Exhausting repetitive piano tasks lead to local forearm manifestation of muscle fatigue and negatively affect musical parameters

Muscle fatigue is considered as a risk factor for developing playing-related muscular disorders among professional pianists and could affect musical performance. This study investigated in 50 pianists the effect of fatiguing repetitive piano sequences on the development of forearm muscle fatigue and on piano performance parameters. Results showed signs of myoelectric manifestation of fatigue in the 42-electromyographic bipolar electrodes positioned on the forearm to record finger and wrist flexor and extensor muscles, through a significant non-constant decrease of instantaneous median frequency during two repetitive Digital (right-hand 16-tones sequence) and Chord (right-hand chords sequence) excerpts, with extensor muscles showing greater signs of fatigue than flexor muscles. In addition, muscle fatigue negatively affected key velocity, a central feature of piano sound intensity, in both Digital and Chord excerpts, and note-events, a fundamental aspect of musicians’ performance parameter, in the Chord excerpt only. This result highlights that muscle fatigue may alter differently pianists’ musical performance according to the characteristics of the piece played.


Results
Group clustering. One participant was excluded for both Digital and Chord tasks because of missing EMG data. In both Digital and Chord tasks (Appendix-A1), two groups were identified using the k-means clustering method, with a silhouette coefficient of 0.90 and 0.88, respectively. The groups were termed as ShortDuration and LongDuration groups, and lasted on average 209.7 ± 99.6 s and 693.2 ± 72.7 s for the Digital task, respectively, and 257.2 ± 97.4 s and 686.2 ± 72.8 s for the Chord task, respectively (Fig. 1). Both groups significantly differed by the time-to-task termination only ( Table 1). As can be seen in Fig. 1 showing participants' individual time-to-task termination, 15 out of the 19 participants and 17 out of the 23 participants of the LongDuration group reached 12 min for the Digital and Chord tasks, respectively. Other stratification parameters such as age, mass, experience, practice time per day, and maximal voluntary contraction (MVC) did not differ between groups (Table 1).  48 . In both tasks, mRPE scores increased for both groups over time, with a greater increase for the ShortDuration than the LongDuration group (Fig. 2). Tukey post-hoc analysis revealed that the mRPE score increased significantly along the entire tasks for the ShortDuration group (ES = 1.38 and ES = 1.26 respectively for the Digital and Chord task), while the mRPE score increased only until I 1 for the Digital task (ES = 2.38) and until I 2 for the Chord task (ES = 1.05) for the LongDuration group. Tukey post-hoc analysis also revealed that the ShortDuration group had a lower mRPE score than the LongDuration group for I 0 and I 1 of the Digital task (ES = 0.76). Additionally, the ShortDuration group had a higher mRPE score than the LongDuration group at the end of both tasks (i.e., I 3 : ES = 1.52, I 4 : ES = 1.76, and I 5 : ES = 1.76) (Fig. 2).
Evolution of electromyographic median frequency. Digital task. There was a significant Group-Time interaction for the EMG median frequency for all the forearm bipolar signals (Fig. 3A). The median fre- Table 1. Demographic data of participants (mean ± SD). *in the past year. **MVC maximal voluntary contraction. Bold values highlight significant differences between groups (α = 0.05). www.nature.com/scientificreports/ quency decreased for both groups over time, with a greater decrease in the ShortDuration group (ES = 1.06) than the LongDuration group (ES = 1.00). Tukey post-hoc analyses revealed that for ShortDuration group, median frequency significantly decreased along the task from I 0 to I 4 for most of the 42 bipolar signals. For the Long-Duration group, median frequency significantly decreased between I 0 and I 1 for 36 out of 42 bipolar signals and between I 1 and I 2 for 7 bipolar signals in the lateral and posterior parts of the forearm, but remained mostly unchanged during the rest of the task ( Fig. 3B; Fig. 2 in Appendix-A4). There was a significant main effect of Column on the variation (I 5 -I 0 ) of EMG median frequency at the task termination. Post-hoc analysis revealed a significantly greater decrease of EMG median frequency in most of the extensor muscles compared to flexor muscles for the ShortDuration and LongDuration groups (Fig. 4).
Chord task. There was a significant Group-Time interaction for the EMG median frequency in all the forearm bipolar signals (Fig. 5A). The median frequency decreased for both groups over time, with a greater decrease in the ShortDuration group (ES = 1.07) than the LongDuration group (ES = 0.18). Post-hoc analyses revealed that for ShortDuration group, median frequency significantly decreased from I 0 to I 1 and from I 1 to I 2 for most of the bipolar signals. For the LongDuration group, median frequency significantly decreased between I 0 and I 1 for most of the bipolar signals, and remained mostly unchanged during the rest of the task ( Fig. 5B; Fig. 3 in Appendix-A4).
There was a significant effect of Column on the variation (I 5 -I 0 ) of EMG median frequency at the task termination. Post-hoc analysis revealed a significantly greater decrease of EMG median frequency in most of the extensor muscles recorded by c2 and c3 columns of electrodes compared to flexor muscles, respectively for the ShortDuration and LongDuration groups (Fig. 6).

Discussion
The objective of the present study was to investigate the evolution of forearm muscle fatigue during two repetitive piano sequences and to assess how muscle fatigue alters piano performance. During the Digital and Chord excerpts, two groups were discriminated based on the time-to-task termination We found that signs of MMF were greater in the ShortDuration than the LongDuration group and that in both groups, signs of MMF were more important in finger and wrist extensors in both Digital and Chord tasks. The note-event errors illustrated by incomplete cycles tended to increase at the end of both tasks with a significant effect in the Chord task. The key velocity variability, related to the intensity of the sound, was higher at the end of both tasks.
The time-to-task termination varied between 1 and 12 min in Digital task and between 2 and 12 min in Chord task. Such variations in time-to-task termination were previously observed 49,50 . Interestingly, the large sample Results of the ShortDuration group are represented above the blue diagonal while results of the LongDuration group are represented below the blue diagonal (α = 0.05). For example, the variation of EMG median frequency (I 5 -I 0 ) of c1 for the ShortDuration group (first row, first column) is not significantly different from c2 (first row, second column), but is significantly different from c4 (first row, fourth column). Similarly, the variation of EMG median frequency (I 5 -I 0 ) of c1 for the LongDuration group (first column, first row) is not significantly different from c2 (first column, second row), but is significantly different from c4 (first column, fourth row). www.nature.com/scientificreports/ size in this study allowed clustering participants into two groups to assess differences in the evolution of signs of MMF according to the time-to-task termination. Although, there were no significant Group effect in terms of years of experience, practice hours per day, gripping force, age, as well as piano performance parameters at the beginning of each excerpt ( Table 1 in Appendix-A3), descriptors of fatigue evolution, i.e., EMG median frequency and mRPE, showed Group-Time interactions. The EMG median frequency decreased quickly for the ShortDuration group during the first half of the task (I 2 /I 3 ), followed by a slower decrease until the end, especially for the Digital task. This pattern was amplified for the LongDuration group that showed a decrease of EMG median frequency mainly at the beginning (I 1 -I 2 ) followed by constant values of EMG median frequency during both tasks. Although it is well accepted that EMG median frequency linearly decreases during fatiguing contractions until task termination 10,11 , this biphasic behavior of MMF was already shown via the decrease of motor unit discharge rate until 40% of time-to-task termination and then followed by a reversal pattern 51 . While several factors may explain this biphasic mechanism such as low-to-moderate contractions during the fatiguing task or muscle specificities 51 , the evolution of EMG median frequency in our study also suggests that signs of MMF may depend on participant-specific endurance capacities. This results highlights the benefit of sub-grouping participants based on their time-to-task termination to better understand neuromuscular processes underlying fatigue. Interestingly, the evolution of EMG median frequency concorded with the evolution of participants' effort perception. Indeed, mRPE of ShortDuration group increased quickly and constantly during the tasks. On the other hand, mRPE of LongDuration group increased during the first 20% and 30% of the tasks (I 2 and I 3 ), respectively for the Digital and Chord tasks, before remaining constant until the end. This observation strengthens the adequacy between the perceived effort and EMG median frequency evolution showed in previous studies [11][12][13]19 . Finally, we could hypothesize that for the same amount of time played, pianists with short time-to-task termination would be at a higher risk of developing PRMDs by an increased muscle fatigue accumulation. Therefore, a higher prevalence of PRMDs could be expected in the ShortDuration group. A follow-up in a longitudinal study of the sample of pianists who participated to this study could answer this hypothesis.
Muscle fatigue was previously proposed as a risk factor of PRMDs 1,2 , with a higher prevalence reported at the forearm and wrist [7][8][9] . The presented results of the Digital and Chord tasks showed that extensor muscles  (I 0 ) representation of the participants' EMG median frequency of the forearm muscles throughout the duration of the task (I 0:5 ), for the ShortDuration (superior panel) and LongDuration (inferior panel) groups. Note that ANOVA were performed on EMG median frequency values and that variation from baseline was for representation only. Dots indicate significant differences revealed by post-hoc analyses. One dot indicates a significant difference between time intervals I i and I i-1 for a given pair of electrode, while two dots indicates a significant difference between time intervals I i and I i−2 . For example, at time interval I 4 for the ShortDuration group, the dot on the left column (c1) of the top row indicates a significant difference between the EMG median frequency at time intervals I 3 and I 4 , while the two dots on the second column (c2) of the last row indicates a significant difference between the EMG median frequency at time intervals I 2 and I 4 . www.nature.com/scientificreports/ were more prone to show signs of MMF than flexor muscles. This observation supports the idea of a higher prevalence of injury in the extensor muscles. The most common injury among pianists is lateral epicondylitis 7,8 which is associated with overuse of the wrist extensor muscles. This result is in agreement with large amount of stress undergone by the extensor muscles 7,52 , and a greater strain on the lateral side of the elbow joint during the repetition of piano sequence [52][53][54] . In addition, previous results suggested that muscle force of wrist flexors was higher than wrist extensors [55][56][57] . Flexor muscles are used to depress keys in the same direction of gravity force, while extensor muscles are used both as wrist stabilizer and to release keys against gravity. Each change in movement direction requires co-activation of agonist-antagonist muscle pairs, and because wrist flexors have a greater moment arm than the extensor muscles, larger forces are required by the extensor muscles to maintain the wrist posture 53 , which can lead to increase fatigue in this muscle group 53,58 . In line with this idea, our results suggest that pianists' wrist and finger extensor muscles located at the forearm, which have an antagonist role in relation to key depression and an agonist role in relation to key release, are more prone to develop muscle fatigue in the context of fast and loud Digital sequences. The metacarpophalangeal joint is the most important finger joint producing the vertical motion of the fingertip during Digital piano task 59 . Consequently, the latters suggested that finger kinematic patterns to press the keys indicated a predominant use of intrinsic hand muscles. Higher contribution on finger flexors muscles located at the hand rather than at the forearm might in part explain the greater signs of MMF in extensor muscles during the Digital task. During Chord sequences, multi-joint upper-limb movements used to perform loud chords are similar to those of isolated loud tones, where fingertip downward velocity is produced first by elbow extension and then wrist flexion 60 . During a Chord piano task, wrist flexors have an agonist role in the depression of keys, while wrist extensors act as continuous stabilizers of this joint 7 . In our study, wrist flexion probably contributed to both loudness and fast repetition of chords at the middle-high and high registers. However, as for the Digital task, the reported evolution of the median frequency during the Chord task suggested a higher incidence of muscle fatigue in extensor muscles, which is probably related to their continuous stabilizer role. www.nature.com/scientificreports/ Concerning musical performance, fatigue, as evidenced by mRPE and EMG median frequency at task termination, significantly altered piano performance parameters. This result support previous findings showing a detrimental effect of fatigue on performance in wind instrumentalists as well as during sport activities, postural adjustment task, or circular horizontal arm movement 40,41,46,[61][62][63] . In our study, key velocity variance increased in both Digital and Chord tasks. As key velocity is directly related to the intensity of the sound produced 37 , our results indicate that muscle fatigue might negatively affect precise control of sound intensity levels. This result is in accordance with the increase of finger force variability in discrete and cyclic force production tasks performed by all four fingers (excluding the thumb) pressing in parallel 64 . This increased variability in fingers' force production could be a strategy to prevent or slower the development of muscle fatigue as previously suggested [65][66][67][68] , and could contribute to the increased variability in key velocity observed in our study.
Although mean values related to note-event accuracy tended to be negatively affected by fatigue in both Digital and Chord tasks, fatigue effect was only significant during the Chord task. Regardless of their group classification, pianists performed significantly more incomplete cycles at the termination of the Chord task than at the initiation. An undershoot of the final position previously reported in extension movement under muscle fatigue condition 42 could be related to note-event accuracy. This result reflects that muscle fatigue affects the accuracy of large amplitude movements such as those in the Chord task involving big leaps across different registers and the execution of fast repetitions of the same notes. The non-significant effect of fatigue in the Digital task may be due to smaller hand range of motion and/or slower rhythm in the excerpt investigated (0.13 s vs. 0.094 s between notes, respectively in the Digital and Chord tasks). Consequently, the effects of fatigue on note-event accuracy may be influenced by the musical style of the selected piece, and further investigation of this aspect should be explored. www.nature.com/scientificreports/ Although mean values of timing variability tended to be higher at the end of the Digital and Chord tasks, absence of a significant effect of fatigue suggests that pianists' control of the inter-onset timing between successive notes is less affected by fatigue than sound intensity and accuracy. Taken together, our results highlight that piano performance parameters were negatively affected by fatigue. Future studies should focus on investigating additional musical parameters that could be affected by fatigue, such as articulation and rhythmic errors.
Our study has some limitations. While the Digital and Chords tasks were based on excerpts from the piano repertoire (Chord task) and an exercise designed to target independent finger activity (Digital task) commonly used in pianists' training 69,70 , their continuous repetition until exhaustion is not representative of typical piano practice. However, this choice allowed to standardise piano performance parameters among participants as in previous studies on the biomechanics of piano performance 24,37 . The arrangement of EMG electrodes is a second limit as the forearm is composed of 20 muscles difficult to record individually. However, we believe that the standard procedure used to place electrodes diminished the precision recording bias. Moreover, it allowed localising forearm area where signs of MMF is more prone to develop (i.e., posterior area). Future studies could then increase the spatial sampling in this particular area to increase precision. Another limitation is that only finger flexion MVC was measured via the handgrip task while finger extension MVC, which may differ between the ShortDuration and LongDuration groups, was not assessed. However, previous studies showed that extensor muscles are highly active during a handgrip test 71 , and that a correlation exists between the isometric and isokinetic torques recorded during wrist flexion and wrist extension 72 that may suggest that both groups had similar wrist flexion and extension MVC. Finally, it must be mentioned that the 60-120-180 Hz ± 1 Hz stop-band second order zero-lag Butterworth filters used to remove 60 Hz electrical contamination and its harmonics also removed EMG components. However, the removal of this narrow components, i.e. 6 Hz out of the 10-400 Hz frequency band of interest of EMG signals, marginally affects the EMG power spectrum as previously shown in Mello et al. 73 . Consequently, this pre-processing step should not have a significant impact on the changes in EMG median frequency as the power spectrum shift towards lower frequencies in conditions causing muscular fatigue would affect all EMG frequency components.
In conclusion, this study showed that continuous repetition of different piano excerpts leads to muscle fatigue. As muscle fatigue caused by repetitive motion of piano could be a precursor of PRMDs, finger and wrist extensors muscles seem more subjected to risk of injuries than flexors muscles, as suggested in epidemiological studies. Finally, muscle fatigue developed through the repetition of movement patterns affected note-event and key velocity musical performance parameters among expert pianists.

Methods
Participants. Fifty expert pianists volunteered to participate in this study. All participants had at least a university (or equivalent) degree or were enrolled in undergraduate or graduate studies in piano performance. Forty three of them were right-handed. Participants were not included if they reported any PRMDs during the year preceding the experimentation. Their characteristics are described in Table 1 ("Results" section). After receiving instructions on the experimental protocol, participants read and signed a written informed consent. The protocol was approved by the Université de Montréal's Ethics Committee (CPER-18-086-D), and all the study was performed in accordance with the STROBE guidelines.
Instrumentation. Participants were equipped with 49 monopolar EMG electrodes of 1.5 mm diameter (TMSi, EJ Oldenzaal, The Netherlands) positioned on the right forearm muscles according to a 7 × 7 array as shown in Fig. 8. Previous studies showed that the hotspot of EMG activation of the forearm depends on the task performed. During wrist extension, radial and ulnar deviation, as well as individual four finger extensions at the metacarpophalangeal joint 74,75 , EMG activation hotspots were distributed over a large surface of the forearm. All these contractions are made by pianists while performing. Therefore, the 49 monopolar EMG electrodes covered the entire forearm. Before electrode placement, forearm skin was scrubbed with 70% isopropyl alcohol pads. A conductive gel was used to improve skin-electrodes conductivity. Electrodes were attached to the skin with circle double-sided tape. Additionally, a medical elastic net bandage (not shown in Fig. 8) was used to avoid electrode detachment. Electrodes were positioned 2 cm apart along columns, which were arranged on the forearm as follows: c1 from the lateral edge of the brachio-radial muscle to the midpoint between the base of index and middle fingers; c2 from the medial-inferior edge of the radial head to the midpoint between the base of middle and ring fingers; c3 from the lateral-inferior edge of the radial head to the lateral-posterior edge of the ulnar head; c4 from the medial edge of the ulnar crest to the medial-anterior edge of the ulnar crest; c5 from the medial-posterior epicondyle to the posterior pisiform; c6 from the medial-anterior epicondyle to the anterior pisiform; and c7 from the medial edge of the bicipital tendon to the anterior aspect of the base of the second metacarpal. The electrode positioning procedure was performed by a trained physiotherapist to ensure reliability and physiological meaning of the data. EMG signals were acquired using a 72-channel Refa amplifier (TMSi, EJ Oldenzaal, The Netherlands) at a sampling rate of 2048 Hz.
An instrumented grand piano (Disklavier DC7X Enspire Pro, Yamaha Canada Music Ltd, Toronto, Canada) was used to record key depression timing and hammer velocity of each key. A digital sound level meter (The 407730 SLM, Extech Instruments, Nashua, USA) was placed at a 1.5 m distance of the back-left side of the piano bench to continuously monitor sound intensity level.
Experimental procedures. Participant's maximal handgrip force was measured using a handgrip dynamometer (Takei Scientific Instruments, Tokyo, Japan). They performed two 5-s maximal voluntary contractions with their right hand interspaced with a 1-min rest period. Verbal encouragement were provided during contractions. www.nature.com/scientificreports/ After a 5-min warm-up piano session, a sound test was performed to familiarize participants with the required sound intensity level. They were then asked to play on a loop without interruption two excerpts named Digital and Chord tasks described thereafter. Both excerpts were played fast and loud to increase muscle activation levels 7,24,76,77 to accelerate the development of muscle fatigue. The music score of the Digital and Chord excerpts were sent to participants at least 5 days prior to the experiment so that they practice them in order to be able to perform them comfortably. 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' spanning a major sixth interval in the middle register of the keyboard (Appendix-A1). Participants were asked to play this excerpt loud (minimum threshold = 78 dB) and fast (quarter note = 112 BPM [beats per minute]) and with no accents in the melody. The Chord task corresponded to a single-chord right-hand sequence on the measure 119 of Franz Liszt's Ballade no 2 in B minor S.171. The chord was composed of three notes repeated five times: once in the middle register, fast repetition in the middle-high register, and fast repetition in the high register of the keyboard (Appendix-A1).
Participants were asked to play this excerpts loud (minimum threshold = 84 dB [middle register] and 94 dB [high register]) and fast (quarter note = 120 BMP) while conserving dynamics and musical intention.
The order of the excerpts was randomly assigned and interspaced with a 15-min rest period. Participants wear an earphone in their left ear that indicated the tempo. A screen was placed in front of the piano to inform participants if sound intensity levels were lower than the minimum loudness threshold. Finally, the rate of perceived exertion was monitored every 30-s via the modified CR-10 Borg scale (mRPE) 48 displayed on the screen. Participants were stopped if they reached twice in a row a score of 7 or after 12 min of continuous playing 78 . Electromyography analysis. Monopolar EMG signals were converted into bipolar EMG signals by subtracting the two closest monopolar signals along columns. EMG data were then filtered using a 10-400 Hz band-pass filter and a 60-120-180 Hz ± 1 Hz stop-band filter to remove 60 Hz electrical contamination and its harmonics. All filters were second order zero-lag Butterworth filters. Data were then zero-aligned by subtracting the mean value, and finally down sampled to 1024 Hz to reduce the time of computation in the subsequent time-frequency analyses. A time-frequency analysis was performed by applying a continuous Morlet wavelet transform (wave number: 7, frequency range: 1-400 Hz in 1 Hz steps) to the pre-processed EMG signals (WavCrossSpec Matlab package 79 ). The instantaneous EMG median frequency was then computed on a time-history basis. The mean of the instantaneous median frequency was calculated for each cycle determined using key depression timing. A cycle refers to each repetition of Digital and Chord excerpts. Data were time normalized, and the mean of cycles' median frequency was calculated during the six intervals described previously. We found 16 EMG bipolar channels out of 2303 (47 muscles × 49 participants) with persistent artifacts in the Digital task and 24 EMG bipolar channels out of 2303 with persistent artifacts in the Chord task. For each interval of abnormal data, median frequency was set to the median value of the group 80 .
Piano performance analysis. Key depression timing and velocity data were used to compute piano performance parameters, which were related to note events, key velocity (a performance feature highly related to sound intensity), and timing (time intervals between the onset of successive notes) [35][36][37][38] . Performance parameters described thereafter were computed for the first and the last 30-s of both Digital and Chord tasks for each participant. The analysis started after the first 10 cycles of each task which were considered as a warm-up period.
Incomplete cycles were defined in relation to the notes/chords events. Due to their different musical characteristics, Digital and Chord tasks were analyzed using a distinct rationale. For the Digital task, a cycle was considered as incomplete when participants missed a note or played a wrong note. For the Chord task, a cycle was considered as incomplete when participants (1) unsuccessfully played the three notes of the chord in all three registers, independently of the successful repetition of the chord in the middle-high and high registers, and (2) when they unsuccessfully performed the repetitions of the highest note of the chord in the middle-high and high registers as the highest note is considered as the most important note of the chord 81,82 .
Key velocity variability was assessed for both Digital and Chord tasks using the key velocity data of the completed cycles. For the Digital task, the key velocity variance was calculated for each of the 16 notes of the excerpt. For the Chord task, the key velocity variance was computed from the velocity of the highest note of each of the 5 chords 81,82 .
Timing variability was assessed for both Digital and Chord tasks using the key depression timing data of the completed cycles. For the Digital task, the timing variance between the 16 notes of the excerpt was calculated. For the Chord task, the timing variance of the highest note of the chord played twice in the middle-high and high registers was computed.

Statistical analysis.
Since the time-to-task termination between participants ranged from 1 to 12 min in Digital task, and from 2 to 12 min in Chord task, we expected different effect of time on the evolution of fatigue and piano performance. Consequently, based on the participants' time-to-task termination, a k-means clustering and a silhouette validity index were used to identify the optimal number of sub-groups for each task. As described in the Results section, the cluster analysis discriminated two groups named ShortDuration and Long-Duration groups. Consequently, in addition to the effect of time, a Group effect and a Group-Time interaction were considered in subsequent statistical analyses. Group comparison was performed for demographic data using t test or F test when appropriate, and a false discovery rate correction was applied to p values.
In terms of fatigue evolution (mRPE and EMG median frequency), a two-way ANOVA on Group (Short-Duration vs LongDuration) × Time (I 0 , I 1 , I 2 , I 3 , I 4 , I 5 ) with repeated measures on the last factor was performed for mRPE and EMG median frequency of each bipolar EMG signal followed by a Tukey post-hoc analysis when appropriate. In the post-hoc analysis, we focused on the differences between I i and I i-1 and between I i and I i-2 only when there was no difference between I i and I i-1 . Finally, to determine the location of the forearm with greater signs of MMF at the end of each task, a one-way ANOVA on Column (c1, c2, c3, c4, c5, c6, c7) was performed on the difference of EMG median frequency between I 5 and I 0 for both groups and both tasks separately. In case of significant effect, a Tukey post-hoc analysis was performed to assess differences between each pair of column. For this analysis, the variation (I 5 -I 0 ) of EMG median frequency was used since the frequency content of the EMG can be affected by factors such as the thickness of the subcutaneous tissues between electrodes and muscles 83 .
In terms of piano performance, a two-way ANOVA on Group (ShortDuration vs LongDuration) × Time (initiation vs termination) with repeated measures on the factor Time was performed for incomplete cycles, key velocity variance, and timing variance.