1. Introduction
Muscular fatigue is a fundamental constraint on sports performance,
particularly when high-intensity actions must be repeated with limited
recovery. Rather than representing a single physiological process,
fatigue arises from interacting peripheral and central mechanisms that
modify the capacity of the neuromuscular system to meet task demands
[1–3].
Importantly, these internal changes do not necessarily produce an
immediate deterioration in observable performance. Skilled athletes
may temporarily preserve force, velocity, or technical outcome despite
fatigue-related physiological alterations, making external performance
alone an incomplete indicator of the underlying neuromuscular state
[3]. This
distinction is particularly relevant in elite sport, where training
status and adaptive motor strategies may delay the point at which
fatigue becomes evident as performance loss.
Surface electromyography (sEMG) provides a non-invasive means of
examining muscle-specific myoelectric manifestations associated with
fatigue during exercise
[4–6].
Amplitude- and frequency-domain features provide complementary, but
not interchangeable, information. Surface-EMG amplitude is influenced
by motor-unit recruitment and discharge behavior, as well as by
action-potential cancellation, electrode position, muscle geometry,
muscle length, and contraction velocity
[6,
7]. Spectral
analysis provides a complementary perspective. During fatiguing
contractions, the EMG power spectrum commonly shifts toward lower
frequencies, reflected by reductions in measures such as median or
mean frequency and associated with changes in muscle-fiber conduction
and motor-unit behavior
[4,
5].
Interpretation is more challenging during dynamic sporting movements
because rapid changes in muscle length, contraction velocity, force,
and electrode–muscle geometry introduce substantial non-stationarity
into the signal
[7].
Time–frequency approaches are therefore particularly useful for
examining spectral behavior within short, movement-specific windows
[8], and spectral
sEMG parameters have shown utility for detecting fatigue-related
myoelectric changes during dynamic swimming exercise
[9]. Accordingly,
spectral changes observed during dynamic exercise are most
appropriately interpreted as fatigue-compatible myoelectric
manifestations rather than as direct quantitative measures of
biological fatigue.
These interpretative challenges are particularly relevant during
repeated explosive sport actions, in which brief maximal efforts are
performed in succession and the external task outcome may remain
stable despite evolving neuromuscular demands. This issue becomes
especially important when performance depends on the integrated
contribution of multiple body segments rather than on the output of a
single muscle or joint. Within this context, repeated maximal shooting
in water polo provides a useful sport-specific model for examining the
potential dissociation between localized myoelectric changes and
external performance. The water polo shot is a high-velocity action
performed without a stable support base and requires coordinated
contributions from body elevation, trunk motion, and upper-limb
acceleration
[10,
11]. Ball
velocity therefore represents the integrated result of a complex
whole-body action. Importantly, increasing fatigue can impair
technical performance without necessarily producing a proportional
decline in shooting velocity
[12].
Previous electromyographic studies have characterized
upper-extremity muscle activation during water polo shooting and
demonstrated differences in activation magnitude, timing, and
proximal-to-distal organization between experienced and inexperienced
performers [13,
14]. Kinematic
studies have additionally documented substantial intraindividual
variability in the organization of the 5-m shot despite comparatively
stable performance outcomes
[15]. However,
these studies have primarily characterized individual shooting actions
or between-group differences. It remains unclear how muscle-specific
myoelectric manifestations evolve during a sequence of repeated
maximal shots, whether fatigue-compatible spectral changes emerge
uniformly or selectively across the throwing musculature, and whether
such changes occur despite preservation of ball velocity. It is
likewise unknown whether repeated shooting is accompanied by
systematic changes in activation amplitude or timing that could
contribute to maintenance of the external performance outcome.
Against this background, the present study examined muscle-specific
myoelectric changes during repeated maximal shooting in professional
water polo players while ball velocity was monitored simultaneously.
Surface EMG was recorded from a 14-muscle throwing-side and trunk
montage across preparation, cocking, acceleration, and deceleration,
with fatigue-compatible spectral changes evaluated within a
standardized late-cocking window. The primary aim was to determine
whether repeated maximal shooting produced selective or generalized
spectral manifestations compatible with fatigue and whether these
changes occurred despite preserved ball velocity. Secondary analyses
examined phase-specific activation amplitude and timing and explored
whether myoelectric change or activation-pattern reorganization was
associated with ball velocity. We hypothesized that fatigue-compatible
spectral changes would be muscle-specific rather than uniform and
could emerge despite preservation of external performance.
2. Materials and Methods
2.1. Study design, participants, and ethics
This single-session repeated-measures study included 14 male
professional field players from one team competing in the Italian A1
league. All athletes were actively training and competing at the
time of testing. Exclusion criteria were acute upper-extremity or
trunk injury, previous surgery to the throwing shoulder,
neurological disorders affecting upper-extremity motor function, or
intolerance to surface-electrode placement. Players with
non-limiting shoulder symptoms were not excluded, provided that they
were able to perform maximal-effort shooting.
The study was conducted in accordance with the Declaration of
Helsinki and was approved by the Comitato Etico per la Ricerca di
Ateneo (CERA; approval no. 2024.55, 14 June 2024). All participants
provided written informed consent before testing.
2.2. Shooting protocol, video acquisition, and ball
velocity
After electrode placement, players completed a warm-up consisting
of freestyle swimming and practice shots. Each player then performed
up to 20 consecutive maximal-effort penalty shots from the official
5-m line. Approximately 5 s separated consecutive attempts, during
which a poolside assistant returned the ball to the player. A shot
was excluded from a specific analysis when the ball bounced before
reaching the goal or when the data required for that analysis were
incomplete. The final EMG dataset comprised 10–19 valid shots per
athlete. The chronological order of retained valid shots was
preserved throughout the analyses; when an observation was
unavailable for a specific downstream analysis, subsequent valid
shots retained their original within-series position and were not
renumbered.
Three GoPro HERO8 cameras (GoPro, San Mateo, CA, USA) recorded
the shooting action at 120 frames·s⁻¹ and were positioned to provide
frontal/goal-line, lateral, and approximately 45° diagonal views.
Before data acquisition, the frontal and lateral cameras were
stereo-calibrated using a checkerboard calibration target. The
synchronized calibrated frontal and lateral recordings were used to
reconstruct the ball trajectory, whereas the diagonal camera
provided an additional view for movement-event identification and
was not used for stereo reconstruction.
Ball release was operationally defined as the first video frame
in which no visible contact between the throwing hand and the ball
remained. Ball velocity was derived from the stereo-reconstructed
trajectory and calculated as the mean of the ball-speed values
obtained over the first four post-release frames, beginning with the
first frame showing complete loss of hand–ball contact. Velocity was
expressed in km·h⁻¹ and assigned to the corresponding athlete and
shot.
Video and EMG recordings were synchronized using a common
physical synchronization event. The cameras recorded the EMG sensor
positioned over the biceps while an operator delivered a brief tap
directly to the sensor. The tap was visually identifiable in the
video recordings and simultaneously generated a sharp transient
artifact in the EMG trace, providing a common temporal reference for
alignment of the video and EMG signals.
Movement events were identified by an operator through
frame-by-frame visual inspection of the synchronized recordings.
Five events were marked: (1) start of preparation, (2) start of
cocking, (3) start of acceleration, (4) ball release/start of
deceleration, and (5) end of deceleration (Figure 1). These events
delimited four movement phases: preparation, cocking, acceleration,
and deceleration. With the exception of ball release, which was
operationally defined as described above, the remaining events were
identified visually from the movement sequence rather than from
predetermined kinematic thresholds.
2.3. Surface EMG acquisition
Surface electromyography was recorded at 2,000 Hz using a MiniX
Waterproof wireless system (Cometa, Italy) and disposable bipolar
Kendall Ag/AgCl surface electrodes. Before electrode placement, the
skin was shaved when necessary, lightly abraded, and cleaned with
alcohol. Electrodes were positioned over the muscle belly and
aligned with the presumed direction of the underlying muscle fibers,
following SENIAM recommendations where applicable and established
recommendations for surface-EMG acquisition and reporting
[16,
17].
Recordings were obtained from 14 muscles on the throwing side:
flexor carpi radialis, biceps brachii, lateral triceps, anterior
deltoid, middle deltoid, posterior deltoid, pectoralis major,
clavicular portion, pectoralis major, sternocostal portion, teres
major, infraspinatus, latissimus dorsi, middle trapezius, lower
rectus abdominis, and thoracic erector spinae. All 14 channels were
included in the phase-specific activation and spectral analyses.
For data acquisition and export, channel labels carried an “R_”
prefix; this designation referred to the throwing side rather than
necessarily to the anatomical right side. Accordingly, EMG was
recorded from the left upper limb in the two left-handed
throwers.
Figure 1: Shooting phases and movement events used to
segment the maximal 5-m water polo shot.
![]()
2.4. Raw-signal quality control and processing
All processing started from the raw, non-rectified sEMG signal.
Samples were classified as invalid when data were missing or when
absolute signal amplitude reached the acquisition-system saturation
threshold of 3,290 digital units. A muscle–shot channel was excluded
when more than 5% of the samples within the analyzed movement were
invalid or when a constant-value segment persisted for at least 100
ms, indicating loss or corruption of the recorded signal. Isolated
invalid samples within otherwise acceptable channels were replaced
by linear interpolation. No channel-level data failing the
predefined quality criteria were imputed.
Raw signals were filtered using a zero-phase frequency-domain
passband from 20 to 450 Hz, with cosine transitions between 15–20 Hz
and 450–500 Hz. Power-line interference was attenuated using a 50-Hz
notch with cosine shoulders extending from 48.5 to 51.5 Hz. A
centered 50-ms RMS envelope was subsequently calculated for
amplitude-domain analyses. These procedures followed established
principles for surface-EMG acquisition, preprocessing, and artifact
control [6,
17].
For each athlete and muscle separately, the normalization
reference was defined as the 95th percentile (P95) of the RMS
envelope pooled across all valid trials. RMS amplitude was then
expressed as a percentage of this athlete- and muscle-specific P95
reference. This task-based normalization provided a robust
within-athlete, within-muscle reference while limiting the influence
of isolated extreme peaks
[18]. Because
a separate P95 reference was calculated for each muscle and athlete,
normalized amplitudes were interpreted within muscle and were not
used as measures of absolute activation magnitude across different
muscles or participants.
2.5. Fatigue-compatible spectral outcome
For each of the 14 recorded muscles, the primary spectral outcome
was the median frequency of the sEMG signal estimated within a fixed
200-ms late-cocking window ending at the onset of acceleration. At
the acquisition frequency of 2,000 Hz, each analysis window
comprised 400 samples. All observed cocking phases were at least
204.5 ms in duration, allowing the same task segment to be examined
across shots.
Median frequency was estimated using a continuous-wavelet
time–frequency approach. Such approaches are particularly suitable
for dynamic contractions in which conventional stationarity
assumptions are difficult to satisfy
[4,
5,
7–9].
Muscle-specific end-to-start changes across the shooting sequence
were modeled while adjusting for within-athlete ball velocity and
concurrent log-transformed RMS amplitude. Inclusion of concurrent
RMS was intended to reduce the possibility that apparent spectral
differences merely reflected concurrent differences in signal
amplitude.
For graphical presentation, shot-by-shot median-frequency changes
were additionally expressed as percentage reduction relative to the
athlete- and muscle-specific mean of the first five valid shots.
These percentage values were used as normalized electrical
indicators of fatigue-compatible change and were not interpreted as
direct percentage measures of biological fatigue.
Robustness of the primary spectral findings was examined using
alternative analytical specifications. These included matched
lower-frequency cutoffs of 20, 30, and 40 Hz, fast Fourier transform
estimation, restriction to active windows with RMS >10% of the
athlete- and muscle-specific P95 reference, low-band trimming,
first-three versus last-three contrasts, leave-one-athlete-out
analyses, and chronological permutation.
2.6. Phase-specific activation, timing, and duration
Preparation, cocking, acceleration, and deceleration were
independently time-normalized from 0% to 100% of their respective
phase duration. For each muscle and shot, mean RMS-envelope
amplitude and the activation-weighted temporal centroid were
calculated separately within each phase.
The temporal centroid represented the relative position within
the phase around which the RMS envelope was temporally concentrated,
with lower values indicating an earlier and higher values a later
distribution of activation. It therefore characterized within-phase
activation timing rather than activation onset, peak timing, or
motor-unit synchronization.
The absolute durations of the cocking and acceleration phases
were additionally retained in milliseconds to evaluate whether
changes in movement-phase duration accompanied repeated shooting or
were associated with ball velocity.
2.7. Integrated activation–fatigue–performance
analysis
For the phase-activation atlas, shot-level values were first
averaged within athlete and then summarized across athletes so that
participants contributing a greater number of valid shots did not
receive disproportionate weight. Changes across the shooting series
were additionally summarized by comparing each athlete’s first five
and last five valid shots for RMS amplitude and activation-weighted
temporal centroid.
To quantify complete-pattern reorganization, each shot was
represented by a 14-muscle × four-phase profile. Amplitude-pattern
reorganization was calculated as the root-mean-square distance
between the shot-specific log-transformed RMS-amplitude profile and
the athlete’s mean first-five profile. Temporal-pattern
reorganization was calculated analogously using phase-normalized
activation-weighted temporal centroids. Larger distance values
therefore indicated greater deviation from the athlete’s
early-series activation pattern. Distances were calculated across
available muscle–phase cells; at least 42 of the 56 possible cells
were required, and missing cells were not imputed.
Two complementary spectral fatigue-burden indices were examined.
The all-muscle burden summarized median-frequency reduction across
all valid recorded muscles relative to the corresponding athlete-
and muscle-specific first-five baseline. Because it incorporated the
complete recorded montage without selecting muscles on the basis of
their individual statistical results, this index was used as the
non-selected complete-montage measure of spectral change.
A secondary localized burden was calculated using only the two
muscles that met the FDR criterion in the preceding 14-muscle
analysis—the clavicular portion of pectoralis major and teres major.
Because these muscles were selected after inspection of the
muscle-specific results, the localized burden was explicitly treated
as a post-identification explanatory measure and not as independent
confirmation of fatigue.
Integrated analyses required at least 10 valid spectral channels
and sufficient phase data to meet the 42-of-56 cell criterion.
Analyses involving the localized burden additionally required valid
spectral data from both FDR-identified muscles. The final integrated
dataset comprised 211 shots from all 14 athletes.
2.8. Statistical analysis
Athletes, rather than individual shots, defined the inferential
clustering unit. Within each athlete, shot position was scaled from
0 for the first retained valid shot to 1 for the last retained valid
shot. When an observation was unavailable for a specific downstream
analysis, subsequent valid shots retained their original position
within the retained shooting sequence and were not renumbered.
Repeated-measures regression models included athlete fixed
effects to account for stable between-athlete differences and linear
and quadratic terms for shot position to allow for non-linear
changes across the shooting series. For each of the 14 muscles,
median frequency was modeled as a function of shot position while
adjusting for within-athlete ball velocity and concurrent
log-transformed RMS amplitude. Muscle-specific end-to-start changes
were obtained from the fitted sequence terms.
Statistical uncertainty for regression estimates was quantified
using cluster-bootstrap resampling at the athlete level, thereby
preserving dependence among repeated shots from the same player.
Bootstrap distributions were used to derive 95% confidence intervals
and two-sided p values for model coefficients.
For the comparison of ball velocity between the first five and
last five valid shots, mean velocity was first calculated within
athlete for each period. The athlete-level mean difference was
reported together with an athlete-cluster bootstrap 95% confidence
interval and an exact two-sided paired sign-flip p value. Because
the confidence interval and p value were obtained from distinct
inferential procedures, their numerical threshold decisions were not
required to coincide exactly.
Multiplicity was controlled using the Benjamini–Hochberg
false-discovery-rate (FDR) procedure
[19]. FDR
correction was applied across the 14 muscle-specific
median-frequency tests and separately within each predefined family
of activation–fatigue–performance models. An FDR-adjusted q value
<0.05 was considered statistically significant.
Phase selectivity was summarized at the athlete level as the
difference between each muscle’s observed highest-amplitude phase
and the mean amplitude of its remaining three phases. These
contrasts were evaluated using paired sign-flip tests, with FDR
correction across the 14 muscles. Athlete-level cluster-bootstrap
95% confidence intervals were additionally calculated for the
phase-selectivity estimates. Because the highest-amplitude phase was
identified from the same observed phase atlas, phase-selectivity
inference was considered secondary and exploratory.
First-five versus last-five changes in phase-specific amplitude
and temporal centroid were treated as descriptive athlete-level
comparisons and summarized with bootstrap confidence intervals.
All analyses were performed in Python using reproducible analysis
scripts. The processed datasets, variable definitions, and complete
statistical outputs are provided in the accompanying analysis
workbook.
3. Results
3.1. Participants and data completeness
The 14 professional players were 25.7 ± 6.1 years old, 184.1 ±
6.2 cm tall, and 89.2 ± 8.9 kg, with 14.6 ± 6.1 years of playing
experience. Twelve were right-handed and two were left-handed
throwers (Table 1).
The final dataset comprised 222 valid EMG shots, of which 219 had
matched ball-velocity measurements. Across the 14-muscle montage,
2,957 of 3,108 possible late-cocking muscle–shot windows (95.1%)
satisfied the predefined signal-quality criteria. The
phase-activation atlas contained 11,876 of 12,432 possible
muscle–phase–shot observations (95.5%), with 9–14 athletes
contributing to individual muscle–phase cells. The integrated
activation–fatigue–performance analyses included 211 shots from all
14 athletes
(Supplementary
Table S1).
Table 1: Participant characteristics.
| Characteristic |
Value |
| Players |
14 male professional field players |
| Age, years |
25.7 ± 6.1 |
| Height, cm |
184.1 ± 6.2 |
| Body mass, kg |
89.2 ± 8.9 |
| BMI, kg·m⁻² |
26.3 ± 2.2 |
| Playing experience, years |
14.6 ± 6.1 |
| Throwing-side dominance |
12 right; 2 left |
3.2. Ball velocity was preserved
Across the 219 velocity-matched shots, the descriptive shot-level
mean ball velocity was 82.07 ± 11.89 km·h⁻¹. At the athlete level,
mean velocity was 79.47 km·h⁻¹ during the first five valid shots and
82.00 km·h⁻¹ during the last five, corresponding to a mean
difference of +2.53 km·h⁻¹ (bootstrap 95% CI 0.28 to 4.76; exact
paired sign-flip p=0.055). Thus, repeated maximal shooting was not
accompanied by a decline in ball velocity.
Across ordinal shot positions, velocity showed an initial rise
followed by later attenuation rather than a progressive monotonic
decrease (Figure 2).
3.3. Phase-specific activation across the 14-muscle
montage
The four-phase atlas demonstrated marked within-muscle phase
dependence across the recorded montage (Figure 3;
Supplementary
Figure S1). Eleven of the 14 muscles met the FDR
criterion in the secondary phase-selectivity analysis.
Seven muscles showed their strongest relative activation during
acceleration: flexor carpi radialis (+59.79 percentage points [pp],
q=0.0012), pectoralis major, clavicular portion (+37.27 pp,
q=0.0012), latissimus dorsi (+36.40 pp, q=0.0014), lateral triceps
(+31.92 pp, q=0.0012), pectoralis major, sternocostal portion
(+31.47 pp, q=0.0012), lower rectus abdominis (+29.99 pp, q=0.0012),
and teres major (+20.91 pp, q=0.0248).
Preparation selectivity was observed for middle deltoid (+18.89
pp, q=0.0056) and middle trapezius (+14.74 pp, q=0.0248), whereas
infraspinatus (+15.44 pp, q=0.0296) and thoracic erector spinae
(+13.24 pp, q=0.0492) showed deceleration selectivity. No other
amplitude phase-selectivity contrast met the FDR criterion
(Supplementary
Table S2).
Because RMS amplitude was normalized using a separate athlete-
and muscle-specific P95 reference, the values displayed in Figure 3A
describe within-muscle phase organization and should not be
interpreted as absolute activation rankings between different
muscles.
3.4. Fatigue-compatible spectral changes across the 14-muscle
montage
The same 200-ms late-cocking window, covariate adjustment, and
FDR-controlled analysis were applied to all 14 recorded muscles
(Figure 4; Table 2;
Supplementary
Table S3). Only two muscles showed a reliable downward
shift in median frequency across the shooting series. Teres major
showed an adjusted end-to-start change of −14.62 Hz (95% CI −22.71
to −5.89; q=0.0042), whereas pectoralis major, clavicular portion
showed a change of −10.08 Hz (95% CI −17.18 to −4.93; q=0.0028).
These were the only two muscles that met the FDR criterion across
the complete 14-muscle analysis. Infraspinatus (−8.32 Hz),
pectoralis major, sternocostal portion (−6.20 Hz), lower rectus
abdominis (−4.78 Hz), and latissimus dorsi (−2.46 Hz) showed
negative point estimates, but their confidence intervals included
zero and none survived FDR correction (all q≥0.243). The remaining
recorded muscles likewise showed no reliable downward
median-frequency shift.
Notably, strong acceleration-phase activation selectivity did not
necessarily coincide with fatigue-compatible spectral change. Flexor
carpi radialis and lateral triceps showed two of the largest
acceleration-selectivity contrasts, yet neither showed a significant
reduction in median frequency. The shot-by-shot evolution of the two
FDR-identified muscles is shown in Figure 5A.
Figure 2: Ball velocity during repeated maximal
shooting. (A) First versus last five valid shots. (B) Velocity
across ordinal shot position; the number of contributing athletes
is shown above each position.
![]()
Table 2: Complete 14-muscle fatigue-compatible
spectral analysis.
| Throwing-side muscle |
Adjusted end-to-start
change |
95% CI |
FDR q |
| Teres major |
−14.62 Hz |
−22.71 to −5.89 |
0.0042 |
| Pectoralis major, clavicular
portion |
−10.08 Hz |
−17.18 to −4.93 |
0.0028 |
| Infraspinatus |
−8.32 Hz |
−23.51 to +7.55 |
0.624 |
| Pectoralis major, sternocostal
portion |
−6.20 Hz |
−12.26 to +0.04 |
0.243 |
| Lower rectus abdominis |
−4.78 Hz |
−10.72 to +0.94 |
0.331 |
| Latissimus dorsi |
−2.46 Hz |
−6.51 to +2.02 |
0.612 |
| Biceps brachii |
−1.30 Hz |
−5.26 to +2.66 |
0.812 |
| Lateral triceps |
−1.05 Hz |
−9.42 to +9.02 |
0.912 |
| Posterior deltoid |
−0.71 Hz |
−5.72 to +2.54 |
0.899 |
| Middle deltoid |
−0.69 Hz |
−7.05 to +3.22 |
0.912 |
| Anterior deltoid |
+0.05 Hz |
−10.18 to +9.15 |
0.958 |
| Flexor carpi radialis |
+1.25 Hz |
−5.96 to +9.02 |
0.899 |
| Middle trapezius |
+1.87 Hz |
−3.59 to +8.44 |
0.812 |
| Thoracic erector spinae |
+4.81 Hz |
−1.07 to +11.30 |
0.331 |
3.5. Integrated activation–fatigue–performance
analysis
The non-selected all-muscle fatigue burden showed no reliable
change across the shooting sequence (β=+0.42 standardized units, 95%
CI −1.36 to +2.06; q=0.946). It was also not associated with
amplitude-pattern distance (β=−0.13, 95% CI −0.36 to +0.12; q=0.946)
or temporal-pattern distance (β=−0.01, 95% CI −0.22 to +0.16;
q=0.946).
In the corresponding velocity model, ball velocity was not
independently associated with the all-muscle fatigue burden (+0.26
km·h⁻¹ per within-athlete SD, 95% CI −0.59 to +1.15; q=0.846),
amplitude-pattern distance (+0.59 km·h⁻¹, 95% CI −0.73 to +1.84;
q=0.846), or temporal-pattern distance (+0.03 km·h⁻¹, 95% CI −1.11
to +1.00; q=0.944). Neither the all-muscle fatigue ×
amplitude-pattern interaction (−0.32 km·h⁻¹, 95% CI −1.58 to +0.66;
q=0.846) nor the all-muscle fatigue × temporal-pattern interaction
(+0.13 km·h⁻¹, 95% CI −1.17 to +1.28; q=0.944) met the FDR criterion
(Table 3;
Supplementary
Table S4).
The secondary localized burden, calculated from the clavicular
portion of pectoralis major and teres major after these muscles had
been identified in the preceding 14-muscle FDR analysis, increased
across shot position (β=+1.78 standardized units, 95% CI +0.43 to
+2.97; q=0.030). Because this index was defined post-identification,
this sequence effect was not interpreted as independent confirmation
of fatigue.
In the localized performance model, neither the localized fatigue
burden (−0.01 km·h⁻¹ per within-athlete SD, 95% CI −0.67 to +0.59;
q=0.997), amplitude-pattern distance (+0.41 km·h⁻¹, 95% CI −0.89 to
+1.76; q=0.817), nor temporal-pattern distance (−0.004 km·h⁻¹, 95%
CI −0.92 to +0.89; q=0.997) was independently associated with ball
velocity. Likewise, neither the localized fatigue ×
amplitude-pattern interaction (+0.50 km·h⁻¹, 95% CI −0.94 to +1.64;
q=0.761) nor the localized fatigue × temporal-pattern interaction
(−1.07 km·h⁻¹, 95% CI −2.46 to +0.28; q=0.277) met the FDR criterion
(Figure 5C; Table 3).
Thus, neither the complete-montage nor the localized analysis
identified a systematic EMG reorganization that statistically
accounted for preserved ball velocity.
3.6. Phase-specific activation timing and phase
duration
The activation-weighted temporal-centroid analysis showed
distinct within-phase timing patterns across the recorded muscles
(Figure 3C). During cocking, pectoralis major, clavicular portion
(73.3% of phase), flexor carpi radialis (70.2%), pectoralis major,
sternocostal portion (69.9%), and lower rectus abdominis (66.9%)
showed activation distributions weighted toward the later portion of
the phase.
During acceleration, middle deltoid (61.9%), posterior deltoid
(60.4%), thoracic erector spinae (58.8%), middle trapezius (58.2%),
and infraspinatus (57.7%) showed later temporal centroids, whereas
both pectoralis major portions were weighted earlier within the
phase (clavicular portion, 42.0%; sternocostal portion, 42.3%).
Acceleration duration did not change reliably across the shooting
series (adjusted end-to-start change −2.58 ms, 95% CI −11.08 to
+6.51; q=0.589). Cocking duration showed a larger negative point
estimate (−29.95 ms, 95% CI −73.29 to +1.86), but this change also
did not meet the FDR criterion (q=0.153).
Neither phase duration was associated with ball velocity. Each
additional 10 ms of acceleration was associated with +0.11 km·h⁻¹
(95% CI −0.56 to +0.73; q=0.729), whereas each additional 10 ms of
cocking was associated with −0.08 km·h⁻¹ (95% CI −0.27 to +0.10;
q=0.595). Thus, preserved ball velocity was not explained by a
systematic change in cocking or acceleration duration. The
corresponding performance-model estimates are summarized in Table
3.
3.7. Phase-specific changes across the shooting
series
First-five to last-five changes in normalized RMS amplitude were
heterogeneous across muscles and phases rather than uniformly
increasing or decreasing (Figure 3B). Descriptively, the largest
late-series increases were observed during deceleration for
latissimus dorsi (+26.4 pp), pectoralis major, sternocostal portion
(+15.4 pp), teres major (+11.7 pp), and lateral triceps (+8.9 pp),
whereas thoracic erector spinae increased during acceleration (+12.1
pp).
In contrast, acceleration-phase amplitude decreased for flexor
carpi radialis (−4.9 pp), middle deltoid (−4.4 pp), biceps brachii
(−3.8 pp), middle trapezius (−3.6 pp), and anterior deltoid (−3.6
pp). Because amplitudes were normalized using athlete- and
muscle-specific P95 references, these changes describe within-muscle
variation across the series and should not be interpreted as
absolute differences between muscles.
Changes in activation-weighted temporal centroid were smaller and
likewise showed no common direction (Figure 3D). The largest later
shifts were observed for flexor carpi radialis during deceleration
(+5.0% of phase) and pectoralis major, clavicular portion during
acceleration (+3.6%), whereas biceps brachii shifted earlier during
cocking (−3.4%).
Overall, the first-five to last-five comparisons showed
distributed phase-specific changes in activation amplitude and
timing without a consistent montage-wide pattern. Figure 5B provides
a focused visualization of the first-five and last-five
phase-specific activation profiles of the two FDR-identified
muscles.
Figure 3: Phase-specific activation and temporal
organization across the 14-muscle montage. (A) RMS amplitude. (B)
First-five to last-five RMS-amplitude change. (C)
Activation-weighted temporal centroid. (D) First-five to last-five
temporal-centroid shift. pp, percentage points.
![]()
Figure 4: Fatigue-compatible spectral changes across
the 14-muscle montage. (A) Shot-by-shot median-frequency reduction
relative to the first-five reference. (B) Adjusted end-to-start
change in median frequency; highlighted muscles survived FDR
correction.
![]()
Table 3: Integrated activation-fatigue-performance and
phase-duration models.
| Outcome |
Estimate |
95% CI |
FDR q |
Interpretation |
| All-muscle fatigue burden → velocity |
+0.26 km·h⁻¹ per SD |
−0.59 to +1.15 |
0.846 |
No independent association |
| Amplitude-pattern distance → velocity,
all-muscle model |
+0.59 km·h⁻¹ |
−0.73 to +1.84 |
0.846 |
No independent association |
| Temporal-pattern distance → velocity,
all-muscle model |
+0.03 km·h⁻¹ |
−1.11 to +1.00 |
0.944 |
No independent association |
| All-muscle fatigue × amplitude-pattern
distance |
−0.32 km·h⁻¹ |
−1.58 to +0.66 |
0.846 |
No interaction |
| All-muscle fatigue × temporal-pattern
distance |
+0.13 km·h⁻¹ |
−1.17 to +1.28 |
0.944 |
No interaction |
| Localized fatigue burden → velocity |
−0.01 km·h⁻¹ per SD |
−0.67 to +0.59 |
0.997 |
No independent association |
| Amplitude-pattern distance → velocity,
localized model |
+0.41 km·h⁻¹ |
−0.89 to +1.76 |
0.817 |
No independent association |
| Temporal-pattern distance → velocity,
localized model |
−0.004 km·h⁻¹ |
−0.92 to +0.89 |
0.997 |
No independent association |
| Localized fatigue × amplitude-pattern
distance |
+0.50 km·h⁻¹ |
−0.94 to +1.64 |
0.761 |
No interaction |
| Localized fatigue × temporal-pattern
distance |
−1.07 km·h⁻¹ |
−2.46 to +0.28 |
0.277 |
No interaction |
| Acceleration duration → velocity |
+0.11 km·h⁻¹ per 10 ms |
−0.56 to +0.73 |
0.729 |
No reliable association |
| Cocking duration → velocity |
−0.08 km·h⁻¹ per 10 ms |
−0.27 to +0.10 |
0.595 |
No reliable association |
3.8. Robustness analyses
The primary spectral findings were robust across alternative
analytical specifications. Across the complete-montage sensitivity
framework, all amplitude-adjusted standardized end-to-start
contrasts remained negative and FDR-significant when
continuous-wavelet and Fourier estimators, lower-frequency cutoffs
of 20, 30, and 40 Hz, active-window restriction, and low-band
trimming were applied (all q≤0.0012;
Supplementary
Table S5).
At the muscle-specific level, the two muscles identified in the
primary analysis remained affected under the conservative ≥40-Hz
analysis with concurrent amplitude adjustment: pectoralis major,
clavicular portion showed an adjusted change of −5.69 Hz (q=0.0008),
and teres major showed −12.79 Hz (q=0.0165).
Leave-one-athlete-out analyses yielded negative end-to-start
estimates in every repetition for the complete-montage standardized
contrast and for both FDR-identified muscles, indicating that the
direction of the primary spectral findings was not driven by any
single athlete.
The chronological permutation analysis was directionally
consistent with the primary findings but did not independently meet
the FDR criterion (q=0.080). It was therefore considered a secondary
robustness check rather than independent statistical
confirmation.
4. Discussion
The present study examined the myoelectric response to repeated
maximal shooting in 14 professional water polo players using a
14-muscle throwing-side and trunk sEMG montage synchronized with
movement-phase identification and ball-velocity assessment.
Fatigue-compatible spectral changes were evaluated within a
standardized late-cocking window, while phase-specific activation
amplitude, temporal organization, movement-phase duration, and
multimuscle pattern reorganization were examined across the shooting
sequence. The main findings were that ball velocity remained
preserved, whereas significant reductions in median frequency emerged
selectively in the clavicular portion of pectoralis major and teres
major. Changes in activation amplitude and timing were heterogeneous
across the recorded musculature, and neither complete-montage nor
localized measures of EMG reorganization were systematically
associated with ball velocity. Together, these findings indicate that
localized myoelectric alterations can develop during repeated maximal
sport-specific actions without a concomitant deterioration in the
primary external performance outcome.
The dissociation between preserved ball velocity and localized
fatigue-compatible spectral change is one of the most relevant
observations of the present study. Maintenance of the task outcome did
not imply stability of the underlying myoelectric state. This
interpretation is consistent with contemporary views of fatigability,
according to which physiological adjustments may develop before they
become evident as task failure or measurable performance loss
[3]. A similar
dissociation has previously been reported in water polo by Royal et
al. [12], who
found that increasing fatigue impaired aspects of shooting performance
and technique while ball velocity remained comparatively resistant to
deterioration. The present results extend this concept by showing that
muscle-specific spectral alterations may also emerge despite
preservation of ball speed. Importantly, the spectral effects
identified here were not dependent on a single analytical
specification: their direction was maintained across alternative
lower-frequency cutoffs, wavelet and Fourier estimators, concurrent
amplitude adjustment, active-window restriction, and
leave-one-athlete-out analyses. The findings therefore support the
possibility that, during a relatively short sequence of maximal
sport-specific efforts, local myoelectric alterations may precede any
detectable reduction in the primary performance variable.
The selective involvement of the clavicular portion of pectoralis
major and teres major is biomechanically plausible, although the
present data do not establish why these two muscles were
preferentially affected. Both contribute to forceful shoulder actions
relevant to overhead propulsion, including humeral adduction and
internal rotation. Classic electromyographic investigations of
throwing have identified pectoralis major among the muscles strongly
activated during the propulsive phase, while skilled throwers appear
to organize shoulder-muscle activity differently from less experienced
performers [20].
More broadly, the throwing kinetic chain transfers motion and force
generated by proximal segments through the shoulder and upper limb to
the ball [21].
Experimental work on shoulder internal-rotation tasks has also
demonstrated coordinated involvement of pectoralis major and teres
major [22], and
sport-dependent adaptations in shoulder-muscle activation during
fatiguing internal-rotation exercise have been described in elite
aquatic athletes, including water polo players
[23]. In the
present phase atlas, both the clavicular portion of pectoralis major
and teres major showed acceleration-selective activation, further
supporting the biomechanical plausibility of their involvement.
However, the spectral outcome was measured during the final 200 ms of
cocking, immediately before acceleration. The findings should
therefore be interpreted as evidence that fatigue-compatible
myoelectric alterations in these muscles were detectable before the
propulsive phase, rather than as evidence that fatigue was generated
specifically during acceleration.
Another important finding was that high relative activation did not
necessarily correspond to greater fatigue-compatible spectral change.
Flexor carpi radialis and lateral triceps showed some of the strongest
acceleration-phase selectivity, yet neither demonstrated a reliable
reduction in median frequency. Conversely, the two muscles showing
significant spectral changes did not exhibit a uniform increase in RMS
amplitude across the shooting sequence. This distinction is
physiologically and methodologically important because amplitude- and
frequency-domain sEMG measures provide complementary but
non-equivalent information. Surface-EMG amplitude is influenced by
motor-unit recruitment and discharge behavior, but also by
action-potential cancellation, electrode–muscle geometry, muscle
length, contraction velocity, and other factors that become
particularly relevant during dynamic movements
[4,
6,
7]. Spectral
variables are likewise not direct readouts of a single physiological
mechanism, because changes in muscle-fiber conduction,
action-potential characteristics, motor-unit behavior, and recording
geometry can all influence their values
[24].
Accordingly, the present findings do not support a simple model in
which repeated shooting produces fatigue that is accompanied by a
uniform increase in RMS amplitude across the throwing musculature.
Rather, the amplitude and spectral domains captured different aspects
of the evolving myoelectric response, reinforcing the value of
examining both during dynamic sport-specific tasks
[4,
9].
The integrated analyses did not identify a single, common
reorganization of the recorded EMG pattern that accounted for
preservation of ball velocity. Neither the non-selected all-muscle
fatigue burden nor amplitude-pattern or temporal-pattern distance was
independently associated with ball velocity, and the corresponding
interaction terms were also null. The secondary localized model led to
the same conclusion. These findings should not be interpreted as
evidence that compensatory adjustments did not occur. Surface EMG
captures only part of the neuromuscular coordination underlying a
complex whole-body task, and inferring neural control strategies from
changes in multimuscle surface signals requires caution
[25]. Moreover,
fatigue-related adaptation does not necessarily require wholesale
reorganization of muscle coordination. Turpin et al.
[26], for
example, showed during cyclic exercise that fatigue-related
adaptations could occur largely through changes in activation level
while broader coordination patterns remained comparatively stable. In
water polo, the range of possible mechanical solutions may be even
greater because ball velocity depends on body elevation, trunk
rotation, scapular control, upper-limb sequencing, and lower-limb
support [10,
11,
21]. The
heterogeneous first-to-last changes observed across muscles and
phases, together with the absence of a montage-wide statistical
pattern, are therefore more consistent with variable or
athlete-specific adjustments than with a single universal compensation
strategy. This interpretation is also compatible with previous
evidence of substantial intraindividual movement variability during
the 5-m water polo shot despite comparatively stable performance
outcomes
[15].
The temporal findings further argue against a simple mechanical
explanation for preserved ball speed. Neither acceleration nor cocking
duration changed reliably across the shooting series, and neither
duration was associated with velocity. Players therefore did not
appear to preserve performance by systematically extending the
acceleration phase or by producing a consistent change in gross
cocking duration. Activation-weighted temporal-centroid changes were
likewise heterogeneous and generally small, with no common shift
across the muscle montage. Ball velocity itself showed an initial rise
followed by later attenuation rather than a progressive monotonic
decline. Several mechanisms could theoretically contribute to such a
trajectory, including familiarization, task-specific potentiation,
progressive optimization of the shooting action, or competing effects
of potentiation and accumulating fatigue. These mechanisms were not
directly measured, however, and cannot be distinguished from the
present dataset. The appropriate interpretation is therefore more
limited: despite repeated maximal efforts and localized
fatigue-compatible spectral changes, neither gross phase duration nor
a common temporal reorganization of recorded muscle activity explained
preservation of ball velocity.
Taken together, the present findings suggest short-term performance
resilience during repeated maximal shooting in professional water polo
players. The observable task outcome remained stable despite localized
alterations in the myoelectric signal, indicating that preserved ball
velocity should not be equated with absence of neuromuscular change.
At the same time, the results should not be interpreted as evidence
that the athletes were globally unfatigued or that the clavicular
portion of pectoralis major and teres major alone determined the
physiological cost of the task. The complete 14-muscle analysis
instead showed that fatigue-compatible spectral manifestations were
selective within the recorded montage, whereas activation amplitude
and temporal organization changed heterogeneously across muscles and
phases. The study therefore supports a distinction between external
performance and local myoelectric state during repeated explosive
sport actions. This interpretation remains physiological rather than
causal: the present design demonstrates concurrent evolution of
myoelectric characteristics and performance, but it cannot determine
the specific mechanisms by which ball velocity was maintained.
4.1. Practical applications
Figure 5: Localized myoelectric change and performance
integration. (A) Shot-by-shot spectral evolution of the two
FDR-identified muscles. (B) Descriptive first-five versus
last-five phase-specific activation profiles. (C) Exploratory
post-identification localized performance model.
![]()
From an applied perspective, ball velocity alone may
underestimate the neuromuscular cost of repeated maximal shooting.
In the present professional cohort, velocity remained preserved
while two shoulder muscles already showed robust fatigue-compatible
spectral changes. This finding does not imply that multichannel sEMG
should routinely be used during training, but it suggests that
stable ball speed should not automatically be interpreted as
evidence of an unchanged muscular state. Shot volume, inter-shot
recovery, technical quality, perceived exertion, body elevation, and
shoulder symptoms may therefore provide useful complementary
information when repeated maximal-shooting sets are prescribed.
The selective involvement of the clavicular portion of pectoralis
major and teres major may also be relevant when considering
conditioning and recovery strategies for repeated overhead shooting.
However, these findings should not be translated into isolated
strengthening or injury-prevention prescriptions. Water polo
shooting depends on an integrated kinetic chain
[21], and
previous work in aquatic athletes indicates that fatigue-related
shoulder activation patterns can differ according to sport-specific
adaptation
[23]. The
present results therefore support individualized monitoring of
repeated-shooting load and recovery rather than a universal
intervention focused on two muscles.
4.2. Strengths and limitations
Several features strengthen the present study. First, the
protocol examined professional athletes performing a sport-specific
maximal action rather than an isolated laboratory contraction.
High-sampling-rate sEMG was synchronized with multi-view video and
stereo-derived ball velocity, allowing local myoelectric behavior to
be examined together with the external performance outcome. Second,
the same 14-muscle montage was evaluated using explicit
signal-quality criteria, and the primary spectral analysis was
applied consistently across all recorded muscles. Third, the models
accounted for concurrent RMS amplitude and within-athlete ball
velocity, controlled multiplicity across the complete muscle set,
and quantified uncertainty at the athlete level rather than treating
repeated shots as statistically independent observations. The
primary findings were also challenged through multiple alternative
spectral specifications and leave-one-athlete-out analyses. Finally,
separating the non-selected all-muscle burden from the
post-identification localized burden reduced the risk of presenting
the latter as independent confirmation of the two muscle-specific
findings.
Several limitations nevertheless constrain interpretation. First,
the study included only 14 male professional players from a single
team. Although the repeated-shot design generated a relatively large
number of observations, the number of independent athletes remained
small, particularly for interaction analyses. The results therefore
cannot automatically be generalized to female athletes, younger
players, lower competitive levels, or athletes from other teams and
training backgrounds. Second, surface EMG recorded during a rapid
aquatic overhead movement is particularly vulnerable to changes in
electrode–muscle geometry, motion artifact, amplitude cancellation,
and cross-talk
[6,
7,
25]. This is
especially relevant when interpreting individual shoulder muscles,
including teres major, and reinforces the need to describe the
observed spectral changes as fatigue-compatible myoelectric
manifestations rather than direct measures of local biochemical
fatigue. Third, the study did not include direct measures of central
or peripheral fatigue, muscle-fiber conduction velocity, metabolic
state, perceived exertion, or local muscle oxygenation. The
physiological mechanisms underlying the observed spectral changes
therefore cannot be determined from the present dataset. Fourth, P95
amplitude normalization was specific to each athlete and muscle and
consequently supports within-muscle comparisons across the shooting
sequence but does not permit interpretation of absolute activation
magnitude between different muscles. Finally, the shooting sequence
represented a relatively brief repeated-effort protocol with
approximately 5 s between shots. The findings should therefore be
interpreted as characterizing short-term responses to repeated
maximal shooting rather than the neuromuscular consequences of an
entire match or prolonged training session.