Fatigue induces altered spatial myoelectric activation patterns in the medial gastrocnemius during locomotion

Sustained voluntary muscle contractions can lead to fatigue, which diminishes the muscle ’ s ability to absorb energy and produce force at a desired level. Prolonged fatigue can lead to a decline in human performance and increase the muscle ’ s susceptibility to injury. In this study, we investigated how localized muscle fatigue affected spatial EMG patterns during locomotion. We recorded high-density electromyography (EMG) from the medial gastrocnemius of 11 healthy subjects while they walked (1.2 m/s) and ran (3.0m/s) on a treadmill before and after performing a task that locally fatigued their ankle plantar ﬂ exor muscles. We applied multivariate signal cleaning methods to remove motion artifacts from the recorded signals. From these data, we com-puted the peak EMG amplitude, spatial entropy, peak EMG barycenter, and mean power frequency content during walking and running before and after subjects fatigued. We also calculated sagittal plane lower limb joint kinematics and kinetics in each condition. We found that peak EMG activity signi ﬁ cantly decreased during walking and running after the fatigue task, and the location of the peak EMG barycenter had shifted proximally compared to its prefatigue location. We also observed an increase in the EMG mean power frequency during locomotion postfatigue. Despite the changes in spatial EMG activation, lower limb biomechanics were similar before and after fatigue. These results suggest that motor unit recruitment was altered to sustain force production and forward propulsion. This may be a protective mechanism to more broadly distribute muscle loads and avoid myo-tendinous injury. frequency aspects of neuromuscular recruitment in the medial gastrocnemius change in response to fatigue, resulting in altered myoelectric patterns during walking and running. These data may help us better understand the adaptations that occur in lower limb muscles to avoid overuse injuries caused by fatigue.


INTRODUCTION
Fatigue is a broadly used term that refers to a reduction in physical and/or cognitive function that limits human performance (1). It is often composed of both central and peripheral components. Central components constitute a diminished ability of the central nervous system to drive motor neurons, whereas peripheral components constitute changes within a limb or muscle that lead to an attenuated output response (2). Although repetitive voluntary muscle contractions, such as those that occur during walking and running, often involve some degree of central fatigue (3), a greater proportion occurs locally within muscles (4).
Muscle fatigue is characterized by a marked decrease in a muscle's capacity to perform its normal physical actions (5). Though recovery from muscle fatigue often occurs within the first hour, repeated voluntary contractions may lead to a prolonged state of weakness known as fatigue of long duration (6,7). In the initial stages of muscle fatigue, additional motor units are recruited to maintain the required muscle forces and distribute loads among muscle fibers (4). With fatigue of long duration, the normal excitation-contraction coupling is disrupted, limiting the muscle's ability to generate the desired forces (8,9). When the muscle is unable to sustain this activity, muscular fatigue can lead to the development of chronic tendon injuries (10,11). Repeated muscle contractions increase tendon compliance and often lead to damage of the tendon structure (12). This is particularly relevant during endurance sports, where athletes often experience alterations in their neuromuscular recruitment and central drive (13,14). During long-distance running, for example, ankle plantarflexor muscle fatigue can lead to increased loads on the Achilles tendon, leaving it susceptible to injury once its load tolerance is exceeded (15).
Electromyographic (EMG) measurements have shown that localized muscle fatigue alters the spatial myoelectric activation patterns in healthy individuals. In separate experiments, Falla et al. (16,17) showed that the centroid, or barycenter, of EMG activation shifts during fatiguing contractions in the upper trapezius and lumbar region. Similar results have been seen in the lower limb. During a stationary fatiguing plantarflexion task, changes in EMG amplitude and frequency occurred in localized spatial regions of the medial gastrocnemius (18). This redistribution of EMG activity may act as a mechanism to prolong an individual's endurance and prevent overload on the active muscle fibers (19). Fatigue of the quadriceps resulted in varying alterations in the vasti and rectus femoris muscles, underscoring that neuromuscular fatigue may need to be analyzed on a muscle-by-muscle basis (20).
Although considerable research has been done to study the effects of muscle fatigue, translation into augmented human performance and injury prevention have not kept pace (1). Altered motor neuron input during fatigue is not well understood (21). Lower limb muscle injuries result from multiple factors (22), and the mechanisms are likely task dependent (20). The mechanisms involved in muscle fatigue are also task dependent (23), so stationary tasks may not fully represent the manifestation of fatigue during locomotion. The effects of fatigue on spatial EMG distribution have not been studied during locomotion, and the results may provide insight into how the neuromuscular system responds to fatigue during dynamic, cyclic tasks. Recent studies using high-density EMG suggest that it may be a viable technology to analyze changes in neuromuscular recruitment during locomotion (24)(25)(26).
In this study, we investigated how localized muscle fatigue of long duration subsequently affected spatial EMG activation patterns in the medial gastrocnemius during locomotion. We measured high-density EMG from healthy individuals during walking and running before and after they performed a repetitive calf raise task that locally fatigued their ankle plantarflexors through eccentric and concentric muscle actions. We tested two major hypotheses in this study. First, after inducing plantarflexor fatigue, we anticipated that during walking and running, peak EMG amplitude would increase, and mean power frequency and spatial EMG homogeneity would decrease (27). We also expected a proximal shift in the barycenter of peak EMG activity compared to its prefatigue location based on evidence that localized fatigue alters the spatial EMG pattern of a muscle (18). Second, we predicted that sagittal plane lower limb kinematics and kinetics would not significantly change during locomotion when postfatigue cadence is constrained to match prefatigue cadence. Previous research has shown that ankle plantarflexor fatigue has very little effect on lower limb biomechanics during walking (28). With this research, we aim to better understand the spatial neuromuscular adaptations in the medial gastrocnemius muscle in the presence of localized fatigue.

Subjects
Eleven healthy subjects (4 men, 7 women; mean age: 24 ± 5 yr) with no history of major orthopedic injuries or neurological conditions participated in this study. All subjects were right-leg dominant. The University of Florida Institutional Review Board approved all study procedures, and all subjects provided written informed consent in accordance with the Declaration of Helsinki.

Experimental Protocol
We recorded high-density EMG from the medial gastrocnemius of each subject's right leg before and after performing a task that fatigued his or her calf muscles ( Fig. 1; fatigue task described in further detail below in Fatigue Task section). We shaved the hair and cleaned the skin over the muscle using an abrasive paste and then wiped the skin with an alcohol swab. We placed a high-density EMG array (OT Bioelettronica, Turin, Italy) onto the skin using a conductive paste. The array had 64 electrodes arranged into 13 rows and 5 columns (the top left corner of the array was empty), and the interelectrode spacing was 8 mm. To standardize placement of the array across subjects, we aligned the center of the array with the midline of the muscle belly so that the longitudinal axis of the array was aligned with the muscle fibers (29).
Each subject walked (1.2 m/s) and ran (3.0 m/s) on an instrumented treadmill (Bertec FIT, Bertec Corporation, Columbus, OH) in two conditions during the experiment: 1) Before the fatigue task (Prefatigue) and 2) approximately 20 min after performing the fatigue task (Postfatigue). In all conditions, we allowed subjects to become comfortable with the treadmill speed and then recorded data for a duration of 20 steps per leg. During the Postfatigue condition, we compared each subject's cadence during walking and running to his or her Prefatigue cadence. If we found a change in cadence greater than 5%, subjects completed an additional trial so that cadence would not be a confounding factor with regard to changes in EMG activity (30). Three subjects had altered Postfatigue running cadence that exceeded 5% compared to Prefatigue running. These subjects completed an additional running bout at 3.0 m/s using a metronome to match their Prefatigue cadence. None of the subjects had altered walking cadence Postfatigue.
To assess plantarflexor muscle fatigue, we measured the maximum voluntary isometric contraction (MVIC) of each subject's plantarflexor muscles using an isokinetic dynamometer (Biodex Multi-Joint System 4, Shirley, NY). Subjects lay in a supine position with their knee fully extended, thereby ensuring that the gastrocnemii would be recruited during the contractions (31). Their foot was secured to the footplate of the dynamometer and held at a 90 angle with respect to their leg. When the task began, subjects performed five isometric plantarflexion contractions for a duration of 3 s each, with 30 s of rest in between each contraction. We gave each subject a practice trial so they felt comfortable performing the task, and we instructed them to give maximal effort throughout all contractions. We also allowed subjects to rest as long as needed before the recorded trials. We recorded each subject's MVIC at three time points: 1) upon arrival at the laboratory (Prefatigue), 2) immediately following the fatigue task (Postfatigue 1), and 3) following Postfatigue treadmill testing, approximately 1 h after the fatigue task (Postfatigue 2).
We sampled the high-density EMG data at 2,048 Hz with an online bandpass filter (10-500 Hz), and we sampled ground reaction forces from the instrumented treadmill at 1,000 Hz to determine timings of gait events. We also recorded position data of 28 lower body reflective markers on the pelvis, thighs, shanks, and feet using a 20-camera motion capture setup during the walking and running trials (sampling rate: 100 Hz; Optitrack, Corvallis, OR).

Fatigue Task
To induce localized fatigue of the ankle plantarflexors, each subject performed a series of calf raises after their initial walking and running trials (Fig. 1, right). Subjects performed bilateral calf raises to avoid inducing possible gait asymmetries during Postfatigue treadmill testing, though we solely recorded high-density EMG data from the right limb during locomotion. The fatigue task was designed in a pyramid format consisting of sequential bouts of calf raises and rest periods that progressively increased in duration until the apex bout (2.5 min of calf raises, 2.5 min of rest) was reached. After the apex, the duration of each successive bout decreased. To standardize the number of calf raises performed by all subjects, we played a metronome at a rate of 40 beats/min and instructed subjects to perform each calf raise when a beat was played. We instructed subjects to lift their heels as high as possible while forcefully contracting their calf muscles during both the eccentric and concentric muscle actions of the task. If a subject became too fatigued to continue at any point (based on a verbal self-report), we allowed them to rest until they were able to resume the task. Subjects who were able to complete all calf raises performed 500 calf raises over the course of the 25-min task.

Data Processing and Analysis
Each MVIC time point (Prefatigue, Postfatigue 1, Postfatigue 2) included five individual plantarflexion contractions. During each of these trials, we identified the three contractions with the greatest peak torque values. We averaged these values to calculate an average peak plantarflexion torque for each subject at each time point.
We processed the high-density EMG signals using canonical correlation analysis to decompose and clean each of the individual signals (25). We downsampled the monopolar EMG channel data to match the sampling frequency of the force plates (1,000 Hz). We decomposed these resampled time series data into canonical components and used the Fast Fourier Transform (FFT) to convert all components into the frequency domain. We identified outlier components by statistically evaluating the skewness, kurtosis, and standard deviation of all components. If outlier components were detected, we performed spectral noise cancellation on all frequencies in these components based on specific criteria (frequencies whose amplitude were >10 times or <2 times the median spectral amplitude) (32,33). After cleaning the components, we performed an inverse FFT and reconstructed the monopolar EMG data using the cleaned components. We converted the 64 clean monopolar channels into 59 differential channels along the longitudinal axis of the EMG array. We then repeated the same steps to decompose, evaluate, and clean the components from the differential channel data. We then high-pass filtered the differential channel data (4th-order Butterworth, 20 Hz cutoff, zero lag). Finally, we used objective criteria to check for any differential channels with excessively poor signal quality (34,35). If any channels were flagged, we removed them from further analyses.
Since the medial gastrocnemius is primarily active during the stance phase of locomotion, we used the ground reaction force data to epoch the EMG data into successive stance phases during walking and running. We calculated the average root mean square (RMS) value during stance for all differential channels and averaged the values from the 20 strides in each condition. We normalized the RMS values to the peak value in the Prefatigue condition for walking and running individually. Fn1 1 We calculated the modified entropy (36) of the EMG array RMS values to assess variations in the homogeneity of the spatial EMG activation before and after fatigue. Modified EMG entropy is defined as: where p 2 (i) represents the square root of the RMS value at each electrode i normalized to the sum of the squares of all 59 RMS values. Previous research has shown that EMG amplitude in the distal portion of the gastrocnemius is much greater than other areas of the muscle (25,26,37). Therefore, to quantify changes in the spatial location of the peak EMG activity, we calculated the EMG amplitude barycenter (38) for RMS values !75% of the peak RMS value (top 25%). This EMG barycenter location for the x and y coordinates (B x and B y , respectively) is defined as: where RMS TOTAL is the sum of the RMS values from all electrode locations, RMS i is the RMS value at the current electrode (from i = 1:59) that corresponds to coordinates on the electrode array (x i , y i ). To assess spatial spectral differences in the EMG signals Pre-and Postfatigue, we calculated the mean power frequencies of all differential channels of the EMG array. We used these data to spatially map the frequency content in each condition during stance, as well as calculate the overall mean power frequency in each condition.
We filtered the ground reaction forces at 6 Hz (4th-order Butterworth, zero lag) and synchronized the data with the kinematic motion capture data before inverse dynamics calculations (39). We calculated joint angles, moments, and power for the ankle, knee, and hip of the right leg during walking and running using standard inverse dynamics calculations in Visual 3 D (C-Motion, Germantown, MD). We calculated all joint parameters across each subject's relative gait cycle timing, and we normalized joint moments and power to each subject's body mass. Due to motion capture data loss in a few trials, the number of subjects included in some of the biomechanical analyses is fewer than 11. Walking

Statistical Analysis
We used a one-way repeated measures ANOVA (a = 0.05) to test for differences in subjects' plantarflexion MVIC Prefatigue, Postfatigue 1, and Postfatigue 2. We applied post hoc pairwise comparisons with a Bonferroni correction in the presence of a significant main effect. We used paired t tests (a = 0.05) to test for differences in the peak EMG-RMS value, mean RMS entropy value, x-and y-coordinate barycenter values, and mean power frequency before and after the fatigue task. We also used paired t tests to compare peak joint angles, moments, and power between the Pre-and Postfatigue conditions (separately for walking and running).

Pre-and Postfatigue: MVIC
After performing the fatigue task, subjects' ankle joint plantarflexion MVIC values significantly decreased compared to their Prefatigue trial (Fig. 2). Peak plantarflexion torque decreased by an average of 21.5% immediately after the fatigue task (P = 0.002). Subjects' maximum plantarflexion torque also did not recover during the second Postfatigue Figure 2. Peak plantarflexion torque for all subjects (n = 11) before the fatigue task (Prefatigue), immediately following the fatigue task (Postfatigue 1), and approximately 1 h after the fatigue task (Postfatigue 2). We observed a significant decrease from the Prefatigue peak plantarflexion torque during each Postfatigue measurement. Ã Significant differences were determined by a one-way repeated measures ANOVA (P < 0.05). measurement, as there was still a 21.1% average decrease compared to the Prefatigue trial (P = 0.001).

Pre-and Postfatigue: EMG Spatial Activation
Medial gastrocnemius EMG activation patterns during walking and running showed unexpected amplitude decreases during stance on the right limb after the fatigue task (Fig. 3). Despite the large amplitude decreases, the overall muscle activation timing was similar Pre-and Postfatigue for both walking and running. In two subjects, we found large spikes in the differential channel data during early swing phase of running in the Postfatigue condition resulting from mechanical artifact of the EMG ground cable (a representative example is shown in Supplemental Fig. S2). Since we only analyzed the stance phase in this study, our results were unaffected by this occurrence.
We found peak EMG-RMS values at the distal portion of the medial gastrocnemius muscle underlying the electrode array during the Prefatigue walking and running trials (Fig.   4, left). After the fatigue task, we observed large decreases in EMG activity. The mean peak EMG-RMS value was 38.6% lower during walking (Prefatigue: 1.0 ± 0.0; Postfatigue: 0.61 ± 0.30; P = 0.002) and 41.1% lower during running (Prefatigue: 1.0 ± 0.0; Postfatigue: 0.59 ± 0.24; P = 0.001) compared to the Prefatigue condition. However, fatigue did not have an effect on the mean RMS entropy value during stance in either walking (Prefatigue: 5.59 ± 0.15; Postfatigue: 5.53 ± 0.20; P = 0.545) or running (Prefatigue: 5.65 ± 0.14; Postfatigue: 5.52 ± 0.19; P = 0.068), indicating that the spatial homogeneity of muscle activation was unchanged before and after fatigue. The spatial muscle barycenter y-coordinate shifted proximally after subjects' plantarflexor muscles were fatigued (Fig. 4, black circles). During walking, there was a mean proximal shift of 16 ± 13.4 mm among subjects in response to fatigue (P = 0.003), whereas the mean barycenter shifted 11.3 ± 14.2 mm proximally during running (P = 0.025). The barycenter shifted proximally Postfatigue for all subjects during walking, whereas eight of the 11 subjects experienced Figure 3. Mean ensemble medial gastrocnemius EMG activation patterns throughout the gait cycle for all subjects (n = 11) during walking (left) and running (right). Data were aggregated from the 59 differential channels of the EMG array and averaged to provide the mean myoelectric activity throughout the surface recording area. Blue lines represent data from the Prefatigue condition, whereas orange lines represent data from the Postfatigue condition. EMG amplitudes were normalized within each subject and locomotion speed to the peak value in the Prefatigue condition. Vertical dashed lines designate toe-off (TO) for the right limb, and 0% and 100% indicate foot-ground contact. EMG, electromyography. . Group average EMG spatial activation plots (n = 11) during stance phase of walking (top) and running (bottom) in the Pre-and Postfatigue conditions (left and middle, respectively) normalized to the peak value in the Prefatigue condition. On the right, the EMG spatial activation plot from the Postfatigue condition normalized to the peak value from the Postfatigue condition to better show the distribution of RMS values. The group average barycenter for the top 25% of RMS values in each condition is indicated by a black circle with an "X" in the center. EMG, electromyography; RMS, root mean square. a proximal shift during running Postfatigue. We did not find any differences Pre-and Postfatigue in the x-coordinate barycenter in either walking or running (P > 0.546).
In the Prefatigue walking and running conditions, we observed the greatest mean power frequencies in the myoelectric activity from the central portion of the muscle underlying the electrode array (Fig. 5, left column). We found the lowest mean power frequencies in the distal portion of the muscle. In the Postfatigue walking and running conditions, the greatest mean power frequency occurred in the proximal portions of the muscle (Fig. 5, right column). The group mean power frequency across the EMG array increased Postfatigue during running (6.7% increase, P = 0.027) and tended to increase Postfatigue during walking, though the result was not statistically significant (8.3% increase, P = 0.075).

Pre-and Postfatigue: Lower Limb Biomechanics
Sagittal plane biomechanical profiles at the ankle, knee, and hip joints remained largely consistent Pre-and Postfatigue during walking and running when constraining step cadence (Figs. 6 and 7, respectively). Subjects' peak ankle, knee, and hip joint flexion and extension angles were similar Pre-and Postfatigue in both walking and running (P > 0.203 for all). We did not observe any significant changes in peak ankle or knee torque Pre-and Postfatigue during walking or running (P > 0.216 for all), though subjects showed a slight trend toward greater peak hip joint moments in late stance during running Postfatigue (P = 0.081). Peak joint power at the ankle, knee, and hip during walking and running were all unaffected by the presence of fatigue (P > 0.133 for all).

DISCUSSION
We found that healthy individuals exhibited changes in their medial gastrocnemius spatial myoelectric patterns during locomotion in response to localized muscle fatigue. After successfully inducing localized, long-duration fatigue in the ankle plantarflexor muscles through repetitive eccentric and concentric muscle actions, peak EMG activity amplitude significantly decreased in the medial gastrocnemius during walking and running, along with a proximal shift in the peak EMG-RMS barycenter. The mean power frequency of the spatial EMG activity also increased after the fatigue task, with the greatest changes occurring in the proximal portion of the muscle. Despite these fatigue-induced neuromuscular adaptations, lower limb biomechanics were virtually unchanged during locomotion when constraining step cadence.

EMG Spatial Activation
During submaximal exercise, such as walking or running, motor unit activation in a muscle occurs selectively depending on the mechanical requirements of the activity (40). Motor units consisting of slow-twitch (Type I) muscle fibers, which provide greater efficiency during low-velocity cyclical motions, are often recruited in greater proportions than fasttwitch (Type II) fibers during walking (41). Slow-twitch fibers are also more resistant to fatigue than fast-twitch fibers (42).
In the presence of local muscle fatigue, the normal active proportion of muscle fibers is impaired, so additional motor units are recruited to sustain the required power output (43,44). In our results, we saw a shift in the peak EMG barycenter after fatigue, as well as an increase in EMG mean power frequency. During walking and running Prefatigue, the distal portion of the muscle had the greatest EMG activity amplitude (Fig. 4). EMG activity in this region decreased Postfatigue, possibly signaling that different proportions of slow-and fast-twitch fibers were recruited to maintain the required force production. Increased motor unit synchronization is associated with higher surface EMG amplitudes and decreased mean power frequencies (45,46). We observed opposite effects (decreased amplitude and increased frequencies) Postfatigue, which may point to lower synchronization among activated motor units. Localized activation in the distal portions of the muscle may be explained by spatial organization of motor units into small territories within the medial gastrocnemius (47), though contrasting evidence suggests that these motor units may occupy larger territories (48). However, Gallina et al. (18) showed that the effects of fatigue in the medial gastrocnemius manifest as regional differences in amplitude and frequency. Thus, the spatial variations we observed during walking and running Pre-and Postfatigue may represent changes in neuromuscular recruitment that counteract the effects of fatigue and work to sustain the force production for continued locomotion.
We hypothesized that the mean power frequencies would decrease during walking and running Postfatigue, as muscle fatigue is often indicated by a shift toward lower frequencies in the EMG power spectrum (42). Instead, we observed sizeable increases in each parameter Postfatigue. One possible explanation for this discrepancy is that muscle fatigue is often measured during continuously repeated muscular contractions, while the muscle is actively fatiguing (i.e., during the "fatigue task"). Rather, we measured how spatial myoelectric activity is altered when the plantarflexor muscles are already experiencing fatigue of long duration. We confirmed this fatigue via MVIC measurements: nearly one hour after the fatigue task ended, our subjects' maximum plantarflexion torque was still significantly lower than Prefatigue levels  (Fig. 2). The eccentric component of our fatigue task may have caused acute muscle damage and an uncontrolled release of calcium into the sarcoplasm (49). Calcium regulates muscle contraction and relaxation (50), and impairments to its normal release from the sarcoplasmic reticulum could alter the normal patterns of motor unit firing, perhaps leading to increased mean EMG frequencies Postfatigue. de Ruiter et al. (51) found that vastus medialis motor unit discharge rates significantly increased during submaximal contractions after fatigue was induced. They hypothesized that additional motor units were recruited to maintain the levels of force production they measured prefatigue (51). Additional research has also shown that the mean power frequency increases when individuals are fatigued during submaximal activities (52,53). Given that our subjects were able to walk and run with no significant changes to their biomechanics, it is likely they were fatigued at a submaximal level. In this context, the increased spectral power Postfatigue is expected. The shift to a higher mean power frequency Postfatigue appears to be due to a decrease in spectral power at lower frequencies (a representative example is shown in Supplemental Fig. S3). To better understand these results and measure the instantaneous effects of fatigue, it may be beneficial to measure spatial EMG activation during an exhaustive run performed at a faster speed or for a longer duration. Alternative signal processing methods, such as the Choi-Williams transform, may also be better suited to analyze specific time-frequency content of the EMG signals during fatigue (54).

Lower Limb Biomechanics
Despite the large decrease in the medial gastrocnemius EMG activity amplitude after subjects fatigued (Figs. 3 and  4), the average lower limb sagittal plane biomechanics were nearly identical to the Prefatigue condition for both walking and running (Figs. 6 and 7). This supports previous findings that moderate levels of localized fatigue do not have significant effects on lower limb kinematics and kinetics (28,55). We also constrained our subjects' Postfatigue cadences to ±5% of their Prefatigue trials. Minor increases in step rate reduce the loading at the hip and knee during running (56). Eight of the 11 subjects ran within ±5% of their Prefatigue cadence naturally, so any effects of the constrained cadence were presumably minimal. It is possible that the soleus muscle played a greater functional role in the Postfatigue trials to maintain similar gait mechanics as the Prefatigue trials. Cronin et al. (57) found increased soleus activation after a long bout of prolonged walking with a corresponding decrease in medial gastrocnemius activity, possibly highlighting the ability of the neuromuscular system to maintain a constant gait pattern in the presence of neural and mechanical changes to muscles. Ankle plantarflexion is primarily generated through combined force production by the triceps surae muscles (58), though these muscles play different roles based on locomotion speed. The soleus muscle operates with similar force-length properties during walking and running (59), whereas the contractile properties of the medial gastrocnemius tend to vary across different walking speeds (60). The soleus has a higher proportion of slowtwitch fibers than the medial gastrocnemius, so it may be able to resist fatigue during prolonged exercise more effectively than the gastrocnemii (60). We did not measure spatial EMG activity from the soleus in this study because the majority of the muscle body lies beneath the gastrocnemii, thereby reducing the accessibility of recording with a highdensity electrode array. Future studies investigating plantarflexor spatial EMG activity should consider measuring soleus activation with bipolar surface electrodes or smaller linear EMG arrays to measure its relative contributions during fatiguing exercise. Other muscle groups, such as the quadriceps and hamstrings, also have altered EMG activity during locomotion when the ankle plantarflexors are fatigued (61). EMG measurements from these muscle groups, in conjunction with gastrocnemius and soleus measurements, may provide a comprehensive analysis of the compensatory mechanisms used to maintain normal biomechanics despite plantarflexor fatigue.
There were some limitations with this study. We did not equally recruit numbers of subjects with different foot strike patterns (fore foot, heel) during running. Runners with fore Figure 7. Mean ensemble joint angle, torque, and power (rows) for the ankle, knee, and hip joints (columns) during running in the Prefatigue (blue) and Postfatigue (orange) conditions. Joint torque and power are normalized within each subject to their body mass. The dashed blue lines denote the standard deviation envelope for the Prefatigue condition, whereas the shaded orange areas denote the standard deviation envelope in the Postfatigue condition. Vertical dashed lines designate toe-off (TO) for the right limb, whereas 0% and 100% indicate foot-ground contact. Due to data loss in some trials, the number of subjects included in each analysis varies. Ankle-Joint Angle (n = 7), Joint Torque (n = 7), Joint Power (n = 7); Knee-Joint Angle (n = 5), Joint Torque (n = 5), Joint Power (n = 6); Hip-Joint Angle (n = 6), Joint Torque (n = 5), Joint Power (n = 5).
foot strike patterns have increased medial gastrocnemius activity and plantarflexion at the moment of foot contact compared to those who heel strike (62,63). However, our final subject group included only one individual who ran with a fore foot strike pattern, so our results are likely unaffected. We averaged EMG data during stance, as this is the primary gait phase that the medial gastrocnemius is active during locomotion. It is possible that spatial EMG activation may vary within the smaller subphases of stance (e.g., braking, propulsion). There is also increased preactivation of the plantarflexors during late swing phase of running. Within a single stride, motor units are selectively recruited based on the demands of the muscle (40). Load sharing across muscles during cyclic fatiguing activities may also occur in very short phases (64). Investigating changes in spatial muscle activation in gait subphases and/or within smaller time windows may yield more specific information regarding changes in neuromuscular recruitment. We also chose to average the EMG activation patterns and lower limb biomechanics across strides for each individual, potentially masking any variability that occurred due to fatigue. Future studies should consider the effects of long duration fatigue on both intra-and intersubject variability. Finally, the muscle fibers in the gastrocnemius are not parallel to the skin across the entire length of the muscle. The gastrocnemius fiber pennation angles affect the surface EMG spatial amplitude and frequency content, particularly in regions where the fibers are more obliquely oriented with respect to the skin (65). In the presence of fatigue, EMG amplitude and frequency variations are greater, suggesting different motor unit recruitment strategies among individuals (65). Though we were able to quantify the regional activation of the medial gastrocnemius during locomotion, we could not directly measure the recruitment of specific motor units nor the individual frequencies of these motor units when each subject was fatigued.
Investigating neuromuscular recruitment during dynamic activities may increase our understanding of how lower limb musculoskeletal injuries develop. Our fatigue task consisted of a series of continuous calf raises, a task with both concentric and eccentric muscle contractions. The eccentric phase may have had the greatest effect, as it is known to cause more detrimental effects to muscle function than concentric contractions (66). Eccentric contractions are also associated with greater risk of injury to muscle fibers during exercise (22), so the spatial EMG activation we observed during locomotion Postfatigue may have been a mechanism to avoid further damage to the muscle. We analyzed spatial muscle activation patterns in healthy individuals, but individuals with chronic pain and/or injuries have altered patterns of spatial EMG activity during prolonged exercise and fatigue in stationary experiments (17,67,68). Individuals with chronic Achilles tendon injuries also demonstrate altered load sharing and myoelectric activity in the triceps surae muscles during locomotion (69,70). It is possible that individuals with lower limb musculoskeletal injuries may not display the same pattern of spatial EMG alteration that we observed in healthy individuals. Understanding these differences may provide insight into how fatigue of long duration contributes to the onset of injuries. High-density EMG is a useful tool to measure these altered spatial EMG activation patterns during dynamic activities. Expanding these analyses to environments outside the laboratory may provide us with a better understanding of how real-world lower limb neuromuscular injuries occur, as well as strategies to mitigate their development.

ACKNOWLEDGMENTS
We thank Dr. Jennifer Nichols for the use of her isokinetic dynamometer throughout this study.

GRANTS
This study was funded by the Cognition and Neuroergonomics Collaborative Technology Alliance ARL W911NF-10-2-0022.