Table of Contents

Temporally Coordinated Activity among Motor Cortex, Thalamus, and Thalamic Reticular Nucleus Neurons during Rat Forelimb Movements

  1. Yoshikazu Isomura2,6
  1. 1Department of Psychiatry and Behavioral Sciences, Graduate School of Medical and Dental Sciences, Institute of Science Tokyo, Tokyo 113-8519, Japan
  2. 2Department of Physiology and Cell Biology, Graduate School of Medical and Dental Sciences, Institute of Science Tokyo, Tokyo 113-8519, Japan
  3. 3Department of Epileptology, National Center of Neurology and Psychiatry, Tokyo 187-8551, Japan
  4. 4Department of Molecular Genetics, Institute of Biomedical Sciences, Fukushima Medical University, Fukushima 960-1295, Japan
  5. 5Department of Biochemistry, Faculty of Medicine, University of Yamanashi, Yamanashi 409-3898, Japan
  6. 6Center for Brain Integration Research, Institute of Science, Tokyo, Tokyo 113-8519, Japan

  1. Author contributions: M.S., H.T., and Y.I. designed research; M.S. performed research; M.S., A.R., M.K., S.K., S.H., T.O., and K.K. contributed unpublished reagents/analytic tools; M.S., A.R., and M.K. analyzed data; M.S., H.T., and Y.I. wrote the paper.

 
Next Section

Abstract

Thalamic reticular nucleus (TRN) neurons receive excitatory inputs from both cortical and thalamic neurons and, in turn, inhibit thalamic relay neurons, thereby modulating thalamocortical signaling. Recent studies suggested that the TRN in the motor system contributes to precise temporal control within the cortico-thalamo-cortical loop. However, few studies have examined how TRN neurons temporally coordinate with cortical and thalamic neurons during discrete limb movements. Here, we investigated how TRN neurons collaborate with other neurons during forelimb movements depending on task demands. We trained head-fixed rats (male and female) to perform a forelimb lever-pull task at one of the two difficulty levels (low- or high-demand task group). Spike activity was recorded from TRN neurons, some identified optogenetically, and from neurons in the primary motor cortex and ventral thalamus during task performance. Among these neurons, putative excitatory cortical neurons had the highest proportion of tonic spike activity before lever pull (Hold-type) and the lowest proportion of phasic pull-related activity (Pull-type), with the highest directional specificity. In contrast, TRN neurons exhibited more frequent Pull-type activity with lower directional specificity. Furthermore, rats in the high-demand group adopted a more uniform and robust strategy compared with those in the low-demand group. Notably, TRN neurons showed the greatest adaptive-like activity differences in association with task demands among those neurons: the proportion of Pull-type neurons was higher, and their activity was greater, earlier, and more direction-specific. These results suggest that TRN neurons in the cortico-thalamo-cortical loop play a crucial role in dynamically controlling movements by adapting to circumstances.

Previous SectionNext Section

Significance Statement

The thalamic reticular nucleus (TRN), a key component of the cortico-thalamo-cortical loop, is thought to be critical for motor information processing. However, it remains unclear how this activity contributes to skilled limb movements when task demands change. We found that when the motor task was highly demanding, a population of TRN neurons showed adaptive-like activity differences. These neurons exhibited stronger, faster, and more directionally specific responses during movement execution. Our findings suggest that the TRN may act as a pivotal hub for optimizing the cortico-thalamo-cortical loop, enabling the adaptive motor control required for skilled movements.

Previous SectionNext Section

Introduction

Between the cerebral cortex and the thalamus, cortical neurons send excitatory corticothalamic (CT) projections to thalamic neurons, which then send excitatory thalamocortical (TC) projections back to the cortex, forming a reciprocal excitatory loop. The thalamic reticular nucleus (TRN) is a thin, shell-like structure that overlays the dorsolateral aspects of the thalamus and is composed of inhibitory GABAergic neurons. TRN neurons receive excitatory inputs through axon collaterals from TC and CT neurons and provide inhibitory feedback to TC neurons in a topographically organized manner, shaping the processing of sensory and motor information within the cortico-thalamo-cortical loop (Pinault, 2004; Jones, 2007; Halassa and Acsády, 2016; Crabtree, 2018).

In sensory systems, with the exception of olfaction, information is transmitted to each primary sensory area of the cerebral cortex through TC neurons. Excitatory TC neurons receive descending inputs from cortical areas, which amplify and process sensory signals. In contrast, inhibitory TRN neurons are thought to exert gating control to regulate appropriate sensory timing (Crick, 1984). This “gating mechanism” has been supported by recent optogenetic studies, showing that it plays a pivotal role not only in individual sensory modalities but also in multisensory integration (McAlonan et al., 2000, 2008; Yu et al., 2009; Wimmer et al., 2015; Clemente-Perez et al., 2017; Liu et al., 2021).

Although studies are limited, TRN neurons in the motor system may also contribute to motor control at the appropriate time within the cortico-thalamo-cortical loop (Lam and Sherman, 2015; Carroll et al., 2022; Hádinger et al., 2023). During locomotion in cats, the gating mechanism of motor-related TRN neurons depends on the body part being controlled and the phase of movement (Marlinski et al., 2012; Marlinski and Beloozerova, 2014). In visually guided reaching movements of monkeys, rostrodorsal TRN sector neurons show different activities related to sensory stimuli, motor preparation, and motor execution, which may provide predominantly spatially nonselective inhibitory input to motor thalamic nuclei (Saga et al., 2017). In humans, thalamic infarction involving the TRN is associated with movement disorders (Lehéricy et al., 2001). These findings suggest that the TRN plays a role in regulating appropriate timing of movements. If so, the timing of motor control should be determined by the temporally coordinated dynamics of neuronal activity in the motor cortico-thalamo-cortical loop, namely, excitatory and inhibitory cortical neurons, excitatory TC neurons, and inhibitory TRN neurons. However, previous studies have not recorded identified TRN (TRNi) neurons and compared their activity changes with those of cortical and thalamic neurons to elucidate their coordination in motor control.

Recent advances in the identification and manipulation of specific neuronal subtypes have revealed functional heterogeneity and causal roles of TRN neurons (Halassa et al., 2014; Dong et al., 2019; Lee et al., 2019). For example, selective suppression of synaptic inputs from TRN neurons impairs action sequence learning in mice (Ohno-Shosaku et al., 2023). Likewise, unilateral lesions of the rostral TRN disrupt action selection in rats (Weese, 2022). However, these experiments in freely moving rodents were designed to assess action selection involving whole-body movements, such as locomotion and postural adjustments. In such settings, it is difficult to observe discrete movements of specific body parts, such as the forelimb. Therefore, how the activity dynamics of TRN neurons differ from those of cortical and thalamic neurons during the execution of specific movements remains unclear.

To address this limitation, we recently established behavioral tasks in which head-fixed rats manipulated a lever with their forelimb while maintaining a stable posture, allowing us to monitor discrete forelimb movements with high spatiotemporal resolution (Isomura et al., 2009; Kimura et al., 2012). We combined these behavioral tasks with electrophysiology and optogenetics to record spike activity in specific neuronal subtypes (Soma et al., 2017, 2019; Nonomura et al., 2018; Rios et al., 2023). In this study, we aimed to elucidate how motor cortical, thalamic, and TRN neurons coordinate motor information processing as head-fixed rats control forelimb movements under different task demands.

Previous SectionNext Section

Materials and Methods

Animals and surgery

All experiments were approved by the Animal Care and Use Committee of the Institute of Science Tokyo (A2021-041, A2023-116) and were conducted in accordance with the Fundamental Guidelines for the Proper Conduct of Animal Experiments and Related Activities in Academic Research Institutions (Ministry of Education, Culture, Sports, Science, and Technology of Japan) and the Guidelines for Animal Experimentation in Neuroscience (Japan Neuroscience Society). All surgical procedures were performed under isoflurane anesthesia, and efforts were made to minimize suffering. The animal experimental procedures were described in our previous studies (Isomura et al., 2009, 2013; Kimura et al., 2012, 2017; Saiki et al., 2014; Soma et al., 2017, 2019; Rios et al., 2019, 2023; Mitani et al., 2022).

Ten adult vesicular GABA transporter (VGAT)-Cre transgenic rats (Long–Evans, 271 ± 37 g; six males and four females; Soma et al., 2019) were kept in their home cages under an inverted light schedule (lights off at 9 A.M., lights on at 9 P.M.). The rats were briefly handled by the experimenter (10 min, twice) in advance to adapt to the experimental environment.

For headplate implantation, the animals were anesthetized with isoflurane (4.5% for induction and 2.0–2.5% for maintenance; Pfizer) using an inhalation anesthesia apparatus (Univentor 400 anesthesia unit; Univentor) and placed on a stereotaxic frame (SR-10R-HT; Narishige). Lidocaine (Aspen) was administered through a surgical incision. During anesthesia, the rectal temperature was maintained at 37°C using an animal warmer (BWT-100, Bio Research Center). A headplate (CFR-2, Narishige) was attached to the skull using small stainless-steel anchor screws (M1, 2 mm in length) and dental resin cement (Super-Bond C&B, Sun Medical; Unifast II, GC Corporation). Reference and ground electrodes (Teflon-coated silver wires, A-M Systems; 125 μm diameter) were implanted on the dura mater above the cerebellum. The skull surface was covered with silicone sealant (DentSilicone-V, Shofu). Analgesics and antibiotics were applied postoperatively as required (meloxicam, 1 mg/kg s.c., Fujita Pharmaceutical; gentamicin ointment, 0.1%, Takata Pharmaceutical).

After full recovery from surgery (≥5 d later), the rats had ad libitum access to water during the weekends, but during the rest of the week, they obtained water only when they performed the task correctly. When necessary, an agar block containing water was provided to rats in their home cages to maintain them at >80% of their original body weight (Saiki et al., 2014; Soma et al., 2017, 2019, 2023; Rios et al., 2019, 2023; Sakairi et al., 2025).

Vector and tracer injections

For anterograde tracing, 0.4 μl of AAV5-CAG-tdTomato vector (titer 1.0 × 1013 vg/ml; Addgene) was injected into the right primary motor cortex (M1; AP, +1.0; ML, +2.5; DV, +1.4 mm from the bregma). For retrograde tracing, 0.5 μl of CTB-555 (cholera toxin subunit B, Alexa Fluor 555 conjugate, Thermo Fisher Scientific) was injected into the right TRN (AP, −1.8; ML, +3.0; DV, +5.8 mm from the bregma). The rats were killed at least 3 weeks after AAV injection and 7 d after CTB injection. For the optogenetic identification (opto-tagging) of recorded neurons, 0.4–0.5 μl of AAV2-Syn-Flex-rcChrimsonR-tdTomato vector (7.5–8.3 × 1012 vg/ml; Addgene) was injected into the left TRN at two sites (AP, −1.4; ML, +2.1; DV, +5.8 mm; and AP, −1.8; ML, +3.0; DV, +5.8 mm from the bregma). In the high-demand task group (see below), 0.4–0.5 μl of AAVDJ-Syn-hChR2-EYFP vector [1.8 × 1012 vg/ml; modified in Hamada et al. (2021)] was additionally injected into the left M1 (AP, +1.0; ML, +2.5; DV, +1.0 mm from the bregma). The vector solution was slowly injected (25–40 nl/min) through a glass capillary (tip diameter, 50 μm) from a 10 μl Hamilton syringe using a syringe pump (Legato 100; KD Scientific). The capillary was left in place for 10–30 min before and 10 min after injection to allow for diffusion. Optogenetic identification during recording was performed at least 3 weeks after injection.

Behavioral tasks

In the behavioral experiments, rats were head-fixed and placed inside a cylinder in a stereotaxic frame in a dim chamber for several hours per session (Fig. 1). They were categorized into two groups (low-demand and high-demand task groups) and trained to repeat the same forelimb lever-pull movement but with either low-demand (easy) or high-demand (difficult) task requirements under the head-fixed condition. For both groups, initial training sessions without external force were conducted in a common, simplified training apparatus (multirat task-training system, O’hara & Co.) prior to the recording sessions. Although this apparatus did not support neural recording or lever trajectory tracking, all other task parameters were equivalent to the recording setup (Isomura et al., 2009, 2013; Kimura et al., 2012; Saiki et al., 2014).

For the low-demand task group (N = 5 rats, three males and two females; n = 64 sessions), the rats learned to perform a basic lever-pull task, in which they had to pull a “spout lever” in response to a visual stimulus to obtain a reward (Kimura et al., 2012, 2017). For two of the five rats (one male and one female; total n = 15 sessions), opto-tagging and neural recordings were unsuccessful due to technical issues; therefore, these two rats contributed to behavioral data only, and their spike data were excluded from all neural analyses. Behavioral analyses included all rats that performed the task, whereas neural analyses included only animals/sessions with successful opto-tagging and recording. In the final form of this low-demand task, the rats started each trial by pushing the spout lever with their right forelimb in a self-paced manner following a 3 s intertrial interval (ITI). After they remained stable within the push area (0–20% of the full lever range) for 500 ms, a visual stimulus (30 lux) was presented for 1,000 ms as a Go cue by two LED lights in front of them. If they pulled the spout lever to their mouth after the onset of the visual cue and kept it stable within the pull area (80–100% of the full lever range) for another 500 ms, water droplets (5 μl × 2 shots) were given as a reward from the tip of the spout lever. In the low-demand task condition, there was no time limit for them to respond to the cue presentation, and even if they gave a wrong response, they could simply continue trying with neither an error notification nor ITI until they succeeded. Additionally, the low-demand task did not include perturbative force trials (described below). The rats completed task training for recording the experiments when they met the criterion of achieving typically >500 correct trials per daily session (2–3 weeks). For subsequent neural analysis, we selected correct trials in which the lever left the push area within 100 ms after cue onset, never turned back, even slightly, and stayed in the pull area for 500 ms.

The high-demand task group (N = 5 rats, three males and two females; n = 58 sessions) was required to perform a similar but demanding task with time limits, fluctuations, and occasional perturbations by external force. In the final form of this high-demand task, we made the task more demanding by setting the hold time before the cue to 500–1,500 ms, the push area to 0–12%, the pull area to 75–100%, and the ITI to only 500 ms. Moreover, we occasionally applied external force to the spout lever in the direction of pushing back as a sudden perturbation, while they pulled it (“force trials”). In force trials, an external force was applied by the attraction of electromagnets (MH-P20/15, Electrical Kingdom on the lever side; a16101700ux0246jp, Uxcell on the ground side, controlled by Arduino) when the lever reached the 23.5–24.3% threshold position. Subsequently, the force gradually weakened owing to the separation of these magnets when the magnet moved further toward the pull area. The static force exerted on the lever was measured using a tension gauge (EA715BD-1, 0.025–1.0 N; Ōba Instrument Works). Measurements were repeated three times at five normalized positions under both no-force trials (positions, 0, 0.316, 0.504, 0.695, 1) and force trials (positions, 0, 0.232, 0.504, 0.695, 1). Each session began with an initial block of 30 no-force trials and was repeated in two types of trial blocks (alternating and random). Each alternating block consisted of 30 trials, starting with a no-force trial, in which the no-force and force trials alternated repeatedly in the order of no-force, force, no-force, force, and so on. Each random block consisted of 50 trials of randomly ordered force (25%) and no force (75%). This pattern was sufficiently complex to prevent rats from predicting the emergence of perturbations in each trial. Specifically, the reaction time (RT), lever trajectory during the pull, and success rate in the no-force trials did not differ between the alternating and random blocks (data not shown). We categorized the three behavioral errors as false alarm, pushback miss, and bounce miss. The false alarm was an error response in which the lever left the push area before the cue onset. A pushback miss was an error in which the subject was pushed back without reaching the pull area. The bounce miss was an error that was not held for 500 ms in the pull area but exited there. The rats completed task training for the recording experiments when they met the criteria of achieving >700 correct trials per session and a success rate of >65% for two consecutive sessions. For later analysis, we excluded the error trials and used only the correct responses. The entire task control system was operated by the behavioral control system Task Forcer (O’hara & Co.; Kimura et al., 2012, 2017).

Electrophysiological recording

Once the rats completed task learning (three rats in the low-demand task group and five rats in the high-demand task group), they underwent a second surgery under isoflurane anesthesia for later electrophysiological recordings. Two holes were made in the skull and dura mater, above the left M1 and TRN. The skull surface and holes were immediately covered with antibiotic ointment and silicone sealant until the experiments were recorded.

We performed extracellular, multichannel recordings of spike activity from individual neurons and local field potentials (LFPs) in the left M1, ventral thalamus, and TRN, while the head-fixed rats performed behavioral tasks (49 and 58 sessions in the low- and high-demand task groups, respectively). Supported by agarose gel (2% agarose-HGT, Nacalai Tesque) on the brain, the caudally inclined 32-channel silicon probe (Iso_3x_tet-A32; NeuroNexus Technologies; seven tetrodes and four tandem electrodes at 250 µm interval on three shanks with 50 µm thick; 12°) was advanced +5.0 to +6.5 mm from the brain surface (AP, −2.7 to −3.4 mm; ML, +2.2 to +3.3 mm), aiming at the viral injection site in TRN and the adjacent ventral thalamus using a micromanipulator (SMM-200, Narishige) on a stereotaxic frame (SR-10R-HT, Narishige). The other rostrally inclined 32-channel silicon probe (the same type but 15 µm thick; 20°) was advanced +0.5 to +1.0 mm from the brain surface (AP, +1.2 mm; ML, +0.5 to +1.0 mm), aiming at the viral injection site in M1 in the same way. To prevent electrode drift, we waited ∼30 min after inserting the silicon probes before starting the behavioral task and electrophysiological recording.

Wide-band signals were amplified and filtered (FA64I, Multi Channel Systems; final gain, 2,000; bandpass filter, 0.5 Hz to 10 kHz) through a 32-channel head stage (MPA32I, Multi Channel Systems; gain, 10). The signals were digitized at 20 kHz and recorded using three 32-channel hard-disc recorders (LX-120, TEAC) that simultaneously digitized the lever positions tracked by the angle encoders and the behavioral events.

Spike isolation and neuron classification

The raw signal data were processed offline to isolate the spike events of individual neurons in each tetrode of the silicon probes. Spike candidates were detected and clustered using the semiautomatic spike-sorting software, EToS (Takekawa et al., 2010, 2012). Using open-source software (Klusters clustering software and NeuroScope viewing software; Hazan et al., 2006), spike clusters were manually combined, divided, discarded, or combined to refine single-neuron clusters based on two criteria: the presence of refractory periods (>2 ms) in their own autocorrelograms and the absence of refractory periods in cross-correlograms (CCGs) with other clusters.

For each neuron (spike cluster), the basal spiking properties and task-related activity in relation to the behavioral task performance were analyzed using MATLAB (MathWorks) as follows: the ongoing spike rate and width (trough-to-peak duration) for individual spike clusters were defined as described in our previous studies (Soma et al., 2023; Sakairi et al., 2025). Spike clusters recorded in M1 were classified as RS (putative excitatory) and FS (inhibitory) neurons based on the spike width (≥0.5 ms for RS neurons, <0.5 ms for FS neurons), showing a clear bimodal distribution (Fig. 3). Spike clusters in the ventral thalamus and TRN were also classified as putative TC (TCp) and TRN neurons based on the spike width (≥0.4 ms for TC neurons, <0.4 ms for TRN neurons). We established the criteria based on previous reports (Halassa et al., 2011, 2014; Lewis et al., 2015; Huo et al., 2020) and our own optogenetic identification, which showed that TRN neurons exhibited narrow spike widths (Fig. 2).

Task-related neuron analysis

We defined the key timing of the lever movements (onset of pull, end of pull, and push) in each trial based on the normalized lever trajectory velocity. Pull onset was defined as the time point at which the velocity of the smoothed and normalized lever trajectories first exceeded a threshold of 1 (/s) after the hold period in the push area. Pull end was defined as the first time point at which the lever velocity dropped below a threshold of 1 (/s) after the lever had reached the target pull area. Push end was defined as the first time point at which the lever velocity exceeded a threshold of −1 (/s) after the lever reached the push area to start the trial. If multiple push end events were detected within a single trial (∼25.1% of the trials), the event closest to the go cue was selected for analysis. The RT was defined as the interval between the onset of the go cue and the onset of the pull. Pull duration (PD) was defined as the time from pull onset to the pull end. Notably, RT and PD were not determined for the initial training sessions, as these were conducted in a simplified apparatus (as described above) where lever trajectories were not recorded. Consequently, we did not compare behavioral performance across training sessions and recording sessions.

To assess the functional spike activity, we analyzed the spike trains of individual neurons in relation to the timing of lever movements. Only neurons that fired at least 100 spikes across a minimum of 20 valid trials were included in the analysis. These neurons were then classified as “task-related” or “nontask-related” using the task relevance indexes (TRI; Soma et al., 2017; Kawabata et al., 2020). The indices compare the cumulative distribution of spike times during task performance with the theoretical uniform distribution using the Kolmogorov–Smirnov (KS) test. A neuron was considered task-related if the KS test yielded a d value ≥0.02 and a p value ≤10−⁶ (Fig. 3). Using only the p value can lead to the inclusion of neurons with small changes in activity, simply because they have a high number of spikes; therefore, we used the d value in addition to the p value. The task-related neurons were further subclassified into three groups—Hold-type, Pull-type, and Postpull-type—based on the temporal profile of their activity, following a previously described scheme (Isomura et al., 2013; Soma et al., 2017, 2019, 2023; Fig. 4). First, to isolate Hold-type neurons, we focused on a subpopulation of task-related neurons whose peak z-scored spike rate occurred before pull onset. Among these, the Hold-type neurons were defined as those that did not display a significant phasic increase around the pull onset [see Isomura et al. (2013) for details]. Then, the remaining (i.e., not Hold-type) task-related neurons were categorized as either Pull-type or Postpull-type neurons, based on the peak time of spike activity. Postpull-type neurons were defined as those that showed peak activity later than the pull end (median), whereas Pull-type neurons were defined as the remainder. In short, the Hold-type neurons exhibited a sustained increase in spiking activity prior to pull onset. Pull-type neurons have a phasic spike increase in relation to the pull onset. Postpull-type neurons exhibit slow activity, peaking after the pull end. Task-related spike activity was displayed on a perievent time histogram (PETH) by aligning the spike sequences with the task event.

Moreover, we evaluated the directional specificity of Pull-type neurons by calculating the normalized activity difference, which was the peak spike rate during push movements (Rpeak_push) divided by that during pull movements (Rpeak_pull) and subtracting one (Rpeak_push / Rpeak_pull − 1; Fig. 5). Here, the Rpeak_push value was determined in the PETH that was constructed in 10 ms bins by aligning spike sequences with the push end and smoothed using a Gaussian filter with a window size of 3 using the smoothing function in MATLAB. The peak time was identified within a window of −500 to 0 ms from the push end. The Rpeak_push was then calculated as the averaged spike rate around the peak time (±150 ms). The Rpeak_pull value was also calculated in the same manner as for the push movements.

To evaluate the synchrony of spike activity between two neurons, we analyzed the trial-by-trial covariance of their spike counts by calculating Spearman's correlation coefficient (r) during task performance (Zohary et al., 1994; Averbeck et al., 2006). This analysis was performed on spike counts during the pull movement (0 to +500 ms from pull onset) across 200 trials. For comparison in population analysis, shuffled data were prepared for each neuron pair by randomizing the original data into blocks of 10 trials each. This procedure cancels the spurious correlations that can arise from slow changes in spike activity (Saiki et al., 2014; Kimura et al., 2017).

Opto-tagging of TRN neurons

To optogenetically identify TRN neurons (opto-tagging), we inserted an optical fiber vertically above the TRN recording site (AP, −1.6 to −1.4 mm from the bregma; ML, the same as the silicon probe inserted into the TRN; DV, +0.5 to 1.0 mm above the probe) in eight VGAT-Cre rats, Cre-dependently expressing ChrimsonR in GABAergic TRN neurons after the local AAV injection. Single orange light pulses (595 nm, <5 mW at the tip, 1–2 ms duration) were delivered repeatedly every 2 s through the optical fiber (FT400EMT, 400 µm core, NA 0.39, FC; Thorlabs) from an ultra-high-power LED light source (UHP-Mic-LED-595, FC; Prizmatix) triggered by a digital stimulator (SEN-8203; Nihon Kohden). In some cases (five rats in the high-demand task group), we additionally confirmed that TRN neurons received axon collaterals from M1 neurons by cortically stimulating M1 neurons Cre-independently expressing channelrhodopsin (ChR2) with single blue light pulses (460 nm, <125 mW, 1 ms, every 2 s) on the M1 surface through another optical fiber (FT1000EMT, 1,000 µm core, NA 0.39; Thorlabs) and an LED light source (UHP-Mic-LED-460; Prizmatix). The distance between the stimulated M1 site and recorded TRN site was ∼4 mm. At the intensity of the blue light pulses used in this study, the amount of light reaching the TRN was substantially below the activation threshold for ChrimsonR (Klapoetke et al., 2014; Jiang et al., 2022). Therefore, we concluded that it was unlikely that the TRN neurons expressing ChrimsonR would be directly evoked to discharge by blue light stimulation. In both tests, spikes were considered “light-evoked” if they occurred within 20 ms after the light pulse. Furthermore, they were considered to originate from the same neuron if the spike waveform (mean, ±1 ms from the spike trough) of light-evoked spikes was highly correlated (correlation coefficient >0.9) with that of spontaneous spikes of the original neuron (Nonomura et al., 2018; Rios et al., 2023).

Sleep spindle analysis

Sleep spindles are cortical and thalamic oscillatory activity driven by the interplay between the TC and TRN neurons (Fernandez and Lüthi, 2020). Therefore, we aimed to characterize the spike properties of TC and TRN neurons in sleep spindles (Fig. 7). Sleep spindles were detected in the cortical LFPs (M1) obtained during a 15 min rest period before task performance using a threshold-based method as previously described (Gardner et al., 2013). Briefly, the multichannel LFP signals were downsampled to 1 kHz and bandpass filtered between 6 and 18 Hz to cover the spindle band from 8 to 15 Hz. The filtered signal was rectified, and its amplitude envelope was calculated using the Hilbert transform. Spindle candidates were detected if the envelope exceeded the primary threshold of 3.5 standard deviations (SD) above the mean of the entire envelope signal. The spindle event was required to remain above the secondary threshold of 2.0 SD for at least 0.4 s. The onset and offset of the spindle were defined as the time points at which the envelope crossed the 2.0 SD threshold. Spindle events separated by <0.5 s were merged into a single spindle event, and events lasting longer than 3 s were discarded as artifacts. The LFP trace that showed the strongest spindle activity during each session was used. For spindle detection, we used data from all 54 sessions in which cortical LFPs were stably recorded. To analyze the neuronal interplay in the spindles, we used 49 of the 54 sessions that included a sufficient number of TC and TRN neurons.

For pairs of simultaneously recorded neurons (Neurons X and Y, either the TC or TRN), we obtained spikes that occurred within spindle periods. These spike trains were then binned at 10 ms intervals. We performed a logistic regression analysis (using the fitglm function in MATLAB) to examine how well the spiking activity of Neuron X in a preceding time bin (coded as 1 for a spike and 0 for no spike) predicted the current spiking activity of Neuron Y, with a time lag of 50 ms between them. We further calculated the CCGs for neuron pairs that showed a significant relationship in the logistic regression analysis (p < 0.05). The CCGs were computed with a bin size of 2 ms and a maximum lag of ±1,000 ms.

To quantify the phase relationship between the spiking of the two neurons, we first applied a Fourier transform to the CCG to identify the peak frequency of the oscillation, which is typically in the range of 6–18 Hz. The CCG data were then bandpass filtered around this peak frequency, and the Hilbert transform was subsequently applied to extract the instantaneous phase. The phase difference between two neurons was defined as the phase of the Hilbert-transformed CCG at zero lag, t = 0. To assess the quality of oscillatory coupling, a sine wave was constructed using the calculated peak frequency and phase difference. We then computed the Pearson's correlation coefficient (R) between the sine wave and the filtered CCG data. Only neuron pairs with R ≥ 0.25 were included in the final population analysis of phase displacements.

Histology

After the recording experiments were completed, rats were deeply anesthetized with urethane (2–3 g/kg, i.p.) and transcardially perfused with saline, followed by 4% paraformaldehyde (PFA) or 4% formaldehyde (FA) solution in 0.1 M phosphate buffer. Brains were postfixed for 12–48 h at 4°C in 4% PFA or FA. Subsequently, 50 µm coronal sections were cut using a microslicer (VT-1000S, Leica), and the probe track locations were verified by examining the scars under a fluorescence microscope (IX83-DP80, Olympus). The tracer observations were conducted in a similar manner.

For immunohistochemistry, sections were incubated for 30–60 min in a blocking solution containing 0.3% Triton X-100 in PBS and either 5% normal goat serum (for recorded rats) or 5% normal donkey serum (for tracer-injected rats). This step was followed by an overnight incubation at 4°C with primary antibodies (see below) diluted in the blocking solution. The following day, the sections were rinsed three times in PBS for 5–10 min each and then incubated for 2 h at room temperature with secondary antibodies (see below) diluted in the blocking solution. They were rinsed three times in PBS for 5–10 min before being mounted onto slides. Finally, the sections were mounted and counterstained with DAPI-containing mounting medium (DAPI-Fluoromount-G; Southern Biotech).

The primary and secondary antibodies were used as follows: for recorded rats, the primary antibody was rabbit polyclonal anti-RFP (1:1,000; Rockland Immunochemicals, RRID: AB_2209751), and the secondary was goat anti-rabbit IgG, Alexa Fluor 555 (1:300; Abcam, RRID: AB_2722519). For tracer-injected rats, rabbit polyclonal anti-parvalbumin (1:1,000; Thermo Fisher Scientific, RRID: AB_2808011) and donkey anti-rabbit IgG, Alexa Fluor 488 (1:1,000; Thermo Fisher Scientific, RRID: AB_2535792), were used.

Statistics

All data are expressed as median values (first and third quartiles in parentheses) unless otherwise noted. Statistical analyses were performed using the MATLAB Statistics and Machine Learning Toolbox (MathWorks). The specific statistical methods included the Wilcoxon rank-sum test, Wilcoxon signed-rank test, and Kruskal–Wallis test, followed by Dunn's test with the Benjamini–Hochberg (B–H) correction for post hoc analysis. The KS test was used to assess the task relevance of spike activity. For categorical data, the χ2 test was used, with Holm's correction applied for multiple comparisons. Correction for multiple comparisons was applied to the four major subtypes (M1-RS, M1-FS, TC, and TRN). The TRNi neurons were analyzed without correction because they were considered independent hypotheses. For multiple comparisons in other tests, the Bonferroni’s correction or false discovery rate (FDR) control was applied, where appropriate. Correlations were assessed using Pearson's or Spearman's rank correlation coefficients.

Given that our behavioral dataset had a hierarchical structure (multiple sessions nested within each animal) and the spike analysis dataset similarly had a hierarchical structure (units nested within sessions and animals), we first quantified the contribution of interanimal and intersession variability to the variance of behavioral and neuronal indices. We calculated the intraclass correlation coefficients (ICCs) using linear mixed-effect models (LMMs) for continuous variables and generalized LMMs (GLMMs) for categorical variables. For categorical neuronal variables (i.e., Hold-, Pull-, and Postpull-type classifications), the responses for each category were binarized (1, target category; 0, all other categories). For each category, we fitted a logistic GLMM with random intercepts for the animals and sessions. Variance components (σ2animal, σ2session) were extracted from the covariance parameters of the fitted models (using the fitglme function in MATLAB). Based on these variance components, we derived two types of ICCs: (1) the animal ICC, representing the variance explained by individual differences, and (2) the cumulative animal + session ICC, representing the total variance explained by the hierarchical structure. These ICC calculations allowed us to quantify the magnitude of hierarchical dependencies, ensuring that we avoided the inflation of Type 1 error rates that arises from treating nested sessions or units as independent samples. Then, permutation tests were performed at the animal level for behavioral indices and TRN neuronal variables that exhibited an animal or cumulative ICC ≥ 0.4 in at least one of the task conditions (Cicchetti, 1994; Koo and Li, 2016). Specifically, we calculated the mean of session-wise variables for each animal. For comparisons between task groups (low-demand vs high-demand), we randomly permuted task labels across animals. For within-animal comparisons (high-demand NF vs F), we used a paired permutation (sign-flip) test. The B–H procedure was used to control the FDR for multiple comparisons.

All statistical tests were two-sided, and a p value of <0.05 was considered statistically significant. Blinding and randomization were not performed during the experiment.

Previous SectionNext Section

Results

Behavioral performance

To investigate the motor control mechanisms involving the cortico-thalamo-cortical loop, including TRN neurons, we trained head-fixed rats to perform a forelimb lever-pull task with either low- or high-demand requirement (Fig. 1A). In this task, rats were required to push a lever to a designated “push area” and hold it for a certain duration (hold time). When the lever was pulled into the “pull area” in response to a visual cue, drops of water were delivered as a reward. In the high-demand task, additional requirements were introduced: temporal restrictions and variations were imposed on the task procedures for each trial, and in some trials, a force load was applied against the pull direction (Fig. 1A; 0.15–0.20 N at the trigger position, 0.232). The high-demand task therefore consisted of two trial types: force (F) and no-force (NF), which were presented pseudorandomly. In contrast, the low-demand task always consisted of NF trials (0 N).

Figure 1.
View larger version:
Figure 1.

Behavioral performance of forelimb lever-pull task. A, Schematic of the cued forelimb lever-pull task. The head-fixed rats pushed the spout lever into the push area with their forelimbs and held it for the hold time. They must quickly pull it into the pull area in response to the presentation of a visual Go cue and hold it for 0.5 s to obtain water droplets from the tip of spout lever as a reward (low-demand task group, Rats #1–3, a, b). The right panel shows the perturbative load of the magnetic force during the pull movement, which was randomly applied (force trials or no-force trials) in the high-demand task group (Rats #4–8). ITI, intertrial interval; RT, reaction time; PD, pull duration. B, Training progression (number of trials in the first 30 min of each session) for each rat. Different symbols with dashed lines indicate individual rats (also in G–J), and solid lines indicate the mean of each of the two task groups. C, Lever trajectories during pull movements aligned with the Go cue (left) and pull onset (right) in a session of the low-demand task group (Rats #1–3, a, b; no-force trials). The thick line indicates the median trajectory. D, Distribution of RT (left) and PD (right) in the session shown in C. Light gray indicates RT <0 ms (predictive pulls). Arrows indicate median values. E, F, Same as C and D but in the high-demand task group (Rats #4–8). The high-demand task included force (orange) and no-force (gray) trials, which were presented either randomly or alternately. G, H, Summary of the RTs across all sessions (low-demand task group, n = 64; high-demand task group, n = 58). G, Median RT for no-force (NF) trials in the low-demand task group and for no-force and force (F) trials in the high-demand task group. H, Median absolute deviation (MedAD) of the RT. Data represented as group averages (each left) or individual rats (each right). Boxplots with gray background show behavioral data from low-demand task rats (a and b that did not undergo electrophysiological recordings). ***p < 0.001, Kruskal–Wallis test with post hoc Dunn's test. I, J, Summary of the PDs across all sessions. I, Median PD for no-force trials in the low-demand task group and for no-force and force trials in the high-demand task group. J, MedAD for the PD.

In the low-demand task group, the rats successfully learned to perform the forelimb lever-pull task in the absence of an external force within 2 weeks (Fig. 1B–D; N = 5 rats; n = 64 sessions). In the high-demand task group, the rats learned the low-demand task within 10 d and subsequently mastered the high-demand task with an external force in another 2–10 d (Fig. 1B,E,F; five rats, 58 sessions). Notably, trial counts in the high-demand group reached a plateau during the early training phase (Fig. 1B), and MedAD of RT and PD did not show an obvious trend across recording sessions. Compared with the low-demand task group, the high-demand task group showed no significant difference in RT in the session-wise analysis [Fig. 1G; RT (ms) as the median (first–third quartiles): low-demand NF 105.0 (91.5–113.5), high-demand NF 103.5 (99.0–110.0), high-demand F 103.2 (99.0–111.0); Kruskal–Wallis test, p = 0.94, χ2 = 0.12], but significantly lower variability [Fig. 1H; median absolute deviation (MedAD) of RT (millisecond), low-demand NF 38.2 (24.2–47.5), high-demand NF 10.0 (9.0–11.5), high-demand F 9.0 (8.0–11.0); p < 4.7 × 10−27; χ2 = 1.21 × 102; η2 = 0.67; post hoc Dunn's test following B–H correction, low-demand NF vs high-demand NF, p < 1.05 × 10−20; t = 9.37; r = 0.85; low-demand NF vs high-demand F, p < 1.50 × 10−20; t = 9.41; r = 0.85; high-demand NF vs high-demand F, p = 0.97; t = 0.034; r = 0.11].

The PD in the high-demand task group was significantly shorter [Fig. 1I; median PD (millisecond), low-demand NF 202.2 [165.0–231.5], high-demand NF 105.5 [98.0–122.0], high-demand F 137.2 [120.0–155.0]; p < 3.9 × 10−25; χ2 =  1.12 × 102; η2 = 0.62; low-demand NF vs high-demand NF, p < 9.80 × 10−26; t = 10.6; r = 0.81; low-demand NF vs high-demand F, p < 3.85 × 10−8; t = 5.57; r = 0.66; high-demand NF vs high-demand F, p < 9.40 × 10−7; t = 4.90; r = −0.61] and less variable [Fig. 1J; MedAD of PD (millisecond), low-demand NF 37.0 [28.0–49.0]; high-demand NF 12.5 [10.0–18.0]; high-demand F 24.0 [20.0–26.0]; p < 2.0 × 10−26; χ2 = 1.18 × 102; η2 = 0.66; low-demand NF vs high-demand NF, p < 6.32 × 10−27; t = 10.8; r = 0.82; low-demand NF vs high-demand F, p < 1.04 × 10−7; t = 5.32; r = 0.66; high-demand NF vs high-demand F, p < 1.03 × 10−7; t = 5.39; r = −0.66] than that in the low-demand task group.

Because our behavioral dataset had a hierarchical structure (multiple sessions nested within each animal), we next quantified the contribution of interanimal variability to the dispersion of behavioral indices using a mixed-effect model to compute ICCs (see Materials and Methods for details). Most of ICCs were >0.4, indicating that a substantial proportion of the variance in behavioral indices was attributable to differences between animals rather than within animals (ICCs for median RT, 0.700, 0.611, 0.611 in the order of low-demand, high-demand NF, and high-demand F; MedAD RT, 0.640, 0.611, 0.608; median PD, 0.519, 0.828, 0.689; MedAD PD, 0.454, 0.752, 0.149). These results suggest that behavioral data (ICC > 0.4) are strongly animal-dependent and should be analyzed using animals as the unit of analysis.

To address this hierarchical structure, we compared the behavioral indices between tasks based on their ICC values. Comparisons were performed at the animal level using permutation tests if the index exhibited an ICC value ≥0.4 in any of the low-demand, high-demand NF, or high-demand F task conditions. Specifically, for each animal, we averaged session-wise behavioral indices (median RT, MedAD RT, median PD, and MedAD PD) and used these animal means as independent samples in the permutation tests (n = 10,000 iterations). For between-animal comparisons (low-demand vs high-demand NF; low-demand vs high-demand F), task labels were randomly permuted across animals while keeping animal identity fixed, and p values were computed from the resulting null distribution of mean differences. For within-animal comparisons (high-demand NF vs high-demand F), we used the mean within-animal difference (NF − F) as the test statistic and performed a paired permutation (sign-flip) test. Using hierarchical permutation tests at the animal level, we confirmed significant differences in variability metrics (MedAD of RT and PD) between the low-demand and high-demand tasks, whereas effects on median RT/PD were weaker and did not consistently reach significance across comparisons (median RT, low-demand vs high-demand NF, p = 0.600; T = −7.12; low-demand vs high-demand F, p = 0.645; T = −6.18; high-demand NF vs high-demand F, p = 0.503; T = 0.944; MedAD RT, low-demand vs high-demand NF, p = 0.0123; T = 23.8; low-demand vs high-demand F, p = 0.0123; T = 24.3; high-demand NF vs high-demand F, p = 0.684; T = 0.451; median PD, low-demand vs high-demand NF, p = 0.0456; T = 81.4; low-demand vs high-demand F, p = 0.0659; T = 51.0; high-demand NF vs high-demand F, p = 0.0659; T = −30.4; MedAD PD, low-demand vs high-demand NF, p = 0.0246; T = 24.4; low-demand vs high-demand F, p = 0.0338; T = 14.2; high-demand NF vs high-demand F, p = 0.0647; T = −10.2). Thus, even when accounting for the nested structure of sessions within animals, the rats performing the high-demand task exhibited more consistent movement initiation and execution than those performing the low-demand task.

Overall, these observations suggest that rats performing the high-demand task adopted a more uniform and solid motor strategy, exhibiting lower variability in both movement initiation (RT) and execution (PD) than those performing the low-demand task. Although we acknowledge that the maturity of skill learning may partially contribute to behavioral stability, the robust differences between groups are unlikely to be explained solely by learning progression during the recording phase.

Identification of motor TRN neurons

To examine spike activity in motor TRN neurons, we first identified the motor-related sector of the TRN anatomically. Neurons in the motor TRN sector receive inputs from both the motor cortex (e.g., M1) and motor thalamic nuclei (e.g., the ventral lateral nucleus, VL) and send inhibitory outputs back to the thalamus (Pinault and Deschênes, 1998; Jones, 2007). As most TRN neurons express parvalbumin (PV; Hou et al., 2016; Li et al., 2020), we were able to depict the boundaries of the TRN and others using immunostaining for PV. Using an anterograde tracer injected into M1, we identified the motor TRN region receiving its axonal projections, which overlapped with the PV-positive area (Fig. 2A). Conversely, retrograde tracing from the same TRN region revealed labeled projection neurons, primarily in the deep layers of M1 and M2 (Fig. 2A). This region is located in the rostral part of the TRN and projects mainly to the motor VL nucleus. Thus, we successfully confirmed the motor sector of the TRN (Pinault and Deschênes, 1998).

Figure 2.
View larger version:
Figure 2.

Identification of motor TRN neurons. A, Identification of the TRN region connected to the primary motor cortex (M1). Left, Anterograde tracing from the M1 to TRN by injecting the AAV-CAG-tdTomato vector into M1. PV is in green; tdTomato is in red. Arrowheads indicate M1 axonal collaterals projecting to the TRN. Right, Retrograde tracing from the TRN to M1 by injecting CTB-555 (red) into the motor TRN region. Arrowheads indicate M1 neuronal cell bodies projecting to the motor TRN. VL, ventral lateral nucleus of the thalamus; IC, internal capsule. Scale bars, 500 µm. B Left, Optogenetic identification of TRN neurons (opto-tagging). In the low-demand task rats (left), AAV2-Syn-Flex-rcChrimsonR-tdTomato was injected into the left motor TRN region of the vesicular GABA transporter (VGAT)-Cre transgenic rats. The TRN was stimulated by orange light via an optical fiber, and the spike activity and LFPs were recorded in the left TRN and M1. In the high-demand task rats (second from the left), ChrimsonR was expressed in TRN neurons in a Cre-dependent manner, and ChR2 was expressed in ipsilateral M1 neurons in a Cre-independent manner using AAVDJ-Syn-hChR2-EYFP. M1 was illuminated by blue light from the brain surface. Middle, Electrode alignment of three-shank and 32-channel silicon probe. Right, Probe tip tracks (yellow dots) and expression of ChrimsonR-tdTomato in motor TRN neurons (arrowheads). Scale bars, 500 and 100 µm (magnified view). The magnified area is indicated by a yellow square. C, Spike raster plot and PETH of an TRNi neuron. Optical stimulation (orange) evoked spiking in the TRN neurons. D, Correlation coefficients of spike waveforms (from each 4-channel tetrode) between spontaneous (black) and light-evoked (orange) spikes plotted against the energy of light-evoked spikes for all light-excited neurons. The filled symbols indicate the opto-tagged neurons that met the criterion (correlation coefficient > 0.9). The red symbol (with an arrow) corresponds to the same neuron as shown in C. E, Distribution of the peak latencies of light-evoked spikes from the TRNi neurons in the low-demand task group. F, Classification of all isolated TRN and thalamic neurons into TRNi (filled), TRNp (gray), and TCp (empty) neurons based on spike width (trough-to-peak duration). The lower vertical axis (lower) is on the logarithmic scale of the ongoing spike rate. G, Spatial CCG showing the medial–lateral position of TRNi/TRNp or TCp neurons relative to the reference TRN neurons (0 mm) simultaneously recorded (top, using only TRNi and TC neurons; bottom, using all TRN and TC neurons). ***p < 0.001, Wilcoxon rank-sum test. H, Top, Distribution of peak latencies of direct light-evoked spikes from the TRNi neurons in the high-demand task group. Stimulation from TRNi (black) units before and after the low-demand task. Bottom, Distribution of peak latencies of spikes evoked synaptically by M1 stimulation in identified (black) and putative (gray) TRN neurons. ***p < 0.001, Wilcoxon rank-sum test. I, Representative examples of recording position (silicon probe tips) across five sessions in a low-demand task rat and a high-demand task rat.

We then combined electrophysiology and optogenetics to identify individual TRN neurons. We injected an AAV vector Cre-dependently expressing the red-shifted channel rhodopsin ChrimsonR into the previously identified motor TRN in VGAT-Cre transgenic rats (Fig. 2B). This strategy ensured that the Cre-dependent expression of ChrimsonR was largely restricted to GABAergic TRN neurons because the adjacent rodent motor thalamus contains few inhibitory interneurons (Arcelli et al., 1997; Jager et al., 2021). We confirmed brief (1–2 ms) orange light pulses reliably evoked spikes in multichannel recordings from the TRN region (Fig. 2C). In the low-demand task group, we obtained 161 TRNi neurons out of 1,142 recorded neurons using an opto-tagging technique. Each TRNi neuron showed almost identical spike waveforms between the evoked and spontaneous spikes (Fig. 2D). The median latency of light-evoked spike responses in these TRNi neurons was 5.00 ms (Fig. 2E), consistent with previous reports (Halassa et al., 2011, 2014; Lewis et al., 2015). It is possible that thalamic neurons were included in our multichannel recordings using the three-shank silicon probe (Fig. 2B). In fact, the analysis of spike properties revealed a bimodal distribution in spike width (trough-to-peak duration) among all 1,142 neurons recorded in the TRN region and adjacent thalamus (Fig. 2F). The 161 TRNi neurons consistently exhibited narrow spike widths [0.22 (0.20–0.26) ms], which aligns with previous findings (Halassa et al., 2011, 2014; Lewis et al., 2015; Huo et al., 2020). Based on this bimodal distribution, we classified all unidentified neurons as either putative TRN (TRNp) or TCp neurons, using a spike width of 0.4 ms (TRNp, n = 769, 0.22 [0.18–0.26] ms; TCp, n = 212, 0.55 [0.50–0.60] ms). The TRNp neurons had significantly higher spike rates than the TCp neurons (Fig. 2F; TRNp 4.83 [2.46–9.26] Hz; TCp 2.91 [1.65–5.40] Hz; Wilcoxon rank-sum test, p < 1.32 × 10−9; z = 6.07; r = 0.19).

To validate our classification, we examined the relative anatomical positions of neuronal pairs recorded simultaneously using a three-shank silicon probe. As expected, TCp neurons were located significantly more medially to both identified and TRNp neurons (Fig. 2G). The relative distance of TRNi neurons was 0.25 mm lateral (mode) to TCp neurons, whereas it was 0.00 mm to other TRNi neurons (Wilcoxon rank-sum test, p < 1.92 × 10−63; z = 16.8; r = 0.43). Similar results were obtained among TRNp and TC neurons (p < 6.61 × 10−234; z = 32.7; r = 0.21).

In the high-demand task group, we injected two different AAV vectors to express ChR2 in M1 neurons and ChrimsonR in GABAergic TRN neurons. We identified 303 TRN neurons by opto-tagging for ChrimsonR under orange light (Fig. 2H). We further tested whether the TRNi neurons exhibited synaptically evoked spike responses to optogenetic M1 stimulation by blue light via ChR2 (see Materials and Methods for details). We found that optogenetic M1 stimulation evoked spike responses in 77 of 303 TRNi neurons and 147 TRNp neurons; that is, 224 TRN neurons responded to M1 stimulation (Fig. 2H). As expected, the latency of the spike response to M1 stimulation was significantly longer than to direct TRN stimulation for opto-tagging (Fig. 2H; TRN stim. n = 303, 6.0 [4.0–9.0] ms; M1 stim. n = 224, 11.0 [8.0–13.0]; Wilcoxon rank-sum test, p < 2.04 × 10−25; z = −10.4; r = 0.45). The anatomical locations of the silicon probe tips were confirmed using an MRI atlas of the P80 rat brain (Calabrese et al., 2013), which was accessed via the Scalable Brain Atlas (Bakker et al., 2015). We confirmed that the locations of sampled neurons were comparable between the two task groups (Fig. 2I). Taken together, our anatomical and electrophysiological analyses provide converging evidence for the reliable identification and classification of motor TRN and TC neurons.

Here, the vast majority of TRNi neurons satisfied the criterion of narrow spike width of <0.4 ms (Fig. 2F). However, a small fraction of the TRNi neurons (9 of 161, 5.6%) failed to meet the criterion of a narrow spike width. Although this criterion would cause a slight misclassification of true TRN neurons, we considered that it was negligible overall and that the criterion was robust enough to draw conclusions. Therefore, we merged the identified and TRNp neurons and subjected all the TRN neurons to subsequent neuronal activity analyses.

Classification of functional neuron activity

Spike activity was recorded in the TRN, TC, and M1 neurons using two silicon probes simultaneously for each session in eight rats performing a forelimb lever-pull task. We isolated 2,607 M1 neurons from the low-demand task group (Fig. 3). These M1 neurons were classified into excitatory regular-spiking (RS) and inhibitory fast-spiking (FS) neurons based on the bimodal distribution of spike width (Fig. 3A; RS, n = 1,625, 0.80 [0.74–0.94] ms; FS, n = 982, 0.28 [0.24–0.36] ms). M1-FS neurons had significantly higher spike rates than M1-RS (RS, 2.05 [1.00–4.27] Hz; FS, 3.75 [1.55–8.25] Hz; Wilcoxon rank-sum test, p < 4.08 × 10−32; z = −11.79; r = 0.23), consistent with previous studies (Saiki et al., 2014; Soma et al., 2017, 2019).

Figure 3.
View larger version:
Figure 3.

Classification of neuronal activity in relation to forelimb lever-pull task. A, Classification of isolated M1 neurons into RS (empty) and FS (gray) based on spike width (trough-to-peak duration). B, Definition of task-related neurons using TRI (d and p). For each neuron, the p value of the KS test for spike sequences aligned with the onset of the pull was plotted against its d value (the criteria in the box; see Materials and Methods for details). Black dots indicate task-related neurons (p < 10−6, blue solid line; d ≥ 0.02, blue dashed line), and gray dots indicate nontask-related (discarded) neurons. Red letters indicate example neurons shown in C. C, Lever trajectories, spike raster plots, and PETHs (10 ms bins) of three representative task-related neurons (Hold-type, Pull-type, and Postpull-type) aligned to the push end (orange), Go cue (green), and pull onset (blue) at 0 s.

For the M1-RS, M1-FS, TC, and TRN neurons, we selected functional (task-related) neurons using the TRI calculated from the spike sequences aligned with the pull onset. Figure 3B shows the distribution of TRI (d and p values) for all recorded M1 neurons. We further classified the task-related neurons into three activity patterns, Hold-, Pull-, and Postpull-types (Fig. 3C; see Materials and Methods for details), according to our previous studies (Isomura et al., 2009; Soma et al., 2017, 2019). Hold-type neurons showed a sustained increase in spike activity during the hold period, which decreased during the pull movement (Fig. 3C, Neuron a). Pull-type neurons showed an increase in phasic activity just before or during the pull movement (Fig. 3C, Neuron b). Postpull-type neurons showed increased activity, with a peak after the end of the pull movement (Fig. 3C, Neuron c). We applied the same criteria of activity-type classification to all M1-RS, M1-FS, TC, and TRN neurons in both the low- and high-demand task groups.

Different repertoires of functional activities across neuron subtypes

Figure 4A shows functional (task-related) spike activities in relation to pull onset (i.e., Hold-, Pull-, and Postpull-types) among cortical, thalamic, and TRN neurons in the low-demand (M1-RS, n = 648; M1-FS, n = 381; TC, n = 86; TRN, n = 379 including 87 TRNi) and high-demand task groups (M1-RS, n = 1,458; M1-FS, n = 385; TC, n = 132; TRN, n = 1,222 including 223 TRNi). These neuron subtypes showed different repertoires of task-related activity in the low-demand task group (Fig. 4B, top; Table 1; 3 × 4 χ2 test, p < 2.0 × 10−10; χ2 = 56.9; Cramer's V = 0.14). The proportion of Hold-type in M1-RS neurons was significantly larger than that of M1-FS neurons, and the proportion of Pull-type was smaller and vice versa (post hoc residual analysis, M1-RS vs M1-FS, Hold-type p < 0.001; Pull-type p < 0.001; Postpull-type p < 0.01), as previously reported (Isomura et al., 2009, 2013; Saiki et al., 2014; Kimura et al., 2017; Soma et al., 2017, 2019). The proportion of Pull-type neurons in the TRN was significantly smaller than that in the M1-FS and larger than that in the M1-RS (M1-RS vs TRN, Hold-type p = 0.05; Pull-type p < 0.001; Postpull-type p = 0.10; M1-FS vs TRN, Hold-type p < 0.01; Pull-type p < 0.0001; Postpull-type p = 0.17).

Figure 4.
View larger version:
Figure 4.

Repertoires of task-related activities across neuron subtypes. A, Three types of task-related spike activity in the cortical (M1-RS, M1-FS), thalamic (TC), and TRN neuron subtypes aligned with the pull onset (0 s). Each top panel shows z-scored PETHs for individual task-related neurons, grouped into Hold- (blue bar on the right), Pull- (red), and Postpull-type (green). Each row represents a single neuron sorted by the peak time within the activity type. Each lower panel shows the average PETHs of Hold- (blue), Pull- (red), and Postpull-type (green) neurons (shaded areas, the standard error of the mean). Top and bottom panels show data from the low- and high-demand task groups, respectively. Insets (gray) in A and B show the data from the TRNi subset. B, Proportions of task-related activity types in the M1-RS, M1-FS, TC, and TRN neurons in the low-demand (top) and high-demand (bottom) task groups. Color codes and statistical symbols are the same as those in A. ***p < 0.001; **p < 0.01; see Results for details regarding statistical comparisons among neuron subtypes and between task groups. C, Peak amplitude (z-score) of Pull-type activity around the pull onset among the four neuronal subtypes (rightmost, TRNi subset) in both groups. Box plots show the distribution of individual neurons. Each symbol represents the median value of a single session, with shapes corresponding to individual rats (as in Fig. 1B,G–J). Arrows indicate the median values for individual rats across all sessions. ***p < 0.001; **p < 0.01; *p < 0.05; see Results for details.

View this table:
Table 1.

Proportion of activity types for each cell type

The high-demand task group displayed similar trends in the task-related activity repertoire as the low-demand task group, except for a few differences (Fig. 4B, bottom; Table 1). The proportion of Hold-type decreased in all four neuron subtypes of the high-demand task group (M1-RS, 3 × 2 χ2 test; p < 2.1 × 10−4; χ2 = 16.9; Cramer's V = 0.090; M1-FS, p < 1.1 × 10−5; χ2 = 23.0; Cramer's V = 0.17; TC, p < 3.2 × 10−3; χ2 = 11.5; Cramer's V = 0.23; TRN, p < 1.3 × 10−13; χ2 = 59.4; Cramer’s V = 0.19; post hoc residual analysis, Hold-type, M1-RS p < 0.01; M1-FS p < 0.001; TC p < 0.01; TRN p < 0.001). The proportion of Postpull-type was higher in the M1-RS and M1-FS neurons but lower in the TRN (Postpull-type, M1-RS, p < 0.001; M1-FS p < 0.001; TC, p = 0.76; TRN, p < 0.01). Notably, the proportion of Pull-type neurons differed predominantly in the TRN neurons (Pull-type, M1-RS, p = 0.16; M1-FS, p = 0.14; TC, p = 0.51; TRN, p < 0.001). The Pull-type population of TRNi neurons showed a similar proportion to the total TRN neurons in low-demand task rats and even larger in the high-demand task rats (Fig. 4B, insets; low-demand, 3 × 2 χ2 test; p = 0.862; χ2 = 0.296; Cramer's V = 0.025; high-demand, p < 9.8 × 10−3; χ2 = 9.25; Cramer's V = 0.080; post hoc residual analysis, Hold-type p < 2.4 × 10−3; Pull-type p = 0.12; Postpull-type p = 0.28).

For unit-level analyses, we calculated the ICCs using GLMMs to quantify the variance structure, specifically the animal-level ICC (animal ICC) and animal + session-level ICC (animal + session ICC; see Materials and Methods for details). All the ICCs in TRN neurons were <0.1, suggesting that only a little proportion of the variance was attributable to differences among animals or sessions (low-demand task rats, animal ICC, Hold-type, 0.0228; Pull-type, 0.0136; Postpull-type, 1.39 × 10−10; animal + session ICC, Hold-type, 0.0658; Pull-type, 0.0616; Postpull type, 0.0504; high-demand task rats: animal ICC, Hold-type, 1.47 × 10−10; Pull-type, 0.0190; Postpull-type, 0.0261; animal + session ICC, Hold-type, 2.71 × 10−10; Pull-type, 0.0874; Postpull-type, 0.0926). Therefore, permutation tests were not performed in this section.

Focusing on Pull-type activity, we found significant differences in peak amplitudes among neuron subtypes and task groups. The Pull-type activity of TRN neurons was lower than that of M1-RS and TC neurons in the low-demand task group (Fig. 4C; M1-RS, 3.1 [2.0–5.0] in z-score; M1-FS, 3.0 [1.9–5.0]; TC, 4.1 [2.5–5.3]; TRN, 2.7 [1.6–4.5]; Kruskal–Wallis test, p < 1.5 × 10−2; χ2 = 10.5; η2 = 0.011; post hoc Dunn's test following B–H correction, M1-RS vs TRN, p < 2.95 × 10−2; t = 2.58; r = 0.13; TC vs TRN, p < 4.38 × 10−2; t = 2.68; r = 0.18). Notably, the Pull-type activity was higher in the two inhibitory subtypes, M1-FS and TRN neurons, in the high-demand task group (Wilcoxon rank-sum test with Bonferroni’s correction, M1-RS, p = 1; z = −0.30; r = 0.011; M1-FS, p < 4.59 × 10−7; z = −5.30; r = 0.26; TC, p = 1; z = −0.51; r = 0.048; TRN, p < 2.33 × 10−20; z = −9.40; r = 0.30). In addition, the TRNi neurons also showed a significantly higher activity difference in Pull-type activity between the low- and high-demand task rats (Fig. 4C, inset; low-demand, 3.3 [1.9, 5.0] in z-score; high-demand, 5.8 [3.5, 9.4]; Wilcoxon rank-sum test, p < 6.55 × 10−6; z = −4.51; r = 0.316). The Pull-type activity of TRN and TRNi neurons did not differ significantly in the low-demand task rats (Wilcoxon rank-sum test, p = 0.14; z = −1.47; r = −0.100) but differed significantly in the high-demand ones (p < 7.0 × 10−3; z = −2.70; r = −0.086).

We calculated the ICCs of the peak amplitudes to assess animal and session identities. In the low-demand task rats, the ICC values were generally low (<0.4) across neuron subtypes (e.g., TRN: animal ICC = 0.0333, animal + session ICC ≈ 0), supporting the independence of the neuronal samples. In the high-demand task rats, the ICC values for TRN neurons were elevated (animal ICC = 0.302, animal + session ICC = 0.316), suggesting that spike activity became more clustered by animal and session identities under high cognitive demands; however, the explained variance remained moderate at ∼30%.

Characterization of functional activities across neuron subtypes

Figure 5A compares the temporal profiles (peak latency from pull onset) of Pull-related activity (combining Pull- and Postpull-types) among the four neuron subtypes. In the low-demand task group, the M1-FS neurons responded fastest among them relative to the pull onset, whereas the M1-RS responded most slowly; thus, the TRN neurons were third (Fig. 5A, left; M1-RS, 0.18 [0.04–0.35]; M1-FS, 0.08 [0.03–0.218]; TC, 0.07 [0.03–0.26]; TRN, 0.11 [0.04–0.29] in seconds; Kruskal–Wallis test, p < 5.2 × 10−6; χ2 = 27.3; η2 = 0.022; post hoc Dunn's test following B–H correction, M1-RS vs M1-FS, p < 1.53 × 10−5; t = 4.70; r = 0.18; M1-FS vs TRN, p < 3.0 × 10−2; t = 2.58; r = −0.092). In contrast, pull-related TRN neurons were visibly the fastest among them in the high-demand task group (Fig. 5A, right; M1-RS, 0.13 [0.05–0.30]; M1-FS, 0.08 [0.02–0.20]; TC, 0.06 [0.02–0.19]; TRN, 0.04 [0.01–0.11 s]; p < 1.15 × 10−62; χ2 = 2.90 × 102; η2 = 0.11; M1-RS vs M1-FS, p < 6.07 × 10−10; t = 6.36; r = 0.17; M1-RS vs TC, p < 1.03 × 10−5; t = 4.56; r = 0.14; M1-RS vs TRN, p < 3.18 × 10−53; t = 15.5; r = 0.37; M1-FS vs TC, p = 0.67; t = 0.42; r = 0.020; M1-FS vs TRN, p < 1.03 × 10−5; t = 4.50; r = 0.18; TC vs TRN, p < 1.8 × 10−2; t = 2.44; r = 0.083). Importantly, only the TRN neurons exhibited a significant difference in peak latency between the low- and high-demand task groups (Fig. 5B; Wilcoxon rank-sum test with Bonferroni’s correction, M1-RS, p = 0.28; z = 1.82; r = 0.047; M1-FS, p = 0.35; z = 1.71; r = 0.067; TC, p = 0.49; z = 1.55; r = 0.11; TRN, p < 4.14 × 10−18; z = 8.83; r = 0.24). The peak latency of Pull-type activity was also quantified for the TRNi neurons (Fig. 5B, inset; low-demand, 0.11 [0.05–0.33] s; high-demand, 0.04 [0.02–0.10] s), and there was no significant difference between TRN and TRNi neurons. Again, we found the latency to be significantly shorter in the high-demand task group than the low-demand group (Wilcoxon rank-sum test, p < 3.51 × 10−⁶; z = 4.64; r = 0.280).

Figure 5.
View larger version:
Figure 5.

Characterization of pull-related activities across neuron subtypes. A, Cumulative distributions of the peak time from pull onset for all pull-related activities (Pull- and Postpull-types) in the M1-RS, M1-FS, TC, and TRN neurons. Left and right panels show the low- and high-demand task groups, respectively. ***p < 0.001; **p < 0.01; *p < 0.05; Kruskal–Wallis test with post hoc Dunn's test. B, Comparison of the peak time distributions of pull-related activities in the four neuron subtypes (rightmost, TRNi subset) in the two task groups. Box plots, symbols, and arrows in B and E are shown as in Figure 4C. ***p < 0.001, Wilcoxon rank-sum test with Bonferroni’s correction applied among the four neuron subtypes. C, Spike activities for push and pull lever movements (aligned to the push and pull ends, respectively) in the Pull-type neurons (M1-RS, M1-FS, TC, and TRN). The push activity was normalized to the peak of the counterpart pull activity (set to 1). The top and bottom represent the low- and high-demand task groups, respectively. D, Plots of the peak activity for push and pull lever movements in individual Pull-type neurons. Symbols represent single neurons from low-demand (cyan, Rats #1–3) and high-demand (magenta, #4–8) task groups. The inset shows data from the TRNi subset. E, Comparison of the normalized activity differences of Pull-type neurons among the four neuron subtypes (rightmost, TRNi subset) in the two task groups. The normalized activity difference (i.e., directional specificity) goes toward −1 when the activity for push is smaller than the activity for pull, and it is 0 when both are equal. ***p < 0.001; **p < 0.01; *p < 0.05; see Results for details.

We calculated the ICCs of the peak latencies for animal and session identities. In the high-demand task rats, the animal-level ICCs were relatively higher but remained below commonly used criteria (<0.4), suggesting that the overall impact of identity was limited (low-demand task, animal ICC = 0.00427; animal + session ICC = 0.0564; high-demand task, animal ICC = 0.0125; animal + session ICC = 0.0763).

Furthermore, we examined the directional specificity of the Pull-type activity in the four neuron subtypes by comparing movement-related activities between the push and pull directions (Fig. 5C; push activity normalized by pull activity), although the analysis was simple (see Materials and Methods for details). In the low-demand task group, excitatory M1-RS and TC neurons showed larger directional specificity (toward −1) than inhibitory M1-FS and TRN neurons, and the M1-FS neurons also showed larger specificity than the TRN neurons (Fig. 5D,E). Thus, the TRN neurons had the smallest directional specificity in Pull-type activity among the four neuron subtypes (Fig. 5E; normalized activity difference, M1-RS, −0.3 [−0.5 to 0.1]; M1-FS, −0.2 [−0.4 to 0.0]; TC, −0.3 [−0.5 to 0.1]; TRN, −0.1 [−0.2 to 0.1]; Kruskal–Wallis test, χ2 = 50.8; p < 5.5 × 10−11; η2 = 0.072; post hoc Dunn's test following B–H correction, M1-RS vs M1-FS, p < 2.35 × 10−4; t = 3.95; r = −0.20; M1-RS vs TRN, p < 2.97 × 10−11; t = 6.91; r = −0.34; M1-FS vs TRN, p < 2.04 × 10−3; t = 3.20; r = −0.16; TC vs TRN, p < 6.88 × 10−4; t = 3.58; r = −0.26).

In the high-demand task group, the directional specificity became significantly smaller (toward zero) in the M1-RS neurons and larger (toward −1) in the TRN neurons compared with the low-demand task groups (Fig. 5E; Wilcoxon rank-sum test with Bonferroni’s correction, M1-RS, p < 1.22 × 10−6; z = −5.12; r = 0.19; M1-FS, p = 0.24; z = −1.88; r = 0.091; TC, p = 0.35; z = −1.71; r = 0.16; TRN, p < 6.28 × 10−6; z = 4.8; r = 0.15). The directional specificity of TRNi neurons was −0.1 [−0.4, 0.1] in the low-demand task group and −0.2 [−0.4, −0.1] in the high-demand task group (Fig. 5E, inset). The directional specificity was not significant between TRN and TRNi neurons, whereas it was larger in the high-demand task group (Wilcoxon rank-sum test, p < 8.3 × 10−3; z = 2.64; r = 0.185).

The ICCs for directional specificity were generally low to moderate (<0.4) in the TRN neurons (low-demand task, animal ICC = 0.0125; animal + session ICC = 0.262; high-demand task, animal-level ICC = 0.177; session ICC = 0.166).

Finally, we examined whether these Pull-type neurons participated in the pull movement in coordination with nearby Pull-type neurons in a trial-by-trial manner. For example, a pair of Pull-type neurons showed a positive correlation in trial-by-trial variability of spike activity during the pull period (0.5 s from the pull onset), which completely disappeared in the shuffled data (Fig. 6A). We found significant but only slight increases in such synchrony of Pull-type activity in the M1-FS and TRN–TRN pair populations in the high-demand task group (Fig. 6B; Wilcoxon signed-rank test with Bonferroni’s correction, M1-RS pairs, n = 1,410; p = 1; z = 1.35; r = 0.036; M1-FS pairs, n = 299; p < 1.4 × 10−7; z = 5.51; r = 0.32; TC pairs, n = 57; p = 1; z = 3.20; r = 0.069; TRN pairs, n = 5,341; p < 1.5 × 10−33; z = 12.2; r = 0.17). Similarly, TRNi–TRNi pairs showed a significant increase in spike synchrony (Wilcoxon signed-rank test, n = 2,064; p < 9.0 × 10−9; z = 5.75; r = 0.127). The ICC values were low at the animal and session levels (TRN–TRN pairs, animal ICC = 0.0183; animal + session ICC = 0.0225). Therefore, inhibitory M1-FS neurons and TRN neurons displayed synchronous activity fluctuations in a trial-by-trial manner; however, the synchrony was probably not strong enough to drive network synchronization during movement.

Figure 6.
View larger version:
Figure 6.

Trial-by-trial variability of spikes in neuron pairs during pull movements. A, An example pair of Pull-type TRN neurons (#1, 2) showing a positive correlation in trial-by-trial variability of spike counts during the pull movement. They were simultaneously recorded from different tetrodes. Top, original data; bottom, shuffled data. The correlation coefficient is shown as the r value. B, Cumulative distribution of r values for all pairs of neurons of the same subtype with Pull-type activity (high-demand task group). Black, original; gray, shuffled. P values by Wilcoxon signed-rank test with Bonferroni’s correction.

Neuronal interplay during sleep spindles

Sleep spindles are characteristic oscillatory activities generated by interactions between the thalamus and the TRN during slow wave sleep. Therefore, we verified the validity of the classification of TRN and TC neurons by comparing the spiking characteristics during sleep spindles between TRNi, TRNp, and TCp neurons. We obtained sleep spindles in the cortical LFP data from M1, along with spike activity of the TC and TRN neurons, during a waiting period before performance of the low-demand task (Fig. 7A; 3 rats, 54 sessions, 59.7 ± 13.2 spindles per session; see Materials and Methods for details). In an example of spike-triggered average (STA) analysis of sleep spindles (Fig. 7B), a trigger TC neuron discharged preferentially at each trough of spindle oscillations in the LFP signal and at the peak of spiking in another TC neuron. In contrast, a TRN neuron always discharged after the spikes of these TC neurons in each sleep spindle cycle. For quantification, we employed a pairwise logistic regression analysis, followed by a cross-correlation analysis, to identify neuron pairs with significantly correlated spiking during sleep spindles (see Materials and Methods). This procedure yielded a substantial number of significant pairs (logistic regression, 743 TRN–TC pairs, 2,398 TRN–TRN pairs) and correlated pairs (cross-correlation, Pearson's r ≥ 0.25; 267 TRN–TC pairs, 964 TRN–TRN pairs). At the population level, TRNp neurons discharged significantly later compared with TCp neurons, and TRNi neurons showed a similar tendency [Fig. 7C; TRNi–TCp pairs, n = 73, 0.2 (−0.1 to 0.6) in radian; TRNi–TRNi pairs, n = 57, 0.0 (−0.3 to 0.5); TRNp–TCp pairs, n = 194, 0.5 (0.0–0.9); TRNp–TRNp pairs, n = 907, 0.0 (−0.4 to 0.5); Wilcoxon rank-sum test, TRNi–TCp pairs vs TRNi–TRNi pairs, p < 0.088; z = 1.71; r = 0.15; TRNp–TCp pairs vs TRNp–TRNp pairs, p < 1.87 × 10−11; z = 6.72; r = 0.20]. No significant difference in phase lag was observed between TRNp–TRNp pairs and TRNi–TRNi pairs (Wilcoxon rank-sum test, p = 0.087; z = 1.71; r = 0.15), but TRNp–TCp pairs exhibited significantly larger phase lags than TRNi–TCp pairs (p < 2.35 × 10−11; z = 6.68; r = 0.19). The ICCs were generally low, suggesting acceptable variability due to individual differences (animal ICCs, TRNp–TCp pairs, 0.0820; TRNp–TRNp pairs, 0.0161; TRNi–TCp pairs, 0.104; TRNi–TRNi pairs, 0; animal + session ICCs: TRNp–TCp pairs, 0.225; TRNp–TRNp pairs, 0.0606; TRNi–TCp pairs, 0.179; TRNi–TRNi pairs, 0.118). This observation of delayed spindle spikes in TRN neurons is consistent with the established theory of TC–TRN neuron interplay (Steriade et al., 1993; von Krosigk et al., 1993; Bonjean et al., 2011), strongly supporting the validity of our classification of TRNp and TCp neurons (Fig. 2). It is also worth noting that the order of spike synchronization during sleep spindles (Fig. 7C; TC to TRN) differed from that of Pull-type activity in the high-demand task (Fig. 5A; TRN to TC).

Figure 7.
View larger version:
Figure 7.

Functional interactions between TC and TRN neurons during sleep spindles. A, An example of a sleep spindle event in the cortical (M1) LFP and neuronal activity during a nap. Power spectrograms of cortical LFP, raw LFP, and filtered LFP traces (6–18 Hz bandpass filter). Blue shading indicates the detected spindles (see Materials and Methods). Below, Spike raster plots from two TC and one TRN were recorded simultaneously. B, STA and CCG analyses of the three neurons. Top, STA of the M1 LFP triggered by TC1 spikes. Middle CCG between TC2 and TC1 spikes. The two TRN neurons were discharged coincidentally at the spindle oscillation trough. Bottom, CCG between TRN and TC1 spikes, indicating that TRN spikes lagged behind TC1 spikes. C, Spike phases of TC and TRN neurons relative to the spikes in other TRN neurons. In TCp–TRNp pairs, TRN neurons tended to be discharged with a slight phase delay relative to TC neurons (top). In contrast, putative TRN–TRN (TRNp–TRNp) pairs showed almost no phase shift in spike timing (bottom). Box plots show the distribution of individual neurons. Each symbol represents the median value of a single session, with shapes corresponding to individual rats. Arrows indicate the median values for individual rats across all sessions. ***p < 0.001 between them, Wilcoxon rank-sum test. A similar trend was observed in pairs limited to the TRNi neurons (although not significant).

Previous SectionNext Section

Discussion

In this study, we compared the temporal characteristics of activity in four distinct neuronal populations during voluntary movements in rats: excitatory RS and inhibitory FS neurons in the primary motor cortex (M1), excitatory TC neurons in the motor thalamus, and inhibitory neurons in the TRN. To achieve this, we employed a head-restrained condition in which rats repeatedly moved the right forelimb rather than a freely moving condition, thereby enabling temporally precise measurements of behavior (Isomura et al., 2009; Soma et al., 2017, 2019; Fig. 1). We used a combination of transgenic rats and AAV vectors to optogenetically and electrophysiologically identify and classify each neuron subtype (Fig. 2). Task-related neuronal activity was categorized into Hold-, Pull-, and Postpull-types based on the temporal profile of peak activity (Fig. 3). Among the four neuronal subtypes, M1-RS neurons exhibited the lowest proportion of Pull-type activity, the broadest distribution of peak activity timing, and the highest directional selectivity (Figs. 4, 5). In contrast, TRN neurons—similar to inhibitory M1-FS neurons—showed the highest proportion of Pull-type activity, with spike rates peaking near the pull onset and the lowest directional selectivity (Figs. 4, 5). Furthermore, pairs of M1-FS neurons and pairs of TRN neurons displayed a tendency for synchronous spiking (Fig. 6). The activity of TC neurons generally showed properties that were intermediate between those of other neuronal subtypes. We further compared neural activity between rats trained in a “low-demand” task condition with a constant motor load and those trained in a more difficult “high-demand” task condition that introduced a variable load (Fig. 1). Rats that learned the high-demand task exhibited less variability in pull latency and execution time, indicating more uniform movement execution. Interestingly, when comparing the low- and high-demand task groups, TRN neurons showed the most substantial differences among the four neuron subtypes: a greater proportion of Pull-type activity, higher spike rates (Fig. 4), a significant advancement in peak timing, and, paradoxically, higher directional selectivity (Fig. 5) in the high-demand condition. This movement-related activity profile contrasted with that observed during sleep spindles, in which TRN activity lagged behind TC activity (Fig. 7). The observation that the TRN leads TC activity during movements, particularly under high-demand task conditions, suggests that the mechanisms underlying voluntary movement representation are distinct from those governing sleep spindle generation. Taken together, our findings elucidate the spatiotemporal dynamics of M1, the motor thalamus, and TRN during voluntary movement. Furthermore, we provide the first evidence that the TRN is involved in demand-dependent motor control, as if adjusting its activity strength and timing to meet task demands. Consequently, a low-demand task may result in more variable and complex behavior, whereas a high-demand task may yield more controlled and stereotyped performance (Fig. 1).

Isomura et al. (2009) and Soma et al. (2017, 2019) reported neuronal activity in the primary and secondary motor cortices (M1 and M2) and the posterior parietal cortex during each phase of a similar lever/pedal movement task performed by head-fixed rats. We also observed three response patterns (Hold-, Pull-, and Postpull-type) across four neuronal populations: M1-RS, M1-FS, TC, and TRN. The activity characteristics of M1-RS and M1-FS neurons were consistent with previous reports. Saga et al. (2017) demonstrated the existence of “movement-related” TRN neurons in macaque monkeys that showed increased or decreased activity around movement onset. Although this study showed that the TRN had a higher proportion of neurons with increased (i.e., Pull-type) activity than the premotor cortex, we found that the proportion of Pull-type neurons was highest in M1-FS neurons, followed by the TRN and M1-RS. Thus, differentiating these neuronal subtypes enabled us to elucidate their response characteristics, including activity strength and directional selectivity. Notably, many inhibitory TRN neurons, similar to M1-FS neurons, were activated during movement execution (exhibiting Pull-type activity) rather than during movement cessation. This observation suggests that TRN neurons do not primarily function as gating mechanisms for motor commands. One study on locomotion in cats (Marlinski et al., 2012; Marlinski and Beloozerova, 2014) demonstrated that TRN neuron activity during walking was distributed across different phases of the gait cycle, depending on the limb being controlled and its interaction with TC neurons. This may be related to a gating mechanism for motor commands. Locomotion is the rhythmic and repetitive movement driven by a central pattern generator in the spinal cord. Its neural mechanisms are likely different from those of discrete forelimb movements, which are controlled by the cerebral cortex and thalamus. The concurrent activation of cortical, thalamic, and TRN neurons observed here appears to result from the coactivation of the direct and indirect pathways (excitatory and inhibitory effects, respectively) of the basal ganglia (Cui et al., 2013; Isomura et al., 2013). Such coordinated activation of excitatory and inhibitory systems for motor execution may be a general operating principle in the brain.

Next, we examined the neural circuit mechanisms underlying motor execution. It is well known that CT neurons in layers 5 and 6 (including M1-RS neurons) activate TC neurons, which provide feedback to the cortex. TRN neurons receive excitatory inputs from both TC and CT neurons through axonal collaterals, thereby providing inhibitory feedback to TC neurons (Pinault and Deschênes, 1998; Halassa and Acsády, 2016; Crabtree, 2018). Lam and Sherman (2015) demonstrated a thalamo-reticulo-thalamic closed–loop circuit, including the ventral anterior and lateral thalamic nuclei (VA/VL), by recording biphasic depolarizing and hyperpolarizing responses to laser-scanning photostimulation in VA/VL neurons. Hádinger et al. (2023) and Carroll et al. (2022) described two inhibitory pathways from layer 5 of the mouse frontal cortex to the thalamus via the TRN. The first pathway is feedforward inhibition, in which cortical neurons (M1/M2) and their target TRN neurons innervate the same thalamic nuclei—ventral medial (VM) and parafascicular (PF)—likely for signal gain control. The second pathway is lateral inhibition, in which M1/M2 neurons and their target TRN neurons innervate the VM and VL nuclei. This target mismatch is thought to filter information and enhance signal contrast. The TRN modulates the magnitude, scope, and timing of motor thalamic activity to suppress excessive excitation and unnecessary noise, thereby stabilizing the cortico-thalamo-cortical loop and enhancing the fidelity of motor commands. Future studies will reveal the specific connectivity of motor TRN neurons—namely, their source layer in the motor cortex and their target nuclei in the thalamus—and determine whether TRN neurons are responsible for lateral or feedforward inhibition in the cortico-thalamo-cortical loop.

A particular point of interest in this study was the observation that TRN neurons exhibited adaptive-like activity patterns depending on the task demands for motor precision. In the high-demand task group, TRN neurons showed a greater proportion and magnitude of Pull-type activity (Fig. 4) and earlier peak latency (Fig. 5) compared with other neuron subtypes. While inhibitory TRN and M1-FS neurons generally displayed low directional specificity, TRN neurons showed higher directional specificity under high-demand conditions in contrast to M1-RS neurons (Fig. 5). Moreover, we found significant spike synchrony in pairs of TRN and M1-FS neurons with Pull-type activity (Fig. 6). Such inhibitory synchronization can filter irrelevant activity in the loop, ultimately improving the signal-to-noise ratio of motor commands. Thus, these inhibitory TRN and M1-FS neurons appear not to encode a specific motor command (signal) but instead participate in coordinating excitatory and inhibitory neurons to improve temporal precision and suppress noise.

Several studies on attention and effort allocation have suggested that rats change their behavioral strategies depending on task difficulty or motor demands, both of which increase uncertainty. The risk-sensitive foraging theory suggests that an animal's internal state influences whether it adopts a “bold strategy” or a “cautious strategy” in an uncertain environment (Caraco et al., 1980). In reinforcement learning, when the environment is highly uncertain, animals increase “exploration” to find the optimal behavioral strategy, and when the environment becomes stable and predictable, they shift to “exploitation” (Daw et al., 2006). It has been shown that neurotransmitters such as acetylcholine and noradrenaline also regulate responses to uncertainty (Hasselmo and McGaughy, 2004; Yu and Dayan, 2005). Therefore, task demand-dependent modulation of TRN neuronal activity might result from top–down control by these neuromodulators in response to environmental uncertainty (Pita-Almenar et al., 2014).

Based on our observations, we propose a novel functional role for TRN neurons: optimizing the activity of the entire cortico-thalamo-cortical loop through local inhibition and signal selection and contributing to precise motor output by suppressing unnecessary activity, particularly when adaptive motor control is required. To the best of our knowledge, this is the first report demonstrating that the CT–TRN circuit is dynamically involved in regulating demand-dependent motor execution. However, our findings lack causal evidence, and future manipulation experiments and within-subject analyses after full skill acquisition will be needed to clarify the causal roles of the TRN circuit more precisely. In addition, abnormal activity in the CT–TRN circuit may be related to pathological conditions such as dystonia, tremor, and epilepsy. Therefore, future studies should address not only its physiological role in healthy animals but also its pathophysiological significance in disease models.

Previous SectionNext Section

Footnotes

    • Received September 26, 2025.
    • Revision received February 2, 2026.
    • Accepted February 24, 2026.

  • This work was supported by Grants-in-Aid for Scientific Research (B; JP19H03342 and JP23H02589 to Y.I.), for Transformative Research Areas (A; JP21H05242 to Y.I.), and for Challenging Research (Exploratory; JP24K21999 to Y.I.) from MEXT and JSPS; by CREST (JPMJCR1751 to Y.I.) and SPRING (JPMJSP2120 to M.S.) from JST; by the Takeda Science Foundation (to H.T. and Y.I.); and by the Center for Brain Integration Research (CBIR), Institute of Science, Tokyo. We thank all members of the Isomura Laboratory for their technical assistance and helpful discussions. We thank Editage (www.editage.jp) for English proofreading.

  • The authors declare no competing financial interests.

  • Correspondence should be addressed to Yoshikazu Isomura at isomura.y.b2f5{at}m.isct.ac.jp.
Previous Section
 

References

  1. ↵
    1. Arcelli P,
    2. Frassoni C,
    3. Regondi MC,
    4. De Biasi S,
    5. Spreafico R
    (1997) GABAergic neurons in mammalian thalamus: a marker of thalamic complexity? Brain Res Bull 42:27–37. https://doi.org/10.1016/s0361-9230(96)00107-4
    CrossRefMedline
  2. ↵
    1. Averbeck BB,
    2. Latham PE,
    3. Pouget A
    (2006) Neural correlations, population coding and computation. Nat Rev Neurosci 7:358–366. https://doi.org/10.1038/nrn1888
    CrossRefMedline
  3. ↵
    1. Bakker R,
    2. Tiesinga P,
    3. Kötter R
    (2015) The scalable brain atlas: instant web-based access to public brain atlases and related content. Neuroinformatics 13:353–366. https://doi.org/10.1007/s12021-014-9258-x
    CrossRefMedline
  4. ↵
    1. Bonjean M,
    2. Baker T,
    3. Lemieux M,
    4. Timofeev I,
    5. Sejnowski T,
    6. Bazhenov M
    (2011) Corticothalamic feedback controls sleep spindle duration in vivo. J Neurosci 31:9124–9134. https://doi.org/10.1523/JNEUROSCI.0077-11.2011
    Abstract/FREE Full Text
  5. ↵
    1. Calabrese E,
    2. Badea A,
    3. Watson C,
    4. Johnson GA
    (2013) A quantitative magnetic resonance histology atlas of postnatal rat brain development with regional estimates of growth and variability. Neuroimage 71:196–206. https://doi.org/10.1016/j.neuroimage.2013.01.017
    CrossRefMedline
  6. ↵
    1. Caraco T,
    2. Martindale S,
    3. Whittam TS
    (1980) An empirical demonstration of risk-sensitive foraging preferences. Anim Behav 28:820–830. https://doi.org/10.1016/S0003-3472(80)80142-4
    CrossRef
  7. ↵
    1. Carroll BJ,
    2. Sampathkumar V,
    3. Kasthuri N,
    4. Sherman SM
    (2022) Layer 5 of cortex innervates the thalamic reticular nucleus in mice. Proc Natl Acad Sci U S A 119:e2205209119. https://doi.org/10.1073/pnas.2205209119
    CrossRefMedline
  8. ↵
    1. Cicchetti DV
    (1994) Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology. Psychol Assess 6:284–290. https://doi.org/10.1037/1040-3590.6.4.284
    CrossRefMedline
  9. ↵
    1. Clemente-Perez A, et al.
    (2017) Distinct thalamic reticular cell types differentially modulate normal and pathological cortical rhythms. Cell Rep 19:2130–2142. https://doi.org/10.1016/j.celrep.2017.05.044
    CrossRefMedline
  10. ↵
    1. Crabtree JW
    (2018) Functional diversity of thalamic reticular subnetworks. Front Syst Neurosci 12:41. https://doi.org/10.3389/fnsys.2018.00041
    CrossRefMedline
  11. ↵
    1. Crick F
    (1984) Function of the thalamic reticular complex: the searchlight hypothesis. Proc Natl Acad Sci U S A 81:4586–4590. https://doi.org/10.1073/pnas.81.14.4586
    Abstract/FREE Full Text
  12. ↵
    1. Cui G,
    2. Jun SB,
    3. Jin X,
    4. Pham MD,
    5. Vogel SS,
    6. Lovinger DM,
    7. Costa RM
    (2013) Concurrent activation of striatal direct and indirect pathways during action initiation. Nature 494:238–242. https://doi.org/10.1038/nature11846
    CrossRefMedline
  13. ↵
    1. Daw ND,
    2. O'Doherty JP,
    3. Dayan P,
    4. Seymour B,
    5. Dolan RJ
    (2006) Cortical substrates for exploratory decisions in humans. Nature 441:876–879. https://doi.org/10.1038/nature04766
    CrossRefMedline
  14. ↵
    1. Dong P, et al.
    (2019) A novel cortico-intrathalamic circuit for flight behavior. Nat Neurosci 22:941–949. https://doi.org/10.1038/s41593-019-0391-6
    Medline
  15. ↵
    1. Fernandez LMJ,
    2. Lüthi A
    (2020) Sleep spindles: mechanisms and functions. Physiol Rev 100:805–868. https://doi.org/10.1152/physrev.00042.2018
    CrossRefMedline
  16. ↵
    1. Gardner RJ,
    2. Hughes SW,
    3. Jones MW
    (2013) Differential spike timing and phase dynamics of reticular thalamic and prefrontal cortical neuronal populations during sleep spindles. J Neurosci 33:18469–18480. https://doi.org/10.1523/JNEUROSCI.2197-13.2013
    Abstract/FREE Full Text
  17. ↵
    1. Hádinger N,
    2. Bősz E,
    3. Tóth B,
    4. Vantomme G,
    5. Lüthi A,
    6. Acsády L
    (2023) Region-selective control of the thalamic reticular nucleus via cortical layer 5 pyramidal cells. Nat Neurosci 26:116–130. https://doi.org/10.1038/s41593-022-01217-z
    CrossRefMedline
  18. ↵
    1. Halassa MM,
    2. Acsády L
    (2016) Thalamic inhibition: diverse sources, diverse scales. Trends Neurosci 39:680–693. https://doi.org/10.1016/j.tins.2016.08.001
    CrossRefMedline
  19. ↵
    1. Halassa MM,
    2. Siegle JH,
    3. Ritt JT,
    4. Ting JT,
    5. Feng G,
    6. Moore CI
    (2011) Selective optical drive of thalamic reticular nucleus generates thalamic bursts and cortical spindles. Nat Neurosci 14:1118–1120. https://doi.org/10.1038/nn.2880
    CrossRefMedline
  20. ↵
    1. Halassa MM,
    2. Chen Z,
    3. Wimmer RD,
    4. Brunetti PM,
    5. Zhao S,
    6. Zikopoulos B,
    7. Wang F,
    8. Brown EN,
    9. Wilson MA
    (2014) State-dependent architecture of thalamic reticular subnetworks. Cell 158:808–821. https://doi.org/10.1016/j.cell.2014.06.025
    CrossRefMedline
  21. ↵
    1. Hamada S,
    2. Nagase M,
    3. Yoshizawa T,
    4. Hagiwara A,
    5. Isomura Y,
    6. Watabe AM,
    7. Ohtsuka T
    (2021) An engineered channelrhodopsin optimized for axon terminal activation and circuit mapping. Commun Biol 4:461. https://doi.org/10.1038/s42003-021-01977-7
    Medline
  22. ↵
    1. Hasselmo ME,
    2. McGaughy J
    (2004) High acetylcholine levels set circuit dynamics for attention and encoding and low acetylcholine levels set dynamics for consolidation. Prog Brain Res 145:207–231. https://doi.org/10.1016/S0079-6123(03)45015-2
    CrossRefMedline
  23. ↵
    1. Hazan L,
    2. Zugaro M,
    3. Buzsáki G
    (2006) Klusters, NeuroScope, NDManager: a free software suite for neurophysiological data processing and visualization. J Neurosci Methods 155:207–216. https://doi.org/10.1016/j.jneumeth.2006.01.017
    CrossRefMedline
  24. ↵
    1. Hou G,
    2. Smith AG,
    3. Zhang ZW
    (2016) Lack of intrinsic GABAergic connections in the thalamic reticular nucleus of the mouse. J Neurosci 36:7246–7252. https://doi.org/10.1523/JNEUROSCI.0607-16.2016
    Abstract/FREE Full Text
  25. ↵
    1. Huo Y,
    2. Chen H,
    3. Guo ZV
    (2020) Mapping functional connectivity from the dorsal cortex to the thalamus. Neuron 107:1080–1094.e5. https://doi.org/10.1016/j.neuron.2020.06.038
    CrossRefMedline
  26. ↵
    1. Isomura Y,
    2. Harukuni R,
    3. Takekawa T,
    4. Aizawa H,
    5. Fukai T
    (2009) Microcircuitry coordination of cortical motor information in self-initiation of voluntary movements. Nat Neurosci 12:1586–1593. https://doi.org/10.1038/nn.2431
    CrossRefMedline
  27. ↵
    1. Isomura Y,
    2. Takekawa T,
    3. Harukuni R,
    4. Handa T,
    5. Aizawa H,
    6. Takada M,
    7. Fukai T
    (2013) Reward-modulated motor information in identified striatum neurons. J Neurosci 33:10209–10220. https://doi.org/10.1523/JNEUROSCI.0381-13.2013
    Abstract/FREE Full Text
  28. ↵
    1. Jager P, et al.
    (2021) Dual midbrain and forebrain origins of thalamic inhibitory interneurons. Elife 10:e59272. https://doi.org/10.7554/eLife.59272
    CrossRefMedline
  29. ↵
    1. Jiang S,
    2. Wu X,
    3. Rommelfanger NJ,
    4. Ou Z,
    5. Hong G
    (2022) Shedding light on neurons: optical approaches for neuromodulation. Natl Sci Rev 9:nwac007. https://doi.org/10.1093/nsr/nwac007
    CrossRefMedline
  30. ↵
    1. Jones EG
    (2007) The thalamus, Ed 2. Cambridge, UK: Cambridge University Press.
    Google Scholar
  31. ↵
    1. Kawabata M,
    2. Soma S,
    3. Saiki-Ishikawa A,
    4. Nonomura S,
    5. Yoshida J,
    6. Ríos A,
    7. Sakai Y,
    8. Isomura Y
    (2020) A spike analysis method for characterizing neurons based on phase locking and scaling to the interval between two behavioral events. J Neurophysiol 124:1923–1941. https://doi.org/10.1152/jn.00200.2020
    CrossRefMedline
  32. ↵
    1. Kimura R,
    2. Saiki A,
    3. Fujiwara-Tsukamoto Y,
    4. Ohkubo F,
    5. Kitamura K,
    6. Matsuzaki M,
    7. Sakai Y,
    8. Isomura Y
    (2012) Reinforcing operandum: rapid and reliable learning of skilled forelimb movements by head-fixed rodents. J Neurophysiol 108:1781–1792. https://doi.org/10.1152/jn.00356.2012
    CrossRefMedline
  33. ↵
    1. Kimura R,
    2. Saiki A,
    3. Fujiwara-Tsukamoto Y,
    4. Sakai Y,
    5. Isomura Y
    (2017) Large-scale analysis reveals populational contributions of cortical spike rate and synchrony to behavioural functions. J Physiol 595:385–413. https://doi.org/10.1113/JP272794
    CrossRefMedline
  34. ↵
    1. Klapoetke NC, et al.
    (2014) Independent optical excitation of distinct neural populations. Nat Methods 11:338–346. https://doi.org/10.1038/nmeth.2836
    CrossRefMedline
  35. ↵
    1. Koo TK,
    2. Li MY
    (2016) A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med 15:155–163. https://doi.org/10.1016/j.jcm.2016.02.012
    CrossRefMedline
  36. ↵
    1. Lam YW,
    2. Sherman SM
    (2015) Functional topographic organization of the motor reticulothalamic pathway. J Neurophysiol 113:3090–3097. https://doi.org/10.1152/jn.00847.2014
    CrossRefMedline
  37. ↵
    1. Lee JH,
    2. Latchoumane CV,
    3. Park J,
    4. Kim J,
    5. Jeong J,
    6. Lee KH,
    7. Shin HS
    (2019) The rostroventral part of the thalamic reticular nucleus modulates fear extinction. Nat Commun 10:4637. https://doi.org/10.1038/s41467-019-12496-9
    CrossRefMedline
  38. ↵
    1. Lehéricy S,
    2. Grand S,
    3. Pollak P,
    4. Poupon F,
    5. Le Bas JF,
    6. Limousin P,
    7. Jedynak P,
    8. Marsault C,
    9. Agid Y,
    10. Vidailhet M
    (2001) Clinical characteristics and topography of lesions in movement disorders due to thalamic lesions. Neurology 57:1055–1066. https://doi.org/10.1212/wnl.57.6.1055
    Medline
  39. ↵
    1. Lewis LD,
    2. Voigts J,
    3. Flores FJ,
    4. Schmitt LI,
    5. Wilson MA,
    6. Halassa MM,
    7. Brown EN
    (2015) Thalamic reticular nucleus induces fast and local modulation of arousal state. Elife 4:e08760. https://doi.org/10.7554/eLife.08760
    CrossRefMedline
  40. ↵
    1. Li Y, et al.
    (2020) Distinct subnetworks of the thalamic reticular nucleus. Nature 583:819–824. https://doi.org/10.1038/s41586-020-2504-5
    CrossRefMedline
  41. ↵
    1. Liu PF, et al.
    (2021) Modulation of itch and pain signals processing in ventrobasal thalamus by thalamic reticular nucleus. iScience 25:103625. https://doi.org/10.1016/j.isci.2021.103625
    Medline
  42. ↵
    1. Marlinski V,
    2. Beloozerova IN
    (2014) Burst firing of neurons in the thalamic reticular nucleus during locomotion. J Neurophysiol 112:181–192. https://doi.org/10.1152/jn.00366.2013
    CrossRefMedline
  43. ↵
    1. Marlinski V,
    2. Sirota MG,
    3. Beloozerova IN
    (2012) Differential gating of thalamocortical signals by reticular nucleus of thalamus during locomotion. J Neurosci 32:15823–15836. https://doi.org/10.1523/JNEUROSCI.0782-12.2012
    Abstract/FREE Full Text
  44. ↵
    1. McAlonan K,
    2. Brown VJ,
    3. Bowman EM
    (2000) Thalamic reticular nucleus activation reflects attentional gating during classical conditioning. J Neurosci 20:8897–8901. https://doi.org/10.1523/JNEUROSCI.20-23-08897.2000
    Abstract/FREE Full Text
  45. ↵
    1. McAlonan K,
    2. Cavanaugh J,
    3. Wurtz RH
    (2008) Guarding the gateway to cortex with attention in visual thalamus. Nature 456:391–394. https://doi.org/10.1038/nature07382
    CrossRefMedline
  46. ↵
    1. Mitani K,
    2. Kawabata M,
    3. Isomura Y,
    4. Sakai Y
    (2022) Automated and parallelized spike collision tests to identify spike signal projections. iScience 25:105071. https://doi.org/10.1016/j.isci.2022.105071
    Medline
  47. ↵
    1. Nonomura S, et al.
    (2018) Monitoring and updating of action selection for goal-directed behavior through the striatal direct and indirect pathways. Neuron 99:1302–1314.e5. https://doi.org/10.1016/j.neuron.2018.08.002
    CrossRefMedline
  48. ↵
    1. Ohno-Shosaku T,
    2. Yoneda M,
    3. Maejima T,
    4. Wang M,
    5. Kikuchi Y,
    6. Onodera K,
    7. Kanazawa Y,
    8. Takayama C,
    9. Mieda M
    (2023) Action sequence learning is impaired in genetically modified mice with the suppressed GABAergic transmission from the thalamic reticular nucleus to the thalamus. Neuroscience 532:87–102. https://doi.org/10.1016/j.neuroscience.2023.09.019
    Medline
  49. ↵
    1. Pinault D
    (2004) The thalamic reticular nucleus: structure, function and concept. Brain Res Brain Res Rev 46:1–31. https://doi.org/10.1016/j.brainresrev.2004.04.008
    CrossRefMedline
  50. ↵
    1. Pinault D,
    2. Deschênes M
    (1998) Projection and innervation patterns of individual thalamic reticular axons in the thalamus of the adult rat: a three-dimensional, graphic, and morphometric analysis. J Comp Neurol 391:180–203. https://doi.org/10.1002/(sici)1096-9861(19980209)391:2<180::aid-cne3>3.0.co;2-z
    CrossRefMedline
  51. ↵
    1. Pita-Almenar JD,
    2. Yu D,
    3. Lu HC,
    4. Beierlein M
    (2014) Mechanisms underlying desynchronization of cholinergic-evoked thalamic network activity. J Neurosci 34:14463–14474. https://doi.org/10.1523/JNEUROSCI.2321-14.2014
    Abstract/FREE Full Text
  52. ↵
    1. Rios A,
    2. Soma S,
    3. Yoshida J,
    4. Nonomura S,
    5. Kawabata M,
    6. Sakai Y,
    7. Isomura Y
    (2019) Differential changes in the lateralized activity of identified projection neurons of motor cortex in hemiparkinsonian rats. eNeuro 6:ENEURO.0110-19.2019. https://doi.org/10.1523/ENEURO.0110-19.2019
  53. ↵
    1. Rios A, et al.
    (2023) Reward expectation enhances action-related activity of nigral dopaminergic and two striatal output pathways. Commun Biol 6:914. https://doi.org/10.1038/s42003-023-05288-x
    Medline
  54. ↵
    1. Saga Y,
    2. Nakayama Y,
    3. Inoue KI,
    4. Yamagata T,
    5. Hashimoto M,
    6. Tremblay L,
    7. Takada M,
    8. Hoshi E
    (2017) Visuomotor signals for reaching movements in the rostro-dorsal sector of the monkey thalamic reticular nucleus. Eur J Neurosci 45:1186–1199. https://doi.org/10.1111/ejn.13421
    CrossRefMedline
  55. ↵
    1. Saiki A,
    2. Kimura R,
    3. Samura T,
    4. Fujiwara-Tsukamoto Y,
    5. Sakai Y,
    6. Isomura Y
    (2014) Different modulation of common motor information in rat primary and secondary motor cortices. PLoS One 9:e98662. https://doi.org/10.1371/journal.pone.0098662
    CrossRefMedline
  56. ↵
    1. Sakairi T,
    2. Kawabata M,
    3. Rios A,
    4. Sakai Y,
    5. Isomura Y
    (2025) Hippocampal sharp-wave ripples decrease during physical actions including consummatory behavior in immobile rodents. J Neurosci 45:e0080252025. https://doi.org/10.1523/JNEUROSCI.0080-25.2025
    Abstract/FREE Full Text
  57. ↵
    1. Soma S,
    2. Saiki A,
    3. Yoshida J,
    4. Ríos A,
    5. Kawabata M,
    6. Sakai Y,
    7. Isomura Y
    (2017) Distinct laterality in forelimb-movement representations of rat primary and secondary motor cortical neurons with intratelencephalic and pyramidal tract projections. J Neurosci 37:10904–10916. https://doi.org/10.1523/JNEUROSCI.1188-17.2017
    Abstract/FREE Full Text
  58. ↵
    1. Soma S, et al.
    (2019) Ipsilateral-dominant control of limb movements in rodent posterior parietal cortex. J Neurosci 39:485–502. https://doi.org/10.1523/JNEUROSCI.1584-18.2018
    Abstract/FREE Full Text
  59. ↵
    1. Soma S,
    2. Ohara S,
    3. Nonomura S,
    4. Suematsu N,
    5. Yoshida J,
    6. Pastalkova E,
    7. Sakai Y,
    8. Tsutsui KI,
    9. Isomura Y
    (2023) Rat hippocampal CA1 region represents learning-related action and reward events with shorter latency than the lateral entorhinal cortex. Commun Biol 6:584. https://doi.org/10.1038/s42003-023-04958-0
    Medline
  60. ↵
    1. Steriade M,
    2. McCormick DA,
    3. Sejnowski TJ
    (1993) Thalamocortical oscillations in the sleeping and aroused brain. Science 262:679–685. https://doi.org/10.1126/science.8235588
    Abstract/FREE Full Text
  61. ↵
    1. Takekawa T,
    2. Isomura Y,
    3. Fukai T
    (2010) Accurate spike sorting for multi-unit recordings. Eur J Neurosci 31:263–272. https://doi.org/10.1111/j.1460-9568.2009.07068.x
    CrossRefMedline
  62. ↵
    1. Takekawa T,
    2. Isomura Y,
    3. Fukai T
    (2012) Spike sorting of heterogeneous neuron types by multimodality-weighted PCA and explicit robust variational Bayes. Front Neuroinform 6:5. https://doi.org/10.3389/fninf.2012.00005
    Medline
  63. ↵
    1. von Krosigk M,
    2. Bal T,
    3. McCormick DA
    (1993) Cellular mechanisms of a synchronized oscillation in the thalamus. Science 261:361–364. https://doi.org/10.1126/science.8392750
    Abstract/FREE Full Text
  64. ↵
    1. Weese GD
    (2022) Action selection is impaired by unilateral lesions in the rostral thalamic reticular nucleus. Behav Neurosci 136:230–242. https://doi.org/10.1037/bne0000507
    CrossRefMedline
  65. ↵
    1. Wimmer RD,
    2. Schmitt LI,
    3. Davidson TJ,
    4. Nakajima M,
    5. Deisseroth K,
    6. Halassa MM
    (2015) Thalamic control of sensory selection in divided attention. Nature 526:705–709. https://doi.org/10.1038/nature15398
    CrossRefMedline
  66. ↵
    1. Yu AJ,
    2. Dayan P
    (2005) Uncertainty, neuromodulation, and attention. Neuron 46:681–692. https://doi.org/10.1016/j.neuron.2005.04.026
    CrossRefMedline
  67. ↵
    1. Yu XJ,
    2. Xu XX,
    3. He S,
    4. He J
    (2009) Change detection by thalamic reticular neurons. Nat Neurosci 12:1165–1170. https://doi.org/10.1038/nn.2373
    CrossRefMedline
  68. ↵
    1. Zohary E,
    2. Shadlen MN,
    3. Newsome WT
    (1994) Correlated neuronal discharge rate and its implications for psychophysical performance. Nature 370:140–143. https://doi.org/10.1038/370140a0
    CrossRefMedline

Navigate This Article

  1. Published online before print 6 March 2026, doi: 10.1523/JNEUROSCI.1827-25.2026 The Journal of Neuroscience, 46(14): e1827252026
  1. AbstractFree
  2. » Full Text
  3. Full Text (PDF)

ORCID

  1. Profile for Rios, A.https://orcid.org/0000-0002-2849-0609
  2. Profile for Isomura, Y.https://orcid.org/0000-0003-0393-3124
Responses to This Article
Be the first to respond to this article
Submit a Response
View Similar Articles
Similar articles in this journal»
Advertisement