Neuromotor variability partially explains different endurance capacities of expert pianists

During fatiguing piano tasks, muscle fatigue develops differently between expert pianists. Differences in neuromotor strategies employed could explain a slower rate of fatigue development. The objective was to compare muscle activation and kinematic variabilities between ShortDuration (i.e., pianists with less endurance) and LongDuration groups. Results from 49 pianists showed that EMG activation variability of most shoulder and upper limbs muscles was greater for the ShortDuration group with time during two piano fatiguing tasks, namely Digital and Chord tasks. Segment acceleration variability, assessed using inertial measurement units, was also greater with time for the ShortDuration group at the right arm during the Digital task, and at the thorax and head during the Chord task. Finally, thorax lateroflexion variability increased with time for the LongDuration group (but not the ShortDuration group) during the Digital task. During the Chord task, wrist flexion variability was higher for the LongDuration group compared to the ShortDuration group. These results showed a direct effect of time on the pianists’ acceleration variability and EMG activation variability. In contrast, a protective effect of fatigue development could be attributed to kinematic variability. Results also suggest a higher risk of injury among pianists in the ShortDuration group.

Participants.Fourty-nine expert pianists were recruited.All participants were enrolled in undergraduate or graduate studies in piano performance or had at least a university (or equivalent) degree.Participants had to be free of any playing-related muscular disorders during the year preceding the experimentation.After receiving instructions on the protocol, participants read and signed written informed consent.The protocol was approved by the Université de Montréal's Ethics Committee (CPER-18-086-D), and the research was performed in accordance with the Declaration of Helsinki.As described in our previous study, participants were categorized into two groups, ShortDuration and LongDuration, based on their duration on a Digital (repeating a right-hand 16-tone sequence) and a Chord (repeating a right-hand chord sequence) task 3 , as described in full details hereafter.Briefly, a k-means clustering and a silhouette validity index were used to identify the optimal number of subgroups, i.e., two, based on the participants' time-to-task termination, which varied across participants in the tasks described hereafter.Characteristics of participants are shown in Table 1.

Instrumentation.
As in Goubault et al. 3 , participants were equipped with 49 monopolar EMG electrodes of 1.5 mm diameter (TMSi, EJ Oldenzaal, The Netherlands) separated by 2 cm and positioned on the right wrist flexor and extensor muscles according to a 7 × 7 array (Fig. 1).Before electrode placement, forearm skin was scrubbed with 70% isopropyl alcohol pads.A conductive gel was used to improve skin-electrode conductivity.Electrodes were attached to the skin with circular double-sided tape, and a medical elastic net bandage was used to hold the electrodes in place and minimize cable movements.Additionally, participants were equipped with two monopolar electrodes separated by 2 cm and positioned on the right biceps brachii, triceps brachii, anterior and lateral deltoids, and superior trapezius.Electrode locations were determined according to the Surface ElectroMyoGraphy for the Non-Invasive Assessment of Muscles project (SENIAM) recommendations 22 .Hair was removed with a razor when necessary, and skin was cleaned with alcohol swabs.The electrode positioning procedure was performed by a trained physiotherapist to ensure the reliability and physiological meaning of the data.It was then validated by a series of 10 submaximal voluntary contractions during which EMG signals were visually inspected in real-time on a display monitor.Briefly, the production of a burst of activation during a submaximal contraction was checked to ensure proper positioning of the electrodes with an appropriate signal-to-noise ratio.EMG signals were acquired using a 72-channel Refa amplifier (TMSi, EJ Oldenzaal, The Netherlands) at a sampling rate of 2048 Hz.
Participants were also equipped with 17 inertial measurement units (IMUs) (XSENS, Enschede, Netherlands) positioned on feet, shanks, thighs, pelvis, trunk, head, shoulders, arms, forearms, and hands, as recommended by the manufacturer.The IMUs positioning procedure was performed by the same researcher to ensure interparticipant reliability.IMU signals were acquired at a sampling rate of 60 Hz.
The participants played a grand piano (Disklavier DC7X Enspire Pro, Yamaha Canada Music Ltd, Toronto, Canada).A digital sound level meter (407,730 SLM, Extech Instruments, Nashua, USA) was placed at a 1.5 m distance from the back-left side of the piano bench to continuously monitor the sound intensity level.
Experimental procedures.The experimental procedure was identical to Goubault et al. 3 .The music scores were sent to the participants at least five days before the experiment.Participants were asked to practice the two excerpts to play accurately and comfortably during the experiment.Before collecting data, participants' maximal handgrip force was measured using a handgrip dynamometer (Takei Scientific Instruments, Tokyo, Japan).Participants performed two 5-s maximal voluntary contractions with the right hand, interspersed with a 1-min rest period.Verbal encouragement was provided during contractions.Participants then performed a 5-min warm-up session on the piano, followed by a sound check to familiarize themselves with the required sound intensity level.Participants were then asked to perform the Digital and Chord tasks described hereafter without interruption.The order of the two piano tasks was randomly assigned, and a 15-min rest period was provided to participants in-between tasks.Both tasks were performed fast and loud to increase muscle activation levels [23][24][25] and accelerate the development of muscle fatigue 3 .To this end, the tempo was established using a metronome connected to an earphone placed in the participants' left ear.A screen was positioned in front of the piano for the pianist to adjust the playing volume when necessary.The Digital task corresponded to a 16-tone right-hand digital sequence on the first two measures of Exercise no 7 from C.L. Hanon's 'The Virtuoso Pianist' (Appendix-A1), spanning a major sixth interval in the middle register of the keyboard.Every note was a sixteenth (very short in duration) with no rests between each note.Participants were asked to play this task loudly (minimum threshold = 78 dB) and fast (quarter note = 112 beats per minute [BPM]) while maintaining an accurate sound intensity.The Chord task corresponded to a right-hand chord sequence based on bar 119 of Franz Liszt's Ballade no 2 in B minor S.171 (Appendix-A1).The chord sequence was composed of three notes played five times: once in the middle register (a staccato [detached] eight note) and twice (fast repetitions) in each of the middle-high and high registers of the keyboard.The rhythm of the two chords for each fast repetition was different, with the first chord as a sixteenth note and the second chord as a staccato eight note.There was the equivalent of an eighth note rest between all the registers which, given the performance tempo, was a very small period.After the high-register chords, the rest period was slightly longer (one quarter rest and one Table 1.Characteristics of participants (mean ± SD). *In the past year.Bold values highlight significant differences between groups (α = 0.05).Group comparisons were performed for demographic data using a t-test or F-test when appropriate.Term.(i.e., Termination) is the average time instant at which the ShortDuration group stopped the task.Mid.(i.e., Midpoint) is the average midpoint instant at which the LongDuration group was analyzed, as it best matches the time instant at which the ShortDuration stopped the task (i.e., Termination).Kinematics analysis.Free-of-gravity segment accelerations (x, y, z) of the pelvis, thorax, neck, right scapula, shoulder, elbow, and wrist were filtered using a second-order zero-lag Butterworth band-pass filter with www.nature.com/scientificreports/0.5-15 Hz cut-off frequencies.The magnitude of segment acceleration was then calculated, and cycles were time-normalized.Joint angles were directly extracted from XSENS MVN software (XSENS, Enschede, Netherlands), except for the pelvis and thorax, since only specific spine segments (i.e., L5, L3, T12, T8) are estimated with XSENS MVN software using interpolation of the pelvis, thorax, and head data.Therefore, pelvis and thorax angles were reconstructed using quaternion data and 'yxz' Euler angle sequence.Joints of interest were located at the pelvis (3 degrees-of-freedom [DoFs] in tilt-elevation-rotation), thorax ( For each DoF, the angle cycles were time-normalized.

Measure of variability.
For the ShortDuration group, the interquartile range (IQR) was calculated for each muscle activation, segment acceleration, and joint angle at each frame for the first and last 15 cycles (task Initiation and Termination, respectively), as illustrated on Fig. 2. For the LongDuration group, the interquartile range (IQR) was calculated for each muscle activation, segment acceleration, and joint angle at each frame for the first (task Initiation) and mid (task Midpoint) 15 cycles, i.e., corresponding approximately to the task Termination time of the ShortDuration group.For each participant, the median IQR at Initiation and Term./Mid.were used as the intra-participant variability measure.

Statistical analysis.
A two-way ANOVA on Group (ShortDuration vs. LongDuration) × Time (Initiation vs. Term./Mid.) with repeated measures on the Time was performed on the variability of EMG activation level, segment accelerations, and joint angles.T-test analysis was performed post-hoc as necessary.Data processing and statistical analyses were performed with Matlab R2019a (The MathWorks Inc., Natick, MA, USA).

Results
Three and two out of fourty-nine participants were removed from analysis for the Digital and Chord tasks, respectively, because of missing data.In addition, for angle analysis of the Digital task, five participants were removed because of important drift observed in angle data (i.e., more than 5°).

Muscle activation variability. Digital task.
There was a significant Group-Time interaction for the EMG activation variability for 30 out of 42 forearm bipolar signals (Fig. 3).Paired t-test analysis showed that the EMG activation variability increased with Time for the ShortDuration group (p < 0.01), while it did not change for the LongDuration group (p > 0.05).The EMG activation variability significantly increased with Time for 6 out of 42 forearm bipolar signals.It was significantly higher for the ShortDuration than the LongDuration group for 5 out of 42 forearm bipolar signals.
In addition, there was a significant Group-Time interaction for the EMG activation variability of the biceps, anterior and lateral deltoids.Paired t-test analysis showed that the EMG activation variability increased with Time for the ShortDuration group (p < 0.001), while it did not change for anterior and lateral deltoids (p > 0.05) and slightly increased for the biceps (p = 0.01) for the LongDuration group.Moreover, the EMG activation variability of the triceps and superior trapezius significantly increased with Time and was significantly higher for the Short-Duration than the LongDuration group.Statistical results are shown in supplementary table S1 (Appendix-A2).

Segment acceleration variability.
Digital task There was a significant Group-Time interaction on the variability of the right arm acceleration.Paired t-test analysis showed that the variability of the right arm acceleration increased significantly with Time for both ShortDuration and LongDuration groups (p < 0.001 and p = 0.008, respectively), with a greater increase for the ShortDuration group.Moreover, there was a significant increasing acceleration variability of the pelvis, thorax, head, shoulder, forearm, and hand with Time (Fig. 5).

Statistical results are shown in supplementary table S3 (Appendix-A2).
Chord task There was a significant Group-Time interaction on the variability of the thorax and head accelerations.Paired t-test analysis showed that the acceleration variability of the thorax and head increased significantly with Time for the ShortDuration group (p = 0.02 and p = 0.04, respectively), while it did not change for the LongDuration group (p = 0.41 and p = 0.15, respectively).Moreover, there was a significant decreasing acceleration variability of the right forearm and hand with Time (Fig. 6).Statistical results are shown in supplementary table S4 (Appendix-A2).

Joint angle variability. Digital task
There was a significant Group-Time interaction on the variability of the thorax lateroflexion.Paired t-test analysis showed that the variability of the thorax lateroflexion increased significantly with Time for the LongDuration group (p < 0.004), while it did not change for the ShortDuration group (p = 0.81).Moreover, there was a significant effect of Time on the angle variability of the pelvis (tilt, elevation, and rotation), thorax (flexion and rotation), neck (lateroflexion), scapula (elevation and rotation), shoulder (flexion and abduction), and elbow (flexion and pronation).For all these DoFs, the angle variability increased with Time (Fig. 7

). Statistical results are shown in supplementary table S5 (Appendix-A2).
Chord task There was a significant effect of Group on the angle variability for wrist flexion, with a higher angle variability for the LongDuration group compared to the ShortDuration group.Additionally, there was a significant effect of Time on the angle variability of the thorax flexion, and neck lateroflexion; angle variability increased with Time (Fig. 8).Statistical results are shown in supplementary table S6 (Appendix-A2).

Discussion
The objective of the present study was to understand the neuromotor strategies that slow down the development of fatigue in piano performance.Results showed that EMG activation variability increased more for the ShortDuration group with Time during both Digital and Chord tasks, while it did not change, or decreased for the LongDuration group.Additionally, EMG activation variability of the biceps, anterior and lateral deltoids increased more for the ShortDuration group as compared to the LongDuration group during the Digital task.In terms of kinematics, segment acceleration variability increased more with fatigue for the ShortDuration group at distal segments (i.e., right arm) during the Digital task, and at proximal segments (i.e., thorax and head) during the Chord task.Moreover, proximal joint angle (i.e., thorax lateroflexion) variability increased with time for the LongDuration group during the Digital task, while it did not change for the ShortDuration group.During the Chord task, distal joint angle (i.e., wrist flexion) variability was higher for the LongDuration group compared to the ShortDuration group.

Variability and time-to-task termination.
Results of the present study highlight a higher EMG activation variability with time for the ShortDuration group (i.e., the fatigued group), which partially contradicts previous studies suggesting either a protective effect of EMG variability against fatigue development 7 , or a decreasing EMG variability with fatigue 8 .However, in tasks that require a high level of accuracy, such as piano performance, motor variability impacting performance variables is undesirable.This could explain why the non-fatigued pianists of the LongDuration group maintained a low-level of EMG activation variability, whereas the fatigued pianists of the ShortDuration group had an increasing EMG activation variability, possibly to impede further fatigue development 5,7 .The increased EMG activation variability was accompanied by a greater increase of the right arm acceleration variability during the Digital task, and a greater increase of the thorax and head acceleration variability during the Chord task for the ShortDuration group, as compared to the LongDuration group.This possibly explains the decreased control of key-attack velocity due to fatigue for the ShortDuration group in Goubault et al. 3 .Interestingly, Madeleine et al. 16 showed that experienced butchers had a larger variability than novices for several kinematic variables, but a reduced EMG variability compared to novices in simulated cutting tasks.This suggests that higher EMG variability among novices could be seen as a lack of adaptation to the work task, as seen among computer workers 28 .In the present study, the two groups had similar levels of piano experience and practice, and only differed in the time to exhaustion and myoelectric manifestation of fatigue 3 .Nonetheless, pianists from the LongDuration group adopted specific motor programmes, with an increased thorax lateroflexion and wrist flexion variability during the Digital and Chord task, respectively, that could have reduced the level of activity of some of the upper-limb muscles, and could possibly explain the greater time-to-task termination.An increased thorax angle variability was previously shown in repetitive fatiguing tasks 27,29,30 suggesting a possible adaptation strategy at proximal joints used to compensate for upper-limb fatigue or fatigue-related discomfort during repetitive dynamic tasks.As pianists' movements are individualized 31 , trunk movement could be a compensatory mechanism that potentially reduces the risk of injury.Nevertheless, these results suggest that pianists of the ShortDuration group could be at higher risk of injury than pianists of the LongDuration group, since they presented similar increasing EMG activation variability and decreasing joint angle variability as in Madeleine et al. 16 .However, future studies are needed to verify this hypothesis.

Variability and fatigue.
Results of previous studies suggest a task dependency in the evolution of EMG activation variability with fatigue.Indeed, an increasing EMG activation variability has previously been observed in arm muscles (i.e., upper trapezius, anterior deltoid, and biceps) during a repetitive fatiguing task 32,33 .In contrast, a decreasing EMG activation variability was observed with fatigue in biceps brachii and brachialis during isokinetic, concentric/eccentric elbow flexion 8 .In the present study, the time affected both ShortDuration and LongDuration groups during the Digital and Chord tasks.The EMG activation variability increased with time for some wrist and finger flexor and extensor muscles and for the triceps and superior trapezius during the Digital tasks.During the Chord task, a decreasing EMG activation variability was observed with time for some wrist and finger flexor and extensor muscles as well as for the biceps.The notion of task dependency is reinforced with kinematics adaptations, with an increasing acceleration variability with time for most segments during the Digital task, and a decreasing acceleration variability with time for the right forearm and hand during the Chord task.It is also reportable that acceleration variability of the thorax, for instance, was qualitatively four times higher   during the Chord task compared to the Digital task, suggesting that pianists used more their trunk motion during the Chord task to produce keystrokes.To produce keystrokes, expert pianists may use multi-joint upper-limb movements 34,35 .These multi-joint movements can also involve trunk motion, which can contribute to creating hand and finger keystroke velocities 36 , to facilitating and initiating upper-limb movements 37 as well as to increasing upper-limb joint angle variability 38 .The same observation can be made for other segments, i.e., acceleration variability was higher at the upper-limb(s) and the head during the Chord task compared to the Digital task.The different movement characteristics of both tasks 3 could explain the differences observed in the evolution of segment acceleration variability.The Chord task (1) requires movements of larger amplitude than the Digital task, and ( 2) is composed of sections formed by different fast rhythms with a short rest between each section (demanding therefore the production of higher bursts of keystroke force in short time periods).Therefore, during the Chord task, fatigue could be at the origin of the decreasing acceleration variability of distal segments because of less movement amplitude or a negative impact on the musical intention with the development of fatigue.On the contrary, the Digital task (1) requires movements of less amplitude with notes very closed to each other on the keyboard, and ( 2) is only composed of continuous fast (sixteenth) notes without rest between each note.This means that for keeping a high level of performance (i.e., no error and the right sound intensity and timing between notes as assessed in our previous study 3 ), the room for maneuver of pianists in terms of acceleration variability is very low.Therefore, an increase of acceleration variability of the segments could be interpreted as a partial "loss of control" due to fatigue in that particular case, one that probably had a direct incidence on the variance of key velocity observed in our previous study 3 .Overall, even if the acceleration variability evolved in either direction with fatigue during both Digital and Chord tasks, it may be interpreted as a consequence of fatigue development instead of a strategy to impede further fatigue progress.

Music and variability.
High inter-trial repeatability of EMG measurements was observed in string players (violinists) with less than 10% deviation of the mean muscular activities from their averages over the repetitions [39][40][41] .Another study showed low variability in maximum mouthpiece forces and kinematics among trumpet players 42 .The highly repeatable muscle activation and kinematic patterns in musicians, reflected in the present study by marginal variability of expert pianists muscle activation and movement patterns, could be explained by the years of practice required to become an expert 40,43 .Ericsson et al. 44 estimated that expert musicians spent over 10,000 h of musical practice by the age of 21, leading to reorganization in certain sensory and motor systems and their interface 45 .The present study highlights that expert pianists can reproduce muscle activity and movement patterns and partially use variability with the development of fatigue when performing repetitive fatiguing Digital and Chord tasks.
Limitations.Some limitations have been taken into consideration when interpreting the results.First, the maximal voluntary contraction was not performed for all muscles.The EMG normalization was therefore determined using the average of the 10 maximum amplitude values recorded during each task.However, this should have a limited impact on the intra-participant EMG activation variability results as each bipolar signal is compared to itself within task Initiation and Term./Mid.Then, angular joints measured using IMUs should be interpreted with caution since fusion algorithms used to estimate sensor orientations can be subject to error accumulation over time caused by environmental magnetic perturbations and drift [46][47][48] .Nonetheless, drift should have a limited impact on the interpretation of the results since joint angle variability was calculated over 15 cycles, representing approximately 30-s of data recording.

Conclusion.
Overall, this study highlights a direct effect of time on the EMG activation and segment acceleration variability of pianists, whereas a higher joint angle variability was observed in pianists having greater time-to-task termination.This suggests a direct effect of fatigue on EMG activation and segment acceleration variability, while a protective effect of fatigue development could be attributed to joint angle variability.Also, expert pianists with lower time-to-task termination during standard repeated Digital and Chord tasks exhibited an increased EMG activation variability, a greater increased acceleration variability and a lower wrist and trunk angle variability with time as opposed to expert pianists having higher time-to-task termination.This highlights different neuromotor strategies between pianists of different endurance characteristics, that could have an impact on the prevalence of injury.Future research should verify this hypothesis by assessing the EMG activation and kinematic variability effects from a training intervention aimed at increasing pianists' endurance.
www.nature.com/scientificreports/eight rest) before the pianist had to return to the middle register for the subsequent repetition.Participants were asked to play this task loudly (minimum threshold = 84 dB [middle register] and 94 dB [high register]) and fast (quarter note = 120 BPM) while maintaining dynamic and musical intention.The rate of perceived exertion was monitored every 30 s while participants were playing via the CR-10 Borg scale26 .Participants were stopped if they reached two consecutive scores of 7 or more at the CR-10 Borg scale or after 12 min of continuous playing27 .Two videos extract of both Digital and Chord tasks are available in supplemental files.Data processing.Electromyography analysis.Monopolar EMG signals were converted into bipolar EMG signals by subtracting the two closest monopolar signals along the columns.EMG data were then filtered using a 10-400 Hz band-pass filter and a 60-120-180 ± 1 Hz stop-band filter to remove 60 Hz electrical contamination and its harmonics3 .All filters were second-order zero-lag Butterworth filters.Data were then zero-aligned by subtracting the mean value.Muscle activation level was calculated using a 9 Hz low-pass filtering of the full-wave rectified EMG signals, normalized to the average of the 10 maximum amplitude values recorded during each task.Cycles, referring to each repetition of Digital and Chord excerpts, were time-normalized to handle possible variations between cycles before intra-participant variability calculation.

Figure 1 .
Figure 1.(A) close-up of the electrodes positioned on the forearm before medical elastic net bandage positioning, and (B) schematic view of the 7 × 7 array of electrodes positioned on the forearm and the underlying muscles, by Goubault et al. 3 , CC BY.

Figure 2 .
Figure 2. Illustration of IQR calculation at Initiation for one EMG signal of one participant.The median value of the calculated IQR curve was then used as the intra-participant variability measure.

Figure 3 .
Figure 3. (A) colormap representation of ANOVA results' p-values for Group-Time interaction (left), main effects of Group (middle) and Time (right) for the Digital task.Columns of electrodes are labeled c1 to c7 as illustrated in Fig. 1.Notes.Non-significant values, i.e., p-values above or equal to 0.05 are represented in white.(B) Median value of activation levels' IQR at Initiation (left) and Term./Mid., (right) for the ShortDuration (upper panel) and LongDuration (inferior panel) groups.Notes.*indicates significant Group-Time interaction; O indicates significant main effect of Group; + indicates significant main effect of Time; BI Biceps brachii; TR Triceps brachii; AD Anterior deltoid; LD Lateral deltoid; ST Superior trapezius.

Figure 4 .
Figure 4. (A) colormap representation of ANOVA results' p-values for Group-Time interaction (left), main effects of Group (middle) and Time (right) for the Digital task.Columns of electrodes are labeled c1 to c7 as illustrated in Fig. 1.Notes.Non-significant values, i.e., p-values above or equal to 0.05 are represented in white.(B) Median value of activation levels' IQR at Initiation (left) and Term./Mid., (right) for the ShortDuration (upper panel) and LongDuration (inferior panel) groups.Notes.*indicates significant Group-Time interaction; O indicates significant main effect of Group; + indicates significant main effect of Time; BI Biceps brachii; TR Triceps brachii; AD Anterior deltoid; LD Lateral deltoid; ST Superior trapezius.

Figure 5 .
Figure 5. IQR of acceleration magnitude at Initiation and Term./Mid.For the Digital task.The white dots represent the mean, the first whisker interval represents the 95% confidence interval of the mean, and the second whisker interval represents the standard deviation of the mean.Notes.*indicates significant Group-Time interaction; + indicates significant main effect of Time; no main effect of Group was observed.

Figure 6 .
Figure 6.IQR of acceleration magnitude at Initiation and Term./Mid.For the Chord task.The white dots represent the mean, the first whisker interval represents the 95% confidence interval of the mean, and the second whisker interval represents the standard deviation of the mean.Notes.* indicates significant Group-Time interaction; + indicates significant main effect of Time; no main effect of Group was observed.

Figure 7 .
Figure 7. IQR of angle segment at Initiation and Term./Mid.for the Digital task.The white dots represent the mean, the first whisker interval represents the 95% confidence interval of the mean, and the second whisker interval represents the standard deviation of the mean.Notes.*indicates significant Group-Time interaction; + indicates significant main effect of Time; no main effect of Group was observed.

Figure 8 .
Figure 8. IQR of angle segment at Initiation and Term./Mid.for the Chord task.The white dots represent the mean, the first wisker interval represents the 95% confidence interval of the mean, and the second wisker interval represents the standard deviation of the mean.Notes.+ indicates significant main effect of Time; o indicates significant main effect of Group; no Group-Time interaction was observed.