Dissociable roles of central striatum and anterior lateral motor area in initiating and sustaining naturalistic behavior

SUMMARY

Understanding how corticostriatal circuits mediate behavioral selection and initiation in a naturalistic setting is critical to understanding behavior choice and execution in unconstrained situations. The central striatum (CS) is well poised to play an important role in these spontaneous processes. Using fiber photometry and optogenetics, we identify a role for CS in grooming initiation. However, CS-evoked movements resemble short grooming fragments, suggesting additional input is required to appropriately sustain behavior once initiated. Consistent with this idea, the anterior lateral motor area (ALM) demonstrates a slow ramp in activity that peaks at grooming termination, supporting a potential role for ALM in encoding grooming bout length. Furthermore, optogenetic stimulation of ALM-CS terminals generates sustained grooming responses. Finally, dual-region photometry indicates that CS activation precedes ALM during grooming. Taken together, these data support a model in which CS is involved in grooming initiation, while ALM may encode grooming bout length.

Graphical abstract

In brief

Corbit et al. studied patterns of activity in corticostriatal circuits during naturally occurring complex behavior in mice. In contrast to models in which cortex leads striatum, these findings suggest that central striatum activity occurs first, at grooming initiation, while anterolateral cortex activity peaks later and is associated with grooming cessation.

INTRODUCTION

To navigate our world, it is essential to string together series of movements into complete actions that can be used to attain goals. Carrying out actions in the appropriate context, for the appropriate amount of time, is essential for adaptive behavior. However, although much is known regarding the neural substrates underlying these processes,^1–5 the mechanisms underlying both appropriate initiation and sustainment of spontaneous actions are still incompletely understood.

Prior work has implicated striatal circuits in behavioral selection. The activity of dorsal striatal spiny projection neurons (SPNs) correlates with the first and/or last action in a sequence of movements in highly trained animals, suggesting striatum encodes movement sequences as chunked actions.^4–6 Furthermore, lesions in the lateral or medial regions of dorsal striatum cause well-trained animals to exhibit goal-directed or habitual behavior, respectively, in a lever press task.^7,8 These data suggest that specific regions of dorsal striatum and/or their upstream inputs are important for modulating the extent to which behavior is intentional (e.g., value based) or automatic (e.g., value independent).

While these operant-based paradigms have yielded great insight into the striatum’s role in learning and performing trained actions, investigation of naturalistic behavior is important for understanding the neural mechanisms of key behaviors for survival. For instance, mice spontaneously engage in rearing and locomotion to explore their surroundings, obtain food, and avoid threats. Rodents also perform grooming and nest building to maintain hygiene and care for pups. Unfortunately, despite the importance of these behaviors, only a few studies have investigated the striatum’s role in behavioral selection in unrestrained, untrained settings. Using both fiber photometry^2,9 and one-photon microendoscopy,^10,11 several of these studies demonstrated activation of dorsal striatal SPNs prior to or at the onset of locomotion,^2,9–11 turning,^2,9,11 and rearing.^9,11 These results suggest that diverse types of spontaneous movements are associated with increased activity in dorsal striatum, although specific SPN temporal-activation profiles for different behaviors in relation to movement onset demonstrate some variation. However, because these studies were associational, and prior causal studies focused solely on the dorsal striatum’s role in locomotion,¹² further work is needed to determine whether particular striatal activity patterns can directly generate species-typical behaviors.

The striatum exhibits topographical cortical inputs following a motor-limbic dorsal-to-ventral gradient.^13,14 Interestingly, this positions the central striatum (CS) at a crucial nexus of corticostriatal inputs along this topography,^15,16 making it well suited to link motivational factors and sensorimotor responses. Consistent with this idea, in prior work, we showed that CS receives projections from both the lateral orbitofrontal cortex (LOFC) and the anterior lateral motor area (ALM),¹⁷ regions implicated in behavioral flexibility¹⁸ and motor preparation and tongue movements,^19,20 respectively. Furthermore, early studies showed that activating central striatal regions via picrotoxin disinhibition caused tic-like behavior in rodents.²¹ More recent work has replicated this effect in CS (and dorsolateral striatum) and further showed that tic production depends on striatal, and not cortical, activity.²² Production of tics by CS disinhibition highlights a potential role for this region in the initiation of movement fragments.

Despite this convergent evidence hinting at an important role for CS in the selection of spontaneous behavior, there is little investigation of this hypothesis in vivo. One prior study showed CS hyperactivity at baseline and during persistent conditioned grooming behavior in the SAPAP3-KO mouse model of compulsive behavior,²³ suggesting a correlation between abnormal CS activity and abnormal behavioral selection. Burguière and colleagues also showed that activating LOFC-CS inputs was sufficient to reduce this grooming behavior, suggesting LOFC serves to inhibit abnormal behavior generated in CS.²³ However, CS inputs from ALM, a region associated with (1) tongue movements important for grooming and (2) preparation, sustainment, and sequencing of trained behavior,^{19,20,24–27} have not yet been examined. We therefore investigated whether CS and its inputs from ALM mediate spontaneous naturalistic behavior using fiber photometry, optogenetics, and microendoscopy for in vivo observation and manipulation.

Surprisingly, we found a selective increase in CS activity at the initiation of grooming but not at the onset of other spontaneous behaviors. Direct optogenetic stimulation of CS SPNs (to mimic this activity increase) evoked immediate-onset grooming-related fragmented movements that were short in duration and distinguishable from normal grooming behavior using a support vector machine (SVM) classifier. Using fiber photometry, we found that ALM is also selectively activated during grooming behavior. Surprisingly, peak ALM activity correlated with grooming bout length, suggesting ALM may encode duration of sequenced behaviors. In addition, stimulation of ALM terminals in CS caused long-latency evoked grooming bouts resembling naturalistic grooming when using an SVM classifier. Finally, dual-color, dual-region photometry revealed that CS activation precedes ALM activation at naturalistic grooming bout initiation. Taken together, these results suggest CS selects or initiates naturalistic grooming behavior, while ALM could play a role in sustaining effective bouts.

RESULTS

Increased activity in CS is specifically associated with groom start

To investigate the role of CS in naturalistic behaviors, we unilaterally injected AAV-jRGECO1a into CS and implanted a 200 μm fiber photometry probe 300 μm above the injection site (Figures 1A and S1). Bulk calcium activity was recorded while animals performed unconstrained, spontaneous behaviors in an observation chamber. CS calcium activity was then analyzed at the onset of three distinct spontaneous behaviors comprising a majority of non-immobility time: grooming, rearing, and locomotion (Figure 1B). Strikingly, CS showed a rapid increase in activity at grooming onset (Figure 1C, repeated-measures [RM]-ANOVA, time × virus interaction, p = 0.03). To further delineate which aspect of grooming is encoded by CS, we separated bouts into face grooming (primarily involving paw movements) or body grooming (primarily involving tongue/body movements). This analysis demonstrated that increased CS activity at grooming onset is primarily associated with body grooming (Figure 1D, RM-ANOVA, time × behavior-type interaction, p = 0.002). In contrast, when we aligned activity to rearing onset, we did not observe increased CS activity, instead seeing a significant reduction in activity (RM-ANOVA, time × virus interaction, p < 0.001, Figure 1E). Similarly, CS activity did not increase at locomotion onset (RM-ANOVA, time × virus interaction, p = 0.55, Figure 1F). Together, these data suggest CS activity may specifically be related to initiation of grooming, particularly movements related to body grooming such as licking the flank or torso, but do not prove a causal role for CS in generation of grooming.

Figure 1. CS calcium activity selectively increases at grooming onset.

(A) AAV8-syn-jRGECO1a was injected unilaterally into the central striatum (CS). After ≥3 weeks recovery, the mice were handled for 30 s/day for ≥3 days and then habituated for ≥3 days to scruffing, cable attachment, and the bottom-up observation chamber. The mice were then placed in the chamber, and CS bulk calcium activity was recorded using fiber photometry for 30 min during spontaneous behavior.

(B) (Left) CS calcium activity example trace aligned to three distinct behaviors: grooming, rearing, and locomotion. (Right) Proportions of each behavior subclass in six example mice. Number of behavior bouts (average ±SEM): grooming, 68.5 ± 15.4; rearing, 24.2 ± 4.7; and locomotion, 208.5 ± 22.5.

(C) Normalized CS calcium activity aligned to grooming onset (jRGECO1a n = 9 [teal], mCherry control n = 5 [gray], two-way RM-ANOVA, time × virus interaction, p = 0.03; significant post hoc contrasts [p < 0.05] for bins 1–3, 4–6, and 7–8 s relative to baseline [1 s before onset]).

(D) Normalized jRGECO1a activity for grooming subtypes: face (light teal) and body (dark teal) (two-way RM-ANOVA, time × behavior interaction, p = 0.002; significant post hoc contrasts [p < 0.05] for bins 0–1 and 6–8 s relative to baseline [1 s before onset]). mCherry traces for face and body grooming are shown in light and dark gray, respectively.

(E) Normalized CS calcium activity aligned to rearing (jRGECO1a n = 9 [teal], mCherry control n = 5 [gray], two-way RM-ANOVA, time × virus interaction, p < 0.001; significant post hoc contrasts [p < 0.05] for bins 2–8 s relative to baseline [1 s before onset]).

(F) Normalized CS calcium activity aligned to locomotion onset (jRGECO1a n = 9 [teal], mCherry control n = 5 [gray], two-way RM-ANOVA, time × virus interaction, p = 0.55).

See Figure S1.

Optogenetic stimulation of CS evokes partial grooming movements

To test whether increased activity in CS could directly cause grooming behavior, we bilaterally injected hSyn-ChR2-EYFP or EYFP control virus into CS of wild-type (WT) mice and implanted optical fibers above the injection site (Figure 2A). Because our photometry data suggested that grooming-related CS activity entails an increase from baseline, we developed a stimulation paradigm that enabled the capture of sufficient trials during which animals were quiescent and therefore presumably at baseline CS activity levels (10 s constant light pulses; pseudorandom intertrial interval [30 ± 5 s]) (Figure 2B). To control for potential heating effects from prolonged stimulation,²⁸ we performed identical laser stimulation in EYFP-virus controls.

Figure 2. CS optogenetic stimulation evokes grooming-like movements with short latency.

(A) AAV5-syn-ChR2-EYFP was bilaterally injected into CS and fiber optics were implanted above the injections.

(B) Fifty-one stimulation trials (10 s constant light) were separated by 25 to 35 s pseudorandom intertrial intervals with 5 s jitter. Grooming/grooming-related movement fragments were manually scored and aligned to laser presentations; peri-laser grooming probability was calculated.

(C) Probability of grooming/grooming-related movements in ChR2 mice (n = 9) was significantly greater than in EYFP mice (n = 6) at laser pulse onset (two-way RM-ANOVA, significant time × virus interaction, significant in bins 1–3.5 s after laser onset, p < 0.01).

(D) SLEAP was used to track body posture (upper left). After the positions of five body parts were obtained, position data were rotated around the body axis to align every frame’s data (bottom left). Manual scoring was used to isolate bouts of face, body, and pseudo-grooming. Two bouts/grooming type are shown for two example mice (right; one bout/plot). Mouse 4821 showed pseudo-grooming that looked similar to body grooming via SLEAP; mouse 4825 showed pseudo-grooming that looked similar to face grooming.

(E) The SVM classifier was trained/tested on the ability to differentiate face, body, and pseudo-grooming based on a feature set extracted from body position data. The classifier distinguished three types of grooming with equally good performance for each class (one-way RM-ANOVA, p = 0.062).

(F) A separate classifier was trained to distinguish face grooming, body grooming, and locomotion (Loc) with the same extracted feature set. Pseudo-grooming instances were then entered into the trained model, and percentage of pseudo-grooming instances classified as each of the three trained classes is shown for six responding animals (purple, body-grooming-type responders; yellow, face-grooming-type responders).

Figure S2 shows an additional classifier explanation; Figure S3 shows an additional characterization of optogenetically evoked behavior.

The primary response to CS stimulation was short-latency initiation of grooming-like movements (Figure 2C and Video S1)−i.e., evoked behavior resembled components of grooming behavior, such as a paw swiping the face, but appeared incomplete (i.e., the paw swipe never touched the face). To quantify these evoked behaviors, which were manually labeled as “pseudo-grooming,” we used automated postural tracking (SLEAP²⁹) combined with manual behavior annotation (Figure S2). Posture tracking of forepaws, snout, body center, and tail base aligned to manually labeled behavior bouts demonstrated that naturalistic face grooming and body grooming consisted of distinct body-part positions, but there was some variability in position across bouts. In contrast, evoked “pseudo-grooming” was more stereotyped within a mouse (although different across mice, Figure 2D).

To quantify the relationship between evoked pseudo-grooming and naturalistic grooming, we first identified a set of key postural features in the tracking data based on a priori knowledge of the visual appearance of grooming behavior: paw distance to snout, snout distance to body center, snout angle off body axis, snout angular velocity, and overall body velocity (Figure S2). We then used manual annotation data to label every time point in the tracking data as face grooming, body grooming, pseudo-grooming, or not grooming.

Using these annotated data, we first investigated the degree of difference between evoked grooming-like movements and naturalistic grooming by training an SVM classifier on data from face grooming, body grooming, and pseudo-grooming. To assess model performance, we used individual class (behavior) F1 scores, an aggregate score of precision and recall. In a mouse that exhibited all three behaviors, we observed statistically equal classification performance (face grooming, F1 = 0.860 ± 0.035; body grooming, F1 = 0.784 ± 0.061; pseudo-grooming, F1 = 0.923 ± 0.016; one-way RM-ANOVA, main effect p = 0.062). We did observe a trend for better classification of pseudo-grooming relative to body grooming (post hoc t test, p = 0.057, Figure 2E). Given that these differences were not statistically significant at our chosen alpha level, we concluded that all three types of behaviors were classified equally well, with an average F1 score of 85%. Thus, the three types of grooming-related movements are distinct and identifiable both by a human observer and by a machine learning model.

Next, to evaluate how similar the evoked behavior was to natural behavior, we trained a new classifier to distinguish face grooming, body grooming, and locomotion in the same mice. We then tested the model on manually annotated pseudo-grooming instances (which the model was not trained on) and calculated the proportion of pseudo-grooming instances classified as each type of behavior that the model was trained on (recall as face grooming, body grooming, or locomotion). This analysis allowed us to determine which frequently occurring natural behavior the pseudo-grooming most resembled. For each animal that showed evoked pseudo-grooming, we visually determined whether the behavior appeared closer to face grooming (Figures 2D–2F, yellow) or body grooming (Figures 2D–2F, purple) by looking at joint positions and videos. Consistent with our visual observations, animals whose evoked behavior most resembled face grooming showed a high proportion of pseudo-grooming classified as face grooming (Figure 2F, yellow lines), whereas animals whose evoked grooming resembled body grooming showed a high proportion of pseudo-grooming classified as body grooming (Figure 2F, purple lines). We did observe that there was a high proportion of pseudo-grooming classified as locomotion in one animal, presumably because the animal’s pseudo-grooming resembled a turn toward the flank, which could resemble turning during walking. Taken together, these two classifiers demonstrate that the observed evoked behavior was similar to specific types of natural grooming (Figure 2F), but still quantifiably distinct from fully naturalistic grooming (Figure 2D).

We next investigated the temporal dynamics of pseudo-grooming. Laser-evoked grooming movements were initiated with a short latency after CS stimulation, averaging +1.03 s (SEM = 0.95 s). Interestingly, bouts of grooming-like movements evoked by direct CS stimulation did not last throughout the 10 s stimulation period (average bout length 4.07 ± 0.13 s). As a proxy for reliability of stimulation-evoked behavior within a mouse, we also assessed grooming probability as “# stim trials with evoked behavior/# total stim trials.” ChR2 mice had a significantly greater reliability of laser-evoked grooming or grooming-like behaviors (44.0% ± 0.1%) than EYFP controls (0.13% ± 0.04%, t test, p = 0.04) This effect was specific for grooming-like movements, as we did not see a similar increase in rearing or locomotion during CS stimulation (Figure S3). Note that we did not see a behavioral effect from pulsed (20 Hz) stimulation (Figure S3), consistent with our observations during simultaneous optogenetic stimulation and single-cell calcium imaging (Inscopix) showing that constant stimulation was more effective at recruiting CS activity than 20 Hz pulses (Figure S3).

Together, these data demonstrate that sustained CS stimulation selectively evokes short-latency grooming-related movements. However, the absence of complete grooming bouts after CS stimulation, and the fact that evoked grooming-like behavior was not sustained throughout stimulation, suggested that an alternative brain region is necessary for appropriate sequencing of grooming-related movements into sustained bouts.

ALM shows increased activity specifically at grooming onset

Prior work has demonstrated that anterior M2/ALM, one of the major cortical inputs to CS,¹⁷ is associated both with appropriate sequencing of movements during trained tasks and with generating licking movements.^20,24,30 ALM may therefore be uniquely suited to guiding selection of spontaneously generated sequenced behaviors such as grooming. To investigate the relationship between ALM activity and generation of untrained behaviors, we performed unilateral fiber photometry to record calcium activity during grooming, rearing, and locomotion in spontaneously behaving mice (Figures 3A, 3B, and S4).

Figure 3. ALM calcium activity increases selectively at grooming behavior onset.

(A) AAV9-GCaMP6m was injected unilaterally (ALM), and photometry fiber optics were implanted 0.10 mm above injection. Habituation was as in Figure 1A.

(B) ALM photometry example trace; grooming events are highlighted in gray.

(C) Normalized ALM calcium activity aligned to grooming onset (GCaMP6m n = 9 [magenta], EYFP control n = 6 [gray], two-way RM-ANOVA, time × virus interaction, p < 0.0001; significant post hoc contrasts [p < 0.05] for bins 4.5–6 s relative to baseline [1 s before onset]).

(D) Normalized GCaMP6m activity for face grooming (light magenta) and body grooming (dark magenta). EYFP traces for face and body grooming are in light and dark gray, respectively (time × virus interaction: face grooming, p = 0.11; body-grooming, p < 0.0001, but no significant bins after post hoc Sidak’s correction for multiple comparisons).

(E) Normalized ALM calcium activity aligned to rearing onset (GCaMP6m n =9 [magenta], EYFP control n = 6 [gray], two-way RM-ANOVA, time × virus interaction, p = 0.89).

(F) Normalized ALM calcium activity aligned to locomotion onset (GCaMP6m n = 9 [magenta], EYFP control n = 6 [gray], two-way RM-ANOVA, time × virus interaction, p = 0.174).

Figure S4 shows additional fiber photometry characterization.

In contrast to the proposed role of ALM and supplementary motor cortices in action preparation,^20,31,32 we found that ALM activity increased gradually at the onset of, not prior to, grooming behavior (Figure 3C; RM-ANOVA, time × virus interaction, p < 0.0001). When we separated calcium traces based on grooming subtype (face vs. body), there were no overall significant differences between subtypes (Figure 3D; RM-ANOVA, time × groom subtype interaction, p = 0.99; time main effect, p < 0.0001). Similar to our observations in CS, we did not see increased calcium activity at rearing (Figure 3E; RM-ANOVA, time × virus interaction, p = 0.89) or locomotion bout onset (Figure 3F; RM-ANOVA, time × virus interaction, p = 0.051). Thus, these photometry data suggest ALM activity is preferentially associated with grooming compared to other spontaneous behaviors.

Peaks in ALM activity correlate with grooming cessation

Close examination of temporal dynamics of grooming-related calcium activity in ALM revealed that the average fluorescence peak occurred several seconds after grooming onset (3.21 ± 2.77 s post-onset). This long latency to peak fluorescence suggested that, in contrast to predictions, ALM activity may be related to termination, and not planning, of a grooming bout.

To investigate this hypothesis, we first inspected ALM photometry data on a trial-by-trial basis, comparing calcium activity on a given grooming trial with timing of initiation and termination of grooming (Figure 4A). Interestingly, ALM activity typically increased at grooming bout initiation and remained elevated until that bout was terminated (see example traces in Figure 4B). To further probe this phenomenon, we separated grooming trials into quartiles (2 s intervals) and calculated average ALM calcium activity traces from each quartile (Figure 4C). This analysis revealed a pattern of longer time to peak fluorescence and increased peak height as grooming bout duration increased (Figure 4C). When we restricted these plots to face or body grooming (Figure S4), we observed similar but weaker patterns of activity, likely due to reduced bout numbers. However, when we instead aligned activity during all grooming bouts to grooming cessation, we observed that peak activity in each quartile occurred shortly before groom stop, with ramping activity occurring during the grooming bout (Figure 4D).

Figure 4. ALM calcium activity correlates with grooming bout length.

(A) ALM normalized fluorescence heatmap for individual grooming trials (baselined to 3 s prior to groom onset). Yellow star: each trial’s grooming offset.

(B) Individual ALM representative traces show ramping between groom onset (dashed line) and offset (gold hash). Peaks were automatically identified using MATLAB’s “prominence” measurement (see STAR Methods) and labeled in purple. Units: normalized dF/F (divided by whole-trace standard deviation).

(C) Separation of grooming bouts into four quartiles (0–2 s [n = 103 bouts], 2–4 s [n = 53 bouts], 4–6 s [n = 29 bouts], and 6–8 s duration [n = 10 bouts]) reveals a pattern of increasing peak amplitude and trend toward increased peak time in longer grooming bouts. Each dotted line marks average bout length for the corresponding quartile (denoted by different shades of pink). Bouts of >8 s were excluded from the plot because of insufficient n (<5 bouts for each remaining 2 s bin).

(D) Aligning the same bout quartiles to grooming offset shows that peaks occur shortly before grooming offset, with gradual increase in activity leading up to the peak.

(E and F) Detected peak time (E; Pearson correlation, R² = 0.40, p = 5.7 × 10⁻²³) and amplitude (F; Pearson correlation, R² = 0.13, p = 2.0 × 10⁻⁷) significantly correlate with bout duration on trial-by-trial basis. The x axis is cropped at 10 s for clarity, although bouts of >10 s were included in the analysis.

Figure S4 shows additional fiber photometry characterization.

To better quantify this phenomenon, we identified the first prominent peak after groom onset in each trial and measured time to peak and amplitude (see STAR Methods). Consistent with our quartile analysis, bout length was significantly correlated with ALM fluorescence peak time, suggesting that ALM grooming-related activity peaks at groom offset (Figure 4D, R² = 0.40, p = 1 × 10⁻¹⁴). Interestingly, when we statistically compared peak times grouped within each quartile, we found only trend-level significance (p = 0.09, one-way ANOVA on quartiles’ peak times), suggesting that “time to ALM activity peak” correlates with grooming bout length on a continuum rather than in discrete time bins. We also observed a weaker but significant correlation of bout length with peak amplitude (Figure 4E, R² = 0.13, p = 4.8 × 10⁻⁵). In addition, we found a significant difference between peak amplitudes across the four quartiles (p = 5 × 10⁻⁵, one-way ANOVA, significant post hoc comparisons between quartile 1 [0–2 s] and quartiles 3 [4–6 s] and 4 [6–8 s]). These data suggest that both peak time and amplitude of ALM activity may be good predictors of grooming bout length. In contrast, only a trend-level correlation was observed between grooming bout length and CS calcium activity, and the ALM correlation was significantly stronger (Figure S4). To determine whether ALM calcium activity features could decode bout length, we ran a multiple linear regression on ALM peak time and amplitude during grooming bouts. Using these features, we predicted bout length with an average error of 0.92 s. Together, these data indicate that time to peak activity in ALM correlates with length of naturally occurring grooming bouts, suggesting that ALM activity has potential to play a role in sustaining grooming bouts but is not directly responsible for planning or initiating grooming behavior.

ALM-CS stimulation evokes more natural grooming behavior at long latency

These combined data demonstrate that CS activity is related to grooming initiation, while ALM activity is significantly associated with grooming bout length, suggesting ALM could be responsible for sequencing and sustaining fragments of grooming-related movements generated via CS activation. According to this model, activating ALM-CS terminals should initiate sustained grooming bouts, in contrast to grooming fragments generated by direct CS activation. To test this hypothesis, we bilaterally injected ChR2-EYFP (or EYFP control virus) into ALM, implanted optical fibers, and performed ALM-CS terminal stimulation with a 20 Hz pulsed stimulation paradigm (Figures 5A and S5 and STAR Methods).

Figure 5. Stimulation of ALM-CS terminals causes full grooming with latency paralleling timing of CS activation.

(A) AAV2-syn-ChR2-EYFP or control AAV2-syn-EYFP was bilaterally injected into ALM; fiber optics were simultaneously implanted in CS.

(B) Heterogeneous responses observed following bilateral optogenetic stimulation (ChR2 n = 15).

(C) Splitting ChR2 mice into responders (n = 10) and non-responders (n = 5) revealed a significant difference between responders and EYFP controls (n = 8, two-way RM-ANOVA; time × virus interaction, p < 0.0001, p < 0.05 [Sidak’s correction for multiple comparisons] for time bins 15–20 s). No difference was seen between non-responders and EYFP controls (two-way RM-ANOVA, time × virus interaction, p = 0.009, but no significant bins with Sidak’s correction).

(D) Aligned joint position (see Figures 2 and S2) examples for two mice, showing snout, body center, tail base, and forepaws from bottom-view camera for face-, body-, and evoked-grooming instances. Evoked grooming sometimes resembled face grooming and other times resembled body grooming.

(E) F1 scores from SVM classifier trained to distinguish face, body, and evoked grooming. Evoked grooming classification had significantly lower F1 scores than face (p = 0.004) and body grooming (p = 0.006).

(F) Classifier trained to identify face/body grooming (grouped as natural grooming [AsNaturalGroom]), stereotypies (AsStereo), and locomotion (AsLoc) was tested on held-out evoked-grooming data and predicted evoked grooming to be “natural grooming” significantly more than stereotypies (p = 0.005) or locomotion (p = 0.000002).

(G) nVoke 1.0 microscope (Inscopix) was used to simultaneously stimulate ALM terminals in CS (AAV8-syn-Chrimson) and record CS single-cell calcium activity (GCaMP6m).

(H) Heatmap of responding cells’ average activity (20 trials). ALM terminal stimulation evoked long-latency CS activation (152/827 cells from nine mice showed significant activation). Yellow dots, average onset activation for each cell, calculated over 20 LED trials.

(I) Average response for all activated cells shows temporal profile similar to ALM-CS evoked-grooming response (C).

Figures S5 and S6 show additional characterization of ALM-CS terminal stimulation effects.

In contrast to direct CS activation (Figure 2), ALM-CS stimulation yielded more natural-looking grooming bouts (Video S2). However, laser-evoked grooming was noticeably heterogeneous between mice (Figure 5B). To better capture this phenomenon, we classified mice into grooming “responders” (n = 10) or “non-responders” (n = 5) based on ≥2 standard deviations for increased grooming probability over the baseline mean probability during laser stimulation periods. There was a significant difference in grooming probability during laser stimulation between responders and EYFP controls (Figure 5C, RM-ANOVA, time × virus interaction, p < 0.001), but not between non-responders and EYFP controls (Figure 5C, RM-ANOVA, time × virus interaction, p = 0.009, no significant bins with Sidak’s multiple comparison correction). In addition to grooming-like behavior, a “stereotypy” behavior consisting of repetitive licking of the floor or walls was observed in 6/15 mice (5 of whom were grooming responders; Figure S5). However, no increases in rearing or movement velocity were observed (Figure S5).

To quantify whether ALM-CS stimulation-evoked behavior was distinguishable from naturalistic grooming, we implemented another SVM using manual annotation of naturalistic and evoked grooming and automated position tracking (Figures 5D–5F). In contrast to stereotyped behavior evoked by CS stimulation, ALM-CS terminal stimulation evoked diverse grooming-related behaviors within an individual mouse (Figure 5D). The classifier was able to distinguish natural face and body grooming with good (>80%) F1 scores (face grooming F1 = 0.84 ± 0.02, body grooming F1 = 0.82 ± 0.02), suggesting each of those behaviors appears distinct. However, F1 scores for evoked grooming were significantly lower, indicating that evoked grooming was not as accurately classified by the model as a separate behavioral entity (evoked grooming F1 = 0.70 ± 0.04, one-way ANOVA, p = 0.008; post hoc paired t tests show lower F1 scores for evoked grooming compared to face [p = 0.009] and body grooming [p = 0.043], Figure 5E). This analysis supports the notion that ALM-CS terminal stimulation-evoked grooming contains instances that resemble both naturalistic face and body grooming, contrasting with our observations from the CS stimulation classifier (Figure 2E).

To further understand how similar evoked grooming was to natural grooming, we trained a new classifier to distinguish face grooming, body grooming, stereotypies, and locomotion based on our set of posture features. We then tested the model’s predictions on held-out evoked-grooming data to see which behaviors the evoked grooming resembled. The ALM-evoked grooming resembled naturalistic grooming bouts−i.e., in a given mouse we observed evoked movements that resembled face and body grooming. Thus, after classification, we grouped face and body grooming into one “natural grooming” category for further assessment (Figure 5F). We found that the model predicted evoked grooming to be natural grooming more frequently than it predicted evoked grooming to be stereotypies or locomotion (Figure 5F, one-way RM-ANOVA, F(18) = 27.5, p =3 × 10⁻⁶; post hoc paired t tests, groom vs. stereotypies, p = 0.005; groom vs. locomotion, p =2 × 10⁻⁶), supporting our visual observations that evoked behavior resembled natural grooming.

Having demonstrated that behavior evoked by ALM-CS stimulation resembled natural grooming, we aimed to further characterize heterogeneity in grooming responses across animals. We first quantified ALM projection fluorescence in CS and observed a weak, non-significant correlation of peak grooming probability with strength of ALM projection fluorescence in CS (R² = 0.17, p = 0.16, Figure S6). Although this correlation was not significant, this analysis prompted us to further investigate how ALM terminal activation recruits downstream CS neurons. Unlike CS stimulation-evoked behavior, ALM terminal stimulation-evoked behavioral responses had a surprisingly long onset latency (average latency from laser onset to evoked-grooming onset = 11.23 ± 4.24 s), suggesting ALM-evoked activation of CS may also occur at long latency. To directly investigate how ALM terminal stimulation affects CS activity, we employed unilateral simultaneous optogenetic stimulation of ALM terminals and endoscopic single-cell calcium imaging through a GRIN lens in CS (Figures 5G and S6). Because stimulation was unilateral, it was not surprising that we did not see a grooming response in these mice. However, consistent with our behavioral findings with bilateral stimulation, activation of ALM terminals did not immediately evoke calcium activity in most CS cells (Figures 5H, 5I, and S6). Instead, 78% of activated cells had an activation latency of at least 5 s after the onset of light-emitting diode (LED) stimulation (average latency in activated cells = 9.42 ± 0.39 s), similar to what was recently shown in CS during ALM cell body stimulation.³³ The average activity trace of all responding cells (Figure 5I) showed temporal dynamics similar to those of the behavioral response (Figures 5B and 5C). These results suggest that, while artificial activation of ALM terminals can evoke more natural-looking grooming bouts in some circumstances, initiation of grooming behavior may require sufficient downstream activation of CS cells.

ALM and CS display different temporal dynamics related to grooming behavior

Together these results suggest that CS may initiate movements required for grooming, while ALM may be involved in sustaining grooming bouts. To explore this model, we simultaneously recorded calcium activity in ALM (GCaMP6m) and CS (jRGECO1a) during spontaneous grooming behavior (Figure 6A). Visual observation of the two time-varying signals showed similar changes in activity across the two regions (Figure 6A). To characterize the relationship between activity in these two regions, we calculated the normalized cross-correlation between ALM and CS mean-centered calcium signals at grooming onset (Figure 6B). Quantification of the area under the curve for each side showed that the cross-correlation was significantly weighted toward the right (t(174) = −4.61, p = 7.63 × 10⁻⁶), indicating that CS activity leads ALM activity at grooming onset.

Figure 6. CS activation plays a role in grooming initiation.

(A) Dual-color photometry was conducted using GCaMP6m (ALM) and jRGECO1a (CS); fiber optics were implanted unilaterally above the virus injection. Example traces show similarly varying signals in ALM and CS. Units are normalized dF/F (divided by whole-trace standard deviation).

(B) Cross-correlation of simultaneously recorded signals at grooming onset shows that CS precedes ALM (comparison of area under the curve on either side of time = 0, t(174) = −4.61, p = 7.63 × 10⁻⁶).

(C) Derivatives of each signal were calculated to better compare across different indicator kinetics. Aligning the derivative signal in ALM or CS to grooming onset showed that the greatest rate of change in CS occurs before grooming onset, while the greatest change in ALM activity occurs after grooming onset (significant difference in peak time between ALM and CS, WRST, p = 0.004).

Figure S6 shows an elaboration on dual-color calcium data collection/analysis.

This evidence that CS precedes ALM in grooming-relevant activity was unexpected, given previous literature suggesting ALM plays a role in motor planning. Thus, we conducted an additional test on the simultaneous signals to confirm our results. First, we performed a derivative transformation (shown to be an estimation of spiking activity⁹) to directly compare rates of change of each signal. This transformation was then aligned to grooming onset. Paralleling our initial findings, CS exhibited the highest rate of change immediately before grooming−i.e., CS shows its primary activity increase immediately prior to grooming onset (Figure 6C). In contrast, ALM displayed the highest rate of change following grooming onset, consistent with our initial findings that ALM activity starts to increase after groom initiation (Figure 6C). Among simultaneously recorded animals, there was a significant difference between the time of peak change in ALM (1.46 ± 0.59 s after grooming onset) and CS activity (1.99 ± 1.30 s before grooming onset) (WRST [Wilcox rank sum test], p = 0.004). These experiments provide additional evidence that CS activity changes precede ALM activity changes at the initiation of grooming behavior.

DISCUSSION

Our data reveal that the ALM-CS circuit plays a role in evoking and potentially sustaining spontaneous grooming behavior in mice. First, we found that increases in CS calcium activity are associated with initiation of grooming bouts and not other spontaneous behaviors (Figure 1). In addition, mimicking this activity in CS is sufficient to evoke short, fragmented movements resembling grooming behavior (Figure 2). Next, we found that ALM, a prominent input to CS, also shows increased activity at groom onset (Figure 3); however, this activity ramps up during each bout, with a peak that is highly correlated with grooming bout length (Figure 4). Furthermore, optogenetically activating ALM projections in CS evokes natural-looking grooming behavior with a long latency that parallels the time course of CS activation (Figure 5). Analysis of simultaneous recordings in ALM and CS demonstrates that the increase in CS activity leads the increase in ALM activity at grooming onset (Figure 6). Taken together, these data suggest that striatal activation plays a role in initiating naturalistic grooming behavior, but not sustaining it. Furthermore, these findings may support a general model in which striatum initiates spontaneous movements, while reverberating activity in corticobasal ganglia-thalamic loops is required to properly sequence and sustain them. However, alternative models to explain our data are possible, as discussed below.

CS may uniquely integrate motor, cognitive, and limbic information to generate species-typical behaviors essential for survival

Although it is somewhat surprising that activation of CS results in a very specific type of movement, this is reminiscent of classic striatal topographical models from the primate literature. For instance, striatal disinhibition via bicuculline causes different behavioral effects depending on which subregion is injected.³⁴ While disinhibition of the dorsolateral striatum causes hyperactivity and dyskinetic movements of specific body parts, bicuculline injections into more central regions cause stereotypies, including perseverative grooming.³⁴ Thus, our data are broadly consistent with the idea that striatum plays a key role in movement generation and that the particular movement generated may be dictated by striatal topography.

As discussed above, the ability of CS to generate grooming movements suggests this region may be specialized for behaviors critical for survival. Interestingly, in rodents, CS receives input from several key regions in addition to ALM, including fore- and hindlimb motor cortical regions¹⁵ and associative orbitofrontal cortex.^17,23 CS can therefore integrate information about action value with different body parts that can be used to adapt behavioral responses. This unique convergence of inputs could explain why CS may be important for self-regulating ethologically important behaviors, like grooming, that are highly related to an animal’s level of stress and anxiety.³⁵

CS activation generates movement fragments

Our data show that CS activity is sufficient to produce short-latency grooming-like movements (Figure 2), suggesting CS may contribute to initiation of naturalistic grooming behavior via generation of movement fragments. Importantly, these data support the broad possibility that initiations of specific movement fragments are represented by unique activity patterns in CS, consistent with past work showing that disinhibition of CS (via mistargeted dorsolateral striatum [DLS] injections) causes stereotyped tic-like behavior.²² This contrasts with prior work in DLS and dorsomedial striatum (DMS) showing similar activation profiles for several different types of movements, including grooming.^2,9–11 This divergence in results could be reconciled if CS represents specific behavioral repertoires crucial for survival, while DLS/DMS instead represent activation of specific muscles and postures widely used across multiple behaviors.

Interestingly, the evoked-grooming-related movements observed during CS optogenetic stimulation did not last the entire stimulation period (Figure 2). This could theoretically be explained if striatal neurons enter depolarization block after sustained stimulation. However, it has been shown that striatal neurons can maintain elevated firing rates for at least 5 s during in vivo optogenetic stimulation,¹² and our combined optogenetic and calcium imaging experiment suggests that, while evoked activity does diminish after ~2–3 s, activity remains elevated over baseline for the entire stimulation period (Figure S3). Thus, an alternative possibility is that striatal activation is more involved in initiating, rather than sustaining, grooming movements. These data support a model in which activity in another region−possibly ALM (but see discussion of alternative models)−is important for sustaining grooming behavior.

ALM may facilitate smooth, sustained movements through appropriate sequencing of behavior syllables

How do movement fragments become organized into smoothly sequenced behaviors? Our data could support a corticostriatal model of action selection in which ALM contributes to the organization and sustainment of sequences of grooming movements that are initiated by activity in CS. A role for ALM in generating behavioral sequences is supported by previous work showing that supplementary motor regions contribute to trained sequenced behavior.^24,36,37 Consistent with a role for ALM in linking together a series of movements, our data could indicate that sustained ramping activity in ALM encodes the length of a self-initiated grooming bout (Figures 4 and S6). Similar ramping activity has been shown in ALM between a sample and the subsequent lick response in a delay-match-to-sample task.^19,26 Furthermore, supplementary motor regions in humans have activity related to waiting period duration in a timed motor production task.^38–40 However, our results contrast with previous findings that supplementary motor areas (SMAs) show preparatory activity before a movement.^{19,20,41–43} One important distinction between our results and this prior work is that previous studies all looked at trained tasks in which the anticipated response was well known. Thus, the temporal coding observed in these studies was associated with a sustained waiting period between a cue and a movement.^20,38–42 This is likely to be a fundamentally different process from the generation of spontaneous, internally generated movements like those investigated in this study.

One theory that could merge these seemingly disparate findings is that SMAs could encode sustainment of various behavioral states, including continuous repetitive movement (naturalistic grooming) or prolonged waiting postures (periods between trained cues and responses). Similar to our findings that longer grooming bouts showed greater ALM activation (Figure 4), work in primates supports this theory by showing that longer waiting periods evoke greater activity in the pre-SMA.⁴² These data together suggest that supplementary motor regions across species are important for temporal encoding of sustained behavioral sequences or waiting periods.

Specialization of ALM-CS circuit for grooming may partially reflect role in licking behavior

Our results indicate that ALM-CS activity is relatively specific for body grooming (Figures 1 and 3), which may reflect ALM’s known role in anticipatory and ingestive licking in trained tasks that require consumption of liquid.^20,30 Here, in an untrained context, we observed evoked grooming and grooming-like behaviors that also involved licking. These convergent data suggest the ALM-CS circuit in mice may have evolved to be selective for behaviors involving licking, a behavior critical for mouse survival via hygiene, eating, and drinking.

ALM terminal stimulation results in delayed recruitment of CS cells and accompanying grooming behavior

One notable finding was that stimulation of ALM-CS terminals caused grooming behavior in a subset of mice, and initiation of this behavior was delayed to ~11 s after stimulation started (Figure 5). One possible mechanism is that ALM → CS terminal stimulation caused antidromic activation of upstream cortical cells, leading to activation of an unknown, multisynaptic circuit that caused the delayed grooming. Given that we observed occasional seizures in some of our mice (which has previously been reported with high-intensity cortical/corticostriatal optogenetic stimulation⁴⁴), there is evidence of at least some antidromic cortical activation. However, correlative analyses suggest that instead of (or in addition to) antidromic activation, this heterogeneous response may be due to insufficient and/or delayed recruitment of CS cells, likely through both mono- and polysynaptic activation. Namely, we show that in vivo activation of ALM-CS terminals produces delayed CS activation that parallels the delayed behavioral effect. Although this long latency between ALM terminal stimulation and CS cell activation was initially surprising, this striatal activation profile was also reported previously during both ALM and OFC terminal stimulation³³ and may be related to local multisynaptic activity. Furthermore, we observed an evoked-grooming behavioral effect even in mice that did not exhibit seizures, suggesting the grooming response is separable from the potential antidromic activation (although we acknowledge that antidromic activation does not always produce seizure activity). In addition, in previous ex vivo slice electrophysiology work in WTs and Sapap3-KOs,¹⁷ we showed that ALM projects monosynaptically to CS, contacting both SPNs and fast-spiking interneurons (FSIs). Though ALM inputs to CS cells were present in WTs, they were relatively weak compared to those in Sapap3-KOs. This is consistent with the observed delayed activation of CS cells when we stimulate ALM terminals in vivo in WTs and raises the possibility that it may be easier for CS cells to reach the firing threshold and trigger grooming in Sapap3-KOs due to significantly strengthened ALM-CS synapses. Finally, we observed that the ALM-CS-evoked grooming response was reduced after repeated stimulation periods. This incidental finding is consistent with a reduction in strength of ALM synapses onto CS cells leading to weaker responses after repeated stimulation.⁴⁵ Overall, we conclude that our heterogeneous and delayed grooming effects are likely reflective of the normal mechanisms through which ALM terminal activity recruits CS cells.

Interestingly, ALM-CS terminal stimulation caused more naturalistic-looking grooming behavior than directly stimulating CS cells. Although there are several possible explanations, we believe the most likely is that ALM terminal stimulation may preferentially recruit CS SPNs that are endogenously related to grooming behavior, consistent with our observed activation of the ALM-CS circuit during naturalistic grooming behavior. In contrast, direct non-specific stimulation of a broad population of CS SPNs may cause activation of multiple (potentially competing) circuits in CS, impeding expression of a “naturalistic” grooming response. A remaining point of interest is that different levels of light power caused different levels of grooming behavior. While we believe this is due to differential recruitment of CS cells, future experiments will need to use in vivo electrophysiology to thoroughly characterize how different levels of ALM terminal activation impact ALM-CS circuit activity and resulting evoked grooming behavior.

A model of grooming generation in the ALM-CS circuit

Our combined data could support a model in which grooming movements can be initiated through striatal activation, while sequencing and sustainment of these fragments into normal grooming behavior is encoded by ALM (although see alternative models below). Specifically, we show that increases in CS activity precede increases in ALM activity during initiation of spontaneous grooming bouts (Figure 6). Furthermore, optogenetic and photometry experiments demonstrate that CS activity is more temporally related to grooming initiation than ALM activity. This presents the possibility that CS activation may be a driver of grooming initiation and that ALM may be activated after grooming initiation to support sequencing of movement fragments, through either CS or different downstream targets. This contrasts with classic models positing that action plans are represented in the cortex and gated through the striatum.³ A potential circuit to support this mechanism exists: the downstream output nucleus of the striatum, the substantia nigra, projects to the ventromedial (VM) thalamus,⁴⁶ a region shown to directly affect ALM activity by supporting persistent activity.⁴⁷ Thus, CS activity could prompt grooming, and downstream VM thalamus activation could then cause reverberant, persistent activity in ALM to sustain it. An optogenetic experiment could test this model by stimulating CS first (to initiate grooming), followed by delayed ALM activation (to sustain grooming). Note that alternative models to explain our data are possible, as we do not causally demonstrate the necessity of ALM-CS for sustaining behavior (see “limitations of the study”). For example, once grooming has begun, ALM could be activated by VM thalamus but then project to globus pallidus externa or other alternative regions that ultimately sustain grooming.⁴⁸ Alternatively, it is possible that the initial and middle phases of grooming comprise different movements that require engagement of distinct circuits. For example, non-ALM cortex could kick off grooming initiation and facial grooming, acting in part through CS. As grooming proceeds, mice would start to body groom, which would increasingly recruit ALM because of its known role in licking behavior. Thus, ALM activity would not sustain grooming; rather, it would be needed to encode one of the movements required for its execution. However, our observations in mice do not support the idea that licking occurs only in later grooming phases, since we observe licking during both face grooming (wetting paws) and body grooming (licking fur). Ongoing experiments are testing these alternative models.

A remaining question is which region(s) is responsible for grooming termination. One candidate region is the subthalamic nucleus (STN).^49–51 Because it receives direct inputs from the cortex⁵² and is thought to inhibit thalamocortical transmission downstream,⁵⁰ STN is well placed to potentially terminate grooming behavior. For example, the ALM activity peak toward the end of a grooming bout could cause STN to cross an activation threshold, leading to dampening of thalamocortical transmission and grooming termination. Supporting this hypothesis, activation of posterior ALM-STN projections is sufficient to cause termination of ongoing locomotion.⁴⁹ However, this work investigated externally triggered stops, rather than internally generated behavior termination. To investigate the possibility that STN also terminates spontaneous, naturalistically generated movements, substantial work must be done.

Limitations of the study

Several limitations must be considered when interpreting our work. First, fiber photometry is a bulk calcium imaging technique that best captures synchronous events from neurons firing together, which is rare in cortical regions. There is also ongoing debate regarding whether striatal calcium activity reflects action potentials or dendritic potentials.^9,53 Thus, future studies must examine single-neuron activity during grooming to understand how our observed CS initiation signals and persistent ramping activity in ALM are associated with activity at the individual neuron level. In addition, our work does not include a causal demonstration of the necessity of the ALM-CS circuit for generating grooming behavior, due to the challenges of inhibiting a relatively infrequent spontaneous behavior. Formal testing of the model of ALM sustaining grooming must be carried out in future experiments. Finally, we cannot definitively conclude that we are recording from SPNs, since we used non-specific synapsin promoters. Since SPNs make up 89%–95% of the cells in the striatum, we believe most of our recorded striatal cells are SPNs, but since we did not specifically record striatal interneurons, we cannot verify this via direct comparison of observed firing rates and interneuron firing rates. However, we have evidence from both literature and our lab that striatal interneurons typically display characteristic calcium dynamics that we do not observe in this dataset−i.e., a gradual rise and fall in calcium activity, as opposed to the discrete events that we observed here. Despite these limitations, these data are an important step in understanding the role of an understudied region of the striatum and one of its primary cortical inputs in the generation of sequenced behaviors.

Conclusions and future directions

Our data present a corticostriatal framework for understanding how spontaneous behavioral components may be initiated and sequenced into smooth movements. Complementary recent studies showing the involvement of CS in compulsive behavior^54,55 provide support for our conclusions that the ALM-CS circuit is important for motivated, repetitive behaviors. Furthermore, our results suggest that supplementary motor regions may be a useful target for treatment of abnormal sequencing and sustainment of behaviors in illnesses such as obsessive-compulsive disorder (OCD). Consistent with this idea, the pre-SMA has been identified as a promising target for transcranial magnetic stimulation treatment in OCD.^56,57 In contrast, disorders characterized by the generation of abnormal movement fragments may be better treated by targeting the striatum and downstream regions such as thalamus or STN, which have been identified as promising targets for Tourette syndrome and dyskinesia.^58,59 Future studies investigating the neural bases of initiation and sustainment of repetitive behaviors are critical for development of individualized treatments to target specific symptom subtypes.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources and reagents should be directed to and will be fulfilled by the lead contact, Susanne Ahmari (ahmarise@upmc.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• All data reported in this paper will be shared by the lead contact upon request.

• All original code used throughout the paper has been deposited on GitHub or Zenodo and is publicly available as of the date of publication. DOIs are listed in the key resources table.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Fluoromount	ThermoFisher	00–4959-52
Anti-GFP	Abcam	ab13970; RRID:AB_300798
Anti-mCherry	Novus Biologicals	NBP2–25158; RRID:AB_2636881
Bacterial and virus strains
AAV9-Synapsin-GCaMP6m-WPRE-SV40	Addgene	100841; RRID:Addgene_100841
AAV2-hsyn-EYFP	UNC Vector Core	AV4876D
AAV1-syn-NES-jRGECO1a-WPRE-SV40	Penn Vector Core	CS1311
AAV2-hsyn-mCherry	UNC Vector Core	AV5033D
AAV8-syn-ChrimsonR-tdT	Addgene	59171; RRID:Addgene_59171
AAV2-hSyn-ChR2-EYFP	UNC Vector Core	AV4384H
Experimental models: Organisms/strains
C57BL/6J	Jackson Labs	000664; RRID:IMSR_JAX:000664
Sapap3-KO	Guoping Feng laboratory
Software and algorithms
Observer XT	Noldus	10
Ethovision	Noldus	10
Graphpad	Prism	8
SLEAP	SLEAP	1.2.2
SPSS	IBM	26
sklearn	python	1.2.0
Inscopix Data Processing Software	Inscopix Inc, Palo Alto, CA USA	v1.3.0
TurboReg	Ghosh et al.60	N/A
Constrained Non-negative Matrix Factorization (CNMFe)	Zhou et al.61	N/A
Analysis code	This paper	https://github.com/vcorbit/alm-cspaper2024
Single-cell Calcium Data analysis code	Piantadosi et al.55	https://doi.org/10.5281/zenodo.10790735
Illustrator 20	Adobe	N/A
ImageJ	National Institutes of Health	1.54d

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Male and female wild-type (WT) C57BL6 background mice were used for all experiments. Most WT mice were genotype-confirmed littermates of Sapap3-KO mice, a colony initially established at MIT by Dr. Guoping Feng. One cohort of C57BL6/J WT mice (part of ALM-CS Stim, Figure 5) was purchased directly from Jackson Laboratory. Mice were group housed with 2–5 mice per cage except when noted. All mice had ad libitum access to food and water. Animals were randomly assigned to experimental or control groups, balancing for sex and cage mates. All experiments were approved by the Institutional Animal Use and Care Committee at the University of Pittsburgh in compliance with National Institutes of Health guidelines for the care and use of laboratory animals.

METHOD DETAILS

Stereotaxic surgeries

Mice underwent stereotaxic surgery between the ages of 4 and 8 months. Stereotaxic surgeries were performed under isofluorane anesthesia (2%). Burr holes were drilled over the target location for subsequent virus injection or implant. Virus was injected using a syringe pump (Harvard Apparatus) fitted with a syringe (Hamilton) connected to PE10 tubing and a 30 gauge cannula and allowed to incubate for at least 3 weeks before experiments.

All recording (single-cell calcium imaging, fiber photometry) experiments were conducted using unilateral virus injections and implants. For fiber photometry, AAV9-Synapsin-GCaMP6m-WPRE-SV40 (or EYFP control virus, 250nL, Addgene) was injected into ALM (AP 2.80–2.90, ML 1.55, DV .75mm) and/or AAV1-syn-NES-jRGECO1a-WPRE-SV40 (or mCherry control virus, 500nL, Addgene) was injected into CS (AP .50, ML 1.95, DV 3.00mm). Optical fibers (NA = .37) were implanted into ALM and CS at the same AP and ML coordinates, but were 0.15mm above the injection site. For ALM-CS stimulation and calcium imaging experiments, AAV8-syn-ChrimsonR-tdT (300nl, Addgene) was injected into ALM and AAV9-hSyn-GCaMP6m (800nl, Addgene) was injected into CS. A 6mm long 500um diameter GRIN lens was implanted over the injection site in CS, to simultaneously stimulate ALM terminals and record calcium activity from CS soma. For simultaneous stimulation and calcium imaging from CS cells, AAV8-syn-ChrimsonR-tdT and AAV9-hSyn-GCaMP6m (800nl total, mixed 1:1) were injected into CS of WT mice. A 6mm long 500um diameter GRIN lens was implanted over the injection site in CS to allow for simultaneous stimulation and activation of CS cells.

All optogenetic behavioral manipulations were conducted with bilateral virus injections of AAV2-hSyn-ChR2-EYFP or AAV2-hSyn-EYFP (500nL, Addgene) into either ALM (AP 2.80–2.90, ML 1.55, DV 0.75mm) or CS (AP 0.70, ML 2.00, DV 3.00mm, fibers at 2.60–2.85mm). For ALM terminal stimulation, AAV2-hSyn-ChR2-EYFP or AAV2-hSyn-EYFP (350nL, Addgene) was injected into ALM and fibers were implanted into CS (AP 0.70, ML 2.00, DV 2.60–2.85mm). For CS stimulation, fibers were implanted 0.15–0.40mm above the viral injection site (AP 0.70, ML 2.00, DV 2.60–2.85mm). ChR2 mice and EYFP control mice were treated identically for the remainder of the experiments.

Freely-moving behavior observations

Behavioral observation sessions were optimized to minimize animal anxiety and maximize the expression of spontaneous grooming behavior. This included at least 3 days of handling and at least 3 days of habituation to scruffing, fiber cable attachment, and exposure to the behavioral chamber. This habituation period ensured that mice were comfortable with the experimenter and the behavioral paradigm. To further reduce anxiety, tests were conducted under dim light and the observation chamber used was smaller than a typical open field (10in x 10in).

Behaviors (grooming, rearing) were manually tracked offline using frame-by-frame scoring of behavior start and end times (Noldus, Observer XT). Grooming was defined as unilateral or bilateral swipes of the paw at the face, turning the head to the side and licking the flank, or hindleg scratching of the body. Rearing was defined as a mouse balancing itself on its hindlegs and stretching its body upwards. The start of grooming or rearing was marked as the first frame when the animal moved both paws off the ground before a grooming or rearing instance. The end of a bout was marked as the frame when the animal placed both paws back on the ground. If an animal’s resting posture did not involve paws on the ground before and/or after the behavior bout, then start and end were marked as the frame when the animal’s paws moved from or returned to the rest position, respectively. In a set of practice videos, scoring of grooming and rearing between two independent blinded scorers had a significant correlation (R = .89, p = .00001) with a linear slope of 1.07.

Locomotion was tracked automatically using Noldus Ethovision. Locomotion was defined as velocity exceeding a threshold of 10cm/s, and immobility defined as velocity below a threshold of 3cm/s.

Optogenetic behavioral manipulations

After 4–6 weeks of virus incubation and recovery, mice were handled for several days prior to behavior experiments. All mice were habituated to the observation chamber and optical fiber tethering for three days prior to behavioral manipulation. On experiment day, mice were scruffed and attached to optical cables and placed in a 10x10 inch clear plexiglass observation chamber. A Point Grey camera was fixed beneath the chamber and behavior was filmed from below.

For CS stimulation experiments, 5mW 470nm laser light was used. Fifty-one 10s trials of constant light were presented with a pseudorandom inter-trial interval with an average of 30s (25–35s, 5s jitter, TTLs triggered via an Arduino). Constant prolonged optogenetic stimulation has previously been shown to produce spiking in striatal neurons¹². All experimenters were blinded to experimental condition.

Stimulation of ALM terminals was altered due to the propensity of cortical optogenetic stimulation to cause seizure activity. 20Hz (10ms pulse width) pulsed 470nm laser light was triggered via TTLs from an Arduino sent through a Master-8 Pulse Stimulator (AMPI) to generate the pulse pattern. Initial experiments were conducted at 10mW light power, and animals were monitored for seizure or pre-seizure activity. Pre-seizure activity was operationally defined as interruption of ongoing behavior and initiation of bilateral synchronized rhythmic movement of the paws. Seizure activity was operationally defined as evolution of bilateral synchronized paw movement into tonic-clonic generalized seizure activity. For animals that did show seizure activity, light was lowered first to 7mW and then to 5mW if seizure activity persisted. Sessions included in analysis were the highest light power used that did not cause seizure activity; additionally, any trials within the session that had pre-seizure activity were excluded from analysis. Pilot testing showed that 1) 20s light periods were more likely to show grooming behavior than 10s light periods, and 2) grooming behavior was reduced as trial number increased (Figure S3). Thus, data are presented from experiments conducted with 20s light pulses, and only the first usable 20 trials (e.g. without pre-seizures) of this experiment were analyzed.

Fiber photometry

Fiber photometry experiments were conducted in freely behaving mice in a 10x10 inch clear plexiglass chamber. A Neurophotometrics 3-color, 2-site system was used to collect imaging data (Neurophotometrics). Three LEDs (415nm, 470nm, 560nm) were pulsed at 30Hz in an interleaved manner to obtain 1) isosbestic motion signal, 2) GCaMP6m activity, and 3) jRGECO1a activity. The recorded trace was then separated to obtain activity for each channel individually.

Four cohorts of mice were used in the collection of photometry data (Cohort 1: GCaMP6m in ALM, total N=9 with ALM signal; Cohort 2: GCaMP6m in ALM, JRGECO1a in CS, total N = 3, 1 with CS signal; Cohort 3: GCaMP6m in ALM, JRGECO1a in CS, total N = 7, 3 with CS signal; Cohort 4: GCaMP6m in ALM, JRGECO1a in CS, total N = 5, 5 with CS signal). Reasons for no signal in some animals include poor targeting of virus and/or fiber, and fibers being pulled out of the cement headcap. All mice that had CS signal were used in Figure 1 (N=9). The ALM-only imaging cohort (Cohort 1) was used for data in Figure 3. To gain additional data for the bout length analyses (Figure 4), additional ALM-GCaMP6m trials from the dual-color photometry cohorts were added to the analysis (Cohorts 2–4). Finally, for the dual-color photometry analyses (Figure 6), any animal with signal in both ALM and CS was used (N = 6/15).

Freely moving microendoscopy

Mice were habituated to microscope attachment and the grooming chamber for three consecutive days prior to testing. For recording single cell activity in the CS, analog gain of the image sensor was set between 1 and 4 while the 470 nm LED power was set between 10 to 30% transmission range. Stimulation of ALM terminals or CS cells infected with ChRimson was achieved through the delivery of 600nm amber light through the objective lens as GCaMP-positive cells were simultaneously being excited with 460nm blue light.⁶² After mice were placed into the chamber and calcium imaging began, the 600 nm optogenetic LED (OG-LED) was turned on. For ALM-CS stimulation experiments, OG-LED stimulation consisted of 15 pulse trains of 20 Hz stimulation for 20s. Each 20 Hz train was followed by a 30s interval of no stimulation. Each session lasted 15 minutes in total. For CS stimulation experiments, OG-LED stimulation consisted of two sessions, one in which mice received 14 trials of 10s of constant OG-LED stimulation followed by 50s of no stimulation. A second session was conducted in which mice received 14 trials of 10s 20hz OG-LED stimulation followed by 50s of no stimulation.

Histology

After experiments were completed, mice were transcardially perfused using 4% paraformaldyhyde (PFA) and post-fixed in PFA for 24 hours. Post-hoc confirmation of viral and implant targeting was conducted on 35um slices from the harvested brains. Slices were mounted with DAPI coverslipping media and inspected for relevant fluorophores (e.g. EYFP, GFP, or mCherry). Fiber implants were identified by finding damage tracks in the brain at the specific location.

QUANTIFICATION AND STATISTICAL ANALYSIS

Behavior quantification and classification

Behavioral data were manually scored by blinded raters using Noldus Observer. Grooming probability was binned into 500ms time bins and analyzed with two-way RM-ANOVAs and post-hoc t-tests using Sidak’s p-value correction (Prism, Graphpad). Both stimulation cohorts (Main Figures 2 and 5) were replicated in an additional cohort of mice.

To quantify evoked grooming in the optogenetic stimulation experiments (Figures 2 and 5), all bottom-up view behavioral videos were run through a trained SLEAP tracking network (CS stimulation network trained on 1201 frames, ALM-CS network trained on 956 frames; Figure S2). The tracking network was trained to identify snout, body center, tailbase, forepaws, and hindpaws, though the hindpaw locations were not used in subsequent analyses. After SLEAP tracking, joint x/y positions were extracted for each video and center and rotated along with tailbase to body-center axis. The following features were then calculated: x/y paw distance to snout for each paw, x/y snout distance to body center, snout angle off body axis, snout angular velocity, and overall body-center velocity. These features were chosen a priori as factors important for identifying grooming behavior. For the first classifier to determine if evoked-grooming was distinct from face- and body-grooming (Figures 2E and 5E), manual labeling was used to isolate times of face-grooming, body-grooming, and evoked-grooming. The data were split into training data (60%) and test data (40%) using the train_test_split function in the sklearn python package, including the stratify option to ensure that equal amounts of each type of grooming were included in each dataset. A support vector machine was trained on the training data (sklearn.svm.SVC, C = .5), and then classes were predicted on the test data. F1 scores (2*precision*recall/(precision+recall) were reported as an overall measure of classifier performance.

For the second classifier to determine which behaviors the evoked grooming most resembled (Figures 2F and 5F), a new classifier was trained to distinguish face-grooming, body-grooming, and locomotion (and stereotypies in Figure 5F). Locomotion periods were identified using a velocity threshold of 10cm/s, and other behaviors were manually labeled. For this model, all evoked-grooming times were withheld from training. After the model was trained to identify the different behaviors, evoked-grooming was input to the model to obtain predictions for what behavior class the evoked-grooming was most similar to. To quantify this, the recall of evoked-grooming (proportion of evoked-grooming timepoints that were classified as face-grooming, body-grooming, stereotypy, or locomotion) was calculated. In Figure 2F, animals were classified as having evoked behavior that looked most similar to face-grooming vs body-grooming via inspection of the joint positions and the experimental videos. In Figure 5F, because evoked-grooming could look like face- or body-grooming within a mouse, the recall proportions were combined for these two classes.

Fiber photometry analysis

After separating the 3 channels, a linear fit of the isosbestic signal to the activity channels (470nm or 560nm) was calculated. This linear fit was then subtracted from each corresponding activity channel to remove baseline fluorescence and motion artifacts. An additional moving minimum baseline (2min sliding window) was subtracted from each resulting trace to account for slow fluctuations in activity, such as additional decay from bleaching. Finally, each activity trace was normalized by dividing by the standard deviation. Manually (grooming, rearing) or automatically (locomotion) scored behaviors were then aligned to the activity traces using an initial “session start” LED signal for alignment.

Unless otherwise noted, analyzed grooming bouts were restricted to bouts that did not show grooming for 3s prior to grooming bout onset. Trials were then zeroed to the 3s baseline period by subtracting the mean activity in that interval from the overall trace. Statistical analysis of time-varying calcium traces was performed using two-way repeated-measures ANOVAs with post-hoc contrasts comparing the activity at 1s before behavior onset to all other time bins (SPSS, IBM). All statistical tests were conducted across individual animals’ average trial data unless otherwise noted.

For peak detection analysis, grooming trials were further excluded if they had any additional grooming initiations in the 10s following grooming bout onset, to avoid grooming onset contamination in the detection of peaks. Peak analysis was conducted using the MATLAB function findpeaks. This function finds the local maxima in a trace and provides information about time, amplitude, and “prominence”– a measure that takes into account the amplitude of a given peak, the slopes on either side, and the amplitude of the peaks surrounding the peak of interest. The first peak reaching the prominence threshold that occurred after grooming onset was the automatically detected peak. Time was extracted relative to grooming onset time for each trial, and absolute amplitude was calculated from the baselined trials.

The cross-correlation analysis for the simultaneously recorded ALM and CS data in Figure 6 was conducted by first identifying grooming-initiation-centered fragments of the traces for each animal and then computing the cross-correlation between the ALM and CS trace fragments. Cross correlation traces were then combined across trials and animals and plotted. Area under the curve on either side of the groom-initiation point (time=0) was calculated and compared using a paired t-test.

Freely moving microendoscopy analysis

Following acquisition, raw calcium videos were spatially downsampled by a binning factor of 4 (16x spatial downsample) and temporally downsampled by a binning factor of 2 (down to 10 frames per second) using Inscopix Data Processing Software (v1.3.0, Inscopix Inc, Palo Alto, CA USA). Lateral brain motion was corrected using the registration engine TurboReg,⁶⁰ which uses a single reference frame to match the XY positions of each frame throughout the video. Motion corrected 10 Hz video of raw calcium activity was then saved as a .TIFF and used for cell segmentation.

Using custom MATLAB scripts, the motion corrected .TIFF video was then processed using the Constrained Non-negative Matrix Factorization approach (CNMFe), which has been optimized to isolate signals from individual putative neurons from microendoscopic imaging.⁶¹ The CNMFe method is able to simultaneously denoise, deconvolve, and demix imaging data⁶³ and represents an improvement over previously used algorithms based on principle component analysis.⁶¹ Putative neurons were identified and manually sorted according to previously established criteria.⁶⁰ For each individual cell, the raw fluorescence trace was Z-scored to the average fluorescence and standard deviation of that same trace. Thus, fluorescence units presented here are referred to as “Z-scored fluorescence”.

To identify central striatal cells modulated by ALM terminal activation or direct CS activation, the Z-scored fluorescence response for each cell to 20s of OG-LED stimulation was averaged across presentations (15 total presentations). This average was then compared with an unpaired t-test to the periods immediately preceding (10s before) and during (20s) stimulation (Bonferroni multiple comparison correction p ≤ 0.0001).

The response latency for the ALM-CS nVoke experiment (Figure 5H) was determined by first calculating an average response across stimulation trials for all activated cells. Baseline levels were calculated by determining the average level of activity in the 10s preceding the stimulation pulse. Then, for each cell, the threshold for when the response reached 30% of its maximum difference from baseline was calculated. The latency was recorded as the time when the cell crossed that 30% threshold.

Histology analysis

To create histology maps (Figures S1, S3, and S4), microscope images of the fluorophone and DAPI channels were imported into Adobe Illustrator and aligned to the best-fitting atlas image (Paxinos and Watson). All microscope images were captured using the same exposure settings, and no adjustments were made to the color balances. Using the pencil tool, the shape of viral spread was traced. The ferrule tip location was identified by looking for damage tracks.

ALM projection fluorescence quantification in the ALM-CS terminal stimulation experiment (Figure S10) was conducted using ImageJ (NIH). First, the grayscale images of the EYFP fluorescence were obtained and the slice that contained the ferrule tip spot was identified. Second, a circular region-of-interest under the ferrule tip spot was drawn (500um diameter). Finally, the total intensity of the pixels was calculated for each hemisphere and the sum was used to determine “ALM projection intensity” for that animal.

Highlights.

• ALM cortex and central striatum (CS) are probed during spontaneous behavior in mice

• CS Ca²⁺ peaks at groom start, and CS optoexcitation causes rapid grooming-like behavior

• ALM Ca²⁺ rises gradually, and peaks in activity correlate with grooming cessation time

• Optogenetic excitation of ALM-CS causes delayed, longer-lasting grooming-like behavior