A non-canonical striatopallidal “Go” pathway that supports motor control

In the classical model of the basal ganglia, direct pathway striatal projection neurons (dSPNs) send projections to the substantia nigra (SNr) and entopeduncular nucleus to regulate motor function. Recent studies have re-established that dSPNs also possess “bridging” collaterals within the globus pallidus (GPe), yet the significance of these collaterals for behavior is unknown. Here we use in vivo optical and chemogenetic tools combined with deep learning approaches to dissect the roles of bridging collaterals in motor function. We find that dSPNs projecting to the SNr send synchronous motor-related information to the GPe via axon collaterals. Inhibition of native activity in dSPN GPe terminals impairs motor activity and function via regulation of pallidostriatal Npas1 neurons. We propose a model by which dSPN GPe collaterals (“striatopallidal Go pathway”) act in concert with the canonical terminals in the SNr to support motor control by inhibiting Npas1 signals going back to the striatum.


Abstract Introduction
In 2014, we observed that the density of bridging collaterals in the GPe is highly plastic in the adult animal, being 23 regulated by dopamine D2Rs and neural excitability. High levels of bridging collaterals also lead to a stronger reduction 24 in GPe firing rate following dSPN optogenetic stimulation 6 . Recent work extends our findings, showing that neural 25 activity and 6-OHDA dopamine lesions modulate bridging collateral density or connectivity to the GPe 7, 15,24 . These data 26 argue for a role of bridging collaterals in shaping the output of the BG circuitry and motor function.

27
To more directly understand the significance of bridging collaterals for behavior, it is essential to record their activity 28 dynamics in awake-behaving mice as well as to inhibit their activity acutely during natural behavior. Here, we overcame 29 existing technical challenges to address these questions. We combined terminal-specific in vivo calcium recording or 30 manipulation techniques, with in vivo physiology, closed loop approaches, and deep learning-based behavioral tracking 31 to dissect the role and relative dynamics of dSPN GPe bridging collaterals in motor function (summarized in 32 Supplementary Fig. S1). Specifically, we wanted to test two alternative hypotheses: (1) bridging collaterals functionally 33 diverge from SNr terminals, acting like a second "NoGo" pathway to inhibit the GPe and locomotion; or (2) they act in 34 convergence with canonical SNr projections working as a second "Go" pathway to promote locomotion and support   S2). We found that GCaMP6s expression was largely restricted to dSPN VGAT+ puncta and not fibers of passage 43 (Fig. 1B,C) and that dSPN terminals in the GPe accounted for more than half the density of SNr terminals (Fig. 1D).
expressing AAV 47 into the DMS and implanted with optic fibers above the GPe and SNr ( Fig. 2A

20
Supplementary Video S1). We used body center speed to compute the onset and offset of locomotor movements and 21 averaged all motor bouts aligned to the onset and offset (Fig. 2D-E). This confirmed that mouse speed increased at the 22 onset of movements and decreased at their offset. Like mouse speed, we found that dSPN activity in the GPe and SNr 23 (quantified using the z-score of deltaF/F; 'dFF') increased at movement onset and decreased at movement offset ( Fig.   24 2F). Moreover, total GPe and SNr activity correlated significantly with mouse speed (compared to shuffled control) with 25 a maximal Pearson correlation around 0.5 (GPe: r=0.58; SNr: r=0.53) (Fig. 2G). These data showed that dSPN axons 26 in the GPe and SNr continuously encode mouse speed during locomotion, consistent with findings at the cell body 43,48 .

27
Importantly, activity of dSPN axons in the GPe and SNr were highly correlated with each other (r=0.81) (Fig. 2H), 28 suggesting they encode copies of the same neuronal information during self-paced locomotion. These data align with 29 existing models emphasizing a role for striatal SPNs in representing the speed (or vigor) of body movements in a 30 continuous manner 43 and indicate that such information is transmitted down to synaptic terminals. More importantly,  Body speed (paired t-test p<0.001) and GPe/SNr dFF (ANOVA: main effect: p<0.001) significantly increase at movement 10 onset vs. offset G. GPe (r= 0.58) and SNr (r= 0.53) dFF significantly correlate with mouse speed when compared to 11 phase-shuffled data (Mann-Whitney GPe p<0.001, SNr p<0.001) H. GPe and SNr dFF are highly correlated with each 12 other (Pearson r= 0.81) vs. phase-shuffled data (Mann-Whitney p<0.001). Data is mean±SEM.

7/29
dSPN GPe and SNr axons track the temporal boundaries of motor jumping bouts 1 We next determined whether the activity of bridging collaterals is regulated by more complex motor behaviors. Indeed, 2 striatal SPNs are known to track multiple types of motor variables beyond body speed; for instance they show sustained 3 activity throughout the execution of motor sequences or track the temporal boundaries (onset, offset) of individual 4 movements 40,48,49 . Mice were subjected to a rotarod motor assay known to engage the striatum 32,39-46 which allows to 5 impose repetitive motor patterns allowing experimental control on running speed and trial averaging. First, mice were 6 subjected to 10 rotarod trials at accelerating speed (5 to 40 rpm). We found that the activity of GPe and SNr axons was 7 sustained throughout the rotarod epoch compared to 'pre' and 'post' rest periods (Supplementary Fig. S3B-C), showing 8 a higher baseline fluorescence and area under the curve (AUC) (Fig. S3D). Sustained dSPN terminal activity would be 9 consistent with evidence for certain striatal units showing sustained activation during rotarod running 40 . When zooming 10 in onto individual peaks, we also noticed that the properties of peaks arising from dSPN GPe or SNr axons strongly 11 differed across task epochs, showing higher frequency and lower amplitude (taken from the local baseline) during rotarod running (Supplementary Fig. S3D). Since no learning-related differences emerged across the 10 trials, trials 13 were pooled. The interval between peaks was also significantly shorter in the rotarod 'running' epoch as opposed to 14 'pre'/'post' rest periods (Supplementary Fig. S3E). Together, these data indicated that GPe and SNr axons are 15 activated by running and likely track running-related motor parameters in the rotarod task.

16
We hypothesized that dSPN axons track the temporal boundaries (onset and offset) of task-specific body 17 movements. To address this, we monitored the individual trajectories of mouse body parts during running using 18 DeepLabCut 50 (Supplementary Fig. S4A). Although we did observe foot stepping behavior, it was highly variable and 19 foot tracking quality in our setup was not good enough to allow behavior/calcium cross-analyses ( Supplementary Fig   20   S4A). However, we observed that, while performing the rotarod, all mice adopted a behavioral strategy to "jump" up the 21 rotarod then slide back down (Supplementary Video S2). This was made evident by tracking the vertical position (Y axis) of the lower body, which regularly alternated between the lower and upper bounds of the rotarod (also seen in 51 ) 23 ( Fig. 3D). To see if the duration of jumps decreased as the rotarod speed increased, we exposed animals to rotarod 24 trials at different constant speeds (5, 10 or 15 rpm) ( Fig. 3A-C). We computed the interpeak intervals between 25 consecutive jumps, and as expected there was a significant decrease in interpeak interval with increased rotarod speeds 26 ( Fig. 3E), We then aligned the Ca2+ signal to the onset of jumps to determine if GPe/SNr axons track the temporal 27 boundaries of jumping bouts. When averaging >1000 jumping bouts per trial type (5, 10, 15 rpm) we found that dSPN 28 GPe and SNr signals were time-locked to the boundaries of the jumps (up at onset and down at offset) (Fig. 3F,

29
Supplementary Fig. S4B). Moreover, as expected, jumps became more frequent as the rotarod speed increased.

30
Similarly, GPe/SNr jump-related transients became more frequent at increasing rotarod speeds (Fig. 3G). We then 31 conducted statistical analyses on >4000 individual jumps, to determine if the temporal properties of individual GPe/SNr 32 transients are adjusted on a trial-by-trial basis. We found that within individual motor bouts, the jump bout duration 33 (duration between two consecutive jumps) correlated significantly with the GPe or SNr Ca2+ transient duration (time 34 interval between two consecutive peaks) at different rotarod speeds (GPe: Pearson r= 0.75, SNr: r= 0.67) (Fig. 3H).

35
This confirmed our initial hypothesis, that dSPN GPe and SNr axons track the temporal boundaries of individual motor 36 bouts during rotarod running on a trial-by-trial basis, consistent with previous observations at the cell body 40,48,49 . Lastly,

37
we determined the degree of correlation between dSPN GPe and SNr axonal activity in the rotarod. Activity of GPe and 38 SNr terminals were highly correlated with each other during rest (Pearson r = 0.90), similar to the open field. However, 39 the correlation was significantly reduced (r= 0.70) during the running epoch (Fig. 3I). This indicated that although dSPN cell bodies send axonal copies to the GPe and SNr, differences emerge during motor behavior in a task-dependent the running phase of a rotarod task, showing both sustained activity during the entire task and acute activation at the

12
GPe and SNr dFF increases at jump onset, showing shorter and smaller transients with increasing rotarod speed G.

13
Duration of jumping bouts (peak-to-peak) and duration of GPe/SNr dFF transients (peak-to-peak) significantly decrease        axon branches. In principle, collateral/terminal calcium signals should dominate since voltage-gated calcium channels 20 are concentrated at terminals, and previous work found that calcium transients in primary axon branches are minimal 21 as compared to calcium transients in terminals 52 . Still, to get confirmation that the activity of dSPN GPe bridging collaterals is regulated by motor tasks, we performed GPe terminal-specific recordings in Drd1-cre mice expressing the 23 calcium indicator GCaMP6s tethered to the presynaptic vesicle protein synaptophysin (Synaptophysin-GCaMP6s) 25 .

24
However, as published before, we found that Synaptophysin-GCaMP6s has a low signal-to-noise ratio and higher rate 25 of photobleaching 53 (data not shown) and thus was not adequate for correlational analyses.
10/29 enriched in GPe dSPN terminals, showing 6x higher fluorescent optical density in terminals vs. axons; contrasting with 1 regular jGCaMP7s showing a 2x terminal:axon ratio (Fig 4C). Like for jGCaMP7s (Fig. 3, Supplementary Fig. S3), we 2 detected significant increases in calcium activity in GPe and SNr presynaptic terminals during running, evidenced by an 3 increased baseline and AUC ( Fig. 4E-F). This confirms that dSPN GPe terminals are engaged during motor tasks.

4
Importantly, like for jGCaMP7s, activity of GPe and SNr terminals were highly correlated with each other during rest 5 (Pearson r = 0.81), and the correlation was significantly reduced (r= 0.60) during running (Fig. 4G). Since jGCaMP7s 6 and SyGCaMP8s results concord, this confirms our finding that dSPN cell bodies send axonal copies to GPe and SNr 7 terminals, where local activity diverges in a task-dependent manner.

21 dSPN bridging collaterals in the GPe are necessary for motor function
Since dSPN GPe terminals encode locomotion and rotarod motor variables, we next asked if they are necessary for 23 normal locomotion and motor function. Selectively manipulating dSPN bridging collateral activity is not trivial. Using 24 classical excitatory opsins such as ChR2 is precluded since they would affect anterograde and retrograde action 25 potential propagation in dSPN passing fibers in the GPe (Fig. 1B). Moreover many inhibitory opsins have off-target 26 effects 25 . We therefore used the inhibitory DREADD hM4D, as it was previously shown to inhibit synaptic release with 27 minimal effects on action potentials in axons 55 and used to target dSPN collaterals in the ventral pallidum during cocaine 28 seeking 56 . Drd1-cre mice were bilaterally injected with a flexed AAV expressing hM4D or mCherry into the dorsal 11/29 inhibit GPe terminals. First, we verified that locally-infused radioactive [ 3 H]-CNO (300 nL; 7 µCi/mL) stayed restricted in 1 the GPe (Fig. 5C). We then locally infused the same volume (300 nL) of CNO (at 1mM) in the GPe 20-30 min before 2 behavior. We found that chemogenetic inhibition of bridging collaterals reduced locomotor speed in the open field ( Fig.   3   5D), shown by a significant reduction in mouse speed in hM4D but not in mCherry control mice. The same manipulation 4 also impaired rotarod motor performance, shown by a non-significant decrease in latency to fall and a significant 5 increased number of falls in hM4D but not mCherry mice (Fig. 5E). These data suggest that bridging collaterals support 6 and are necessary for motor control.

7
Of note, in this experiment dSPN passing axons fibers in the GPe going to the SNr are physically exposed to the 8 locally infused CNO. We therefore set to verify that our hM4D results could not be explained by unspecific effects in the 9 SNr. Indeed, although previous work showed that CNO+hM4D inhibits synaptic release with minimal effects on action 10 potentials in axons 55 , this was done in cortical neurons, which may have different biophysical properties than dSPNs.

11
Drd1-cre mice were injected with a mix of flexed AAVs expressing ChR2 and hM4D into the DMS. We used our 12 previously validated setup 6 to record single-unit responses in the GPe and SNr after acute optogenetic stimulation of 13 dSPN somas at increasing durations (0, 250, 500, 1000 ms) in anesthetized mice. We also locally infused Saline or 14 CNO (300 nL, 1mM) above the GPe 25 min before recording to inhibit synaptic release at dSPN GPe terminals ( Fig.   15 6A-B). Consistent with previous work 6,17 , in Saline control mice dSPN opto-stimulation led to an inhibition of spike firing 16 frequency in the GPe (Fig. 6C) and SNr (Fig. 6E). The dSPN opto-induced inhibition of GPe spike firing was blunted when dSPN GPe terminals were chemogenetically inhibited via local GPe CNO infusion (Fig. 6D). This confirmed that 18 local GPe CNO infusion in hM4D-expressing Drd1-cre mice (Fig. 5) inhibits synaptic release at dSPN GPe terminals, in 19 line with the established role of hM4D as a presynaptic release inhibitor 55 . Importantly, local CNO infusion into the GPe 20 did not affect opto-induced inhibition of SNr spike firing activity (Fig. 6F). Upon quantification, we found that local GPe 21 CNO infusion significantly reduced the number of inhibited units in the GPe (Fig. 6G), but not in the SNr (Fig. 6H).
Similarly, local GPe CNO infusion significantly blunted the opto-induced inhibition of GPe units ( Fig. 6I-

20
To confirm these findings and gain higher temporal resolution, we next used the recently developed Gi/o mosquito 21 rhodopsin eOPN3 shown to selectively inhibit synaptic release while maintaining action potential fidelity in axons 58 .
Drd1-cre mice were bilaterally injected with a flexed AAV expressing eOPN3 or GFP into the DMS and implanted with 23 GPe optic fibers ( Fig. 7A-B, Supplementary Fig. S7). As expected, optogenetic inhibition of dSPN GPe terminals 24 impaired rotarod motor performance, as shown by a significant decreased latency to fall detected in eOPN3 but not 25 GFP controls (Fig. 7C). We next performed a closed-loop open field task to inhibit bridging collaterals during ongoing 14/29 speed was compared in laser-on vs laser-off epochs. We found that closed-loop optogenetic inhibition of dSPN GPe 1 terminals reduced ongoing locomotion speed, as shown by a significant reduction in mouse speed detected in eOPN3 2 but not GFP mice (Fig. 7D, Supplementary Video S3). We then classified behaviors into motor states to dissect the 3 fine motor patterns induced by dSPN GPe inhibition. Trajectories of mouse body parts (obtained with DeepLabCut) were 4 used to classify frames into 3 categories: locomotion, motionless and fine movements. We found that opto-inhibition of dSPN GPe terminals promoted motor states consistent with decreased motion but not with behavioral arrest, as shown 6 by our observations of a significantly increased time spent in fine movements and a decreased time spent locomoting, 7 but no change in time spent motionless (Fig. 7E-F). There were no differences in time spent in center vs. periphery 8 zones, suggesting no effects on anxiety (Supplementary Fig. S5A). Altogether, these results show that dSPN GPe 9 terminals support locomotion and motor performance in the rotarod, suggesting they act as a second "Go" pathway. What could be the circuit mechanisms by which dSPN GPe terminals support motor function in the GPe? We 12 hypothesized that dSPN "Go" axons functionally inhibit ongoing motor-related activity in arkypallidal GPe "stop" neurons.

13
In slice experiments 15 15 . Since arkypallidal neurons are the primary target of dSPNs 15,17 , and since Npas1 18 in the GPe is almost exclusively expressed in arkypallidal neurons 15 we used Npas1-cre mice to capture cell-specific Npas1 signals. Drd1-cre;Npas1-cre mice were injected with a flexed AAV expressing ChrimsonR or mCherry into the 20 DMS and a flexed AAV expressing GCaMP6s into the GPe. An optic fiber was implanted above the GPe to 21 optogenetically stimulate dSPN axons in the vicinity of GCaMP+-Npas1 neurons ( Fig. 8A-B, Supplementary Fig. S7).
Behavior was quantified using body positions obtained from DeepLabCut (Fig. 8C, Supplementary Fig. S6A). We used 23 unilateral stimulation of dSPN GPe axons to minimize effects on behavior, and disentangle them from effects on calcium 24 activity. In a first approach, we asked if dSPN axons inhibit ongoing Npas1 activity. Here closed-loop dSPN axon 25 stimulation was triggered when Npas1 activity reached a maxima (see methods), based on previous work showing that 26 Npas1 activity is high at locomotor onset 59 . We found that dSPN axon stimulation for short or long durations (3 or 10 s; 27 20 Hz) led to the inhibition of Npas1 activity in a graded manner (stronger inhibition with 2mW vs. 0.5mW) and observed 28 in all 6 ChrimsonR mice recorded (Fig. 8D, Supplementary Fig. S6, Supplementary Video S4), but not in mCherry 29 controls (Supplementary Fig. S6C). Stimulating dSPN axons at 20Hz but not 10Hz significantly inhibited Npas1 30 neurons, suggesting Npas1 inhibition occurrs only when activation of dSPNs reaches a certain threshold (Fig. 8F).

31
Importantly, these neural effects were decorrelated from effects on behavior. Indeed, as expected, unilateral stimulation 32 did not affect mouse speed (Fig. 8E, H). Significant increases in rotational behavior emerged after 10 but not 3 s 33 stimulation protocols (Supplementary Fig. S6B). This suggested that the Npas1 neural effects of dSPN stimulation 34 could not be solely explained by changes in mouse behavior. In a second approach, we asked if dSPN axons inhibit 35 motor-related Npas1 signals. Here closed-loop dSPN unilateral axon stimulation was triggered when the mouse was 36 actively locomoting (see methods) (Fig. 8I). Stimulation was done at ultra-low ChrimsonR-LED power (0.2 mW), which 37 had no effects on mouse speed (Fig. 8J) or rotations (Supplementary Fig. S6B). As expected 59 , before stimulation 38 Npas1 dFF activity increased concurrently with mouse speed (Fig. 8J-K). We found that dSPN axon stimulation was 39 sufficient to inhibit motor-related Npas1 calcium activity, even at this low optogenetic power (Fig. 8K). Finally, we also verified the absence of crosstalk between the LEDs required to activate ChrimsonR and GCaMP, respectively 15/29

GPe Npas1 but not ChAT neurons mediate the effects of bridging collaterals on motor function
Our data suggest a mechanism by which dSPN bridging collaterals support motor function by inhibiting their primary 23 GPe target, Npas1 neurons, during ongoing behavior. This would align with past research showing that Npas1 neurons 24 are locomotor-suppressing 15,60 . Since bridging collateral inhibition reduces locomotion speed and impairs rotarod motor 25 performance, we wondered whether disinhibition of Npas1 neurons could recapitulate these phenotypes. Npas1-cre 26 mice were bilaterally injected with a flexed AAV expressing ChR2 or YFP into the GPe and implanted with GPe optic 27 fibers ( Fig. 9A-B, Supplementary Fig. S7). We stimulated Npas1 neurons at 20 Hz around their firing frequency during 28 ongoing locomotion 59 . Optogenetic stimulation of Npas1 neurons inhibited locomotion speed, as shown by a significant 29 reduction in mouse speed in ChR2 mice but not YFP controls (Fig. 9C), confirming the locomotor-suppressing role of 17/29 phenotypes observed with bridging collateral manipulation. Previous work also showed that dSPN bridging collaterals 1 provide monosynaptic connections to a small population of ChAT neurons located at the caudal-ventral GPe border 2 near the basal forebrain (BF) 61 . The role of these neurons in behavior is, however, unknown. We determined whether 3 GPe ChAT neurons are also motor-suppressing and could mediate the motor effects of bridging collaterals. ChAT-cre 4 mice were bilaterally injected with a flexed AAV expressing ChR2 or YFP into the caudal-ventral GPe and implanted 5 with GPe optic fibers (Fig. 9A-B). We stimulated ChAT neurons at 10 and 20 Hz in the open field, firing frequencies that 6 BF ChAT neurons reach in vivo 62 . Optogenetic stimulation of ChAT neurons did not affect locomotion speed in the open 7 field (Fig. 9G) or rotarod motor performance (Fig. 9H). Since GPe ChAT neurons project to the cortex, reticular thalamus 8 and amygdala 61 , they may mediate other behavioral functions of dSPNs and warrant further study. Altogether, these 9 data indicate that dSPN bridging collaterals support motor function via GABAergic inhibition of their Npas1 targets in 10 the GPe, rather than via their ChAT target cells.

13
Within the vertebrate brain, motor behavior is controlled not by one descending projection but by the coordinated activity Here we address this question focusing on striatal dSPNs by selectively recording or inhibiting dSPN axon collaterals in the GPe. We find that dSPN axon collaterals in the GPe bear an axonal copy of motor signals sent to the SNr. This 23 projection supports motor control by inhibiting its own GPe circuit involving motor-suppressing Npas1 neurons in vivo.

24
We propose a model by which dSPN GPe terminals act in concert with the canonical terminals in the SNr to control 25 motor function via a striatopallidostriatal subcircuit, which we term the striatopallidal "Go" pathway. Specifically, we find

18/29
These data converge onto an overarching model in which the direct pathway controls motor function via its 1 simultaneous influence on 3 brain regions: the classical targets of dSPNs (i.e., EP, SNr) and the GPe. Here our pathway-2 specific inhibition manipulations showed that dSPN GPe terminals are necessary for motor control in healthy mice.

3
These results argue against the notion that the different dSPN outputs are redundant, rather suggesting they act in a 4 complementary fashion, influencing motor output by leveraging their own distinct subcircuits 38 . Together with recent 5 studies using slice physiology, whisker stimulation or optogenetic manipulations under anesthesia 15,17,19 , our findings 6 indicate that dSPN GPe terminals control motor function by inhibiting Npas1 neurons during behavior. Thus, via this 7 monosynaptic connection, dSPNs are granted unique access to a neural population, Npas1 neurons, with potential 8 broad impacts on striatal outflow 21-23,67 . Since Npas1 neurons target both dSPN and iSPN dendrites in the striatum,

9
Npas1 neurons have been proposed to work as a gain control to filter weak synaptic inputs to the striatum 21 . Here we 10 speculate that bridging collaterals could act as a fine-tuning knob for this gain control, inhibiting Npas1 neurons to

21
Because axon collaterals are abundant throughout the central and peripheral nervous systems 63,64 and profuse within the BG itself 8,11 , it is intriguing to speculate on the potential advantages of axon collaterals as opposed to separate 23 neural populations. One of the most studied functions of axon collaterals is their ability to send an "efference copy" of 24 motor instructions to two brain regions simultaneously, one to instruct motor output and one to inform sensory brain

35
Similarly, speculating that dSPN-Npas1-dSPN circuits are topographically organized, feedforward loops generated 36 among these neurons could delineate the spatial boundaries of recruited striatal dSPNs. Understanding the impact of 37 dSPN-Npas1 circuits on the spatial and temporal organization of striatal activity could be addressed in future work.

39
While we did not analyze the specific information encoded by dSPN GPe vs. SNr terminals, we found that the GPe/SNr 40 axonal copy is not exact, since correlation coefficients between GPe and SNr presynaptic terminal activity were 41 significantly reduced during the running phase of a rotarod motor task. This modulation dependent on the task condition 42 suggests the existence of local regulatory mechanisms specific to the GPe or SNr that may allow region-specific 43 divergence of activity in a behaviorally relevant manner. This could arise from differential presynaptic regulation. For 1 from Wiley and Springer. Remaining authors declare no competing interest.

3
Materials & Correspondence 4 Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead 5 contact, Christoph Kellendonk (ck491@cumc.columbia.edu).

1
Reagents 2 Reagent and equipment information is given in Supplementary Table S2     Synaptophysin-GFP+ terminal optical density in the region was quantified in ImageJ using two random counting frames 2 per section positioned above the GPe or SNr (average 5.5 sections/brain region/animal); values reported are in 3 percentage of striatal optical density as in 6 . For Synaptophysin-GCaMP8s vs. jGCaMP7s bouton vs. axon quantification, 4 brains were kept unstained to compare native GFP fluorescence (GFP contained in GCAMP); optical density was 5 calculated by selecting ROIs of boutons and axons (average 3.5 ROIs/section/brain region; 3.5 sections/animal); values 6 reported are ratios of boutons vs. axons in same section.

8
Fiber photometry during behavior 9 Fiber photometry equipment was set up using two 4-channel LED drivers connected to two sets of a 405 and a 465 nm

27
Opto-photometry during behavior in closed loop 28 Animals were tested in the dark phase and mildly food-deprived to elicit locomotion. In a first experiment, dSPNs were 29 stimulated in closed-loop based on Npas1 activity. Animals explored the open field for 1h or when 12 trials per condition 30 were completed. The optogenetic LED was driven at various powers (0.5, 1, 2mW), duration (3, 10s) and frequencies 31 (10, 20Hz). At least 8 trials per condition were completed. Optogenetic stimulation was triggered when Npas1 calcium 32 activity (dFF) reached a peak and was above a minimal threshold (estimated post-hoc to be around 2-zscores) for 33 250ms. This was done using a custom-written Synapse program. dFF was estimated online using the following equation:   5-s) was triggered only after a brief rest period (5-s) followed by a longer 'high mobility' bout (average speed threshold acceptable radioactivity and EtOH levels, but the same volume (300 nL) was used. 30 min after infusion (=time of 23 behavioral testing), brains were collected, flash frozen and stored at -80C. Tissue was sectioned (20 μm) on a cryostat 24 and thaw mounted onto ethanol-washed slides. Slides were air dried overnight, placed in a Hypercassette™ and 25 covered with a BAS-TR2025 Storage Phosphor Screen. Slides were exposed to the screen for 12-14 days and imaged 26 using a phosphorimager (Typhoon FLA 7000).

28
In vivo electrophysiology 29 Anesthetized mice (chloral hydrate) were locally infused with Saline or CNO (300 nL, 1 mM) at an average rate of intertrial interval (maximal test duration 90-min). The long intertrial interval was necessary to allow eOPN3 to recover 16 as shown in 58 , which we also observed here (Fig. 7D). Stimulation was done in a closed-loop fashion using a custom-17 written Anymaze protocol and the Anymaze AMi-2 optogenetic interface. The trial started only after a brief rest period 0.03m/s for 5-s). See increase in locomotion in GFP and eOPN3 mice in the pre-stimulation epoch (Fig. 7D).

37
Event duration was computed from event start to peak maxima (for movement 'onsets') or peak minima (for movement 38 'offsets') detected by looking for the closest local maxima/minima to the event in the 0-5 sec timewindow. Pearson 39 correlations were computed between individual body parameters (jump duration, from peak to the next peak) and dFF Relative body center mouse speed was calculated by substracting the baseline speed calculated in the pre-stimulation 3 epoch (-60 to -5 sec). DeepLabCut body points with likelihood <0.9 were interpolated with inpaint_nans function in falling into the other categories. Frame category were used to create behavioral maps in the 'pre-stimulation' and 'during-7 stimulation' periods and percent time in each category quantified. The arena was subdivided into a periphery zone (most 8 outer 10 cm square) vs. center to classify frames (periphery or center) as a proxy for anxiety.

11
Full statistical results including p-values and F-values can be found in Supplementary Table 3