Mean EDA, expressed in microSiemens (μS), was extracted during time windows encompassing the NTV and electrical stimulation periods. Force plate signals were digitally low pass filtered (dual-pass) at 10 Hz (4th order low-pass Butterworth filter) and COP was calculated from moments (Mx and My) and vertical force (Fz). The root mean squared (RMS) displacement and mean power frequency (MPF) of COP in the anteroposterior (AP) and mediolateral (ML) directions were calculated during each trial. EMG data were DC removed and full wave rectified for NTV analysis. Signals were digitally filtered at 1,000 Hz (dual-pass, 4th order, low-pass Butterworth filter). Background muscle activity was determined for the SOL, MG, and TA during H-reflex and NTV assessment by extracting the RMS amplitude during time windows encompassing the stimulation period plus one second on either end. For H-reflex assessment in the SOL, EMG signals were trigger-averaged to the stimulus delivery onset within a window of 10 ms preceding to 100 ms following tibial nerve stimulation. This resulted in a trace representing the mean response from the five H-waves evoked during each condition trial. Values were extracted for stimulus-triggered average peak-to-peak amplitude and latency from the stimulus to the first peak in the response wave (Figure 2). For NTV assessment in the SOL, coherence analysis was performed using the NeuroSpec2.0 software package (Rosenberg et al., 1989; Halliday et al., 1995) for MATLAB (Mathworks). This approach was adopted from prior studies investigating the time and frequency characteristics of stochastic stimuli-evoked physiological responses (Dakin et al., 2007, 2010, 2011; Mildren et al., 2017). Coherence values represent normative estimates of the frequency coupling strength between two signals (Mildren et al., 2017). Thus, coherence functions were used to assess the strength of the linear relationship between the input (NTV magnitude) and the output (SOL EMG) signals in the frequency domain. Coherence was estimated as the magnitude of the input-output signal cross spectra squared divided by the product of the input and output autospectra (Dakin et al., 2007; Mildren et al., 2017). Temporal characteristics of coherent frequencies were detected by using the inverse Fourier transform of the input-output signal cross spectra to compute cross-covariance (cumulant density), which was normalized by the product of the vector norms of the input and output signals. Resulting cross covariance values were normalized to a range of −1 to 1 (Dakin et al., 2010) and approximate the signal coupling strength in the time domain. Therefore, a positive correlation could be produced by either increased probe magnitude being coupled with an increase in EMG, or by decreased probe magnitude being coupled with a decrease in EMG. Independence of the two signals (input and output) are assumed under the constructed 95% confidence limits for coherence (positive threshold) and cross covariance (positive and negative thresholds) (Halliday et al., 1995). Values exceeding these limits are indicative of a significant linear relationship between the input (stimulus) and output (response) signals. Coherence spectra and cumulant densities were computed for each individual participant, as well as by pooling the data from each condition across all 20 participants (FFT windows = 1s; frequency resolution 1 Hz). Peak-to-peak and the timing of the cumulant density peak was extracted for statistical analysis. Additionally, we conducted pairwise Difference of Coherence tests (Amjad et al., 1997) to determine significant differences in pooled coherence spectra (see Figure 3).

Analysis was performed using custom MATLAB (MathWorks, MA, USA) and SPSS software (version 26.0; IBM SPSS Statistics Inc., Chicago, Il, USA) with α set to 5%. General linear mixed effects models were used to determine the main effect of visual condition on dependent variables. This statistical model provided comparisons between all conditions in real (REO and REC) and virtual (VRL and VRH) environments with pairwise estimates using sequential Bonferroni adjustments for multiple comparisons. Dependent variables in both real and VR environments included mean EDA (assessed independently during NTV and electrical stimulation), mean rectified background root mean squared (RMS) EMG in each muscle (SOL, MG, and TA), RMS and mean power frequency (MPF) of the anterior-posterior (AP) and medial-lateral (ML) COP, H-reflex peak-to-peak amplitude in the SOL, and NTV-reflex cumulant density peak-to-peak amplitude in the SOL. Pairwise difference of coherence tests compared coupling strength as a function of frequency for the identical planned comparisons. Subjective balance confidence, fear, stability, and anxiety data from questionnaires were assessed for VRL versus VRH using (one-tailed) paired t-tests. Mean EDA during each type of stimulus delivery (electrical vs. NTV) was compared for each condition using paired t-tests to determine if participant arousal was consistent during H-reflex and NTV-reflex assessment. Cohen’s d effect sizes were calculated, and magnitude of the effects may be interpreted as small (ES = 0.2), medium (ES = 0.5), or large (ES = 0.8) (Cohen, 1988).

Force plate analysis was conducted using data from 19 subjects; due to technical issues, these data were unavailable for one participant. Data from one participant was also excluded from the final H-Reflex analysis due to equipment malfunction. Of the remaining 19 participants, all had strong SOL muscle responses to H-reflex, while a weaker or absent response was observed in MG (18 responders) and TA (15 responders) muscles. Similarly, all 20 participants showed strong SOL and MG reflex responses (coherence) during NTV, while none definitively produced reflex responses in the TA. Since SOL consistently demonstrated the strongest response and was present in all individuals during H-reflex and NTV delivery, we focused the analysis exclusively on SOL. This approach is aligned with prior H-Reflex (Horslen et al., 2013; Young et al., 2018) and mechanical tendon stimulation (Horslen et al., 2013, 2018; Mildren et al., 2016, 2017) studies which focused on SOL for analysis.

There were significant main effects for condition on EDA during both H-reflex assessment (F(3, 76) = 6.769; p < 0.001) and NTV (F(3, 76) = 7.003; p < 0.001). Pairwise contrasts, adjusted for multiple comparisons using the sequential Bonferroni correction, determined differences between specific conditions. As shown in Figure 2A, during NTV, significantly greater EDA was observed in the VRH condition compared to REO (41.6%; t(76) = 3.557; p = 0.003; CI: 1.045 – 7.08), REC (51.9%; t(76) = 4.126; p = 0.001; CI: 1.63 – 7.818), and VRL (38.4%; t(76) = 3.358; p = 0.005; CI: 0.912 – 6.756). There were no significant differences in EDA across stimulation method (electrical vs. NTV) for any condition. During electrical stimulation (H-reflex assessment), there was a similar increase in sympathetic arousal in the VRH condition compared to REO (39.6%; t(76) = 3.741; p = 0.002; CI: 1.324 – 7.692), REC (40.9%; t(76) = 3.826; p = 0.002; CI: 1.346 – 7.875), and VRL (35.1%; t(76) = 3.422; p = 0.004; CI: 1.041 – 7.207).

Questionnaires were administered prior to and following VR conditions only. We observed significant mean differences between the high and low conditions across all psychosocial measures. Compared to the VRL condition, participants reported significantly lower confidence in their ability to maintain balance prior to VRH (14.3%; t(19) = 2.63, p = 0.008, ES = 0.588). Similarly, during the VRH versus the VRL condition, participants felt more unstable (11.5%; t(19) = 2.87, p = 0.005, ES = 0.642) and more fearful of falling (17.3%; t(19) = −2.89, p = 0.005, ES = 0.646), while experiencing greater levels of anxiety (10.6%; t(319) = −7.3, p < 0.001, ES = 0.408; Figure 2B).

A general pattern of increased muscle activity emerged within the VR environment, most notably in VRH, in each muscle measured. Soleus: In SOL, there was a significant main effect for condition on background muscle activity (F(3, 76) = 2.832; p = 0.044). Pairwise contrasts, adjusted for multiple comparisons using the sequential Bonferroni correction, did not identify any significant difference between any conditions; however, non-significant trends for increased EMG were observed in the VR environment, with both VRL (3.8%; t(76) = 2.49; p = 0.075; CI: 0.000 – 0.004) and VRH (3.9%; t(76) = 2.557; p = 0.075; CI: 0.000 – 0.004) approaching significance compared to baseline (REO) (Figure 3). Medial Gastrocnemius: In MG, there was a significant main effect for condition on background muscle activity (F(3, 76) = 2.962; p = 0.037). Pairwise contrasts demonstrated an increase in EMG during VRH compared to baseline (REO) (10.4%; t(76) = 2.826; p = 0.036; CI: 0.000 – 0.007). Tibialis Anterior: Similarly, there was a significant main effect for condition on background muscle activity in TA (F(3, 76) = 3.151; p = 0.030). Pairwise contrasts demonstrated an increase in EMG during VRH compared to baseline REO (6.5%; t(76) = 2.953; p = 0.025; CI: 0.000 – 0.004).

Standing at virtual height impacted postural sway characteristics in the AP direction, indicated by a significant main effect for condition on AP COP MPF (F(3, 72) = 4.605; p = 0.005). Increases in AP COP MPF versus baseline (REO) were significant during the VRH condition only (31.1%; t(72) = 3.693, p = 0.003, CI: 0.021 – 0.135). There were no differences in COP RMS in the AP or ML direction, or in ML COP MPF, between any conditions. Figure 3 illustrates COP responses across conditions.

There was a significant main effect for condition on H-reflex P2P amplitude (F(72) = 13.829; p < 0.001). Significant pairwise contrasts demonstrate patterns of reduced H-reflex P2P amplitude within the virtual environment versus during real world conditions. Indeed, both VRL (40.8%; t(72) = −4.934, p < 0.001; CI: −1.315 – −0.397) and VRH (43.4%; t(72) = −5.247, p < 0.001; CI: −1.381 – −0.44) evoked smaller H-reflexes compared to baseline (REO). Similarly, H-reflex P2P amplitude was also reduced compared to REC during both VRL (34%; t(72) = −3.676, p = 0.001; CI: −1.063 – −0.212) and VRH (36.9%; t(72) = −3.988, p = 0.001; CI: −1.136 – −0.247). There was no difference in H-reflex peak-to-peak amplitude between VR height conditions (t(72) = −0.312, p = 0.756; CI: −0.4 – 0.292). A mean latency of 36 ms was observed, defined as the time between the stimulus delivery and the initial peak of the bi-phasic reflex wave (Figure 4).

A mean latency of 44.3 ms was observed, defined as the time between the stimulus delivery and the initial peak of the bi-phasic reflex wave. However, there were no significant condition effects on NTV-reflex P2P amplitudes. However, pairwise difference of coherence tests found that coherence between NTV magnitude and SOL muscle activity increased for a subset of frequency components (see Figure 5) for VRH vs. VRL, VRL vs. REO, and VRH vs. REO comparisons.

The primary objective of this study was to investigate adaptations in arousal, postural control, and muscle stretch reflexes under different conditions of real and virtual visual feedback. We observed that mechanically evoked muscle stretch reflex amplitude remained unchanged across visual conditions, however, there were significant reductions in H-reflex amplitude in virtual reality. This phenomenon occurred not only when exposed to virtual height-induced threat, but also while standing at virtual ground level (VRL), a condition mimicking the REO baseline condition. Thus, merely altering visual input from a real setting to a virtual environment appears to alter the excitability of spinal reflex pathways. This lends support to existing literature suggesting that a specific adaptation may occur within the muscle stretch reflex circuity, such that, in VR, there is a global decrease in neuro-excitability at the level of the spine. Our results also suggest there may be a concomitant increase in muscle spindle sensitivity, which offsets this decrease in spinal excitability, keeping mechanically evoked muscle stretch reflexes unchanged (or even slightly increased) in amplitude. To appropriately address our primary research objective, this study also aimed to elicit physiological and psychological responses to postural threat using VR. A physiological arousal response to postural threat was observed, with significant EDA increases occurring with transition to VRH from baseline (REO), REC, and VRL conditions. Similarly, subjective anxiety and fear of falling increased, while balance confidence decreased in the VRH condition compared to VRL. Postural muscles of the lower limb also demonstrated increased co-contraction, which reached significance when transitioning to the VR height environment, likely explaining the increase in the mean power frequency of postural sway observed during this condition.

Significant changes in physiological and psychological responses in the virtual height condition compared to ground level are indicative of increased sympathetic arousal. Skin conductance (EDA) in the VRH condition was elevated compared to VRL, and concomitant reductions in perceived stability and balance confidence, as well as increased fear of falling were also documented. These findings support prior indications that VR height simulations can stimulate sympathetic arousal such as increased skin conductance, heart rate, and perceived anxiety and fear (Meehan et al., 2002; Cleworth et al., 2012) to a similar degree to real-world elevation. Additionally, no differences were observed in EDA between the real (REO) and virtual (VRL) baseline conditions. This indication that transitioning into VR does not increase participant arousal has been previously reported (Simeonov et al., 2008; Cleworth et al., 2012). The invariability in arousal between REO and VRL suggests that participants were not exposed to physiological stress during this transition, alluding that any associated changes in reflex excitability may be attributed to visual feedback rather than perceived threat. Overall, this study was successful in using VR to generate measurable physiological and psychosocial changes in sympathetic arousal.

Here we observed a general pattern of increased muscle activity with the addition of virtual reality induced threat. Patterns of increased muscle activity in VR were significant during the height condition (VRH) compared to baseline (REO) for medial gastrocnemius and tibialis anterior. Increases in activity for these muscles at virtual height is consistent with the established increase in lower-limb co-contraction observed in real-world height induced postural threat. Simultaneously, postural sway frequency was also affected by VR; compared to baseline (REO), AP MPF increased in the VRH condition, indicative of a tensing response at the ankle joint. This likely represents a neural adaptation strategy to reduce instability (maintain balance) under conditions where the risk or consequence of a fall is heightened. A linear increase in COP frequency with height-induced threat has been previously reported (Adkin et al., 2000). In real height environments, increased muscular co-contraction alongside greater frequency of AP sway is observed.

Previous studies, which directly compared standing at ground level and height in both real and VR environments did not report significant differences in postural control between matched real and virtual conditions when standing on a firm surface (Simeonov et al., 2005; Cleworth et al., 2012), although Simeonov et al. (2005) did observe increased postural sway when standing on an unstable surface in VR versus a real environment for both ground level and height conditions. However, VR has been reported to increase postural instability during quiet stance on both firm and foam surfaces compared to similar real-world conditions; in fact, Horlings et al. (2009) observed that when compared to real-world quiet standing, the effects of VR on postural sway were similar to an eyes closed condition, consistent with the current study, which found no change in postural control between REC and VRL. It is possible that differences in technology or virtual visual environment account for the contradicting findings regarding postural control in VR. For example, the former studies created VR scenes that very closely matched their real-world laboratory (Simeonov et al., 2005; Cleworth et al., 2012), while Horlings et al. (2009) and the current study did not match the VR scene to the real-world eyes open condition. Participant VR experience may have also played a role in this discrepancy. While other studies do not report on this, participants entered the current study with very limited prior exposure to VR. It is possible that the novel nature of the VR experience could affect the results.

Other studies also did not report on changes in muscle activity in VR; however, the changes in postural control in this study are supported by the associated increase in muscular co-contraction described above. A review by Adkin and Carpenter (2018) identified eight studies which investigated height-induced postural and neurophysiological outcomes. A pattern emerged in the literature that is supported by the present research – each study reported evidence of increased COP frequency (Adkin and Carpenter, 2018). These previous studies also reported a reduction (6 studies) or no change (2 studies) in sway amplitude when threatened with height (Adkin and Carpenter, 2018). Recent research has substantiated these findings of altered balance control in response to standing in virtual elevation. Raffegeau et al. (2020) recorded postural deviations using inertial sensors while subjects were elevated in VR from ground level to 15 m height. The authors reported that AP postural responses (sway acceleration) were significantly greater at height, but only when standing parallel to the threat (i.e., on the edge of the platform compared to standing perpendicular to the edge).

We observed that vibration evoked stretch reflexes remained unchanged, or marginally increased with virtual reality induced postural threat. Given that H-reflexes, and presumably, spinal excitability was diminished in virtual reality, our NTV results suggest that a compensatory neural mechanism is at play to maintain nearly constant mechanically evoked stretch responses. Indeed, in conditions where H-reflexes were reduced, NTV reflexes remained unchanged or slightly increased on average. This is consistent with previous reports of threat-induced modulations to postural control. Indeed, T-reflex amplification has been demonstrated to occur during static (Davis et al., 2011; Horslen et al., 2013) and dynamic reactive (Horslen et al., 2018) standing at height, as well as alternative sources of arousing stimuli (Bonnet et al., 1995; Both et al., 2003, 2005; Hjortskov et al., 2005; Kamibayashi et al., 2009).

As mentioned, an opposing trend to vibration reflex responses emerged for electrically evoked muscle stretch reflexes. H-reflex amplitudes were attenuated in VR conditions (VRL and VRH). No difference was observed in H-reflex amplitude between virtual height conditions in the present study. Interestingly, compared to REO, H-reflex amplitudes were reduced in the VRL condition, suggesting a more general effect of changing visual input from real to virtual, despite these conditions being seemingly similar. This demonstrates that the neural control of standing is altered in VR, even in the absence of any virtual height manipulation, which to our knowledge has not been reported previously (i.e., standing on ground level in the real- and virtual-world results are not equivalent). Reduced H-reflex amplitudes in response to threatening scenarios is consistent with previous reports. Subjects have exhibited weaker H-reflex amplitudes in response to an increased risk (Llewellyn et al., 1990; Horslen et al., 2013; Miranda et al., 2019) or consequence (Sibley et al., 2007; Horslen et al., 2013) of a fall, as well as during alternative scenarios stimulating sympathetic arousal (Kamibayashi et al., 2009; Tanaka, 2015). This suggests an inhibitory response at the spinal level that may be facilitated by descending commands from cortical or supraspinal structures (Llewellyn et al., 1990; Adkin et al., 2008). Tanaka et al. (2013) support this theory by using single-pulse transcortical stimulation to demonstrate increased corticospinal excitability to the internal oblique muscle during postural-challenging tasks at height. It is feasible that, when encountered with added sympathetic arousal or altered visual input, in general, the central nervous system downregulates spinal excitability to stifle excess motor outputs which may interfere with its ability to produce appropriate and efficient sensory-evoked responses to the stimuli. The net effect of this neural adaptation would be that mechanical reflex excitability remains unchanged, as observed in the present study.

We are under the assumption that both vibration and H-reflexes are short-latency responses using the same spinal circuitry except for initiation method and location. The H-reflex is initiated by stimulating the Ia afferent fibers directly (bypassing activation of muscle spindles), and therefore, its pathway is incomplete compared to its mechanically induced counterpart. Vibration reflex initiation occurs via muscle spindle stimulation resulting from a mechanical stretch applied to the muscle or tendon, encapsulating the entirety of the Ia afferent pathway. It is postulated, based on results of this study, that specific adaptations local to the muscle spindle may occur in the presence of virtual reality induced postural threat. This inference is consistent with prior research (Bonnet et al., 1995; Both et al., 2003, 2005; Hjortskov et al., 2005; Kamibayashi et al., 2009; Davis et al., 2011; Horslen et al., 2013, 2018) suggesting a heightened spindle-specific role in postural control in arousing situations. Furthermore, because significant H-reflex reduction signifies heightened inhibition at the spinal or supra-spinal level, the increase in spindle sensitivity alongside postural threat must be sufficient to overcome and exceed these inhibitions to register a net-positive increase in overall reflex. Therefore, it is likely that the muscle spindle response to threat is underestimated despite the significant enhancements observed here and in previous reports. The role of fusimotor control in posture and movement, and its specific adaptations when threats are perceived, demands further investigation by directly measuring spindle activity via single unit recordings with human microneurography.

Golgi Tendon Organ (GTO) inhibition may also play a role in NTV-reflex facilitation in threatening scenarios. Horslen et al. (2017) used electrical stimulation of the Achilles tendon to demonstrate Ib-inhibition in the triceps surae muscles during upright standing at both low and high height conditions compared to lying prone at ground level. Thus, it seems GTO-Ib reflexes are also dependent on task and threat perception, and their inhibition under these conditions may enable amplified short latency Ia-muscle spindle responses. Similar reductions in Ib inhibition have been previously demonstrated (i.e., Faist et al., 2006; Van Doornik et al., 2011) and are postulated to reflect a shift toward excitatory, rather than inhibitory reflexes during postural tasks compared to unloaded positions; thus, it seems reasonable to suggest this effect may be amplified as basic postural tasks (i.e., standing) advance to more threatening conditions (i.e., standing at height) with greater levels of muscle activation. However, axons of Ib afferents innervating GTOs are known to be smaller in diameter than Ia afferents supplying muscle spindles. Thus, their stimulation threshold requires a greater current – likely much higher than that delivered during this study, which used stimulation intensities in lower ranges to target muscle spindles specifically (Pierrot-Deseilligny and Burke, 2005). The concurrent roles of Ib-GTO inhibition and Ia-muscle spindle excitation in response to sympathetic arousal must be assessed further to confidently infer their contributions to postural tasks in threatening states.

One interpretation of our findings is that postural threat in VR increased fusimotor outflow, which maintained vibration reflex strength during a concomitant increase of spinal inhibition. A limitation of this research, however, is that we are inferring potential changes in fusimotor outflow and muscle spindle feedback indirectly using reflexive muscle activity as a proxy. As such, we cannot presently make firm conclusions regarding muscle spindle responses or potential mechanisms of fusimotor control. In future studies, these results should be bolstered by direct recordings from individual muscle spindles with human microneurography in threatening conditions. Even so, direct recordings from single fusimotor neurons are not feasible in humans; therefore, 𝝲MN activity will still need to be inferred from changes in muscle spindle firing, all while keeping αMN (muscle background activity) constant across conditions. This challenge of inferring fusimotor drive from muscle spindle activation is a well-appreciated in the literature (Burke, 2021; Dimitriou, 2021).

Another limitation to this study is the lack of a real-world high height condition. VR has previously been shown to be capable to eliciting similar responses to real-world conditions involving quiet standing at low and high heights (Cleworth et al., 2012). The current study aimed to build on this work by probing the state of neural circuitry under similar conditions. It would have been beneficial to directly compare these outcomes in real and virtual height conditions as well as in the real and virtual ground level conditions explored in this study. However, due to practical limitations, a real height condition was replaced with an eyes closed condition in this study. Nonetheless, closing the eyes enabled an interesting comparison to virtual environments, and has been used previously in studies involving real-world and VR visual environments (Horlings et al., 2009). Finally, comparisons between REO and VR conditions in current study would have been better controlled had the virtual environment been designed to match the real-world visual scene of the lab. Previous studies comparing real and virtual conditions have implemented near identical visual scenes within their virtual display (Simeonov et al., 2005; Cleworth et al., 2012), and this discrepancy may have played a role in some conflicting findings between studies.

Our data reinforces that VR can elicit sympathetic arousal and subjective fear responses with manipulations in virtual elevation. Future research should investigate how these outcomes scale in different, potentially even more threatening, virtual environments. Beyond basic neural mechanisms of balance control, our findings further suggest that VR may be useful for therapeutic purposes. For example, VR exposure therapy has been presented as a promising concept for managing phobias and anxieties, such as towards heights or flying, by gradually exposing patients to a negative stimulus (Rothbaum et al., 1997; Krijn et al., 2004; Bush, 2008; Albakri et al., 2022). There is already research in this area suggesting benefits in terms of fear of falling and fall risk in older adults (Rendon et al., 2012; Levy et al., 2016), and clinically in people with acrophobia (Coelho et al., 2009; Rimer et al., 2021), or anxiety disorders (Meyerbröker and Emmelkamp, 2011).