Introduction

The process of sensorimotor learning is underpinned by sensory perception and motor control.¹ In the mouse whisker system, tactile sensations are acquired via active sensor (e.g., whisker) movement to obtain relevant environmental information and subsequent processing by the well-characterized primary somatosensory barrel cortex (S1) circuitry.^1,2 This whisker-related information is transmitted from the periphery to S1 via two thalamic nuclei, ventral posterior medial (VPM) and posterior medial (POm), constituting the lemniscal and paralemniscal pathways, respectively.^{3,4,5,6,7,8,9,10} VPM reliably encodes fast-whisking components including self-motion and tactile information.^{7,11,12,13,14} Conversely, POm encodes phase-related whisking activity with relatively lower magnitude responses and higher response failure rates.^{7,12,13,14,15,16,17,18} Recent work has highlighted two behavior-related aspects of POm function: (1) activation during changes in behavioral state, especially related to sensory and nociceptive processing,^{7,11,12,13,17,18,19,20,21} and (2) driving learning-related plasticity at its cortical synapses.^{22,23,24,25,26}

POm receives a plethora of inputs including glutamatergic (S1, primary motor cortex (M1), secondary somatosensory cortex, superior colliculus, and spinal trigeminal interpolaris), ^{15,27,28,29,30,31,32,33,34} GABAergic (ventral zona incerta, anterior pretectal nucleus, and thalamic reticular nucleus),^{16,35,36,37,38,39,40,41,42} and cholinergic (pedunculopontine and laterodorsal tegmental nuclei).^43,44,45,46 Further, the stereotypical POm-cortical projection terminates in S1, specifically layers (L)1 and L5A,^4,47,48,49 and has been studied in the context of driving cortical plasticity/perceptual learning.^{22,23,24,25,26,50,51} In addition to its cortical projection, POm axons pass through and collateralize with terminal synaptic boutons in both thalamic reticular nucleus and posterior dorsolateral striatum (pDLS) as they ascend towards cortex.^{49,52,53,54,55,56,57,58} Here, we focus on the POm-striatal projection as striatal circuitry modulation may have powerful effects on sensorimotor integration and behavior.⁵⁹ However, POm’s influence over striatal microcircuitry and behavioral performance is unresolved.

The striatum is the predominant input nucleus of the basal ganglia and is predominantly composed of GABAergic spiny projection neurons (SPNs) expressing either D1 or D2 dopamine receptors,^60,61,62 but it also contains a rich diversity of interneurons, such as parvalbumin-expressing (PV) fast-spiking interneurons that exert robust modulatory control over SPN output.^{63,64,65,66,67} Within this microcircuitry, the dorsal striatum integrates widespread convergent cortical and thalamic inputs that constitute part of the force driving normal striatal functioning.^{68,69,70,71,72,73,74,75,76} Notably, functionally-related cortical (S1 and M1) and thalamic (POm) inputs converge within shared striatal subregions.^73,74,77 For example, M1 and S1 are heavily interconnected via reciprocal L2/3 and L5 connections,^78,79,80 and their projections overlap within dorsal striatum and even onto the same neuron.^{73,81,82,83,84,85,86} While some studies treat striatal inputs as a uniform entity,^87,88 they have been shown to differ anatomically,^73,74,76 functionally,^89,90 and behaviorally.^89,91,92 Thus, the specific cortical and thalamic origin of striatal inputs likely has significance for understanding how the striatal circuitry integrates sensorimotor information to modulate behavior.

The most prominent thalamostriatal modulation occurs via parafascicular (Pf) thalamus.^{90,93,94,95,96,97,98,99,100,101} Pf is implicated in regulating action flexibility^102,103 and contributing to the initiation and execution of learned sequences of movements^101,104 through its robust innervation of striatal cholinergic interneurons.^95,87,88,90 Conversely, despite direct comparisons to Pf,⁵⁶ POm’s functional innervation pattern and subsequent influence over the striatal microcircuitry and choice behavior is undetermined.⁵⁹ Here, we used ex vivo whole-cell recordings of identified D1-SPNs, D2-SPNs, and PV interneurons to assess the functional connectivity of POm-striatal projections, and in vivo fiber photometry and photoinactivation to identify the contribution of POm-striatal axonal activity on sensory-guided behavioral performance and learning.

Results

POm Equally Innervates Striatal Cell Types With Faster Latency In PV Interneurons

We stereotaxically injected pAAV-ChR2-EYFP unilaterally in POm, permitting channelrhodopsin (ChR2) expression in its thalamostriatal terminals to investigate the relative synaptic strength of POm inputs onto three identified striatal neurons (D1-SPNs, D2-SPNs, and PV interneurons; Figure 1A-B). The injection site was confirmed by verifying the stereotypical POm-cortical projection pattern (S1 L1 and L5a; Figure 1C).^{4,25,26,47,49} In acute ex vivo brain slices, neurons were targeted for patch clamp recordings within pDLS (AP from bregma = −0.34 to −1.22) corresponding with the POm-striatal axonal projection field (Figure S1).^56,105 Striatal neurons were identified and targeted by crossing their respective Cre-recombinase mouse lines with tdTomato-expressing reporter mice and validating their intrinsic electrophysiological properties in response to hyperpolarizing and depolarizing current steps (Figure S1A-D, G-L; see STAR Methods).^64,67,106 Whole-cell current-clamp recordings were performed without inhibitory synaptic blockers to resemble natural physiological responses to optogenetic activation of POm inputs.⁸⁹ After breaking in, patched cells were subjected to a standard set of protocols: (1) hyperpolarizing and depolarizing current steps to define intrinsic firing properties and optogenetic; and (2) single pulse (SP), (3) paired-pulse ratio (PPR), and (4) train stimulation to measure synaptic responses.

Figure 1.

Optogenetic activation of POm terminals readily elicited depolarizing postsynaptic potentials (PSP) in all targeted cell types (Figure 1D). Responses to SP stimulation resulted in relatively equal PSP amplitudes for D1-SPNs (7.05±0.75mV), D2-SPNs (8.79±1.67mV), PVs (10.83±1.91mV), and neighboring unlabeled cells that we termed putative SPNs, based on their intrinsic firing properties (5.54±1.04mV) (F_(4,51) = 2.455, p = 0.4835; Figure 1E). Three PV interneurons and one D1-SPN exhibited action potentials to SP stimulation and were excluded from further analysis. No correlation was observed between PSP amplitude and increasing distance from the injection site (r² = 0.08493, n = 59 cells; Figure S1E-F). However, the latency to maximum PSP amplitude was significantly shorter in PVs (7.29±0.32ms, n = 17 cells from 7 mice) than D1-SPNs (11.05±0.45ms, n = 20 cells from 6 mice), D2-SPNs (10.68±0.56ms, n = 11 cells from 5 mice), and putative SPNs (10.81±0.92ms, n = 7 cells from 4 mice) (F_(4,51) = 29.78, p < 0.0001, PV vs. D1 p < 0.0001, PV vs. D2 p = 0.0003, PV vs. SPN, p = 0.0056; Figure 1F). In a subset of recordings, identified and unidentified cells within the same field of view on the same slice were patched sequentially to control for injection site variability. D1- and D2-SPNs did not differ in PSP amplitude (9.55±2.83mV in D1-SPNs vs. 8.48±2.15mV in D2-SPNs, p = 0.6406, n = 8 pairs, N = 5) or latency (10.13±0.66ms in D1-SPNs vs. 10.92±0.56ms in D2-SPNs, p = 0.3254, n = 8 pairs, N = 5; Figure S1M-O). In contrast, sequentially patched PV and SPNs did not significantly differ in maximum PSP amplitude (10.26±3.05mV in PVs vs. 4.76±0.97mV in SPNs, p = 0.2500, n = 9 pairs, N = 7), but PVs had faster latency (6.85±0.42ms in PVs vs. 10.75±0.87ms in SPNs, p = 0.0006, n = 9 pairs, N = 7; Figure S1P-R), validating the population results. Strikingly, we found relatively equal PSP amplitudes in all recorded cell types within pDLS, indicating that POm provides robust and unbiased synaptic input to all targeted striatal cells.

Short-Term Synaptic Dynamics Are Similar Between Striatal Cell Types, But Synaptic Depression Is Milder In PV Interneurons

The strength of synaptic inputs varies dramatically based on activation frequency and in a cell-type-specific manner with robust synaptic contacts generally exhibiting synaptic depression.^{89,90,107,108} To fully characterize the relative synaptic strength of POm-striatal inputs in a cell-type-specific manner, we assessed short-term plasticity by applying a PPR protocol of five pulses (Figures 1G-H, S2A-B).^109,110 While all cell types exhibited robust synaptic depression overall, no PPR differences were observed (D1-SPNs: 0.73±0.03, n = 20, D2-SPNs: 0.72±0.05, n = 11, PVs: 0.86±0.05, n = 17, SPNs: 0.71 ± 0.08, n = 7) (F_(4,51) = 7.101, p = 0.0688; Figure 1I).

To further characterize short-term synaptic dynamics, we applied a train protocol of thirty pulses at a frequency characteristic of POm-striatal activity (Figure 1J-K).^18,111 Similar to PPR stimulation, train stimulation elicited overall synaptic depression in all cell types, but PV interneurons exhibited a milder synaptic depression relative to both SPN types that occurred predominantly between pulses 5-15 (Figure 1L, S2). PV interneurons (0.49±0.05, n = 17) showed significant differences compared to D1-SPNs (0.30±0.03, n = 20) and D2-SPNs (0.28±0.03, n = 11), but not with SPNs (0.31±0.06) (F_(4,51) = 10.99, p = 0.0118, D1-SPN vs. PV p = 0.0184, D2-SPN vs. PV p = 0.0481, PV vs. SPN p = 0.4940; Figure 1L). Thus, POm-striatal projections provide SPNs and PV interneurons with unbiased and large amplitude synaptic inputs, characterized by milder synaptic depression in PVs, highlighting a potentially significant role in modulating striatal microcircuitry.^89,90

Mice Rapidly Learn To Discriminate Between Two Textures

While specific sensorimotor integrative and learning roles have been proposed and tested for several striatal inputs,^{69,72,87,89,90,104,112,113,114} the role of POm-striatal projections is still unknown. To monitor the activation of POm-striatal projections, we injected pAAV-axon-jGCaMP8s unilaterally into left POm and implanted a 400µm core cannula into left pDLS (Figure 2A-B). Violet light (405nm) and blue light (470nm) were constantly delivered to pDLS throughout the entire session to measure the isosbestic and POm-pDLS axonal calcium signals, respectively. Using a similar protocol from our previous publication,⁸⁹ water-restricted wild-type mice were trained on a whisker-based discrimination (Go/NoGo) paradigm. Mice received water for licking correctly (Hit) to the Go texture (P100 sandpaper). They received a white noise tone and a 12s time-out period for licking incorrectly (False Alarm; FA) to the NoGo texture (P1200 sandpaper; Figure 2C). If mice did not lick to the Go or NoGo texture, the texture retreated to its starting point, and trials were considered Miss and Correct Rejection (CR), respectively (Figure 2D). Additionally, pupil dynamics are a known metric of arousal,^115,116 correlate well with POm activity,^18,26 and exhibit outcome-dependent differences during the Go/NoGo paradigm.¹¹⁷ Therefore, synchronized orofacial video was captured during behavioral performance, and deep-learning^118,119 pupillometry¹²⁰ was applied to assess pupil dynamics during task performance (Figure S3B) in which our results mirrored previously reported outcome-dependent differences.¹¹⁷

Figure 2.

Mice in the fiber photometry (FP) cohort (n = 5) underwent three training phases (Shaping, Learning, and Expert; Figure S3A) that were segmented into five discrete behavioral time points (Shaping, Early Learning, Late Learning, Expert, and Reward; see STAR Methods). During the learning phase, Hit rate increased, and FA rate decreased significantly, leading to markedly increased sensitivity (d’) and slightly decreased bias (Figure 2E, S3A). Mice were considered Expert once they had reached ≥ 0.80 Hit Rate and ≤ 0.30 FA Rate for two consecutive sessions as opposed to a strict sensitivity (d’) threshold. As Hit Rate or FA Rate approach their extremes (0 or 1), the sensitivity metric is artificially increased, giving the appearance of discrimination (Figure S3A). On average, it took this cohort 7.6±0.51 sessions to become Experts (Figure 2F). Thus, the FP cohort rapidly learned to discriminate between the two textures as validated by behavioral responding parameters.

Calcium Activity Markedly Increases And Becomes Stereotyped Across Learning

To elucidate the contribution of POm-striatal projections during the Go/NoGo discrimination task, we measured POm axonal calcium activity along with licking and pupil activity at five discrete behavioral time points (Shaping, Early Learning, Late Learning, Expert, and Reward; Figure 2G-K). Photometry measurements revealed learning-related increases in POm axonal activity, starting before and peaking near texture presentation (Figures 3A, S3A) as measured by two parameters: calcium signal amplitude (ZMAD; median absolute deviation of the z-score) and area under the curve of the receiver-operator characteristic (auROC). First, the average maximum calcium amplitude significantly increased across learning (F_{(1.818,7.271)} = 39.24, p = 0.0001, n = 5; Figure 3B). Further, the average maximum amplitude at Shaping (0.27±0.04) was significantly smaller compared to Early Learning (1.04±0.12, p = 0.0190), Late Learning (1.58±0.18, p = 0.0061), and Expert (1.59±0.17, p = 0.0052). To test whether POm activity was reward-related, a single session was performed following the Expert phase during which the textures were removed and water was automatically provided at the end of the presentation time (PT) window. We observed that calcium activity regressed to Shaping levels and was significantly smaller than Early (p = 0.0306), Late (p = 0.0126), and Expert (p = 0.0098).

Figure 3.

Second, to quantify POm axonal activity within the striatum relative to learning, we employed signal detection theory, utilizing auROC values to compare the basal activity across the five behavioral time points as learning occurs.^121,122 As with the maximum calcium amplitude, auROC values significantly increased across learning (F_{(2.162,8.650)} = 51.17, p < 0.0001, n = 5; Figure 3C). Notably, auROC values at Shaping (0.54±0.01) were significantly lower than Early Learning (0.70±0.03, p = 0.0116), Late Learning (0.83±0.03, p = 0.0017), and Expert (0.85±0.02, p = 0.0007), but not Reward (0.57±0.01, p = 0.4111). Further, significant auROC differences were also noted between the Early Learning vs. Expert (p = 0.0383), Late vs. Reward (p = 0.0084), and Expert vs. Reward (p = 0.0042). Finally, calcium-related events increased longitudinally and became stereotyped within a 4s target window (centered around texture presentation) compared to a non-task-related 1s control window (during the 1s pre-task interval prior to the sound cue and texture movement; Figure S4A-D). Thus, POm-striatal projections represent a learning-related signal that increases prior to and becomes stereotyped to texture presentation.

Licking and Pupil Dilation Also Markedly Increase Across Learning, But Only Pupil Is Correlated With Calcium Activity

Both licking and pupil dilation also exhibited marked increases across learning (Figures 2G-K, S3A-B). Licks occurring before texture presentation decreased dramatically as licking became stereotyped within the PT window (Figure 3D). We assessed whether licking and POm activity were correlated as both increase across learning (Figure 3A, D) by plotting the average calcium activity from the Early, Late, and Expert time points with the grand average reaction time (RT) set to 0 (Figure 3E). Comparison of average calcium activity pre- and post-RT revealed an overall significant reduction (0.79±0.09 in pre-RT vs. 0.00±0.03 in post-RT, p = 0.0221; Figure 3F), indicating no correlation between POm and licking activity, validating previous results.²⁶

Pupil dilation started immediately following the cue sound and peaked near texture presentation before slowly decreasing throughout the rest of the trial (Figure 3G). Similar to previous studies,^18,26 we found that pupil and POm activity were tightly correlated but decoupled during the PT window with POm activity immediately regressing to baseline, while pupil activity remained elevated (Figure 3A, G). Pupil dilation lagged by ~250ms on average (Figure 3H), and the correlation became more consistent across learning and was restricted to before and at texture presentation (Figure 3I). Thus, despite all three activity parameters increasing across learning, POm activity is not related to motor activity and is only correlated with pre-PT pupil dilation.

Increased POm Axonal Calcium Activity Is Independent of Trial Type or Outcome

This whole-session analysis did not account for sensory (texture) or outcome differences. To assess whether POm axonal activity is sensory-related (texture-specific), sessions were segmented by the presented texture (Go or NoGo; Figure S4E-F) and by trial outcome (below). Notably, licking and pupil dynamics have discrete activation patterns based on the presented texture.¹¹⁷ Licking activity in the Go condition occurs squarely within the PT window and is indistinguishable from whole-session licking activity (Figure 3D), while licking peaks near the end of the grace period in the NoGo condition (Figure 3J). Similarly, pupil activity starts to increase at the same point in both conditions but deviates at the end of the grace period. Conversely, POm axonal activity remained strikingly consistent between the Go and NoGo conditions (Figure 3J). This observation was further validated by comparing the average maximum calcium amplitude of each behavioral time point between the presented texture conditions. The main effect of behavioral time points was significant (F _(4,32) = 57.16, p < 0.0001), but the main effect of presented texture was not (F _(1,8) = 0.3797, p = 0.5549; Figure 3K). Post-hoc comparisons found that the maximum calcium amplitude during Shaping (Go: 0.55±0.06; NoGo: 0.53±0.04) was significantly reduced compared to Early Learning (Go: 1.18±0.11, p = 0.0016; NoGo: 1.05±0.11, p = 0.0128), Late Learning (Go: 1.65±0.23, p < 0.0001; NoGo: 1.59±0.16, p < 0.0001), and Expert sessions (Go: 1.68±0.21, p < 0.0001; NoGo: 1.52±0.12, p < 0.0001), but not the Reward session. The same overall effects were noted for the auROC analysis (main effect of behavioral time point: F _(4,32) = 66.43, p < 0.0001; main effect of presented texture: F _(1,8) = 0.01694, p = 0.8997; Figure 3L). As with the maximum amplitude, post-hoc comparison highlighted that the auROC value during Shaping (Go: 0.59±0.02; NoGo: 0.61±0.03) was significantly smaller than Early Learning (Go: 0.72±0.02, p = 0.0035; NoGo: 0.70±0.04, p = 0.0355), Late Learning (Go: 0.84±0.02, p < 0.0001; NoGo: 0.83±0.02, p < 0.0001), and Expert sessions (Go: 0.87±0.02, p < 0.0001; NoGo: 0.85±0.02, p < 0.0001), but not the Reward session. Thus, despite divergent licking and pupil activity as a function of the presented texture, axonal calcium activity remained unchanged, indicating POm-striatal projections do not encode vibrissae-specific sensory information.

We next tested whether POm calcium activity correlated with trial outcome (e.g., Hit, Miss, False Alarm, Correct Rejection; Figure 2C-D). Upon segmentation (Figure S4G-J), licking and pupil dynamics exhibited discrete outcome-dependent activity patterns,¹¹⁷ but calcium activity again remained remarkably consistent (Figure 3M-P). These results were further validated by comparing the average maximum calcium amplitude of each behavioral time point between the outcome conditions: the main effects of behavioral time point (F _{(1.273,5.090)} = 22.09, p = 0.0043) and trial outcome (F_{(1,153, 4.613)} = 10.80, p = 0.0231; Figure 3Q) were significant. However, the only significant post-hoc comparisons were between Hit-Shaping and CR-Shaping (p = 0.0305) and FA-Shaping and CR-Shaping (p = 0.0114). Overall, POm-striatal projections do not appear to encode texture- or outcome-specific information, suggesting that POm may represent a behaviorally relevant arousal-related role.

Optogenetic Suppression of POm-Striatal Axons During Behavior Slows Learning Rate

Our findings indicate that POm-striatal inputs may represent an arousal signal. To investigate this possibility, wild-type mice were divided into two cohorts: a No Stim and a photoinactivation JAWS cohort. For both cohorts, a 200µm fiber optic cannula was unilaterally implanted over left pDLS. However, only the JAWS cohort received an injection of the inhibitory actuator JAWS (Figure 4A-B)^123,124,125 into ipsilateral POm, verified by the stereotypical POm-cortical projection pattern (Figure 4B).^4,47,49 Both cohorts were trained on the Go/NoGo task (Figures S3, 5). However, once the Learning phase started, POm-striatal axons of the JAWS cohort were photoinactivated for ~50% of trials per session until reaching the Expert phase. Photoinactivation occurred via an amber (617nm) LED (~7mW) for 2s of constant illumination centered around the PT window start (Figure 4C-D) where the maximum calcium signals were detected (Figure 3A). Both the No Stim and JAWS cohorts achieved Expert status (Figure S5A-B) by increasing Hit Rate, decreasing FA Rate, and with licking becoming stereotyped across learning (Figures 4E-H, S5A-B). However, photoinactivation of POm-striatal axons during the Learning phase resulted in the JAWS cohort requiring significantly more sessions (12.50±0.50 sessions, n = 4 mice) to attain Expert status compared to the other two cohorts with different experimental conditions (FP with constant illumination across the entire session: 7.60±0.51 sessions, n = 5 mice; No Stim with no illumination: 7.40±0.24 sessions, n = 5 mice) (F_(3,11) = 8.542, p = 0.0041, JAWS vs. FP p = 0.0470; JAWS vs. No Stim p = 0.0202; FP vs. No Stim p > 0.9999; Figure 4I-J). We assessed whether photoinactivation modified behavioral responding parameters between the Learning and Expert phases (Figure 3K). The only significant effect was a decrease in overall responding (Bias) during the Learning phase (Off: 0.56±0.32, On: 0.48±0.32, p = 0.0351). Due to this, we assessed if inhibition resulted in slower RTs and found that the average RT was significantly slower in the On condition (Off: 0.87±0.07, On: 0.93±0.07, p = 0.0124). Thus, the suppression of this behaviorally relevant arousal signal resulted in more learning sessions due to delayed RTs.

Figure 4.

Discussion

POm has received considerable attention for its potential roles in sensory/nociceptive processing,^{7,12,13,14,15,16,17,43} pain signaling,^{19,20,21,126,127} and cortical plasticity mediation.^{22,23,24,25,26, 34,50,51} Its widespread connectivity with sensory and motor cortical areas (S1, S2, and M1),^{2,48,49,54,55,128,129} along with its strong modulation of behavioral state and arousal,^17,18 suggests a significant role in sensory-guided behaviors. While POm-striatal projections are anatomically prominent and robustly overlap with corticostriatal inputs from S1 and M1,^{2,29,31,32,33,49,73,130} their functional innervation pattern and potential role in learned behaviors are unknown. Therefore, we aimed to elucidate the functional innervation pattern of POm onto striatal cell types, and the role of POm-striatal projections on behavioral performance and learning. Slice electrophysiology revealed strong and equal POm-mediated synaptic inputs to D1-SPNs, D2-SPNs, and PV interneurons, with shorter latency responses in PVs. In vivo photometry recordings showed increasing activation of POm-striatal axons across task learning, with axonal signals and pupil dilation in expert mice increasing prior to and becoming stereotyped to stimulus presentation, independent of stimulus type or trial outcome. Photoinactivation of POm-striatal axons resulted in prolonged RTs and delayed learning, with more training sessions required to achieve expert behavioral performance. We propose that POm-striatal projections may play a vital sensorimotor learning role by providing a salience- or arousal-related “priming” signal necessary for action selection. POm projections may have discrete target-specific functions, such that POm-striatal inputs may play a distinct role in sensorimotor behavior compared to POm-cortical inputs.^18,26,51

A striking feature of our POm-striatal projection measures in behaving mice was the early task-related activation, with the strongest increases before and during texture presentation rather than during trial outcome, reward presentation, or licking. This feature, together with our data showing broad innervation of both direct and indirect pathways (D1- and D2-SPNs) and PV interneurons in dorsal striatum, suggest that POm projections are well-positioned to prime striatal circuitry for processing subsequent synaptic input. SPNs require an up-state transition from relative hyperpolarization to near action potential threshold.¹³¹ While this up-state transition is correlated with cortical oscillatory activity,^87,132 the role of thalamostriatal projections has not been fully elucidated.¹³³ POm inputs likely arrive in striatum with shorter latency than cortical inputs following whisker stimulation,^{55,56,105,134} and therefore could be involved in initiating the up-state transition in SPN subpopulations. Indeed, we found that POm provides large amplitude PSPs to D1- and D2-SPNs, similar to previously measured M1 inputs and larger than S1 inputs⁸⁹. This non-discriminant innervation of D1- and D2-SPN is consistent with previous work on the co-activation of SPN populations during natural movements.^135,136,137 We also observed strong innervation of PV interneurons by POm. PV-mediated feedforward inhibition may mediate this priming role in different ways that depend on its effect on SPN membrane potential. Recent work showed that GABAergic input depolarizes SPNs via a relatively positive chloride equilibrium potential, increasing the action potential firing probability with further glutamatergic input.¹³⁸ Even if PV-mediated GABAergic input to SPNs instead caused hyperpolarization,^63,65,66 POm-mediated feedforward inhibition could function to selectively prevent up-state transitions of certain SPN ensembles, effectively increasing the signal-to-noise of striatal population activity, and therefore, the recruitment of SPNs by subsequent cortical inputs. To fully resolve these possibilities, experiments would need to measure POm input dynamics relative to SPN ensemble activation. In either case, we propose that POm-striatal priming could play a key role in the initial stages of action selection via non-discriminant effects over SPN up-state transitions.

We have focused on the initial volley of POm-striatal activity, but POm is part of an interconnected circuitry that could signal to striatum in multiple iterations on a behaviorally relevant timescale. POm-cortical axons bifurcate and contact neurons within pDLS,^{49,52,53,54,55,57,58} but the main terminations continue to cortex where they strongly innervate S1 L1 and L5a.^48,49,54,55 Notably, L5a neurons comprise the predominant S1-striatal projection,^68,69,76 creating a secondary POm-S1-pDLS loop in addition to the direct POm-pDLS projection. Thus, the striatal microcircuitry may be initially engaged via the direct POm projection followed by the secondary loop that recruits L5a pyramidal neurons to provide more processed input to primed SPN ensembles. Next, our hypothesis is simplistic as it does not fully encompass other extrinsic and intrinsic factors that influence striatal functioning. For example, the striatum receives inputs from a plethora of subcortical regions including other thalamic nuclei^73,99 and external globus pallidus.^139,140 Further, local dopamine release has profound effects on striatal functioning directly⁶² and indirectly via cholinergic interneurons.^141,142 However, POm activation early in behavior may set the stage for these other significant inputs to influence action selection at later behavioral stages.

Another crucial factor in POm signal timing is its inhibitory gating by two GABAergic inputs, ventral zona incerta (vZI)^{16,35,36,37,42} and anterior pretectal (APT).^16,41,143 All three nuclei receive ascending spinal trigeminal (whisker-related sensory) inputs, but vZI and APT efficiently shunt incoming sensory information via feedforward inhibition onto POm neurons.^16,39,44 This inhibitory gating is overcome by (1) arousal-related cholinergic suppression of presynaptic GABA release within POm⁴³ or (2) convergence of bottom-up and top-down signals within a specified temporal window.^31,144 Both factors are likely at play during sensory-guided behavior. The involvement of cholinergic signaling suggests that a POm-striatal priming effect would take place during a behaviorally relevant period of sensation, or immediately prior to sensory information becoming perceptible.¹⁴⁵ In addition to cholinergic, cortical, and subcortical GABAergic afferents, POm also receives direct glutamatergic input from superior colliculus (SC) that enhances sensory responses.^27,34 SC, a region implicated in attentional orienting of either somatosensory^34,146 or visually-relevant^147,148 stimuli, bidirectionally modulates POm, further ascribing a potential arousal-related functional role. It is yet to be resolved whether POm-striatal inputs are driven by ascending sensory information, descending cortical-POm feedback, or the convergence of both.

A critical issue is whether other sparse, yet powerful striatal interneurons are recruited by POm-striatal projections. Notably, Pf-striatal projections engage cholinergic interneurons to mediate its robust behavioral effects.^{72,87,90,102,104,149,150} Additionally, other striatal interneurons also respond to thalamic stimulation including tyrosine-hydroxylase (THIN) and neurogliaform (NGF) interneurons.^67,109 Determining the precise circuitry of POm-striatal inputs relative to corticostriatal inputs^86,89,90 will be important for understanding how these inputs are integrated. Other thalamic nuclei, specifically Pf, tend to synapse onto dendritic shafts.^151,152,153 If POm preferentially contacts dendritic shafts, it may induce dendritic filtering, creating a longer temporal window for cortical input to arrive.^154,155 Conversely, if POm preferentially contacts dendritic spines, that temporal window would be restricted, likely requiring near coincident arrival of thalamostriatal and corticostriatal projections onto the same or nearby spines.⁵⁶

A technical limitation is that our photometry measure only captured a bulk axonal signal. POm neurons exhibit heterogeneous responses to direct paralemniscal stimulation,¹⁰ peripheral stimulation,¹⁵ and the suppression of vZI activity,³⁹ which may be due to functionally distinct anterior and posterior subpopulations;^2,49,54,156 therefore, the role of individual POm axons cannot be resolved. It is possible that POm-cortical axons of passage contributed to the recorded signal. However, we contend that even if signal contamination is present, the measures are likely similar as most POm-cortical projections bifurcate within striatum rather than projecting exclusively to striatum or cortex.^49,55 Further, we placed the optical fiber above pDLS to specifically record from axons, rather than somas, as signal modulation can occur within the axon itself.¹⁵⁷ Finally, signal modification is more likely to occur at synapses via presynaptic or postsynaptic mechanisms,^158,159 potentially permitting discrete functional roles (e.g., internal state monitor, cortical plasticity modulation, priming signal) at different neuronal regions.^18,26

A critical issue is how POm activation is related to behavioral state¹⁸ and how this impacts perceptual learning.^26,51 Overall, we observed a learning-related increase in POm-striatal activity that correlated to pupil dilation in many but not all phases of behavioral performance, and a necessity of these projections for efficient behavioral performance and learning, supporting a role for POm-striatal projections in task-related behavioral arousal. First, POm-striatal axonal activity, pupil dilation, and licking markedly increased across learning with photometry signals becoming stereotyped to texture presentation, and licking becoming stereotyped within the PT window. Photometry signals and pupil dilation were tightly correlated between trial onset and texture presentation, but decoupled within the PT window, with photometry signals decreasing prior to licking and reward delivery. Second, when sorting trials for presented texture (stimulus) or trial outcome (response), photometry signals remained elevated, while both licking and pupil dilation exhibited stimulus- or response-specific changes. This suggests that POm-striatal projections do not encode sensory- or outcome-related information, but rather arousal or salience during anticipatory states of a learned behavior. Finally, photoinactivation of these projections delayed RTs on individual trials, resulting in more sessions required to achieve expert task performance.

In summary, we show that POm-striatal projections encode a behaviorally-relevant arousal-related priming signal, which may contribute to inducing the up-state transition of SPN ensembles necessary for action selection by equally engaging three prominent striatal neurons. This finding suggests a previously unknown functional role of POm priming the striatal microcircuitry. It will be important for future studies to investigate whether POm further innervates other striatal interneurons, to assess the timing between POm-striatal and S1-striatal inputs onto both SPNs, and to assess whether POm-striatal synapses undergo synaptic plasticity across learning.

Acknowledgements

The authors would like to thank all members of the Margolis lab for their comments related to manuscript text and figures. This work was generously supported by funding from the National Institutes of Health (F31NS117093, A.J.Y.); (NCATS TL1TR003019, A.J.Y.); (R01NS094450, D.J.M.), the National Science Foundation (IOS-1845355, D.J.M.), and the Rutgers Busch Biomedical Grant Program (I. L./D.J.M.).

Author Contributions

A.J. Yonk and D.J. Margolis conceived and designed the research strategy. A.J. Yonk performed and analyzed electrophysiology experiments with assistance from M.A. Gradwell and A.J. George. A.J. Yonk performed and analyzed fiber photometry behavioral experiments with assistance from I. Linares-Garcia, L. Pasternak, S.E. Juliani, and A.J. George. A.J. Yonk performed and analyzed optogenetic behavioral experiments with assistance from I. Linares-Garcia and A.J. George. A.J. Yonk prepared the figures with input from I. Linares-Garcia, S.E. Juliani, M.A. Gradwell, A.J. George, and D.J. Margolis. A.J. Yonk wrote the paper with all authors contributing to its editing. D.J. Margolis supervised the study. All authors approved the final version of the manuscript.

Declaration of Interests

The authors declare no competing interests.

STAR ★ Methods

Key Resource Table

Resource Availability

Lead Contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, David Margolis (david.margolis@rutgers.edu).

Materials Availability

This study did not generate new unique reagents.

Data and Code Availability

Data are available upon request from the Lead Contact, David Margolis (david.margolis@rutgers.edu).

Experimental Model Details

Animals

All work involving animals including housing, surgery, behavioral experimentation, and euthanasia was approved by the Rutgers Institutional Animal Care and Use Committee (protocol #: 999900197). Mice were group housed in a reverse light cycle room (lights on from 20:00 to 08:00) with food and water available ad libitum with the exception of mice undergoing water restriction during behavioral experiments. All handling and experiments were conducted within the dark phase of this light cycle. Regardless of their experimental designation, all experimental animals underwent a unilateral AAV injection or a simultaneous AAV injection and cannula implant between 45 - 65 days (average: 52.37 days, range: 47 - 61 days). Briefly, male and female double transgenic mice were used for electrophysiology experiments. To identify specific neuronal populations during electrophysiology, Ai14 mice (cre-dependent tdTomato; The Jackson Laboratory, #007914) were crossed with either (1) D1-SPN cre (B6.FVB(Cg)-Tg(DrD1cre)EY262Gsat/Mmucd; MMRRC, #030989), (2) D2-SPN cre (B6.FVB(Cg)-Tg(Adora2a-cre)KG139Gsat/Mmucd; MMRRC, #036158), or (3) PV-cre (B6.129P2-Pvalb^tm1(cre)Arbr/J; The Jackson Laboratory, #017320) mice. This permitted red fluorescence in the specific cell population for visual identification via epifluorescent illumination. Animals designated for electrophysiology were euthanized between 2 - 4.5 months. Both male and female wild-type mice (The Jackson Laboratory #000664) were used for fiber photometry and optogenetic experiments. To motivate behavioral performance, daily water intake was restricted to ~1.5mL per mouse per day. Body weight was carefully controlled and never permitted to drop below 80% of a calculated baseline value.¹⁶⁹

Method Details

Electrophysiology Adeno-Associated Viral (AAV) Injection

Male and female double transgenic mice designated for electrophysiology experiments underwent a unilateral injection of channelrhodopsin-2 (ChR2; pAAV-hSyn-hChR2(H134R)-EYFP; Addgene #26973)¹⁶⁰ targeting the left posterior medial (POm) thalamic nucleus. Briefly, mice were anesthetized using isoflurane (4% induction, 1-2% maintenance) and moved into a stereotaxic apparatus (Stoelting/Kopf Instruments) containing a feedback-controlled heating pad on the base (maintained between 35-37°C; FHC). Ophthalmic ointment (Akorn) was applied to the eyes to prevent them from drying out. Ethiqa XR (3.25 mg/kg; Fidelis Animal Health) and Bupivacaine (0.25%, 0.1mL, Fresenius Kabi) were injected subcutaneously into the right flank and scalp, respectively. After, the scalp was sterilized by three cycles of Betadine (Purdue Products) and 70% ethanol, a midline incision was made. The exposed skull was cleared of fascia and leveled by confirming that bregma and lambda coordinates were on the same dorsoventral plane. A craniotomy was drilled (coordinates with respect to bregma: anteroposterior (AP) = −2.05 mm, mediolateral (ML) = +1.35 mm, dorsoventral (DV) = −3.25 mm) and the micropipette containing ChR2 was slowly lowered to the appropriate depth and permitted to sit for 5 minutes. Following this, ~100 nL of ChR2 viral solution was injected over the course of ~15 minutes via the Nanoject III system (Drummond Scientific). After an additional delay of 12 minutes, the micropipette was slowly raised, the scalp was sutured and secured with tissue glue. Immediately following surgery, mice were transferred to a clean cage on top of a heating blanket until ambulation was observed. Mice were continually monitored for 72 hours post-surgery. After this observation period, mice were permitted to recover for at least three weeks before undergoing electrophysiological experiments, permitting ChR2 expression into POm axon terminals in the striatum.

Whole Cell Patch Clamp Recordings

Mice were briefly induced (via 3% isoflurane), deeply anesthetized with an intraperitoneal injection of ketamine-xylazine (300/30 mg/kg, respectively), and transcardially perfused with recovery artificial cerebrospinal fluid (ACSF), which contains the following (in mM): NMDG 103, KCl 2.5, NaH₂PO₄ 1.2, NaHCO₃ 30, HEPES 20, Glucose 25, HCl (1N) 101, MgSO₄ 10, Thiourea 2, Sodium Pyruvate 3, N-Acetyl-L-Cysteine 12, and CaCl₂ 0.5 (saturated with 95% O₂ and 5% CO₂).^89,109 Following decapitation, the brain was rapidly extracted and submerged in recovery ACSF before being mounted onto a VT-1200S vibratome (Leica). The vibratome chamber was filled with oxygenated recovery ACSF, and 300µm slices were cut. Slices were immediately transferred to recovery ACSF that was heated to 35°C for ~5 minutes. After, slices were transferred to RT external ACSF which contained (in mM): NaCl 124, KCl 2.5, NaHCO₃ 26, NaH₂PO₄ 1.2, Glucose 10, Sodium Pyruvate 3, MgCl₂ 1, and CaCl₂ 2 (saturated with 95% O₂ and 5% CO₂), and slices were allowed to recover for at least ~45 minutes before recording. Once the hippocampus began to appear, vibratome sectioning was terminated, and the posterior tissue block containing the injection site was transferred into 10% neutral-buffered formalin for post-hoc confirmation.

Whole-cell patch clamp recordings were acquired from slices that were constantly superfused (2-4mL/min) with oxygenated external ACSF at ~34°C. Slices and cells were visualized by infrared differential interference contrast (IR-DIC) microscopy using a CCD camera (Hamamatsu) mounted onto a BX-51WI upright microscope (Olympus) fitted with a swinging objective holder containing two switchable lenses: a 4X air lens and a 40X water-immersion lens. Patch pipettes (3-5 MΩ) were pulled from borosilicate glass micropipettes (2mm O.D., Warner Instruments) using a P-1000 horizontal puller (Sutter Instruments).

Current-clamp recordings were obtained from unidentified and identified striatal neurons in mice expressing tdTomato in either D1-SPNs, D2-SPNs, or PV cells within pDLS (−0.34 to −1.22mm relative to bregma), which is known to receive POm projections.^55,56,105 The internal pipette solution for current-clamp experiments contained (in mM): K Methanesulfonate 130, KCl 10, HEPES 10, MgCl₂ 2, Na₂ATP 4, Na₂GTP 0.4, at pH 7.25 and 285-290 mOsm/L. Further, 2% biocytin was freshly dissolved in the internal solution on the recording day. ChR2 in the POm axon terminals was stimulated via illumination with a 2.5 ms, 470 nm LED light pulse (~0.6 mW measured after the objective; Thorlabs) delivered through the 40X objective lens. The illumination spot size had a diameter of ~550 µm. After patching, the internal solution was permitted to dialyze for ~5 minutes. At this point, the resting membrane potential was recorded. All cells were held at −80 ± 2mV to ensure equal driving forces when studying synaptic strength and short-term synaptic plasticity. Baseline voltages that drifted outside of this range were excluded from analysis. Patched cells were held for ~25-35 minutes while being run through a standardized set of protocols: (1) hyperpolarizing/depolarizing current steps, (2) single pulse (SP), (3) paired pulse ratio (PPR), and (4) train stimulation. Occasionally, unidentified cells within the same FOV were sequentially patched following the initial identified cell patch to control for injection site volume and location. After breaking in, the cell was allowed to recover for ~5 minutes before being subjected to hyperpolarizing and depolarizing current steps (−300 pA to +400 pA, 50 pA steps, 500 ms, 15 sweeps) for cell health and intrinsic parameter confirmation. For the SP protocol, a single 2.5 ms blue light pulse was presented once every 15s for 20 sweeps. For the PPR protocol, five 2.5 ms pulses, separated by 50 ms inter-pulse intervals (IPI), were presented once every 15s for 20 sweeps. For the train protocol, thirty 2.5 ms pulses, separated by 64.2 ms IPI (15 Hz), were presented once every 30s for 5 sweeps. While these protocols were being run, biocytin within the internal solution diffused into the cell for subsequent 3D morphological reconstructions. Data were acquired via a EPC10USB amplifier and digitized at 20 kHz in Patchmaster Next (HEKA). Liquid junction potential was not corrected in these traces.

Analysis of Patch Clamp Recordings

All data were analyzed offline using custom-written MATLAB and Python scripts. DAT files from Patchmaster Next (HEKA) were imported, organized, and saved as a mat variable. The mat variable data was imported into Python for post-processing using the electrophysiology feature extraction library (eFEL) created by the Blue Brain Project.¹⁶¹ All pertinent intrinsic value parameters were calculated at the +350pA current step. Key values pertaining to every action potential in a sweep were calculated via eFEL including the action potential threshold value and index, the peak value and index, and the corresponding minimum afterhyperpolarization (AHP) value and index. Each action potential threshold value and index were captured using a derivative threshold method (dV/dt ≥ 10mV/s). Action potential properties were assessed for all spikes in a sweep and averaged together. The action potential peak was defined as the difference between the action potential threshold and its maximum positive peak. Half-height width (HHW) and rise time were both calculated via interpolation. HHW was measured at 50% of the action potential peak, while rise time was measured from 10-90% of the action potential peak. AHP amplitude was calculated as the difference between the action potential threshold value and the minimum AHP value. Maximum frequency

For all optogenetically-evoked postsynaptic potentials (PSPs), a baseline measure was calculated by averaging the first 10,000 sampling points for each individual sweep. This measured value was subtracted from every value in each individual sweep, setting the baseline equal to zero. The maximum PSP amplitude, relative to the zero baseline value, was calculated from the average of 20 sweeps during SP and PPR protocols, and 5 sweeps during Train protocols. The latency to maximum PSP amplitude was measured as the difference between photostimulation onset and the maximum PSP index. For PPR and Train protocols, an exponential function was fitted to the decay component of each PSP, and their amplitudes were extracted after subtracting the decay of preceding PSPs.

Simultaneous AAV Injection and Optical Cannula Implantation Surgery

For male and female wild-type in the designated fiber photometry cohort, a 400µm core optical cannula (ferrule OD = 2.5mm, length = 2mm, nA = 0.50; Thorlabs #CFM15L02) was implanted directly above the left pDLS along with a ~100nL unilateral injection of axon-jGCaMP8s (pAAV-hSynapsin-axon-jGCaMP8s-P2A-mRuby3; Addgene #172921)¹⁶² into the ipsilateral POm. Male and female wild-type mice designated for optogenetic manipulation were split into two cohorts: (1) the No Stim or (2) the JAWS cohort. Both optogenetic cohorts were implanted with a 200µm core optical cannula (ferrule OD = 1.25mm, length = 2mm, nA = 0.50, CFMLC52L02, Thorlabs) above the left pDLS along with a ~100nL AAV injection into the ipsilateral POm. However, the No Stim cohort was injected with the excitatory optogenetic actuator, ChR2 (pAAV-hSyn-hChR2(H134R)-EYFP, Addgene #26973),¹⁶⁰ while the JAWS cohort was injected with the inhibitory optogenetic actuator, JAWS (pAAV-hSyn-JAWS-KGC-GFP-ER2, Addgene #65014).^{123,124,125,163} Further, an HHMI head plate¹⁶⁴ was fitted to each mouse using methods previously described.^{89,117,165,166,167} Briefly, mice were anesthetized with isoflurane (4% induction, 0.8 – 1.5% maintenance) and mounted within a stereotaxic frame (Stoelting/Kopf Instruments) with a feedback-controlled heating pad (FHC) maintaining the body temperature between 35 – 37°C. Following administration of an analgesic (Ethiqa XR, 3.25mg/kg; Fidelis Animal Health) to the left flank and a local anesthetic (0.25% Bupivacaine; Fresenius Kabi) under the scalp, the scalp was sterilized with a triple cycle of Betadine (Purdue Products) followed by 70% ethanol (Fisher). A midline incision was made, and a circular piece of scalp was removed to expose the skull. Both lateral muscles and the nuchal muscle were separated from the skull. The skull was cleaned by gently scraping away the periosteum. A dental etch bonding agent (iBond; Heraeus Kulzer) was applied to the clean skull and cured with blue light for 20s twice. A ring of dental composite (Charisma; Heraeus Kulzer) was applied to the outer edge of the skull and cured with blue light. After ensuring that bregma and lambda coordinates were equal in the dorsoventral plane, two craniotomies were made: one above the left posterior striatum (AP = −0.80mm, ML = +2.8mm, DV = 1.8mm) for cannula implantation and the other above left POm (AP = −2.05mm, ML = +1.35mm, DV = −3.25mm) for the corresponding AAV injection. The AAV injection was always performed first. The micropipette containing axon-jGCaMP8s (fiber photometry cohort), ChR2 (No Stim cohort), or JAWS (JAWS cohort) was slowly lowered to the appropriate depth and permitted to sit for 5 minutes. ~100nL of AAV solution was injected over the course of ~15 minutes via the Nanoject III system (Drummond Scientific). After an additional delay of 12 minutes, the micropipette was slowly raised. Once the optic cannula was secured in the stereotaxic frame and lowered to the appropriate DV coordinate, it was secured with dental composite (Tetric Evoflow; Heraeus Kulzer) that was cured with blue light. The HHMI headpost was delicately secured on the posterior area of the charisma ring with cyanoacrylate and dental composite. Finally, a single layer of dental composite was used to build and secure the rest of the headcap before being cured four times with blue light for 20s each. The scalp was closed around the headcap by using cyanoacrylate.

Immediately after the optical implant and viral injection surgery, mice were placed in a sterile, clean cage that was half-on/half-off a heating pad until ambulation was observed. Mice were diligently monitored for 72 hours post-surgery. After this monitoring period, mice were transferred to a clean cage on a ventilated rack for at least three weeks prior to handling and water restriction.

Handling and Water Restriction

After a three week waiting period, mice were placed under citric acid water restriction¹⁶⁸ during which mice were handled twice daily for one week. The bitterness of citric acid naturally causes mice to reduce their water consumption and, consequently, their weight while still having access to water. Initially, mice were acclimated to handling as researchers placed their hands into the cage for increasing amounts of time (e.g., 5 minutes to 10 minutes to 15 minutes). After mice became comfortable, they were held for increasing amounts of time (e.g., 2 minutes to 5 minutes to 10 minutes). Additionally, mice were acclimated to head fixation by holding their HHMI headposts (e.g., 30 seconds to 1 minute to 2 minutes), and they were allowed to freely explore the behavioral tube for 5 minutes per handling session. Finally, mice were headfixed for increasing amounts of time in the behavioral setup (e.g., 5 minutes to 10 minutes to 15 minutes). Water was provided via transfer pipette to comfortably acclimate mice to head fixation. The head-fix apparatus contained a tube (length = 15cm; inner diameter = 4cm) that was affixed to a custom metal platform (length = 17cm; width = 12cm). The platform also contained HHMI mounting arms and holders for head fixation.¹⁶⁴ During the final day of handling, mice were transitioned to full water restriction as this permits greater control over motivation level. During behavioral testing, daily water intake was limited to ~1.5mL per day to motivate performance on the Go/NoGo paradigm described below. The baseline body weight was measured daily during water restriction, and overall body weight was not permitted to drop below 80% of the baseline weight, consistent with levels of restriction used to motivate performance.¹⁶⁹

Fiber Photometry Setup

Fiber photometry data were collected using a RZ10x lock-in amplifier within the Synapse suite (Tucker-Davis Technologies). This amplifier and accompanying Synapse software was used to control a custom-built optical benchtop through drivers (LEDD1B, Thorlabs) to modulate LED signals. Briefly, this optical benchtop consisted of a self-contained system of four 30mm cage cubes with integrated filter mounts (CM1-DCH/M, Thorlabs). A 405nm LED (M405L4, Thorlabs) and a 470nm LED (M470L3, Thorlabs) were mounted onto the first 30mm cage cube. The 405nm LED was used to extract the calcium-independent isosbestic signal, and the 470nm LED was used to acquire calcium-dependent axonal GCaMP signals during the Go/NoGo paradigm. The 405nm isosbestic signal was modulated at 210Hz, and the 470nm GCaMP signal was modulated at 330Hz. A 425nm dichroic longpass filter (DMLP425R, Thorlabs) in the first cage cube reflected the 405nm excitation light and permitted the 470nm light to pass through. As the light entered the second cage cube, both excitation lights were reflected by a 495nm dichroic longpass filter (495DCLP, 67-079, Edmund Optics) into the third cage cube. A 460/545nm bandpass filter (69013xv2, Chroma) reflected both excitation wavelengths down to the subject via a low autofluorescence patch cable (MAF3L1, core = 400µm, nA = 0.50, length = 1m, Thorlabs). This cable was attached onto the implanted optical cannula (see above) by a ceramic mating sleeve (Thorlabs). Isosbestic and axon-jGCaMP8s emissions were collected via the optic cannula and passed through the 460/545nm bandpass filter (69013xv2, Chroma)into the fourth. Finally, the emission fluorescence passed through the detection pathway to reach the RZ10x photosensors for online observation.

Orofacial Video Recording

POm activity has been well correlated with whisking and pupil activity.^18,26 To analyze these dynamics, an LED driver controlled an IR spotlight that illuminated the contralateral eye and mystacial pad, and facial recordings were captured through an autofocusing USB webcam (NexiGo N660P) at ~20fps within the Synapse software. To limit light pollution from outside sources (e.g., LED illumination within the brain/eye), an IR filter was placed in front of the webcam. Also, a shortpass emission filter at 750 nm (Chroma #ET750sp-2p8) was placed between the third and fourth cage cubes to prevent recorded IR light from overloading the photosensors.

Synchronization of Task-Related Components

The Synapse software suite (Tucker-Davis Technologies) permitted the synchronous recording of both emitted (isosbestic and calcium) signals and video frames. Furthermore, the LabVIEW system controlled paradigm-related components (e.g., texture movement, lick thresholds, trial type, trial outcome, etc.) and recorded the resulting behavioral parameters (e.g., licking activity, trial type, and trial outcome). To synchronize these two data streams for post-hoc analysis, TTL pulses relating to the texture arrival at its endpoint (which indicates the start of the presentation time window) from the LabVIEW system were captured within the Synapse software.

In Vivo Optogenetics

For the JAWS cohort, a high-powered 617nm LED (M617F2, Thorlabs) and current driver (LEDD1B, Thorlabs) were used for photoinactivation of POm axons in the striatum.^{123,124,125,163} LED stimulation was provided at a probability of 0.50 for every session from the start of the Learning phase until mice reached the Expert phase.²⁶ The light intensity was measured at ~7mW at the tip of the fiber. Stimulation intensity was kept consistent between mice and days by measuring the intensity with an optical power meter (PMD-100D, Thorlabs) prior to the first session every day. Light was delivered to the thalamic afferents in pDLS through an optical fiber patchcord (M95L01, fiber core diameter: 200µm, length: 1m, nA: 0.50) connected to a 200µm core optical cannula (described above) via a mating sleeve (Thorlabs). A small piece of black heat shrink tubing (Qualtek) was placed over the cannula during LED testing to prevent stray light from illuminating the presented texture during the task. The LabVIEW system controlled a Pulse Pal¹⁷⁰ that activated the LED current driver to provide constant illumination for two seconds, evenly split 1s before and 1s after texture presentation, corresponding to the increased calcium activity observed during fiber photometry recordings.

For the No Stim cohort, a high-powered 470nm LED (Prizmatix) and current driver (Prizmatix) were used for optogenetic activation. Note that no LED stimulation was provided during the Learning or Expert phase. Testing occurred during the first 4 sessions after the 5 initial Shaping sessions and the last 4 sessions after mice reached Expert status. Sessions consisted of 50 baseline trials followed by 10 alternating blocks of 10 OFF and 10 ON trials. The light intensity was measured at ~5mW at the tip of the fiber and was kept consistent between mice and days by measuring the intensity with an optical power meter (PMD-100D, Thorlabs) prior to the first session on testing days. Light was delivered to the thalamic afferents in pDLS through an optical fiber patchcord (M73L01, fiber core diameter: 200µm, length: 1m, nA: 0.50, Thorlabs) connected to a 200µm optical cannula (described above) via a mating sleeve (Thorlabs). A small piece of black heat shrink tubing (Qualtek) was placed over the cannula during LED testing to prevent stray light from illuminating the presented texture during the task. The LabVIEW system controlled a Pulse Pal¹⁷⁰ that activated the LED current driver to provide pulsed illumination for 2s at 15Hz (matching the electrophysiology train photostimulation paradigm), evenly split 1s before and 1s after texture presentation.

Go/NoGo Whisker-Based Discrimination Paradigm

Headfixed, water restricted mice were trained to perform a whisker-based discrimination paradigm as two textures were presented unilaterally to the right whiskers in a randomized order based on a custom-written LabVIEW code (National instruments). This code used transistor-transistor logic (TTL) pulses to control all aspects of the paradigm including a water delivery spout connected to a piezo film sensor (MSP1006-ND; Measurement Specialties), and a motorized linear stage (T-LSM100A; Zaber) with a stepper motor (T-NM17A04; Zaber) containing two windmill arms holding two different sandpaper textures (Go texture = 100 grit sandpaper, P100; NoGo texture = 1200 grit sandpaper, P1200; 3M) as previously described.^{89,117,165,166,167} Mice were trained to discriminate between these two textures by licking the piezo film sensor spout. Mechanical spout displacement resulted in transient voltage changes, and a lick was defined as voltage changes crossing either an upper or lower threshold once. After a lick was detected within the appropriate response window, the LabVIEW software immediately delivered the appropriate output: for Go trials, mice received a small water reward via a solenoid valve (0127; Buerkert) through the piezo spout, and, for NoGo trials, mice received a timeout period with co-occurring white noise. This paradigm occurred within a darkened room to minimize non-tactile related cues. Both textures and the headfix apparatus were cleaned with 70% ethanol between each mouse. If mice were not performing the task, sessions could be ended early. Water could be automatically delivered (AutoReward or AR) by the experimenter following 20 consecutive trials without a response. Finally, a session could be terminated if a mouse did not lick when water was present on the end of the spout after three ARs.

Trials began with a 1000ms pre-task interval followed by a brief cue tone (100ms, 2039Hz) that accompanied windmill texture movement towards the mice. Once the windmill texture had reached a predetermined distance within the whisking field, mice had to respond within a 2000ms presentation time (PT) window. For the first 500ms of the PT window, a grace period was present where responses did not trigger appropriate outcomes to reduce impulsivity. If mice licked in response to the Go texture, the trial was considered a Hit and resulted in a water reward accompanied by a correct tone (2793Hz). If mice licked in response to the NoGo texture, the trial was considered a False Alarm (FA) and resulted in punishment parameters including a time-out period (12000ms) and an accompanying white noise during the time-out period. Air puffs were eschewed as a punishment parameter to permit continuous pupil dynamic recording. If mice did not lick to the Go or the NoGo texture, nothing occurred, and the trial was considered a Miss or a Correct Rejection (CR), respectively. Immediately after a lick was recorded or the 2000ms PT window had elapsed, the windmill texture retreated to its original position where the current texture either remained or the other texture was rotated into position. Finally, trials were separated by a 2000ms intertrial interval. Behavioral performance was tracked across texture discrimination sessions by computing multiple behaviorally-related parameters including Hit Rate, FA Rate, Sensitivity (d’) and Bias.¹⁷¹ For the fiber photometry cohort, the 405nm and 470nm LEDs provided constant illumination throughout all trials. For the JAWS cohort, LED photoinactivation occurred at a trial probability of 0.50 for every session from the start of the Learning until the end of the Expert phase.²⁶ For the No Stim cohort, LED stimulation did not occur during training (i.e., during either the Learning or Expert phases). It only occurred for 4 sessions after the initial 5 shaping sessions, and 4 sessions after mice reached expert status.

This whisker-based discrimination paradigm lasted up to 3 weeks. The FP cohort were only tested once per day to limit photobleaching. The JAWS and No Stim cohorts were tested twice daily. Each session consisted of 150 trials. All cohorts progressed through three behavioral phases (Shaping, Learning, and Expert) that were segmented into five analytical time points (Shaping, Early Learning, Late Learning, and Expert). The Shaping time point was the same for all mice. For the first three sessions, mice were acclimated to reliably trigger water delivery by licking the water spout. During these sessions, neither texture was presented to the whiskers. After, mice proceeded to texture discrimination training still under the Shaping phase. For the last two sessions, both textures were presented simultaneously with a Go texture probability of 0.90 and 0.75, respectively. Go and NoGo trials were interleaved in a pseudorandom order determined by the LabVIEW software. After the 0.75 probability session, mice progressed into the Learning phase. For all following sessions, the Go texture probability was set to 0.50 with a maximum of three consecutive presentations of the same texture. Discrete behavioral time points were determined as follows. The Early Learning time point was considered the first two sessions of the Learning phase. The Late Learning time point was considered the last two sessions of the Learning phase before achieving expert status. This occurred when mice had a Hit Rate ≥ 0.80 and a FA Rate ≤ 0.30 for two consecutive sessions. A strict sensitivity threshold was not used due to artificially increased sensitivity (discrimination) values as Hit Rate and/or FA Rate approach their extremes (e.g., see FPOm-18 sessions 8 to 15 in Figure 6). After discrimination training (i.e., achieving expert status), mice in the fiber photometry cohort were subjected to a single Reward session to assess calcium, licking, and pupil activity in the absence of texture input and licking-related outcomes. During this session, the Go trial probability remained at 0.50, and the Zaber motor moved the windmill texture holder towards the whisker field, but the textures were rotated out of whisker range. Further, the upper and lower thresholds were set so that licking could not trigger outcomes before the end of the PT window. A water reward was automatically delivered at the end of the PT window during Go trials only. A whisker trim control session was performed in a subset of mice to confirm that mice used their whiskers to discriminate as previously observed.⁸⁹

Histology

For electrophysiology experiments, the tissue block containing the injection site was stored overnight in 10% neutral-buffered formalin at 4°C. After, it was transferred into 0.2% PBS Azide at 4°C until sectioning. The tissue block was mounted onto a stage and sectioned in 0.1M PBS at a thickness of 100µm using a VT-1000 vibratome (Leica). Slices were mounted onto microscope slides using DAPI Fluoromount-G (Southern Biotech #0100-20) and coverslipped before confocal imaging.

Following all behavioral experiments, mice were anesthetized with an intraperitoneal injection of Ketamine-Xylazine (120mg/kg Ketamine, 24 mg/kg Xylazine) and transcardially perfused with PBS followed by 10% neutral-buffered formalin. The brain was delicately extricated and stored in 10% neutral-buffered formalin overnight at 4°C. Tissue was mounted onto a stage and sectioned at 100µm using a VT-1000 vibratome (Leica). Sections were mounted and coverslipped using DAPI Fluoromount-G (Southern Biotech #0100-20). Confocal images were acquired using a LSM800 confocal laser scanning microscope (Zeiss) for injection site location verification, cannula placement, and viral expression in POm axons within pDLS and stereotypical POm-cortical projections in S1 L1 and L5a of all experimental mice.^{4,25,26,47,49} All data were acquired using the Zen software suite (Zeiss).

Quantification and Statistical Analysis

Behavioral Responding Analysis

Behavioral performance was tracked across texture discrimination sessions by computing multiple behaviorally-related parameters including Hit Rate, FA Rate, Sensitivity (d’) and Bias.¹⁷¹ Hit Rate was calculated as follows: [Hit / (Hit + Miss)], where hit is the number of correct Go trials and Miss is the number of incorrect Go trials. The FA Rate formula is similar to the Hit Rate formula except it replaces Hit with FA and Miss with CR: [FA / (FA + CR)]. Sensitivity illustrates the ability to discriminate between the signal (Go texture) and the noise (NoGo texture), and it is derived from signal detection theory.¹⁷¹ It is calculated as follows: [normalized inverse (Hit Rate) - normalized inverse (FA Rate)]. Finally, Bias illustrates the overall responding bias, independent of trial type. It is calculated as follows: [0.5 * (normalized inverse (Hit Rate) + normalized inverse (FA Rate)].

Pupil Analysis

After all data were captured, the recorded orofacial videos were analyzed using DeepLabCut, a deep learning model for pose estimation, to estimate pupil dynamics as mice learned to discriminate between the two textures.^118,119 Briefly, a model¹²⁰ was created with eight markers circumscribing the pupil, permitting the estimation of the approximate pupil area for each frame. All videos were cropped to a smaller dimension (230 x 275 pixels) that focused on each mouse’s face. For each mouse, one video was selected at varying behavioral time points, and 30 frames were extracted and manually labeled with the eight markers. Each marker corresponded to a pupil location: top, top right, right, bottom right, bottom, bottom left, left, and top left. Additionally, another marker was placed on a static location (e.g., the water spout) to label frames when blinking occurred. Overall, the initial training dataset contained 150 labeled frames, and the model was trained on this dataset with a ResNet50-based neural network for 250,000 iterations. After, the initial training videos were analyzed to assess the performance of the model from its marker estimations. 25 outlier frames from the initial training videos were extracted, manually corrected, and merged with the initial dataset. The model was trained again on this 275 labeled frame dataset with a ResNet50-based neural network for 250,000 iterations. After evaluating the network, the train error was calculated at 1.24 pixels, and the test error was measured at 1.25 pixels. Once the model was successfully trained, all behavioral videos of the five mice were analyzed. To estimate pupil area, a python library (scikit-image)¹⁷² was used to fit an ellipse on the estimations of the eight markers generated by DeepLabCut. This ellipse model was used to predict the values of the vertices and co-vertices on the ellipse, permitting the calculation of the major and minor axes. Thus, these measurements were used to approximate the area of the pupil for each frame. To detect blinking behavior, the maximum pupil size in non-blinking conditions was calculated and used as a threshold to identify abnormally high pupil predictions. As such, pupil areas greater than 450 pixels were removed via interpolation. The pupil area data were saved as a csv file for importing into MATLAB.

Fiber Photometry Signal Processing

Custom-written MATLAB scripts were used for post processing of the fluorescent signals. Photobleaching was corrected in both the isosbestic and GCaMP signals using detrended lines of best fit and subtracting the line from all values. After, the isosbestic and GCaMP median absolute deviation of z-score (ZMAD) signals were calculated before subtracting the GCaMP signal from the isosbestic to remove calcium-independent artifacts.

Alignment of Task-Related Components

A custom-written MATLAB script imported the lick-related and overall responding data, the processed ZMAD GCaMP signal, and the processed pupil csv for alignment. The overall responding data (containing trial elements such as time of PT window start which is the TTL flag within Synapse) was converted from UNIX timecode into seconds to match the Synapse (containing the pupil video and processed GCaMP signal) time. Pupil data was resampled to align with GCaMP signal using rational fraction approximation. Furthermore, the length of each trial, as determined by the overall responding data, was recorded and used to capture the GCaMP and pupil data within each trial window. Licks were identified by setting upper and lower thresholds, and detecting when either was crossed. At this point, all parameters (e.g., GCaMP, pupil, and licking activity) were captured and aligned within the time window of each trial. These parameters could now be segmented by trial type (e.g., Go vs. NoGo texture) and by trial outcome (e.g., Hit, Miss, FA, CR) for more advanced analysis.

Calcium Analysis

auROC is an analysis commonly applied to calcium imaging data to characterize the stereotypy of neuronal responses.^121,122 Briefly, a baseline window is set within a non-task-related component of the overall calcium signal and compared to the rest of the signal via a sliding window. The maximal value of each signal is captured at each behavioral time point and averaged across the cohort. auROC values equal to 0.50 indicate no differences between the baseline signal and the task-related signal.

To assess longitudinal changes in calcium activity, two windows were established: a Control Window located from trial start to the sound cue indicating trial start, and a Target Window located 2s before and after PT start (overall = 4s). For each trial, all calcium peaks were measured, and only peaks ≥ 90th percentile with a minimum peak prominence of 2 were selected. Finally, the maximum values within the Control and Target Windows (if present) were captured and averaged for every session.¹⁷³

Statistical Analyses

All data are reported as mean ± SEM unless otherwise noted. Statistical analyses were performed in GraphPad Prism (USA). All data were tested using the Shapiro-Wilk normality test. The means of different data distributions were analyzed and compared using two-tailed Student’s t-test (Figures S1O, S1R, 2E FA Rate, 2E Bias, 3F, 4E Hit Rate, 4E Sensitivity, 4E Bias, 4G FA Rate, 4G Bias, 4K Learning/Expert Hit Rate, 4K Learning/Expert FA Rate, 4K Learning/Expert Sensitivity, 4K Learning/Expert Bias, 4L, S5C Hit Rate, S5C FA Rate, S5C Sensitivity, S5C Bias), Wilcoxon signed rank test (Figures S1N, S1Q, 2E Hit Rate, 2E Sensitivity, 4E FA Rate, 4G Hit Rate, 4G Sensitivity), ordinary one-way ANOVA with Tukey’s multiple comparison correction (Figures S1G, S1L), Kruksal-Wallis with Dunn’s multiple comparison correction (Figures 1E, 1F, 1I, 1L, S1H, S1I, S1J, S1K, 4I), linear regression (Figure S1E), Repeated measures one-way ANOVA with Tukey’s multiple comparison correction (Figures 3B, 3C), Repeated measures mixed-effects analysis with Tukey’s multiple comparison correction (Figures S4D Top, S4D Bottom), Repeated measures two-way ANOVA with Sidak’s multiple comparison correction (Figures 3K, 3L), Repeated measures mixed-effects analysis with Sidak’s multiple comparison correction (Figure 3Q). For all statistical tests, *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.