Patch and matrix striatonigral neurons differentially regulate locomotion

The striatonigral neurons are known to promote locomotion1,2. These neurons reside in both the patch (also known as striosome) and matrix compartments of the dorsal striatum3–5. However, the specific contribution of patch and matrix striatonigral neurons to locomotion remain largely unexplored. Using molecular identifier Kringle-Containing Protein Marking the Eye and the Nose (Kremen1) and Calbidin (Calb1)6, we showed in mouse models that patch and matrix striatonigral neurons exert opposite influence on locomotion. While a reduction in neuronal activity in matrix striatonigral neurons precedes the cessation of locomotion, fiber photometry recording during self-paced movement revealed an unexpected increase of patch striatonigral neuron activity, indicating an inhibitory function. Indeed, optogenetic activation of patch striatonigral neurons suppressed locomotion, contrasting with the locomotion-promoting effect of matrix striatonigral neurons. Consistently, patch striatonigral neuron activation markedly inhibited dopamine release, whereas matrix striatonigral neuron activation initially promoted dopamine release. Moreover, the genetic deletion of inhibitory GABA-B receptor Gabbr1 in Aldehyde dehydrogenase 1A1-positive (ALDH1A1+) nigrostriatal dopaminergic neurons (DANs) completely abolished the locomotion-suppressing effect caused by activating patch striatonigral neurons. Together, our findings unravel a compartment-specific mechanism governing locomotion in the dorsal striatum, where patch striatonigral neurons suppress locomotion by inhibiting the activity of ALDH1A1+ nigrostriatal DANs.


Abstract
The striatonigral neurons are known to promote locomotion 1,2 .These neurons reside in both the patch (also known as striosome) and matrix compartments of the dorsal striatum [3][4][5] .However, the speci c contribution of patch and matrix striatonigral neurons to locomotion remain largely unexplored.Using molecular identi er Kringle-Containing Protein Marking the Eye and the Nose (Kremen1) and Calbidin (Calb1) 6 , we showed in mouse models that patch and matrix striatonigral neurons exert opposite in uence on locomotion.While a reduction in neuronal activity in matrix striatonigral neurons precedes the cessation of locomotion, ber photometry recording during self-paced movement revealed an unexpected increase of patch striatonigral neuron activity, indicating an inhibitory function.Indeed, optogenetic activation of patch striatonigral neurons suppressed locomotion, contrasting with the locomotion-promoting effect of matrix striatonigral neurons.Consistently, patch striatonigral neuron activation markedly inhibited dopamine release, whereas matrix striatonigral neuron activation initially promoted dopamine release.Moreover, the genetic deletion of inhibitory GABA-B receptor Gabbr1 in Aldehyde dehydrogenase 1A1-positive (ALDH1A1 + ) nigrostriatal dopaminergic neurons (DANs) completely abolished the locomotion-suppressing effect caused by activating patch striatonigral neurons.Together, our ndings unravel a compartment-speci c mechanism governing locomotion in the dorsal striatum, where patch striatonigral neurons suppress locomotion by inhibiting the activity of ALDH1A1 + nigrostriatal DANs.

Main
Striatal spiny projection neurons (SPNs) constitute approximately 95% of the neuronal population in the dorsal striatum and play a key role in motor learning, decision-making, and regulating voluntary movement 1,2 .These SPNs can be broadly classi ed into two main subtypes: the dopamine receptor D1 (Drd1)-expressing direct-pathway neurons (dSPNs), which project directly to the globus pallidus internus (GPi) and substantia nigra pars reticulata (SNr); and the dopamine receptor D2 (Drd2)-expressing indirect-pathway SPNs (iSPNs), which project to the globus pallidus externus (GPe) 1,2 .The dSPNs primarily project to the SNr and these SNr-projecting dSPNs are also referred as striatonigral neurons.It has been well recognized that the dSPNs facilitate movement initiation and the iSPNs restrain unwanted movements 2 .Additionally, the SPNs can be further subdivided into two complementary compartments within the dorsal striatum, namely the patch and matrix compartments, identi ed by distinct neurochemical markers and input-output connectivity [3][4][5]7 . Th patch compartment is typically marked by the expression of m-opiate receptor (MOR1) in rodents 3,8 , while the matrix compartment is labeled by the CALB1 expression 6 .Each compartment contains both dSPN and iSPNs.Advances in bulk and single-cell transcriptomics provide additional genetic markers for dSPNs and iSPNs as well as patch and matrix SPNs 9,10 .Previous research has also underscored the signi cance of patch SPNs in modulating mood, decision-making, and reward processing, while matrix SPNs are implicated in action selection 11 .
However, the speci c roles of patch and matrix dSPNs in locomotion control remain to be determined.
Both patch and matrix dSPNs innervate the aldehyde dehydrogenase 1A1-positive (ALDH1A1 + ) dopaminergic neurons (DANs) located in the ventral tier of substantia nigra pars compacta (SNc) 12 .ALDH1A1 + DANs are particularly vulnerable to degeneration in Parkinson's disease 13 .Notably, a portion of patch dSPN axon terminals bundle up to form distinctive striosome-dendron bouquet structures, intertwining with the dendrites of ALDH1A1 + DANs that extend into the SNr, in contrast to the matrix dSPNs 14,15 .This anatomical arrangement likely underlies the more potent presynaptic inhibition of DANs by patch dSPNs in contrast to matrix dSPNs 16 .Moreover, patch dSPNs exert a prolonged inhibitory in uence on DAN neuronal activity via GABA-B receptors, crucially regulating the transition from tonic to burst ring in ALDH1A1 + nigrostriatal DANs, as evidenced by brain slice recordings 14,17 .Despite these insights, the physiological and behavioral implications of this distinctive striatonigral circuit remain to be explored.
One major challenge to study patch SPNs is the di culty in de nitively characterizing and manipulating these neurons, largely due to their less well-de ned neurochemical organizations.In this study, we rst identi ed the Kremen1 gene as a speci c molecular maker for patch SPNs in the dorsal striatum.
Subsequently, we generated a line of Kremen1 2A-Cre knock-in (KI) mice to investigate the distribution of Kremen1-positive (Kremen1 + ) patch dSPN subpopulations and their functional signi cance in locomotor control compared to the Calb1-positive (Calb1 + ) matrix dSPNs.

Kremen1 transcript is enriched in patch compartments
Using patch reporter Nr4a1-eGFP transgenic mice 18 , we isolated the enhanced green uorescent protein (EGFP)-containing and the adjacent tissues in the dorsal striatum by laser capture microdissection for bulk RNA sequencing (NCBI accession: PRJNA870469).In comparison to various patch markers examined previously, such as mu-type Opioid Receptor (Oprm1), Teashirt Zinc Finger Family Member 1 (Tshz1), Prodynorphin (Pdyn)and Selenoprotein W (Sepw1), the expression of Kremen1 mRNA exhibited substantially higher differentials in the patch versus matrix compartments (Fig. 1a, b).
Consistently, Kremen1 was also identi ed as a patch marker by an independent single-nuclei transcriptomic study 10 .Further co-localization analyses with Drd1 and Drd2 using RNAscope in situ hybridization found that approximately 60% of Kremen1 + SPNs corresponded to dSPNs, while 40% were identi ed as iSPNs, resembling the ratio of total dSPNs versus iSPNs in the dorsal striatum (Extended data Fig. 1a-c).Moreover, Kremen1 + SPNs were predominantly distributed in the dorsal striatum (Fig. 1d, Extended data Fig. 1d).Within the striatum, Kremen1 + SPNs accounted for 12.7% of total SPNs, 52% of SPNs within the patch compartments, and 8% of SPNs in the matrix compartments (Extended data Fig. 1e).Notably, the density of Kremen1 + SPNs was substantially higher in patch compartments compared to matrix compartments (Fig. 1c).Together, these ndings identify Kremen1 as a promising genetic marker for distinguishing patch-matrix compartments.
Tshz1 is also enriched in patch compartments 9,10 .We further compared the distribution of Kremen1 + and Tshz1 + dSPNs and total SPNs in the dorsal striatum by RNAScope in situ hybridization and found only a 7-8% overlap between these two dSPN subpopulations (Extended Date Fig. 2).These results suggest that Kremen1 + and Tshz1 + mark two largely different subtypes of patch SPNs in the dorsal striatum.
Kremen1 2A-Cre knock-in mice are useful for studying patch SPNs Given the enrichment of Kremen1 in patch SPNs, we generated a line of Kremen1 2A-Cre knock-in (KI) mice using CRISPR/Cas9-mediated gene editing (Extended data Fig. 3).By crossing these mice with Ai14 reporter mice, which express the red uorescent protein tdTomato under Cre-dependent regulation, we observed distinct tdTomato signals scattered within the dorsal but not ventral striatum of Kremen1 2A- Cre ;Ai14 mice (Fig. 1d).These signals co-localized with the widely used patch marker MOR1 (Fig. 1d, e).
Apart from patch SPNs, tdTomato signals were also detected in pericytes, with no staining detected in the interneurons or glial cells of the dorsal striatum (Extended data Fig. 4a, b).The Kremen1 + patchy structures were restricted to the dorsal striatum but distributed throughout the entire dorsal striatum along the rostral to caudal and medial to lateral axes (Extended data Fig. 5a, b).Beyond the dorsal striatum, tdTomato signals were also visible in the hippocampal regions (Extended data Fig. 5b, c).
To investigate the projection pattern of Kremen1 + SPNs, we employed stereotaxic injection of an adenoassociated viral vector (AAV) co-expressing Cre-dependent tdTomato and synaptophysin-fused EGFP (sypEGFP) into the dorsal striatum of Kremen1 2A-Cre KI mice (Fig. 1f).Both tdTomato and sypEGFP signals were found in the patch compartments (Fig. 1g).Furthermore, sypEGFP speci cally marked the axon terminals of patch SPNs in the GPe and entopeduncular nucleus (EPN), a mouse equivalent to GPi (Fig. 1h), as well as in the SNr and SNc (Fig. 1i).Notably, the incoming tdTomato-positive patch dSPN axons exhibited distinct dendron-bouquet like structures 15 , forming connections with the dendrites of DANs in SNr (Fig. 1i).This striosome-dendron bouquet structure was also observed in the SN of Kremen1 2A-Cre ;Ai14 mice (Extended data Fig. 5d).In summary, Kremen1 2A-Cre KI mice may serve as a useful tool for investigating the neuroanatomy and functions of patch SPNs.Therefore, Kremen1 + SPNs will be referred to as patch SPNs hereafter.

Increased patch dSPN activity precedes the cessation of locomotion
To monitor the activity of patch and matrix dSPNs during self-paced locomotion, we injected AAVs expressing the genetically encoded calcium indicator GCaMP8s (AAV1-FLEX-GCaMP8s) into the dorsal striatum and implanted an optic ber in the SNr of the same hemisphere in Kremen1 2A-Cre or Calb1 IRES2- Cre mice 19 (Fig. 2a, b).The mice were head-xed and allowed to walk freely on a belt treadmill, where movement velocity signal was synchronized with a ber photometry setup for recording calcium transients in the axon terminals of patch or matrix dSPNs simultaneously (Fig. 2c-e).As shown in representative sample recording traces from a Kremen1 2A-Cre and a Calb1 IRES2-Cre mouse respectively, both patch and matrix dSPN axon terminal calcium transients covaried with velocity changes during locomotion bouts (Fig. 2f, g).However, when aligning their mean activity with locomotion onset and offset, it revealed a difference between patch and matrix dSPNs.In the same sample mice, both patch and matrix dSPN activity increased following the onset of locomotion, but there was an unexpected rise of patch dSPN activity preceding the locomotion cessation and continued to rise until shortly after animal stopped moving, in contrast, the activity of matrix dSPN decreased preceding the locomotion cessation (Fig. 2h, j).This pattern was consistent with additional analyses involving more animals (Fig. 2i, k).The timing of peak activity during locomotion onset did not signi cantly differ between patch and matrix dSPNs (p = 0.08, two-tailed unpaired t test, Fig. 2l).However, the timing of peak activity during locomotion offset was notably delayed in patch dSPNs compared to matrix dSPNs (p < 0.0001, Fig. 2m), and in fact patch dSPN activity peaked after locomotion offset (p = 0.04, one-tailed one-sample t test).Additionally, the slope of activity preceding locomotion offset differed signi cantly between patch and matrix dSPNs (p < 0.001, Fig. 2n), with patch dSPNs exhibiting increased (positive) and matrix dSPNs showing decreased (negative) activity, suggesting their differential regulation of motor activity.

Patch and matrix dSPNs exert contrasting roles in locomotor control
To examine the in uence of patch dSPN activity on locomotion, we used optogenetic manipulations by introducing AAV vectors containing Cre-dependent optogenetic activator Channelrhodopsin-2 (AAV1-FLEX-ChR2) into the dorsal striatum of Kremen1 2A-Cre mice and implanted the optic ber in the SNr for light stimulation of dSPN axon terminals (Fig. 3a, b).Optogenetic activation at 3mW/20Hz for 3min led to a notable reduction in locomotion velocity, followed by a subsequent recovery in motor activity after the stimulation period (Fig. 3c).The decrease in walking speed likely resulted from both a reduction in velocity during each bout of ambulatory movement (Fig. 3d) and reduced locomotor activity, as evidenced by both reduced movement frequency and shortened duration (Fig. 3e, f), as well as an increase in immobility frequency (Fig. 3g).This reduction in locomotion velocity was also observed at various stimulation parameters, including continuous 0.25mW power intensities, 10Hz stimulation frequency, and 10s durations (Extended data Fig. 6a-c).In contrast, optogenetic activation of matrix dSPNs of Calb1 IRES2-Cre mice yielded opposite outcomes, resulting in a substantial increase in movement velocity, coupled with an increase of ambulatory speed and duration, as well as a decrease in immobility frequency and duration (Fig. 3h-n, Extended data Fig. 6d-f).Consequently, our ndings unravel the opposing roles of patch and matrix dSPNs in the regulation of ambulatory movement.
To investigate the role of patch iSPNs in the context of locomotion, we introduced AAV vectors containing optogenetic activators that were both Cre and Flp-dependent (AAV1-ConFon-ChR2) into the dorsal striatum of Kremen1 2A-Cre ;A2a Flp double KI mice (Extended data Fig. 7a).Light stimulation was applied at the axon terminals of iSPNs situated in the GPe (Extended data Fig. 7b, c).The high-power stimulation at 3mW/20Hz for 3min resulted in a modest increase in average velocity and duration per bout of movement (Extended data Fig. 7d-h), while at low light power level with 0.25mW constant stimulation we did not observe any notable alteration in locomotion velocity (Extended data Fig. 7i).These results underscore the limited role of patch iSPNs in regulating locomotion.Indeed, chemogenetic activation of both patch dSPNs and iSPNs in Kremen1 2A-Cre mice led to reduced locomotor activity (Extended data Fig. 8), supporting a major role of patch dSPNs in suppressing locomotion.

Patch and matrix dSPNs differently regulate dopamine release
To explore how patch and matrix dSPNs distinctively regulate locomotion at the circuit level, we examined the impact of dSPN neuronal activity on dopamine release in the dorsal striatum.Both patch and matrix dSPNs innervate nigrostriatal DANs and regulate their activity 12,14,16,20 .Using the genetically encoded dopamine sensor GRAB DA3m 21 and Cre-dependent ChR2, we assessed dopamine release in the dorsal striatum of Kremen1 2A-Cre (Fig. 4a, b) and Calb1 IRES2-Cre (Fig. 4a, c) mice through ber photometry following optogenetic stimulation at the SNr for each mouse line.Activation of patch dSPNs led to a gradual reduction in dopamine levels in a stimulation dose dependent manner, with stimulation frequencies of 10 Hz and 20 Hz, and durations of 2, 5, and 15 seconds (Fig. 4d).In contrast, activation of matrix dSPNs induced a biphasic change in dopamine release: an initial transient increase in dopamine signals at the onset of optic stimulation, followed by a subsequent reduction during the stimulation (Fig. 4e).The initial surge of dopamine release following matrix dSPN activation was present across different stimulation frequencies and durations (Fig. 4f).However, the magnitude of dopamine release reduction was far less pronounced with matrix dSPNs stimulation compared to patch dSPNs (Fig. 4g).Additionally, only patch dSPN stimulation resulted in a post-stimulation rebound of dopamine release (Fig. 4h), consistent with a previous observation of DANs rebound ring following patch stimulation in brain slice 14 .These ndings suggest that patch and matrix dSPNs regulate dopamine release differently, with patch dSPN activation leading to a more potent inhibition of dopamine release.
Patch dSPNs suppress locomotion through the GABA-B receptors in ALDH1A1 + DANs We next set to determine whether patch dSPNs affected locomotion through inhibiting nigrostriatal DANs.Considering that ALDH1A1 + nigrostriatal DANs receive prolonged inhibitory inputs mediated by GABA-B receptors from patch dSPNs 14,17 , we genetically deleted the receptor-encoding Gabbr1 gene in ALDH1A1 + nigrostriatal DANs through introduction of Cre-dependent CRISPR/saCas9-Gabbr1sgRNA gene targeting AAVs into the SNc of Aldh1a1 CreERT2 mice 12 (Extended Data Fig. 9a, b).The resulting knockdown (KD) of Gabbr1 in ALDH1A1 + nigrostriatal DANs led to increased ambulatory velocity (Extended Data Fig. 9c-e).These ndings support an important role of GABA-B receptor-mediated inhibitory signaling in ALDH1A1 + DANs in locomotor control.
To investigate the role of postsynaptic GABA-B receptors in patch dSPN-induced locomotor inhibition, we selectively activated the patch dSPNs through optogenetics while simultaneously downregulating Gabbr1 in ALDH1A1 + nigrostriatal DANs of Kremen1 2A-Cre ;Aldh1a1 CreERT2 double KI mice (Fig. 5a, b).Since the Gabbr1-KD and control mice tended to have different baseline locomotion, to compare between groups, we computed a velocity ratio from baseline for individual mice.Remarkably, the genetic KD of Gabbr1 in the ALDH1A1 + nigrostriatal DANs completely abolished the locomotionsuppressing effects induced by patch dSPN activation (Fig. 5c, d).Correspondingly, control mice showed more reduction in velocity compared to Gabbr1-KD mice during light stimulation (Fig. 5e).To further validate the involvement of GABA-B receptors in locomotion inhibition, we stimulated both patch and matrix dSPNs in Kremen1 2A-Cre ;Aldh1a1 CreERT2 double KI mice under the control and Gabbr1-KD conditions (Fig. 5f, g).While the dSPN stimulation markedly increased locomotion in both groups of mice, the absence of GABA-B receptors in ALDH1A1 + nigrostriatal DANs seemed to further augment dSPN-induced ambulatory movement (Fig. 5h, i), although this effect did not reach statistical signi cance (Fig. 5j).Together, our ndings demonstrate that patch dSPNs negatively modulate ambulatory locomotor behavior by regulating ALDH1A1 + nigrostriatal DAN activity through GABA-B receptors.

Discussion
In this study we challenge the canonical model suggesting that striatonigral dSPNs promote locomotion and demonstrate that this model perhaps only applies to the matrix striatonigral dSPNs, while the patch striatonigral dSPNs play an opposing role by suppressing locomotion.This inhibitory function of patch dSPNs is implicated in the rise of neuronal activity during the offset of self-paced movement and is causally validated by optogenetic activation.Furthermore, we showed the critical involvement of GABA-B receptors in ALDH1A1 + nigrostriatal DANs during patch dSPN-induced locomotion inhibition.
The dSPNs and iSPNs are subdivided into patch and matrix compartments based on neurochemical markers and their input-output connectivity; 7 however, it remains challenging to ambiguously de ne the patch compartments.Conventional patch markers, including MOR1, substance P and enkephalin, can only identify a portion of patch compartments along the rostral to caudal axis of the striatum 22 .
Speci cally, MOR1 marks the patch in the rostral and intermediate striatum, while substance P and enkephalin label the patch in the intermediate and caudal striatum 22 .In contrast, Kremen1 marks the patch structures within the entire dorsal striatum, including the dorsolateral striatum, a region heavily involved in motor control and learning 23 .Additionally, Kremen1 marks both Drd1-and Drd2-expressing SPNs in the same ratio as the general SPN population, while MOR1 tends to label more Drd1-expressing dSPNs 24 .In addition to MOR1, Pdyn and Tshz1 are also preferentially mark the patch dSPNs 9,10 .
Notably, there is less than 3% overlap between the Pdyn + and Tshz1 + Drd1-expressing dSPNs 25 , suggesting heterogeneity among patch SPNs based on different gene expression pro les.Furthermore, substantially more Pdyn + and Tshz1 + SPNs are distributed in the matrix compartments compared to the Kremen1 + SPNs.Therefore, Kremen1 serves as a useful patch marker, which unbiasedly labels both dSPNs and iSPNs selectively in the patch compartments.
The Pdyn-positive (Pdyn + ) dSPNs were suggested to promote locomotion, while the Tshz1positive(Tshz1 + ) ones were claimed to suppress locomotion in a place-preference test 25 .Since patch dSPNs have been implicated in discriminating valence during decision-making, changes in movement velocity by patch dSPN activation in valence context (such as place preference test) may re ect alterations in active avoidance or approach choices rather than locomotion itself 11 .Moreover, the roles of SNr-projecting dSPNs were not speci cally examined in the aforementioned study 25 , considering GPi-projecting dSPNs may originate from a distinct population 26 .As Tshz1 + patch dSPNs have also been associated with locomotion suppression 25 , we conducted a comparison of Kremen1 + and Tshz1 + Drd1expressing neurons in the dorsal striatum.Our ndings reveal that Kremen1 and Tshz1 mark two largely different subtypes of patch dSPNs in the dorsal striatum.While we demonstrate that SNrprojecting Kremen1 + dSPNs inhibit ambulatory locomotion, whether SNr-projecting Tshz1 + SPNs serve a similar function remains unclear 25 .Given the signi cant diversity among patch SPNs based on molecular markers and connectivity, future studies should explicitly identify subtypes using genetic markers and projection targets.
Both patch and matrix dSPN activity showed a similar transient increase at the onset of locomotion, but started to diverge as movement ceased, with matrix dSPN exhibiting decreased activity while patch dSPN transiently increased activity.The activity pattern of patch and matrix dSPNs during locomotion mirrors previous ndings of synergistic concurrent activation of opposing dSPNs and iSPNs during movement initiation 24 with their activity decorrelating as movement progresses 27 .Furthermore, manipulating these two populations of dSPNs causally produced opposing locomotor effects.These ndings highlight the presence of opposing motifs within the dSPN population, one originating from the patch and the other from matrix, akin to the dichotomous organization of dSPN and iSPN in the broader dorsal striatum.Our results suggest that patch dSPN activity may facilitate state transition from movement to quiescent.In support of this hypothesis, patch dSPN activity began to ramp hundreds of milliseconds before movement termination, and optogenetic activation of patch dSPNs shortened movement bout duration while increasing instance of immobility.State transition regulation has been identi ed as a fundamental principle of basal ganglia motor control 28 .In addition to receive sensorimotor inputs, patch dSPNs receive more limbic inputs than matrix dSPNs 29 and have robust direct connections with midbrain DANs, suggesting their role in integrating external sensorimotor information with implicit motivation to regulate motor state transitions.Conversely, we found that patch iSPNs exerted a positive, albeit less pronounced, effect on ambulatory movement upon activation compared to patch dSPNs.This aligns with previous ndings suggesting that striatal iSPNs apply inhibitory motor control through their GABAergic axon collaterals to surrounding dSPNs, while the majority of striatopallidal inputs drive negative reinforcement learning 30 .The synaptic strength of iSPN-dSPN connections within the patch remains unclear.Given the relatively smaller number of patch iSPNs, their collaterals within the patch may not be as robust.Further exploration is warranted to elucidate the function of patch iSPNs.ALDH1A1 + nigrostriatal DANs receive the most monosynaptic inputs from the dorsal striatum 12 .Furthermore, these neurons appear to form reciprocal innervation with dSPNs in the dorsal regions of dorsal striatum 12 .This reciprocal connection between ALDH1A1 + DANs and dSPNs may constitute a feedback loop for timely regulation of the dopamine release and SPN activity in motor control and learning.While both patch and matrix dSPNs innervate ALDH1A1 + DANs 12 , the axon terminals of patch dSPNs are the primary presynaptic components in this so-called striosome-dendron bouquet structure with the dendrites of ALDH1A1 + nigrostriatal DANs 14,15 .This arrangement potentially plays a pivotal role in regulating dopamine release in both dorsal striatum and SNr.Accordingly, brain slice recordings indicate that patch dSPNs deliver stronger inhibitory inputs to the DANs and could induce rebound ring in ALDH1A1 + nigrostriatal DANs compared to matrix dSPNs and the other inhibitory inputs 14,16 .Consistent with these in vitro studies, our in vivo recordings reveal a more pronounced inhibition of dopamine release upon stimulating patch dSPNs compared to the matrix dSPNs, with a rebound dopamine release observed upon the cessation of patch dSPN stimulation.Supporting the functional signi cance of the connection between patch dSPNs and ALDH1A1 + nigrostriatal DANs, genetic deletion of GABA-B receptors in ALDH1A1 + nigrostriatal DANs completely abolished the locomotion-suppressing effects mediated by patch dSPNs.This synaptic coupling might dynamically regulate the dopamine supply in both the dorsal striatum and SNr, thereby modulating various motor activities.
Notably, the activation of matrix dSPNs initially triggered a transient increase in dopamine release before a subsequent reduction occurred.Considering that matrix dSPNs project to inhibitory parvalbumin neurons (PVNs) in the SNr and PVNs provide direct and tonic inhibitory inputs to DANs [31][32][33][34] , it's plausible that matrix dSPNs initially disinhibit DANs via PVNs, resulting in an early increase in DAN activity and dopamine release.The prevalence of this transient increase of dopamine release within the rst two seconds of matrix dSPN stimulation suggests that the primary effect of matrix dSPN activation on DANs is disinhibition or excitation.Therefore, the distinct modulation of DAN activity by patch and matrix dSPNs may underlie their differential in uence on motor control.

Mouse work
All mouse studies were conducted in accordance with the guidelines approved by the Institutional Animal Care and Use Committees (IACUC) of the Intramural Research Program of the National Institute on Aging (NIA), NIH.All mouse lines were maintained as heterozygotes in a C57BL/6J background.The Kremen1 2A-Cre KI mice were generated by Shanghai Model Organisms Inc. (Shanghai, China).The Aldh1a1 CreERT2 KI mice were generated as previously described 12  The Kremen1 2A-Cre KI mice were generated using the CRIPSR/Cas9 approach in the C57BL/6J strain by Shanghai Model Organisms Inc.A donor plasmid containing the 2A ribosome skipping sequence, Cre DNA recombinase sequence, and anking Kremen1 mouse genomic DNA sequence was constructed.This was used to insert to the 2A-Cre DNA fragment into exon 9 immediately after the stop codon of Kremen1 gene locus, guided by the gRNA with sequence GTGGGCTTCAGTCACTCACG AGG.One founder mouse was generated, and the correct genomic modi cation was con rmed by sequencing.
Laser capture microdissection and RNA sequencing One-month-old Nr4a1-eGFP transgenic mice were euthanized with CO 2 followed by rapid decapitation.
The brains were immediately dissected and preserved at −80 °C.Cryosectioning of the frozen brains was performed at −20 °C, and the sections were mounted onto PAN membrane frame slides (Applied Biosystems, Foster City, CA).The dorsal striatum was determined based on anatomic landmarks such as the corpus callosum, lateral ventricle and nucleus accumbens.Using the ArturusXT microdissection system with uorescent illumination (Applied Biosystems), the eGFP-positive island-like structures within the dorsal striatum of Nr4a1-eGFP transgenic mice were carefully isolated onto LCM Macro Caps (Applied Biosystems) and designated as "patch", while surrounding tissues of similar size were also isolated onto Macro Caps and designated as "matrix".
Total RNA extraction and puri cation were executed from hundreds of caps using the PicoPure Isolation kit (Applied Biosystems), with subsequent genomic DNA clearance facilitated by RNase-free DNase (Qiagen) following the manufacturer's protocols.Quanti cation of RNA was performed using a NanoDrop spectrophotometer (ThermoFisher), and RNA integrity was evaluated using the Bioanalyzer RNA 6000 pico assay (Agilent).Only RNA samples exhibiting high integrity were selected for subsequent patch and matrix RNAseq library preparation.cDNA libraries were prepared from the puri ed RNA using the TruSeq Stranded Total RNA LT library preparation kit (Illumina) following the manufacturer's protocol.The quality of the libraries was assessed using the Bioanalyzer DNA 1000 assay (Agilent) before sequencing on an Illumina HiSeq 2000 platform.Fastq les were generated using the standard Illumina pipeline.Transcript abundance, annotated by Ensembl, was quanti ed using Salmon in a non-alignment-based mode, and gene-level counts were estimated utilizing the Tximport package (Bioconductor).Normalization of counts and subsequent data analysis were conducted following previously established procedures 35 .The accession number of the striatal tissue RNA-seq is PRJNA870469.

RNA in situ hybridization and image analysis
RNAin situ hybridization (ACDBio, RNAscope) was used to detect the expression of Drd1, Drd2, Kremen1 and Tshz1 mRNAs in the dorsal striatum of adult C57BL/6J mice.For tissue preparation, mice were anesthetized with CO 2 and rapidly decapitated.The brains were fresh-frozen on dry ice and stored at -80 °C before sectioning.Striatal sections (12µm) were collected using a cryostat (Leica Biosystems) and stored at -80 °C until processed.
RNAscope images were analyzed using Imaris (v10.0.0,Bitplane, Belfast Northern Ireland, UK).Surfaces for the dorsal and ventral striatum were created using the Allen Brain Atlas as a reference.The spatial patch was de ned by outlining Kremen1 + SPN clusters (a minimum of 5 SPNs), and the density of Kremen1 + SPNs within a spatial patch is at least 200 cells/mm 2 .Surfaces for individual channels (i.e., Drd1, Drd2, Kremen1, Tshz1 and DAPI) were created with unique parameters, which were saved and applied to all images within the same batch.Minor adjustments were made in subsequent rounds of RNAscope to improve quanti cation accuracy.DAPI surfaces in the striatum were ltered by the mean intensity of Drd1 (dSPNs) and the mean intensity of Drd2 (iSPNs).These Drd1 + and Drd2 + cells were further ltered by the overlapped volume ratio to Kremen1 to quantify patch dSPNs and patch iSPNs, respectively.RNAscope analysis with Tshz1 was conducted in a similar manner, ltering Drd1 + and Drd2 + cells by the overlapped volume ratio to the Tshz1 surface.All data points are presented as the average across striatal sections of both hemisphere for each mouse.

Immunohistochemistry
Mice were anesthetized with pentobarbital and transcardially perfused with precooled PBS, followed by 4% paraformaldehyde (PFA) solution as described previously 12,36 .Brains were stored in 4% PFA at 4  °C overnight and then transferred to 30% PBS buffered sucrose solution for at least 2 days before sectioning.Series of coronal sections (40µm) were collected using a cryostat and stored at 4°C in PBS.

Construction of pAAV-FLEX-SaCas9-U6-sgGabbr1
The reference sequence of Gabbr1 gene was retrieved from the UCSC genome browser database (http://genome.ucsc.edu/),and subsequent identi cation and alignment of exons were performed using the Mouse Genome Informatics (MGI) database.The most 5′ common coding exons were then selected, and the sequence was uploaded to the CRISPOR website (http://crispor.org)for identifying potential sgRNAs and PAM sequences.Three Gabbr1 sgRNAs were synthesized as short oligos (Euro ns Genomics) with a 5′ CACC-3′ overhang on the forward primer, and a 5′ -AAAC 3′ overhang on the reverse primer, facilitating seamless integration into the pX601-AAV-CMV;NLS-SaCas9-NLS-3xHA-bGHpA;U6;BsaI-sgRNA vector, a gift provided by Dr. Feng Zhang (Addgene plasmid # 61591).Sanger sequencing con rmed the insertion of the Gabbr1 sgRNA using the primer TAACCACGTGAGGGCCTATTTC.

Stereotaxic injection and optic ber implantation
The stereotaxic survival surgery was performed as previously described 35 .All surgery were conducted under aseptic conditions, and body temperature was maintained using a heating pad.Brie y, adult mice (2-4 months old) were anesthetized with iso urane (1-2%) and head-xed in a stereotaxic frame (Kopf Instruments).A total volume of 500-700 nL of AAVs was injected unilaterally or bilaterally into the dorsal striatum with two locations per hemisphere (coordinates: 0.5 mm AP to Bregma, ±2.2 mm ML, −2.5 mm DV from dura surface; 1.3 mm AP from Bregma, ±1.8 mm ML, −2.5 mm DV from dura surface) or SN (coordinates: −3.1 mm AP from Bregma, ±1.5 mm ML, −3.9 mm DV from dura surface).The infusion of viruses was controlled by a stereotaxic injector (Stoelting) at a speed of 75 nL/min.After a 5-minute wait following the end of injection, the injector was slowly withdrawn.The scalp was then sutured, and the mice were returned to their home cages.All behavior experiments were performed at least 4 weeks after injection to allow time for full heterologous gene expression.For Aldh1a1 CreERT2 mice, 4-OHT was injected intraperitoneally at a dosage of 10 mg/kg bodyweight for ve consecutive days, beginning one week after surgery, to induce Cre recombinase expression.
To prepare mice for optogenetics or ber photometry experiments, we performed a second surgery to implant optic bers at least 3 weeks after viral injection.For behavioral optogenetics experiments, optical ber stubs (200 μm core, 0.39 NA, Thorlabs) were bilaterally implanted with the tips positioned over the SNc (−3.1 mm AP from Bregma, ±1.5 mm ML, −3.9 mm to −4.1 mm DV from dura surface) or GPe (−0.3 mm AP from Bregma, ±2.0 mm ML, −3.5 mm from dura surface).The optic bers were secured in place with a thick layer of radiopaque adhesive cement (C&B METABOND, Parkell).Once dried, Vetbond tissue adhesive (3M) was applied to seal the head incision with adhesive cement.
For simultaneous photometry recordings and optogenetics experiments, optic ber stubs with a 200 μm core and 0.39 NA (Thorlabs) were unilaterally implanted over the SN (−3.1 mm AP from Bregma, 1.5 mm ML, −3.9 mm to −4.1 mm DV from dura surface) for optogenetic stimulation.In addition, optic ber stubs with a 200 μm core and 0.5 NA (Plexon) were implanted over dorsolateral striatum (+1.0 mm AP from Bregma, 2.0 mm ML, −2.2 mm to −2.5 mm DV from the dura surface) on the same side for striatal rDA3m photometry recording during optogenetic stimulation.
For the treadmill ber photometry experiment, optic ber stubs with a 200 μm core and 0.5 NA (Thorlabs) were unilaterally or bilaterally implanted in the SN (−3.1 mm AP from Bregma, 1.5 mm ML, −4.0 mm to −4.3 mm DV from dura surface) to record the axon calcium signals of patch or matrix dSPNs.The optic bers were xed in place using adhesive cement.A titanium metal head-bar (Labeotech) was mounted on top of the adhesive cement rostral to the ber stubs for head-restraint.The animals were allowed to recover for at least one week after the optic implantation before any optogenetics or ber photometry experiments.

Open-eld spontaneous locomotion
Locomotion in freely moving mice was measured using video tracing analyses.Mice were habituated for 30 min in the behavioral room, wherein a 20W lamp was shielded in a box and positioned in the dark opposite to the testing area to provide a diffuse light source.For the video tracing test, mice were placed in a 50cm × 50cm grey opaque chamber for 30min.Activity was recorded from overhead by a digital camera at a frame rate of 15 Hz.EthoVision XT software (Noldus) was used to track the mice and analyze the video for velocity, time, and distance travelled.
Chemogenetic manipulation JHU37160-dihydrochloride (Hello Bio) was dissolved in water to achieve a stock concentration of 0.3 mg/mL and stored in small aliquots at -20 °C.Prior to each use, the working solution was freshly prepared by diluting the stock 10-fold with 0.9% saline.Mice received 0.3mg/kg bodyweight dosage via intraperitoneal injection 30 minutes before behavioral tests.

Optogenetic stimulation
A LED light source (PlexBright, Plexon) was connected to a multimodal optic ber patch cable (200 μm core, 0.39 NA, Plexon) through ceramic sleeves (Thorlabs) to the ferrule of the optic ber stubs previously implanted in the mouse.Light power was calibrated using a power meter (PM100D, Thorlabs) to achieve the desired output measured at the tip of optic ber.For ChR2 experiments, photo stimulation consisted of 5ms-width blue light pluses (465nm, 3mW) at different frequency or constant blue light (465nm, 0.25mW).Light pulses were generated by a TTL pulse generator (OPTG-4, Doric Lenses).

Optogenetics with open eld
Mice were habituated in the testing room 30min before testing, and the apparatus was cleaned with 50% ethanol between animals.Mice were placed in a 50cm × 50cm clear chamber and their activities were captured by top-view and side-view cameras (Logitech).After a 3min exploration period, mice received either 3min ON and 3min OFF bilateral stimulation or 10s ON and 1min OFF bilateral stimulation.Video and LED light TTL were acquired and synchronized with Synapse software (Tucker-Davis Techonologies).Distanced traveled and velocity during the acquisition period were calculated using EthoVision XT software.Ambulation bouts were scored as periods of movement >2cm/s lasting for > 0.5s and separated by >0.5s.Immobility bouts were scored as periods of <2% pixel change lasting for >0.5s and separated by >0.5s.

Optogenetics with ber photometry
Mice were allowed to freely move in an open eld chamber and received photostimulation of 2s, 5s, or 15s for a total of approximately 10 trials for each period setting, with intervals of at least 45s between trials.The rDA3m signals were recorded simultaneously with videotaping (see "Fiber Photometry section).We established the baseline by determining the average ΔF/F value in the 2 s preceding stimulation.The reduction in dopamine amplitude was computed by subtracting the average ΔF/F value in the 0.2 s prior to stimulus offset from the baseline.For Calb1 IRES2-Cre mice subjected to matrix dSPNs stimulation, the dopamine peak was determined by subtracting the average ΔF/F during a 0.1s interval between 0.35 to 0.45s after stimulus onset from the baseline.Additionally, post-stimulus dopamine change amplitude was calculated by subtracting the average ΔF/F during a 5-sec interval between 3 to 8s after stimulus offset from the baseline.

Fiber Photometry
Dopamine sensor or GCaMP8s uorescence was measured using a locked-in ampli er system (Tucker-Davis Technologies, Model RZ10X with Synapse software).The photometry recordings were conducted with either green (GCaMP8s) or red (rDA3m) uorescence.For GCaMP8s recordings, the blue LED (465nm) was sinusoidally modulated at 330 Hz, and the UV LED (405nm) was modulated at 211 Hz as an isosbestic control channel.For rDA3m recordings, the green LED (560nm) was modulated at 410 Hz, and the UV LED was modulated at 211 Hz.The peak intensity of each LED was calibrated to 20-60 μW, measured at the distal end of the patch cable.Light emissions were ltered through a 6-port uorescence mixing cube (Doric Lens) before being coupled to an optic patch cable (200μm core, 0.5 NA), a xed to the implanted optic ber in each mouse.The emitted uorescent signals were collected by integrated photosensors in RZ10X real-time signal processor equipped with lock-in ampli er.Transistor-transistor logic (TTL) signals were employed to timestamp onset times for each event of interest (e.g., stimulation onset, locomotion onset and offset), which were detected via the RZ10X in the Synapse software.Fiber photometry data was analyzed using custom MATLAB code.Demodulated 465nm, 560nm, and 405nm recording traces were recorded at a sampling rate of 1k Hz.
For the analysis of rDA3m signals during optogenetic stimulation, to capture the large offset change caused by optogenetic stimulation, the demodulated photometry trace of rDA3m was normalized to compute ΔF/F.F was estimated as a referenced rDA3m signal, where a RANSAC ordinary least square linear regression was performed between the demodulated UV isosbestic reference signal and demodulated rDA3m signal to transform the reference signal to account for the differences in gain and offset between the two signals, as well as possible motion artefacts and long-term photo bleaching effects.Then the referenced rDA3m signal (F) was subtracted from the demodulated rDA3m signal to compute ΔF.
To analyze the modulation of GCaMP8s activity change with locomotor behavior, photometry signal normalization followed a published study 37 , using custom Matlab code.Demodulated photometry traces of both GCaMP8s and UV isosbestic reference channels were pre-normalized by rst computing ΔF/F0.F0 was estimated by calculating the 10th percentile of the raw photometry amplitude using a 5-sec sliding window to account for slow, correlated uorescence changes, including photobleaching in both channels.ΔF was calculated as raw photometry amplitude subtracted its respective F0.Both channels were initially normalized with this procedure.An additional referencing procedure was performed to remove the effects of motion or mechanical artefacts from analysis.For this referencing procedure, the ΔF/F0 of the UV reference signal was low-pass ltered with a second-order Butterworth lter with a 3 Hz corner frequency.Then a RANSAC ordinary least squares regression between the ltered reference signal and normalized GCaMP8s signal was used to transform the reference signal to account for the differences in gain and offset between the two signals.Lastly, the transformed references trace was subtracted from the normalized GCaMP8s trace as the nal ΔF/F.The transformed reference trace was then used as UV ΔF/F.Only the experiments where the maximum percentage of DF/F exceeded 1.5 and the GCaMP8s and the UV reference correlation was below 0.6 were included for further analysis.Z-score of ΔF/F was calculated with mean and standard deviation of the entire recording session.

Head-xed voluntary treadmill with ber photometry
The head-xed locomotion was achieved with a compact low friction manually driven treadmill, originally designed by Janelia Research Campus of the HHMI, purchased from LABmaker (Berlin, Germany).Te on belt movement was tracked by a rotary encoder, which sends 0-3.3V analog output of speed and direction to an ADC port at TDT RZ10x to synchronize with photometry recordings.The speed was calibrated with 10 cm/s at 2.5V and sampled at 1k Hz.The head xture was achieved through head xation system along with an implantable titanium head bar from Labeotech (Montreal, Canada).Mice started treadmill habituation training one week after recovering from head-bar implant surgery to get used to walking on treadmill while head was xed.Each habituation session lasted 20min and repeated at most twice a day for 3-4 days until mice were able to spontaneously initiate at least 20 walk bouts during a 20min session.After habituation training, a 30min ber photometry recording session was performed for each individual mouse, while they walked on the treadmill with head xation.
Movement versus rest time bins were de ned with a 0.25 cm/s threshold on the velocity trace.Isolated movement periods with duration shorter than 0.5s or movement periods with average velocity smaller than 0.5cm/s were excluded for analysis.Because mice tended to walk slower on the treadmill, time bins were considered as rest periods only if they lasted longer than 0.8s.A movement bout initiation time was de ned as velocity cross the threshold at the end of the rest period, and a movement termination time was de ned as the time velocity fell below the threshold followed by a rest period.Timestamp of maximum value of z-scored GCaMP activity during 1s before and 1s after locomotion onset or offset was calculate the time of maximum activity for the respective events.Activity slope during locomotion offset was computed as the coe cient of a least-square linear regression t for the z-scored GCaMP activity during the interval of 0.5s before and after locomotion offset.

Statistics
Data were analyzed by Prism 9 software (Graphpad) and custom code written in MATLAB (MathWorks).All statistical details for each experiment can be found in the corresponding gure legends.Data were presented as means and standard errors of the mean (SEM) or median and minimum or maximum.We assessed the statistical signi cance using parametric t-test and two-way analysis of variance (ANOVA).

Discussion
In this study we challenge the canonical model suggesting that striatonigral dSPNs promote locomotion and demonstrate that this model perhaps only applies to the matrix striatonigral dSPNs, while the patch striatonigral dSPNs play an opposing role by suppressing locomotion.This inhibitory function of patch dSPNs is implicated in the rise of neuronal activity during the offset of self-paced movement and is causally validated by optogenetic activation.Furthermore, we showed the critical involvement of GABA-B receptors in ALDH1A1 + nigrostriatal DANs during patch dSPN-induced locomotion inhibition.
The dSPNs and iSPNs are subdivided into patch and matrix compartments based on neurochemical markers and their input-output connectivity; 3 however, it remains challenging to ambiguously de ne the patch compartments.Conventional patch markers, including MOR1, substance P and enkephalin, can only identify a portion of patch compartments along the rostral to caudal axis of the striatum 22 .Speci cally, MOR1 marks the patch in the rostral and intermediate striatum, while substance P and enkephalin label the patch in the intermediate and caudal striatum 22 .In contrast, Kremen1 marks the patch structures within the entire dorsal striatum, including the dorsolateral striatum, a region heavily involved in motor control and learning 23 .Additionally, Kremen1 marks both Drd1-and Drd2-expressing SPNs in the same ratio as the general SPN population, while MOR1 tends to label more Drd1-expressing dSPNs 24 .In addition to MOR1, Pdyn and Tshz1 are also preferentially mark the patch dSPNs 9,10 .Notably, there is less than 3% overlap between the Pdyn + and Tshz1 + Drd1-expressing dSPNs 25 , suggesting heterogeneity among patch SPNs based on different gene expression pro les.Furthermore, substantially more Pdyn + and Tshz1 + SPNs are distributed in the matrix compartments compared to the Kremen1 + SPNs.Therefore, Kremen1 serves as a useful marker for a distinctive population of patch SPNs, which unbiasedly labels both dSPNs and iSPNs selectively in the patch compartments.
The Pdyn-positive (Pdyn + ) dSPNs were suggested to promote locomotion, while the Tshz1-positive (Tshz1 + ) ones were claimed to suppress locomotion in a place-preference test 25 .Since patch dSPNs have been implicated in discriminating valence during decision-making, changes in movement velocity by patch dSPN activation in valence context (such as place preference test) may re ect alterations in active avoidance or approach choices rather than locomotion itself 11 .Moreover, the roles of SNrprojecting dSPNs were not speci cally examined in the aforementioned study 25 , considering the GPiprojecting dSPNs may originate from a distinct population 26 .As Tshz1 + patch dSPNs have also been associated with locomotion suppression 25 , we conducted a comparison of Kremen1 + and Tshz1 + Drd1expressing neurons in the dorsal striatum.Our ndings reveal that Kremen1 and Tshz1 mark two largely different subtypes of patch dSPNs in the dorsal striatum.While we demonstrate that SNr-projecting Kremen1 + dSPNs inhibit ambulatory locomotion, whether SNr-projecting Tshz1 + SPNs serve a similar function remains unclear 25 .Given the signi cant diversity among patch SPNs based on molecular markers and connectivity, future studies should explicitly identify subtypes using genetic markers and projection targets.
Both patch and matrix dSPN activity showed a similar transient increase at the onset of locomotion, but started to diverge as movement ceased, with matrix dSPN exhibiting decreased activity while patch dSPN transiently increased activity.The activity pattern of patch and matrix dSPNs during locomotion mirrors previous ndings of synergistic concurrent activation of opposing dSPNs and iSPNs during movement initiation 24 with their activity decorrelating as movement progresses 27 .Furthermore, manipulating these two populations of dSPNs causally produced opposing locomotor effects.These ndings highlight the presence of opposing motifs within the dSPN population, one originating from the patch and the other from matrix, akin to the dichotomous organization of dSPN and iSPN in the broader dorsal striatum.Our results suggest that patch dSPN activity may facilitate state transition from movement to quiescent.In support of this hypothesis, patch dSPN activity began to ramp hundreds of milliseconds before movement termination, and optogenetic activation of patch dSPNs produced statedependent locomotor effect, shortening movement bout duration while increasing instance of immobility.State transition regulation has been identi ed as a fundamental principle of basal ganglia motor control 28 .In addition to receive sensorimotor inputs, patch dSPNs receive more limbic inputs than matrix dSPNs 29 and have robust direct connections with midbrain DANs, suggesting their role in integrating external sensorimotor information with implicit motivation to regulate motor state transitions.Conversely, we found that patch iSPNs exerted a positive, albeit less pronounced, effect on ambulatory movement upon activation compared to patch dSPNs.This aligns with previous ndings suggesting that striatal iSPNs apply inhibitory motor control through their GABAergic axon collaterals to surrounding dSPNs, while the majority of striatopallidal inputs drive negative reinforcement learning 30 .The synaptic strength of iSPN-dSPN connections within the patch remains unclear.Given the relatively smaller number of patch iSPNs, their collaterals within the patch may not be as robust.Further exploration is warranted to elucidate the function of patch iSPNs.ALDH1A1 + nigrostriatal DANs receive the most monosynaptic inputs from the dorsal striatum 12 .
Furthermore, these neurons appear to form reciprocal innervation with dSPNs in the dorsal regions of dorsal striatum 12 .This reciprocal connection between ALDH1A1 + DANs and dSPNs may constitute a feedback loop for timely regulation of the dopamine release and SPN activity in motor control and learning.While both patch and matrix dSPNs innervate ALDH1A1 + DANs 12 , the axon terminals of patch dSPNs are the primary presynaptic components in this so-called striosome-dendron bouquet structure with the dendrites of ALDH1A1 + nigrostriatal DANs 14,15 .This arrangement potentially plays a pivotal role in regulating dopamine release in both dorsal striatum and SNr.Accordingly, brain slice recordings indicate that patch dSPNs deliver stronger inhibitory inputs to the DANs and could induce rebound ring in ALDH1A1 + nigrostriatal DANs compared to matrix dSPNs and the other inhibitory inputs 14,16 .Consistent with these in vitro studies, our in vivo recordings reveal a more pronounced inhibition of dopamine release upon stimulating patch dSPNs compared to the matrix dSPNs, with a rebound dopamine release observed upon the cessation of patch dSPN stimulation.Supporting the functional signi cance of the connection between patch dSPNs and ALDH1A1 + nigrostriatal DANs, genetic deletion of GABA-B receptors in ALDH1A1 + nigrostriatal DANs completely abolished the locomotion-suppressing effects mediated by patch dSPNs.This synaptic coupling might dynamically regulate the dopamine supply in both the dorsal striatum and SNr, thereby modulating various motor activities.
Notably, the activation of matrix dSPNs initially triggered a transient increase in dopamine release before a subsequent reduction occurred.Considering that matrix dSPNs project to inhibitory parvalbumin neurons (PVNs) in the SNr and PVNs provide direct and tonic inhibitory inputs to DANs [31][32][33][34] , it's plausible that matrix dSPNs initially disinhibit DANs via PVNs, resulting in an early increase in DAN activity and dopamine release.The prevalence of this transient increase of dopamine release within the rst two seconds of matrix dSPN stimulation suggests that the primary effect of matrix dSPN activation on DANs is disinhibition or excitation.Therefore, the distinct modulation of DAN activity by patch and matrix dSPNs may underlie their differential in uence on motor control.The Kremen1 2A − Cre KI mice were generated using the CRIPSR/Cas9 approach in the C57BL/6J strain by Shanghai Model Organisms Inc.A donor plasmid containing the 2A ribosome skipping sequence, Cre DNA recombinase sequence, and anking Kremen1 mouse genomic DNA sequence was constructed.This was used to insert to the 2A-Cre DNA fragment into exon 9 immediately after the stop codon of Kremen1 gene locus, guided by the gRNA with sequence GTGGGCTTCAGTCACTCACG AGG.One founder mouse was generated, and the correct genomic modi cation was con rmed by sequencing.

Laser capture microdissection and RNA sequencing
One-month-old Nr4a1-eGFP transgenic mice were euthanized with CO 2 followed by rapid decapitation.
The brains were immediately dissected and preserved at − 80°C.Cryosectioning of the frozen brains was performed at − 20°C, and the sections were mounted onto PAN membrane frame slides (Applied Biosystems, Foster City, CA).The dorsal striatum was determined based on anatomic landmarks such as the corpus callosum, lateral ventricle and nucleus accumbens.Using the ArturusXT microdissection system with uorescent illumination (Applied Biosystems), the eGFP-positive island-like structures within the dorsal striatum of Nr4a1-eGFP transgenic mice were carefully isolated onto LCM Macro Caps (Applied Biosystems) and designated as "patch", while surrounding tissues of similar size were also isolated onto Macro Caps and designated as "matrix".
Total RNA extraction and puri cation were executed from hundreds of caps using the PicoPure Isolation kit (Applied Biosystems), with subsequent genomic DNA clearance facilitated by RNase-free DNase (Qiagen) following the manufacturer's protocols.Quanti cation of RNA was performed using a NanoDrop spectrophotometer (ThermoFisher), and RNA integrity was evaluated using the Bioanalyzer RNA 6000 pico assay (Agilent).Only RNA samples exhibiting high integrity were selected for subsequent patch and matrix RNAseq library preparation.
cDNA libraries were prepared from the puri ed RNA using the TruSeq Stranded Total RNA LT library preparation kit (Illumina) following the manufacturer's protocol.The quality of the libraries was assessed using the Bioanalyzer DNA 1000 assay (Agilent) before sequencing on an Illumina HiSeq 2000 platform.Fastq les were generated using the standard Illumina pipeline.Transcript abundance, annotated by Ensembl, was quanti ed using Salmon in a non-alignment-based mode, and gene-level counts were estimated utilizing the Tximport package (Bioconductor).Normalization of counts and subsequent data analysis were conducted following previously established procedures 35 .The accession number of the striatal tissue RNA-seq is PRJNA870469.
RNA in situ and image analysis RNA situ hybridization (ACDBio, RNAscope) was used to detect the expression of Drd1, Drd2, Kremen1 and Tshz1 mRNAs in the dorsal striatum of adult C57BL/6J mice.For tissue preparation, mice were anesthetized with CO 2 and rapidly decapitated.The brains were fresh-frozen on dry ice and stored at -80°C before sectioning.Striatal sections (12µm) were collected using a cryostat (Leica Biosystems) and stored at -80°C until processed.
RNAscope images were analyzed using Imaris (v10.0.0,Bitplane, Belfast Northern Ireland, UK).Surfaces for the dorsal and ventral striatum were created using the Allen Brain Atlas as a reference.The spatial patch was de ned by outlining Kremen1 + SPN clusters (a minimum of 5 SPNs), and the density of Kremen1 + SPNs within a spatial patch is at least 200 cells/mm 2 .Surfaces for individual channels (i.e., Drd1, Drd2, Kremen1, Tshz1 and DAPI) were created with unique parameters, which were saved and applied to all images within the same batch.Minor adjustments were made in subsequent rounds of RNAscope to improve quanti cation accuracy.DAPI surfaces in the striatum were ltered by the mean intensity of Drd1 (dSPNs) and the mean intensity of Drd2 (iSPNs).These Drd1 + and Drd2 + cells were further ltered by the overlapped volume ratio to Kremen1 to quantify patch dSPNs and patch iSPNs, respectively.RNAscope analysis with Tshz1 was conducted in a similar manner, ltering Drd1 + and Kremen1 + cells by the overlapped volume ratio to the Tshz1 surface.All data points are presented as the average across striatal sections of both hemisphere for each mouse.
Construction of pAAV-FLEX-SaCas9-U6-sg Gabbr1 The reference sequence of Gabbr1 gene was retrieved from the UCSC genome browser database (http://genome.ucsc.edu/),and subsequent identi cation and alignment of exons were performed using the Mouse Genome Informatics (MGI) database.The most 5′ common coding exons were then selected, and the sequence was uploaded to the CRISPOR website (http://crispor.org)for identifying potential sgRNAs and PAM sequences.Three Gabbr1 sgRNAs were synthesized as short oligos (Euro ns Genomics) with a 5′ CACC-3′ overhang on the forward primer, and a 5′ -AAAC 3′ overhang on the reverse primer, facilitating seamless integration into the pX601-AAV-CMV;NLS-SaCas9-NLS-3xHA-bGHpA;U6;BsaI-sgRNA vector, a gift provided by Dr. Feng Zhang (Addgene plasmid # 61591).Sanger sequencing con rmed the insertion of the Gabbr1 sgRNA using the primer TAACCACGTGAGGGCCTATTTC.

Stereotaxic injection and optic ber implantation
The stereotaxic survival surgery was performed as previously described 35 .All surgery were conducted under aseptic conditions, and body temperature was maintained using a heating pad.Brie y, adult mice (2-4 months old) were anesthetized with iso urane (1-2%) and head-xed in a stereotaxic frame (Kopf Instruments).A total volume of 500-700 nL of AAVs was injected unilaterally or bilaterally into the dorsal striatum with two locations per hemisphere (coordinates: 0.5 mm AP to Bregma, ± 2.2 mm ML, − 2.5 mm DV from dura surface; 1.3 mm AP from Bregma, ± 1.8 mm ML, − 2.5 mm DV from dura surface) or SN (coordinates: −3.1 mm AP from Bregma, ± 1.5 mm ML, − 3.9 mm DV from dura surface).The infusion of viruses was controlled by a stereotaxic injector (Stoelting) at a speed of 75 nL/min.After a 5-minute wait following the end of injection, the injector was slowly withdrawn.The scalp was then sutured, and the mice were returned to their home cages.All behavior experiments were performed at least 4 weeks after injection to allow time for full heterologous gene expression.For Aldh1a1 CreERT2 mice, 4-OHT was injected intraperitoneally at a dosage of 10 mg/kg bodyweight for ve consecutive days, beginning one week after surgery, to induce Cre recombinase expression.
To prepare mice for optogenetics or ber photometry experiments, we performed a second surgery to implant optic bers at least 3 weeks after viral injection.For behavioral optogenetics experiments, optical ber stubs (200 µm core, 0.39 NA, Thorlabs) were bilaterally implanted with the tips positioned over the SNc (− 3.1 mm AP from Bregma, ± 1.5 mm ML, − 3.9 mm to − 4.1 mm DV from dura surface) or GPe (− 0.3 mm AP from Bregma, ± 2.0 mm ML, − 3.5 mm from dura surface).The optic bers were secured in place with a thick layer of radiopaque adhesive cement (C&B METABOND, Parkell).Once dried, Vetbond tissue adhesive (3M) was applied to seal the head incision with adhesive cement.
For simultaneous photometry recordings and optogenetics experiments, optic ber stubs with a 200 µm core and 0.39 NA (Thorlabs) were unilaterally implanted over the SN (− 3.1 mm AP from Bregma, 1.5 mm ML, − 3.9 mm to − 4.1 mm DV from dura surface) for optogenetic stimulation.In addition, optic ber stubs with a 200 µm core and 0.5 NA (Plexon) were implanted over dorsolateral striatum (+ 1.0 mm AP from Bregma, 2.0 mm ML, − 2.2 mm to − 2.5 mm DV from the dura surface) on the same side for striatal rDA3m photometry recording during optogenetic stimulation.
For the treadmill ber photometry experiment, optic ber stubs with a 200 µm core and 0.5 NA (Thorlabs) were unilaterally or bilaterally implanted in the SN (− 3.1 mm AP from Bregma, 1.5 mm ML, − 4.0 mm to − 4.3 mm DV from dura surface) to record the axon calcium signals of patch or matrix dSPNs.The optic bers were xed in place using adhesive cement.A titanium metal head-bar (Labeotech) was mounted on top of the adhesive cement rostral to the ber stubs for head-restraint.The animals were allowed to recover for at least one week after the optic implantation before any optogenetics or ber photometry experiments.

Open-eld spontaneous locomotion
Locomotion in freely moving mice was measured using video tracing analyses.Mice were habituated for 30 min in the behavioral room, wherein a 20W lamp was shielded in a box and positioned in the dark opposite to the testing area to provide a diffuse light source.For the video tracing test, mice were placed in a 50cm × 50cm grey opaque chamber for 30min.Activity was recorded from overhead by a digital camera at a frame rate of 15 Hz.EthoVision XT software (Noldus) was used to track the mice and analyze the video for velocity, time, and distance travelled.
Chemogenetic manipulation JHU37160-dihydrochloride (Hello Bio) was dissolved in water to achieve a stock concentration of 0.3 mg/mL and stored in small aliquots at -20°C.Prior to each use, the working solution was freshly prepared by diluting the stock 10-fold with 0.9% saline.Mice received 0.3mg/kg bodyweight dosage via intraperitoneal injection 30 minutes before behavioral tests.

Optogenetic stimulation
A LED light source (PlexBright, Plexon) was connected to a multimodal optic ber patch cable (200 µm core, 0.39 NA, Plexon) through ceramic sleeves (Thorlabs) to the ferrule of the optic ber stubs previously implanted in the mouse.Light power was calibrated using a power meter (PM100D, Thorlabs) to achieve the desired output measured at the tip of optic ber.For ChR2 experiments, photo stimulation consisted of 5ms-width blue light pluses (465nm, 3mW) at different frequency or constant blue light (465nm, 0.25mW).Light pulses were generated by a TTL pulse generator (OPTG-4, Doric Lenses).

Optogenetics with open eld
Mice were in the testing room 30min before testing, and the apparatus was cleaned with 50% ethanol between animals.Mice were placed in a 50cm × 50cm clear chamber and their activities were captured by top-view and side-view cameras (Logitech).After a 3min exploration period, mice received either 3min ON and 3min OFF bilateral stimulation or 10s ON and 1min OFF bilateral stimulation.Video and LED light TTL were acquired and synchronized with Synapse software (Tucker-Davis Technologies).Distanced traveled and velocity during the acquisition period were calculated using EthoVision XT software.Ambulation bouts were scored as periods of movement > 2cm/s lasting for > 0.5s and separated by > 0.5s.Immobility bouts were scored as periods of < 2% pixel change lasting for > 0.5s and separated by > 0.5s.For behavioral-state dependent effects of optogenetic stimulation experiment, mice received 10s ON and 1 min OFF bilateral 20Hz stimulation.Ambulation state was determined by movement velocity > 2cm/s in the entire 0.5 second immediately preceded either stimulation onset or stimulation offset.Quiescence (includes small movement) state was determined by movement velocity < 1.5cm/s in the entire 0.5 second immediately preceded either stimulation onset or stimulation offset.
Due to kinematics of stimulation on behavior, quanti cation for velocity change used 1 sec interval instead.

Optogenetics with ber photometry
Mice were allowed to freely move in an open eld chamber and received photostimulation of 2s, 5s, or 15s for a total of approximately 10 trials for each period setting, with intervals of at least 45s between trials.The rDA3m signals were recorded simultaneously with videotaping (see "Fiber Photometry section).We established the baseline by determining the average ΔF/F value in the 2 s preceding stimulation.The reduction in dopamine amplitude was computed by subtracting the average ΔF/F value in the 0.2 s prior to stimulus offset from the baseline.For Calb1 IRES2 − Cre mice subjected to matrix dSPNs stimulation, the dopamine peak was determined by subtracting the average ΔF/F during a 0.1s interval between 0.35 to 0.45s after stimulus onset from the baseline.Additionally, post-stimulus dopamine change amplitude was calculated by subtracting the average ΔF/F during a 5-sec interval between 3 to 8s after stimulus offset from the baseline.

Fiber Photometry
Dopamine sensor or GCaMP8s uorescence was measured using a locked-in ampli er system (Tucker-Davis Technologies, Model RZ10X with Synapse software).The photometry recordings were conducted with either green (GCaMP8s) or red (rDA3m) uorescence.For GCaMP8s recordings, the blue LED (465nm) was sinusoidally modulated at 330 Hz, and the UV LED (405nm) was modulated at 211 Hz as an isosbestic control channel.For rDA3m recordings, the green LED (560nm) was modulated at 410 Hz, and the UV LED was modulated at 211 Hz.The peak intensity of each LED was calibrated to 20-60 µW, measured at the distal end of the patch cable.Light emissions were ltered through a 6-port uorescence mixing cube (Doric Lens) before being coupled to an optic patch cable (200µm core, 0.5 NA), a xed to the implanted optic ber in each mouse.The emitted uorescent signals were collected by integrated photosensors in RZ10X real-time signal processor equipped with lock-in ampli er.
Transistor-transistor logic (TTL) signals were employed to timestamp onset times for each event of interest (e.g., stimulation onset, locomotion onset and offset), which were detected via the RZ10X in the Synapse software.Fiber photometry data was analyzed using custom MATLAB code.Demodulated 465nm, 560nm, and 405nm recording traces were recorded at a sampling rate of 1k Hz.
For the analysis of rDA3m signals during optogenetic stimulation, to capture the large offset change caused by optogenetic stimulation, the demodulated photometry trace of rDA3m was normalized to compute ΔF/F.F was estimated as a referenced rDA3m signal, where a RANSAC ordinary least square linear regression was performed between the demodulated UV isosbestic reference signal and demodulated rDA3m signal to transform the reference signal to account for the differences in gain and offset between the two signals, as well as possible motion artefacts and long-term photo bleaching effects.Then the referenced rDA3m signal (F) was subtracted from the demodulated rDA3m signal to compute ΔF.
To analyze the modulation of GCaMP8s activity change with locomotor behavior, photometry signal normalization followed a published study 37 , using custom Matlab code.Demodulated photometry traces of both GCaMP8s and UV isosbestic reference channels were pre-normalized by rst computing ΔF/F0.
F0 was estimated by calculating the 10th percentile of the raw photometry amplitude using a 5-sec sliding window to account for slow, correlated uorescence changes, including photobleaching in both channels.ΔF was calculated as raw photometry amplitude subtracted its respective F0.Both channels were initially normalized with this procedure.An additional referencing procedure was performed to remove the effects of motion or mechanical artefacts from analysis.For this referencing procedure, the ΔF/F0 of the UV reference signal was low-pass ltered with a second-order Butterworth lter with a 3 Hz corner frequency.Then a RANSAC ordinary least squares regression between the ltered reference signal and normalized GCaMP8s signal was used to transform the reference signal to account for the differences in gain and offset between the two signals.Lastly, the transformed references trace was subtracted from the normalized GCaMP8s trace as the nal ΔF/F.The transformed reference trace was then used as UV ΔF/F.Only the experiments where the maximum percentage of DF/F exceeded 1.5 and the GCaMP8s and the UV reference correlation was below 0.6 were included for further analysis.Z-score of ΔF/F was calculated with mean and standard deviation of the entire recording session.
. The Nr4a1-eGFP (Stock No: 036737-UCD) transgenic mice were obtained from Mutant Mouse Resource & Research Centers (MMRRC).The Calb1 IRES2-Cre mice (Stock No: 028523) and Ai14 (Stock No: 007908) were obtained from the Jackson laboratory.The A2a Flp KI mice were generated by the Rodent Transgenic Core of National Institute of Mental Health (NIMH).Both females and males were used for all experiments.Mice used for viral injections were between 2 and 4 months of age.The mice were housed in a 12-hour-light/12-hour-dark cycle in groups of 2-5 animals and had ad libitum water and a regular diet.All the behavioral tasks were performed during the light cycle.Littermates were randomly assigned to different groups prior to experiments.Generation of Kremen1 2A-Cre KI mice All mouse studies were conducted in accordance with the guidelines approved by the Institutional Animal Care and Use Committees (IACUC) of the Intramural Research Program of the National Institute on Aging (NIA), NIH.All mouse lines were maintained as heterozygotes in a C57BL/6J background.The Kremen1 2A − Cre KI mice were generated by Shanghai Model Organisms Inc. (Shanghai, China).The Aldh1a1 CreERT2 KI mice were generated as previously described12 .The Nr4a1-eGFP (Stock No: 036737-UCD) transgenic mice were obtained from Mutant Mouse Resource & Research Centers (MMRRC).The Calb1 IRES2 − Cre mice (Stock No: 028523) and Ai14 (Stock No: 007908) were obtained from the Jackson laboratory.The A2a Flp KI mice were generated by the Rodent Transgenic Core of National Institute of Mental Health (NIMH).Both females and males were used for all experiments.Mice used for viral injections were between 2 and 4 months of age.The mice were housed in a 12-hour-light/12-hour-dark cycle in groups of 2-5 animals and had ad libitum water and a regular diet.All the behavioral tasks were performed during the light cycle.Littermates were randomly assigned to different groups prior to experiments.Generation of Kremen1 2A − Cre KI mice

Figures
Figures

Figure 4 Patch
Figure 4