Frequency characteristics of human muscle and cortical responses evoked by noisy Achilles tendon vibration

We applied noisy (10–115 Hz) vibration to the Achilles tendon to examine the frequency characteristics of lower limb proprioceptive reflexes during standing. Ongoing muscle activity was coherent with the noisy vibration (peak coherence ~40 Hz), and coherence positively scaled with increases in stimulus amplitude. Our findings suggest that noisy tendon vibration, along with linear systems analysis, is an effective novel approach to study proprioceptive reflex actions in active muscles.

Abstract

Noisy stimuli, along with linear systems analysis, have proven to be effective for mapping functional neural connections. We explored the use of noisy (10–115 Hz) Achilles tendon vibration to examine somatosensory reflexes in the triceps surae muscles in standing healthy young adults (n = 8). We also examined the association between noisy vibration and electrical activity recorded over the sensorimotor cortex using electroencephalography. We applied 2 min of vibration and recorded ongoing muscle activity of the soleus and gastrocnemii using surface electromyography (EMG). Vibration amplitude was varied to characterize reflex scaling and to examine how different stimulus levels affected postural sway. Muscle activity from the soleus and gastrocnemii was significantly correlated with the tendon vibration across a broad frequency range (~10–80 Hz), with a peak located at ~40 Hz. Vibration-EMG coherence positively scaled with stimulus amplitude in all three muscles, with soleus displaying the strongest coupling and steepest scaling. EMG responses lagged the vibration by ~38 ms, a delay that paralleled observed response latencies to tendon taps. Vibration-evoked cortical oscillations were observed at frequencies ~40–70 Hz (peak ~54 Hz) in most subjects, a finding in line with previous reports of sensory-evoked γ-band oscillations. Further examination of the method revealed 1) accurate reflex estimates could be obtained with <60 s of low-level (root mean square = 10 m/s²) vibration; 2) responses did not habituate over 2 min of exposure; and importantly, 3) noisy vibration had a minimal influence on standing balance. Our findings suggest noisy tendon vibration is an effective novel approach to characterize somatosensory reflexes during standing.

NEW & NOTEWORTHY We applied noisy (10–115 Hz) vibration to the Achilles tendon to examine the frequency characteristics of lower limb somatosensory reflexes during standing. Ongoing muscle activity was coherent with the noisy vibration (peak coherence ~40 Hz), and coherence positively scaled with increases in stimulus amplitude. Our findings suggest that noisy tendon vibration, along with linear systems analysis, is an effective novel approach to study somatosensory reflex actions in active muscles.

charles sherrington was the first to trace the source of the stretch reflex to muscle spindles and subsequently describe reflexes as a simple expression of the interactive action of the nervous system (48). Since then, reflex excitability has been probed by delivering transient tendon taps or electrical nerve stimuli and observing the muscle response (for reviews, see Refs. 42, 51, 53). Reflex examinations have enhanced our understanding of the nervous system and have proven beneficial for diagnosing and monitoring rehabilitation efforts for many disorders associated with spasticity or other reflex abnormalities, including stroke, polyneuropathies, spinal cord injury, and cerebral palsy (21, 34, 39, 45, 52).

There are, however, some limitations associated with the assessment of reflex pathways using transient stimuli (mechanical or electrical). Generally, many stimuli need to be delivered to obtain a reliable average response (24, 40); this can be somewhat inefficient and multiple abrupt stimuli permit the influence of anticipation and habituation effects (19). Furthermore, due to the large and relatively synchronized volleys of Ia afferent input generated by a tap or nerve stimulation (6), sufficient recovery time must be left between stimuli to allow postactivation depression to dissipate (10, 25). Limitations of previous methods become particularly salient when studying reflex actions in postural muscles during tasks in which they are actively engaged such as standing and walking. During standing, additional recovery time likely must be left between stimuli to allow for recovery of balance following the postural perturbation of each stimulus. Triceps surae reflex responses are also dependent on postural sway, with increased H reflex responses observed during anterior sway and decreased responses during posterior sway (50). If sway is not accounted for during stimulus delivery, results averaged over a limited number of samples could be skewed or highly variable; this observation is supported by the finding that individuals with greater postural sway demonstrate poor reliability in reflex amplitude (44).

Suprathreshold white noise stimulation methods have proven effective for assessing the frequency characteristics of human vestibular reflexes in posturally active muscles (stochastic vestibular stimulation; Ref. 13) and could be adapted to mechanical tendon stimulation to circumvent some of the limitations of traditional tendon tap methods. In general, noisy stimuli delivered to a neural system can be used to assess connectivity through estimates of the amount of frequency variability in ongoing neural activity that can be explained by the frequency content of the stimulus (18, 28). A suprathreshold noisy tendon vibration (NTV) methodology, along with linear systems analysis, could provide researchers and clinicians with a unique tool to unobtrusively examine reflex excitability in posturally active muscles and gain insight into the frequency characteristics of stretch reflex coupling.

Previous research has probed the frequency characteristics of stretch reflexes by applying continuous sinusoidal stretches (at discrete frequencies between 10 and 50 Hz) to upper limb muscles and subsequently measuring the magnitude of EMG modulation and phase lag between the mechanical stretch and muscle response (36). The modulation strength and estimated phase delays provided insight into the operation of reflex circuitry at different frequencies, as well as provided an arguably superior measure of the delays inherent in the reflex pathway (36). The delays measured from the phase estimates take into account the envelope of the response and therefore are more representative of the “average response time of the average unit” (36), whereas the latency measured to the onset of a response evoked by transient stimuli is likely representative of only the fastest conducting axons. Sinusoidal stimuli can also shed light on motorneuron evoked response properties, such as where their response(s) lie within a given stretch cycle and how susceptible they are to coupling with stimuli close to their firing rate to produce a sharp increase in reflex gain (carrier resonance effect; Ref. 37). Although the pure sinusoidal stimulus method imparts numerous benefits, there are several notable drawbacks that include 1) response contamination from voluntary tracking of the predictable stimulus (9), 2) the development of movement illusions or tonic vibration reflex, and 3) the experimental time necessary to test a series of frequencies individually.

Stretch reflexes have also been examined using large amplitude, low-frequency (<25 Hz) pseudorandom joint perturbations (27, 31–33). In particular, low-frequency ankle joint perturbations in humans lying prone have been used to identify the ongoing relationship between ankle movement velocity and muscle activity (31, 32). Using pseudorandom stimuli, these authors identified some key differences in reflex organization between the triceps surae and tibialis anterior (31, 32). Other researchers who have used a pseudorandom muscle stretching protocol also chose to apply primarily low-frequency stimuli that are relevant to voluntary movement and within motorneuron firing rate limits (8, 27, 31–33). Human muscle spindles, however, have the capacity to respond and entrain to higher frequency stimuli (exceeding 100 Hz; Ref. 15), and strong reflex EMG modulation can be observed during high-frequency sinusoidal stretching (50 Hz; Refs. 35, 36). In addition, high-frequency components are present in impulses naturally experienced by the ankle joint, for example, during a trip; therefore, these high-frequency components have functional relevance. Thus we focused our experiment on responses to a broadband noisy mechanical stimulus that contained power up to the highest frequency that could evoke a discernable reflex response.

Similar to spinal stretch reflexes, transient electrical or mechanical stimuli are typically used to evoke cortical potentials to study the ascending transmission of sensory information (14, 17, 38). Thus these methods of evoking cortical activity are subject to similar limitations as tendon tap or H reflexes, such as the potential influence of anticipation, habituation, and postural interference. Estimates of coherence between noisy peripheral sensory input and somatosensory cortex activity may have the additional benefit of providing a useful alternative approach to probe the frequency characteristics of somatosensory-evoked cortical potentials.

The primary objective of our experiment was to explore the use of Achilles NTV to assess the frequency characteristics of triceps surae reflexes and sensorimotor cortex-evoked potentials during standing. We also aimed to examine the scaling of reflex responses to different vibration amplitudes in the soleus and medial and lateral gastrocnemius muscles. Finally, we aimed to establish trial durations that evoke consistent responses and identify whether stretch reflexes evoked by noisy mechanical stimuli are subject to habituation over 2 min of stimulus exposure.

METHODS

Participants.

Eight healthy young adults (age = 27 ± 5.3 yr, 4 male) free of musculoskeletal and neurological disorders participated. Participants provided written informed consent and all procedures were approved by the University of British Columbia Research Ethics Board.

Experimental setup.

Participants stood on a force plate (OR6–7; AMTI) with their stance width normalized to foot length and their gaze directed onto a visual target positioned at eye level ~3 m ahead. A 3-cm diameter probe, attached to a linear motor (model MT-160; Labworks), was positioned against the right Achilles tendon. The linear motor was secured onto two near-frictionless linear slides and was pulled forward onto the tendon by a weighted pulley system (Fig. 1); this setup was decoupled from the force plate and was able to maintain a constant ~1 N preload force on the tendon. A force transducer (model 31; Honeywell) was placed in line with the probe and motor piston and an accelerometer (model 220–010; X Tronics) was secured to the back of the motor piston. Acceleration and force signals were differentially amplified ( ×1 and ×100, respectively) and low-pass analogue filtered at 600 Hz (Brownlee model 440; AutoMate Scientific). All motor command signals were generated using LabVIEW 11 software and output at 5 kHz from a PXI-6225 multifunctional data acquisition board (running with a PXI-8106 real-time controller in a PXI-1031 chassis). Analogue voltage commands were sent to a motor amplifier (PA-141; Labworks) for open-loop control of tendon stimulation.

Fig. 1.

Experimental setup showing the linear motor on frictionless slides pulled forward using a weighted pulley system (A) and a zoomed in view of the probe positioned against the Achilles tendon of a participant standing on a force plate (B). Sample profile of the noisy vibration acceleration over time and the acceleration power spectrum (C). SOL, soleus muscle; MGas, medial gastrocnemius muscle; LGas, lateral gastrocnemius muscle.

Electromyography (EMG) was recorded from the soleus (SOL), medial gastrocnemius (MGas), and lateral gastrocnemius (LGas) muscles using surface electrodes positioned over the muscle bellies in bipolar arrangement (amplified ×2,000, 10-Hz highpass and 1,000-Hz lowpass filter; NeuroLog NL824 preamplifier and NL820 Isolator; Digitimer). The ground electrode was placed on the lateral malleolus. Electroencephalography (EEG) was recorded across the sensorimotor cortex using scalp ring electrodes placed over Cz (active), Fpz’ (reference), and the right mastoid process (ground) (amplified ×20,000; 1-Hz highpass and 1,000-Hz lowpass filter; GRASS P511 AC amplifier; Astromed). EEG electrode impedance levels were maintained below 5 kΩ. Ocular and facial muscle artifacts were identified and displayed to participants, and participants were subsequently instructed to minimize these behaviors during trials. Forces and moments from the force plate were amplified (×1,000–4,000) and sampled at 100 Hz, while motor voltage commands, probe acceleration and force, surface EMG, and EEG signals were sampled at 2 kHz (Power 1401 A/D board and Spike2 software; Cambridge Electronic Design).

Experimental procedures.

Participants completed a 2-min quiet stance trial and mean foot center of pressure (COP) positions were calculated in the mediolateral and anteroposterior directions to determine their neutral position. Participants began subsequent trials at their neutral position, and in the case that their COP drifted, experimenters provided verbal feedback to guide them back to neutral. Four 2-min trials of NTV were conducted where a white noise signal low-pass filtered at 100 Hz was delivered to the right Achilles tendon. In the recorded probe acceleration, power below 10 Hz and above 115 Hz was less than or equal to −13 dB (ref. peak plateau of power spectrum); a sample recorded acceleration profile and power spectrum are displayed in Fig. 1C. This stimulus bandwidth was chosen based on pilot data that demonstrated the maximum frequency that could contain significant coherence was ~90 Hz even when larger stimulus bandwidths were delivered (e.g., 10–300 Hz). The NTV was delivered at four different amplitudes (vibration root mean square accelerations: 5, 10, 15, and 20 m/s²); these amplitudes were also chosen based on pilot data that suggested these amplitudes would fall approximately within a steep, ascending portion of the reflex recruitment curve. Two additional trials of 20 tendon taps (30 Hz raised-cosine bell curve pulses at 25 m/s²; 8- to 12-s interstimulus interval) were conducted to compare the temporal characteristics of the responses elicited by noisy stimulation to the responses elicited by taps. The tendon tap and NTV trials were presented in block-randomized order.

Analyses.

Forces and moments from the force plate were digitally low pass filtered at 10 Hz (5th order dual pass Butterworth filter) and COP was calculated from moments (Mx and My) and vertical force (Fz). For tendon stimulation and quiet stance trials, the frequency spectra of COP in the anteroposterior and mediolateral directions was calculated (frequency resolution 0.0122 Hz) and mean power frequency (MPF) were determined as:

$$\text{MPF}={\displaystyle \sum }_{j=1}^{n}fjPj/{\displaystyle \sum }_{j=1}^{n}Pj$$

where f is frequency and P is power.

EMG, EEG, force, and acceleration data were digitally low pass filtered at 1,000 Hz (5th order dual pass Butterworth filter). For the tendon tap trials, EMG data were full wave rectified and EMG and EEG signals were trigger averaged to the tap stimulus onset within a window from 20 ms preceding to 300 ms following the stimulus. COP was also trigger averaged within a window from 0.5 s preceding to 4.5 s following the tap. For noisy stimulation trials, EMG data were full wave rectified and coherence analysis was performed using the NeuroSpec2.0 software package developed by Rosenberg, Halliday, and colleagues (20, 47) for MATLAB (Mathworks). Our approach was similar to that of previous research conducted to establish the time and frequency characteristics of vestibular responses elicited by stochastic stimuli (11–13). To determine the strength of the linear association between two signals in the frequency domain, coherence functions were calculated between probe acceleration (input signal) and rectified surface EMG of each muscle and EEG (output signals) (13, 20, 47). Coherence was calculated as the magnitude of the input-output signal cross spectra squared divided by the product of the input and output autospectra (13). Thus coherence values provide normative estimates of the frequency coupling strength between two signals. To identify temporal characteristics of coherent frequencies, cross covariance was calculated using the inverse Fourier transform of the input-output signal cross spectra and normalized by the product of the vector norms of the input and output signals. Therefore, cross covariance values are bounded by −1 and +1 and provide an estimate of the signal coupling strength in the time domain (12). Our convention was that acceleration toward the tendon and increased (rectified) EMG were assigned positive polarities. For example, a positive correlation would represent acceleration into the tendon is associated with increased EMG, or acceleration away from the tendon is associated with decreased EMG. As described by Halliday et al. (20), 95% confidence limits for coherence (positive threshold) and cross covariance (positive and negative thresholds) were constructed under the hypothesis of independence between the two signals. Values exceeding these limits provide evidence of a significant linear relationship between the stimulus and response. Phase was also estimated to infer the temporal relationship between the NTV and EMG at all frequencies containing significant coherence (1).

To analyze pooled responses, data were concatenated across participants and stimulus amplitudes to compare between the three muscles; this yielded a total of 3,712 disjoint sections (1.024 s/segments; frequency resolution = 0.9765 Hz). Data were also concatenated across participants and muscles for comparisons between stimulus amplitudes; this yielded a total of 2,784 disjoint segments (1.024 s/segments; frequency resolution = 0.9765 Hz). For analysis of individual 2-min trials, data were sectioned into 116 segments to obtain a frequency resolution of 0.9765 Hz (1.024 s/segments). The number of segments used in the analysis is an important factor in determining the 95% confidence limits constructed around coherence and cross-covariance traces. In addition, we divided each 2-min trial in various ways to answer two questions: 1) what is the minimum trial duration required to obtain reliable reflex measures, and 2) does the reflex response habituate over the trial? To answer the first question, 10-s portions of data (9–10 segments) were incrementally added and normative error in peak-to-peak cross covariance was calculated for each duration as:

$$\text{error}=\sum \frac{\sqrt{{(\left|{\text{C}}_{120}\right|-\left|{\text{C}}_n\right|)}^2}}{{\text{C}}_{120}}\times 100\%$$

where C₁₂₀ is cross covariance for the full trial duration and C_n is cross covariance for different trial lengths between 10–110 s in 10-s increments (4). Since our segment size was 1.024 s, each 10-s addition of data increased the number of segments by either 9 or 10 segments because incomplete segments were removed from the analysis. Signal to noise ratios were also calculated for each trial duration as peak-to-peak cross covariance divided by the width between the 95% confidence limits. To answer the second question, coherence and cross covariance were calculated and compared between the first ~40 s (39 segments, 39.936 s data used) and the last ~40 s of the trial. Comparisons of different trial durations were conducted for the SOL muscle in response to the 10 m/s² RMS acceleration noisy stimulus. We elected to probe trial duration and habituation effects in SOL since it is the most commonly studied lower limb muscle for reflex testing, and we chose the medium-low level NTV based on results that showed this subtle stimulus level evoked strong responses across all participants.

Statistics.

To examine if the tendon stimulation affected the frequency content of postural sway, we performed a one-way repeated-measures ANOVA (5 levels: 4 vibration amplitudes and no vibration) on anteroposterior and mediolateral COP MPF. To examine how reflex responses scaled with stimulus amplitude, we conducted a two-way (muscle × NTV amplitude) repeated-measures ANOVA on peak-to-peak cross covariance. Significant ANOVA effects were followed up with Fisher least significant difference post hoc comparisons. Pooled subject data were concatenated across stimulus amplitudes and a χ² extended difference of coherence (DOC) test was performed to determine whether the independent coherence estimates significantly differed between muscles (1). Similarly, pooled subject data were concatenated across muscles and a χ² extended DOC test was performed to determine if coherence significantly differed between stimulus amplitudes. Significant main effects were followed up with pairwise DOC tests between successive stimulus amplitudes and between each muscle combination. Finally, to determine if the SOL muscle response to NTV habituated throughout the trial, we compared peak-to-peak cross covariance in the first 40 s of the trial to last 40 s of the trial using a paired t-test. Effects were considered significant at an α-level of 0.05, all error bars demonstrate standard error (n = 8).

RESULTS

No participants reported illusory movements in response to the NTV (even when asked to close their eyes) nor did they report any noticeable interference with standing balance. Participants naturally maintained their COP around their neutral position and verbal feedback to correct postural drift was only necessary for two subjects. MPF of COP in both the anteroposterior and mediolateral directions were not affected by the noisy vibration (P = 0.583 and 0.773, respectively; Fig. 2). Perturbations were observed in participants’ COP traces in response to tendon taps; triggered-average anteroposterior COP demonstrated that the taps evoked a directional postural response that required several seconds for recovery (Fig. 2C). Meanwhile, the noisy stimulus did not produce any noticeable change in COP relative to quiet stance.

Fig. 2.

Mean power frequency of center of pressure (COP) in the anteroposterior (A) and mediolateral (B) directions across the four stimulus amplitudes and quiet stance (0 stimulus) trials. Representative traces of anteroposterior (AP) COP during the noisy tendon vibration, quiet stance, and tendon tap trials, as well as a representative trace of AP COP trigger-averaged to the tap stimulus (C).

NTV-muscular coherence.

In all participants, significant (exceeding 95% confidence limits) muscle responses were observed in the SOL and MGas EMG in the frequency (coherence) and time (cross covariance) domains for all NTV stimulus amplitudes. Responses were characterized by a significant coherence band generally within ~10–80 Hz (Fig. 3), with the peak response observed at ~40 Hz in all muscles (SOL 36 ± 4 Hz; MGas 39 ± 5 Hz; LGas 47 ± 9 Hz). Background activity in the LGas muscle during standing was low or absent in the majority of participants, and significant reflex responses were not always observed in LGas EMG, particularly at the lower levels of stimulation (absent in 50% of participants). For all three muscles, the slope of the phase estimate was generally linear, indicating a fixed reflex delay across frequencies (Fig. 4). The magnitude of the slope corresponded to approximately a 40-ms delay. There was, however, a small upward deflection in the phase estimates, accompanied by a reduction in the coherence strength, at ~20 Hz.

Fig. 3.

Representative traces of stimulus triggered-average muscular and cortical responses to tendon taps (A); data are shown for both unrectified and rectified EMG. Representative traces of coherence and cross covariance between the stimulus acceleration and triceps surae EMG and sensorimotor cortex EEG for the 10 m/s² noise stimulus (B). Horizontal lines indicate 95% confidence intervals. SOL, soleus muscle; MGas, medial gastrocnemius muscle; LGas, lateral gastrocnemius muscle.

Fig. 4.

Coherence (A) and phase estimates (B) for each muscle for data pooled across participants and stimulus amplitudes. SOL, soleus muscle; MGas, medial gastrocnemius muscle; LGas, lateral gastrocnemius muscle.

The NTV-EMG cross covariance displayed a short-latency biphasic profile with a positive peak followed by a trough. For the SOL muscle, the first peak in the cross covariance occurred at a lag of 38.1 ± 0.8 ms (range = 34 to 42 ms), and similar lag times were observed for the MGas (38.5 ± 1.9 ms) and LGas (38.4 ± 3.1 ms) muscles. Lag times observed in the cross covariance generally corresponded to the latencies of the first peak stimulus-triggered average response to tendon taps (SOL = 41.3 ± 2.4 ms; MGas = 40.5 ± 2.5 ms; LGas = 41.3.7 ± 1.6 ms; Fig. 3).

Peak-to-peak cross covariance positively scaled with NTV amplitude (Fig. 5); statistically there was a significant main effect of stimulus amplitude on peak-to-peak cross covariance [F_(3,21) = 13.135, P < 0.001], main effect of muscle [F_(2,14) = 7.464, P = 0.006], and muscle × stimulus amplitude interaction [F_(6,42) = 2.464, P = 0.039]. These results indicate that overall the SOL muscle had stronger coupling with the noisy stimulus and scaled more with increases in the stimulus amplitude. In contrast, LGas demonstrated the weakest coupling and the shallowest rate of increase with stimulus amplitude. Post hoc comparisons revealed significant overall differences in peak-to-peak cross covariance between SOL and LGas at all four stimulus amplitudes (5, 10, 15, and 20 m/s²; P range 0.009–0.016) and between LGas and MGas at the 5 m/s² (P = 0.041) and 10 m/s² (P = 0.025) stimulus amplitudes.

Fig. 5.

Noisy tendon stimulation results for the triceps surae muscles showing increases in mean peak-to-peak cross covariance (A) in response to increases in tendon stimulation amplitude. Sample cross covariance traces (B) and coherence traces (C) from the SOL muscle of one participant showing the increases with stimulus amplitude. SOL, soleus muscle; MGas, medial gastrocnemius muscle; LGas, lateral gastrocnemius muscle.

Results for pooled data revealed a general increase in coherence with increases in stimulus amplitude (Fig. 6A). The χ² extended DOC test demonstrated a significant effect of the stimulus amplitude on coherence at frequencies between ~10 and 60 Hz (Fig. 6B). Pairwise DOC tests indicated significant increases in coherence between the 5 and 10 m/s² stimulus amplitudes at frequencies ~20–40 Hz, and significant increases in coherence between the 15 and 20 m/s² stimulus amplitudes at frequencies ~10–30 Hz. Pooled data also revealed generally stronger NTV-EMG coherence for the SOL and MGas muscle compared with the LGas (Fig. 6C). The χ² extended DOC test indicated a significant effect of the muscle on coherence at frequencies between ~20–70 Hz (Fig. 6D). Pairwise DOC tests demonstrated significantly higher coherence in SOL compared with both MGas and LGas at frequencies ~30–50 Hz and significantly higher coherence in MGas compared with LGas at frequencies ~50–70 Hz.

Fig. 6.

Results from data concatenated across participants and muscles to demonstrate overall coherence at each stimulus amplitude (A) as well as pooled difference of coherence (DOC) results across stimulus amplitudes, and pairwise DOC results between successive stimulus amplitudes (B). Results from data concatenated across stimulus amplitudes to demonstrate overall coherence for each muscle (C) as well as pooled DOC results across muscles, and pairwise DOC results between each muscle (D). SOL, soleus muscle; MGas, medial gastrocnemius muscle; LGas, lateral gastrocnemius muscle.

NTV-cortical coherence.

One participant’s EEG data were excluded due to facial muscle and blink artifacts. Clear stimulus triggered-average-evoked potentials were observed in the EEG recording across the sensorimotor cortex in all remaining seven participants in response to tendon taps. However, significant NTV-EEG coherence was only observed in five out of the seven participants, and coherence was often absent during the lower stimulus amplitudes (5 and 10 m/s²). Significant coherence was observed within a frequency range of ~40–70 Hz (within the γ-band), with the peak coherence located at 54 ± 8 Hz (Fig. 3).

The peak EEG response occurred at a slightly longer lag relative to muscle responses, with the trough observed at 50 ± 6 ms and peak observed at 53 ± 6 ms; these latencies generally correspond to early event related potentials from tendon taps in our experiment (peak 47.8 ± 4.5 ms; trough 56.2 ± 8.7) and previous experiments (14). EEG responses were subtle relative to muscle responses and characterized by multiple peaks and troughs. Interestingly, there was no prominent EEG activity in the cross covariance at longer lag times (e.g., ~100 ms and later) that would correspond to later stages of sensory processing.

Trial duration and habituation.

Two-minute trials of the 10 m/s² NTV were subdivided into 10-s additive sections to determine the minimum trial duration necessary to obtain reasonable reflex estimates in the SOL muscle. With ~10 s of data collection (9 segments), mean normative error in the peak-to-peak cross covariance was high (~40%) and there was a large amount of between-participant variability in normative error (Fig. 7). As expected, confidence intervals and background noise decreased as the trial duration (and number of segments used in the analysis) increased. The decline in the mean and variability of normative error with each addition of ~10 s of data began to plateau at the ~40-s data length, signifying the reflex responses measured with ~40 s of data collection approximated responses measured with 2 min of data collection (10% difference). In addition, the signal to noise ratio increased with each addition of data segments, and this increase was steeper between 10 and 40 s (Fig. 7). The signal to noise ratio approximately doubled between ~10 and ~40 s of data, while further increases in data length up to ~120 s only resulted in an increase in the signal to noise ratio by another 1.7-fold.

Fig. 7.

Signal to noise ratios (A) and normative error in peak-to-peak cross covariance between each data length and the full 120-s trial length (B). Mean peak-to-peak cross covariance calculated from the first 40 s vs. the last 40 s of the trial (C).

There was no evidence that the peak-to-peak cross-covariance amplitude differed between the beginning and end of the trial (P = 0.417; Fig. 7C), indicating no significant habituation to the NTV over 2 min of stimulus exposure.

DISCUSSION

The primary objective of our experiment was to examine the frequency responses of triceps surae EMG, and sensorimotor cortex EEG, evoked by noisy (10–115 Hz) Achilles tendon vibration in standing participants. Our results showed surface EMG was significantly coherent with NTV across a broad frequency range, with strong responses observed in the SOL and MGas muscles. Our results also demonstrate that reasonable SOL coherence estimates can be obtained with minimal perturbation to standing balance, and without habituation effects.

NTV-evoked muscle responses.

We observed significant coherence between the NTV and surface EMG that extended from ~10–80 Hz. This bandwidth falls within the muscle spindle vibration sensitivity range (15), but exceeds the upper limit of human individual motor unit firing rate capabilities (maximal soleus motor unit rates ~20 Hz; Ref. 29). In spite of known homosynaptic postactivation depression and motorneuron afterhyperpolarization effects, high-frequency sine wave vibration bursts (3 cycles at 100 Hz) have previously been shown to generate three distinct reflex responses in SOL surface EMG (16). EMG responses to high-frequency vibration likely reflect that high-frequency spindle input can effectively shift the firing probability of individual triceps surae motor units.

The profile of the reflex response evoked by NTV in the time domain exhibited a short latency peak followed by a trough; the lag time between the NTV and muscle response (~38 ms) approximately corresponds to the latency of the T reflex in the lower limb observed in our experiment (~41 ms) as well as in previous experiments (24, 40, 55). Similar to our tap-evoked responses, there was no evidence of responses in the cross covariance at longer delays that could parallel the medium or long latency responses observed in ramp-and-hold stretch reflexes (56). Therefore, we believe our NTV method can be used to characterize the functional short latency reflex coupling between type Ia spindle afferents and lower limb motorneurons. It has previously been shown that an adapted trigger-averaging technique used with sinusoidal stimuli allowed for the identification of a shorter reflex onset latency compared with cross-correlation lag times (30). However, there was no difference between the trigger-averaged latency and cross-correlation lag time measured to the peak of the response (30), as was done in our experiment. This suggests that trigger averaging to either transient or sinusoidal stimuli could provide clearer identification of the earliest reflex onset latency (mediated by the fastest conducting axons) compared with correlational techniques. However, trigger-averaging techniques do not provide important information about the frequency characteristics of responses, which are obtained through estimates of coherence, gain, and phase using the framework developed by Halliday and colleagues (1, 20, 47).

The slopes of the phase estimates for each muscle were linear and suggested a relatively conserved stimulus-response delay of ~40 ms across frequencies containing significant coherence. There was, however, a slight upward deflection in the phase slope at ~20 Hz, accompanied by a discontinuity in the coherence plot at ~20 Hz. This 20-Hz phenomenon was observed in all three muscles, although it was more prominent at higher stimulus amplitudes and in SOL. This pattern of a decrease in EMG modulation with sinusoidal stimuli around 20–25 Hz, accompanied by an upward deflection in the phase estimate, has previously been observed in the flexor carpi radialis (FCR) muscle (35). Matthews (35) suggested that this pattern could arise from the interference between two different latency reflex responses (which traveled through pathways that impart different delays), since the sum of the two responses would create a phase intermediate between them. The appearance of a longer latency interfering response in the FCR was suggested to have a functional role in stabilizing the system to tremor or clonus (35). The interference pattern that we observed in the triceps surae muscles at ~20 Hz requires more detailed investigation.

The amount of EMG that could be explained by the tendon vibration increased as we increased the amplitude of the NTV, and the overall reflex response and amplitude scaling were strongest in the SOL and weakest in the LGas muscle. DOC tests also demonstrated stronger NTV coupling in the SOL muscle ~30–50 Hz compared with MGas and LGas. Response differences within the triceps surae group could reflect differences in their spindle density, where SOL houses a higher overall number (~400) and density (0.94 spindles/g) of muscle spindles compared with the gastrocnemius muscles (~150 spindles total, 0.4 spindles/g; Refs. 2, 54). SOL muscle spindles might also experience a higher stimulus intensity compared with gastrocnemius spindles due to dampening of the vibration as it travels through tissue. Reflex coupling strengths could also reflect differences in their relative contributions to standing balance, where SOL generally provides the majority of plantarflexion torque while LGas remains relatively silent (23). In addition, the higher proportion of slow twitch fibers in the SOL might favor stronger reflex coupling since animal studies have shown higher efficacy of Ia input to low-threshold motorneurons in a way that accentuates orderly recruitment (22).

Postural and perceptual effects.

Across the four stimulus intensities, ~17–34% of the SOL EMG variability could be explained by the tendon stimulation. Despite this strong reflex coupling, there was no notable interference with posture, illusions of forward sway (i.e., illusory muscle lengthening) or presence of a tonic vibration reflex (TVR). Compared with sine wave vibration, the frequency variability of our noisy stimulus (10–115 Hz) seems to impede the generation of illusory movement and TVRs. We speculate that the frequency variability precludes the stimulus from producing a ramping reflex contraction. The absence of any illusory movement or TVR suggests NTV is a more suitable method for the continuous assessment of reflex excitability. Additionally, tendon taps produce a strong unidirectional postural response, whereas noisy stimulation produces a subtle and more bidirectional postural response; thus NTV circumvents some of the limitations of traditional tendon tap methods. The absence of low-frequency content in the NTV likely causes less interference with postural sway, similar to observations from stochastic vestibular stimulation (12).

Trial duration and reflex habituation.

With the use of the RMS 10 m/s² NTV, reasonably accurate reflex excitability estimates could be obtained with a minimum of ~40 s of data collection; the response amplitudes measured with ~40 s of data collection were 10% different from those measured with 2 min of data collection. T-reflex responses between successive stimuli are inherently variable (24, 40); therefore, our method of sampling the ongoing association between the stimulus and muscle activity over a sufficient window (e.g., 40–60 s) has advantages over traditional approaches (e.g., T and H reflex) that only provide discrete snapshots of reflex excitability.

There was no indication that reflex responses to NTV habituated with continuous stimulus exposure over 2 min. Therefore, measurements obtained using this method are not affected by habituation within the intensity levels and durations necessary for reflex testing.

NTV-evoked cortical potentials.

Although tendon tap-evoked potentials were present in all participants throughout the experiment, NTV-cortical coherence was absent in two out of seven participants, and in two of the remaining participants it was only prominent during the highest stimulus amplitude (20 m/s²). When significant coherence was present, NTV-EEG peak coherence values were small and scaled with the stimulus amplitude from r² ~0.04 to 0.07. Our success rate (71%) in extracting NTV-EEG coherence is similar to the success rate previously reported for extracting corticomuscular coherence during voluntary muscle contraction (50–75%; Refs. 41, 43, 46). When present, triceps surae corticomuscular coherence values are also modest (r² < 0.06; Refs. 41, 43). The absence of strong NTV-EEG coherence could be due to nonlinearities between the stimulus and evoked cortical activity and the absence of low-frequency power to associate with the longer latency cortical-evoked responses.

We observed NTV-EEG coherence within the frequency range of ~40–80 Hz, with the peak located at 54 Hz on average. This range corresponds to the γ-band oscillations recorded over the somatosensory cortex induced by mechanical or electrical nerve stimuli (3, 26). Specifically, magnetoencephalography (MEG) recordings have shown that tactile stimulation of the finger evokes γ-oscillations (concentrated around 60–90 Hz) over the sensorimotor cortex ~40–100 ms after stimulus presentation (3). Attention directed toward the spatial features of the stimulus further enhanced these γ-oscillations, which source analysis suggested originated from the primary somatosensory cortex (3). It has been suggested that γ-band oscillations in the somatosensory cortex reflect early stages of processing functionally relevant sensory information and that it is crucial in communicating with other somatosensory areas for higher level processing.

The profile of the NTV-cortical cross covariance was oscillatory, and the initial trough lagged the NTV by ~50 ms. This short delay is generally in alignment with the latency of early event related potentials recorded over the somatosensory cortex in response to mechanical stimuli (5, 14, 49). This NTV-cortical lag time specifically approximates the latencies of the Achilles tendon tap-evoked potentials observed in this experiment as well as in previous experiments (14).

Methodological considerations.

There are some important considerations with regards to the assessment of reflexes using stimuli applied to the tendon. Tendon tap reflex responses (and likely NTV responses by extension) might not reflect the strength of direct monosynaptic connections between triceps surae spindles and motorneurons for several reasons. First, in addition to targeting triceps surae spindles, tendon taps have been shown to evoke multiple spikes in Ia afferent fibers innervating extensor hallucis longus, tibialis posterior, and intrinsic muscles of the foot (6). Tendon taps have also been shown to alter the discharge of some type II spindle afferents as well as Golgi tendon organ and skin afferents (6). Despite this, however, it appears to be the input from type Ia spindle afferents that accounts for the motor response (6). Second, the rise time of composite excitatory postsynaptic potentials is broad enough to permit time for oligosynaptic pathways to contribute to the response; thus it should be considered that these methods might not strictly assess monosynaptic reflex strength (7). It should also be noted that similar limitations regarding the purity of the afferent stimulus and spinal connections tested are present with direct nerve stimulation (H reflex; Refs. 6, 7). In addition, direct nerve stimulation bypasses natural mechanotransduction and generates very artificial, synchronized nerve impulses. Our noisy stimulation methodology has the advantage of mechanically stimulating receptors themselves within a physiological range and subsequently providing temporal and frequency information about somatosensory projections to muscle and to the cortex. Further exploration of tendon tap reflex responses of single motor units during standing using frequency and probability based measures (e.g., peristimulus time histograms and peri-stimulus frequencygrams), along with coherence between sensory and motor spike trains, is necessary to more fully understand the characteristics of the pathways that contribute to tendon tap and noisy vibration reflexes in humans while standing.

Conclusions.

Our findings indicate that noisy vibration of the Achilles tendon is an effective novel approach to study somatosensory reflexes in lower limb muscles during standing. Additionally, NTV-muscular coherence can shed light on short latency communication between sensory receptors and the motorneuron pool across frequencies. Our findings also show promise for the use of NTV to concomitantly assess somatosensory related cortical activity, although this requires further investigation. These NTV methods could enhance researchers’ and clinicians’ ability to assess reflexes in posturally active muscles efficiently and with minimal interference with standing balance.

GRANTS

This work funded by the Natural Sciences and Engineering Research Council of Canada (NSERC) grants (to M. G. Carpenter, J.-S. Blouin, and J. T. Inglis). J.-S. Blouin also received support from the Canadian Institutes of Health Research-Canadian Chiropractic Research Foundation and Michael Smith Foundation for Health Research. R. M. Peters received salary support from NSERC funding granted to J. T. Inglis, and R. L. Mildren also received financial support through an NSERC doctoral research award. A. J. Hill received internal funding from a Work Learn International Undergraduate Summer Research Award.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

R.L.M. and A.J.H. performed experiments; R.L.M. analyzed data; R.L.M., R.M.P., J.-S.B., M.G.C., and J.T.I. interpreted results of experiments; R.L.M. and A.J.H. prepared figures; R.L.M. drafted manuscript; R.L.M., R.M.P., A.J.H., J.-S.B., M.G.C., and J.T.I. edited and revised manuscript; R.L.M., R.M.P., A.J.H., J.-S.B., M.G.C., and J.T.I. approved final version of manuscript.