Dopamine neuron glutamate cotransmission evokes a delayed excitation in lateral dorsal striatal cholinergic interneurons

Dopamine neurons have different synaptic actions in the ventral and dorsal striatum (dStr), but whether this heterogeneity extends to dStr subregions has not been addressed. We have found that optogenetic activation of dStr dopamine neuron terminals in mouse brain slices pauses the firing of cholinergic interneurons in both the medial and lateral subregions, while in the lateral subregion the pause is shorter due to a subsequent excitation. This excitation is mediated mainly by metabotropic glutamate receptor 1 (mGluR1) and partially by dopamine D1-like receptors coupled to transient receptor potential channel 3 and 7. DA neurons do not signal to spiny projection neurons in the medial dStr, while they elicit ionotropic glutamate responses in the lateral dStr. The DA neurons mediating these excitatory signals are in the substantia nigra (SN). Thus, SN dopamine neurons engage different receptors in different postsynaptic neurons in different dStr subregions to convey strikingly different signals. Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review. The Reviewing Editor's assessment is that all the issues have been addressed (see decision letter).


Introduction
Dopamine (DA) neurons send dense projections to the striatum (Str) that are topographically organized. Medially located ventral tegmental area (VTA) DA neurons project to the ventromedial Str; more laterally located substantia nigra (SN) DA neurons project to the dorsolateral Str (Haber et al., 2000;Ikemoto, 2007). DA neuron gene expression profiles identify DA neuron subtypes differentially distributed within the ventral midbrain (Poulin et al., 2014) that likely project differentially to striatal subregions, as well as extrastriatal regions. DA neurons are heterogeneous in their membrane excitability and modulation (Morales and Margolis, 2017;Roeper, 2013), and the synaptic signals they convey .
There is significant heterogeneity in DA release and its modulation in different Str subregions (Sulzer et al., 2016). DA neurons make fast D2-mediated DA synaptic connections to cholinergic interneurons (ChIs) in the dorsal Str (dStr) , while they elicit slower DA actions via volume transmission throughout the Str (Surmeier et al., 2014;. DA neuron glutamate cotransmission contributes striking regional heterogeneity; in the nucleus accumbens (NAc) medial shell, DA neurons elicit glutamate EPSCs in the three major Str cell types, with the strongest in ChIs (Chuhma et al., 2014). In the dStr, DA neuron glutamate EPSCs have not been seen in ChIs (Chuhma et al., 2014), while they have been variably observed in dStr spiny projection neurons (SPNs) (Mingote et al., 2015;Stuber et al., 2010;. The dStr comprises multiple functional subregions, determined by cortical inputs (Hunnicutt et al., 2016;Joel and Weiner, 2000;Liljeholm and O'Doherty, 2012). The medial and lateral subregions correspond roughly to associative and sensorimotor Str, respectively (Joel and Weiner, 2000;Kreitzer, 2009;Liljeholm and O'Doherty, 2012;McGeorge and Faull, 1989;Voorn et al., 2004). DA modulates learning and performance of goal-directed behavior in the medial dStr (mdStr), and formation of habits in the lateral dStr (ldStr) (Faure et al., 2005;Hilário and Costa, 2008;Lerner et al., 2015). Studies of functional synaptic connectivity have addressed differences between the NAc and dStr, but not between dStr subregions. Whether heterogeneity in DA neuron synaptic actions extends to dStr subregions has not been elucidated.
We have compared synaptic responses elicited by DA neurons in identified mdStr and ldStr neurons, focusing principally on synaptic connection to ChIs, since the most prominent direct DA neuron synaptic connections are seen in ChIs (Chuhma et al., 2014). This has revealed subregional heterogeneity in the dStr due to a slow mode of DA neuron glutamate cotransmission in ldStr ChIs, mediated by substantia nigra DA neurons.

Different responses of ChIs to DA neuron terminal stimulation in the mdStr and ldStr
To activate DA neuron terminals impinging on recorded Str neurons, we used mice with DA transporter (DAT) driven channelrhodopsin 2 (ChR2)-enhanced yellow fluorescent protein (EYFP) expression (DAT IREScre ;ChR2-EYFP mice), in which ChR2-EYFP is expressed almost exclusively in ventral midbrain DA neurons (over 98%) , and wide-field photostimulation. DA neuron synaptic responses were recorded from ChIs in the mdStr or ldStr ( Figure 1A), identified by soma size and membrane properties (viz. Chuhma et al., 2014) (Figure 1-figure supplement 1). With train photostimulation of DA neuron terminals, mimicking the phasic firing of DA neurons (five 5-msec pulses at 20 Hz), mdStr ChIs showed the previously reported pause in firing (Chuhma et al., 2014;Straub et al., 2014), while ldStr ChIs showed a shorter pause in firing followed by an increase in firing ( Figure 1B). Average ChI firing frequency (in 100 ms bins) was reduced similarly during photostimulation in mdStr and ldStr, while post-stimulation responses differed; firing was reduced in the mdStr ChIs and increased in the ldStr ChIs ( Figure 1C). To evaluate the changes relative to baseline firing, firing z-scores were calculated. During photostimulation, 0-0.4 s from the onset of the train, firing z-scores in the mdStr (À2.5 ± 0.3) and ldStr ChIs (À1.7 ± 0.3) were both negative, reflecting a reduction in firing, and were not significantly different (p=0.074, independent-sample t-test). During the post-stimulation period, from 0.5 to 0.9 s, firing z-scores in mdStr ChIs remained negative (À2.4 ± 0.4), while they became positive in ldStr ChIs, reflecting an increase in firing (+2.5 ± 0.8) ( Figure 1D).
In voltage clamp recordings done at À70 mV, single-pulse photostimulation of DA neuron terminals (at 0.1 Hz) evoked sub-second PSCs in both mdStr and ldStr ChIs ( Figure 1E). The peak amplitudes of IPSCs (outward currents) did not differ significantly between mdStr ChIs (23.7 ± 4.5 pA) and .9, p=0.000, mixed ANOVA; post-hoc regional comparison during-stim p=0.074, post-stim p=0.000. (E) Under voltage clamp, PSCs were evoked by single-pulse photostimulation, delivered at 0.1 Hz. Traces are the average of 10 consecutive traces. (F) Average amplitude of IPSCs (outward current), EPSCs (inward current), and synaptic response charge transfer during and following stimulation (1 s window). IPSCs showed no regional difference in amplitude (p=0.29, independent sample t-test); EPSCs showed a regional difference in amplitude (p=0.000, Welch's t-test) and charge (p=0.001, independent sample t-test). mdStr: n = 10 cells from six animals. ldStr: n = 12 cells from eight animals. Dots in bar graphs show the average measurement from each recorded cell. Numbers of cells recorded are in parentheses. *, ** and *** indicate p<0.05, p<0.01 and p<0.001, respectively. See also  ldStr ChIs (18.1 ± 2.8 pA). Peak amplitude of slow EPSCs (inward currents) was significantly greater in ldStr ChIs (mdStr 1.8 ± 0.6 pA, ldStr 10.4 ± 1.6 pA) ( Figure 1F). The charge transfer of PSCs (time integration of PCSs) in mdStr ChIs (11.7 ± 1.5 pC) was significantly larger than in ldStr ChIs (3.0 ± 1.6 pC) ( Figure 1F). PSCs were not seen in ChIs of mice with conditional EYFP expression ( Figure 1figure supplement 3). These observations suggest that the shorter firing pause in ldStr ChIs was likely due to shorter IPSCs with the same peak amplitude, which were captured more effectively by measuring charge transfer.

Cell type and regional distribution of slow EPSCs
To examine the distribution of slow EPSCs across cell types and locations, we recorded from ChIs and the two classes of spiny projection neurons (SPNs) in the mdStr and ldStr. SPNs, which are the principal Str neurons, were identified by either D1-tdTomato (direct-pathway SPNs; dSPNs) or D2-EGFP (indirect pathway SPNs; iSPNs) fluorescence in triple mutant mice, produced by breeding the reporter lines with DAT IREScre ;ChR2-EYFP mice. Slow EPSCs were evoked by single pulse photostimulation at 0.1 Hz, pharmacologically isolated with a cocktail of D2, iGluR and GABA A antagonists, and charge transfer measured in the window from 0.2 to 1.7 s after the onset of photostimulation, corresponding to the duration of the EPSC.
Single photostimulation evoked small or no slow EPSCs in mdStr ChIs (2.1 ± 0.5 pC), prominent slow EPSCs in the ldStr ChIs (25.8 ± 4.4 pC), and no PSCs in SPNs in either the mdStr or ldStr (mdStr SPN 0.93 ± 0.24 pC, ldStr SPN 0.64 ± 0.15 pC) ( Figure 3A, B and C). The charge transfer of slow EPSCs ranged from 0.014 to 6.7 pC in mdStr ChIs and from 3.9 to 140.0 pC in ldStr ChIs ( Figure 3B and C). Charge transfer was significantly different between cell types, between regions, with a significant cell type by region interaction. ldStr ChIs showed significantly larger responses than SPNs or mdStr ChIs. mdStr ChIs showed significantly larger responses than SPNs, in either the mdStr or ldStr. No significant difference was observed between mdStr SPNs and ldStr SPNs. When SPNs were split into dSPNs and iSPNs, there were no differences in charge transfer between region or cell type  Figure 3D); slow EPSCs were not seen in either dSPNs or iSPNs. Thus, slow EPSCs were exclusive to ChIs and prominent in ldStr ChIs.

Pharmacological analysis of slow EPSCs
To test monosynapticity of slow EPSCs in ldStr ChIs, DA neuron terminals were stimulated after application of 1 mM tetrodotoxin (TTX) and 0.5 mM 4-aminopyridine (4AP) (Figure 4). In this condition, action potentials are blocked by TTX, and axon terminals slightly depolarized by 4AP, but not enough to cause transmitter release; optical stimulation causes further depolarization only in ChR2-expressing terminals and transmitter release (Petreanu et al., 2009). Thus, monosynaptic connections generate postsynaptic responses, while polysynaptic connections fail. Application of TTX blocked slow EPSCs completely (pre-drug 26.7 ± 5.6 pC, TTX 1.2 ± 0.3 pC), confirming the slow EPSCs were action-potential dependent ( Figure 4A). Addition of 4AP partially recovered slow EPSCs (11.3 ± 2.5 pC), indicating that slow EPSCs have a monosynaptic component ( Figure 4A). To determine whether there was an additional polysynaptic component, the rising phase of EPSCs was compared before and after TTX + 4 AP ( Figure 4B). Traces were superimposable after amplitude scaling; the rise time (10% to 90% of peak amplitude) was not different (pre-drug 0.24 ± 0.03 s, TTX + 4 AP 0.21 ± 0.02 s), indicating that the slow EPSCs are solely monosynaptic.
To confirm that slow EPSCs arose from ventral midbrain DA neurons, we injected a conditional viral vector expressing ChR2 (AAV-DIO-ChR2-EYFP) into the ventral midbrain of DAT IREScre mice and found that the responses did not differ from those recorded in DAT IREScre ;ChR2-EYFP mice ( When G-protein coupled receptor (GPCR) transduction was blocked by GDPbS (0.5-1 mM) in the intracellular solution, slow EPSCs were almost completely blocked 9 min after entering whole cell mode, while EPSCs were not significantly reduced with control intracellular solution (with GTP), showing that the blockade was not due to cell deterioration (comparison to 0-1 min from achieving whole cell, Ctrl 5-6 min 115.2 ± 16.1%, 9-10 min 97.0 ± 15.7%; GDPbS 5-6 min 13.7 ± 8.0%, 9-10 min 6.0 ± 2.9%) ( Figure 5A). To confirm complete blockade of the slow EPSCs, we applied train stimulations after 10 min recording with single stimulation ( Figure 5-figure supplement 1). The slow EPSCs were not observed in GDPbS dialyzed cells (2.5 ± 0.9 pC), while slow EPSCs persisted in control cells (36.1 ± 10.2 pC). Thus, slow EPSCs were GPCR mediated.
Next, we sought to determine the receptors mediating the slow EPSC, which should be (1) GPCRs, (2) capable of exciting neurons within a second, (3) expressed in ChIs, and (4) responsive to neurotransmitters released by DA neurons. Two receptor types fulfill these criteria: DA D1-like (D1/ D5) receptors and group I metabotropic glutamate receptors (mGluRs). D1-like receptors mediate a delayed firing increase in olfactory tubercle ChIs (Wieland et al., 2014), and ldStr ChIs are likely to share the same mechanisms. mGluR1s mediate slow EPSCs in cerebellar Purkinje cells, which have a time course similar to the slow EPSCs in ldStr ChIs (Batchelor and Garthwaite, 1997;Tempia et al., 1998). Bath application of group 1 mGluR agonist elicits inward currents in dStr ChIs (Berg et al., 2007;Takeshita et al., 1996) sufficient to mediate slow EPSCs. Thus, we examined the effects of D1-and group I mGluR (mGluR1/5) antagonists on slow EPSCs.
Bath application of mGluR group I agonists have been shown to depolarize dStr ChIs by activation of transient receptor potential channel (TrpC) 3 and 7 (Berg et al., 2007), pointing to TrpC mediation of the slow EPSC. We blocked D2 and GABA A receptors, but not iGluRs to show that antagonist actions were selective for the slow EPSC, as TrpC antagonists may affect transmitter release by reducing presynaptic excitability (Yau et al., 2010). Bath application of the TrpC3 selective antagonist Pyr3 (20 mM) reduced the slow EPSC to 27.2 ± 5.1% pre-drug, without affecting the fast EPSC (89.8 ± 12.2%) ( Figure 6A and B). The TrpC3/7 antagonist flufenamic acid (FFA, 100 mM) blocked the slow EPSC completely (3.9 ± 1.1% pre-drug), without affecting the fast EPSC (89.5 ± 4.2% pre-drug) ( Figure 6A and B), indicating that the slow EPSCs were mainly TrpC3 mediated. Taken together, these results show that slow EPSCs were mediated by mGluR1 and D1/5R through TrpC3/7 as the effector channel, but mainly by mGluR1 through TrpC3.
To confirm that glutamate responses originated from DA neurons, we examined responses in vesicular glutamate transporter 2 (VGLUT2) conditional knockout mice (cKO; DAT IREScre ;ChR2-EYFP; VGLUT2 lox/lox ) and controls (DAT IREScre ;ChR2-EYFP; VGLUT2 +/+ ). Fast EPSCs were absent in ldStr SPNs in cKOs, but not in controls (cKO 1.1 ± 0.3 pA, Ctrl 26.1 ± 6.0 pA) ( Figure 7D). Slow EPSCs in ldStr ChIs isolated by a cocktail of D2, GABA A and iGluR antagonists were significantly smaller in VGLUT2 cKO mice, although the responses were not completely eliminated (cKO 3.0 ± 0.8 pC, Ctrl 13.7 ± 2.6 pC) ( Figure 7E). When the D1-antagonist SCH23390 was added to the cocktail of antagonists, slow EPSCs were almost completely abolished, confirming glutamate and D1 mediation (cKO 1.3 ± 0.3 pC, Ctrl 10.2 ± 2.4 pC) ( Figure 7F). Thus, in the ldStr, DA neuron glutamate cotransmission engages different glutamate receptors in different postsynaptic target cells -mGluRs mediating slow EPSCs in ChIs and iGluRs mediating fast EPSCs in SPNs; in the mdStr, very little glutamate cotransmission is seen in either ChIs or SPNs. Substantia nigra DA neurons mediate glutamate cotransmission in the ldStr DA neurons project to the Str topographically, with the most medially located VTA neurons projecting to the NAc medial shell and more lateral SN DA neurons projecting to the ldStr (Haber et al., 2000;Ikemoto, 2007). Based on this topography, projections to the ldStr should be from SN DA neurons. However, in adult mice, VGLUT2 expression is prominent in VTA but not SN pars compacta (SNc) DA neurons (Morales and Root, 2014), arguing that non-topographical projections of VTA DA neurons mediate glutamate EPSCs in the ldStr. To evaluate the projections of DA neurons to the mdStr or ldStr, we injected the retrograde axonal tracer cholera toxin B subunit (CTB) into the ldStr, and for comparison into the mdStr ( Figure 8A, top panels). In the ventral midbrain, double immunostaining for CTB (green) and the DA neuron marker tyrosine hydroxylase (TH; magenta) revealed DA neurons projecting to the injection sites ( Figure 8A, middle). CTB is transported anterogradely as well, and so revealed the terminals of Str projection neurons in the SN pars reticulata (SNr) ( Figure 8A, middle, asterisks). The restricted distribution of anterogradely labeled terminals confirmed that the CTB injections were limited to the mdStr or ldStr, and indicated that the wider CTB staining in the Str was not due to the wider spread of the injected CTB, but rather reflected retrograde and anterograde labeling of locally connected Str neurons. We examined the relative distribution of CTB positive (CTB + ) Str-projecting neurons in the VTA and the SN (for delineation, see  Figure 8A, bottom), demonstrating that about the same numbers of DA neurons in the VTA and SN project to the ldStr. The distribution of CTB + /TH + cells in the VTA and SN differed significantly between mdStr and ldStr injections. Thus, there are non-topographical VTA DA neuron projections to the ldStr, but not to the mdStr. However, this did not resolve whether DA neuron glutamate cotransmission arose from non-topographical VTA DA neuron projections.
To address this, we injected the conditional retrograde viral tracer canine adenovirus 2 (CAV2)-FLEX-ZsGreen (Ekstrand et al., 2014) into the mdStr or ldStr of VGLUT2 IREScre mice to label glutamatergic (VGLUT2 + ) neurons projecting to the sites injected. With mdStr injections, no ZsGreen + cells were seen, consistent with the dearth of DA neuron glutamate cotransmission in the mdStr ( Figure 8B, middle, left panels). With ldStr injections, there were 8 ± 4 ZsGreen + cells in the VTA and 157 ± 13 in the SN ( Figure 8B, middle, right panels), demonstrating that glutamatergic neurons projecting to the ldStr are mostly in the SNc. Among them, 2 ± 1 cells (24.6 ± 6.8% of ZsGreen + ) in the VTA and 90 ± 18 cells (57.0 ± 8.5%) in the SN were DA neurons (TH + ) ( Figure 8B, bottom). Thus, topographically projecting SN DA neurons mediate glutamate cotransmission in the ldStr. However, variable expression of CAV receptors on DA neuron terminals may have limited the number of DA/ glutamate neurons identified.
To visualize VGLUT2-expressing DA neurons in their entirety, we used the INTRSECT strategy (Fenno et al., 2014). We injected a cre-on/flp-on ChR2-EYFP virus (AAV-Con/Fon-ChR2-EYFP) into the SN of VGLUT2 IREScre ; TH 2A-flpo mice to express ChR2-EYFP conditionally in DA neurons capable of glutamate cotransmission ( Figure 9A,B). Due to spread of virus, ChR2-EYFP expression was also observed in the VTA ( Figure 9B). In the SN, there were 247 ± 11 TH + /ChR2 + and 890 ± 116 TH + / ChR2neurons (n = 4 animals), corresponding to 20.7% and 74.8% of all immunopositive (TH + and/ or ChR2 + ) neurons, respectively ( Figures 9B and 7C). There were 53 ± 7 TH -/ChR2 + neurons, so the specificity of ChR2 expression in DA neurons was 82.5%. Of TH + neurons, 22% were ChR2 + indicative of VGLUT2 expression, making them capable of glutamate cotransmission ( Figure 9C). Colabeled terminals were denser in the ldStr, than in the mdStr ( Figure 9D). The observation of sparse fibers in the mdStr, indicated that ZsGreen labeling underestimated the number of DA neurons capable of glutamate cotransmission, because of the small injection volume. Thus, a substantial number of SN DA neurons express VGLUT2 and project preferentially to the ldStr, coincident with the glutamate cotransmission in the ldStr.

Expression of slow EPSC mediators in ChIs
While topographic DA neuron projections account for the presence of the slow EPSC in the ldStr, they do not account for the specificity for ChIs (viz. Figure 3). To identify the receptors and channels mediating slow EPSCs in ChIs, we used ChATcre;RiboTag mice, with conditional expression of a hemagglutinin-tag on the last exon of ribosome protein RPL22, under the control of the choline acetyltransferase (ChAT) promoter, to enable immunoprecipitation (IP) of ribosome-associated mRNA (Sanz et al., 2009) from ChIs. We examined expression of the following genes by quantitative PCR (qPCR): mGluR1 and 5, TrpC 3 and 7, as possible determinants of the slow EPSC; D1R, D5R and D2R, as known controls; ChAT and vesicular acetylcholine transporter (VAChT), as IP controls. We used DCt normalized to GAPDH for gene expression, and DDCt to whole Str mRNA to evaluate enrichment of expression in ChIs. Confirming the successful isolation of ChI mRNA, both ChAT and VAChT showed high enrichment in ChIs ( Figure 10A). D5R, TrpC3 and TrpC7 were enriched in ChIs, while mGluR1, mGluR5 and D1R were reduced ( Figure 10A). This suggests that increased expression of TrpC3/7 and D5R, but not mGluR1, determines the discrete observation of slow EPSCs in ChIs.
When we compared gene expression relative to GAPDH between the mdStr and ldStr, D1, mGluR5 and TrpC3 expression showed significant differences between the mdStr and ldStr ( Figure 10B). Subregional differences in gene expression were not due to differences in the numbers of ChIs, as their density was the same in the mdStr and ldStr (Figure 10-figure supplement 1). Considering the low expression of D5R and TrpC7 ( Figure 10B), reduced expression of D1R in ChIs ( Figure 10A) and the minor contribution of D1-like receptors and TrpC7 to the slow EPSC (Figure 4), high expression and enrichment of TrpC3 in ChIs appears to be crucial for the slow EPSC. However, the subregional difference in TrpC3 expression between mdStr and ldStr was significantly smaller than the difference in the size of the slow EPSC (viz. Figure 3), so differential TrpC3 expression was not the determinant of the prominence of the slow EPSC in the ldStr. Thus, the principal determinant of the medial-lateral difference appears to be the projections of VGLUT2 + DA neurons, while the discrete observation of the slow EPSCs in ChIs appears to be due to postsynaptic TrpC3 expression.

u R 5 T r p C 3 T r p C 7
Enrichment in ChIs * * * * * * * * * * * * * * * * * Figure 10. Expression of slow EPSC mediators in ChIs. qPCR measurements of ChAT, VAChT, DA receptors, mGluR1/5 and TrpC 3/7 expression in dStr ChIs. (A) Enrichment in ChIs is shown relative to whole Str RNA (input sample). A ratio of 1 indicates that expression is the same in ChIs and all dStr cells. In box plot, the middle lines and horizontal outlines of boxes indicate means and SEM, respectively. Statistical significance was examined with one-sample t-test to 1 (no enrichment). See Figure 10-source data 1 for exact p values for each gene. *, ** and *** indicate p<0.05, p<0.01 and p<0.001, respectively, for differences from a ratio of 1. n = 5 replicates (three animals per replicate). (B) Subregional differences in expression in ChIs in the mdStr (blue) and ldStr (green) are shown, relative to a housekeeping gene (GAPDH). Regional differences were examined using a general linear model multivariate analysis. See Discussion DA neurons differentially control ChIs in the dStr. They inhibit mdStr ChIs, while they inhibit and excite ldStr ChIs. The inhibition is D2 mediated, and shorter in ldStr ChIs due to an ensuing slow EPSC. The slow EPSC is mediated monosynaptically by glutamate cotransmission activating mGluR1 receptors, coupled to activation of TrpC3/7, along with DA activation of D1-like (D1/5) receptors. mGluR1 responses are observed only in ChIs in the ldStr, while SPNs show iGluR responses. Topographic projections of SN DA neurons to the ldStr mediate the glutamate responses in ChIs. Slow DA neuron driven EPSCs are seen discreetly in ldStr ChIs, due presynaptically to SN DA neuron glutamate cotransmission and postsynaptically to enriched expression of TrpC3, that together determine the unique responses of ldStr ChIs to DA neuron activity ( Figure 11).

Slow EPSCs in lateral dorsal striatum cholinergic interneurons
Optogenetic activation of DA neurons projecting to the dStr evokes a delayed excitation, of unclear mediation (Straub et al., 2014), which corresponds in magnitude and timing to the slow EPSC. Pharmacological isolation of the slow EPSC together with use of selective and more potent antagonists of group I mGluRs revealed mGluR1 and D1 as mediators of the slow EPSC. We used the highly potent and selective mGluR1 antagonist JNJ, at the maximum mGluR-selective concentration (Fukunaga et al., 2007). Less specific mGluR antagonists require higher concentrations that may mask mGluR1 effects through opposing actions at other mGluRs. While the JNJ and MTEP cocktail showed significant mGluR1 effects after 10 min, the D1 antagonist SCH23390 required 20 min to show significant action. SCH loses D1 selectivity at higher concentrations due to Ca2 + channel blockade (Guenther et al., 1994), so longer application times are required. The slow EPSC was not completely blocked by mGluR1 and D1 antagonists; the small residual EPSC could be due to incomplete antagonist action, or could be a minor still unspecified response mediated by a G-protein and TrpC3/7, as it was blocked by GDPbS or FFA. FFA has multiple actions besides TrpC3/7 antagonism; it is also a TrpC6 agonist and affects other ion channels. The lack of significant effects on fast glutamate EPSCs excluded effects on transmitter release through other ion channels, and complete blockade of slow EPSCs excluded a TrpC6 contribution to the slow EPSCs.
Optogenetic activation of DA neurons projecting to olfactory tubercle (OT) evokes a delayed D1mediated excitation (Wieland et al., 2014). While the delayed excitation in the OT is solely D1 mediated, D1-like receptors mediate only part of the response in dStr ChIs. Since DA neuron glutamatergic projections to the mdStr are sparse, the small slow EPSCs seen in some mdStr ChIs are likely D1R mediated, pointing to slow D1-mediated excitation of ChIs across the Str. Both D1 and D5 receptors are expressed in ChIs, but DA application produces the same depolarization in wild type and D1R KO mice (Centonze et al., 2003), arguing that the D1 component of slow EPSCs is mainly D5R mediated. Although our observation of enriched D5R expression and reduced D1R expression in ChIs agree with previous observations (Bergson et al., 1995;Lim et al., 2014;Yan et al., 1997), we found a higher absolute D1R expression in ChIs than previously reported (Bergson et al., 1995;Yan et al., 1997); this could be due to methodological differences, species (rat vs. mouse) or animal age. Immunostaining would not necessarily reveal D1Rs in ChIs given the high D1R expression in dSPNs, which greatly outnumber ChIs (Bergson et al., 1995).
The slow EPSC in dStr ChIs is mainly mGluR1 mediated. Slow mGluR1 EPSCs were originally described in cerebellar Purkinje cells (Batchelor and Garthwaite, 1997) and ventral midbrain DA neurons (Fiorillo and Williams, 1998;Shen and Johnson, 1997); however, these slow EPSCs are seen only with spike trains, suggesting that they are mediated by glutamate spillover acting at extrasynaptic receptors. In contrast, slow EPSCs in ldStr ChIs are reliably evoked with single spikes, at low frequency, suggesting that mGluR1 receptors are closer to synaptic release sites in ChIs than in Purkinje cells or DA neurons. ChIs, as well as SPNs, express both mGluR1 and iGluRs, and iGluR responses are evoked in SPNs by stimulation of other glutamate inputs (Ding et al., 2010;Johnson et al., 2017), but not mGluR1 responses. Furthermore, mGluR1 is not enriched in ChIs. Thus, the different glutamate responses in ChIs and SPNs -mGluR1 in ChIs and iGluRs in SPNsappear to depend on the differential distribution of the receptors at postsynaptic sites of DA neuron glutamate cotransmission in the two cell types.
Perfusion of group I mGluR agonists evokes a TrpC3/7 mediated depolarization in dStr ChIs in rat brain slice (Berg et al., 2007); our expression studies show that dStr ChIs express both mGluR1 and TrpC3. TrpC3 is also a major mediator of mGluR1-mediated slow EPSCs in cerebellar Purkinje cells (Hartmann et al., 2008), suggesting that ldStr ChIs share with Purkinje cells the same mGluR EPSC mechanism. mGluR1 is also expressed in ventral midbrain DA neurons and their terminals (Fiorillo and Williams, 1998;Zhang and Sulzer, 2004), and presynaptic mGluR1s reduce DA release in the Str (Zhang and Sulzer, 2004). Since activation of presynaptic receptors requires transmitter spillover (Scanziani et al., 1997), activation of presynaptic mGluR1 on DA neuron terminals by single spikes is less likely. However, when DA neurons fire in bursts, mGluR1-mediated presynaptic inhibition would limit glutamate cotransmission in the ldStr, and likely reduce temporal summation.

Glutamate cotransmission from substantia nigra dopamine neurons
While DA neuron cotransmission consistently evokes iGluR EPSCs in the NAc medial shell, reports of EPSCs in the dStr have varied (Chuhma et al., 2014;Mingote et al., 2015;Stuber et al., 2010;. Our present results reveal that iGluR EPSCs vary based on recording location, with small or no responses in the mdStr and larger responses in the ldStr. mGluR1 responses in ChIs are prominent in the ldStr. Thus, the ldStr -in addition to the NAc medial shell -is a hotspot of DA neuron glutamate cotransmission. Given the topography of DA neuron projections (Haber et al., 2000;Ikemoto, 2007), the ldStr should receive input solely from the SNc. Although our CTB results showed the existence of non-topographical projections from VTA DA neurons to the ldStr, as has also been observed in rats (Maurin et al., 1999;Pennartz et al., 2009), these non-topographical VTA projections do not mediate glutamate cotransmission in the ldStr. Considering the very sparse expression of ZsGreen in the VTA following ldStr injections, the non-topographic projection appears to arise from non-glutamate cotransmitting DA neurons, which may contribute to D1-or D2-like receptor responses in ldStr ChIs. Our retrograde viral tracing showed that most glutamate cotransmission in the ldStr ChIs arises from SN DA neurons, where VGLUT2 expression has not been reported in adult mice (Morales and Root, 2014). This discrepancy could be due to the difficulty of detecting a small population of neurons with relatively low expression levels by in situ hybridization. The INTRSECT strategy revealed that about a fifth of SN DA neurons coexpress VGLUT2, which was more than found by in situ hybridization (Morales and Root, 2014) or CAV2-FLEX-ZsGreen retrograde labeling. Similarly, the INTRSECT strategy revealed a few fibers in the mdStr, but there was no retrograde labeling with ZsGreen. This could be due to a lack of CAV receptors on mdStr projecting DA neurons, but was more likely due to the sparse distribution of mdStr projecting DA neurons and the more restricted volume of the CAV injection. While DA neurons capable of glutamate cotransmission constitute a minority of SN neurons, they elicit substantial excitation due to their relatively restricted projection area. The INTRSECT strategy showed TH/VGLUT2 colocalization in the VTA as well. Since TH + /VGLUT2 + neurons in the VTA made little contribution to the non-topographic projection to the ldStr, these coexpressing VTA neurons presumably project topographically to the NAc. There were ChR2 + /THneurons in the SN, apparently showing lower specificity of the INTRSECT strategy. This could be due to a combination of slightly higher non-specific expression of TH-driven gene expression than DAT-driven gene expression (Lammel et al., 2015), recombination efficacy/specificity of the particular INTRSECT virus, and limitations in identifying colocalization of cytosolic TH and membrane-targeted ChR2 (Stuber et al., 2015). Some non-specific expression, notwithstanding, the INTRSECT results showed substantial SN DA neuron glutamate projections to the ldStr, and sparse projections to the mdStr, consistent with the CAV2 results, showing that DA/ glutamate neurons in the SN project to the ldStr.
Our retrograde tracing studies revealed that about 40% of mdStr or ldStr projecting neurons are non-DAergic, presumably GABA-only or glutamate-only neurons (Morales and Root, 2014;Morello and Partanen, 2015). Glutamate-only neurons are seen in both the VTA and SNc (Hnasko et al., 2012;Morales and Root, 2014), and indeed we observed 30-40% TH -/VGLUT2 + neurons in both the VTA and the SN (SNc/SNr). Both TH + /VGLUT2 + and TH -/VGLUT2 + (glutamateonly) neurons projecting to the ldStr reside in the SN, and do not project to the mdStr, consistent with the recognized segregation of DA neurons projecting to the medial and lateral dStr (Lerner et al., 2015). Thus, glutamate cotransmitting DA neurons and glutamate-only neurons appear to project similarly to the dStr.

Functional implications
DA neurons may engage ChIs as a hub to control Str circuits. Although ChIs comprise only 1-2% of striatal neurons, they exert a strong control of striatal circuits with their large axonal fields (Kawaguchi et al., 1995;Kreitzer, 2009). In response to reward-related cues or outcomes, dStr ChIs frequently show a burst-pause-burst firing pattern, coincident with DA neuron burst firing (Morris et al., 2004;Schulz and Reynolds, 2013). While the early burst is mediated by thalamic inputs (Ding et al., 2010), the pause appears to be principally D2R mediated. The subsequent rebound in firing is highly variable, suggesting that it is generated and modulated independently of the D2-mediated pause (Schulz and Reynolds, 2013). Multiple mechanisms likely contribute to the rebound firing, one of which is the DA neuron mediated slow EPSC. The presence of the slow EPSC in the ldStr, and not the mdStr, likely contributes to subregional variability in ChI rebound firing.
Mice used for the experiments were on a C57BL6J background or a C57BL6J/129Sv mixed background, more than five times back crossed to C57BL6J and kept inbred. D2-EGFP mice, originally on a FVB background, were backcrossed to C57BL6J at least eight times. VGLUT2 IREScre mice, originally on a mixed C57BL6J;FVB;129S6 background, were backcrossed 3-5 times to C57BL6J. RiboTag mice were on a C57BL6NJ background.
In DAT IREScre :Ai32 mice, ChIs were identified visually by large soma size, confirmed by spontaneous firing, shallow resting membrane potentials (around À60 mV) and voltage sag by À400 pA current injection (700 msec duration) (Chuhma et al., 2014). Recording patch pipettes were fabricated from standard-wall borosilicate glass capillary with filament (World Precision Instruments). Pipette resistance was 3-7 MW. Composition of intracellular solution for firing and the slow EPSC recording was (in mM): 135 K + -methane sulfonate (MeSO 4 ), 5 KCl, 2 MgCl 2 , 0.1 CaCl 2 , 10 HEPES, 1 EGTA, 2 ATP and 0.1 GTP, pH 7.25. For G-protein coupled receptor blockade, GTP was replaced with 0.5-1 mM GDPbS. GDPbS pipette solution was made from powder just before recording, kept on ice and discarded after 2 hr. For fast glutamate EPSC recording, Cs + -based pipette solution was used; K + -MeSO 4 was replaced with Cs + -MeSO 4 , with QX314 (lidocaine N-ethyl bromide) 5 mM. Both voltage and current clamp recordings were performed with an Axopatch 200B amplifier (Molecular Devices). Holding potential was À70 mV. Current clamp recordings were done in fast current clamp mode. Series resistance (8-26 MW) was compensated online by 70-75%. Liquid junction potentials (10-12 mV) were adjusted online. Synaptic responses were evoked with five msec field illumination with a high-power blue LED (Thorlabs) delivered either as a single pulse at 0.1 Hz, or in a train of five pulses at 20 Hz, repeated at 30 s intervals. Since some slow EPSC recording showed a build up with repeated stimulation, the control pre-drug responses were recorded after the EPSC had reached a plateau.
For pharmacological studies, drugs were delivered by perfusion. For wash off of the effects of mGluR antagonists and D1-antagonist on isolated slow EPSCs, slices were perfused with regular ACSF for 15-25 min, followed by control antagonists (CNQX, D-APV, SR95531 and sulpiride) for at least 5 min. Recording from VGLUT2 cKO mice and control mice were done genotype blind. All the recordings were done at 32-34˚C (TC 344B Temperature Controller, Warner Instruments). Data were filtered at 5 kHz using a 4-pole Bessel filter, digitized at 5 kHz (Digidata 1550A, Molecular Devices) and recorded using pClamp 10 (Molecular Devices; RRID:SCR_011323).

Data analysis for electrophysiology
Electrophysiological data were analyzed with Axograph X (Axograph Science; RRID:SCR_014284). Firing z score during or post train stimulation was calculated as the difference of average firing frequency during (0-0.4 s from train onset) or post train (0.5-0.9 s from train onset), minus the average baseline firing frequency, divided by the standard deviation of baseline firing frequency. Plus score indicates increase of firing from the baseline, while minus score indicates decrease of firing. Since ChIs were firing continuously and it was hard to identify resting state of membrane, we regarded averaged membrane potentials after action potentials were truncated as the resting membrane potentials. PSC evaluation was done from averages made from 10 consecutive traces. Slow EPSCs were evaluated by measuring charge transfer in a 1 s or 1.5 s window ('area under the curve') for more reliable comparison. Since synaptic responses to the first stimulus after a long interval were artificially large, we discarded them. Data are shown as mean ± S.E.M.

Fast-scan cyclic voltammetry in slice
Fast-scan cyclic voltammetry (FSCV) recordings were done in P72-93 DAT IREScre ;ChR2-EYFP mice. Slice preparation and recording environment were the same as described in the Slice electrophysiology section. DA release was evoked with either a single pulse (five msec duration) or a train (5 pulses at 20 Hz; 5 ms duration pulses) of field illumination of blue LED. Photostimulation trains was applied at intervals of 1-2 min. DA release was measured using carbon fiber electrodes placed 30-100 mm below the slice surface. Carbon fibers was calibrated in 1 mM DA after each experiment. A triangular voltage wave (À400 to + 800 mV at 300 V/sec vs. Ag/AgCl) was applied to the electrode at 10 Hz and the resulting currents were recorded using an Axopatch 200B (Molecular Devices), filtered at 10 kHz with a 4-pole Bessel filter. Offset currents were not applied to avoid saturation of the amplifier. DA currents were confirmed by examination of the cyclic voltammograms for the signature DA oxidation current waveform. Traces of DA release were obtained by plotting current measured at the potential of peak DA oxidation for each voltage command as a function of time, in Igor Pro (Wavemetrics; RRID:SCR_000325) using custom routines (Eugene Mosharov, sulzerlab.org). Data were analyzed with MATLAB R2014b (Mathworks; RRID:SCR_001622).

Viral vector and retrograde tracer injection
For selective ChR2 expression in ventral midbrain DA neurons, we injected AAV genotype five encoding ChR2 fused to EYFP (AAV5-EF1a-DIO-ChR-EYFP; UNC Vector Core) into DAT IREScre mice. Mice (P33-36, n = 2 animals) were anesthetized with the ketamine/xylazine mixture. A glass pipette (PCR micropipettes, Drummond), pulled to a tip diameter~20 mm, was lowered to just above the ventral midbrain (coordinates relative to bregma: À3.3 mm antero-posterior (AP), À4.3 mm dorsoventral (DV), and 0.5 mm lateral (L) for the VTA and L 1.3 mm for the SN) and pressure injections of 0.5 ml of AAV5-EF1a-DIO-ChR2-EYFP (titer 1.5 Â 10 12 virus molecules/ml) were made unilaterally in two locations. The pipette was left in place for~3 min to minimize back flow along the injection tract, then withdrawn, and the mouse allowed to recover. Slice recordings were done 28-30 days post injection.

Immunohistochemistry
Wild type C57BL6J mice (P72-93) for ChI cell counts, mice 3 days after injection of CTB, 3 weeks after injection of CAV2, or 4 weeks after injection of INTRSECT virus were anesthetized with ketamine/xylazine and perfused with cold phosphate buffered saline (PBS), followed by 4% paraformaldehyde (PFA). Brains were removed and post-fixed for 2-16 hr in 4% PFA. Coronal sections, 50 mm the beads, which were then washed three times with high-salt buffer (Tris pH 7.4 50 mM, KCl 300 mM, MgCl2 12 mM, NP-40 1%, DTT 1 mM, cycloheximide 100 mg/l). After the final wash, each sample of beads was resuspended in 350 ml RLT buffer (RNeasy Micro Kit, Qiagen) with b-mercaptoethanol (bME; 10 ml/ml, Gibco). The suspension was then vortexed at full speed for 30 s, and put on the magnetic rack again to remove the beads, and the supernatant was then used as the immunoprecipitation (IP) fraction. Similarly, 350 ml RLT buffer with bME was added to the input fraction, which was vortexed for 30 s and the RNA extracted. Both IP and input samples were eluted in 17 ml water.
After extraction, RNA was quantified using the Quant-iT RiboGreen RNA Assay Kit (ThermoFisher Scientific). The measured amount of RNA, in a volume of 17 ml, was in the range of 1.7-22.4 ng for IP samples, and 104-609 ng for input samples. RNA was reverse transcribed, from 16 ml of the 17 mL RNA solution, with the RT2 First Strand Kit (Qiagen). The resulting cDNA was stored at À20˚C pending quantitative PCR (qPCR) determinations. qPCR was performed in Custom RT2 Profiler PCR Arrays (Qiagen, 96 well, #330171, CLAM23840) using RT2 SYBR Green qPCR Mastermix (Qiagen). In addition to the genes of interest, mGluR1, mGluR5, TrpC3 and TrpC7, other genes analyzed included ChAT and VAChT as IP controls, and D1, D2 and D5 receptors as genes of known differential expression in ChIs. GAPDH and b-actin were measured as housekeeping genes. RT controls included a positive PCR control and negative genomic DNA control. cDNA from IP samples was used for PCR without dilution, while cDNA from input samples was diluted 1:1 (with water). PCR was done with a CFX96 Touch thermocycler (BioRad), following a cycle protocol of 95˚C for 10 min, 40 cycles of 95˚C for 15 s and 60˚C for 1 min, followed by a melting curve. Genomic DNA controls were not amplified in any samples. Expression was normalized to GAPDH using Ct values (DCt). Differences in expression between dStr regions were expressed as 2 -DCt . For analysis of enrichment of expression in ChIs, the DCt's of the IP sample and the input sample were compared using DDCt, and the fold-change 2 -DDCt calculated. A fold-change greater than one reflects enrichment in the IP sample. One replicate was omitted as it had more than a 3 SD deviation in 2 -DDCt .

Statistical analysis
Sample size estimation was done with G*Power 3.1 (Heinrich Heine University, Dusseldorf; RRID: SCR_013726), setting a=0.05, and power = 0.9. Effect size was estimated from previous experiments. For t-tests, effect size for experiments anticipating complete blockade (e.g. CNQX effects on fast EPSCs) was set at 2.5. For experiments anticipating partial blockade or enhancement (e.g. sulpiride effects on PSCs), effect size was set at 1.5. These analyses required an n per group of 5 or 7, respectively. For repeated measures ANOVA (including mixed ANOVA), effect sizes for complete blockade and partial blockade were set at 0.8 and 0.5, giving a required n per group of 4 or 7, respectively. For regional comparisons (independent sample comparisons), we set an effect size of 0.8, based on previous recordings from different striatal subregions, giving a required n per region of 10. We did not use non-parametric tests, because (1) generally non-parametric tests are less sensitive with small sample numbers (e.g. less than 20), (2) the variables we measured were continuous numeric variables (not ranked variables), which are likely to show a normal distribution, and (3) nonparametric alternatives do not exist for some parametric tests (e.g. mixed ANOVA).
Statistical analysis was done with SPSS 23 (IBM; RRID:SCR_002865) and JASP ver 0.8.6 (JASP Team, 2018; jasp-stats.org; RRID:SCR_015823). Comparisons of two values were done using a t-test. When sample size (biological replicate number) was smaller than 10 or variances were not equal, a t-test without assumption of equal variances (Welch's t-test) was used. For evaluation of drug effects, comparison was made on a percent basis to the pre-drug response (100%) using a one-sample t-test. For more than three variable comparisons, ANOVA was used. In repeated-measures ANOVA, when sphericity was violated, Greenhouse-Geisser correction was conducted. For gene expression studies, regional differences were compared using a general linear model multivariate analysis. For one-way or two-way ANOVA, Scheffe's post-hoc test was used to identify significant differences. For mixed ANOVA, when significant interactions were found, post-hoc t-tests were done for betweensubject effects. For CTB cell counts, c 2 test was used. For CAV2 counts, a mixed ANOVA was used, as c 2 failed with zeros in some cells. p values smaller than 0.05 were regarded as significant. Data are reported as mean ± S.E.M., unless otherwise noted. In the graphs, dots show the average measurements for each biological replicate, and bars show the mean and S.E.M. for all biological replicates. Exact values of n, what n represents, p values, and F values for ANOVA tests are presented in the figure legends. Numbers of animals used for electrophysiological recordings are indicated in the figure legends. For the main electrophysiological experiments, no more than 3 cells were recorded per animal; for the VGLUT2 cKO experiments and AAV-DIO-ChR2 injection experiments, 3 to 6 cells were recorded per animal. p values are shown to the third decimal place, so p=0.000 reflects p<0.001. Data availability Source data are uploaded as supplements to figures.