurons. Given the fundamental importance of the contrast between low level tonic and high level phasic DA signals (Berke, 2018), it is therefore plausible that these two modes of DA signaling to downstream circuitry remain at least partially intact following loss of Tsc1.
We next examined the paired-pulse ratio of release events evoked by single stimuli delivered three seconds apart, in the presence of DHßE to remove cholinergic modulation and 5 mM Ca 2+ to maximize release probability ( Figure 2K). The second-to-first peak ratio was significantly higher in DA-Tsc1 KO slices compared to DA-Tsc1 WT slices ( Figure 2L). This indicates reduced suppression of DA release with repeated stimulation, again pointing to a reduction in initial dopamine P r with Tsc1 loss.
During the FCV experiments, we observed that clearance of [DA] o , as measured by the falling phase of the DA transient, was faster in DA-Tsc1 KO slices ( Figure S2I). Termination of the DA signal in the striatum is primarily mediated by uptake of released neurotransmitter back into the axon by the dopamine active transporter (DAT) (Hoffman et al., 1998;Nirenberg et al., 1996). Since uptake is exponentially related to the substrate concentration and DAT expression can vary by striatal region (Rice and Cragg, 2008;Rice et al., 2011), we identified concentration-and striatal region-matched transients from DA-Tsc1 WT and DA-Tsc1 KO slices and examined their decay kinetics. Curve-fit analysis of these transients showed that DA reuptake was significantly faster in DA-Tsc1 KO animals ( Figure S2J). Combined, these findings suggest that Tsc1 loss from DA neurons strongly impairs their ability to influence striatal targets through the release of DA.

DA neurons survive, innervate striatum, and have increased DA production following loss of Tsc1
To test whether the reduction in evoked DA release was a result of decreased numbers of DA neurons in DA-Tsc1 KO mice, we performed cell counts of tdTomato-labeled neurons in the SNc and VTA ( Figure 3A,B). At three months of age, the number of tdTomato-expressing neurons was not significantly different between littermate DA-Tsc1 WT and DA-Tsc1 KO mice (Figure 3C,p=0.4343,unpaired,. Therefore, deletion of Tsc1 around embryonic day 15, when the first recombination events are detectable in the midbrain (Bäckman et al., 2006), does not strongly impact the development or survival of DA neurons into young adulthood.
To test whether innervation of the striatum by DA axons might be impacted by loss of Tsc1, we analyzed tdTomato and TH expression in the dorsolateral striatum by immunofluorescence ( Figure 3D-M). While tdTomato+ axons were present throughout the striatum in DA-Tsc1 KO mice, tdTomato fluorescence intensity was reduced compared to WT ( Figure 3D-H). By contrast, levels of TH, the rate limiting enzyme necessary for DA synthesis, were significantly increased in DA-Tsc1 KO striatal axons ( Figure 3I-M). To further explore these findings, we quantified tdTomato and TH levels in DA neuron cell bodies and found similarly decreased tdTomato and increased TH expression following loss of Tsc1 ( Figures 3N-Q). Together, these results suggest that 1) the expression of tdTomato and TH are differentially regulated in the context of high mTORC1 signaling and 2) while dopaminergic axons still innervate the striatum in DA-Tsc1 KO mice, precise quantification of axon density is confounded by changes in axonal protein expression.
To investigate whether striatal axons in DA-Tsc1 KO mice have the capacity to synthesize and package DA, we harvested whole striata for western blot analysis and examined expression levels of TH and the vesicular monoamine transporter-2 (VMAT-2), which is responsible for loading DA into vesicles (Cartier et al., 2010). Consistent with the immunofluorescence results, we observed a significant increase in TH expression in the striatum of DA-Tsc1 KO mice compared to WT littermates, while VMAT-2 levels were unchanged (Fig 3R,S). These results indicate that DA-Tsc1 KO striatal axons have two of the key components necessary to synthesize and package DA into synaptic vesicles.
Higher TH levels suggest that DA-Tsc1 KO neurons may have increased capacity to synthesize DA. To examine this, we analyzed DA tissue content in the dorsal striatum using high-performance liquid chromatography (HPLC) with electrochemical detection. Consistent with high levels of TH, the total striatal tissue content of DA was significantly elevated DA-Tsc1 KO mice compared to WT ( Figure 3T). However, levels of the primary DA metabolite DOPAC were not significantly affected ( Figure 3T). Since the ratio of DOPAC to DA per mouse was unchanged by loss of Tsc1 ( Figure 3T), we conclude that the high DA levels were due to increased DA synthesis rather than reduced turnover. The primary serotonin metabolite 5hydroxyindoleacetic acid (5HIAA) could be detected in a subset of samples and was unchanged by Tsc1 deletion (DA-Tsc1 KO=101.6% ± 15.25% of WT, p=0.966, unpaired, two-tailed t test), indicating selective alterations to the DA system.

Loss of Tsc1 increases DA axon terminal size and reduces vesicle density
Our analyses showed a significant impairment in striatal DA release in DA-Tsc1 KO mice despite elevated striatal DA tissue content. To investigate whether structural changes to the axon terminals of DA neurons could be responsible for the release deficits, we performed electron microscopy (EM) analysis of DA terminals in the dorsolateral striatum identified by TH immunogold labeling (referred to as "profiles", Figure 4A-D). DA neuron axons arborize extensively in the striatum, releasing DA at both synaptic and non-synaptic sites (Matsuda et al., 2009;Moss and Bolam, 2008;Rice and Cragg, 2008). We found that the majority of THexpressing axon profiles were non-synaptic ( Figure 4A,B), defined by the lack of a discernable synaptic membrane specialization. The percentage of synaptic profiles was similar between DA-Tsc1 WT and KO mice (11.25% and 10.25%, respectively) and both synaptic and non-synaptic axon profiles were included in the main analysis. Table S3 provides a separate analysis of synaptic profiles.
Consistent with the somatodendritic hypertrophy we observed (see Figure 1), axonal profile area was significantly enlarged in DA-Tsc1 KO mice ( Figure 4E and Table S3). We quantified the number of vesicles associated with each TH+ profile and found that vesicle number was significantly decreased in DA-Tsc1 KO profiles ( Figure 4F). The combination of increased profile area, together with reduced vesicles per profile, led to a significant decrease in vesicle density DA-Tsc1 KO axons ( Figure 4G). To determine the spatial arrangement of vesicles in each sampled axon, we measured the shortest distance of each vesicle to the profile perimeter (i.e. the plasma membrane), and the distance of each vesicle to every other vesicle within the profile ( Figure 4H). We found that vesicle distance to the perimeter was significantly higher in DA-Tsc1 KO terminals compared to WT ( Figure 4I). In addition, the average intervesicle distance was increased in DA-Tsc1 KO axons (DA-Tsc1 WT: 273.1 ± 0.14 nm, DA-Tsc1 KO: 303.7 ± 0.19 nm, p<0.0001, Kolmogorov-Smirnov test), indicating that vesicles were less clustered. When broken down into 50 nm bins, the histogram of inter-vesicle distances was shifted to the right for DA-Tsc1 KO terminals revealing significantly fewer vesicles within 100-200 nm of each other compared to WT terminals ( Figure 4J). Thus, with reduced vesicle density and clustering, and vesicles further removed from the plasma membrane, the EM analysis of DA-Tsc1 KO axons revealed structural correlates of the observed deficits in DA release.

Locomotor, affective, and social behaviors are normal while cognitive flexibility is reduced in DA-Tsc1 KO mice
DA is involved in a variety of behaviors, from movement to reward learning to cognition (Schultz, 2005;Steinberg et al., 2013;Wickens et al., 2007). To determine whether the marked perturbations in the functional properties of DA neurons affected the behavior of DA-Tsc1 KO mice, we performed a panel of behavior assays that are sensitive to dopaminergic function. Table S4 reports all behavior results and Table S5 shows behavioral analysis by sex.
To examine general exploratory behavior and locomotor activity, DA-Tsc1 KO mice and WT littermates were tested in the open field ( Figure 5A). Deletion of Tsc1 from DA neurons did not affect exploratory locomotor behavior, measured as the distance traveled in the first 10 minutes of the test (DA-Tsc1 WT: 23.44 ± 2.57 m, DA-Tsc1 KO: 22.62 ± 2.08 m, p=0.8046, unpaired, two-tailed t-test). The total distance traveled over 60 minutes and the mean speed were also similar between DA-Tsc1 WT and DA-Tsc1 KO mice ( Figure 5B,C). We observed no changes in rearing behavior or time spent grooming between the two groups ( Figure 5D,E). Therefore, despite a ~60% reduction in evoked DA release, locomotor activity was unaffected by DA neuron-specific loss of Tsc1.
To test whether motor coordination or motor learning were altered in DA-Tsc1 KO mice, we performed the accelerating rotarod test (Rothwell et al., 2014). On days one and two of training, the acceleration ramped from 5 to 40 rpm over the course of five minutes and on days three and four, the acceleration ramped from 10 to 80 rpm. We found no differences between genotypes in either initial motor coordination (defined as the intercept of the line of best fit from trial 1 to 9; DA-Tsc1 WT 10.55 ± 1.19 rpm, DA-Tsc1 KO 13.37 ± 1.39 rpm, p=0.2904, Mann-Whitney test) or motor learning, measured as the slope of performance from the first to last trial for each mouse (DA-Tsc1 WT 1.92 ± 0.25, DA-Tsc1 KO 1.32 ± 0.19, p=0.0637, unpaired t test, Figure 5F). Together, our data indicate that these aspects of motor behavior are robust in the face of changes in the DA system caused by Tsc1 deletion.
In addition to motor behavior, dopamine signaling modulates affective behaviors including anxiety and sociability (Manduca et al., 2016;Zweifel et al., 2011). We found no differences in measures of anxiety in the open field test and elevated plus maze between DA-Tsc1 WT and DA-Tsc1 KO mice ( Figure S3A-H). Sociability was also normal in DA-Tsc1 KO mice as assessed by the three-chamber social approach test ( Figure S3I-K).
The lack of motor deficits in DA-Tsc1 KO mice suggests that the motor system can compensate for developmental alterations in DA function, which is reflective of findings in Parkinson's disease (PD) patients that motor impairments are only manifest when there is a >60-70% loss of DA terminals in the caudate-putamen (Cheng et al., 2010). However, cognitive dysfunction can be observed early in PD (Watson and Leverenz, 2010;Williams-Gray et al., 2007). In addition, cognitive flexibility, the ability to adapt behavior in response to a changing environment, is disrupted both in PD and in a range of neuropsychiatric conditions involving dopaminergic dysfunction (Cools et al., 2001;Klanker et al., 2013).
To examine associative learning and cognitive flexibility in DA-Tsc1 KO mice, we used the four-choice odor-based reversal task ( Figure 5G) (Johnson et al., 2016), in which reversal learning is sensitive to changes in DA neuron function (Luo et al., 2016). Only male mice were used for this test, consistent with a prior study (Luo et al., 2016). In this test, mice learn to dig for a buried food reward in a pot of scented wood shavings. In the acquisition phase, mice learn an initial odor-reward pairing and the number of trials to reach criterion (8 out of 10 sequential trials correct) is a measure of discrimination learning. Following overnight consolidation, mice undergo a reversal learning phase in which the reward contingency is changed and a previously unrewarded odor is rewarded.
We found that DA-Tsc1 KO mice exhibited normal acquisition learning on the first day of testing, reaching criterion in a similar number of trials as littermate controls ( Figure 5H). By contrast, when the rewarded odor was changed, DA-Tsc1 KO mice made significantly more total errors ( Figure 5I) and tended to require more trials to learn the new odor-reward pairing ( Figure 5J). For both genotypes, the majority of errors were perseverative, meaning that the mice continued to choose the initially learned odor even when it was no longer rewarded.
However, DA-Tsc1 KO mice made significantly more perseverative errors than littermate controls ( Figure 5K). The number of novel errors (choosing a newly introduced odor), irrelevant errors (choosing an odor that was never rewarded), or omissions was not significantly different between genotypes ( Figure 5K). Thus, DA-Tsc1 KO mice exhibited a selective deficit in cognitive flexibility, wherein they were slower to update their behavioral strategy in the face of changed environmental conditions.

Genetic reduction of Rptor constrains mTORC1 signaling in DA-Tsc1 KO neurons
Our data show that loss of Tsc1 from DA neurons strongly alters their structure, impairs their ability to release DA, and leads to reduced cognitive flexibility. To test whether these phenotypes could be prevented by suppression of mTORC1 signaling, we conditionally deleted one or two copies of the Rptor gene encoding the mTOR binding protein Raptor from DA signaling in DA neurons in the context of Tsc1 loss; however, this partial reduction of mTORC1 activity was not sufficient to prevent somatic hypertrophy.

Heterozygous Rptor deletion prevents DA release deficits in DA-Tsc1 KO mice and increases cognitive flexibility
To test whether the functional deficits in DA release in DA-Tsc1 KO mice could be prevented by genetic reduction of mTORC1 signaling, we performed FCV experiments in paired DA-Tsc1-KO/Rptor-WT and DA-Tsc1-KO/Rptor-Het striatal slices. We found that DA-Tsc1-KO/Rptor-Het slices had significantly higher evoked [DA] o throughout the dorsal striatum compared to DA-Tsc1-KO/Rptor-WT slices ( Figure 6J,K). On average, evoked DA levels were two-fold higher in the dorsal striatum of DA-Tsc1-KO/Rptor-Het mice compared to DA-Tsc1-KO/Rptor-WT ( Figure 6L,M). We compared the 4p/1p ratio between the two experimental groups to estimate DA release probability, and found that DA-Tsc1-KO/Rptor-Het slices had significantly lower 4p/1p ratio compared to DA-Tsc1-KO/Rptor-WT ( Figure 6N). Therefore, heterozygous Rptor deletion alleviated DA release deficits, likely via a significant increase in dopamine P r .
Given the functional improvements in DA release that we observed in DA-Tsc1-KO/Rptor-Het mice, we tested whether these mice had improved cognitive flexibility, as assessed by their performance in the four-choice reversal learning task. We observed no significant differences between DA-Tsc1-KO/Rptor-Het and DA-Tsc1-KO/Rptor-WT male littermate mice during acquisition learning ( Figure 6O). However, during reversal learning, DA-Tsc1-KO/Rptor-Het animals made fewer total errors ( Figure 6P), reached criterion in significantly fewer trials ( Figure 6Q), and made fewer perseverative errors ( Figure 6R).
Together, these data demonstrate that cell type-specific genetic reduction of mTORC1 signaling in DA-Tsc1-KO mice can improve evoked DA release and counter reductions in cognitive flexibility.

Discussion
Dopaminergic circuits play a central role in the cognitive functions and behaviors that are impacted in mTOR-related neurodevelopmental disorders. Here we used genetic mouse models to investigate how deregulation of mTORC1 signaling affects the functional properties of DA neurons. We find that DA neuron-specific loss of Tsc1, a key negative regulator of mTORC1, leads to constitutive activation of mTORC1 signaling in DA neurons, profoundly affecting their somatodendritic and axonal architecture, intrinsic excitability, and ability to release DA in the dorsal striatum. Mice lacking Tsc1 selectively in DA neurons exhibit reduced cognitive flexibility in the absence of changes to motor function, consistent with patients with mutations in TSC1 or TSC2 (de Vries et al., 2015). We show that DA release deficits and cognitive inflexibility in DA-Tsc1 KO mice can be prevented by genetic reduction of mTORC1 signaling. Together these data provide insight into the cellular origins of the behavioral manifestations associated with deregulated mTORC1 signaling, and suggest a potential locus for intervention.

mTOR signaling in DA neurons
The impact of mTOR signaling on DA neuron physiology has primarily been studied in the context of addiction. Specifically, drugs of abuse have been shown to activate mTORC1 signaling in midbrain DA neurons (Collo et al., 2012(Collo et al., , 2013Dayas et al., 2012;Neasta et al., 2014;Wu et al., 2011), and pharmacologic or genetic inhibition of mTOR changes synaptic function in VTA neurons and attenuates the behavioral effects of drugs of abuse (Dayas et al., 2012;Liu et al., 2018b;Neasta et al., 2014;Wu et al., 2011). mTORC2 signaling has also been implicated in reward tolerance to opiate drugs via changes in VTA neuron size and excitability (Mazei-Robison et al., 2011). mTORC1 signaling has also been studied in the context of PD, where mTOR activation has been shown to acutely promote DA neuron survival, possibly via increased protein synthesis at the site of injury and/or inhibition of autophagic processes (Cheng et al., 2011;Diaz-Ruiz et al., 2009;Domanskyi et al., 2011;Kim et al., 2012;Lan et al., 2017;Malagelada et al., 2010;Xu et al., 2014;Zhou et al., 2015). Here, we build upon this literature and show, for the first time, that chronic developmental activation of mTORC1 signaling arising from Tsc1 loss leads to altered DA neuron structure at multiple levels, inducing somatic, dendritic, and axonal hypertrophy. These morphological changes are associated with decreased intrinsic excitability and impaired DA release. Together with prior studies, this work establishes mTOR as an essential regulator of DA neuron output, which influences multiple aspects of DA neuron function, including membrane excitability and pre-and post-synaptic properties.

DA neuron function requires balanced mTORC1 signaling
We found that chronic developmental activation of mTORC1 signaling profoundly affected DA neuron structure and significantly impaired neurotransmitter release. We attempted to prevent these alterations by inhibiting mTORC1 signaling via deletion of the gene encoding the obligate mTORC1 protein Raptor in DA-Tsc1 KO mice. However, we found identical deficits in evoked DA release in DA-Tsc1-KO/Rptor-KO mice as with Tsc1 deletion alone. Release deficits persisted despite complete reversal of Tsc1 KO-induced somatic hypertrophy. The failure of complete mTORC1 suppression to prevent release deficits is consistent with prior reports showing that acute treatment of striatal slices with the mTOR inhibitor rapamycin or Cre virus-mediated deletion of mTOR from VTA neurons both decrease evoked DA release (Hernandez et al., 2012;Liu et al., 2018b). By contrast, we found that heterozygous loss of Rptor in DA-Tsc1 KO mice, while not sufficient to prevent somatic hypertrophy, significantly improved DA release and countered reduced cognitive flexibility. Together, these results indicate that tight regulation of mTORC1 signaling is essential for DA neuron output as too much or too little mTORC1 signaling is similarly detrimental to DA release. In addition, our findings show that changes in DA neuron soma size can be decoupled from alterations in axonal DA release

Mechanisms of DA release deficits in DA-Tsc1 KO mice
DA neurons have specialized intrinsic and extrinsic mechanisms to control DA release (Liu et al., 2018a;Sulzer et al., 2016). We investigated several potential mechanisms for the DA release deficits in DA-Tsc1 KO mice. We first ruled out DA neuron death, striatal denervation, and non-cell-autonomous effects on the striatal cholinergic system as potential explanations.
We also excluded changes in D2 autoreceptor expression or occupancy. In addition, we found that increasing extracellular calcium was not sufficient to normalize release, suggesting that mechanisms other than (or in addition to) calcium sensitivity or coupling to the release machinery were involved. Importantly, we showed that the axons of DA neurons in DA-Tsc1 KO mice had key molecular machinery to synthesize and package DA into vesicles and, in fact, had higher TH levels, as reported previously for DA neurons with increased mTORC1 signaling due to Pten deletion (Diaz-Ruiz et al., 2009;Domanskyi et al., 2011). High TH levels implied greater capacity to synthesize DA. Indeed, we found that the striatal tissue content of DA was significantly increased in DA-Tsc1 KO mice compared to controls. This may have been driven by increased mTORC1-dependent protein synthesis of TH. Alternatively, TH expression could have been increased independent of mTORC1, as a homeostatic mechanism to compensate for decreased DA release.
The most plausible mechanisms for impaired DA release in DA-Tsc1 KO mice were ultrastructural changes in DA neuron axons. Specifically, we found that Tsc1 loss highly enlarged TH+ axon terminals in the dorsal striatum. Axon profiles in DA-Tsc1 KO mice also had fewer vesicles. The net result of these changes was a marked reduction in vesicle density. In addition, the vesicles were, on average, further removed from the plasma membrane and were less tightly clustered together. Given the stringent structural requirements for efficient vesicle release, for example vesicle proximity to docking sites (Park et al., 2012), it is likely that the observed ultrastructural changes in DA-Tsc1 KO axons led to reduced P r and impaired DA release. Further, in a mouse model of PD overexpressing alpha-synuclein, abnormal clustering of vesicles alone caused a 30% reduction in evoked striatal DA release (Janezic et al., 2013).
A recent study demonstrated that a subset of striatal DA terminals have active zone protein scaffolds, which contain RIM as an essential component and are required for DA release (Liu et al., 2018a). It will be interesting for future studies to explore the potential contribution of changes in active zone proteins to mTOR-dependent effects on DA release.

Behavioral consequences of mTOR-related dopaminergic dysfunction
A notable finding of our study is that despite severe impairments in dorsal striatal DA release, DA-Tsc1 KO mice did not exhibit gross motor impairments. This was assessed by multiple parameters in the open field (total distance traveled, average speed, rearing) and accelerating rotarod test, which measures motor coordination and learning. While initially surprising, the lack of motor deficits is reminiscent of observations in PD where motor symptoms are only evident after 30-40% of SNc neurons have degenerated and striatal DA axons are reduced by 60-70% (Cheng et al., 2010). However, cognitive deficits can be observed in PD patients even at early stages of the disease, suggesting that cognitive functions may be particularly vulnerable to changes in DA (Watson and Leverenz, 2010;Williams-Gray et al., 2007). In addition, in DA-Tsc1 KO mice, Tsc1 is deleted embryonically, therefore, the basal ganglia circuit develops in the context of reduced dopaminergic output and compensatory mechanisms are likely engaged in dopaminoceptive neurons.
We also found that anxiety and social approach behavior were unchanged in DA-Tsc1 KO mice. These behaviors are modulated by mesolimbic DA projections originating from the VTA (Manduca et al., 2016;Zweifel et al., 2011). The lack of change in these behaviors is consistent with the fact that DA release in the nucleus accumbens (a major projection target of VTA neurons) was not significantly altered by Tsc1 deletion. It is also possible that manipulation of mTORC1 signaling in DA neurons alone is not sufficient to strongly affect social behavior.
This contrasts with cerebellar Purkinje cell-specific deletion of Tsc1, which does impair social behavior in a rapamycin-sensitive manner (Tsai et al., 2012). Together, these studies reveal that different cell types are differentially vulnerable to changes in mTORC1 signaling and that distinct cell types and circuits are likely responsible for specific aspects of mTOR-related disorders.
Cognitive flexibility, defined as the ability to adapt behavior in the face of changing environmental demands, is commonly affected in psychiatric disorders, in particular ASD and ADHD (D'Cruz et al., 2013;Dajani and Uddin, 2015;Sergeant et al., 2002). TSC patients frequently present with impairments in executive control processes, which include cognitive flexibility (Prather and de Vries, 2004;de Vries et al., 2015). Here, we found that DA neuronspecific deletion of Tsc1 was sufficient to reduce cognitive flexibility in a reversal-learning task.
This impairment was specific to reversal learning as initial discrimination learning was intact.
Notably, during reversal learning, DA-Tsc1 KO mice made more perseverative errors, in which they continued to return to the previously learned odor even though it was no longer rewarded.
Perseverative errors reflect an inability to flexibly adapt and "update" a previously learned association. Such a deficit may contribute to the inflexible behaviors often observed in patients with TSC (de Vries et al., 2015).
Our findings are consistent with prior reports showing that brain-wide heterozygous deletion of Tsc2 or disruption of the translational regulator eIF4E, which increases protein synthesis, impair reversal but not acquisition learning in water maze tasks (Potter et al., 2013;Santini et al., 2013). Our work suggests that these types of impairments may be attributable to dopaminergic dysfunction, consistent with the involvement of DA in cognitive flexibility in rodents and humans (Darvas and Palmiter, 2011;Klanker et al., 2013). Our findings are also in agreement with studies showing that moderate reductions in dorsal striatal DA levels in rats (as achieved by toxin injections) can impair cognitive flexibility without changing motor behavior (Grospe et al., 2018;O'Neill and Brown, 2007).
Taken together, our work demonstrates that developmental deregulation of mTORC1 signaling profoundly alters key functional properties of DA neurons. These changes lead to reduced dopaminergic output and a selective deficit in cognitive flexibility. Our findings have implications for cognitive dysfunction in neurodevelopmental disorders associated with altered mTORC1 signaling and suggest that these phenotypes may be driven by dopaminergic dysfunction.

Acknowledgments
We thank the members of the Bateup lab for their feedback on this project. We thank Dr. Linda Wilbrecht and members of the Wilbrecht lab for assistance with the four-choice reversal learning task. We thank Dr. Bernardo Sabatini for the original idea to investigate Tsc1 loss in dopamine neurons. We thank Ben Micklem for his technical support in maintaining the electron microscope facility at the MRC Brain Network Dynamics Unit. This work was supported by a

Declaration of Interests
The authors declare no competing financial or non-financial interests.   Figure S1 and Tables S1 and S2 for complete electrophysiology results.      Table S3 for complete EM analysis.   Figure S4 and Table S4 for all p values for behavior tests.

CONTACT FOR REAGENT AND RESOURCE SHARING
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr. Helen Bateup (bateup@berkeley.edu)

Mice
All animal procedures were carried out in accordance with protocols approved by the Mice were switched to the reverse light cycle at least two weeks prior to testing and were tested during the dark phase.  Passive membrane properties were recorded in voltage clamp with the membrane held at -70 mV. Positive current steps (2 s, 25-600 pA) were applied to generate an input-output curve from a baseline membrane potential of -70 mV as maintained by hyperpolarizing current injection.
For whole-cell recordings, series resistance was <30 MΩ and liquid junction potential was not corrected.

Fast-Scan Cyclic Voltammetry (FCV)
DA release was monitored using fast-scan cyclic voltammetry (FCV) in acute coronal slices as described previously (Rice and Cragg, 2004;Threlfell et al., 2010Threlfell et al., , 2012. Paired sitesampling recordings were performed in one reference and one experimental animal in a sexand age-matched mouse pair recorded on a given day.

Electrical stimulation and drug application
Following 30 min slice equilibration in the recording chamber, DA release was evoked, using square wave pulses (0.6 mA pulse amplitude, 2 ms pulse duration) controlled by a

Confocal Microscopy
Images of 30 μm sections processed for immunohistochemistry were acquired using a Zeiss LSM 710 AxioObserver confocal microscope fitted with a motorized XY-stage for tile scanning. A 20x/0.8 N.A. air objective was used to generate tile scans (424x424 m per tile) of one hemisphere (5x3 grid) or the entire midbrain (10x3 grid). 405 nm, 488 nm, 561 nm and 633 nm laser were used. Z-stack images captured the entire thickness of the slice at 1.10-1.25 m steps. Laser power settings and acquisition parameters were kept constant for all experimental conditions.

High-Performance Liquid Chromatography (HPLC) with Electrochemical Detection
Tissue DA content was measured by HPLC with electrochemical detection in tissue punches from dorsal striatum. Male and female mice (P60-P75) were deeply anesthetized by isoflurane and decapitated. 300 m thick coronal slices of striatum were prepared on a vibratome (Leica VT1000 S) in ice cold high Mg 2+ ACSF containing in mM: 85 NaCl, 25 NaHCO 3 , 2.5 KCl, 1.25 NaH 2 PO 4 , 0.5 CaCl 2 , 7 MgCl 2 , 10 glucose, and 65 sucrose. Slices were recovered for 1 hour at RT in ACSF containing in mM: 130 NaCl, 25 NaHCO 3 , 2.5 KCl, 1.25 NaH 2 PO 4 , 2 CaCl 2 , 2 MgCl 2 , and 10 glucose. All solutions were continuously bubbled with 95%O 2 and 5% CO 2 . Following slice recovery, tissue punches from the dorsal (2.5 mm diameter) striatum from two brain slices per animal were taken and stored at -80°C in 200 µl 0.1 M HClO4. On the day of analysis, samples were thawed, homogenized and centrifuged at 16,000 x g for 15 min at 4°C. The supernatant was analyzed for DA, 3,4-Dihydroxyphenylacetic acid (DOPAC), and 5-Hydroxyindoleacetic acid (5HIAA) content using HPLC with electrochemical detection. Analytes were separated using a 4.6 x 150 mm Microsorb C18 reverse-phase column (Varian or Agilent) and detected using a Decade II SDS electrochemical detector with a Glassy carbon working electrode (Antec Leyden) set at +0.7 V with respect to a Ag/AgCl reference electrode. The mobile phase consisted of 13% methanol (v/v), 0.12 M NaH2PO4, 0.5 mM OSA, 0.8 mM EDTA, pH 4.8, and the flow rate was fixed at 1 ml/min.
Analyte measurements were normalized to tissue punch volume (pmol/mm 3 ). HPLC analysis was repeated in two independent experiments.

Western Blotting
Male and female mice (P60-P90) were deeply anesthetized by isoflurane and decapitated. Bilateral striata were rapidly dissected on ice, flash-frozen in liquid nitrogen and stored at -80°C. On the day of analysis, frozen samples were sonicated until homogenized (QSonica Q55) in 500 l lysis buffer containing 1% SDS in 1x PBS with Halt phosphatase inhibitor cocktail (Fisher: PI78420) and Complete mini EDTA-free protease inhibitor cocktail (Roche: 4693159001). Sample homogenates were then boiled on a heat block at 95°C for 10 min, allowed to cool down to RT, and total protein content was determined by BCA assay (Fisher: PI23227). Following BCA assay, protein homogenates were mixed with 4x Laemmli sample buffer (Bio-Rad: 161-0747).10-15 μg of protein were loaded onto 4-15% Criterion TGX gels (Bio-Rad: 5671084). Proteins were transferred to PVDF membrane (BioRad: 1620177) at 4°C overnight using the Bio-Rad Criterion Blotter (12 V constant voltage). The membranes were blocked in 5% milk in 1x TBS with 1% Tween (TBS-T) for 1h at RT, and incubated with primary antibodies diluted in 5% milk in TBS-T overnight at 4°C. The following day, after 3 x 10 min washes with TBS-T, the membranes were incubated with HRP-conjugated secondary antibodies for 1h at RT. Following 6 x 10 min washes, the membranes were incubated with chemiluminesence substrate (Perkin-Elmer: NEL105001EA) for 1 min and exposed to GE its viscosity, and sections were placed on microscope slides, coverslipped, and the resin cured at 65°C for ~70 h (Doig et al., 2010(Doig et al., , 2014. Experimenters were blind to genotype throughout tissue processing and data analysis.

Electron microscopy
All sections were examined under a light microscope and areas from the dorsolateral striatum were cut from the slide, glued to the top of a resin block, and trimmed with razor blades. For each mouse, striatal regions from two sections across the rostro-caudal plane were examined. Serial sections, ~50 nm thick, were cut using an ultramicrotome (Leica EM UC6), collected on pioloform-coated, single-slot copper grids (Agar Scientific), and lead-stained with Reynold's lead citrate to improve contrast for electron-microscopic examination. For each region (block), ultrathin sections from at least two grids were examined. A Philips CM100 electron microscope was used to examine the sections. Analyses of pre-embedded immunogold sections were performed at a minimum of 5 µm from the tissue-resin border (i.e., section surface). The maximum distance from the tissue-resin border examined was determined by the penetration of the gold conjugated antibody together with the angle at which the tissue-resin was sectioned, and was therefore variable. TH-positive structures were systematically analyzed in one of the serial sections on an electron-microscopic grid. At a magnification at which it is not possible to clearly visualize synapses, an area was chosen at random, the magnification was then increased and the first structure positively labeled for TH was digitally recorded (GatanCCD UltraScan US1000 camera; Gatan). TH-positive structures were identified and imaged, in this way, continuing systematically in straight lines across the section, keeping the identified TH-positive structure central within the image frame. For immunogold-labeled structures, the criterion for an immunopositive structure was five or more silver-intensified immunogold particles. The process was continued within the same ultrathin section until 15 THpositive structures were identified and imaged, and was then repeated using a different ultrathin section, on a different grid, from the same block. This was repeated for the second block until a minimum of 60 TH-immunopositive structures were identified and imaged per animal (491 dopaminergic profiles in total, n=4 mice per genotype). Symmetric synapses (Gray's type II) were identified by the presence of presynaptic and postsynaptic membrane specializations, a widened synaptic cleft and cleft material. Any TH-positive structures seen to be forming such symmetric synapses were imaged (Anwar et al., 2011;Janezic et al., 2013).

Behavioral Tests
Behavior testing was carried out in the dark phase of the light cycle under red lights, except the rotarod and four-choice reversal tests, which were carried out in the dark phase under white lights. Mice were habituated to the behavior room for at least 30 min prior to testing and covered by a black-out curtain. Mice were given at least one day in between different tests.
All behavior equipment was cleaned between each trial/mouse with 70% ethanol, and additionally rinsed in diluted soap followed by water at the end of the day. If male and female mice were to be tested on the same day, male mice were run in behavior first, returned to the husbandry room, after which all equipment was thoroughly cleaned, and female mice were brought in for habituation. Behavioral tests were performed with P60-80 male and female mice except for the four-choice test for which P70-130 male animals were used. The experimenter was blind to genotype throughout the testing and scoring procedures.

Open Field Test
Exploratory behavior in a novel environment and general locomotor activity were assessed by a 60 min session in an open field chamber (40 cm L x 40 cm W x 34 cm H) made of transparent plexiglas. Horizontal photobeams to measure rearing were positioned at 9 cm of height. The mouse was placed in the bottom right hand corner of the arena and behavior was recorded using an overhead camera and analyzed using the ANY-maze (Stoelting Co) behavior tracking software. An observer manually scored scratching and grooming behavior during the first 20 minutes of the test.

Rotarod Test
The accelerating rotarod test was used to examine motor coordination, balance and motor learning. Mice were run on a rotarod apparatus (Ugo Basile: 47650) for 4 consecutive days. Three trials were completed per day with a 5 min break between trials. The rotarod was accelerated from 5-40 revolutions per minute (rpm) over 300 s for trials 1-6 (days 1 and 2), and from 10-80 rpm over 300 s for trials 7-12 (days 3 and 4). On the first testing day, mice were acclimated to the apparatus by being placed on the rotarod rotating at a constant 5 rpm for 60 s and returned to their home cage for 5 minutes prior to starting trial 1. Latency to fall, or to rotate off the top of the rotarod barrel, was measured by the rotarod stop-trigger timer.

Four-Choice Odor-Based Reversal Task
The four-choice odor-based reversal task (Johnson et al., 2016) was used to assess learning and cognitive flexibility in adult male mice. Animals were food restricted for 5-6 days in total with unrestricted access to drinking water, and maintained at 85-90% of ad lib feeding body weight. Food was given at the end of the day once testing was completed. Food restriction began 24 h before pre-training. Pots were cleaned with 70% ethanol followed by water. The apparatus was cleaned with diluted soap and water at the end of each testing day.
On the first habituation day of pre-training (day 1), animals were allowed to freely explore the testing arena for 30 minutes and consume small pieces of Cheerios placed inside pots positioned in each of the four corners. On the second shaping day of pre-training (day 2), mice learned to dig to find cereal pieces buried in unscented coarse pine wood shavings (Harts Mountain Corporation, Secaucus, NJ). On the shaping day, animals who failed to consume Cheerio pieces were switched to Fruit Loops, which they consumed successfully. A single pot was used and increasing amounts of unscented wood shavings were used to cover each subsequent cereal reward. The quadrant containing the pot was alternated on each trial and all quadrants were rewarded equally. Trials were untimed and consisted of (in order): two trials with no shavings, two trials with a dusting of shavings, two trials with the pot a quarter full, two trials with the pot half full, and four trials with the cereal piece completely buried by shavings.
The mouse was manually returned to the center cylinder between trials.
On the testing days for odor discrimination ( The reversal phase of the task was completed the following day (day 4). Mice first performed the task with the same rewarded odor as the discrimination day to ensure they learned and remembered the task. After reaching criterion on recall (8 out of 10 consecutive trials correct), the rewarded odor was switched. Perseverative errors were choices to dig in the previously rewarded odor that was no longer rewarded. Novel errors were choices to dig in the pot with the newly introduced odor for reversal testing. Irrelevant errors were choices to dig in the pot that had never been rewarded. Omissions were trials in which the mouse failed to make a digging choice within three minutes from the start of the trial. Total errors are the sum of perseverative, irrelevant, novel and omission errors. Criterion was met when the mouse completed 8 out of 10 consecutive trials correctly.

Elevated Plus Maze Test
Exploratory behavior in a novel environment and generalized anxiety were assessed with elevated plus maze test (Walf and Frye, 2007).

Three Chamber Social Approach Test
Social approach behavior was assessed using the three chamber social approach test (Yang et al., 2011). The testing apparatus was made of clear Plexiglas (60 cm L× 40 cm W × 22 cm H, Stoelting). Two dividing walls had doorways allowing access to three chambers. Age and sex matched C57BL/6J mice were used as novel mice and were habituated to placement inside the wire cups (8 cm diameter, 11 cm tall) prior to testing in either three 10-minute or two 15minute sessions. C57BL/6J mice that displayed persistent agitation, including climbing or biting behavior inside the wire cups, were not used for testing.
The test mouse was placed in the center chamber of the apparatus and allowed to explore and habituate to all three empty chambers for 10 min. The mouse was then placed in the center chamber again for 5 min with access doors closed. During this time, a novel C57BL/6J mouse was placed under a wire cup in one of the side chambers and an empty wire cup was placed on the opposite side (novel object). The chamber doors separating the three chambers were then removed to open access to both left and right chambers. The mouse was allowed to explore the three chambers for 10 min, and allowed to freely interact with the novel object or with the stranger mouse inside the wire cup. The location of the stranger mouse and the novel object were alternated between mice. Behavior was recorded using a video camera positioned directly above the center of the apparatus and analyzed using the ANY-maze (Stoelting Co) behavior tracking software. An observer blind to genotype manually scored time spent sniffing the novel mouse and the novel object during the test. Sniffing was defined as when the mouse's nose was perpendicular to the vertical plane and less than 1 cm away from the wire cup. If the mouse reared on the wire cup, climbed on the wire cup, or bit the wire cup, it was not scored as sniffing. Mice showing a strong side bias during the habituation phase of the test were excluded. Controls were DAT IRES Cre negative littermates of DA-Tsc1 KO mice.
Comparisons were made within genotype.

QUANTIFICATION AND STATISTICAL ANALYSES
Whenever possible, quantification and analyses were performed blind to genotype. All statistical analyses and graphing were performed using GraphPad Prism 6 software. All datasets were first analyzed using D'Agostino and Pearson omnibus normality test, and then parametric or non-parametric two-tailed statistical tests were employed accordingly to determine significance. Significance was set as *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001 unless otherwise indicated in the figure legend. P values were corrected for multiple comparisons. The statistical tests used for each type of experiment are indicated below.
Statistical details for each experiment are reported in the figure legends.

Electrophysiology Data
Data were analyzed in Igor Pro (Wavemetrics) using custom scripts. Evoked action potential properties were measured from the first AP in the train of evoked APs. AP threshold was calculated by taking the second derivative of the action potential. Passive properties and action potential properties were analyzed for significance using unpaired t-tests. Excitability curves were analyzed for significance using a two-way ANOVA with Sidak's multiple comparison test.

Dopamine neuron reconstruction
Morphological parameters, including Sholl intersections and total dendritic length, were extracted using the IMARIS (Bitplane) statistics feature from masks. Total dendrite length was analyzed using an unpaired t-test and Sholl intersections using a repeated measures two-way ANOVA.

Voltammetry Data
Data were pre-processed using AxoScope 10.5 (Molecular Devices) software and analyzed using custom written Excel macros and scripts (Prof S. Cragg  were compared using curve-fit analyses. Current-amplitude and frequency-response data were compared using a two-way ANOVA with Sidak's multiple comparisons test.

Immunohistochemistry
Analyses were performed blind to genotype. Quantification was performed on maxprojected z-stack images. For soma size and p-S6 measurement, soma boundaries were traced manually using Image J (NIH) software in the tyrosine hydroxylase channel to include all clearly identifiable single cells in the SNc or VTA (minimum 100 cells for each area in each slice).
Soma cross-sectional area and fluorescence intensity for the p-S6 channel were extracted automatically in Image J. Cumulative distributions were analyzed using the Kolmogorov-Smirnov test.
For DA neuron counts, every clearly identifiable tdTomato+ cell in both hemispheres was manually counted using the Cell Counter Image J (NIH) plug-in, and analyzed using an unpaired t test. Three anatomically matched sections in the rostro-caudal plane per mouse were counted in three age-and sex-matched mouse pairs.

Western Blotting
Western blot analysis was performed blind to genotype. Bands were quantified by densitometry using Image J (NIH) software. Proteins were normalized to DARPP-32 and expressed as percentage of control within a given experiment. TH data were analyzed using an unpaired t-test with Welch's correction and VMAT-2 data were analyzed using a Mann-Whitney test.

HPLC
Analysis was performed blind to genotype. Data were normalized and expressed as a percent of control within a given experiment before collating across experiments (samples were collected and analyzed in two batches). Data were analyzed for statistical significance using unpaired t-test.

Electron Microscopy Data
Analysis was performed blind to genotype. Digital images were analyzed using Image J (NIH) and the Image J plug-ins PointDensity and PointDensitySyn were also analyzed. Analysis of immunogold-labeled TH-positive structures was performed using PointDensity; after the perimeter of the central structure within the frame was delineated, a point marker was placed within the center of each vesicle. Vesicles were marked if at least 50% of the vesicle membrane was visible. TH-positive profiles that were forming synapses were analyzed using PointDensitySyn, the plasma membrane of the structure was traced, the active zone(s) were delineated, and points were placed in the center of the vesicles (Anwar et al., 2011;Janezic et al., 2013). Cumulative distributions for EM data were analyzed using the Kolmogorov-Smirnov test. Inter-vesicle distance data binned at 50 nm interval was analyzed using a two-way ANOVA with Sidak's multiple comparisons test.

Behavior Data
Testing was performed by an observed blind to genotype. Datasets with non-normal distributions were analyzed using a Mann-Whitney test, data sets with a normal distribution were analyzed using unpaired t-tests. Rotarod and four-choice reversal test were analyzed using a two-way ANOVA with Sidak's multiple comparisons test. The three chamber social approach test was analyzed within genotype to test for significant differences between mouse and object with paired t-tests.    Table S4 and  Table S5 for behavior analysis by sex.