Modulation of Hamstrings Neural Drive During Isometric Contractions of the Hip Extensors and Knee Flexors

ABSTRACT

The aim of the study was to compare motor‐unit discharge characteristics across regions of the hamstring muscles during hip‐extension and knee‐flexion. Fourteen healthy adults (2 females) performed isometric contractions with the hip extensors and knee flexors at 10% and 40% of maximal voluntary contraction (MVC) torque. High‐density surface electromyography was recorded from proximal and distal regions of biceps femoris and semitendinosus. Motor units were decomposed and tracked between tasks using a filter reapplication approach. The outcome measures included mean discharge rate (MDR), coefficient of variation for interspike interval (CoV_ISI), and standard deviation of the filtered cumulative spike train (SD of fCST). Task differences in common synaptic input were assessed at 10% MVC using within‐muscle coherence derived from CSTs in the delta band. Maximal torque was ~24% higher in hip‐extension than knee‐flexion, accompanied by a ~31% increase in torque variability. Across all the identified motor units and those that were tracked, MDR (~5%–14%), CoV_ISI (~7%–11%) and SD of fCST (~10%–16%) were greater during hip‐extension, whereas delta‐band coherence was greater across muscles and regions during knee‐flexion (p = 0.02–p < 0.001; d = 1.11–1.94). Moreover, proximal regions generally exhibited greater values for MDR (~11%) during knee‐flexion, whereas modulation was greater in distal regions during hip‐extension (~4%, p < 0.01). The neural‐drive variability was greater proximally than distally for both tasks (~16%, p = 0.01). The neural control of motor units differs between proximal and distal regions of the hamstrings during hip‐extension and knee‐flexion. Hamstring testing and exercise selection need to consider the task and the muscle region being examined.

1. Introduction

Force generation depends on the recruitment and rate coding of motor units [1]. Discharge rate and its variability reflect the temporal structure and modulation of synaptic input to motor neurons in response to task‐specific mechanical demands [1]. The effective neural drive to muscle is shaped by synaptic inputs shared across the motor neuron pool—termed common synaptic input—which can be quantified using cumulative spike trains (CST) and coherence analyses [2, 3]. Because shared input is closely associated with force fluctuations [2, 3, 4], the combined assessment of motor‐unit discharge properties and the strength of common input provides key insight into the neural control of force. Moreover, common synaptic input facilitates coordination across motor neuron pools by reducing the dimensionality of control [2, 3]. Its magnitude, however, likely varies across tasks, especially for biarticular muscles. Tasks with fewer degrees of freedom or more constrained joint actions may rely on a more unified neural drive than those with more options, such as knee flexion (KF) vs. hip extension (HE). Although common synaptic input can vary across tasks, this modulation is not consistently observed [5, 6].

The hamstrings contribute to both HE and KF, but they do not operate as a homogeneous muscle group [7]. Owing to their biarticular anatomy, they function with varying combinations of muscle length and moment arm across the hip and knee joints [7]. Moment arms are typically greater at the hip than at the knee joint, which likely contributes to the greater net torque during HE than KF [7]. In addition, inter‐ and intramuscular differences in architecture [8] and tendon properties [9] indicate that the semitendinosus (ST) and biceps femoris (BF), as well as regions within each muscle, may not contribute equally to the net torque during different tasks [7].

Regional differences in surface electromyographic (EMG) amplitude have been reported during hip‐ and knee‐based hamstring exercises [10, 11]. However, EMG amplitude is only an indirect surrogate of neural drive [12]. Comparisons of neural drive at the motor unit level across these actions remain equivocal. For example, the gastrocnemii may exhibit comparable motor‐unit activity across tasks (plantar flexion vs. KF) [5], whereas task‐dependent differences in characteristics and intramuscular coherence have been reported in upper‐limb muscles [6].

Recent work suggests regional and intermuscular differences in hamstring motor unit activity [13, 14]. Proximal and distal compartments of the ST [13], as well as synergistic muscles (ST vs. BF), appear to display distinct discharge characteristics [14]. However, it is unclear whether such compartmentalization is modulated by task demands. HE differs from KF in both torque production and moment arms [7], factors that likely influence the relative mechanical contribution and neural drive of different muscle regions [15]. For example, regions located further from the joint axis may require greater activation to contribute to torque generation [13]. Despite this rationale, task‐dependent modulation of regional motor‐unit activity in the hamstrings has not been directly examined. Although dos Santos Albarello et al. [16] described motor‐unit discharge properties in BF during both HE and KF, a direct comparison between tasks was not performed.

In this study, we compared the neural control of the hamstrings between isometric HE and KF performed at the same relative torque. The first aim was to compare motor‐unit discharge characteristics using metrics derived from decomposed high‐density surface EMG (HD‐sEMG) signals. Based on evidence that modulation of motor‐unit activity varies with task demands [6], we hypothesized that HE would exhibit greater discharge rates and variability than KF. The second aim was to compare the strength of shared low‐frequency synaptic input across tasks using delta‐band (0–5 Hz) coherence estimated from within‐muscle CST. Given the proposed role of common synaptic input in reducing control dimensionality [2, 3], we hypothesized that delta‐band intramuscular coherence would be greater during KF than HE. The third aim was to compare proximal and distal regions within each hamstring muscle. In line with architectural heterogeneity [7, 8] and prior findings [13], we hypothesized that proximal regions would exhibit relatively greater discharge rates during KF, whereas distal regions would exhibit relatively greater discharge rates during HE.

Clarifying whether motor‐unit discharge characteristics and low‐frequency shared input vary with tasks and across muscle regions will improve the interpretation of hamstring assessments. If these outcomes are task and region specific, measurements should not be generalized across conditions, thereby limiting direct comparisons across testing sessions and studies.

2. Materials and Methods

2.1. Participants

No a priori power analysis was performed because no directly comparable data were available to estimate effect sizes for task‐, region‐, and torque‐dependent differences in hamstring motor‐unit discharge characteristics. Therefore, the sample size was defined using a feasibility‐based justification, given the technical demands of HD‐sEMG decomposition and motor‐unit tracking across multiple repeated‐measures conditions [17]. Eighteen healthy, physically active adults were recruited (6 females). Four females were excluded because HD‐sEMG decomposition yielded too few motor units in ST and/or BF. Therefore, fourteen participants were included in the final analyses (mean ± SD, 22.9 ± 1.1 years; 1.88 ± 0.07 m; 87.2 ± 5.6 kg; 1810 ± 650 MET·min·week⁻¹). Volunteers were screened using a standardized self‐report questionnaire and excluded if they had sustained any lower‐limb injury (such as ligament‐related knee injury or hamstring strain) in the past 12 months, or if they had a history of knee surgery or neurological disorder during the same period. Throughout the study, participants continued usual daily activities but did not undertake any structured training program. The protocol complied with the Declaration of Helsinki and was approved by the Ethics Committee of the Aristotle University of Thessaloniki (ERC‐016/2025). Written informed consent was obtained before testing.

2.2. Experimental Design

Participants performed two isometric tasks (HE and KF) with the left limb while lying prone on an isokinetic dynamometer (Humac Norm, CSMI, MA, USA). The left limb was selected to standardize data collection. The hip was held in a neutral position for the HE task and the knee was fixed at 0° (full extension) (Figure 1). The hip was maintained in a neutral position for the KF task and the knee was fixed at 30°, which corresponds to the joint angle at which peak torque is typically produced [7] (see Methodological Considerations). The pelvis was secured with a strap. Hip and knee joint angles were monitored continuously using twin‐axis goniometers (TSD 130B, Biopac Systems, Goleta, CA, USA). Any trial with visible trunk or hip displacement during data collection was discarded and repeated.

FIGURE 1.

(A) Overview of the protocol for the two tasks (hip extension and knee flexion). Measurement of maximal voluntary contraction (MVC) torque was followed by torque steadiness trials at 10% and 40% MVC, performed in a randomized order. (B) Spatial distribution of motor‐unit action potentials recorded across the high‐density surface EMG grid for Semitendinosus (light green) and for Biceps Femoris (light purple). (C) Electrode placement over the biceps femoris and semitendinosus with representative ultrasound images at rest. Proximal and distal HD‐sEMG grids were positioned under ultrasound guidance to align with muscle borders and fascicle orientation, with ~1 cm spacing maintained between grids along the thigh. (D) Participant position during hip extension and knee flexion on the dynamometer with visual torque feedback displayed on the monitor.

2.3. Experimental Protocol

Participants completed a standardized warm‐up (~5 min light cycling) followed by ~3–5 min static and dynamic stretching of the knee flexors. After electrode placement, which took ~20 min, participants practiced both the submaximal and maximal tasks on the isokinetic dynamometer. Peak torque was determined during maximal voluntary contractions (MVCs) for the KF and HE tasks. Participants performed two to three MVCs of 3–5 s duration, separated by 3–5 min recovery periods and were supported by verbal encouragement [13, 14].

After 5 min of rest, participants performed, in a randomized order, isometric contractions at 10% and 40% MVC torque for both tasks (Figure 1). Each trial comprised a linear ramp‐up to the target torque at 5% MVC·s⁻¹, a steady isometric phase (30 s at 10% MVC and 15 s at 40% MVC), and a controlled ramp‐down. Three trials were completed at each target level with ~1 min rest between trials. Real‐time visual feedback of the target and applied torques was provided on a monitor positioned at eye level. Perceived discomfort was monitored throughout testing using a 0–10 numerical rating scale (0 = no discomfort; 10 = extreme discomfort) [13].

2.4. High‐Density Surface Electromyography

HD‐sEMG signals were recorded from the proximal and distal regions of ST and BF using two‐dimensional grid electrodes comprising 64 gold‐coated pins (5 × 13 configuration, with one pin absent at a corner; 8 mm between pins; OT Bioelettronica, Italy). Monopolar HD‐sEMG signals were acquired with a multichannel amplifier (EMG‐Quattrocento, OT Bioelettronica), sampled at 2048 Hz, amplified (×150), band‐pass filtered (10–500 Hz), and digitized at 16‐bit resolution. Reference electrodes were positioned over the ipsilateral patella and medial malleolus, and a ground electrode was placed on the contralateral wrist. Data were recorded using OTBioLab+ software.

Prior to grid placement, the skin was shaved, lightly abraded, and cleaned with water. Each grid was mounted on an adhesive foam layer filled with conductive paste and was secured with an elastic band. Electrode position was guided by longitudinal and transverse ultrasound scans to identify muscle borders and fascicle orientation and to verify separation of ST from the adjacent semimembranosus both at rest and during contraction [18, 19]. Ultrasound was used to identify the tendinous inscription of ST as a reference for electrode placement. The proximal electrode for ST was placed over the compartment above the tendinous inscription (proximal compartment), whereas the distal electrode was placed below the tendinous inscription (distal compartment). The HD‐sEMG electrodes for BF were placed in a similar position to ST (Figure 1). There was ~1‐cm distance between the grid electrodes to minimize crosstalk [13, 14].

2.5. Data Analysis

2.5.1. Torque Analysis

The torque signal was low‐pass filtered using a fourth‐order, zero‐phase (zero‐lag) Butterworth filter (20 Hz cut‐off) to remove high‐frequency noise. Peak torque was then defined as the greatest value obtained from the filtered signal using a 500‐ms moving‐average window.

A 10‐s window with the lowest coefficient of variation (CoV) for torque during the isometric plateau was identified during each submaximal contraction. The torque in this window was subsequently band‐pass filtered with a second‐order Butterworth filter (0.75–5 Hz) to isolate the low‐frequency oscillations underlying force steadiness [20]. The fluctuations in torque were quantified as the CoV (SD/mean × 100%). The trial with the lowest CoV for torque was selected for analysis for each task and target.

2.6. HD‐sEMG Processing and Motor‐Unit Decomposition

All offline analyses were performed in MATLAB using custom‐written scripts. HD‐sEMG signals were band‐pass filtered (20–500 Hz), after which channels with artifacts or poor signal quality were removed based on visual inspection (median [IQR]: 2 [1, 2]; range: 1–3 channels). Motor‐unit spike trains were then extracted with a validated convolutive blind source separation method implemented in MUedit [21] (Figure 2). Only motor units meeting a silhouette criterion of ≥ 0.90 were accepted and analyzed. Each motor‐unit spike train was subsequently reviewed and manually corrected by an experienced investigator. Motor units that did not show continuous activity during the steady contraction as well as motor units that exhibited non‐physiological interspike intervals (5–30 pps, < 33.3 ms or > 200 ms) were discarded.

FIGURE 2.

Representative example from the distal region of biceps femoris during knee flexion (top) and hip extension (bottom) at 10% MVC torque. Left panels show HD‐sEMG signals recorded from a 13‐channel column during the steady phase of the isometric contractions. Signals were decomposed to identify individual motor units. Middle panels display raster plots of motor‐unit discharge times, with the superimposed force/torque trace illustrating the steady contraction phase used for analysis. Right panels show intramuscular coherence spectra (0–50 Hz) used to quantify common oscillations in discharge rate. Coherence was estimated from motor‐unit (MU) spike trains during the steady plateau. The trains were randomly split into two equal groups and summed to form two cumulative spike trains (CST1 and CST2). This split‐and‐sum procedure was repeated across 100 permutations, and coherence was computed between CST1 and CST2 for each permutation; spectra were then averaged across permutations. To ensure comparable CST size, the number of MUs contributing to each CST was fixed to one‐half of the MU count from the task with fewer identified units. Colored curves depict coherence spectra computed using increasing numbers of MUs (from the minimum to the maximum available). The horizontal line indicates the significance threshold.

2.7. Motor‐Unit Discharge Variables and Crosstalk Control

The discharge characteristics of both all identified and tracked motor units were quantified from the steadiest 10‐s epoch of each contraction. Mean discharge rate (MDR; pps), CoV of the interspike interval (CoV_ISI; %), and SD of the filtered cumulative spike train (SD of fCST; au) were extracted for each motor unit. Each train of motor unit discharge times was low‐pass filtered with a 400‐ms Hanning window and subsequently summed to form the CST. To ensure that CST estimates could be compared across muscles, regions, torque levels, and tasks, motor‐unit yield was matched within each participant by identifying the minimal number of motor units across all conditions (muscle × region × torque level × task) and randomly subsampling that number from each cell. CST metrics were averaged across repeated draws to minimize subsampling bias.

After manual optimization, discharge times were used to compute spike‐triggered averages of motor‐unit action potentials across all electrode grids [13, 14]. The maximal peak‐to‐peak amplitude of the action potential for each motor unit was identified across all channels of each grid. A motor unit was classified as crosstalk when its maximal peak‐to‐peak amplitude was greater in a grid corresponding to an adjacent muscle or region than in the grid from which it was originally identified and was subsequently excluded from further analysis. We further screened for duplicates by examining temporal coincidence of discharge times; motor units with ≥ 30% overlap in discharge times between muscle/regions were removed from the analyses.

2.8. Motor‐Unit Tracking Across Tasks

After manual editing of discharge times, motor units were tracked across the two tasks to enable within‐unit comparisons of discharge characteristics using a filter‐reapplication procedure (Figure 3). In brief, separation vectors derived from blind‐source separation decomposition in one task were reapplied to the HD‐sEMG signals recorded during the other task to reconstruct spike trains [6]. Because the separation vector for each motor unit was unique, it functioned as a spatiotemporal filter that permitted the identification of the same motor units across tasks. Tracking was performed bidirectionally (KF to HE in both directions) to limit directional bias and to verify consistency. Only motor units that were identified in both directions were retained for the final analyses.

FIGURE 3.

(A) Schematic overview of the motor‐unit (MU) tracking and coherence analysis. HD‐sEMG signals were decomposed into MU discharge times during steady isometric knee flexion (KF) and hip extension (HE) contractions. Raster plots of discharge times illustrate the data for all and the tracked MUs for a representative trial at 10% MVC torque from the distal biceps femoris region (HE, light purple; KF, dark purple). MUs were tracked by reapplying separation vectors for each MU extracted during the plateau phase of the steady contraction for one task to the HD‐sEMG signals for the other task. Common synaptic input was estimated from the level of coherence between two cumulative spike trains (CST1 and CST2). These were constructed by randomly splitting the available MU spike trains into two equal groups and summing those within each group. The random split was repeated across 100 permutations and coherence spectra were averaged across permutations. CST size was fixed to one‐half of the MU count from the condition (task) with fewer available units to standardize estimates across comparisons. Coherence was computed either from (i) all MUs within each task (left) or (ii) the tracked MUs common to both tasks (right); significance was assessed against the corresponding threshold. Intramuscular coherence spectra (z‐transformed) were averaged across groups for all (B) and tracked (D) MUs for biceps femoris (BF, purple) and semitendinosus (ST, green). Dark lines indicate KF and light lines show HE. Individual and group values for area under the curve (A_UC) for the delta‐band (0–5 Hz) for all (C) and tracked (F) MUs above bias. Lines connect task conditions within participants.

2.9. Motor‐Unit Coherence

To quantify adjustments in common synaptic input to the motor‐neuron pool across tasks, we performed coherence analysis on motor‐unit spike trains [5, 22, 23] (Figure 2). Coherence describes the frequency‐specific linear coupling between two signals, ranging from 0 (no coupling) to 1 (perfect coupling) at a given frequency. Analyses were conducted within each muscle (BF, ST) and within each region (proximal, distal). We focused on the delta‐band, which appears to reflect common synaptic oscillations relevant to torque modulation [5, 23, 24].

Because coherence estimates require at least ~20 s of data [5, 23, 24], coherence was computed only for the 10% MVC condition during the 30‐s plateau (Figure 2). Motor‐unit discharge times were converted to binary spike trains, and CSTs were formed by summing the spike trains of randomly selected motor units. Two CSTs of equal size were created for each iteration and coherence was calculated between them. To avoid inflating coherence due to unequal motor‐unit counts, the number of motor units summed per CST was set to one half of the total number of detected motor units in the task with fewer identified units, and this same group size was used for both directions (Figure 2). This random sampling procedure was repeated for up to 100 iterations, and coherence was averaged across iterations for subsequent analyses. Trials were required to have at least four motor units available between the two tasks for inclusion, as coherence was estimated between two equally sized CSTs, requiring a minimum of two motor units per CST [6]. This criterion was satisfied for all participants for the set of all motor units, but the sample sizes were reduced for the tracked motor units: BF proximal (n = 7), BF distal (n = 8), ST proximal (n = 7), and ST distal (n = 7).

Coherence was computed in MATLAB for each iteration between the two detrended CSTs using mscohere and cpsd, with a 1‐s Hanning window and 75% overlap [24]. The resulting coherence spectra were transformed to z‐scores [25]. A bias level was defined as the mean z‐coherence between 250 and 500 Hz [5, 23, 26], and only z‐coherence values exceeding this bias were retained. Differences in common synaptic input between tasks were summarized by calculating the area under the curve (AuC) in the delta band, including only z‐coherence values above the bias.

2.10. Statistics

All analyses were performed in R (v4.3.0; Vienna, Austria). Data distribution was evaluated using quantile–quantile plots and the Kolmogorov–Smirnov test. Differences in maximal torque between the two tasks were examined with a paired‐samples t‐test. Task‐ and torque‐level effects were then assessed with linear mixed models ( nlme package), specifying task (KF vs. HE) and target torque as fixed effects and participant as a random effect. Motor‐unit discharge characteristics were analyzed using linear mixed models including muscle (ST vs. BF), region (proximal vs. distal), task (HE vs. KF), and target torque (10% and 40% MVC) as fixed effects, with participant as a random effect. Fixed effects were evaluated using analysis of variance. When significant main effects or interactions were detected, post hoc pairwise comparisons were performed using least‐squares contrasts with Tukey adjustment (lsmeans package). To compare the synaptic input between the two tasks, paired‐samples t‐tests were applied to the AuC of z‐coherence.

Effect sizes were expressed as partial eta squared (ηp²) and Cohen's d for post hoc contrasts. Statistical significance was set at p < 0.05. Post hoc results are reported as the mean estimate with 95% confidence interval, otherwise stated. For reporting and visualization, SD of fCST values are expressed as arbitrary units multiplied by 100 (a.u. × 100).

To examine whether task‐related differences in delta‐band coherence were related to differences in MDR, Pearson correlation analyses were conducted at 10% MVC. MDR was first averaged across all identified motor units within each task for each participant. A task‐related change ratio between tasks was then calculated for MDR and delta‐band coherence, and the corresponding change ratios were correlated.

3. Results

Participants reported discomfort ratings ≤ 3 throughout the experiment, indicating that the protocol was well tolerated.

3.1. Maximal Torque

Analysis indicated a significant main effect of task (t ₍₁₃₎ = 4.09, p = 0.002), with greater maximal torque observed during HE (150.5 Nm [119.3, 181.8]) compared with KF (121.7 Nm [100.0, 143.4], d = 1.18) (Figure 4).

FIGURE 4.

Maximal torque and torque steadiness during knee flexion and hip extension. Left panel: Maximal torque (Nm) obtained during knee flexion (KF) and hip extension (HE). Middle and right panels: Torque steadiness quantified as the coefficient of variation (CoV, %) for torque at 10% MVC and 40% MVC, respectively. Individual participant data are shown as colored circles with connecting lines between tasks. Shaded bars represent group means ± SEM. Asterisks indicate a significant difference between KF and HE within the same condition (p < 0.05).

3.2. Torque Steadiness

There were significant main effects for task (F = 25.09, p < 0.001, ηp² = 0.69) and target torque (F = 20.45, p < 0.001, ηp² = 0.65), as well as a significant task × target torque interaction (F = 6.81, p = 0.02, ηp² = 0.38). Post hoc tests indicated that CoV for torque was greater during HE (10%: 3.3% [2.8, 3.80] and 40%: 2.2% [1.9, 2.6]) than during KF (10%: 2.3% [2.0, 2.6] and 40%: 1.9% [1.5, 2.2], d = 0.71–1.28). Additionally, CoV for torque decreased from 10% (2.8% [2.5, 3.2]) to 40% MVC torque (2.0% [1.8, 2.3], d = 1.31) (Figure 4).

3.3. Motor‐Unit Decomposition and Crosstalk

After crosstalk and duplicate evaluation, seven BF and eight ST motor units were identified as crosstalk, and six BF and seven ST were regarded as duplicates; all were excluded from further analyses. The total number of identified motor units varied across task and region (Table 1). For BF at 10% MVC, 84–100 motor units were identified in the proximal region and 147–148 in the distal region (tracked: 47–66). At 40% MVC, 62–66 (proximal) and 82–92 (distal) motor units were identified (tracked: 27–35). For ST, 61–72 (proximal) and 88–121 (distal) motor units were identified at 10% MVC (tracked: 43–47), and 47–48 (proximal) and 39–44 (distal) at 40% MVC (tracked: 29–33).

TABLE 1.

Number of motor units identified and tracked across the different tasks, regions, and muscles (n = 14).

	Target torque (% MVC)	Proximal region			Distal region
		HE	KF	Tracked	HE	KF	Tracked
Biceps femoris
All MUs	10%	100	84	47	147	148	66
	40%	66	62	35	92	82	27
Average	10%	7.1 ± 3.2	6.0 ± 2.5	3.4 ± 2.5	10.5 ± 4.8	10.6 ± 5.0	4.7 ± 3.5
	40%	4.7 ± 3.3	4.4 ± 3.2	2.5 ± 2.2	6.6 ± 4.2	5.9 ± 4.0	1.9 ± 1.8
Semitendinosus
All MUs	10%	72	61	47	121	88	43
	40%	48	47	33	44	39	29
Average	10%	5.1 ± 1.8	4.4 ± 1.3	3.4 ± 2.5	8.6 ± 4.0	6.3 ± 3.0	3.1 ± 2.3
	40%	3.4 ± 2.8	3.4 ± 2.7	2.4 ± 2.1	3.1 ± 2.6	2.8 ± 2.4	2.1 ± 1.9

Note: Tracked: number of MUs tracked between the two tasks at the same target torque. Data are presented as the total number of MUs and the mean ± SD number of MUs per participant, calculated across the final sample of 14 participants.

Abbreviations: HE, hip extension task; KF, knee flexion task.

3.4. All Identified Motor Units

3.4.1. Mean Discharge Rate

There was a significant main effect for MDR in task (F = 17.81, p < 0.0001, ηp² = 0.01), muscle (F = 36.11, p < 0.0001, ηp² = 0.02), and target torque (F = 741.42, p < 0.0001, ηp² = 0.29). Additionally, the task × region × target torque interaction (F = 6.58, p = 0.01, ηp² = 0.004) was significant (Figure 5A).

FIGURE 5.

Discharge characteristics of all motor units during hip extension and knee flexion. (A) Mean discharge rate (MDR, pps), (B) coefficient of variation for interspike interval (CoV ISI, %), and (C) standard deviation of the filtered cumulative spike train (SD of fCST, a.u. ×100) for motor units recorded with proximal and distal electrodes from the biceps femoris (BF, purple) and semitendinosus (ST, green) during steady contractions at 10% and 40% MVC torque. As indicated in the legend at the bottom of the figure, data are shown separately for hip extension (HE, light colors) and knee flexion (KF, dark colors). Boxes represent the interquartile range (25th–75th percentile), the central line denotes the median, whiskers indicate minimum and maximum values, and individual points correspond to single motor units (A, B) or participant means (C). Statistical annotations indicate: * significantly greater than KF in the same region; # greater than the proximal region within the same task; $ greater than the distal region within the same task; ^ greater than the other muscle within the same condition.

At 10% MVC torque, post‐hoc tests indicated that distal regions exhibited greater MDR during HE compared with KF in both BF (HE 10.7 pps [9.9, 11.5] vs. KF 9.4 pps [8.6, 10.2], d = 0.63) and ST (HE 9.5 pps [8.7, 10.3] vs. KF 8.6 pps [7.8, 9.4], d = 0.45). Similar results were also found at 40% MVC torque for both BF (HE 13.8 pps [13.0, 14.6] vs. KF 13.1 pps [12.3, 13.9], d = 0.34) and ST (HE 12.8 pps [11.8, 13.7] vs. KF 11.8 pps [10.9, 12.7], d = 0.45).

In contrast, MDR for the proximal region was similar between the two tasks (p ≥ 0.05). Also, MDR was less for the proximal region (10.7 pps [10.4, 11.0]) than the distal region (11.1 pps [10.8, 11.3]) during the HE task, whereas the opposite was found for the KF task (proximal: 11.2 pps [10.9, 11.4]; distal: 10.1 pps [9.8, 10.3]; p < 0.01).

Additionally, overall MDR was greater during HE (11.5 pps [10.8, 12.3]) than during KF (11.0 pps [10.3, 11.8], d = 0.24) and increased from 10% to 40% (d = 1.43). Also, MDR for BF (11.6 pps [10.9, 12.3]) was greater than that for ST (11.0 pps [10.2, 11.7], d = 0.32).

3.4.2. Discharge Variability (CoV_ISI )

There were significant main effects for CoV _ISI in task (F = 8.45, p = 0.004, ηp² = 0.005), muscle (F = 48.61, p < 0.0001, ηp² = 0.03), and target torque (F = 55.73, p < 0.0001, ηp² = 0.03). Task × region × target torque (F = 6.57, p = 0.01, ηp² = 0.005) and muscle × region × target torque interactions (F = 5.04, p = 0.02, ηp² = 0.003) were also significant (Figure 5B).

Post hoc tests indicated that CoV_ISI was greater in the distal region during HE compared with KF at 10% MVC torque for BF (HE 16.5% [14.9, 18.1] vs. KF 15.4% [13.8, 17.0], d = 0.21) and ST (HE 18.2% [16.5, 19.8] vs. KF 17.0% [15.4, 18.6], d = 0.21). Similarly, CoV_ISI was greater for the proximal region during HE compared with KF at 10% MVC torque for BF (HE 16.5% [14.9, 18.2] vs. KF 15.3% [13.6, 17.0], d = 0.22) and ST (HE 17.9% [16.2, 19.7] vs. KF 16.7% [15.0, 18.4], d = 0.22). There were no significant differences across tasks at 40% MVC torque.

Furthermore, the overall CoV_ISI was greater during HE (18.2% [16.6, 19.7]) than during KF (17.4% [15.8, 18.9], d = 0.15) and increased with target torque (d = 0.39). Also, overall CoV_ISI was greater for ST (18.8% [17.0, 20.6]) than BF (16.8% [15.1, 18.5], d = 0.36).

3.4.3. Variability in Neural Drive (SD of fCST)

There were significant main effects for SD of fCST in task (F = 4.38, p = 0.04, ηp² = 0.02), muscle (F = 67.80, p < 0.0001, ηp² = 0.27), region (F = 6.55, p = 0.01, ηp² = 0.03), and target torque (F = 4.44, p = 0.04, ηp² = 0.02) (Figure 5C).

Across conditions, SD of fCST was greater during HE (0.21 au [0.17, 0.25]) than KF (0.19 au [0.15, 0.23], d = 0.32) and increased from 10% to 40% MVC torque (d = 0.31). Also, the SD of fCST for BF (0.23 au [0.19, 0.28]) was greater than for ST (0.16 au [0.13, 0.20], d = 1.17). Moreover, the SD of fCST for the proximal region (0.22 au [0.18, 0.26]) was greater than that for the distal region (0.19 au [0.15, 0.23], d = 0.41).

3.5. Tracked Motor Units

3.5.1. Mean Discharge Rate

There were significant main effects for MDR in task (F = 7.93, p = 0.005, ηp² = 0.01), region (F = 4.84, p = 0.03, ηp² = 0.007), and target torque (F = 269.05, p < 0.0001, ηp² = 0.3). Additionally, interactions for task × region (F = 12.30, p < 0.001, ηp² = 0.02) and region × target torque (F = 34.51, p < 0.0001, ηp² = 0.05) were significant (Figure 6A).

FIGURE 6.

Discharge characteristics of tracked motor units during hip extension and knee flexion. (A) Mean discharge rate (MDR, pps), (B) coefficient of variation for interspike interval (CoV ISI, %), and (C) standard deviation of the filtered cumulative spike train (SD of fCST, a.u. ×100) for motor units recorded with proximal and distal electrodes from biceps femoris (BF, purple) and semitendinosus (ST, green) during steady contractions at 10% and 40% MVC torque. As indicated in the legend at the bottom of the figure, data are shown separately for hip extension (HE, light colors) and knee flexion (KF, dark colors). Boxes indicate the interquartile range (25th–75th percentile), the central line denotes the median, whiskers represent minimum and maximum values, and individual points correspond to single tracked motor units (A, B) or participant means (C). Statistical annotations indicate: * significantly greater than KF within the same region; † significantly greater than HE within the same region; # significantly greater than the proximal region within the same task and muscle.

Post‐hoc tests showed that MDR was greater in distal regions during HE than during KF at 10% MVC for both BF (HE 10.5 pps [10.0, 11.0] vs. KF 9.6 pps [9.0, 10.1], d = 0.49) and ST (HE 10.0 pps [9.5, 10.6] vs. KF 9.3 pps [8.8, 9.8], d = 0.37). Similar results were also found at 40% MVC for both BF (HE 14.2 pps [13.5, 14.8] vs. KF 13.0 pps [12.3, 13.6], d = 0.63) and ST (HE 13.6 pps [12.9, 14.2] vs. KF 12.6 pps [11.9, 13.2], d = 0.51). In contrast, the task effect was not significant for the proximal region.

Overall, MDR was greater for HE than for KF (HE 11.6 pps [11.3, 11.9] vs. KF 11.2 pps [10.9, 11.5], d = 0.23) and increased from 10% to 40% MVC torque (d = 1.32). Also, MDR was greater in the distal region (11.6 pps [11.4, 11.8]) than the proximal region (11.2 pps [11.0, 11.4], d = 0.18).

3.5.2. Discharge Variability (CoV_ISI )

There were significant main effects for CoV _ISI in task (F = 4.85, p = 0.028, ηp² = 0.008) and muscle (F = 32.52, p < 0.0001, ηp² = 0.05). Additionally, the interactions for task × muscle × target torque (F = 9.33, p = 0.002, ηp² = 0.01) and task × region × target torque (F = 3.86, p = 0.04, ηp² = 0.003) were significant (Figure 6B).

Post hoc tests revealed that the CoV _ISI was greater at 10% MVC torque in the distal region during HE compared with KF for both BF (HE 17.1% [14.4, 19.7] vs. KF 15.3% [13.8, 16.8], d = 0.49) and ST (HE 19.5% [18.0, 21.0] vs. KF 18.2% [16.2, 20.2], d = 0.30). However, the task effect was reversed at 40% MVC torque in BF (HE 14.9% [13.3, 16.5] vs. KF 18.4% [16.5, 20.3], d = 0.78, p = 0.02), whereas ST remained greater in HE (18.2% [16.5, 19.9]) than KF (16.7% [15.0, 18.3], d = 0.52). In contrast, there were no significant differences for the proximal region at 10% and 40% MVC torque for BF and ST.

Overall CoV _ISI was greater in HE (17.9% [17.2, 18.6]) than KF (17.1% [16.4, 17.8]; d = 0.18), and was greater in ST (18.6% [17.0, 20.1]) than BF (16.5% [14.9, 18.0]; d = 0.48).

3.5.3. Variability in Neural Drive (SD of fCST)

There were significant main effects for SD of fCST in task (F = 4.67, p = 0.03, ηp² = 0.02) and muscle (F = 35.12, p < 0.0001, ηp² = 0.16). Additionally, the muscle × target torque interaction was significant (F = 10.05, p = 0.002, ηp² = 0.05) (Figure 6C).

Post hoc comparisons showed that the SD of fCST was greater for BF than ST at 10% (BF 0.15 au [0.13, 0.18] vs. ST 0.11 au [0.08, 0.13], d = 0.42–1.17). Additionally, SD of fCST was greater in HE (0.14 au [0.13, 0.15]) than KF (0.12 au [0.11, 0.13], d = 0.28).

3.6. Coherence Between the Two Tasks

3.6.1. All Identified Motor Units

Both the proximal (t = 4.26, p < 0.0001, d = 1.23) and distal regions (t = 4.47, p < 0.0001, d = 1.29) in BF exhibited greater normalized (z‐coherence) delta‐band AuC values during KF than HE (Figure 3). The results were similar for the proximal (t = 5.78, p < 0.0001, d = 1.67) and distal regions (t = 6.72, p < 0.0001, d = 1.94) of ST. The correlation analysis indicated no significant association between the task‐related change ratios in coherence and MDR (r = 0.19–0.32, p ≥ 0.25).

3.6.2. Tracked Motor Units

Delta‐band z‐coherence was greater in both proximal (t = 4.97, p = 0.002, d = 1.76) and distal regions of BF (t = 2.94, p = 0.02, d = 1.11) during KF compared with HE. Similar results were observed in ST, with greater delta‐band z‐coherence in the proximal (t = 3.63, p = 0.01, d = 1.37) and distal regions (t = 6.40, p < 0.001, d = 1.26) during KF than during HE (Figure 3). Correlation analyses showed no significant association between the task‐related change ratios in coherence and MDR across tasks (r = 0.15–0.36, p ≥ 0.29).

4. Discussion

The main purpose of this study was to compare motor‐unit discharge characteristics across regions of the hamstring muscles during isometric contractions for HE and KF. Maximal torque and torque variability were greater during HE than KF. In parallel, MDR, discharge‐time variability, and neural‐drive variability were greater, whereas delta‐band coherence was lower, during HE. Regionally, MDR was greater in proximal regions during KF, but greater in the distal regions during HE.

4.1. Hip Extension Produced Higher MVC but Lower Torque Steadiness Than Knee Flexion

Maximal torque was greater during HE than during KF (Figure 4) due to the hamstrings operating at longer MTU lengths in HE. Based on in vivo estimates and musculoskeletal modeling, these joint configurations likely shift the operating range of hamstring sarcomeres from the ascending limb of the force–length relation during KF toward a more optimal region near the plateau during HE [7]. Additionally, the superior torque capacity during HE is associated with a larger physiological cross‐sectional area (PCSA) [27] and greater mechanical advantage [7, 28]. Whereas KF involves several smaller muscles, such as gastrocnemius (30.8 cm²), biceps femoris short head (5.1 cm²), sartorius (1.7 cm²), and gracilis (2.0 cm²), HE recruits a substantially larger muscle mass, including the gluteus maximus (33.4 cm²) and adductor magnus (20.5 cm²) [7, 27]. Combined with larger hamstring moment arms at the hip, these factors could potentially explain the greater torque observed during HE [7, 29, 30].

Despite this higher torque capacity, torque steadiness was reduced during HE (Figure 4). This finding aligns with the motor unit data, which showed greater neural‐drive variability (Figures 5 and 6) and lower delta‐band coherence (Figure 3), suggesting a less stable low‐frequency common input to the motor neuron pool. Mechanically, HE involves greater redundancy, as torque can be produced through multiple combinations of hip extensors and stabilizing actions at the pelvis and trunk [30]. Such redundancy allows small shifts in activation patterns to alter load‐sharing muscles, thereby increasing output variability [31]. This interpretation is consistent with evidence showing lower force fluctuations in tasks involving fewer muscles compared with multi‐muscle tasks [32, 33].

4.2. Task‐Dependent Modulation of Motor Unit Activity

Across both all identified (Figure 5) and tracked motor units (Figure 6), HE was associated with greater MDR, discharge‐time variability, and neural‐drive variability compared with KF. The consistency of these findings suggests that they are not attributable to differences in sampled motor units. Increased neural‐drive variability during HE is undoubtedly responsible for the reduced torque steadiness observed in this task [20, 34]. This result is further supported by the higher delta‐band coherence during KF, indicating stronger shared low‐frequency input.

The greater MDR during HE may reflect two factors. First, despite the normalized target torques, differences in joint configuration (e.g., moment arms and muscle length) likely reduce the relative contribution of the hamstrings to the net torque [7]. Second, the hamstrings contribute less to net HE torque compared with other hip extensors [35]. Consequently, achieving the same relative torque may require greater hamstring activation, consistent with the observed increase in discharge rates.

Differences in discharge timing, neural‐drive variability, and shared synaptic input likely reflect task‐dependent modulation of descending and afferent input at the spinal level. Group I afferent input is integrated with descending commands through interneuronal networks and is strongly regulated by presynaptic inhibition [36, 37]. This organization may allow flexible adjustment of both excitation level and the degree of shared synaptic input [37]. Additionally, spinal networks project to multiple muscles, coordinating activity across synergists and potentially influencing how shared input is distributed across the motor neurons innervating different regions of the hamstrings [38, 39]. Importantly, simulations indicate that non‐uniform distribution of shared input across a motor‐neuron pool can reduce coherence without decreasing the overall level of shared synaptic input [40]. Thus, task‐dependent differences in descending drive and afferent feedback may alter the composition of shared input, contributing to the observed differences in coherence. As both tasks involve multiple synergistic muscles and the present analysis was limited to the hamstrings, these findings should be interpreted as muscle‐specific rather than reflective of the entire neural strategy underlying torque production.

4.3. Task Selection Influences Region‐Specific Motor‐Unit Activity

Significant regional differences were observed (Figures 5 and 6), with distinct patterns across tasks. Proximal regions exhibited greater MDR during KF than distal regions, whereas the opposite pattern was found during HE. Across both tasks, neural‐drive variability (SD of fCST) remained greater in proximal regions. Task‐related modulation of MDR and discharge variability was more pronounced distally than proximally. These findings may reflect regional differences in mechanical contribution and operating fascicle length across joint configurations [41, 42]. Changes in hip and knee angles may shift different regions along the force–length relation, thereby altering the motor‐unit activity required to achieve a given torque.

This interpretation aligns with previous evidence of proximo‐distal differences in ST activity during KF [13]. In contrast, some studies have reported no regional differences [16], potentially due to methodological factors such as limited electrode coverage or fewer identified motor units. Moreover, previous studies have reported no differences in shear modulus between proximal and distal ST compartments under passive conditions [43], and have shown that morphology and adaptations are strongly coupled [44]. These discrepancies may arise from differences between passive and active conditions, as passive measurements do not capture the task‐dependent mechanical demands present during contraction.

Regional differences in motor‐unit activity are plausible given known variations in innervation [45, 46] and muscle architecture along the length of the hamstrings [8], supporting partially independent functional subregions [13, 15]. Motor neurons innervating different territories may also be spatially organized within the motor pool, enabling selective recruitment by descending input and premotor neurons to bias excitation toward particular regions when necessary [47, 48, 49, 50]. From a mechanical perspective, regional activation differences may reflect variations in effective moment arms and distance from the joint axis, with regions further from the axis requiring greater activation to contribute to torque production [7]. Although animal studies support task‐dependent regional activation [51], direct evidence in humans remains limited.

4.4. Methodological Considerations

HE and KF were performed in different joint configurations to enable near‐maximal torque production in each task [7]. A knee angle of 30° was selected for KF to capture the range of peak torque during isometric contractions. Alternative configurations compromised task isolation and stability, justifying the chosen setup. However, joint angle influences muscle length, force production, and motor‐unit activity [13, 14]. Therefore, these findings are specific to the tested configurations and may not generalize to other joint positions. Nonetheless, prior evidence suggests that relative intra‐ and intermuscular differences are preserved across joint angles [13, 14], indicating that the observed patterns likely reflect underlying neural organization. Torque‐matching was used to standardize task demand, as it provides a direct measure of mechanical output and allows comparison of motor‐unit activity at equivalent relative torque levels [6, 32, 33]. In contrast, EMG‐based matching is less suitable across joint configurations due to its sensitivity to geometric and task‐dependent factors (e.g., muscle length, electrode–muscle distance, and amplitude cancellation) [12].

Several limitations should be noted. First, EMG was not recorded from other hip extensors (e.g., gluteus maximus, adductor magnus), limiting interpretation to the hamstrings. Second, the use of isolated isometric tasks does not capture the biarticular function of the hamstrings during dynamic movements. Third, coherence was assessed only at 10% MVC, restricting conclusions about shared input to low‐contraction intensities. Additionally, coherence estimates can be influenced by discharge rate, although no significant associations were found between MDR and coherence changes. Fourth, only one repetition per condition was analyzed, which may introduce selection bias, despite consistent motor‐unit yield across repetitions. Fifth, the small number of female participants limits generalizability and precludes sex‐based comparisons. Finally, the absence of an a priori power analysis warrants caution when interpreting marginal effects, although the repeated‐measures design and use of mixed‐effects models enhance statistical sensitivity.

5. Conclusion

Neural control of the hamstring muscles during submaximal isometric contractions differs between hip extension and knee flexion. The main findings indicated that hip extension is characterized by greater motor‐unit discharge rates, greater variability in discharge times and neural drive, and lower delta‐band coherence compared with knee flexion. Additionally, discharge characteristics varied between proximal and distal regions, indicating non‐uniform activity along the muscle. These findings highlight the importance of considering both task demands and regional muscle activity in neuromuscular assessments of the hamstring muscles.

6. Perspectives

Future research should determine whether the task‐ and region‐specific modulation of motor‐unit activity in the hamstrings occurs during dynamic, multi‐joint actions. Combining HD‐sEMG with imaging techniques and musculoskeletal modeling would clarify the relations between regional neural drive, local muscle–tendon mechanics, and net torque. Moreover, recording from additional synergistic muscles would enable a more comprehensive understanding of intermuscular coordination during hip extension. Furthermore, multi‐site recordings along the muscle length are recommended to characterize spatial variability in neural drive. Extending this approach to populations with altered neuromuscular function, such as individuals with prior hamstring injury or different training backgrounds, may help establish the clinical and functional relevance of regional motor‐unit activity.

Funding

Roger M. Enoka was supported by an award from the National Multiple Sclerosis Society in the USA (project RG‐2206‐39688).

Ethics Statement

This study was conducted in accordance with the ethical standards of the University ethics committee and received approval (ERC‐016/2025). All procedures involving human participants were performed following ethical guidelines, ensuring confidentiality, informed consent, and compliance with relevant regulations.

Consent

Informed consent was obtained from all individual participants included in the study. Participants were provided with comprehensive information about the study objectives, procedures, potential risks, and benefits before obtaining their consent.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

Source data for this study are not publicly available due to privacy or ethical considerations. However, the source data will be made available to verified researchers upon request by contacting C.S.