Bilateral H-waves were recorded on 16 young subjects without any clinically detectable neurological damage. Eight subjects had no prior sports background (sedentary individuals); i.e., they did not practice any sports activity before their participation in the experiments. The other eight subjects had practiced track and field and other sports for several years (athletes). The type of track and field and sports practiced by these athletes is illustrated in Table 1. Sedentary individuals: 19.37 ± 1.3 years; 61.52 ± 14.36 kg; 1.63 ± 0.07 m. Athletes: 20.25 ± 1.83 years; 63.44 ± 13.83 kg; 1.67 ± 0.87 m.

All subjects signed informed consent, and the protocol was approved by the Local ethics committee of Medical School, National Polytechnic Institute. The experiments were designed and conducted according to the Helsinki declaration.

Subjects were asked not to drink coffee or cola drinks during 24 h prior to the experiment because caffeine increases the excitability of the neuromuscular system (Walton et al., 2003). EMG recordings were carried out on both medial gastrocnemius muscles simultaneously, with subjects in a supine position. To avoid the cancelation of signals (Tucker et al., 2005), the electrodes were located above or below the innervation region (Childers, 2004; Sheverdin et al., 2009). Before recording, the skin around the popliteal fossa and the belly of the medial gastrocnemius muscle were cleaned with alcohol and gauze. After this procedure, the surface electrodes for stimulation and recording were carefully placed on the skin. The array of these two recording bipolar electrodes had the same inter-electrode distance as the stimulation electrodes (2.0 cm). The inter-electrode distance was defined as the center to center distance between the conductive area of two bipolar electrodes (according to SENIAM project; Surface ElectroMyoGraphy for the Non-Invasive Assessment of Muscles). The recording and stimulation electrodes were placed on the medial gastrocnemius according to Tucker et al. (2005). To guarantee a similar placement of the bipolar electrodes pre- and post-training we used a picture of the electrode positions and their slight adjustment to obtain H-waves with the lower intensity of stimulation (Hermens et al., 2000).

The stimuli were simultaneously applied to both legs with two independent stimulator units, using surface electrodes placed on the left and right popliteal fossa. The stimulation electrodes had an inter-electrode distance of 2.0 cm. The optimal stimulation site was the site in the medial gastrocnemius muscle in which we obtained H-waves with the lowest stimulation current. In all the experiments we performed slight adjustments in the position to obtain H-waves with the lower intensity of stimulation (Pierrot-Deseilligny and Mazevet, 2000). In this form, such stimuli activated the posterior tibial nerves and we obtained bilateral H-wave responses recorded in the right medial gastrocnemius (MGR) and left medial gastrocnemius (MGL) muscles (Figure 1A). First, recruitment curves were obtained (Figures 1B,C) for both muscles, and the stimulation intensities were adjusted to obtain bilateral H-waves with amplitudes of about 20% of the maximal M-wave amplitude (M_max). This fact is clearly illustrated in our Figures 1B,C, which shows two H-wave recruitment curves, in which, it is clearly illustrated that at 20% of Mmax the H-wave amplitude is in the ascending part of the H-wave (see blue and magenta curves). Any reader could trace an imaginary horizontal line in 20% of Mmax in Figures 1B,C and could verify that the lower intensity to produce the H-wave at 20% of Mmax is in the ascending part of the recruitment curve. We used the intensity associated with the ascending part of the H-wave curve and not the descending part, which produces an antidromic collision. In this context, the use of 20% of Mmax in our experiments is in line with the recommendation to use H-wave lying in the ascending part of the recruitment curve (as described by Crone et al., 1990). Table 2 shows the range of intensities employed for the recruitment curves and the intensity at 20% of Mmax. Under these conditions, 130 bilateral H-waves were obtained (Figures 1B,C). Moreover, the time course of the peak to peak amplitude of the H-waves was measured, detrended and plotted (Figure 1D). Subsequently, the FD was computed using Higuchi's method.

The training modality in all subjects consisted of running three times per week (for 13 weeks) on a concrete road of 5 km. The subjects were monitored by a running coach. Table 3 shows the timeline for this training program. The intensity of the exercise was set according to the handbook by Macneill and The Sport Medicine Council of British Columbia (2012). In this book, the run plan was designed to turn anyone into a runner, without injury. Sports medicine physicians designed this training program. The subjects were instructed to run faster every week, and the tempo training was monitored. Table 4 describes the training duration in minutes (Min), the rating of perceived exertion scale (RPE) and the training load (TL) for athletes and sedentary subjects. The RPE is a reliable indicator to monitor exercise intensity. To obtain the RPE the subjects were instruted to report the subjective rate of the level of exertion during training. We used the RPE scale from 1 to 10. The number 1 was very light activity. The numbers 2–3 represented light activity. The numbers 4–5 indicated moderate activity. The numbers 6–7 represented vigorous activity. The numbers 8–9 very hard activity, and the number 10 represented maximal effort. To calculate the TL we multiplied the RPE by session duration in minutes. Table 4 shows the RPE and TL the first and the last day of the training program. Note that the TL evolved during the 13 weeks after training. The justification of this training program is that it is commonly employed for beginning runners to prevent injuries and to obtain health benefits.

Higuchi's method is an algorithm for computing the fractal dimension in the time domain (Higuchi, 1988). We chose this method because it is suitable for calculating the FD during short time spans (Accardo et al., 1997). We only analyzed stationary series of H-waves. Recordings with movement artifacts were not considered for the analysis. We validated Higuchi's method with an analysis of the FD of the surrogate data obtained by a random shuffling of the original data (shuffled surrogates of the H-wave time series). Higuchi's method and these analyses were performed on MATLAB (MathWorks, Natick, MA, USA).

The stimulus to produce H-waves was applied simultaneously in both posterior tibial nerves to activate the MGR and MGL muscles (Figure 1A). Such stimuli consisted of constant current rectangular pulses, of 1 ms duration at 0.166 Hz (i.e., one pulse every 6 s). The current intensity was narrowed between 1.4 and 13.9 mA. The stimuli were generated with two independent stimulators (Digitimer, DS5 and DS7) commanded by another stimulator (Master-8), which also generated a trigger pulse to activate the data acquisition system (see the vertical green line, Figure 1A).

The recorded electromyography signals, H-wave and M-wave (see superimposed recordings, Figure 1A, right panel), were amplified 500 times with two separate amplifiers (GRASS Astro-medic LP511), one for each leg, and filtered with a frequency range from 10 Hz to 1 kHz. The basal electrical activity of the muscles was monitored online with a digital oscilloscope (Tektronix TDS 2014). Finally, the signals were digitized with a Digidata 1440A interface (Axon CNS Molecular Device) having a sampling rate of 50 kHz, and stored for further analysis. Subsequently, the latencies and amplitudes (peak to peak) of the H-waves were measured with the AxoScope software version 10.2 (Molecular Devices, Silicon Valley, USA).

To avoid the depression transient due to “homosynaptic post-activation depression” for repetitive stimulation (see Hultborn et al., 1996) we used a very low stimulation frequency of 0.166 Hz. With this stimulation frequency each H-wave was produced every 6 s, and therefore we did not observe the post-activation depression (see Figure 1D). In other studies employing higher stimulation frequencies around 1 Hz (e.g., see Mezzarane and Kohn, 2002) the authors discarded from the analysis the first 10 H-waves to avoid in their analysis the depression transient due to the homosynaptic post-activation depression. In our case, it was not necessary to discard any H-wave because we employed a very low frequency of stimulation and we did not observe this phenomenon. We followed the recommendations made by Rossi-Durand et al. (1999), who employed stimulation time intervals higher than 3 s. This is consistent with studies by Nielsen et al. (1993) who employed stimulation time intervals of 4 s (i.e., 0.25 Hz).

We defined the H-wave latency as the time interval between the stimulus artifact and the first peak of the H-wave. We measured the H-wave latency for all the subjects before and after training. We used the H-wave latency as a criterion to detect a possible abnormality in the spinal function after training. With this index of measurement, we can examine whether the studied subjects suffered a lesion during the training program. Furthermore, because the H-wave latency is not as variable as the H-wave amplitude, we justify that it can be employed as a control to validate changes in the H-wave amplitude between two experimental conditions (before and after training).

In a preliminary protocol, the intensity of the test stimuli was increased to obtain a recruitment curve (Funase et al., 1994; Hilgevoord et al., 1994). The stimulation intensities were increased until the disappearance of the H-wave by the collision of orthodromic and antidromic action potentials on efferent fibers. Such recruitment curves provide information about the excitability state of the spinal cord (Funase et al., 1996). The recruitment curves were also useful for determining the 100% activation of all the motor units (M_max) in the region where the recording electrode was placed. For some subjects, it was not possible to obtain recruitment curves; therefore, these subjects were excluded from the study. Such subjects exhibited high amplitude M-waves and most of them were athletes with high muscle mass. Due to the collision, it was not possible to obtain the H-wave in these subjects. In fact, this has been a very common observation in our laboratory along several years of experience recording the H-wave.

Data from the recruitment curves were fitted to sigmoid functions, and the slope of the ascending branches was computed (Klimstra and Zehr, 2008) (Figures 2C,D). We calculated the slope at 50% of the maximum H-wave amplitude (H_slp), because this is the best way to obtain a fair value for spinal excitability (Klimstra and Zehr, 2008). We also computed the slope of the recruitment curves of the M waves (such slope were defined as: M_slp). This method is shown for a sedentary subject (Figure 2). Moreover, we calculated the ratio H_max/M_max.

The FD was calculated with Higuchi's method with the algorithm written in MATLAB. The data employed to calculate the FD were the peak to peak amplitudes of the H-waves, as previously described. The analysis was performed from series of 130 H-waves, which were recorded during 13 min. Finally, a detailed statistical analysis was applied to examine whether there are significant differences between the mean FD obtained from different conditions.

We computed the CCV of H-wave amplitude sequences recorded from both legs (simultaneous bilateral recordings). To compute the CCV we used a Matlab routine developed by Mezzarane and Kohn (2002). The time series of 130 H-wave amplitudes from both legs were detrended by subtraction of the best straight line fit to the time series. Such detrending was useful to avoid artifacts in the CCV analysis. We computed the 95% confidence interval using the formula 1.96/(N)^1/2, where N = 130 (the number of samples in the H-wave sequence) (Brockwell and Davis, 1991; Mezzarane and Kohn, 2002).

To test for any statistically significant difference in the H-wave and M-wave in sedentary individuals and athletes, we considered the maximum responses, slope and FD of M-wave and H-waves. Moreover, we tested for any statistically significant difference in the CCV of bilateral H-wave amplitude fluctuations. To compare the after vs. before training conditions for every item (M_max, H_max, M_slp, H_slp, H_max/M_max, FD, and CCV), we performed statistical analysis. Because some data are not normally distributed (Kolmogorov-Smirnov normality test, p < 0.05) and some have no homogeneity of variances (Levene test, p < 0.05), we used Friedman's ANOVA for non-parametric parameters. Where the differences were statistically significant, we performed the Wilcoxon signed-rank test with the null hypotheses that the difference between the aforementioned conditions was zero. The test of significance was one-tailed.

The mean latencies of the H-wave recorded in both MGR and MGL muscles in the athletes before and after training were 32.9 ± 0.4 and 33.1 ± 0.38 ms (mean ± standard error), respectively. The corresponding mean latencies of the H-waves in the sedentary individuals before and after training were 32.4 ± 1.5 and 32.7 ± 1.5 ms, respectively. The Student's t-test resulted in t = 1.34 and p = 0.2 for the athletes, while the values were t = 0.95 and p = 0.36 for the sedentary individuals. These data show that in both the athletes and sedentary individuals the H-wave latencies were not statistically modified after track training.

To analyze recruitment curves we used two parameters: the maximal amplitudes of those waves (M_max and H_max, which were measured from the curves, as illustrated in Figures 2A,B), as well as the slope of the ascending branch in the recruitment curves of the M and H waves (H_slp and M_slp). The results are shown for the H_slp and M_slp obtained from a sedentary subject before and after training (Figures 2C,D). The slopes for the M-wave recruitment curves are illustrated in magenta, and the slopes for the respective H-wave recruitment curves in black (Figures 2C,D). Similar graphs were obtained for all the subjects (both the athletes and sedentary individuals) and the data were collected for their statistical analysis. Table 5 shows raw data per subject, for athletes and sedentary subjects. Such table shows the data in mV for the left and right legs, for H-waves, M-waves and H_max/M_max ratio before and after training.

To examine the statistical significance of the comparison of mean values for M_max and H_max (and H_max/M_max) in sedentary individuals and athletes before and after training, we performed two nonparametric Friedman tests among the four conditions (M_max before, M_max after, H_max before and H_max after) for all subjects. For sedentary individuals, the results show highly significant differences between the four conditions [χ(3)2 = 25.15, p = 0.00001]. A subsequent post hoc analysis using the Wilcoxon sign test revealed significant differences between M_max before and M_max after (p = 0.008) (see black arrow in Table 6), but not between H_max before and H_max after (p = 0.361). Similarly, the data for athletes showed significant differences between the four conditions [χ(3)2 = 16.96, p = 0.001]. However, the post hoc analysis using the Wilcoxon sign test revealed no significant difference between M_max before and M_max after (p = 0.779), nor between H_max before and H_max after (p = 0.99) (see Table 6).

Likewise, we examined the statistical significance of the comparison of mean values of the slopes, M_slp and H_slp, for sedentary individuals and athletes before and after training. We performed two nonparametric Friedman tests between the four conditions (M_slp before, M_slp after, H_slp before and H_slp after) for all subjects. The data analysis showed no significant differences between the four conditions [χ(3)2 = 2.52, p = 0.472 for sedentary individuals, and χ(3)2 = 14.87, p = 0.091 for athletes; see Table 6].

Moreover, we examined the statistical significance of the comparison of Hmax/Mmax in sedentary and athletes, before and after training (see Table 5). We performed a nonparametric Friedman test to the data obtained from the four conditions: “sedentary before,” “sedentary after,” “athletes before,” and “athletes after.” The data analysis showed no significant differences among the four conditions [χ2₍₃₎ = 5, p = 0.172]. See Table 6.

Overall, the results strongly suggest that training did not produce changes in spinal excitability related to the gain in amplitude of the recorded H-waves in either group of subjects.

We also explored the qualitative behavior of H-wave amplitude fluctuations for sedentary individuals and athletes before and after training. Return maps were constructed for the incoming peak-to-peak H-wave amplitude on the abscissa (j) vs. the previous peak to peak H-wave amplitude (j-1) on the ordinate (example shown in Figure 3 for one athlete).

Data were recorded on the right leg (Figures 3A,B) and left leg (Figures 3C,D), before training (Figures 3A,C) and after training (Figures 3B,D). In general, we observed a mild change in the data dispersion after training in both groups of subjects, without significant changes in the mean plus standard deviation for any subjects or conditions. This qualitative finding led us to examine whether there are changes in other measurements obtained from the data before and after training, such as the fractal dimension (FD) of the H-wave amplitude fluctuations illustrated in the return maps.

The return maps were not sufficient for detecting changes in the behavior of amplitude fluctuations of the H-wave. Therefore, we computed the FD of such amplitude H-wave fluctuations for all subjects before and after training. The main idea behind the FD value is that it represents a relative index of complexity. Therefore, we employed this index to detect changes in the complexity of the H-wave amplitude fluctuations before and after track training.

To illustrate how we obtained the FD, an example is shown of the FD analysis for one athlete (Figure 4), obtaining data from the right leg (Figures 4A,B) and left leg (Figures 4C,D), before training (Figures 4A,C) and after training (Figures 4B,D). Note that for this athlete, the FD in the amplitude fluctuations of the right H-wave changed from 2 to 1.92 (before vs. after track training), and similarly the FD for the amplitude fluctuations of the left H-wave changed from 2 to 1.93. In general, we systematically observed this type of change in the FD for all the athletes.

Graphs similar to the one illustrated in Figure 4 were obtained for all subjects (athletes and sedentary individuals). The data were collected for statistical analysis.

We employed the nonparametric Friedman test to examine the statistical significance of the FD measurements for the four conditions: sedentary individuals before, sedentary individuals after, athletes before and athletes after. The analysis of data showed significant differences between the four conditions [χ(3)2 = 19.45, p < 0.0002]. A subsequent post hoc Wilcoxon sign test revealed significant differences between sedentary individuals before and after training (p = 0.006), athletes before and after training (p = 0.001), sedentary individuals before and athletes before training (p = 0.001), but not between sedentary individuals after and athletes after training (p = 0.391). It is important to point out that although the absolute value of the FD between sedentary individuals and athletes was similar after training, the direction of the change in the FD was opposite between the two groups of subjects. Whereas the FD index significantly increased for the sedentary individuals, from 1.88 ± 0.12 to 1.94 ± 0.05 (p = 0.006, n = 14), this parameter significantly decreased for the athletes, from 1.98 ± 0.01 to 1.93 ± 0.04 (p = 0.001, n = 13) (see Table 6).

We calculated the CCV from time series of bilateral H-wave amplitude fluctuations. This could be useful for examining whether there are common synaptic inputs modulating the spinal alpha-motoneuron pools from the two legs. A significant change in the CCV amplitude at a particular lag would indicate a stronger modulation of the alpha-motoneuron pools from both legs or a stronger modulation from common central inputs on the alpha-motoneurons. We computed such CCVs before and after training in the two groups of subjects, athletes and sedentary individuals. Therefore, the aim of this analysis was to examine whether the CCV could be useful as a biomarker to detect plastic changes produced by track training in healthy subjects. The CCV analysis is shown for one athlete before and after training (Figure 5). Similar analyses were performed for all the subjects. The detrended time series of the H-wave amplitudes for the right (MGR) and left (MGL) legs is represented (Figure 5A). The horizontal lines illustrate the corresponding reference baselines associated with the detrending. The corresponding CCV obtained from such time series is presented (Figure 5B). The two horizontal lines indicate the 95% confidence interval. A clear CCV peak at lag 0 emerges above the confidence band (Figure 5B). We observed that this CCV exhibited a significant increase for athletes (comparing such CCV peaks in Figures 5B,D). Similar results were obtained for all the CCVs computed for athletes. In sedentary individuals, contrarily, we did not observe a significant change in the CCV peak after track training.

The mean amplitude of the CCV peak at lag 0 for the athletes before training was 0.21 ± 0.19, changing to 0.47 ± 0.13 after training. Meanwhile, the mean amplitude of the CCV peak at lag 0 for the sedentary individuals before training was 0.08 ± 0.12, changing to 0.24 ± 0.21 after training. We employed the nonparametric Friedman test to examine the statistical significance of the CCV between the bilateral H-wave amplitudes for the distinct conditions: athletes before, athletes after, sedentary individuals before and sedentary individuals after. The results show significant differences between the four conditions [χ(3)2 = 9, p < 0.029]. In the CCV, the Wilcoxon sign test revealed a significant difference between athletes before and after (p = 0.02). In contrast, no statistically significant difference was found between sedentary individuals before and after (p = 0.08), sedentary individuals before and athletes before (p = 0.225), or sedentary individuals after and athletes after (p = 0.08) (see Table 6).

We herein highlight the statistically significant changes for the FD and CCV obtained for athletes and sedentary individuals before and after track training. In Table 6 (last column), we point out such significant changes with gray and white arrows and the illustrated asterisks. Note that the FD for the athletes goes down (gray arrow). Conversely, the FD for the sedentary individuals goes up (gray arrow). In both cases, the changes are statistically significant (see asterisk). The conclusions from this statistical analysis for athletes and sedentary individuals before and after track training is illustrated with a cartoon (Figure 6). The CCV for both the athletes and sedentary individuals goes up (white arrows, Table 6, last column). However, only the athletes exhibited a statistically significant increase in the CCV, whereas for the sedentary individuals this increase did not reach statistical significance. The change in the CCV of the bilateral H-waves for the athletes and sedentary individuals is also presented (Figure 6).

We decided to perform the recordings in the gastrocnemius muscle instead of the soleus muscle because it is composed of approximately 50% of rapid and 50% of slow muscle fibers. However, the soleus muscle is composed by 70–90% of slow muscle fibers and only 10–30% of rapid muscle fibers (Johnson et al., 1973; Edgerton et al., 1975; Green et al., 1981). Given the type of track training in our study, subjects running only during 13 weeks, we assumed that plastic changes could be due to rapid muscle fibers.

Plasticity in the spinal cord is a mechanism of the nervous system useful to resolve challenges in motor behavior as well as in learning new motor skills. In this context, we observed that in athletes the TL before and after training changed from 131 ± 34.2 to 83.8 ± 6.94 and in sedentary subjects from 366 ± 18.2 to 176 ± 14.2. These results suggest that the training program employed in these subjects improved the motor performance in both athletes and sedentary subjects, thus suggesting that plastic changes could occur after track training. It is known that with the voluntary training, the neurons show evidence of dendritic grow, increase in the synthesis of new proteins, increase in protein axonal transport and changes in the intrinsic properties of motoneurons (see revision in Gardiner et al., 2006). It is now widely accepted that the H-wave is a tool for studying such plastic changes in the spinal cord (Wolpaw and Tennissen, 2001; Zehr, 2002). Indeed, skills are adaptive behaviors that are acquired through practice and plastic changes in the spinal cord (Grillner and Wallen, 2004; Chen et al., 2005). Moreover, the spinal cord is the first place where plastic changes associated with motor behavior can occur, since this structure has a large quantity of switching circuits susceptible to neuronal synaptic efficacy (Wolpaw and Tennissen, 2001). Changes in the synaptic transmission efficacy may result in plastic events in the same synapse, or in other synapses throughout the selective presynaptic control (Quevedo et al., 1997; Lomelí et al., 1998). In this context, it would be interesting to examine whether the selective presynaptic control could also change after training in our experimental paradigm.

Changes in the synaptic plasticity of the monosynaptic reflex pathway occur not only after training but also after spinal cord injury. Lee et al. (2009) demonstrated that a better functional outcome of compression spinal cord injury in mice is related to alterations of the monosynaptic reflex pathway which facilitate mononeuron recruitment. These authors showed that such facilitation occurred spontaneously after the spinal cord lesion and it was associated with an increase of the H-wave amplitude 6–12 weeks after the injury. In our case we observed a differential change in the FD and CCV of the H-wave fluctuations without significant changes in the H-wave amplitude after training in both athletes and sedentary subjects. A possible explanation for this discrepancy could be that the plastic changes that we observed could occur at the level of mechanisms controlling the H-wave fluctuations via presynaptic actions and not exclusively at the level of the Ia-motoneuron synapse.

In future experiments, it would be interesting to explore the effects of diverse substances that affect H-wave fluctuations, including caffeine, which produces profound alterations in neural function through known mechanisms (Rivera-Oliver and Diaz-Rios, 2014). The possibility of studying the FD and CCV changes in this context could be of interest for physiologists and physicians. However, caution must be taken because it is difficult to establish cause-effect associations in this emergent field known as the neuroscience of exercise (Machado et al., 2015). We conclude that the observed changes in the CCV between bilateral H-waves, as well as the differential changes in the FD of the H-wave amplitude fluctuations, are associated with the previous fitness history of the subjects. Further studies on animals and humans are necessary to describe the physiological mechanisms underlying such associations in detail (see also Reyes et al., 2007).

Previously, it was demonstrated in cats that the intraspinal branches of afferent fibers function as a dynamic assembly that can be fractionated to convey information to selected neuronal targets in the different spinal nucleus (Lomelí et al., 1998) or inside of the same neuronal nucleus (Quevedo et al., 1997). Therefore, it is plausible that in humans, the selective presynaptic action of such afferent fiber collaterals on different neuronal targets could also contribute to changes in the FD index and the CCV after a track training program. In future experiments, it would be interesting to examine whether the strength of presynaptic actions of afferent fiber collaterals and their selectivity on diverse spinal neuronal groups also change in cats after a track training program.

In the present experiments, the intensity of stimulation was very low for the activation of group II fibers. However, because the electrical threshold for activating Ia afferents is too close to that for Ib afferents, we suggest that any physiological explanation of the results must include the participation of Ib afferents and their target interneurons. This possibility is in line with the well-documented evidence that the H-wave is not exclusively monosynaptic and it may be influenced by changes in spinal excitability (Burke et al., 1983, 1984).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.