Dopamine oppositely modulates state transitions in striosome and matrix direct pathway striatal spiny neurons

Summary

Corticostriatal synaptic integration is partitioned among striosome (patch) and matrix compartments of the dorsal striatum, allowing compartmentalized control of discrete aspects of behavior. Despite the significance of such organization, it’s unclear how compartment-specific striatal output is dynamically achieved, particularly considering new evidence that overlap of afferents is substantial. We show that dopamine oppositely shapes responses to convergent excitatory inputs in mouse striosome and matrix striatal spiny projection neurons (SPNs). Activation of postsynaptic D1 dopamine receptors promoted the generation of long-lasting synaptically-evoked “up-states” in matrix SPNs but opposed it in striosomes, which were more excitable under basal conditions. Differences in dopaminergic modulation were mediated, in part, by dendritic voltage-gated calcium channels (VGCCs): pharmacological manipulation of L-type VGCCs reversed compartment-specific responses to D1 receptor activation. These results support a novel mechanism for the selection of striatal circuit components, where fluctuating levels of dopamine shift the balance of compartment-specific striatal output.

Graphical Abstract

eTOC

Prager et al. show that dopamine promotes the maintenance of dendritically-evoked “up-states” in mouse direct pathway matrix SPNs, but opposes it in striosomes. This requires postsynaptic D1 receptors, and involves differential engagement of L-type Ca²⁺ channels. These findings reveal a mechanism where fluctuations in dopamine may constrain compartment-specific striatal output.

Introduction

The dorsal striatum integrates convergent afferents from diverse cortical and subcortical brain regions to guide movement and motor-decision making. This involves the coordinated synaptic engagement of two complementary populations of striatal spiny projection neurons (SPNs), which ground the direct and indirect pathways of the basal ganglia (Cui et al., 2013; Gerfen and Surmeier, 2011). But there is a deeper and less understood organization embedded within the striatal network: SPNs are segregated into two anatomically- and chemically- defined compartments, known as striosomes (patches) and matrix (Gerfen, 1984; Graybiel and Ragsdale, 1978; Pert et al., 1976). Direct synaptic communication between striosome and matrix SPNs is minimal, and compartment-specific activity has been associated with unique behavioral tasks (Brimblecombe and Cragg, 2017; Canales and Graybiel, 2000; Crittenden and Graybiel, 2011; Friedman et al., 2017; Friedman et al., 2015; Jenrette et al., 2019; Nadel et al., 2020; Yoshizawa et al., 2018). Despite growing evidence that such compartmentalization is a crucial component of basal ganglia network function, our understanding of the mechanisms that regulate striosome vs matrix output is limited.

All striatal output is carried by SPNs. As striosomes account for only ~10–15% of striatal volume (Brimblecombe and Cragg, 2017; Crittenden and Graybiel, 2011), and few studies have attempted to make compartment-targeted recordings (Banghart et al., 2015; Crittenden et al., 2017; Kawaguchi et al., 1989; McGregor et al., 2019; Salinas et al., 2016; Smith et al., 2016), most of what we know about SPN function likely pertains to the matrix. SPNs are silent in the absence of synaptic drive, requiring coordinated glutamatergic innervation to overcome their hyperpolarized resting membrane potential and fire action potentials (Blackwell et al., 2003; Gerfen and Surmeier, 2011; Stern et al., 1998). In presumptive matrix SPNs these criteria can be fulfilled by the convergent activation of only about a dozen glutamatergic synapses, provided they are 1) spatially close to each other and 2) located on a distal dendrite (~100 μm or more from the soma) (Dorman et al., 2018; Du et al., 2017; Plotkin et al., 2011). Such synaptic synchronization induces a regenerative dendritic plateau potential that pushes the soma into a long-lasting depolarized “up-state” from which action potentials can fire (Du et al., 2017; Plotkin et al., 2011).

How are SPNs in striosomes and matrix differentially and task-dependently engaged? One mechanism involves preferential innervation by certain limbic-associated cortical regions (Friedman et al., 2017; Friedman et al., 2015; McGregor et al., 2019). But as a singular, generalizable explanation, this is problematic for two reasons: 1) recent work suggests that overlap of afferents to striosome and matrix SPNs is more substantial than previously appreciated (Smith et al., 2016) and 2) neuromodulation is likely a crucial determinant (Banghart et al., 2015; Brimblecombe and Cragg, 2015, 2017; Crittenden et al., 2017; Inoue et al., 2016; Miura et al., 2007; Prager and Plotkin, 2019; Salinas et al., 2016), which is highlighted by the fact that chronic intermittent administration of psychostimulants or dopamine agonists induces compartment-specific changes in immediate early gene expression (Canales and Graybiel, 2000). An unexplored possibility is that there are intrinsic differences in how striosome and matrix SPNs transform converging excitatory inputs into somatic state-transitions, and how this process is modulated by dopamine. Here we show that responses to convergent glutamatergic inputs are differentially modulated by dopamine: activation of D1 dopamine receptors (D1Rs) in D1R-expressing SPNs prolonged the duration of dendritically evoked up-states in the matrix but shortened the duration in striosomes. The opposing actions of D1R signaling were not due to differences in dopamine receptor activation and could be reversed by pharmacologically blocking or enhancing postsynaptic L-type voltage-gated Ca²⁺ channel (VGCC) activity.

Results

Since elevated striosome activity is often associated with hyperkinetic behavioral changes (Canales and Graybiel, 2000; Crittenden et al., 2017), and activity of D1R-expressing direct pathway SPNs (dSPNs, as opposed to D2R-expressing indirect pathway SPNs) is associated with action initiation (Gerfen and Surmeier, 2011; Kravitz and Kreitzer, 2012), we restricted our study to dSPNs residing in each compartment. To visualize dSPNs in striosome and matrix compartments, we crossed transgenic mice expressing eGFP under the control of the Nuclear receptor subfamily 4, group A, member 1 promoter (Nr4a1-eGFP; labeling striosomes) (Davis and Puhl, 2011) with drd1-tdTomato transgenic mice (labeling dSPNs) (Shuen et al., 2008) (Fig 1A, B). Consistent with previous reports using different transgenic mouse lines (Crittenden et al., 2017; McGregor et al., 2019; Smith et al., 2016) striosome dSPNs were more excitable than their matrix counterparts, as reflected by higher input resistance, lower rheobase current and larger miniature excitatory postsynaptic potential (mEPSP) amplitudes (Fig 1C, S1A–H). Though morphologically similar, total dendritic length was reduced in striosome dSPNs (S1I–L), a feature known to result in increased SPN somatic excitability (Gertler et al., 2008), and likely a reflection of the spatial constraints imposed by strict compartmental boundaries (Brimblecombe and Cragg, 2017; Fujiyama et al., 2011).

Figure 1. Matrix and striosome dSPNs support dendritically-evoked somatic up-states.

(A) Immunohistochemical localization of the striosome marker μ-opioid receptor (MOR; red) in a Nr4a1-eGFP (green) mouse (top). 2-photon maximum intensity projection (2P-MIP) image showing recording electrode placement in a parasagittal brain slice from a Nr4a1-eGFP × drd1-tdTomato mouse (bottom; scale bars = 500 μm).

(B) 2P-MIPs of recorded matrix and striosome dSPNs (scale bar = 30 μm).

(C) Corresponding current-voltage (IV) traces of matrix (top) and striosome (bottom) dSPNs.

(D-E) Representative somatic voltage traces in response to 2PLUG at a single distal dendritic spine (red) or multiple distal dendritic spines stimulated at threshold for up-state induction. 2P-MIPs of distal dendrites showing locations of 2PLUG (yellow circles) (right, scale bar = 2 μm).

(F) Somatic voltage traces of dendritically-evoked up-states aligned to the end of 2PLUG stimulation and normalized to peak amplitude.

(G) Box plots showing somatic up-state durations. m = matrix; s = striosome; ns = not significant.

(H) Box plots showing somatic membrane potentials of “down-” and “up-” states. *p<0.05, ***p<0.001. See also Figures S1–4 and Table S2.

Striatal output is ultimately dependent upon the capacity of SPNs to integrate coordinated synaptic events (Gerfen and Surmeier, 2011; Goldberg et al., 2003; Stern et al., 1998; Wilson and Kawaguchi, 1996). We therefore determined if dSPNs residing in striosomes vs matrix differ in their capacity to support dendritically-evoked up-states, a direct assay of SPN-intrinsic glutamatergic synaptic integration (Du et al., 2017; Plotkin et al., 2011). We evoked up-states in dSPNs by using 2-photon laser uncaging of glutamate (2PLUG; assisted by 2-photon laser scanning microscopy) to convergently activate 10–18 neighboring distal dendritic spines, independent of presynaptic identity (Fig 1D, E; S2). While both striosome and matrix dSPNs supported the generation of up-states, striosome dSPNs achieved a significantly more depolarized up-state membrane potential than their matrix counterparts (Fig 1F–H). Elevated up-state amplitude correlated with higher somatic input resistance (r = 0.3475; p = 0.0004), and was not due to differences in 1) synaptic strength (single spine 2PLUG-evoked EPSPs used to evoke up-states had similar amplitudes), 2) postsynaptic glutamate receptor composition (the amplitudes, kinetics and ratios of NMDA- and AMPA- receptor mediated 2PLUG-evoked current components, as well as miniature excitatory postsynaptic currents (mEPSCs), were similar) or 3) synaptic density (S3–4). Differences in up-state amplitudes did not translate to differences in the induction threshold or maintenance of up-states: no differences were observed in the number, density or dendritic location of spines required to evoke up-states, or the duration of up-states once evoked (Fig 1G, H; S2).

To predict how these intrinsic neuronal differences may impact compartment-specific striatal output we created data-derived computational models of striosome and matrix dSPNs. Detailed biophysical models were built from reconstructions of dSPNs (selected to capture the relative differences in dendritic length reported in S1), endowed with axospinous synaptic inputs and ion channels known to be integral to SPN function, and parametrically tuned using an iterative random selection process to match experimentally generated data (Fig 2A; S5) (Dorman et al., 2018; Jedrzejewski-Szmek et al., 2018). Corroborating our experimental results, simulated somatic up-state membrane potentials were significantly more depolarized in striosome vs matrix dSPNs (S5B, C). Further corroborating experimental results, a random forest regression revealed that the main determinants of up-state amplitude were voltage-gaged ion channels, not synaptic AMPA or NMDA receptors (S5E). Dynamic estimates of rheobase current demonstrated that membrane potential mapped directly onto excitability (Fig 2B, C), suggesting that ongoing excitatory synaptic drive will more readily evoke spiking during up-states in striosome vs matrix dSPNs. Indeed, applying randomly distributed synaptic inputs during simulated up-states showed this to be the case (Fig 2D, E). Moreover, these simulations suggest that striosome and matrix dSPNs respond differently to varying levels of ongoing afferent activity: lower frequencies of synaptic drive preferentially increased the firing rate of striosome dSPNs during up-states.

Figure 2. Biologically realistic computational models predict dSPN-intrinsic differences in compartment-specific output activity.

(A) Reconstructions of SPNs used to create striosome and matrix dSPN models.

(B) Box plots showing dynamic measures of rheobase currents during simulated down- and up- states, determined by 100 ms somatic current injection.

(C) Box plots showing firing rate induced by indicated current injections (100 ms) during simulated up-states.

(D) Simulations showing somatic responses to either dispersed excitatory synaptic inputs alone (at the indicated frequency) or dispersed excitatory synaptic inputs during a dendritically-evoked up-state.

(E) Box plots showing the change in firing rate caused by the addition of a dendritically-evoked up-state during dispersed excitatory synaptic inputs at indicated frequencies (timing same as in D). ns = not significant, *** p<0.001. See also Figure S5 and Tables S1 and S2.

While cell-intrinsic differences lay the groundwork for similar convergent synaptic stimulation to favor striosomal output, it cannot by itself explain bi-directional and task-dependent differences in compartment-specific activity. Both striatal circuit function and the behaviors it encodes are shaped by dopamine (Gerfen and Surmeier, 2011; Kravitz and Kreitzer, 2012; Surmeier et al., 2010). Furthermore, D1R activation enhances both neuronal excitability and the duration of dendritically-evoked up-states in uncategorized dSPNs (Lahiri and Bevan, 2020; Plotkin et al., 2011). Might elevations in dopamine bias the engagement of dSPNs in a compartment-specific manner? To test this, we pharmacologically activated D1Rs and determined the effect on dendritically-evoked up-states in striosome and matrix dSPNs. Strikingly, application of the selective D1R agonist dihydrexidine hydrochloride (DHX, 10 μM) had opposite effects on the maintenance of up-states, increasing up-state duration in matrix dSPNs (~30% median increase) and decreasing it in striosome dSPNs (~12% median decrease) (Fig 3A, B). This effect was blocked by pre-application of the selective D1R antagonist SCH 23390 hydrochloride (3μm) in both matrix and striosome dSPNs (S6A, B). Somatic up-state amplitudes were unaffected by DHX in matrix dSPNs (N=9, pre-drug median = −69.22 mV, post-drug median = −68.41 mV, W = −5, p = 0.8203) but were slightly reduced in striosome dSPNs (N=13, pre-drug median = −64.38 mV, post-drug median = −66.59 mV, W = −67, p = 0.0171). This alone cannot account for the DHX-induced shortening of up-states in striosome dSPNs, however, as there was not a significant correlation between changes in up-state amplitude and duration in these neurons (r=0.2207, p=0.4686).

Figure 3. D1R activation oppositely modulates dSPN up-state durations in matrix vs striosomes.

(A-B) Somatic up-state traces induced in (A) a matrix and (B) a striosome dSPN before and after application of the D1R agonist DHX. Arrows indicate direction of change in duration caused by DHX. Box plots show somatic up-state durations before and after DHX.

(C) Schematic of experimental design used in F-G.

(D) Line scan measuring changes in fluorescence of the genetically encoded dopamine sensor dLight 1.1 in dorsal striatum neuropil.

(E) Schematic of stimulation protocol used to coordinate 2PLUG-evoked up-states with endogenous dopamine release. Example somatic up-state from 1D shown above to illustrate timing.

(F-G) Somatic up-state traces induced in (F) a matrix and (G) a striosome dSPN before and after optogenetic stimulation of nigrostriatal dopaminergic (DAergic) fibers. Arrows indicate direction of change in duration caused by D1R activation. Box plots show somatic up-state durations before and in response to optogenetic stimulation.

(H) Somatic up-state traces in the presence of the D1R antagonist SCH 23390, before and in response to optogenetic simulation of nigrostriatal DAergic fibers.

(I) Box plots showing the percent change in somatic up-state duration caused by pharmacological activation of D1Rs in Nr4a1-eGFP × Drd1-tdTomato mice (from A-B) and WT × Drd1-tdTomato mice (from Fig S6D, E), or by transient optogenetic stimulation of DAergic fibers (from F-G). *p<0.05, **p<0.01, ***p<0.001. See also Figure S6 and Table S2.

Because transgenic expression of GFP has been linked to alterations in SPN electrical properties and responsiveness to dopamine (Chan et al., 2012; Cirnaru et al., 2019; Crittenden et al., 2014), we examined compartment-specific dopaminergic modulation in an alternate mouse model that does not rely on striosome- or matrix- specific transgenic markers. Compartments were visualized by unilaterally injecting an adeno-associated virus (AAV) expressing ChR2-EYFP into the left anterior cingulate cortex (ACC) of drd1-tdTomato mice. This produces robust axonal labeling in the ipsilateral striatum that largely avoids striosomes (Friedman et al., 2015). The location of each recording was later affirmed by post-hoc mu opioid receptor staining of recovered brain slices (S6C). Confirming our findings in Nr4a1-eGFP mice, we observed that DHX (10 μM) caused a ~22% median increase in up-state duration in matrix dSPNs and a ~41% median decrease in striosome dSPNs (S6D, E).

The data above show that prolonged pharmacological activation of D1Rs oppositely modulates the duration of synaptically-evoked up-states in striosome vs matrix dSPNs. The obvious question that arises is whether or not transient release of endogenous dopamine from substantia nigra pars compacta (SNc) neuron axons produces the same effect. To address this, we prepared acute brain slices from mice that received bilateral injections of an AAV expressing ChR2-mCherry into the lateral SNc. Local optogenetic stimulation of SNc axon fibers was optimized to evoke phasic elevations of dopamine that began just prior to and lasted during synaptically-evoked up-states (Fig 3C–E). Consistent with pharmacological experiments, phasic release of endogenous dopamine significantly increased the duration of synaptically-evoked up-states in matrix dSPNs (~61% median increase) and decreased the duration in striosome dSPNs (~9% median decrease) (Fig 3F, G). The effect of optogenetically evoked dopamine release on up-state duration was blocked by the D1R antagonist SCH 23390 in both neuron types (Fig 3H). Together these data demonstrate that D1R activation oppositely modulates state-transition durations in matrix and striosome dSPNs, and that this mechanism is 1) not due to the compartment labeling strategy used and 2) engaged by transient release of endogenous dopamine (Fig 3I)

How does activation of D1Rs produce opposite effects in dSPNs residing in different compartments? As these experiments were conducted in the presence of tetrodotoxin (1 μM) to prevent axonal propagation of presynaptic activity, the mechanism is likely postsynaptic and intrinsic to the SPNs themselves. SPN dendritic plateau potentials, and resulting up-states, are NMDA receptor-dependent (Plotkin et al., 2011). In presumptive matrix dSPNs, activation of D1Rs positively modulates NMDA receptor function (Cepeda et al., 1998), promoting an increase in up-state duration (Plotkin et al., 2011). D1R modulation of NMDA receptors in SPNs requires the activation of L-type voltage-gated Ca²⁺ channels (L-VGCCs) (Liu et al., 2004), which are enriched in SPN dendrites (Carter and Sabatini, 2004; Plotkin et al., 2013) but not required for the generation or maintenance of up-states under basal conditions (Plotkin et al., 2011). Could the failure of D1R activation to prolong up-state duration in striosome dSPNs be due to lower engagement of dendritic L-VGCCs? Two obvious mechanisms could lead to such a scenario: 1) basal engagement of L-VGCCs is lower or 2) D1R-enhancement of L-VGCC engagement is impaired. We tested the first hypothesis by comparing depolarization-evoked dendritic Ca²⁺ transients in the two compartments. Somatically-delivered depolarizing voltage steps evoked similar Ca²⁺ transients in distal dendritic spines of striosome and matrix dSPNs (Fig 4A, B). The contribution of L-VGCCs to these Ca²⁺ transients, however, was nearly 50% less in striosome dSPNs, as determined by application of the L-VGCC blocker isradipine (5 μM) (Fig 4A, C). Depolarization-evoked Ca²⁺ influx into SPN dendrites is mediated by a combination of L-, T- and R- type VGCCs (Carter and Sabatini, 2004). To determine if D1R modulation of L-VGCCs is impaired in striosome dSPNs we isolated the L-VGCC- mediated component of spinous Ca²⁺ transients by delivering a cocktail of ML 218 hydrochloride (1 μM) and SNX 482 (0.3 μM) to block T- and R- type VGCCs, respectively (Carter and Sabatini, 2004; Xiang et al., 2011). Confirming the results of the isradipine blockade experiments, the remaining L-VGCC-mediated component was significantly lower in striosome dSPNs (Fig 4D). Application of DHX enhanced L-VGCC- mediated Ca²⁺ transients similarly in both neuron populations (matrix median increase = 13.9%; striosome median increase = 12.5%), indicating that the signaling linkage between D1Rs and L-VGCCs is intact in striosome dSPNs (Fig 4E). Despite this, Ca²⁺ influx through L-VGCCs remained lower in striosome vs matrix dSPNs even in the presence of the D1R agonist. In fact, D1R-enhanced L-VGCC Ca²⁺ transients in striosome dSPNs were still smaller than unmodulated transients in matrix dSPNs (Fig 4E).

Figure 4. Bidirectional D1R-modulation of striosome and matrix dSPN somatic up-state duration is sensitive to L-VGCC engagement.

(A) Average distal dendritic spine Ca²⁺ transients (shaded areas are SEM) in the presence or absence of the L-VGCC channel antagonist isradipine. 2P-MIPs showing location of line scan (yellow) to the left (scale bars = 2 μm, 500 ms, 0.4 ΔG/Gsat; voltage step is to 0 mV).

(B) Boxplots showing depolarization-evoked Ca²⁺ transient amplitudes in distal dendritic spines.

(C) Boxplots showing the percent block of distal dendritic spine Ca²⁺ transients by isradipine.

(D) Boxplots showing distal dendritic spine Ca²⁺ transient amplitudes in the presence of the T- and R-type VGCC antagonists ML 218 and SNX 482.

(E) Boxplots showing distal dendritic spine Ca²⁺ transient amplitudes in the presence of ML 218 and SNX 482 before and after application of DHX. Even after application of DHX, Ca² transients in striosomes remained lower than pre-DHX transients in the matrix.

(F-G) Somatic up-state traces induced in a matrix dSPN in the presence of isradipine (F) and a striosome dSPN in the presence of the L-VGCC agonist Bay K 8644 (G), before and after application of DHX. Box plots show somatic up-state durations.

(H-I) Somatic up-state traces induced in a matrix dSPN in the presence of Bay K 8644 (H) and a striosome dSPN in the presence of isradipine (I), before and after application of DHX. Box plots show somatic up-state durations.

(J) Box plots showing the percent change in somatic up-state duration induced by DHX in matrix and striosome dSPNs under control conditions (from Fig 3A-B), in the presence of isradipine and in the presence of Bay K 8644. *p < 0.05, **p < 0.01; ***p < 0.001. See also Table S2.

If the opposing effects of D1R activation on up-state duration are due to differences in dendritic L-VGCCs, then reversing compartmental differences in L-VGCC engagement should be able to reverse the compartment-specific responses to dopamine. Indeed, this was the case. Increasing L-VGCC activation with the L-VGCC agonist Bay K 8644 (1 μM) caused DHX to increase up-state duration in striosome dSPNs (14% median increase) and blocking L-VGCC activation with isradipine caused DHX to decrease up-state duration in matrix dSPNs (47.6% median decrease), essentially flipping the compartmental phenotypes (Fig 4F, G). Accentuating compartmental differences in L-VGCC engagement, on the other hand, further altered the responses to D1R activation: blocking L-VGCC activation with isradipine eliminated the effect of DHX on up-state duration in striosome dSPNs and increasing L-VGCC activation with Bay K 8644 caused DHX to decrease up-state duration in matrix dSPNs (16% median decrease) (Fig 4H, I). Together, these data suggest that not only is L-VGCC engagement an important determinant of compartment-specific responses to D1R activation, but graded engagement of L-VGCCs does not shape the response to dopamine in a purely linear way (Fig 4J). These data also demonstrate that while both populations of dSPNs possess the requisite machinery for dopamine to suppress up-state maintenance, the mechanism is likely compartment-specific, normally masked by L-VGCCs in matrix dSPNs, and only L-VGCC-dependent in striosome dSPNs.

Discussion

Our results describe a novel framework that resolves how regulation of compartment-specific striatal output may be achieved, independent of the identity or overlap of excitatory afferents. Our data suggest that under basal conditions dSPNs in striosome and matrix compartments integrate convergent excitatory synaptic inputs similarly, though ensuing somatic up-states reside closer to spike threshold in striosome dSPNs. Elevations in dopamine will reduce the time window during which direct pathway striosome SPNs can fire action potentials, while prolonging the time window in matrix dSPNs, biasing ongoing striatal output away from striosomes and towards the matrix. While the full complexity of the underlying mechanism must still be unraveled, the engagement of L-VGCCs plays a key role in determining the opposing responses to dopamine, offering a molecular target through which this process can be manipulated and tuned.

The present study demonstrates that the “associative capacity” (ability to generate and support dendritically-evoked up-states in response to coincident synaptic activation) of dSPNs is independent of the compartment in which they reside. This is consistent with our observations that the density and relative glutamate receptor contributions of dSPN dendritic spines are similar across compartments. Though the associative capacity of individual dendrites may be similar, our data imply that the frequency of dendritic plateaus and total resulting spikes per neuron may differ between compartments under physiological conditions. This is because 1) while proximal dendritic length and arborization are similar between dSPN types, the total real-estate of the dendritic tree that is capable of supporting synaptically-driven plateaus is approximately 55% less in striosomes (total measurable dendritic length >90 μm from soma: 143.4 μm, median, matrix; 79.7 μm, median, striosomes; see S1I–L), 2) while up-state durations are similar between dSPN types, up-state amplitudes reside closer to spike threshold in striosomes, increasing the likelihood of spike generation triggered by synaptic inputs arriving during the up-state (Plotkin et al., 2011; Stern et al., 1998) and 3) simulations in biologically realistic computational models revealed that lower frequencies of ongoing excitatory synaptic inputs are required to drive spiking during up-states in striosome vs matrix dSPNs. This sets a scenario where, under basal conditions, dSPNs in striosomes may display fewer dendritic plateaus (on a whole cell basis) but track ongoing cortical and thalamic activity with higher fidelity during each plateau.

Activation of D1Rs had opposite effects on up-state duration in striosome and matrix dSPNs. Since 1) D1R activation enhances dendritic NMDA receptor function and up-state duration in presumptive matrix dSPNs (Higley and Sabatini, 2010; Plotkin et al., 2011), 2) this enhancement requires Ca²⁺ entry through L-VGCCs (Cepeda et al., 1998; Liu et al., 2004) and 3) dendritic L-VGCC engagement is lower in striosome dSPNs, one would predict that D1R activation would simply have no effect on up-state duration when the requisite signaling elements for NMDA receptor enhancement are absent. So how does D1R activation decrease up-state duration in striosome dSPNs and in matrix dSPNs when L-VGCCs are blocked? Our data suggest that under permissive conditions D1R activation can decrease up-state duration in both neuron types, though the underlying mechanisms likely differ (requiring low engagement of L-VGCCs in striosome dSPNs and a L-VGCC-independent mechanism in matrix dSPNs). While dissecting the mechanisms is beyond the scope of this study, several possibilities may explain the reducing effect of D1Rs that is unmasked in matrix dSPNs when L-VGCCs are blocked. One potential mechanism involves modulation of other VGCCs. Inhibition of T-Type VGCCs reduces up-state duration in presumptive matrix dSPNs (Plotkin et al., 2011). Dopaminergic modulation of NMDA receptors in SPNs requires activation of PKA (Higley and Sabatini, 2010; Snyder et al., 1998). PKA activation can suppress T-type VGCC conductance in heterologous systems and isolated smooth muscle cells (Harraz and Welsh, 2013; Hu et al., 2009), though such modulation in SPNs remains to be determined. A second possibility is that D1R-induced PKA activation leads to the enhancement of a tonic GABA current (Janssen et al., 2009) that may attenuate the generation of dendritic plateau potentials (Du et al., 2017). Compartmental differences in GABAergic signaling have been reported (Banghart et al., 2015; Friedman et al., 2017), but as the MNI caging compound used in this study diminishes GABAergic signaling (Palma-Cerda et al., 2012), the impact that such a mechanism may have under our experimental conditions remains to be determined.

Activation of L-VGCCs was required for both the enhancing and reducing effects of D1R activation in striosome dSPNs, and the degree of L-VGCC activation determined the direction of the dopaminergic effect. Somewhat surprisingly, pharmacologically enhancing the already higher L-VGCC engagement in matrix dSPNs caused D1R activation to diminish up-state duration, rather than further augment it. Together this suggests that there is a nonlinear or “inverted U” relationship between L-VGCC engagement and response to D1R activation in both dSPN types. Such nonlinear effects of Ca²⁺ have been well described in numerous synaptic and behavioral phenomena and are typically attributed to Ca²⁺ concentration-dependent activation of phosphatases vs kinases (Dash et al., 2007; Lisman, 2001). But why does further enhancing L-VGCC engagement in matrix dSPNs reverse the effect of dopamine? Though the precise mechanism and physiological relevance of this remain to be determined, one possibility is that the boosted level of intracellular Ca²⁺ achieved in matrix dSPNs by Bay K 8644 application overwhelms spatial restrictions that normally confine the engagement of Ca²⁺-activated hyperpolarizing conductances (Bloodgood and Sabatini, 2007).

Manipulating L-VGCC engagement is sufficient to determine the directional effects of dopamine on dSPN up-state duration. That said, there are myriad targets of D1R signaling in SPNs that may also be involved (Surmeier et al., 2009). Taking advantage of our biologically realistic SPN models we used a random forest regression, which identifies what features are most important for predicting the dependent variable, to search for other possible candidates that may regulate up-state duration (S5F). Simulations revealed 3 potential targets: NMDA receptors, R-type VGCCs and Kv1.2 potassium channels. NMDA receptors and R-type VGCCs are already known to control SPN dendritic plateau duration (though modulation of R-type VGCCs by D1R signaling in SPNs is unclear) (Plotkin et al., 2011). Interestingly, D1R modulation of Kv1.2 has recently been shown to enhance the excitability of dSPNs during up-state transitions (Lahiri and Bevan, 2020) - our simulations suggest that dopaminergic modulation of this channel may also regulate the duration of up-states, though whether this contributes to compartmental differences remains to be determined.

Previous studies have shown that striosome dSPNs send direct axonal projections to dopaminergic neurons of the SNc, as well as other basal ganglia nuclei (Crittenden and Graybiel, 2011; Fujiyama et al., 2011). Though matrix dSPN axons may share a grossly similar trajectory (Smith et al., 2016), recent studies have demonstrated that striosome-originating SPN axons form tightly woven integrative units encapsulating dopaminergic neurons within the SNc, which in turn send dopaminergic projections throughout the dorsal striatum (Crittenden et al., 2016; Davis et al., 2018). As dSPN-SNc synapses are GABAergic (Brazhnik et al., 2008), it’s tempting to speculate that such projections may serve as a homeostat to keep striatal dopamine levels in check (Crittenden et al., 2016). If this is the case, our data suggest that elevations in striatal dopamine would decrease the duration of striosomal inhibition of SNc dopaminergic neurons, ultimately shunting such a feedback mechanism. This has fundamental implications for understanding how fluctuations in dopamine modulate basal ganglia network activity during goal-directed and habit learning (Darvas et al., 2014; Leventhal et al., 2014; Palmiter, 2008; Pennartz et al., 2009; Yin et al., 2004).

Overall, our data demonstrate that dopaminergic signaling through D1Rs can modulate the balance of compartment-specific striatal output. The opposing effects of D1R activation on synaptic integration in striosome and matrix dSPNs are 1) due, at least in part, to intrinsic differences in the compliment of dendritic VGCCs and 2) not dependent upon the presynaptic identity of converging axo-spinous inputs. As previous studies have suggested that up-state generation and action potential initiation are regulated by separate populations of synaptic inputs (Plotkin et al., 2011; Stern et al., 1998), dopamine may serve to act as a filter which temporally constrains the synaptic cues that can evoke dSPN spiking in striosomes following an associative event, while loosening this constraint in the matrix. These findings suggest that the gating of compartment-specific striatal output, a phenomenon now appreciated to be crucial for task-dependent behaviors (Bloem et al., 2017; Friedman et al., 2017; Friedman et al., 2015), represents an important parameter in the study of dopamine-dependent behavior and pathology involving the basal ganglia.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents may be directed to and will be fulfilled by Lead Contact, Dr. Joshua Plotkin, Joshua.plotkin@stonybrook.edu.

Materials Availability

This study did not generate new unique reagents.

Data and Code Availability

Detailed datasets supporting the current study are available from the corresponding author on request. Source code is available in ModelDB at http://modeldb.yale.edu/266807.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Animals

All experimental procedures were performed in accordance with the United States Public Health Service Guide for Care and Use of Laboratory Animals (Institute of Laboratory Animal Resources, National Research Council), were approved by the Institutional Animal Care and Use Committee at Stony Brook University and conformed with the ARRIVE guidelines. Animals (n = 79; 33 males, 46 females) were housed 2–5 per cage (1–2 per cage if animals underwent surgical procedures) in an environmentally controlled room (20–23 °C, 30–70% humidity, 12 h light/dark cycles [350–400 lux]) with food and water available ad libitum. Cages were cleaned once a week and animal handling was minimized to reduce additional animal stress (Prager et al., 2011). No physical enrichment was provided. Experiments were conducted in 1.5–3-month-old male and female double-transgenic mice created by crossing Tg(Nr4a1-eGFP139G_SAT) (Davis and Puhl, 2011) and Drd1a-tdTomato BAC (Ade et al., 2011; Shuen et al., 2008) transgenic mouse lines. Nr4a1-eGFP mice were visually genotyped at weaning (~postnatal day 21) with a blue LED and GFP filter-equipped goggles (Electron Microscopy Sciences) (Salinas et al., 2016) and Drd1a-tdTomato mice genotyped as previously described (Shuen et al., 2008). Randomization and blinding were not applicable to this study. Researchers attempted to experiment from both striosome and matrix dSPNs in each animal, when possible.

METHOD DETAILS

Surgical Procedures

Male and female mice were induced and maintained under anesthesia via inhalation of vaporized isoflurane (3% for induction and 1–2% for maintenance during surgery). After complete sedation was confirmed by tail pinch, animals were placed in a stereotaxic apparatus, the scalp was shaved and disinfected and eyes were covered with eye-protective gel (Puralube® Ophthalmic Ointment, Dechra Veterinary Products). Identical custom 1 μl Hamilton micro-volume syringes with a 50° bevel point (Hamilton, ref 65458–1) were used for viral injections targeting the Substantia nigra pars compacta (SNc; ±1.49 mm ML, −3.08 mm AP, and −4.11 mm DV), dorsal striatum (±2.40 mm ML, 0.50 mm AP, and −3.06 mm DV) or anterior cingulate cortex (ACC; +0.5 mm AP, +0.28 mm ML, −1.8 mm DV) and manipulated via a motorized stereotaxic injector (Quintessential Stereotaxic Injector QSI™, Stoelting Co., ref 53311) and Angle Two software (Leica Biosystems). 200 nL pAAV.CAG.hChR2(H134R)-mCherry.WPRE.SV40 (titer of 1×10⁻¹³ vg/ml) (sourced from Karl Deisseroth; Addgene viral prep # 100054-AAV5; http://n2t.net/addgene:100054; RRID: Addgene_100054) was slowly injected into the SNc bilaterally over 5 minutes; an additional 5 minutes were allowed for diffusion before slowly withdrawing the needle over 5 minutes. To qualitatively measure optogenetically-evoked dopamine release, wild type C57BL/6J mice received an additional injection of pAAV.CAG.dLight1.1 (200 nL at titer ≥7×10¹² vg/ml) (sourced from Lin Tian; Addgene viral prep # 111067-AAV5; http://n2t.net/addgene:111067; RRID: Addgene_111067) into the dorsal striatum. To fluorescently label the matrix in WT × Drd1a-tdTomato BAC mice, 200nL of pAAV-hSyn-hChR2(H134R)-EYFP (titer: 1×10¹³ vg/mL) (Addgene, ref 26973-AAV9, RRID: Addgene_26973) was injected unilaterally into the left ACC. Throughout all surgical procedures and during recovery immediately after the procedure, animals were kept on a heating pad to stabilize body temperature at 37 °C. Animals were administered a postoperative analgesic, meloxicam (5mg/kg), via subcutaneous injection at the conclusion of the procedure and once daily thereafter as needed for 7 days. Animals receiving SNc viral injections were sacrificed 2–3 weeks following the surgical procedures; animals receiving ACC injections were sacrificed 10–14 days after surgical procedures. Cells from ACC injected animals were included in the final analysis only if tissue could be successfully processed for μ-opioid receptor staining post hoc to confirm electrode placement.

Electrophysiological Recordings

Ex vivo slice recordings.

Para-sagittal brain slices (275 μm) containing the dorsolateral striatum were obtained, as previously described (Plotkin et al., 2011). Briefly, mice were anesthetized with a mixture of ketamine (50mg/kg) and xylazine (4.5mg/kg), transcardially perfused with ice-cold artificial cerebral spinal fluid (ACSF) containing (in mM): 124 NaCl, 3 KCl, 1 CaCl₂, 1.5 MgCl₂ 26 NaHCO₃, 1 NaH₂PO₄ and 14 glucose and continuously bubbled with carbogen (95% O₂ and 5% CO₂). Slices were cut using a VT-1000s vibratome (Leica Microsystems, Buffalo Grove, IL) and transferred to a holding chamber where they were incubated at 32°C for 45 min in ACSF containing (in mM) 2 CaCl₂ and 1 MgCl₂, after which they were stored at room temperature (~21 °C) until recording (Plotkin et al., 2011).

Patch electrodes (3.5–5 MΩ) were pulled from thick-walled borosilicate glass using a Sutter P-1000 puller (Sutter Instruments, Novato, CA). Current clamp experiments were performed using an internal recording solution containing (in mM): 135 KMeSO4, 5 KCl, 10 HEPES, 2 Mg²⁺-ATP, 0.5 Na⁺-GTP, 5 Tris-phosphocreatine; 5 sodium phosphocreatine, 0.05 Alexa Fluor 568 hydrazide sodium salt (Invitrogen), and 0.2 Fluo-4 pentapotassium salt (Invitrogen), pH 7.25, 270–280 mOsm. Voltage clamp experiments were performed using an internal recording solution containing (in mM): 120 CsMeSO₃, 5 NaCl, 10 TEA-Cl (tetraethylammonium-Cl), 10 HEPES, 5 Qx-314, 4 Mg²⁺-ATP, 0.3 Na⁺-GTP, 0.25 EGTA, and 0.05 Alexa Fluor 568 hydrazide sodium salt (Invitrogen), pH 7.25, 270–280 mOsm. For Ca²⁺ imaging experiments EGTA was omitted and replaced with 0.3 mM Fluo-5F pentapotassium salt (Invitrogen). Alexa 568 fluorescence was used to visualize cell bodies, dendrites and spines (Plotkin et al., 2011). Slices were transferred to a submersion-style recording chamber (Scientifica, Uckfield, UK), and electrophysiological recordings were made using a Multiclamp 700B amplifier (Molecular Devices, Jan Jose, CA), digitally sampled at 30 kHz and filtered at 1 KHz.

Following patch rupture, the internal solution was allowed to equilibrate for 10–15 min before imaging and the commencement of experimental procedures. In voltage-clamp mode (V_hold = −70mV), a single 20 ms, 5 mV depolarizing step was evoked to determine passive membrane properties. In current-clamp mode, rheobase was determined from 25 pA increments of depolarizing current injections (500 ms each, from −200 pA). All experiments except for those measuring membrane properties and optogentically evoked up-states were conducted in the presence of 1 μM tetrodotoxin citrate (TTX). mEPSCs were recorded for 120 s while voltage-clamped at −80 mV and mEPSPs were recorded for 360 s in current-clamp mode at V_rest. Access resistance was monitored regularly during recordings and cells were rejected if the resistance changed by more than 15% during the experiment. All recordings were performed at room temperature (21 ± 1°C) unless otherwise noted. Stimulation, acquisition and display were accomplished using Prairie View v5 (Bruker Nano, Inc., Middleton, WI).

Two-photon laser scanning microscopy.

Matrix/striosome compartments and dSPNs were identified by somatic enhanced-GFP and tdTomato two-photon-excited fluorescence using an Ultima Laser Scanning Microscope System (Bruker Nano, Inc., Middleton, WI). A Universal Illumination System with Transmitted Detector (Bruker Nano, Inc., Middleton, WI) was used to provide a bright-field transmission image in registration with the fluorescent images. Green and red signals were acquired using 810 nm pulsed light (90 MHz) excitation (Chameleon Ultra II, Coherent, Inc., Santa Clara, CA), unless otherwise noted. SPNs were patched under bright-field transmission using a high-sensitivity CMOS camera (Thor Labs, Newton, NJ) and a 60x/1.0 N.A. Olympus LUMPFL water-dipping lens.

Ca²⁺ imaging.

Dendritic Ca²⁺ transients were measured using Fluo-5F, as previously described (Shen et al., 2015). SPNs were voltage clamped, and dendritic VGCCs activated by a 300 ms depolarizing somatic voltage step from −80 to 0 mV. Green and red fluorescent line scan signals were acquired from distal dendritic spines at 5.342 ms and 512 pixels per line with 0.0776 μm² pixels and 10 μs pixel dwell time. Ca²⁺ transients were expressed as ΔG/G_Sat to allow inter-cell comparisons (Bloodgood and Sabatini, 2007). Line scans started 500 ms before the stimulation protocol and continued ~1.5 s after the stimulation pulse ended, to obtain the baseline fluorescence and the decay of the optical signal after stimulation.

Two-photon laser uncaging.

2-photon laser uncaging of MNI-glutamate (2PLUG) was achieved using a second Chameleon-Ultra II laser tuned to 720nm, as previously described (Plotkin et al., 2011; Plotkin et al., 2013; Plotkin and Surmeier, 2014). 5 mM MNI-glutamate (Tocris Bioscience, Bristol, UK) dissolved in HEPES-based ACSF (containing (in mM): 124 NaCl, 3 KCl, 2 CaCl₂, 1 MgCl₂, 26 NaHCO₃, 1 NaH₂PO₄, 26 HEPES and 14 glucose) was superfused over the slice using a perfusion syringe pump (World Precision Instruments, Sarasota, FL) and a multibarreled perfusion manifold (Cell MicroControls, Norfolk, VA) at a rate of 0.4ml/h. MNI-glutamate was uncaged using 1 ms pulses of 720 nm light (typically 10–20 mW at the sample plane, controlled by a dedicated Pockels cell modulator).

NMDA/AMPA receptor ratio measurements.

MNI-glutamate was uncaged adjacent to a single dendritic spine (1 ms duration) and uncaging laser power tuned to produce an EPSC of 10–20 pA from a holding potential of −80 mV. The holding potential was then changed to +40 mV and the same spine stimulated again. NMDA/AMPA ratios were calculated as previously described (Plotkin et al., 2014). Briefly, the AMPA receptor-mediated current component was measured as the peak inward current evoked at −80 mV (averaged from 3 trials), and the NMDA receptor-mediated component measured 50 ms after the peak outward current recorded at +40 mV (averaged from 3 trials). Unresponsive spines were excluded from analysis. A total of 3–4 distal spines (>100 μm from soma, measured by a straight line from the stimulation site to the origin of the dendrite at the soma) were measured per neuron.

Induction of dendritic plateau potentials (up-states).

Dendritic plateau potentials were evoked as previously described (Plotkin et al., 2011). Briefly, dSPNs were patch clamped in current clamp mode in the presence of 1 μM TTX (unless otherwise noted), and uncaging laser power was tuned to produce a somatic EPSP of 1–1.5 mV in response to 2PLUG adjacent to a single distal (>100 μm from soma) dendritic spine. Dendritic plateaus (and corresponding somatic up-states) were evoked by stimulating 10–18 neighboring distal dendritic spines in rapid succession (500 Hz), with this stimulation pattern repeated 1–4 times. Up-state duration was quantified as the time between the last uncaging stimulus and the return of the somatic membrane potential to 25% of the peak up-state amplitude. All cells included in the final analyses had calculated up-state durations of at least 70 ms, a duration above the empirically determined “break-point” between subthreshold and supralinear synaptic responses in both striosome and matrix dSPNs (Supplementary Fig. 2A–B). All pharmacological comparisons were performed before and after 10 min bath application.

Optogenetic release of endogenous dopamine.

Acute brain slices were prepared from Nr4a1-eGFP × Drd1a-tdTomato mice that received bilateral injections of AAV5-ChR2-mCherry into the SNc as described above. dSPNs were patch clamped and maintained at ~32 ± 1 °C, and a 473 nm solid state laser was used to locally stimulate nigrostriatal axons along the distal dendritic segment at which a 2PLUG-evoked plateau potential was previously induced. Optogenetic stimulus parameters were based on Yagishita et al. (Yagishita et al., 2014). Transient dopamine release was evoked immediately before and during a plateau potential: 10 sequential 473 nm laser stimuli (1 ms duration, 30 Hz, ~3 mW at sample plane) were delivered along the dendrite segment, the plateau-evoking 2PLUG stimuli initiated 33 ms after the last 473 laser stimulus, and the 10 sequential 473 nm laser stimuli delivered again beginning 33 ms after the final 2PLUG stimulus (Fig. 3E). Experiments were performed in the absence of TTX. Release of endogenous dopamine was confirmed using the genetically encoded dopamine sensor dLight1.1 (Patriarchi et al., 2018). Acute brain slices were prepared from WT mice receiving bilateral injections of AAV5-ChR2-mCherry into the SNc and AAV5-dLight1.1 into the dorsal striatum. Spiral line scans (920 nm, 21.2 ms with 0.0776 μm² pixels and 10 μs pixel dwell) were used to measure changes in green fluorescence in dorsolateral striatum neuropil directly surrounding 10 sequential 473 nm laser stimuli (1 ms duration, 30 Hz, ~3 mW at sample plane). Line scans began 2000 ms before stimuli and continued for 5600 ms after the stimuli ended. A high-speed shutter was used to block signals to the green photomultiplier tube during optogenetic stimulation.

Image acquisition, reconstruction and analysis

At the conclusion of each experiment, whole cell z-series containing the soma and dendrites were acquired with 0.3662 μm² pixels and 10-μs pixel dwell time; ~60–80 images were taken with 1 μm focal steps following each experiment. High magnification z-series of dendrites were acquired with 0.0776 μm² pixels and 10 μs pixel dwell time; ~50–60 images were taken with 0.2 μm focal steps. Z-series were rendered as maximum intensity projections using PicViewer software (John Dempster, Strathclyde University). After recording, the location of the recorded cell was confirmed to be in a striosome or matrix compartment using a low-magnification (4X) objective.

Z-series images were deconvolved using a 60-iteration adaptive blind deconvolution algorithm with AutoQuant X software (Media Cybernetics, Rockville, MD). Neurons were semi-automatically reconstructed from whole cell z-series images using NeuronStudio (RRID:SCR_013798) (Rodriguez et al., 2008). All traces were manually inspected, including verifying that neurites trace back to the soma and reflect correct branching patterns. Sholl analysis was then performed to determine the number of dendritic crosses at concentric circles around the soma (intersections), cumulative dendritic length and branch points (Gertler et al., 2008). Data generated by NeuronStudio were confirmed by repeating Sholl analysis using ImageJ (RRID:SCR_003070). Spine density measurements were semi-automatically calculated from high magnification images of distal dendrites using NeuronStudio (Rodriguez et al., 2008). Dendrites were included in analysis only if blebbing was not present. Whole cell images were used only if cells could be fully visualized and the total dendritic length was greater than 650 μm.

µ-opioid receptor immunohistochemistry

Slices from ACC injected mice were washed once in 1x phosphate-buffered saline (PBS), transferred to 4% paraformaldehyde (PFA) for 4 hours, then to 30% sucrose overnight. Slices were re-sectioned from 275μm to 50μm on a cryostat (Leica Biosystems, ref # CM1950). Free-floating sections were washed in 1x PBS and incubated in blocking solution (3% Normal Goat Serum, 0.1% Triton-X100 in 1x PBS) for 2 hours at room temperature. Sections were then incubated in μ-opioid receptor (MOR) primary antibody (1:2000) (Abcam, ref ab10275, RRID: AB_2156356) overnight at 4°C, and goat anti-rabbit IgG H&L secondary antibody coupled to Alexa Fluor 647 (1:1000) (Abcam, ref 150083, RRID AB_2714032) for 2 hours at room temperature. Sections were counterstained with DAPI (Southern Biotech, ref 0100–20). Slide-mounted sections were imaged using a widefield fluorescence microscope (Olympus VS120 Virtual Slide Microscope, ref VS120-S6-W) and processed using ImageJ.

Computational Modeling

Modeled striosome and matrix dSPNs were based on a biophysically detailed SPN model that we previously published (Dorman et al., 2018). Morphologies were modified using two dSPNs from the Luebke repository on neuromorpho.org (ID: NMO_101636, NMO_101639) (Ascoli et al., 2007; Goodliffe et al., 2018). A neuron with shorter total dendritic length (but otherwise similar morphology) was used for the striosome dSPN as compared to the matrix dSPN, matching experimental observations (see Supplementary Fig. 1I–L). Spines were explicitly modeled as a cylindrical head (0.5 μm diameter, 0.5 μm length) and neck (0.12 μm diameter, 0.5 μm length) (Dorman et al., 2018) and were distributed on dendritic branches (greater than 25 μm from the soma) with a density of 1 spine/μm.

The models were endowed with voltage gated ion channels identified in SPNs (Supplementary Table 1), and Ca²⁺ dynamics were controlled by Ca²⁺ binding proteins, membrane pumps and radial diffusion in dendrites and spines (Dorman et al., 2018). Synaptic channels consisted of NMDAR and AMPAR channels with parameters consistent with Dorman et al. (2018) and conductances set to produce EPSPs close to 1 mV for a single synaptic input. The model was described using a declarative model specification (github.com/neurord/moose_nerp) and simulated using MOOSE with a timestep of 0.01 ms. Simulations were run on the Neuroscience Gateway Portal supercomputer (Sivagnanam et al., 2013).

Model Parameter Optimization.

Channel kinetic equations and parameters are similar to our previously reported model (Dorman et al., 2018), except the half activation voltages and time constants were allowed to be adjusted by the parameter optimization algorithm. Passive parameters (membrane resistivity, membrane capacitivity, and axial resistivity) and channel conductances were determined by fitting the model to hyperpolarizing and depolarizing current injection data from identified striosome and matrix dSPN electrophysiological traces using the automatic parameter optimization software ajustador (Jedrzejewski-Szmek et al., 2018) (http://github.com/neurord/ajustador).

Model parameter set variations.

To identify channel parameters that could modulate the amplitude or duration of up-states, a generic SPN model was used to generate 60,000 parameter sets by randomly varying channel conductance values for all voltage-gated and synaptic channels. Up-states were simulated for each parameter set as described below. Random forest regression was conducted in Python3.6 using the Scikit-Learn package to identify the channels that most strongly predicted either up-state amplitude or up-state duration (Supplementary Fig. 5E–F).

Populations of striosome and matrix dSPN specific models were randomly generated by using the optimized version of each compartment-specific model and randomly modulating channel conductances for the channels previously predicted to affect up-state amplitude and duration (A-type potassium channels, NMDA and AMPA synaptic channels, L-type, R-type, and T-type Ca²⁺ channels, and inward rectifying potassium channels). These parameter sets were each simulated with the set of simulations described below.

Model simulations.

Up-states were simulated by stimulating 14 neighboring distal spines twice each at 500 Hz. Rheobase simulations were conducted at rest or during up-states, using 25pA current steps and a 100 ms duration (the shortened duration was selected to be consistent with upstate durations). IV curves were generated for each parameter set to compute the model optimization fitness. For simulations with dispersed synaptic inputs, 100 inputs were randomly selected throughout the dendritic tree (except for the branch receiving clustered upstate-inducing inputs), and activation of dispersed inputs began 50ms after upstate stimulus onset and lasted for 300 ms. Dispersed input frequencies were varied from 0 to 500 Hz. Simulations for randomly varied parameter sets were included for statistical analysis strictly based on whether their optimization fitness value based on IV curves was sufficiently good (less than 2) and whether the parameter set actually produced an up-state. While the models captured experimental data and cell-type similarities and differences well, one parameter that was not fully captured was a lack of difference in plateau duration (Supplementary Fig. 5D). This likely reflects an unknown channel(s) not represented in the models, or randomly varying parameter values in the same ranges differentially affected plateau duration between the models. Comparisons that could be affected by this difference in duration were avoided; measures of spiking were therefore expressed as frequency or change in frequency.

QUANTIFICATION AND STATISTICAL ANALYSIS

Electrophysiological recordings and Ca²⁺ transients were analyzed using Igor Pro (WaveMetrics, Inc., Lake Oswego, OR; RRID:SCR_000325). Miniature EPSCs and EPSPs were analyzed (from 120 s or 360 s traces, rejected if holding current changed >20%) using the Mini Analysis program (Synaptosoft, Inc., Fort Lee, NJ; RRID:SCR_002184). Population data are reported as medians and presented as box and whisker plots, where boxes outline upper and lower quartiles and whiskers are upper and lower 10% values. Statistical differences were examined using 2-sided Mann-Whitney U, Wilcoxin signed-rank (for paired comparisons) or Kruskal Wallis (for multiple comparisons; followed by Mann-Whitney U) non-parametric tests of significance; normality was not assumed for comparisons. The Kolmogorov-Smirnov test was used to compare cumulative probabilities of mEPSCs. Because no obvious sex differences were observed (data available upon request), data from male and female mice were pooled. The ROUT method (Q = 1%) in GraphPad Prism was used to identify potential outliers (Motulsky and Brown, 2006). Data were analyzed prior to and again after exclusion criteria were enforced. Computational modeling analyses was performed in Python 3.6, using the Numpy, SciPy, Pandas, Matplotlib, and Scikit-Learn python packages. Random forest regression was performed using Scikit-Learn to identify channels associated with up-state duration and amplitude. Model simulation and analysis files are available in ModelDB (http://modeldb.yale.edu/266807). Statistical analysis of simulations from populations of striosome and matrix models followed the same statistical approach for experimental data unless otherwise described. Differences were considered statistically significant if p < 0.05. All statistical analyses were conducted using GraphPad Prism v7.0e (GraphPad Software, La Jolla, CA; RRID:SCR_002798) and for Supplementary Fig 4C Matlab R2020a (MathWorks, Natick, MA); figures were generated using Adobe Illustrator (San Jose, CA; RRID: SCR_010279).

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-mu opioid receptor antibody	Abcam	RRID: AB_2156356
Goat anti-rabbit IgG H&L (Alexa Fluor® 647)	Abcam	RRID: AB_2714032
DAPI	Southern Biotech	0100–20
Bacterial and Virus Strains
pAAV5.CAG.hChR2(H134R)-mCherry.WPRE.SV40	Gift from Karl Deisseroth	Addgene plasmid # 100054-AAV5; http://n2t.net/addgene:100054 RRID Addgene_100054
pAAV5.CAG.dLight1.1	Gift from Lin Tian Patriarchi et al., 2018	Addgene plasmid #: 111067-AAV5; http://n2t.net/addgene:111067 RRID: Addgene_111067
AAV9-hSyn-hChR2(H134R)-EYFP	Gift from Karl Deisseroth	Addgene plasmid # 26973-AAV9; Http://n2t.net/addgene:26973 RRID: Addgene_26973
Experimental Models: Organisms/Strains
Mouse: Tg(Drd1a-tdTomato)6Calak	Jackson Laboratories	RRID:IMSR_JAX:016204
Mouse: Tg(Nr4a1-eGFP139Gsat)	GENSAT; Davis & Puhl, 2011	Donated by Dr. Margaret Davis
Mouse: C57BL/6J	Jackson Laboratories	RRID:IMSR_JAX:000664
Software and Algorithms
adjustador	Jȩdrzejewski-Szmek et al. 2018	https://github.com/neurord/ajustador
Adobe Illustrator	Adobe	RRID: SCR_010279
AutoQuant X	Media Cybernetics	RRID: SCR_002465
GraphPad Prism V7.0e	Graphpad Software	RRID: SCR_002798
Igor Pro	WaveMetrics, Inc.	RRID: SCR_000325
ImageJ	National Institute of Health	RRID: SCR_003070
Matlab R2020a	MathWorks	RRID: SCR_001622
Mini Analysis	Synaptosoft	RRID: SCR_002184
MOOSE	Upinder Bhalla, NCBS	https://moose.ncbs.res.in/
Moose_nerp	Kim Blackwell, George Mason Univ	https://github.com/neurord/moose_nerp
NeuronStudio	Rodriguez et al., 2008	RRID: SCR_013798
Neuroscience Gateway Portal	Sivagnanam et al., 2013	RRID:SCR_008915
PicViewer V2.5.2	John Dempster, Strathclyde University	http://spider.science.strath.ac.uk/sipbs/software_imaging.htm
Prairie View v5	Bruker Nano, Inc	RRID: SCR_017142
Python 3.6	Python Software Foundation	RRID:SCR_008394

Highlights.

• Both striosome and matrix dSPNs support dendritically-evoked somatic “up-states”

• Dopamine oppositely modulates up-state length in striosome vs matrix dSPNs via D1Rs

• Compartment-specific responses to D1R activation involve L-type Ca²⁺ channels

• Changes in striatal dopamine may shift the balance of striosome vs matrix output