Evaluation of the Electromyography Test for the Analysis of the Aerobic-Anaerobic Transition in Elite Cyclists during Incremental Exercise

Featured Application: Determination of neuromuscular fatigue threshold in elite cyclists during an incremental test. Abstract: (1) Background: The aim of this study was to investigate the validity and reliability of surface electromyography (EMG) for automatic detection of the aerobic and anaerobic thresholds during an incremental continuous cycling test using 1 min exercise periods in elite cyclists. (2) Methods: Sixteen well-trained cyclists completed an incremental exercise test (25 W/1 min) to exhaustion. Surface bipolar EMG signals were recorded from the vastus lateralis, vastus medialis, biceps femoris, and gluteus maximus, and the root mean square (RMS) were assessed. The multi-segment linear regression method was used to calculate the first and second EMG thresholds (EMG T1 and EMG T2 ). During the test, gas exchange data were collected to determine the first and second ventilatory thresholds (VT 1 and VT 2 ). (3) Results: Two breakpoints (thresholds) were identified in the RMS EMG vs. time curve for all muscles in 75% of participants. The two breakpoints, EMG T1 and EMG T2 , were detected at around 70%–80% and 90%–95% of VO 2MAX , respectively. No significant differences were found between the means of VT 1 and EMG T1 for the vastii and biceps femoris muscles ( p > 0.05). There were no significant differences between means of EMG T2 and VT 2 ( p > 0.05). (4) Conclusions: It is concluded that the multi-segment linear regression algorithm is a valid non-invasive method for analyzing the aerobic-anaerobic transition during incremental tests with 1 min stage durations.


Introduction
In recent decades, great effort has aimed at identifying the threshold that delineates the aerobic-anaerobic transition during an incremental workout on a cycle ergometer [1,2]. Such "fatigue threshold" has been calculated on the basis of metabolic, ventilatory, and surface electromyographic (EMG) variables. In particular, EMG-based thresholds are especially attractive because they can be used to characterize neuromuscular fatigue and can be assessed non-invasively [3]. In the time domain, the increase in neuromuscular fatigue during an incremental continuous exercise is manifested by an increase in EMG amplitude as a result of recruitment of additional motor units, increased firing obtained from all participants before inclusion. Participants were asked to not take part in vigorous physical activity for two days prior to their test date.
The participants were road cyclists engaged in regular training and amateur road races. On average, all cyclists trained at least four times a week covering a weekly distance ranging between 400 and 600 km, plus competition or Sunday training. Cyclists had a national competitive experience of 4.3 (1.7) mean (SD) years and had ridden for an average of 20,000 km (range 16,000-24,000 km) during the last season. None of the participants reported any injuries or pathologies of limb muscles or joints.

Screening and Familiarization Session
Cyclists underwent a blood test (biochemical and hematological parameters) prior to participation to check for anemia and possible infections. Participants were asked to attend an orientation session to become familiarized with the testing apparatus and procedures. All tests were performed on a custom-made cycle ergometer. The participants were required to bring in the saddle, pedals, and cycling shoes from their normal road bicycle. The saddle and pedals were installed on the cycle ergometer. A submaximal incremental test was then performed on the cycle ergometer to familiarize the cyclists with the experimental protocol.

Maximal Cycle Ergometer Test
Participants performed an incremental test to exhaustion on a SRM powermeter (science SRM, SRM GmbH, Germany). Pedal cadence was maintained at 70 rpm during the test, similar to that used in previous studies [7,16,19]. Before the incremental exercise test started, cyclists performed 5 min of unloaded cycling. The test was initiated at a workload of 125 W and the load was increased by 25 W every 1 min [1] until exhaustion or until pedal cadence could not be kept at 70 rpm. For the validity study, each of the 16 subjects performed a single bicycle ergometer test following the above protocol. For the reliability study, 10 subjects performed two exercise tests on a bicycle ergometer on different occasions and separated from each other by a period of no more than five days.

Analysis of Expired Gas and Determination of Ventilatory Thresholds
During the incremental exercise, breath-by-breath analysis was performed using a turbine flow-meter connected to a face mask (dead space: 30 mL). A side pore of the facemask was connected to fast-response differential paramagnetic oxygen and infrared carbon dioxide analyzers. Throughout the incremental test, the software (Oxycon PRO, Carefusion, Germany) averaged every five consecutive seconds data of oxygen uptake (VO 2 ) and carbon dioxide production (VCO 2 ) and ventilatory parameters, as well as the ventilatory equivalents for oxygen (EqO 2 = VE/VO 2 ) and carbon dioxide (EqCO 2 = VE/VCO 2 ). The first ventilatory threshold (VT 1 ) corresponded to the power output value at which VE/VO2 exhibited a systematic increase without a concomitant increase in VE/VCO 2 [20]. The second ventilatory threshold (VT 2 ) corresponded to the minimal power output at which the increase in VE/VO 2 was accompanied by a parallel increase of VE/VCO 2 [20].

Electromyography
Surface EMG signals were recorded using a pair of circular electrodes Ag/AgCl electrodes (Kendall Meditrace 100, Tyco, Canada) arranged in bipolar configuration. The electrodes had a recording diameter of 10 mm and were separated by a distance of 20 mm (measured from the nearest lateral borders). The electrodes were placed on the dominant leg over the vastus lateralis (VL), vastus medialis (VM), biceps femoris (BF), and gluteus maximus (GLM) according to the SENIAM (Surface EMG for Non-Invasive Assessment of Muscles) guidelines [21]. Before electrode placement, the skin was adequately prepared (light abrasion with sandpaper and cleansing with rubbing alcohol). The electrodes and cables were secured with surgical tape and cloth wrap to avoid movement-induced artifacts. Surface EMG signals were recorded using MP150 equipment (BIOPAC, Goleta, CA, USA).
Raw EMG signals were amplified, filtered (band-pass 10-500 Hz) and digitized (sampling frequency 1000 Hz) using the analog-to-digital conversion system of MP150.

Data Analysis
Data were first analyzed with commercially-available software (AcqKnowledge, BIOPAC Systems, Goleta, CA, USA) to monitor any abnormality in EMG traces. Then, data were exported to Matlab (version R2012b; The Math-Works, Natick, MA, USA) for quantitative analysis using a number of custom scripts. The root mean square voltage (RMS) was used as an index of the global EMG amplitude [15]. For each muscle, the RMS EMG was calculated during the active period of the muscle (EMG burst) of each pedal thrust (64-ms window). Then, the RMS-EMG was averaged every six crank revolutions (corresponding to about 5 s for 70 rpm), as previously described [15].

Determination of EMG Thresholds
The multi-segment linear regression method was employed to establish whether one or two breakpoints (thresholds) existed in the EMG amplitude response. All muscles were analyzed independently. This method was implemented as described in the following steps: 1.
The root mean square (RMS) amplitude of each pedal thrust was calculated and represented vs. time (see the data points in Figure 1A).

2.
A single-segment regression curve (a regression line) was fitted to all data points, and then the residual sum of square (RSS) was calculated. The results of this linear regression were used for later statistical analysis (see Figure 1B).

3.
The best two-segment regression of the EMG data was computed. To do this, the algorithm calculated all the possible two-segment regressions, and the two-segment regression yielding the least pooled RSS was chosen as representing the best fit. It was observed that the intersection point (breakpoint) between these two segments was very close to the maximum intensities in all subjects. The intersection of both segments defined the instant when EMG T2 occurred. An analysis of variance was conducted to determine whether the adjustment using two-segment regression led to a significant (p < 0.05) reduction in the total sum of squares when compared to the adjustment using a single-segment regression (see Figure 1C).

4.
Finally, the best three-segment regression of the EMG data was computed in order to detect an additional breakpoint in the EMG curve. The third middle segment was obtained by methodically breaking the first (left) segment obtained in the previous step into two segments, and keeping the regression that yielded the lower sum of squares (see Figure 1D). Finally, an analysis of variance was conducted to determine whether the adjustment using three-segment regression led to a significant (p < 0.05) reduction in the total sum of squares when compared to the adjustment using two-segment regressions. The two points resulting from the intersection of the above three segments determined the instants when the first (EMG T1 ) and second (EMG T2 ) EMG thresholds occurred (see Figure 2).

Statistics
Kolmogorov-Smirnov tests confirmed that each parameter analyzed in the study (EMG T1 , EMG T2 , VT 1 , and VT 2 ) was normally distributed. For the validity study (16 subjects), one-way repeated-measures ANOVA was used to determine whether there were significant differences between, on the one hand, EMG T1 and VT 1 , and on the other hand, between EMG T2 and VT 2 , when these variables were expressed in W, VO 2 (ml/kg/min), and % VO 2MAX . Pearson correlation (r) was used to determine the possible relation between EMG T1 and VT 1 , as well as between EMG T2 and VT 2 . To further analyze the validity of the EMG method, the Bland-Altman method was followed [22]. For this analysis, the mean differences (bias) and SD of the differences between the mean values (W) obtained with the two methods (EMG T1 vs. VT 1 , and EMG T2 vs. VT 2 ) were calculated. The mean difference plus or minus two SD is shown in Figure 3. In this way, the bias and precision of the EMG technique could be calculated [6]. For the group of six subjects that performed the test on two different days (reliability study), a paired Student's t test was used to compare mean values of EMG T1 and EMG T2 obtained with both tests. Intraclass Pearson's correlation coefficients were also calculated to determine the degree of correlation between mean values of EMG T1 and EMG T2 reported with repeated tests. Significance was set at p < 0.05. technique could be calculated [6]. For the group of six subjects that performed the test on two different days (reliability study), a paired Student's t test was used to compare mean values of EMGT1 and EMGT2 obtained with both tests. Intraclass Pearson's correlation coefficients were also calculated to determine the degree of correlation between mean values of EMGT1 and EMGT2 reported with repeated tests. Significance was set at p < 0.05.   technique could be calculated [6]. For the group of six subjects that performed the test on two different days (reliability study), a paired Student's t test was used to compare mean values of EMGT1 and EMGT2 obtained with both tests. Intraclass Pearson's correlation coefficients were also calculated to determine the degree of correlation between mean values of EMGT1 and EMGT2 reported with repeated tests. Significance was set at p < 0.05.    Table 1 provides the physiological characteristics of the participants calculated from the incremental test. As expected, well-trained cyclists reached high values of VO 2 at the end of exercise: 71.6 ± 4.1 mL/kg/min (range 62.3-81.0). The maximal power output of 406 ± 36 W (range 325-450) was also high. The duration of the test was 11.1 ± 1.3 min (range 8-13 min). Table 1. Physiological (mean ± SD) characteristics of the participants (N = 16) calculated from the incremental test.

Variables
Mean (SD) Range  Figure 2 provides an example of an EMG response during the incremental exercise of one cyclist. As can be seen, three regression lines with different slopes were fitted to the data points, which allows for identification of two breakpoints in the linearity of the RMS EMG curve. This two-EMG-threshold pattern was observed in 75% of subjects for all muscles investigated, with EMG T1 and EMG T2 occurring, respectively, around 70%-80% and 90%-95% of VO2 MAX .
The second threshold, EMG T2 , was detected in all 16 subjects for all muscles. However, EMG T1 could not be detected in the VL, VM, and BF in 20%-25% of the cases, whereas it was identified in the GLM for all subjects. It should be noted that EMG T1 was always detected in at least two of the muscles tested. Table 2 summarizes the average values of VT 1 and EMG T1 expressed in W, VO 2 (ml/kg/min), and %VO 2MAX for all muscles. There were no significant differences between the means of EMG T1 and VT 1 for the VL, VM, and BF muscles. However, for the GLM muscle, the means of EMG T1 were significantly greater than those of VT 1 (p < 0.05).  Table 3 summarizes the average values of VT 2 and EMG T2 expressed in W, VO2 (ml/kg/min), and %VO2 MAX for all muscles. There was no significant differences between means of EMG T2 and VT 2 (p > 0.05) for four muscles. To further examine the validity of the EMG method, the Bland-Altman method was applied to assess the degree of agreement between the EMG thresholds and ventilatory thresholds (see Figure 3). This figure shows that, for all muscles, the bias and error of the multi-segment regression method was relatively low. Note that the levels of agreement were especially low for the comparison between VT2 and EMG T2 (Figure 3b). Moreover, for all comparisons, at least 90% of the individual values were within the limits of agreement.  To further examine the validity of the EMG method, the Bland-Altman method was applied to assess the degree of agreement between the EMG thresholds and ventilatory thresholds (see Figure  3). This figure shows that, for all muscles, the bias and error of the multi-segment regression method was relatively low. Note that the levels of agreement were especially low for the comparison between VT2 and EMGT2 (Figure 3b). Moreover, for all comparisons, at least 90% of the individual values were within the limits of agreement.

Correlation between Ventilatory and Electromyography Thresholds
In all muscles, a significant correlation was found between the power outputs (W) at VT1 and EMGT1 (Table 4). Likewise, for all muscles, there was a significant correlation between the power outputs at VT2 and EMGT2 (Table 4). Table 4. Correlations between the power output (W) obtained at the first electromyographic threshold (EMGT1) and at the first ventilatory threshold (VT1), and between the power output obtained at the second electromyographic threshold (EMGT2) and at the second ventilatory threshold (VT2).

Reliability
No significant differences were found between mean values of either EMGT1 or EMGT2 obtained in the first and second tests (p < 0.05). Intraclass correlation coefficients (ICC) between repeated measurements were significant (p < 0.05) and high (Table 5).

Correlation between Ventilatory and Electromyography Thresholds
In all muscles, a significant correlation was found between the power outputs (W) at VT 1 and EMG T1 (Table 4). Likewise, for all muscles, there was a significant correlation between the power outputs at VT 2 and EMG T2 (Table 4). Table 4. Correlations between the power output (W) obtained at the first electromyographic threshold (EMG T1 ) and at the first ventilatory threshold (VT 1 ), and between the power output obtained at the second electromyographic threshold (EMG T2 ) and at the second ventilatory threshold (VT 2 ).

Reliability
No significant differences were found between mean values of either EMG T1 or EMG T2 obtained in the first and second tests (p < 0.05). Intraclass correlation coefficients (ICC) between repeated measurements were significant (p < 0.05) and high (Table 5).

Discussion
The main findings of our study were the following: (1) the multi-segment linear regression method was able to detect two breakpoints (thresholds) in the RMS EMG vs. time curve from all muscles in 75% of participants during an incremental cycling test of 60 s stages; (2) the two breakpoints, EMG T1 and EMG T2 , were detected at around 70%-80% and 90%-95% of VO 2MAX , respectively; and (3) the two EMG breakpoints, EMG T1 and EMG T2 , occurred at similar power outputs than VT 1 and VT 2 , respectively.

Possible Mechanisms for the Non-Linear Increases in EMG Amplitude
Most studies on cycling performance attribute the occurrence of breakpoints in the linearity of the EMG curve to the recruitment of additional motor units and/or to the change in the type of motor units recruited [5]. However, to date, the mechanisms underlying the abrupt increases in EMG amplitude during an incremental test are not entirely clear, and could involve both local and central (global) factors.
At the local (muscle) level, it has been shown that as fatigue develops, there is a progressive accumulation of metabolic byproducts of muscular contraction (i.e., lactate, hydrogen ions, potassium, and inorganic phosphate), which could result in impairment of excitation-contraction coupling [23,24]. To counteract this deficit in the contractile apparatus, muscle power output during an incremental test might be increased by (1) recruiting more motor units; and (2) recruiting type IIa and type IIb motor units to obtain energy through anaerobic glycolysis. Based on this reasoning, it has been hypothesized that the first breakpoint would correspond to the moment when a large number of type IIa (fast twitch oxidative-glycolytic) motor units come into play, whereas the second threshold would occur when a large population of type IIb (fast twitch glycolytic) motor units are recruited [6]. These hypotheses, however, have not been experimentally verified, and thus remain speculative. Indeed, EMG amplitude could increase due to factors other than the number and type of active motor units, such as broadening of the trans-membrane action potential [25], crosstalk [4], and even signal cancellation [4]. Thus, motor unit recruitment is not the only peripheral factor that can account for the non-linear increases in EMG amplitude.
Central (and not only local) factors could also explain the occurrence of abrupt increases in EMG amplitude. Indeed, impairment of the contractile machinery during the course of an incremental test could be compensated by an enhancement of motor unit rate coding. Specifically, it might be speculated that one or two of the EMG thresholds could be related to increases in motor unit discharge rate and/or synchronization [4]. The possible involvement of neural mechanisms in the EMG breakpoints is further supported by the experiments of Airaksinen et al. [26], who demonstrated that, during an incremental cycling test, both working and non-working muscles showed a shift in EMG at the same load. Airaksinen's observations are suggestive of a global change in the neural activation of muscles. It is therefore plausible that the abrupt increases in EMG activity may be initiated not only by impairment of excitation-contraction coupling, but also by increases/adjustments in the neural activity [26]. We conclude by stating that, because multiple factors influence EMG amplitude, the assumption that EMG breakpoints are caused by changes in motor unit recruitment should be treated with caution, as it remains empirically unsupported.

Comparison of the Multi-Segment Linear Regression Method with Stage Durations of 12 s and 60 s
To date, the multi-segment linear regression method had only been tested during an incremental test with 12 s stage durations [6], and the applicability of this algorithm to incremental protocols with longer stage durations (one or two minutes) remained to be verified. Indeed, the 12 s stage test employed by Lucía et al. resembles a ramp-like protocol, in which muscular effort is continuously increased to cope with the quick increases in workload. Such test differs noticeably from the graded steady state protocol employed here, where subjects were asked to maintain a fixed power output for 1 min. Based on this, it was predicted that the development of fatigue, the pattern of motor unit recruitment, and/or the sensation of muscular effort may be different in protocols with 12 s and 60 s stage durations, and that this may result in differences in the detection of EMG thresholds and in the % of VO 2MAX at which the EMG thresholds are reached. Interestingly, we found that the two breakpoints encountered, EMG T1 and EMG T2 , occurred at higher percentages of VO 2MAX (around 70%-80% and 90%-95% of VO 2MAX , respectively), as compared to those found during the 12-s stage protocol (around 60%-70% and 80-90% of VO 2MAX , respectively). One possible explanation for these differences may be that, during exercise periods of longer duration, adjustments in motor unit rate coding may "delay" the recruitment of a significant number of motor units to higher relative intensities of exercise. Specifically, it may be that, with longer stage durations, exercise intensity is maintained at the expense of changes in motor unit discharge rate and/or synchronization and less so due to recruitment of additional motor units [27]. Due to this, great recruitment of additional motor units would be "delayed/postponed" until higher relative intensities of effort are reached, which explains why EMG thresholds occur at higher percentages of VO 2MAX . Alternatively, it might be hypothesized that, with longer stage durations, the sensation of muscular effort appears at higher relative intensities, and thus the signal that triggers the increases in neural activity occurs at a higher percentage of VO 2MAX [26].
It is also worth noting that, with the 60 s stage duration protocol, we detected the first EMG breakpoint in 75% of participants, whereas, with the 12-s stage duration protocol, the first breakpoint was identified in 90% of subjects by Lucía et al. [6]. This small discrepancy may be attributed to slight differences in the development of fatigue in these protocols. Indeed, it is possible that, with the 60 s stage test, the sensation of muscular effort/fatigue increased gradually at the beginning and middle of the exercise, but increased more steeply towards the end of the test. Such progression of fatigue would lead to a EMG time course in which the first breakpoint is not so well marked (for an example, see Figure 1 of Latasa et al. [28]). In contrast, it is likely that, with the 12 s stage test, fatigue increased more progressively, thus resulting in an EMG time course in which two breakpoints are more easily recognized.

Comparison with Previous Studies
The mathematical method utilized here has clearly identified two breakpoints in the linearity of the EMG recordings in most subjects. However, in reviewing the literature there is clearly a disparity of results regarding the number of EMG breakpoints encountered. The existence of a non-linear increase in EMG amplitude at around the aerobic-to-anaerobic transition intensity was reported by some authors since the 1980s [1,12], but it was not until 1999 that Lucia et al. noted the existence of a second EMG breakpoint occurring at near to maximum intensities [6]. Studies prior to Lucía's findings overlooked and missed this breakpoint, which was most likely due to the great interest in determining the transition between the aerobic and anaerobic metabolism, known to be at lower intensities. For this reason, investigations prior to Lucía's work only detected one non-linear increase in EMG variables at about 60%-70% of VO 2MAX [1,29]. After the key study of Lucía et al., there continues to be a discrepancy in the number and interpretation of EMG breakpoints identified during incremental tests [30,31]. Remarkably, automatic methods used by most authors for EMG-threshold detection had the limitation that only one EMG breakpoint could be possibly identified [8,9,32,33]. Essentially, these methods are based on calculating all the two-segment regressions for the data, and then choosing the two-segment regression with the least residual sum of squares, thus yielding a single intersection point. In contrast, the multi-segment linear regression method presented here offers the possibility to fit the data with three linear segments, which allows better adjustment of data and, most importantly, determination of two EMG breakpoints. It might be argued that the two EMG thresholds do not really exist and that they are artificially determined by the algorithm. However, various authors have confirmed the existence of two EMG thresholds by visual analysis [10,11].
The question remains of why the majority of studies have reported only one EMG threshold instead of two. Several factors may contribute to the conflicting results. First, it has been suggested that the second EMG breakpoint can only be detected when subjects are capable of reaching high intensities during the incremental tests, possibly higher than VT 2 . This would explain why EMG T2 is more easily found in professional cyclists: They have the capability of recruiting a sufficient number of motor units at high intensities so as to induce an abrupt detectable increase in EMG activity [6,11,33]. The two-threshold response reported here was obtained for well-trained cyclists, which lends further support to this theory. A second factor may be the choice of the fixed pedaling cadence. Indeed, with a fixed cadence, the increase in exercise intensity is reached at the expense of increasing muscular effort, which favors the recruitment of fast motor units and possibly facilitates threshold detection at EMG T2 [6,34]. Finally, it might be possible that previous studies had problems in finding the both EMG breakpoint because they only assessed one muscle, typically the vastus lateralis. In the present study, we have shown that EMG T2 is more frequently detected in the gluteus maximus than in the vastus, and that EMG T2 is more easily detected when various leg muscles are examined. It is known that the contribution of gluteus maximus to total force is higher at highest workloads [35], and that hip extensors muscles, like gluteus maximus, compensate the loss of force production of knee extensors [36,37]. As a consequence, the onset of fatigue would be delayed in the gluteus maximus compared to the knee extensors; this would also imply a wider range of EMG amplitudes at the highest workloads, thereby explaining the easier detection of both thresholds in the gluteus maximus.
The present findings suggest that the aerobic-anaerobic transition may take place in two stages, thus reinforcing the view that such transition may be more complex than expected. The occurrence of two breakpoints in the EMG vs. time curve would be in accordance with the existence of three exercise intensity domains (moderate, heavy, and severe), and may signify that, during a continuous incremental ergometer test, the increase in fatigue takes place in various phases.

Comparison between EMG Fatigue Thresholds and Ventilatory Thresholds
Our results showed a correspondence between the occurrence of EMG T1 and VT 1 , on the one hand, and between EMG T2 and VT 2 on the other. This correlation between EMG and ventilatory thresholds could be due to the enhanced activation of respiratory centers triggered by intramuscular accumulation of potassium and hydrogen ions, although this possibility is not based on the present results and thus is speculative. Indeed, it has been suggested that group III-IV muscle afferents are sensitive to changes in muscle acidosis and extracellular potassium concentration [38,39]. The chemo-sensitive activation of these afferents could elicit a marked stimulation of the respiratory neurones [40,41], thus leading to an abrupt increase in minute ventilation. Based on this sensorimotor reflex loop, it might be hypothesized that ventilation exhibits a first breakpoint (VT 1 ) when an abrupt increase in the number of active motor units (EMG T1 ) occurs, and a second breakpoint (VT 2 ), coinciding with the change in motor unit recruitment from predominantly slow twitch motor units to fast twitch motor units (EMG T2 ). In agreement with this theory, there are many reports showing a concomitant parallel increase in ventilation and EMG activity during an incremental cycling test [2,6,[12][13][14]28,33].
Alternatively, there are studies suggesting that changes in ventilation observed during incremental exercise are not triggered by accumulation of potassium and hydrogen ions, but that such changes are mediated by an increase in neural activity originating from the subthalamic motor region or indirectly via α-γ coactivation of motor neurons innervating exercising limbs [42]. Such increase in neural activity may occur in response to the need to progressively recruit fast twitch muscle fibers as exercise intensity is increased. Finally, there are reports suggesting that ventilation variables are not influenced by motor unit recruitment, as evidenced by the dissociation between the occurrence of EMG and ventilation thresholds [2,43].

Methodological Limitations
The present experiments had various limitations that deserve comment. First, we did not assess the isometric maximal voluntary contraction and, as a consequence, we had no means to quantify the level of muscle activation and the possibility to normalize EMG data. Moreover, apart from the multi-segment linear regression method, no other techniques have been employed to examine neuromuscular fatigue during the incremental test, such as a decline in maximal voluntary force, and measures of conduction velocity or spectral EMG analysis. As a result, it was impossible to relate the occurrence of the two thresholds encountered (EMG T1 and EMG T2 ) in the EMG RMS response with other indexes of neuromuscular fatigue. A third limitation is that EMG-based thresholds are not fully reliable indexes of the level of muscle activation because such EMG measures are affected by numerous uncontrollable factors (for a review, see Reference [4]). Finally, many factors influence the relation between EMG amplitude and force, including the thickness of subcutaneous tissue, the recruitment strategy, and the peak discharge rates of the different MUs [4]. Due to many factors that can influence this relation, the correlation between surface EMG amplitude the energy system used during continuous cycling may be lower than expected.

Conclusions
In conclusion, it has been shown that, during an incremental cycling test with 60 s stage durations, the multi-segment linear regression method is able to detect a two-threshold response in the RMS EMG curve from four lower limb muscles in 75% of participants. The first and second EMG thresholds occurred at exercise intensities of 70%-80% and 90%-95% of VO 2MAX , respectively. The correspondence between, on the one hand, the first EMG breakpoint and the first ventilatory threshold, and on the other hand, between the second EMG breakpoint and the second ventilatory threshold, indicates that the multi-segment linear regression algorithm is a valid non-invasive method for analyzing the aerobic-anaerobic transition during incremental tests with 1 min stage durations.