Transient effects in corticospinal and reticulospinal tract excitability induced by motor skill and isometric resistance training

Abstract

Coordinated movement relies on the proper integration of multiple neural circuits. Motor training can alter the excitability of neural circuits controlling movement, but the pathway-specific effects to the lower limb of motor skill versus isometric resistance training remain unclear. Here, we tested how single 30-minute sessions of cue-paced motor skill and isometric resistance training modulate corticospinal, reticulospinal, and spinal excitability in unimpaired adults (N = 23). Using motor-evoked potentials via transcranial magnetic stimulation, we found motor skill training increased corticospinal excitability, while isometric resistance training did not. In contrast, by assessing reticulospinal tract excitability by StartReact responses and measuring spinal excitability with H/M ratios, F-wave response amplitude, and persistence, we found that each tract’s excitability remained largely unchanged. These results suggest that short-term motor skill training selectively enhances corticospinal tract excitability without a measurable impact on spinal or reticulospinal circuits. These results highlight the influence of task complexity on distal lower limb excitability and provide a framework for evaluating neural adaptations across corticospinal, reticulospinal, and spinal circuits.

1. Introduction

Neural excitability in the human motor system is shaped by both the intrinsic properties of descending motor pathways and their adaptive responses to experience (Perez et al., 2004; Sangari and Perez, 2019; Siddique et al., 2020). Activity-dependent reorganization of neural pathways after various training paradigms (Angeli et al., 2018; Bilchak et al., 2021; Gill et al., 2018; Seáñez et al., 2022; Wagner et al., 2018) can alter the excitability of circuits that contribute to skilled movement, posture, and coordination. Therefore, identifying the neural pathways that are prone to changes in excitability via different forms of activity is essential for understanding motor control and for developing training approaches that selectively engage targeted circuits.

The corticospinal and reticulospinal tracts are major descending white matter pathways with descending projections onto spinal motoneurons and interneurons in mammals (Baker et al., 2015; Jankowska and Edgley, 2006; Lemon, 2008). The corticospinal tract has been shown to contribute to the voluntary control of walking (Capaday et al., 1999; Schubert et al., 1997) and distal movements (Welniarz et al., 2017). While corticospinal tract excitability has been shown to increase after motor skill training (Perez et al., 2004), it appears to remain unchanged after specific resistance (Ansdell et al., 2020) and balance training (Bakker et al., 2021), and reduced after exercise-induced fatigue training (Clos et al., 2020) in the lower limbs. However, a recent study in upper limbs has shown that synchronizing strength training with an external cue, such as a metronome, may add a skill or complexity component to the training, thereby increasing corticospinal tract excitability (Leung et al., 2017). The reticulospinal tract has been shown to contribute to postural adjustments (Nonnekes et al., 2015; Prentice and Drew, 2001; Takakusaki, 2017), gross hand function (Baker and Perez, 2017), selecting the appropriate force level during movements (Buford and Davidson, 2004; Glover and Baker, 2020), modulating flexor and extensor motor neurons to maintain balance (Drew et al., 2004; Mackinnon, 2018), and performing bimanual tasks (Maslovat et al., 2020). The spinal cord serves as the integrative center where corticospinal and reticulospinal projections interact with spinal circuits controlling movement (Bilchak et al., 2021; Moreno-López et al., 2016; Pierrot-Deseilligny and Burke, 2012; Siddique et al., 2020). However, studies directly comparing training-induced changes between the corticospinal, spinal, and reticulospinal tracts after similar activity-based training are lacking, especially in the lower limb, limiting our understanding of where neural adaptations occur and which motor circuits are most responsive to different training modalities.

In this study, we investigated the effect of single sessions of cue-paced motor skill and cue-paced isometric resistance training on changes in corticospinal, spinal, and reticulospinal tract excitability for lower limb muscles in twenty-three unimpaired individuals. Corticospinal tract excitability was measured using motor-evoked potentials (MEPs) elicited by transcranial magnetic stimulation (TMS) (Eisner-Janowicz et al., 2023; Perez et al., 2004; Spampinato et al., 2023), specifically targeting the right tibialis anterior muscle. Spinal excitability was evaluated through H-reflexes, M-waves (Lagerquist et al., 2006; Poon et al., 2008; Scaglioni et al., 2002) and F-waves (Eisner-Janowicz et al., 2023; Kumru et al., 2021; Milanov, 1992; Weber, 1998) elicited by peripheral nerve stimulation of the same muscle. Reticulospinal tract excitability was assessed using the StartReact response, defined as a shortening in reaction time following a startling auditory stimulus that engages the reticulospinal tract (Fisher et al., 2013), with all measures collected before and after a 30-minute training session. We hypothesized that single-session motor skill training would selectively enhance corticospinal tract excitability without affecting reticulospinal excitability, whereas resistance training would enhance reticulospinal tract excitability without influencing corticospinal excitability.

2. Methods

2.1. Participants

Thirty-two unimpaired participants provided informed consent to participate in this study, which was reviewed and approved by Washington University in St. Louis’ Institutional Review Board, and 32 began participation in the experiments. Nine individuals were excluded from analysis for the following reasons: protocol change (N=3), participant personal scheduling issue (N=1), TMS motor threshold too high (N=1), could not tolerate TMS (N=1), and identified as outliers (N=3; one due to abnormal MEP exceeding 100% M_max, two due to EMG recording errors affecting reaction time measurement). Participant demographics, group designation, and excluded participants are listed in Supplementary Table S1.

The study was divided into two distinct groups. Group A underwent evaluations of corticospinal and spinal excitability using transcranial magnetic stimulation (TMS) and peripheral nerve stimulation of the posterior tibial and common peroneal nerves, respectively (Mesrati and Vecchierini, 2004; Milanov, 1992; Misiaszek, 2003; Spampinato et al., 2023) (Figure 1a). Group B underwent reticulospinal tract excitability evaluations using a StartReact evaluation (Fisher et al., 2013; Sangari and Perez, 2020) (Figure 1b). The right tibialis anterior muscle was selected as the focal point of this neurophysiological investigation due to its functional relevance in dorsiflexion and its proven reliability when assessed with our non-invasive measurement techniques (Brangaccio et al., 2024; Eisner-Janowicz et al., 2023; Hayman et al., 2025). Both groups underwent the same cross-over design, which included a 30-minute motor skill training session and a 30-minute isometric resistance training session on separate days. The order of these sessions was randomized for each participant. To avoid carryover effects, these sessions were scheduled with a minimum of 24 hours between them, exceeding reported durations of acute training-induced neurophysiological changes (Latella et al., 2017; Perez et al., 2004). Group A participants had an average interval of 12.0 ± 14.8 days (range: 1–49 days), while Group B participants had an average of 6.3 ± 6.2 days (range: 1–23 days).

Figure 1. Neurophysiology experiments to evaluate cortico-reticulo-spinal excitability following motor skill and isometric resistance training.

(a) Assessment of corticospinal and spinal excitability in Group A. Participants in Group A (N = 11) underwent transcranial magnetic stimulation (TMS) to assess corticospinal tract excitability and peripheral nerve stimulation of the posterior tibial and common peroneal nerves to assess spinal excitability. (b) Assessment of reticulospinal tract excitability in Group B. Participants in Group B (N = 12) completed StartReact evaluations to measure reticulospinal tract excitability. Excitability measurements were taken immediately before and after each training session. Both groups participated in a randomized crossover trial involving a 30-minute motor skill training session using a body-machine interface (BoMI) and a 30-minute isometric resistance training session performed on separate days.

We performed a power analysis with 4 pilot participants to determine the sample size required to detect a significant change in motor-evoked potentials at 180% of resting motor threshold with an alpha level of 0.0125 to account for comparisons across multiple comparisons, and a power of 95%. With an effect size (Cohen’s d) of 1.788, we determined that a minimum of 11 participants would be necessary. Therefore, we aimed to complete testing on 12 participants for Groups A and B.

2.2. Motor skill training

During motor skill training, participants sat in a Biodex (System 4 Pro^™, Biodex Medical Systems, USA) isokinetic dynamometer, with both legs hanging freely. They used a body-machine interface (BoMI) to control a computer cursor (Figure 2a). The BoMI comprised of four non-invasive, wireless inertial measurement units (IMUs) (3-Space^™ Sensors, Yost Labs, USA) secured to adjustable straps on the participant’s right foot to record foot and ankle movements.

Figure 2. Improvements in movement coordination and control during motor skill training using an ankle-operated body-machine interface.

(a) BoMI set up for lower-limb motor skill training. Participants used right foot and ankle movements captured by IMUs to control a 2D cursor on the screen. A decoder trained with PCA translated 8-dimensional kinematic data into 2-dimensional cursor movements. (b) Representative cursor trajectories and velocity profiles during training. Cursor paths from an example participant (MS006) in blocks 1 and 5 illustrate straighter, more directed movements as training progressed. Average velocity traces show smoother and more coordinated movements over time, with fewer peaks and reduced variability. (c–f) Group-level improvements in movement quality, shown as mean values with error bars representing the standard deviation. (c) Jerk. A measure of movement smoothness, defined as the rate of change of acceleration, decreased across training. (e) Movement time. The time taken to move from the center target to the outer target was reduced. (e) Path length. The total distance traveled by the cursor from the center to the target decreased across training blocks. (f) End-point error. The final position of the cursor at the 4-second mark, relative to the target, remained consistent. Asterisks above the group averages in (c-f) denote Bonferroni-corrected statistical significance from paired comparisons between performance measures of first and last training blocks: *p < 0.05, **p < 0.01, *** p < 0.001, ‘n.s.’ p > 0.05.

During the calibration phase, participants were instructed to perform “free movements” within a comfortable range for 55 seconds. A PCA was performed on the Euler angles obtained from the IMUs to identify the principal components that accounted for the greatest amounts of variance (Seanez-Gonzalez et al., 2017). The top two principal components were used to control the horizontal and vertical movements of a computer cursor so that the 8-dimensional body motion vector (roll and pitch from 4 sensors) was projected into a 2-dimensional cursor control vector (cursor x and y). Participants could then use their ankle movements to control the computer cursor.

After calibration, participants familiarized themselves with the BoMI before completing five center-out-reaching tasks over the 30-minute session (Casadio et al., 2010; Pierella et al., 2017; Seáñez-González et al., 2016; Thorp et al., 2016). Participants wore headphones (U UFO Over-Ear Headphones, Bluedio, China), which provided a 1 Hz auditory cue to guide them to time their target reaches to conclude at four seconds, encouraging slow and steady movements. If the tasks were completed before the 30-minute mark, participants engaged in a maze game of increasing difficulty to maintain consistent training duration across training paradigms.

2.3. Isometric resistance training

In the isometric resistance training session (Figure 3a), participants sat in the Biodex chair with the left leg resting on a footrest and the right leg positioned with the foot attached to the dynamometer. The hip was positioned at 120°, the knee at 160°, and the ankle at 110° (measured from neutral), with a permissible variance of ±10° to ensure participant comfort and prevent muscle stretch (Pierrot-Deseilligny and Burke, 2012). Before the training began, the maximum torque for plantar flexion and dorsiflexion was obtained by asking participants to perform three consecutive maximal effort ankle joint pushes (for plantar flexion) or pulls (for dorsiflexion). During the training, participants executed slow-ramped movements to reach 30% of their maximum torque in both directions. The training comprised of three blocks of 12 contraction trials, each trial including both dorsiflexion and plantar flexion isometric contractions, for a total training time of 30 minutes. Participants took a 30-second rest between trials and a 2-minute break between blocks. To help maintain consistent velocity during the ramped movements, a 1 Hz auditory cue was provided, similar to the one used in motor skill training. Participants used the cue to guide a sequence of 3-second dorsiflexion and plantarflexion ramps to 30% of their maximum voluntary contraction, each followed by 3-second holds and returns to rest. The auditory cue provided clear timing for when to initiate, sustain, and release each contraction.

Figure 3: Isometric resistance and motor skill training result in similar reductions in EMG amplitude.

(a) Isometric resistance training protocol. Participants completed three blocks of slow-ramped dorsiflexion and plantar flexion contractions at 30% of their maximum voluntary contraction (MVC), paced by a 1 Hz auditory cue to ensure consistent timing. (b) Representative torque traces during training. Torque recordings from participant MS005 during the first and last training blocks show consistent force generation across the session. (c) EMG burst and RMS traces for fatigue assessment. The top trace shows example tibialis anterior EMG bursts from participant MS018 during maximum voluntary contractions (MVCs); the bottom trace shows the root mean square (RMS) of the full-wave rectified EMG signal. The area under the RMS curve over 2.25 seconds after onset was used to quantify EMG amplitude. (d, e) Group-level data are presented as mean values with error bars representing the standard deviation. (d) Training-specific reductions in EMG amplitude. Group-level analysis revealed a decrease in EMG amplitude following both motor skill and isometric resistance training sessions. (e) Comparison of change in MVC amplitude across training types. Reductions in EMG amplitude were comparable between the two training protocols, indicating similar levels of neuromuscular fatigue. Asterisks above the group averages in (d, e) denote Bonferroni-corrected statistical significance from paired comparisons of AUC before and after individual training and between trainings: *p < 0.05, **p < 0.01, ‘n.s.’ p > 0.05.

2.4. Electromyography data acquisition

Wireless surface electrodes (Trigno^® Avanti, Delsys Inc., USA) were used to record electromyography (EMG) data with sensors placed bilaterally according to SENIAM guidelines covering the rectus femoris, vastus lateralis, tibialis anterior, medial gastrocnemius, and soleus. Prior to electrode placement, the skin over the muscle belly was shaved, if necessary, and cleaned using abrasive gel (NuPrep^®, Weaver and Co., USA) applied with a Q-tip^®, followed by wiping with alcohol pads. An additional EMG sensor with analog input (DC-X06 Analog Input, Delsys Inc., USA) was used to capture stimulation onset for offline alignment between stimulation pulses and EMG data. EMG data was amplified using a data acquisition system (Trigno^® Avanti Research+, Delsys Inc., USA; gain: 300; bandwidth 20–450 Hz), and sampled at either 2148 Hz or 2000 Hz, with the difference in sampling rate due to a lab software update that occurred mid-study. All EMG data was displayed in real-time using custom-built software written by our group in Python v3.11 (Bryson et al., 2023).

2.5. Transcranial magnetic stimulation

TMS targeting the left motor cortex was used to evoke motor-evoked potentials in the right tibialis anterior to assess the excitability of the corticospinal tract (Lotze, 2003; Perez et al., 2004) (Figure 4a). A Magstim 200² (MOP01-EN, Magstim, UK) stimulator with a 110 mm double-cone coil (Kesar et al., 2018) (MOP21-EN, peak magnetic field of ≥ 1.2T) and a monophasic (100 ms) waveform was employed for TMS (Figure 4b). Participants sat in the Biodex isokinetic dynamometer with their hips positioned at 120°, knee at 160°, and ankle at 110°, with the right foot strapped securely to the dynamometer to ensure consistent positioning during stimulation. The Cz point, representing the scalp vertex, was identified by finding the intersection between the line from the nasion to the inion and the line between the left and right pre-auricular points. The intersection point, Cz, was marked using a skin-safe marker (Tondaus^®, Model T3023, Surgical skin marker, China).

Figure 4. Increases in corticospinal tract excitability following motor skill and isometric resistance training measured via transcranial magnetic stimulation.

(a) TMS was applied to the left motor cortex to elicit MEPs in the right tibialis anterior. (b) NeuroNavigation setup for hotspot targeting. A custom-built NeuroNavigation software, using infrared camera–based motion capture markers, tracked the position and orientation of a double-cone coil to consistently target the tibialis anterior hotspot before and after each training session. A superior view of the head illustrates the coil rotated 45° in the counterclockwise (CCW) direction from the posterior midsagittal plane (Ni et al., 2011; Richter et al., 2013), with the coil handle oriented perpendicular to the scalp (Thomas and Gorassini, 2005). Arrows shown in both the three-dimensional and superior views indicate the direction of the current flowing in the coil. (c) Average MEP responses at rest before and after training for a representative participant (MS004) illustrate increases in response amplitude following motor skill training. (d) Corresponding recruitment curve of (c). MEP amplitudes increased across multiple TMS intensities (100%, 150%, 180%, 200%) after training. (e) Percent change in MEP amplitude during pre-activation and rest conditions. The left panel shows percent changes in MEP amplitude during the pre-activation condition (15% dorsiflexion), and the right panel shows percent changes during rest, shown as mean values with error bars representing the standard deviation. Both plots represent group-level responses across all participants following motor skill and isometric resistance training. (f) Training-induced changes across evaluation conditions. Values are estimated marginal means (EMM) of MEP amplitudes, back-transformed from the log scale; error bars indicate 95% CIs. MEP amplitude increases following motor skill training. Asterisks above the bars denote post-hoc Holm-adjusted significance values for interactions between training types and evaluation conditions: *p < 0.05, **p < 0.01, *** p < 0.001, ‘n.s.’ p > 0.05.

The TMS coil and participant’s head 3D position and orientation were tracked via infrared cameras (Miqus Hybrid, Qualysis, Sweden) and reflective markers (Super-spherical Markers, Qualysis, Sweden) using a custom-built NeuroNavigation software designed in MATLAB (2024a, Mathworks, USA) (Figure 4b). At the start of each session, a large grid search was performed to identify the hotspot for the right tibialis anterior muscle, which is approximately 1.6 cm lateral and 0.8 cm posterior to C_z (Sivaramakrishnan et al., 2016). The TMS coil was positioned at 45° in the counterclockwise (CCW) direction from the posterior midsagittal plane to induce posterior-anterior current flow (Ni et al., 2011; Richter et al., 2013) (Figure 4b). Stimulation intensity was gradually increased from 30% maximum stimulator output at 5% increments until a short-latency response (20–30 ms) was observed in the contralateral tibialis anterior with a peak-to-peak amplitude of at least 50 μV. The stimulator intensity was recorded as the general motor threshold, and the TMS coil 3D position and orientation with respect to the participant’s head were saved as the initial hotspot location for the tibialis anterior. A hotspot search was then conducted using 120% of the general motor threshold to identify the optimal location for the tibialis anterior targeting based on the largest MEPs. MEPs were evoked at 9 locations spaced 1 cm apart following a 3×3 grid pattern around the initial hotspot, while the peak-to-peak amplitudes of the MEPs were quantified and displayed in real-time using the custom-built software. A higher-precision search was conducted around the location with the highest MEP amplitude by moving the coil 0.3 cm in the posterior, anterior, medial, and lateral directions. The 3D position and head orientation with the highest MEP amplitude were saved as the optimum tibialis anterior hotspot, and the resting motor threshold was determined at this location.

To find the tibialis anterior resting motor threshold at the identified hotspot, the stimulator intensity was reduced in 1–2% increments until the lowest intensity that could evoke an MEP in at least 3 out of 5 stimuli with a peak-to-peak amplitude ≥ 50 μV (Groppa et al., 2012; Perez et al., 2004; Rossini et al., 1994). Recruitment curves for MEP responses (Figure 4c) were collected at intensities of 100%, 150%, 180%, and 200% of the resting motor threshold presented in a randomized order with ten repetitions for each intensity. These recruitment curves were obtained before and after training, with the higher-precision search conducted post-training around the hotspot identified before training (Figure 4d). TMS evaluations were conducted both at rest and during 15% dorsiflexion contraction using a custom-built real-time visualization software displaying the moving average of tibialis anterior EMG feedback (Python v3.11). In the pre-activation condition, the experimenter manually delivered the TMS trigger as soon as the participant’s tibialis anterior EMG exceeded the 15% MVC threshold target. Using EMG allows for a more consistent evaluation of neuromuscular drive, even if participant force production declines due to fatigue after training (Hill et al., 2016; Kamimura and Muramatsu, 2018). Participants were able to rest between pulses, and the order of active vs. passive evaluations was randomized for each participant. As some participants could not tolerate all TMS intensities, the number of samples varied across TMS intensities.

2.6. Peripheral nerve stimulation

Peripheral nerve stimulation targeting the posterior tibial and common peroneal nerves was used to assess spinal synaptic transmission efficacy of Ia sensory fibers to the homonymous motor fibers (Palmieri et al., 2002; Pierrot-Deseilligny and Mazevet, 2000). Supramaximal electrical stimulation was used to assess alpha-motoneuron excitability through the antidromic activation of motor neurons, serving as a complementary method to evaluate the lower neuronal circuits (Mesrati and Vecchierini, 2004; Milanov, 1992) (Figure 5a). Stimulation was delivered using a biphasic constant-current stimulator (Digitimer DSR8, Digitimer Ltd., U.K.) with round surface electrodes (PALS Neurostimulation Electrodes, Axelgaard Manufacturing Co., Ltd., USA) of a 3.2 cm diameter and a National Instruments USB-6001 (NI USB 6001, National Instruments, USA) to trigger 1 ms (per phase) biphasic stimulation pulses. The longer pulse duration was chosen to optimize H-reflex elicitation, as activation of efferent fibers is still possible with 1 ms pulses (Keesey et al., 2024; Panizza et al., 1992). The stimulating cathode was placed on the popliteal fossa of the right leg, with the return anode positioned 2 cm proximally to the cathode (Figure 5a). The electrode configuration allowed for the elicitation of H-reflex, M-waves, and F-waves from the tibialis anterior and medial gastrocnemius simultaneously (Goulart and Valls-Solé, 2001; Poon et al., 2008). A recruitment curve was created using a non-linear sweep of sixteen stimulation amplitudes, ranging from 50% of the lowest H-reflex threshold to 110% of the highest stimulation amplitude that produced the maximal M-wave response (M_max) (Keesey et al., 2024) (Figure 5b). Three repetitions were performed at each stimulation amplitude with a 7-second interval between pulses. F-waves were elicited by delivering 60 pulses at 125% of M_max amplitude, with a 1-second interval between pulses (Figure 5f).

Figure 5. Effects of motor skill training and isometric resistance training on synaptic spinal excitability and spinal motor neuron excitability.

(a) Peripheral nerve stimulation setup, response pathways, and representative responses. (1) Two 3.2 cm diameter surface electrodes were positioned on the popliteal fossa over the common peroneal nerve, with the anode placed 2 cm proximal to the cathode. (2) The M-wave arises from the direct activation of motor axons, leading to a short-latency response. The H-reflex follows stimulation of Ia afferent fibers, which synapse monosynaptically onto alpha motor neurons in the spinal cord, producing a long-latency, reflex-mediated response. The F-wave results from antidromic activation of alpha motor neurons, leading to depolarization of the soma and a recurrent orthodromic response via the same efferent motor pathway. (3) Representative H-reflex, M-wave, and F-wave responses are shown from participant MS018. The M-wave has been removed from the F-wave response for visualization purposes. (b) H-reflex and M-wave recruitment curves of participant MS018. (c-e, g, h) Group-level data are presented as mean values with error bars representing the standard deviation. (c) Percent change in H/M ratio. (d) Percent change in M_max amplitude. A significant increase in M_max was observed following isometric resistance training. (e) Percent change in H_max. (f) Representative F-wave responses of participant MS014. (g) Percent change in F-wave persistence. (h) Percent change in F-wave amplitude. Asterisks above the group averages in (c-e) and (g, h) denote Bonferroni-corrected statistical significance from paired comparisons between training types: *p < 0.05, **p < 0.01. (Sig.) ‘X’ markers below the plot indicate Bonferroni-corrected significant training effects within each condition of percent change based on one-sample tests against μ = 0 (‘X’ p < 0.05, ‘XX’ p < 0.01, ‘n.s.’ p > 0.05).

2.7. StartReact evaluation

StartReact refers to an accelerated release of a prepared motor action that is initiated by a startling stimulus (Fisher et al., 2013; Honeycutt et al., 2013; Valls-Solé et al., 2008). This phenomenon occurs when a startling stimulus is introduced during the preparation phase of a voluntary movement, resulting in the execution of the planned action occurring significantly faster than non-startling auditory or visual cues (Valls-Solé et al., 2008). A custom-built software (Python v3.11) was used to collect synchronized torque and EMG data stream and display them as biofeedback in real-time (43-inch Class FULL HD LCD Display, Panasonic, Japan). During the StartReact task, participants sat in the Biodex (System 4 Pro^™, Biodex Medical Systems, USA) as described above with their right foot secured to the ankle attachment plate (Figure 6a). Headphones rated at 120 dB SPL (U UFO Over-Ear Headphones, Bluedio, China) were connected to an amplifier (AV Surround Receiver AVR-S510BT, Denon, Japan) and used to deliver auditory cues. Upon receiving a ‘go’ cue, participants were instructed to perform an isometric contraction by dorsiflexing their right ankle as quickly as possible. The cue types included a visual cue, a visual + auditory cue (80 dB, 500 Hz, 50 ms), and a visual + startle cue (120 dB, 500 Hz, 50 ms) (Sangari and Perez, 2020). The contraction target was set to 30% of their MVC (Figure 6a). Ten trials of each cue type were performed in randomized order for a total of 30 trials per block, with 1–3 second rest periods (randomized time) between cues.

Figure 6: Reticulospinal tract excitability assessed via the StartReact paradigm remains unchanged following motor skill and resistance training.

(a) StartReact paradigm to assess RST excitability. Participants sat in a Biodex chair and performed a rapid isometric dorsiflexion reaction task in response to visual and auditory cues. Real-time torque biofeedback was provided during testing to guide performance. (b). Schematic of the auditory startle pathway. The StartReact response assesses RST output gain non-invasively. Auditory signals from the cochlear nucleus activate giant neurons in the pontomedullary reticular formation, which then transmit signals via the RST to motoneurons for rapid motor responses. (c) Representative StartReact trial and RST gain calculation for representative participant ASR001. Torque (top) and tibialis anterior EMG profiles (bottom) show responses to each cue type: visual, visual + auditory (80 dB), and visual + startling stimulus (120 dB). The target contraction was set to 30% of MVC. (d) Individual reaction times from the trial shown in (c), plotted for each cue condition. Reaction times and variability decrease in response to the startling auditory cue, reflecting a startle response. (e, f) Group-level data are presented as mean values with error bars representing the standard deviation. (e) Reticulospinal tract gain. Group-level RST gain values before and after motor skill and resistance training. ‘X’ marker below plots indicates Bonferroni-corrected significant contribution of the RST at each timepoint based on one-sample tests against μ = 1 (‘X’ p < 0.05, ‘XX’ p < 0.01, ‘n.s.’ p > 0.05). (f) Training-induced percent changes in RST gain. Neither training protocol measurably altered the influence of the reticulospinal tract on the rapid initiation of a prepared movement in response to a startling stimulus.

2.8. Motor learning analysis

Motor learning was assessed by comparing reaching performance in center-out tasks using several performance metrics: path length, movement time, jerk, and endpoint error (Pierella et al., 2017). Path length refers to the total distance traveled by the cursor during the center-out reaching tasks, normalized to the minimal straight-line distance required to reach the target. Movement time is the duration from the initiation of movement at the center target until the final target is reached. Dimensional jerk, which quantifies movement smoothness, is calculated from the third derivative of position (Hogan and Sternad, 2009). Endpoint error is defined as the Euclidean distance between the final cursor position and the target location after four seconds. Performance metrics were averaged across 24 movements per block after outlier detection and removal using the interquartile range (IQR) method, defined as values exceeding 1.5 times the IQR above the 75th percentile or below the 25th percentile (Kaltenbach, 2011; Steele et al., 2021). Motor learning was quantified by comparing improvements in movement performance from the first to the last block after the 30-min training session.

2.9. Torque and muscle capacity before and after training

Participants were instructed to perform three repetitions of maximal voluntary contraction to assess muscle capacity before and after training. The EMG signal was first full-wave rectified by removing the mean and taking the absolute value. The root mean square (RMS) of the rectified signal was then calculated using a sliding window of 100 ms. The onset and offset of muscle activation were established by detecting instances when the RMS amplitude surpassed a manually set threshold, typically around 0.02 mV. Manual threshold selection was incorporated into the data analysis to address baseline signal variability, ensuring the robustness of burst identification for all individuals. The area under the curve (AUC) for each burst was computed using the trapezoidal rule, applied to the RMS signal between the identified onset to 2.25 seconds later (Figure 3c). The muscle capacity was quantified as the average AUC across the three MVC bursts recorded before and after the training.

2.10. Corticospinal tract excitability analysis

Corticospinal tract excitability was evaluated by examining changes in MEP amplitude from baseline to post-training (Griffin and Cafarelli, 2007; Latella et al., 2017, 2016; Leung et al., 2017; Perez et al., 2004; Wilson et al., 2023). For the right tibialis anterior, MEP peak-to-peak amplitudes were normalized to the maximal M-wave amplitude (M_max) recorded from peripheral nerve stimulation at the corresponding time point (Alibazi et al., 2022; Mason et al., 2019; Woodhead et al., 2024) (pre- or post-training), with responses expressed as a percentage of M_max (Latella et al., 2017; Perez et al., 2004). Following normalization, values that exceeded 100% were excluded, as M_max represents the excitation of the entire motoneuron pool (Palmieri et al., 2002), making MEPs (% of M_max) larger than 100% unreasonable and indicative of a potential recording error. Outliers were identified and removed using the interquartile range (IQR) (Kaltenbach, 2011; Latella et al., 2017; Quintero et al., 2024). For non-target muscles, peak-to-peak amplitudes were similarly evaluated, but normalization was performed within each subject and training day relative to the maximum MEP recorded before training. This approach allowed normalized values to exceed 100% if post-training MEPs were larger than baseline, reflecting facilitation rather than error. Outliers in this analysis were again identified and removed using the IQR method (Kaltenbach, 2011).

2.11. Synaptic and spinal motor neuron excitability

The H/M ratio was used to assess synaptic spinal excitability before and after training (Lagerquist et al., 2006; Scaglioni et al., 2002). This ratio serves as an estimate of the excitability of the spinal reflex pathway by evaluating the activation of the motor neuron pool via the Ia afferents, in relation to the maximum capacity of the entire motor neuron pool (Dishman and Burke, 2003; Misiaszek, 2003; Palmieri et al., 2002). The peak-to-peak amplitudes of the M-wave and H-reflex responses were averaged across the three repetitions at each stimulation level. This process created a recruitment curve for both the M-wave and H-reflex (Figure 5b). H_max and M_max values were identified from this curve and were used to calculate the H/M ratio. To assess training effects, the percent change in the H/M ratio from pre- to post-training was computed for each participant.

F-wave amplitude and persistence were used to assess spinal motor neuron excitability (Kumru et al., 2021; Milanov, 1992; Weber, 1998). EMGs were filtered using a second order Bessel high-pass filter (200 Hz) to improve the clarity of F-wave evaluations (Eisner-Janowicz et al., 2023). An F-wave response was deemed valid if it occurred at least 30 ms after the stimulus and had a peak-to-peak amplitude greater than or equal to 20 μV (Eisner-Janowicz et al., 2023). F-wave persistence was calculated as the percentage of stimuli out of 60 pulses that elicited a valid response. To assess training effects, the percent change in average amplitude and persistence from pre- to post-training was computed for each participant (Kumru et al., 2021).

2.12. Reticulospinal tract excitability analysis

Reticulospinal tract (RST) excitability was evaluated by comparing the StartReact response before and after training (Sangari and Perez, 2020). The StartReact response, or RST gain, is calculated as follows (Figure 6c):

$$\mathit{RST}\mathit{Gain}=\frac{{\mathrm{t}}_{\mathit{visual}}-{\mathrm{t}}_{\mathit{startle}}}{{\mathrm{t}}_{\mathit{visual}}-{\mathrm{t}}_{\mathit{auditory}}}$$

The ratio is the difference in reaction time between the startling (t_startle) and visual cues (t_visual) divided by the difference in reaction time between the auditory (t_auditory) and visual cues (Baker and Perez, 2017; Fisher et al., 2013). Reaction time was defined as the time between cue onset and movement onset. Movement onset was identified from EMG traces (Figure 6c) offline using a Python library for change point detection (ruptures v1.1.9), specifically employing a binary segmentation search method (model = “l2”, Binseg method, predict 8 changepoints) (Truong et al., 2020). Reaction times less than 75 ms were excluded as false starts, and reaction times greater than 700 ms were discarded as indicating a loss of participant focus (Sangari and Perez, 2020). Additionally, reaction times identified as outliers based on the IQR were excluded from analysis (Kaltenbach, 2011). To assess training effects, the percent change in RST Gain from pre- to post-training was computed for each participant.

2.13. Statistics

Statistical analyses were performed using Python (v3.11) using the SciPy and StatsModels libraries. To assess the normality of all data sets, the Shapiro-Wilk test was applied. Given the non-normal distribution of several datasets, both parametric and non-parametric statistical tests were used, with significance set at p < 0.05. Group data are reported as mean ± standard deviation (SD), unless stated otherwise.

To test for motor learning in the BoMI, paired t-tests (for normally distributed data) and Wilcoxon signed-rank tests (for non-normally distributed data) between the first and last blocks of training were conducted under the null hypothesis that there would be no significant reduction in jerk, movement time, path length, or end-point error, following motor skill training (i.e. no motor learning occurred). To assess neuromuscular fatigue, paired t-tests (for normally distributed data) and Wilcoxon signed-rank tests (for non-normally distributed data) were used to compare pre- and post-training MVC EMG amplitudes. The null hypothesis was that there would be no difference in EMG amplitude before and after training (i.e. neither training would induce a reduction in EMG amplitude). These tests also assessed whether the degree of fatigue, indicated by changes in EMG amplitude, differed between the two training modalities, with the null hypothesis stating no difference in training effects.

Prior to statistical analysis, distribution of both raw MEP amplitudes (mV and % of M_max) and log-transformed MEP data (Nielsen, 1996) was validated using Shapiro–Wilk tests, which confirmed non-normality in all cases for the right tibialis anterior. Given that MEP amplitudes are continuous, strictly positive, and right-skewed, we implemented Generalized Linear Mixed Models (GLMMs) with a Gamma distribution and log link function in R (Sasaki et al., 2024). Subject-specific random intercepts were included to account for repeated measures within participants. MEPs recorded during muscle activation and at rest represent physiologically distinct states and exhibited clearly bimodal distributions, we analyzed these conditions in two separate GLMMs. This approach avoids conflating distinct neurophysiological responses and, by modeling trial-level data rather than subject averages, preserves within-subject variability and maximizes statistical power. For each activation condition (rest and active), the GLMM examined the effects of training condition (motor skill training vs. isometric resistance training), timepoint (pre- vs. post-intervention), interaction, and motor threshold (MT) intensity on raw MEP amplitudes (% of M_max). Training condition and timepoint were included as fixed categorical factors, and MT intensity was treated as a fixed continuous variable. If the GLMM indicated a significant increase in excitability due to a protocol, post-hoc comparisons before vs. after training were performed using the ‘emmeans’ package in R, applying Holm’s adjustment for multiple comparisons to control family-wise error. This approach allowed testing whether MEP amplitudes significantly changed following each training paradigm.

For non-target muscle MEPs, the same analytic pipeline was applied. In these models, muscle was additionally included as a fixed effect to account for inter-muscle variability, while training condition, timepoint, and MT intensity remained as predictors. This allowed us to evaluate training-induced changes across multiple muscles simultaneously while maintaining the same trial-level GLMM framework and post-hoc procedures.

All statistical analyses for reticulospinal, synaptic spinal, and spinal motor neuron excitability followed a standardized workflow. Data were first assessed for normality using the Shapiro–Wilk test. Depending on normality, either paired t-tests or Wilcoxon signed-rank tests (with Bonferroni correction for multiple comparisons) were used to compare measurements between the two training conditions. These tests evaluated whether one training modality elicited significantly greater or lesser changes in excitability relative to the other, with the null hypothesis being that there was no difference between training effects. Additionally, one-sample t-tests or Wilcoxon signed-rank tests were applied to assess within-condition percent changes from baseline. For these, the null hypothesis was that the mean percent change was equal to zero, indicating no change in excitability following training. Corrected effect sizes were calculated to complement inferential statistics with Hedge’s g to account for the small sample size bias (Lakens, 2013).

3. Results

3.1. Improvement in cursor control suggests learning of a novel motor task

A novel BoMI was used to perform leg motor skill training, requiring participants to learn precise, coordinated movements to control a cursor in a visuomotor task (Figure 2a). Representative cursor positions and velocity profiles from the first and last blocks showed improvements in body coordination and cursor control over the training session (Figure 2b). Straighter cursor trajectories and smoother velocity profiles, with fewer peaks and shorter durations, characterize these improvements. Quantitative analysis of cursor control metrics revealed significant reductions in jerk (Figure 2c: −16.66% ± 83.98 [mean ± SD], t(22) = 2.84, p = 0.010, Hedge’s g(22) = −0.71, 95% CI [0.25, 1.22]), movement time (Figure 2d: −18.21% ± 38.20, W = 37.00, p = 0.001, Hedge’s g(22) = −0.80, 95% CI [0.37, 1.18]), and path length (Figure 2e: −35.58% ± 39.82, W = 24.00, p < 0.001, Hedge’s g(22) = −0.92, 95% CI [0.61, 1.23]) following five blocks of motor skill training. These findings suggest that participants exhibited smoother, faster, and more coordinated movements during the session. Although the end-point error (distance of the cursor from the target at the 4-second mark) reduced slightly, the change was not statistically significant (−4.40% ± 28.73, W = 99.00, p = 0.247, Hedge’s g(22) = −0.41, 95% CI [−0.14, 0.82]) (Figure 2f). This result is expected, as the task instructions emphasized temporal precision over spatial accuracy, thereby limiting the potential for improvements in end-point errors. Together, these findings highlight the effectiveness of the motor skill training protocol in promoting motor learning of a novel task through BoMI practice (Casadio et al., 2010; Pierella et al., 2017; SeáñezGonzález et al., 2016).

3.2. Comparable neuromuscular fatigue indicates similar physical demands across protocols

In the 30-minute isometric resistance protocol, participants executed twelve slow-ramped isometric contractions of dorsiflexion and plantar flexion at 30% of their maximum voluntary contraction (MVC), organized into three blocks (Figure 3a). Torque traces from both the initial and final blocks exhibited stable performance throughout the training session (Figure 3b). To evaluate the neuromuscular fatigue and effort levels exerted by participants during both training types, MVCs were measured before and after training. Examples of MVCs for the tibialis anterior are illustrated in Figure 3c. Group analysis revealed a significant reduction in EMG amplitude following training for both motor skill (−12.97% ± 17.21, W = 31.00, p = 0.001, Hedge’s g(22) = −0.28, 95% CI [0.14, 0.42]) and isometric resistance training (−6.29% ± 14.39, W = 60.00, p = 0.033, Hedge’s g(22) = −0.15, 95% CI [0.01, 0.24]) (Figure 3d). EMG amplitude was comparable across pre-training timepoints (t(24) = −0.17, p = 1.000, Hedge’s g(24) = 0.03, 95% CI [−0.25, 0.38]) and across post-training (t(23) = −1.14, p = 0.270, Hedge’s g(23) = 0.17, 95% CI [−0.12, 0.47]). Reductions in EMG amplitude were also comparable across training modalities (W = 90.00, p = 0.453, Hedge’s g(22) = −0.41, 95% CI [−0.89, 0.07]) (Figure 3e), suggesting similar levels of neuromuscular fatigue.

3.3. Motor skill training leads to higher increases in corticospinal tract excitability compared to isometric resistance training

Training-induced changes in corticospinal tract excitability were assessed by comparing MEP amplitudes at rest and during 15% maximal voluntary EMG activation of the tibialis anterior (Figure 4e). Evaluation of MEPs at rest and during pre-activation provides complementary information, as rest captures baseline corticospinal tract excitability, while voluntary measures reflect pathway function during movement, albeit influenced by spinal excitability (Siddique et al., 2020; Spampinato et al., 2023). For each activation condition (rest and pre-activation), an individual generalized linear mixed model (GLMM) captured the effects of training condition (motor skill [1] vs. isometric resistance [2]), timepoint (pre- [1] vs. post- [2] intervention), their interaction, and TMS intensity [100%, 150%, 180%, 200% of resting motor threshold] on MEP amplitudes (% of maximal M-wave response or M_max). At rest, MEP amplitude increased significantly after training (Timepoint 1 effect: β = 0.16, SE = 0.04, z = 4.36, p < 0.001), with a significantly smaller increase following resistance training compared to motor skill training (Training 2 × Timepoint 1 interaction: β = −0.11, SE = 0.05, z = −2.17, p = 0.03, see Supplementary Table S2 for details). During pre-activation, baseline MEP amplitudes were significantly higher in both training conditions (Training 1: β = 3.42, SE = 0.06, z = 56.31, p < 0.001; Training 2: β = 3.44, SE = 0.06, z = 56.54, p < 0.001). At pre-activation, MEP amplitude also increased significantly after training (Timepoint 1 effect: β = 0.07, SE = 0.02, z = 3.66, p < 0.001), with a smaller increase following resistance training relative to motor skill training (Training 2 × Timepoint 1 interaction: β = −0.07, SE = 0.03, z = −2.32, p = 0.02, see Supplementary Table S3 for details). The model results indicate that MEP amplitude significantly increased after training overall, but this effect was driven primarily by motor skill training, which produced greater increases than isometric resistance training. Figure 4f shows the estimated marginal means (EMM) of MEP response amplitudes for motor skill and resistance training during rest and pre-activation (15% dorsiflexion) evaluations.

Post-hoc pairwise comparisons of estimated marginal means, with Holm adjustment, were conducted to test significant differences in MEP amplitude between pre- and post-intervention within each training condition (Figure 4f). These analyses revealed that motor skill training elicited a significant increase in MEP amplitude from before to after training during both rest (ratio = 0.85, SE = 0.03, 95% CI [0.79, 0.92], z = −4.36, p < 0.001) and pre-activation (ratio = 0.93, SE = 0.02, 95% CI [0.89, 0.97], z = −3.66, p < 0.001). In contrast, isometric resistance training did not produce a significant increase in MEP amplitude at either rest (ratio = 0.95, SE = 0.03, 95% CI [0.89, 1.02], z = −1.24, p = 0.18) or pre-activation (ratio = 0.99, SE = 0.02, 95% CI [0.95, 1.03], z = −0.32, p = 0.75). These results confirm that motor skill training leads to a robust facilitation of corticospinal excitability, while resistance training results in minimal change. Moreover, our results indicate that while a dorsiflexion pre-activation level of 15% can indeed increase MEP amplitudes in the tibialis anterior, changes in MEP amplitude due to training were less pronounced than when evaluated at rest.

Given that peripheral adaptations (seen in Figure 5d) may alter MEPs, we conducted an analysis in which MEP amplitudes were normalized to the M_max recorded before training, enabling comparison with our primary analysis that used the timepoint-specific M_max (Supplementary Figure S2; Supplementary Tables S4–S5). Importantly, post-hoc Holm-adjusted pairwise comparisons of timepoint effects within each training condition revealed a similar pattern. For the at-rest condition, MEP amplitudes remained significantly higher after both motor skill training (ratio = 0.89, SE = 0.03, 95% CI [0.82, 0.95], z = −3.31, p = 0.002) and resistance training (ratio = 0.91, SE = 0.03, 95% CI [0.85, 0.98], z = −2.54, p = 0.011). In the pre-activation condition, no significant change was observed following motor skill training (ratio = 0.96, SE = 0.02, 95% CI [0.93, 1.00], z = −1.78, p = 0.085) or resistance training (ratio = 0.96, SE = 0.02, 95% CI [0.92, 1.00], z = −2.02, p = 0.085). Compared with the primary analysis (Figure 4f), which showed robust MEP facilitation after motor skill training and no significant effects of resistance training, these findings indicate that attenuating effects of motor skill training specifically under pre-activation, while at the same time revealing significant increases after resistance training in only the rest condition and not the pre-activation condition.

To complement these findings, we conducted an additional analysis of MEPs recorded from non-target lower limb muscles using GLMMs that incorporated muscle as a fixed effect (Supplementary Table S6 for at-rest and Supplementary Table S7 for pre-activation results). This analysis revealed distinct differences between the trained (right) and untrained (left) limbs. In the left leg (Supplementary Figure S1b), significant increases in MEP amplitude at rest were observed in the tibialis anterior after both motor skill and resistance training, and in the vastus lateralis after resistance training. During pre-activation, decreases emerged in the medial gastrocnemius, soleus, and vastus lateralis following motor skill training, while resistance training produced increases in the medial gastrocnemius and vastus lateralis. In the right leg (Supplementary Figure S1c), motor skill training facilitated the medial gastrocnemius at rest, but led to reductions during pre-activation in the medial gastrocnemius, rectus femoris, and soleus. Resistance training also reduced MEP amplitude in the right rectus femoris during pre-activation. As all training was performed with the right ankle, these left–right asymmetries suggest that compensatory mechanisms during training may have contributed to the observed bidirectional changes in excitability. Notably, the right medial gastrocnemius, the primary plantar flexor engaged during training, still showed increased MEPs after motor skill training at rest, consistent with the facilitation observed in the right tibialis anterior.

3.4. The majority of spinal motoneuron and monosynaptic reflex excitability measures do not show training-induced changes

The corticospinal tract contains descending projections from the motor cortex to motoneurons and interneurons in the spinal cord (Jankowska and Edgley, 2006; Lemon, 2008). Therefore, increases in corticospinal tract excitability could be due to either an increase in the excitability of the motor cortex and its descending projections or increases in the excitability of the motoneurons projecting to the muscles and their associated synapses and interneurons. We sought to investigate whether increases in corticospinal tract excitability could be partially accounted for by increases in excitability of spinal circuits.

We used the Hoffman reflex (H-reflex), M-waves, and F-waves elicited via peripheral nerve stimulation of the mixed tibial nerve to compare the excitability of spinal neural circuits before and after training (Mesrati and Vecchierini, 2004; Milanov, 1992; Palmieri et al., 2002; Pierrot-Deseilligny and Mazevet, 2000) (Figure 5a). The H/M ratio is commonly used to assess monosynaptic reflex excitability (Figure 5b) (Lagerquist et al., 2006; Scaglioni et al., 2002). One-sample tests revealed no significant changes in the H/M ratio of the tibialis anterior after motor skill training (25.41% ± 47.35, t(10) = 1.78, p = 0.316, Hedge’s g(10) = 0.54, 95% CI [−0.06, 1.30]) or isometric resistance training (−3.72% ± 19.95, t(10) = −0.62, p = 1.000, Hedge’s g(10) = −0.19, 95% CI [−0.90, 0.42]) (Figure 5c). Moreover, training-induced changes in H/M ratio were comparable across motor skill and isometric resistance training (t(10) = 1.85, p = 0.282, Hedge’s g(10) = −0.77, 95% CI [−1.65, 0.08]). These findings suggest that neither motor skill training nor isometric resistance training significantly increases the excitability of the tibialis anterior monosynaptic reflex.

We next sought to investigate training-induced changes in the maximum M-wave and H-reflex amplitudes (M_max and H_max, respectively) of the tibialis anterior. M_max reflects the activation of muscle fibers via motor efferent axons and acts as a reference for normalizing reflex measures, yet it can either be potentiated or inhibited by exercise (Hicks et al., 1989; Latella et al., 2017; Palmieri et al., 2002). One-sample tests revealed a significant decrease in M_max after motor skill training (−5.67% ± 8.84, W = 7.00, p = 0.028, Hedge’s g(11) = −0.64, 95% CI [−0.96, −0.23]), whereas isometric resistance training produced a significant increase in M_max (5.49% ± 4.61, t(11) = 4.12, p = 0.005, Hedge’s g(11) = 1.19, 95% CI [0.61, 2.04]). Paired comparisons further demonstrated that training-induced changes in M_max were significantly larger following isometric resistance training compared with motor skill training (W = 1.00, p = 0.003, Hedge’s g(11) = 1.53, 95% CI [0.96, 2.12]) (Figure 5d). H_max is an indirect measure of the percentage of motor neurons that can be activated trans-synaptically (Dishman and Burke, 2003; Misiaszek, 2003; Palmieri et al., 2002), although it can be influenced by factors such as presynaptic inhibition, afferent activity, and post-activation depression (Misiaszek, 2003; Palmieri et al., 2002). As with the H/M ratio, one-sample tests revealed that H_max did not significantly change after motor skill training (16.31% ± 41.85, t(10)= 1.29, p = 0.676, Hedge’s g(10) = 0.39, 95% CI [−0.23, 1.10]) or isometric resistance training (1.20% ± 19.51, t(10) = 0.20, p = 1.000, Hedge’s g(10) = 0.06, 95% CI [−0.63, 0.73]). Training-induced changes in H_max were also consistent across training paradigms (t(10) = 1.08, p = 0.913, Hedge’s g(10) = −0.45, 95% CI [−1.27, 0.43]) (Figure 5e).Together, our results suggest that while efferent fiber recruitment increases after isometric resistance training and decreases after motor skill training, this is not enough to impact the H/M ratio.

We used F-wave amplitude and persistence to compare changes in motoneuron excitability due to training (Kumru et al., 2021; Milanov, 1992; Weber, 1998). F-waves are recurrent discharges of antidromically activated motoneurons, integrating both segmental and suprasegmental influences (Milanov, 1992). They appear at supramaximal stimulation amplitudes with variable latencies and amplitudes across repeated stimuli (Milanov, 1992) (Figure 5f). Individual one-sample tests revealed that F-wave persistence did not significantly change after motor skill training (5.39% ± 49.40, t(10) = 0.36, p = 1.000, Hedge’s g(10) = 0.11, 95% CI [−0.58, 0.70]) or isometric resistance training (−8.28% ± 55.32, t(10) = 0.50, p = 1.000, Hedge’s g(10) = 0.15, 95% CI [−0.71, 0.75]). Training-induced changes in F-wave persistence were also comparable across training paradigms (t(10) = −0.11, p = 1.000, Hedge’s g(10) = 0.05, 95% CI [−0.95, 1.02]). (Figure 5g). F-wave amplitude showed no significant changes after motor skill training (12.11% ± 29.90, t(10) = 1.34p = 0.627, Hedge’s g(10) = 0.41, 95% CI [−0.24, 1.00]) or isometric resistance training (−12.06% ± 28.87, t(10) = −1.39, p = 0.588, Hedge’s g(10) = −0.42, 95% CI [−1.29, 0.22]). Training-induced changes in F-wave amplitude were also comparable across motor skill and isometric resistance training (t(10) = 1.73, p = 0.346, Hedge’s g(10) = −0.79, 95% CI [−1.71, 0.08]) (Figure 5h). As a secondary analysis, we examined training-induced changes in the medial gastrocnemius, given that both training paradigms engaged the plantar flexors and found no significant changes in all metrics of spinal and motoneuron excitability (Supplementary Figure S3).

3.5. Reticulospinal tract excitability remains consistent after motor skill and isometric resistance training

We used the StartReact response to evaluate reticulospinal tract excitability before and after training by comparing reaction times in response to visual, auditory, and startling stimuli (Figure 6a,c) (Sangari and Perez, 2020). The StartReact response refers to an accelerated release of a prepared motor action that is initiated by a startling stimulus (Honeycutt et al., 2013; Valls-Solé et al., 2008), and it has been used as a non-invasive method to assess the output gain of the reticulospinal tract, which operates through the auditory startle reflex (Figure 6b) (Fisher et al., 2013; Honeycutt et al., 2013). Specifically, the reticulospinal tract gain (RST) is calculated by dividing the difference in reaction time between the startling stimulus and the visual stimulus by the difference in reaction time between the auditory stimulus and the visual stimulus (Figure 6f) (Baker and Perez, 2017; Fisher et al., 2013). We found a significant contribution of the reticulospinal tract for both motor skill and isometric resistance training. For motor skill training (Figure 6e), pre-training RST gain was significantly greater than 1 (1.69 ± 1.08, W = 7.00, p = 0.009, Hedge’s g(13) = 0.64, 95% CI [0.40, 0.89]), and post-training values remained elevated (1.21 ± 0.31, W = 12.00, p = 0.034, Hedge’s g(13) = 0.69, 95% CI [0.35, 0.99]). Similarly, isometric resistance training demonstrated significant RST contributions both before training (1.29 ± 0.33, t(11) = 3.08, p = 0.042, Hedge’s g(11) = 0.89, 95% CI [0.43, 1.36]) and after training (1.28 ± 0.28, t(11) = 3.51, p = 0.020, Hedge’s g(11) = 1.01, 95% CI [0.63, 1.44]). However, one-sample tests showed no significant changes in RST gain after motor skill training (Figure 6f: −13.09% ± 30.00, t(11) = −1.51, p = 0.477, Hedge’s g(11) = −0.44, 95% CI [−0.97, 0.12]) or isometric resistance training (4.52% ± 36.60, W = 32.00, p = 1.000, Hedge’s g(11) = 0.12, 95% CI [−0.84, 0.56]). Furthermore, paired comparisons demonstrated that training-induced changes in RST gain were consistent across training types (t(11) = −1.31, p = 0.648, Hedge’s g(11) = −0.51, 95% CI [−1.12, 0.27]) (Figure 6f). These results suggest that while there was a significant contribution from the reticulospinal tract for each task, the strength of this contribution was not impacted by training.

4. Discussion

4.1. Motor, but not resistance, training increased corticospinal tract excitability

Our findings demonstrate distinct short-term facilitations in corticospinal tract excitability following single sessions of cue-paced motor skill training. Consistent with our hypothesis, motor skill training enhanced corticospinal excitability, whereas resistance training did not. This distinction aligns with Krakauer’s operational definition of motor skill learning, which emphasizes acquisition and maintenance processes further divided into de novo learning, motor acuity, and cognitive involvement (Krakauer et al., 2019). Our short-term motor skill training task fits well within these definitions, as the BoMI maps high-dimensional body movements to a novel two-dimensional cursor control space (Wang et al., 2014). In this context, participants construct an inverse internal model between body movements and cursor motion (Casadio et al., 2010), assembling a new skill de novo rather than adapting an existing skill to perform a new task. Through five blocks of training, participants demonstrate improvements in accuracy, precision, and smoothness – hallmarks of motor acuity. In line with Krakauer, task performance depends on participants’ cognitive strategies (Krakauer et al., 2019); in BoMI studies, including ours, participants reorganize their body motions to achieve the desired control (Casadio et al., 2010; Pierella et al., 2017). While our single-session design does not address skill maintenance, BoMI training demonstrates both de novo skill construction and refinement of movement execution quality through implicit and explicit mechanisms (Casadio et al., 2010; Pierella et al., 2017; SeáñezGonzález and Mussa-Ivaldi, 2014).

Our study employed a low-to-moderate intensity (~30% MVC) bi-directional isometric contraction as the resistance training paradigm, distinct from conventional high-intensity resistance training protocols typically conducted at ~80% MVC (Latella et al., 2018; Leung et al., 2017). This choice was deliberate to better balance the physical demands of the resistance task with the 30 min training duration and to accommodate the functional capabilities of future clinical populations, for whom sustained high-intensity contractions for long periods may not be feasible. Studies investigating similar low-intensity paradigms have employed protocols such as 30% maximal knee extensor strength training, plantar-flexor contractions at approximately 20% MVC, and light-load strength training at 20% of 1 repetition maximum, all reflecting comparable levels of muscle activation (Alibazi et al., 2022; Lagerquist et al., 2012; Mason et al., 2019).

Critically, while prior evidence has emphasized the vital role of external pacing, particularly auditory cues, in facilitating corticospinal excitability (Leung et al., 2017; Mason et al., 2019), our study did not observe a significant increase in corticospinal excitability following resistance training paired with an auditory cue. Leung et al. (2017) demonstrated that metronome-paced strength training at high intensity significantly increased corticospinal excitability, whereas self-paced training at the same intensity did not, underscoring the contribution of sensory feedback and task intensity. Mason et al. (2019) further showed that externally paced, light-load strength training elicited increases in corticospinal excitability comparable to heavy-load training and visuomotor skill tasks, suggesting that under certain conditions, auditory pacing can substitute for load or complexity. Conversely, studies lacking external pacing or sensory feedback reported no changes in corticospinal excitability at similar intensities (Alibazi et al., 2022; Lagerquist et al., 2012; Schmidt et al., 2011).

Electrical stimulation can also serve as effective sensory feedback to induce cortical changes with low-level intensity training. (Schmidt et al., 2011) found that a motor-only task at 35% maximal effort did not increase corticospinal excitability, while their motor-and-stimulation task did. However, the influence of electrical stimulation on corticospinal excitability could potentially be muscle-specific; (Lagerquist et al., 2006) reported increases in spinal but not corticospinal excitability for plantar-flexors after neuromuscular electrical stimulation combined with low-intensity voluntary contractions. These findings highlight how differences in muscle group targeted, contraction intensity, and sensory feedback modality can influence corticospinal excitability outcomes. Our findings suggest that low-to-moderate intensity resistance training, even when paired with auditory pacing, may be insufficient to engage corticospinal pathways without the added challenge of greater task complexity.

4.2. Intracortical inhibitory circuit modulation as a potential mechanism underlying skill-dependent corticospinal tract facilitation

Although intracortical mechanisms were not directly assessed in this study, the selective increase in corticospinal excitability following ankle motor skill training, paired with the absence of change after low-intensity resistance training, suggests that these two forms of practice likely engage distinct intracortical processes within M1. Prior work using similar ankle-based skill paradigms (Perez et al., 2004; Tatemoto et al., 2019) showed that short-term motor learning is accompanied by a reduction in short-interval intracortical inhibition (SICI), reflecting decreased responsiveness of GABA_A-mediated inhibitory interneurons immediately after training (Gómez-Feria et al., 2023; Ho et al., 2022). This release of inhibition is associated with increases in MEP amplitude and is thought to enable a greater net excitatory drive to corticospinal neurons during early skill acquisition (Ho et al., 2022; Perez et al., 2004; Rosenkranz et al., 2007). Notably, intracortical facilitation (ICF) appears largely unchanged in these leg-area motor learning tasks, suggesting that the early phase of skill acquisition primarily modulates inhibitory rather than facilitatory circuits in M1 (Latella et al., 2017; Perez et al., 2004; Rosenkranz et al., 2007).

In contrast, the absence of corticospinal facilitation following our low-intensity resistance protocol is consistent with evidence that intracortical plasticity is sensitive to task demands and intensity thresholds (Alibazi et al., 2022; Leung et al., 2017; Mason et al., 2019). This lack of change in inhibitory circuitry likely explains why our resistance training paradigm did not elicit measurable corticospinal tract facilitation, suggesting the stimulus was insufficient to reduce the inhibitory synaptic efficacy between intracortical interneurons and corticospinal neurons (Mason et al., 2019). Together, these studies support the interpretation that motor skill training engages inhibitory circuit reorganization within M1, whereas low-intensity resistance training is insufficient to drive comparable intracortical modulation.

4.3. Spinal excitability and peripheral adaptations

The H-reflex response, our primary measure for spinal excitability, and F-wave characteristics, considered a secondary outcome reflecting motoneuron excitability, both exhibited no significant alterations following training (Figure 5c–e, g, h). This aligns with previous findings suggesting that significant spinal plasticity typically requires sustained, high-intensity, or repetitive training paradigms (Kumru et al., 2021; Latella et al., 2017; Palmieri et al., 2002). We observed an increase in the M_max following isometric resistance training, suggesting that this training paradigm potentially influenced peripheral neuromuscular properties, such as increased motoneuron recruitment, or increased muscle fiber conduction velocity allowing for more efficient and synchronous summation processes (Lentz and Nielsen, 2002; Rodriguez-Falces et al., 2015). In contrast, the maximal M-wave was significantly decreased after the motor skill training, potentially reflecting reduced muscle membrane excitability or impaired neuromuscular propagation (Behm and St-Pierre, 1997; Hicks et al., 1989; Lentz and Nielsen, 2002; Rodriguez-Falces et al., 2015). M-wave responses appear muscle- and intensity-dependent, as prior ankle-based motor skill training showed no change in tibialis anterior M_max, whereas high-intensity resistance protocols targeting the leg extensors produced acute M-wave depression following training (Latella et al., 2017). As illustrated in Supplementary Figure S2, when MEPs are normalized only to the pre-training M_max (removing the post-training change), increased MEP amplitude following motor skill training persisted. Although M_max decreased after motor skill training, no spinal reflex measures changed, suggesting that while some peripheral contributions may be present, the predominant effect reflects central adaptations. In contrast, resistance training did not initially show increased MEPs; however, MEP facilitation emerged once post-training M_max increases were removed, suggesting resistance training may induce similar adaptations at both peripheral and central levels (Latella et al., 2017; Siddique et al., 2020). Additionally, secondary analyses of the medial gastrocnemius, evaluated using the same peripheral nerve stimulation metrics as the tibialis anterior, revealed comparable patterns (see Supplementary Figure S3). Specifically, there were no significant changes in H_max or the H/M ratio following either training paradigm. M_max responses did not differ between training groups nor change in response to either training, unlike the tibialis anterior. There were no F-wave amplitude or persistence changes observed in the medial gastrocnemius, consistent with findings in the target dorsiflexor muscle.

These observations echo previous findings reporting increases in corticospinal tract excitability following 4 weeks of visuomotor skill learning but not heavy strength training (Jensen et al., 2005). Similarly to our findings, no substantial short-term changes in spinal excitability were observed, suggesting central nervous system plasticity primarily occurred at higher cortical rather than spinal levels after motor skill training. However, Jensen et al. reported a decrease in corticospinal excitability after several weeks of strength training, whereas in our study a single session of resistance training produced no significant change. One potential reason for this discrepancy is that Jensen et al. targeted upper limb muscles (biceps brachii), whereas our study focused on lower limb muscles (tibialis anterior). Therefore, differential involvement of neural tracts (corticospinal vs. reticulospinal) due to anatomical and functional distinctions between upper and lower limb movements, or training time, could underline differences in neural excitability outcomes.

4.4. Variability and cross-muscle effects

Our observation of corticospinal excitability changes after the multidirectional movements in our training paradigms implies that the adaptations are occurring in a manner relevant to the functional demands of the task. Importantly, even non-target muscles showed significant changes after training, suggesting compensatory or synergistic adaptations (Supplementary Figure S2); similar cross-muscle effects have been reported previously, with (Thomas and Gorassini, 2005) showing training-induced increases in antagonist muscles including the soleus and hamstrings despite training targeting the tibialis anterior and vastus lateralis. Bidirectional changes observed in some participants likely reflect differences in movement strategies during the motor skill task, such as greater reliance on plantar flexors versus dorsiflexors. Additionally, MEP measures exhibit inherent variability, stemming from biological factors (e.g., genetics, neurophysiological state, prior training) (Humphry et al., 2004; Latella et al., 2018; Spampinato et al., 2023), methodological influences (e.g., TMS coil placement, stimulus parameters, background EMG) (Grill and Mortimer, 1996; Lagerquist et al., 2012; Siddique et al., 2020; Spampinato et al., 2023), and muscle- or task-specific properties. While simplifying tasks to unidirectional or isolated muscle contractions can clarify neurophysiological mechanisms, it may reduce ecological validity and limit relevance for populations with neuromotor impairments. Rehabilitation programs commonly focus on restoring coordinated movements essential for daily living—such as transferring, standing, and walking—and often involve training that requires the integration of multiple muscle groups and neural pathways (Franz et al., 2018). These complex movements involve overlapping excitatory and inhibitory influences on the corticospinal tract, with motor evoked potentials reflecting that net balance (Siddique et al., 2020). Evidence shows that externally paced, skill-based training paradigms are the key driver of corticospinal excitability increases and intracortical inhibition reductions (Leung et al., 2017; Mason et al., 2019). Thus, it is likely the pacing and complexity of the task, more than movement directionality, that facilitate neural excitability changes potentially relevant to functional recovery (Christiansen et al., 2020).

4.5. Lack of increase in reticulospinal tract excitability following distal motor training may reflect insufficient engagement or limitations in StartReact paradigm sensitivity

The StartReact response has been used as a non-invasive method to assess the output gain of the reticulospinal tract, which operates through the auditory startle reflex (Fisher et al., 2013; Honeycutt et al., 2013). The auditory pathway, beginning at the cochlear nucleus, directly activates giant neurons in the pontomedullary reticular formation (Valls-Solé et al., 2008). Following this activation, the signal is transmitted to motoneurons in the brainstem and spinal cord via the reticulospinal tract, facilitating rapid motor responses (Valls-Solé et al., 2008). Although the reticulospinal tract contribution to dorsiflexion is generally small in unimpaired individuals, it was significant across conditions (Figure 6e). The StartReact paradigm has been demonstrated as a feasible indirect method to assess reticulospinal tract function in the tibialis anterior, with evidence showing a measurable reduction in reaction times during startling stimuli compared to non-startling cues (Hayman et al., 2025). While the tibialis anterior primarily depends on corticospinal input, this small but consistent StartReact effect suggests that the reticulospinal tract plays a minor yet present role in dorsiflexion (Hayman et al., 2025).

While the StartReact paradigm effectively measured a significant contribution from the reticulospinal tract during the assessed tasks (Figure 6e), neither motor skill nor the isometric resistance training induced changes in reticulospinal tract excitability (Figure 6f). Given previous reports highlighting reticulospinal involvement in gross motor tasks, coordinated finger movements, and strength generation (Baker and Perez, 2017; Honeycutt et al., 2013; Sangari and Perez, 2019), our findings have several possible explanations. First, as the reticulospinal tract is thought to contribute more to balance and strength production (Glover and Baker, 2020; Prentice and Drew, 2001), the distal joint nature of our training tasks may inherently limit reticulospinal engagement. Second, while the reticulospinal gain is consistent across upper limb muscles in unimpaired individuals, it has been shown to be heightened in people with SCI in extensor muscles compared to flexor muscles (Sangari and Perez, 2020) or in the case of spasticity (Sangari and Perez, 2019). Therefore, it remains possible that training-induced changes in reticulospinal tract contributions could be different for people with damage to the central nervous system, such as in SCI. Third, the lack of measurable changes by any training intervention may suggest that the StartReact response is not sufficiently sensitive to detect changes in the excitability of the reticulospinal tract or that a single session is not sufficient to induce these changes. Fourth, changes in excitability after single training sessions may be less likely in subcortical structures compared to the cortex.

4.6. Reticulospinal sparing and corticospinal plasticity may have important clinical implications for targeted neurorehabilitation

Biological repair strategies may traditionally focus on regenerating corticospinal projections (Freund et al., 2007; Grill et al., 1997; Kucher et al., 2018) because of their high contribution to dexterous tasks in humans (Bunday et al., 2014; Perez and Rothwell, 2015). However, reticulospinal circuits contribute to recovery after corticospinal tract lesions or spinal cord injuries (Baker and Perez, 2017; Zaaimi et al., 2012), and their wide distribution in the white matter (Ballermann and Fouad, 2006; Nathan et al., 1996) and direct and indirect projections to spinal motor neurons (Riddle et al., 2009) make reticulospinal circuits more prone to be partially spared after SCI in humans (Kakulas, 1999). Moreover, reticulospinal tract neurons can display equal or greater responses to regeneration strategies after injury compared to corticospinal tract axons (Kadoya et al., 2016; Vavrek et al., 2007; Zörner et al., 2014), leading to more potential for recovery after SCI (Sangari et al., 2023), despite reticulospinal sparing being associated with increased spasticity (Sangari and Perez, 2019). Our findings suggest that single session motor skill and isometric resistance training protocols may be insufficient to engage the reticulospinal tract in the short term, highlighting a possible limitation of these approaches. Studies in SCI populations or across extended rehabilitation timelines could provide valuable insights into how to more effectively recruit reticulospinal pathways through training.

Additionally, skill-based training should be prioritized in rehabilitation programs targeting corticospinal plasticity (Christiansen et al., 2020). Individuals with limited mobility can still perform skill-based training, potentially benefiting from its neuroplastic effects. Our findings enhance the understanding of motor control and neural plasticity, supporting the emerging perspective that rehabilitation strategies should target specific neural pathways according to functional goals (Kumar et al., 2023; Maggio et al., 2024; OuYang et al., 2025). Examining variations in training intensity, duration, and task complexity will help identify optimal interventions for promoting neural plasticity. Longitudinal studies are crucial for evaluating long-term adaptations. Furthermore, integrating neurophysiological measures with functional assessments could significantly enhance translational potential and clinical applicability. In conclusion, while our study provides valuable insights into the effects of motor skill and resistance training on corticospinal, reticulospinal, and spinal excitability, future work is needed to refine training protocols and assessment methods to maximize rehabilitation outcomes.

5. Limitations of study

Our study advances previous research by concurrently examining multiple neural substrates involved in motor control—namely corticospinal, reticulospinal, and spinal circuits—in response to distinct training paradigms. This comprehensive approach provides broader insight into neural adaptations and interactions across different levels of the nervous system. However, the relatively small sample size (N = 23) and homogeneous population (26-year-old average unimpaired participants) restrict the generalizability to clinical populations.

Although motor-evoked potentials are a valuable non-invasive measure of corticospinal tract excitability (Eisner-Janowicz et al., 2023), they do not reflect intracortical excitability or inhibition, which may be important contributors to training-induced changes. Future studies employing paired-pulse TMS techniques to evaluate short-interval intracortical inhibition facilitation could further reveal specific intracortical processes underlying training-induced changes in excitability (Gómez-Feria et al., 2023; Ho et al., 2022). A further limitation is the optimization of TMS only over a single muscle. Identifying an optimal stimulation hotspot and conducting a full sweep of stimulation intensities to generate an input–output curve required approximately 10–20 minutes per muscle per timepoint, making multi-muscle testing impractical within the time constraints of the experimental protocol. For clarity and feasibility, we therefore focused on the tibialis anterior as a representative muscle, enabling simultaneous evaluation of corticospinal, reticulospinal, and spinal pathways from a single recording site. Likewise, although short-term changes in corticospinal and intracortical excitability remain present 75 min after the intervention (Benavides et al., 2020), it is unclear how long they remain stable. Moreover, although recruitment curves modeled with a Boltzmann equation could have allowed the estimation and comparison of additional parameters, such as S50, slope, and MEP_max (Kukke et al., 2014), we selected stimulation intensities of 100%, 150%, 180%, and 200% of resting motor threshold to specifically sample the range where training-induced increases in excitability are most reliably observed (Griffin and Cafarelli, 2007; Latella et al., 2017, 2016; Leung et al., 2017; Perez et al., 2004; Wilson et al., 2023), while also limiting the number of intensity steps to maintain a feasible experimental duration given that ten repetitions were collected at each level. Therefore, increasing the number of evaluations and, thus, the experiment time could play a role in the potential observed effects.

Similarly, our approach to assessing spinal excitability introduced certain limitations. The H-reflex served as the primary measure of spinal excitability, with peripheral nerve stimulation parameters optimized using long-duration pulses, 1 ms, to maximize its elicitation. Although shorter pulses (~100 μs) and stimulation over the fibular head are the standard for F-wave elicitation in the tibialis anterior, efferent fiber activation remains achievable with 1 ms pulses (Panizza et al., 1992). Our use of longer pulses at a non-standard stimulation site may introduce limitations; such as, if Ia afferents from the tibial nerve elicit disynaptic reciprocal inhibition on the tibialis anterior motoneuron pool, this could result in a smaller and more complex (less purely monosynaptic) H-reflex response than typically expected (Pierrot-Deseilligny and Mazevet, 2000). Nevertheless, M-waves, H-reflexes, and F-waves were successfully elicited in the tibialis anterior in 11 of 12 participants. In the medial gastrocnemius, M-waves and H-reflexes were observed in 11 of 12 participants, while F-waves were detected in 10 participants, with only 5 showing F-waves on both testing days. All responses were obtained without altering the stimulation protocol, demonstrating the feasibility of this approach within approximately a 3-hour experimental session.

It is important to note that although training time was limited to 30 min, experimental sessions lasted approximately 2.5 to 3 hours due to EMG preparation time (~30 min) and neurophysiological evaluations (~1 hour before training and ~1 hour after training). Additionally, continuous EMG activity during task execution was not recorded restricting our ability to directly evaluate fatigue effects of training. For the motor skill task in particular, interpretation of EMG could be less clear, as reductions may reflect improved efficiency rather than fatigue. Therefore, although evaluating neural excitability on additional pathways, muscles, time points, and under diverse conditions would certainly improve the clarity of changes in excitability (Benavides et al., 2020; Kumru et al., 2021; Sangari and Perez, 2020), it is crucial to consider participant comfort and avoid excessive session durations, as the 3-hour sessions could already approach the practical limits of participant endurance, attention, and retention.

Supplementary Material

Supplementary v5

Figure S1. Change in motor evoked potentials for non-target lower limb muscles following isometric resistance training and motor skill training.

Figure S2. Increases in corticospinal tract excitability following training measured via transcranial magnetic stimulation with secondary MEP normalization procedure.

Figure S3. Spinal excitability results of medial gastrocnemius.

Table S1: Participants’ demographics information for the neurophysiology experiments.

Table S2: GLMM results of MEP at rest for Figure 4.

Table S3: GLMM results of MEP at pre-activation for Figure 4.

Table S4: GLMM results of MEP at rest for Figure S2.

Table S5: GLMM results of MEP at pre-activation for Figure S2.

Table S6: GLMM results of non-target MEP at rest for Figure S1.

Table S7: GLMM results of non-target MEP at pre-activation for Figure S1.

Appendix A. Supplementary Data

Document S1.