Accepted Manuscript

Muscle synergies during bench press are reliable across days

Mathias Kristiansen, Afshin Samani, Pascal Madeleine, Ernst Albin Hansen

PII: S1050-6411(16)30052-9
DOI: http://dx.doi.org/10.1016/j.jelekin.2016.06.004
Reference: JJEK 1982

To appear in: Journal of Electromyography and Kinesiology

Received Date: 25 September 2015

Revised Date: 11 May 2016
Accepted Date: 8 June 2016

Please cite this article as: M. Kristiansen, A. Samani, P. Madeleine, E.A. Hansen, Muscle synergies during bench
press are reliable across days, Journal of Electromyography and Kinesiology (2016), doi: http://dx.doi.org/10.1016/
j.jelekin.2016.06.004

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Muscle synergies during bench press are reliable across days

Authors

Mathias Kristiansen1, Afshin Samani1, Pascal Madeleine1, Ernst Albin Hansen1

Institution
1
Center for Sensory-Motor Interaction (SMI), Department of Health Science and Technology,
Aalborg University, Aalborg, Denmark

Corresponding Author:
Mathias Kristiansen, MSc, Physical Activity and Human Performance group, Center for
Sensory-Motor Interaction (SMI), Department of Health Science and Technology, Aalborg
University, Fredrik Bajers Vej 7 E2, DK-9220, Aalborg, Denmark.
Tel +45 26803461

Fax +45 98154008

E-mail: mvk@hst.aau.dk

Keywords:

Bilateral arm movement, Motor modules, Muscle coordination, Neural adaptations,

Repeatability, Strength training

1
1. Introduction

For successful execution of movement, both timing and pattern of activation of all involved

muscles need to be well coordinated. However, at the moment there are unknown aspects of

how the central nervous system controls the muscles involved in movements. One theory on

how the central nervous system controls human movement, infers that movements to a large

extent are controlled by a combination of a few basic activation patterns, also known as

motor modules or muscle synergies [Bernstein, 1967; Ivanenko et al, 2006; Lacquaniti et al,

2012]. A muscle synergy can be characterized as a low dimensional organizational structure

controlling multiple muscles. These neural coordinative structures are thought to be located at

spinal level, and to be controlled by motor cortical areas and affected by afferent systems

[Bizzi and Cheung, 2013]. Muscle synergies have thus been suggested to provide a simplified

strategy for the nervous system to control movements [Bernstein, 1967; D'Avella and Bizzi,

2005; Hug, 2010; Ivanenko et al, 2006; Torres-Oviedo et al, 2006; Torres-Oviedo and Ting,

2007]. Indeed, a few basic patterns have been shown to adequately describe a number of

various movements in humans such as reaching in the horizontal plane [Muceli et al, 2010;

Muceli et al, 2014] standing [Krishnamoorthy et al, 2003; Torres-Oviedo and Ting, 2007],

walking [Ivanenko et al, 2004; MacLellan et al, 2014; Oliveira et al, 2014], running

[Cappellini et al, 2006], pedaling [Dorel et al, 2009; Hug et al, 2010; Hug et al, 2011], rowing

[Turpin et al, 2011a; Turpin et al, 2011b], bench press [Kristiansen et al, 2015b], and

backward giant swing [Frre and Hug, 2012].

To study inter-muscular coordination during movements, muscle synergies can be

extracted from multiple surface electromyography (EMG) signals using a nonnegative matrix

2
factorization algorithm. The extraction of muscle synergies offers unique insight into the

combined timing and activation patterns of multiple muscles during movements. Muscle

synergies are reported to be consistent across a variety of postural perturbations in cats

[Torres-Oviedo et al, 2006] and to be robust in human balance control across different

biomechanical contexts in humans [Torres-Oviedo and Ting, 2010]. Consistency has also

been reported for pedaling [Hug et al, 2011].

However, there are to the best of the authors knowledge not any reports available on

within- and between-day reliability of the method. Thus, the aim of the study was to evaluate

the between-day reliability of applying nonnegative matrix factorization to EMG data

collected during bench press. Considering the consistency and robustness of muscle

synergies, we hypothesized that muscle synergies describing bench press would be reliable

across days. Investigations reporting both the absolute and relative reliability are needed to

provide evidence that muscle synergy analyses based on nonnegative matrix factorization are

clinically and experimentally sound. The presentation of this reliability study follows the

guidelines for reporting reliability and agreement studies [Kottner et al, 2011].

2. Materials and methods

2.1. Participants

Healthy male individuals (n=21, age 24.52.2 years (meanstandard deviation (SD)), height

1.810.07 m, body mass at first and second test session 88.513.1 kg and 89.012.8 kg, three

repetition maximum (3RM) in bench press at first and second test session 109.226.1 kg and

109.425.9 kg) volunteered for participation in the current study. At the time of data

collection, all participants had performed full body strength training for 2-3 times per week

for at least two years. The number of participants was determined using an level set to 0.05,

3
 level to 0.20, p0 to 0.7, p1 to 0.9, and n to 2 [Walter et al, 1998]. p0 and p1 denotes the

minimally acceptable level of reliability and the expected level of reliability, respectively.

level and level denotes the probability of making a type 1 and type 2 error, respectively. All

participants gave their written informed consent after having been explained the experimental

methods and risks. The study was approved by the local ethics committee of North Denmark

Region (N-20120036).

Table 1 near here

2.2. Experimental procedure

As a means to study the reliability of extracting muscle synergies during a strength training

movement, we applied the common exercise of bench press. For trained individuals, bench

press is a demanding and complex bilateral arm movement. Proper execution requires the

coordinated activation of all major muscle groups in the legs, back and upper body, while

handling a heavy load. The study consisted of three sessions. The purpose of the first session

was to familiarize the participants with the test protocol, laboratory environment and test

equipment, as well as to minimize learning effects in the two subsequent sessions.

Approximately one week after the familiarization session, all participants performed two test

sessions for investigation of between-day reliability of muscle synergies during bench press.

The time interval between the first and the second test session was on average 8.22.9 days.

In the test sessions, the participants performed the following: warm up, a 3RM test in bench

press, one set of 3 repetitions at 75% of the 3RM load for normalization, and then three sets

of eight repetitions at 60% of the 3RM load for investigation of reliability. It was necessary to

apply a submaximal intensity level during the bench press that was used for investigation of

reliability as a minimum of 20-40 cycles are required during cyclic tasks to obtain

representative EMG data [Hug, 2010; Oliveira et al, 2014]. As there are no conclusive

recommendations on how to normalize EMG data, we used a task-specific submaximal

4
dynamic normalization procedure, which we have previously used in a similar experimental

setup [Kristiansen et al, 2015b]. Briefly, the task-specific submaximal dynamic normalization

procedure consisted of recording the maximal surface EMG envelope on the below

mentioned muscles during the execution of a submaximal bench press at 75% of 3RM. The

obtained value in this procedure was then used as a normalization factor.

2.3. Test sessions

First, the involved skin areas were shaved and cleaned with alcohol. Then, surface EMG

electrodes (Ambu Neuroline 720 01-K/12, Ag/AgCl, inter electrode distance 20 mm, Ambu

A/S, Ballerup, Denmark) were placed on the skin over the following muscles on the right side

of the body: pectoralis major (PM), anterior deltoideus (AD), biceps brachii (BB), triceps

brachii, lateral head (TBL), triceps brachii, medial head (TBM), latissimus dorsi (LD),

erector spinae (ES), rectus femoris (RF), biceps femoris (BF), gastrocnemius lateral head

(GML), soleus (SOL), vastus lateralis (VL), and vastus medialis (VM). The electrodes were

mounted along the muscle fiber direction in a bipolar configuration. Most of the electrodes

were mounted according to the SENIAM recommendations [Hermens et al, 2000]. For PM

and LD, which are not listed by SENIAM, the electrodes were mounted four fingerbreadths

below the clavicle, medial to the anterior axillary border and 3 fingerbreadths distal to and

along the posterior axillary fold, parallel to the lateral border of scapula [Lehman et al, 2006],

respectively. All electrodes were mounted by the same researcher in both test sessions. A

reference electrode was mounted on the ankle, at the lateral malleolus.

After the placement of the electrodes, participants performed a progressive warm up

regimen by lifting increasingly heavier loads in bench press. Bench press was performed

using an ER-Equipment power rack (ER Equipment, Albertslund, Denmark). For the 3RM

test, the load was increased by 2.5-10 kg per set of 3 repetitions, until 3RM was found. On

5
average 4 sets were required for the determination of the 3RM. Four min rest was applied

between all sets.

After successful completion of the 3RM test, the load was decreased to 75% of the

3RM load and 3 repetitions were performed. Participants were instructed to perform the

eccentric phase in approx. 1 s and the concentric phase as fast as possible for the 3RM test

and the normalization set. The EMG data obtained during these repetitions was used for

normalization purpose.

Finally, the load was further decreased to 60% of the 3RM load, and the last 3 sets of

8 repetitions were completed. This resulted in a total of 24 repetitions with a cyclic pattern

consisting of approx. 1 s eccentric phase and 1 s concentric phase. The data recorded during

these repetitions were used to study the betweenday reliability of muscle coordination. A

potentiometer (model KS60, NTT Nordic Transducer, Hadsund, Denmark) was connected to

the middle of the barbell for measurement of the vertical position. A bench press cycle was

defined as the period between two successive top positions. To avoid any initial transition,

the first bench press cycle of each set was removed from the data resulting in the

concatenation of 21 cycles per participant. In the second test session, the exact same

procedure was applied.

Table 2 near here

2.4. Data recording and processing

The surface EMG signals were recorded using a 128-channel surface EMG amplifier (EMG-

USB, LISiN - OT Bioelectronica, Turin, Italy) where they were amplified using an

individual-specific gain factor (100-500) and band-pass filtered [10-750 Hz] before being

sampled at 2048 Hz. Furthermore, a notch filter (4th order Butterworth band stop with

rejection width of 1 Hz centered at the first three harmonics of the power line frequency of 50

Hz) was used to remove line interference. The linear envelopes of the EMG measurements

6
across each of the bench press cycles were obtained by low pass filtering (zero-lag

Butterworth, 2nd order, 4 Hz) of the rectified EMG. Each of the EMG envelopes was then

interpolated into 100 time points. During the normalization set performed at 75% of 3RM,

the first and last half second of the EMG measurements were excluded. The linear envelopes

of the EMG measurements for normalization purposes were then computed and averaged

across 100 ms non-overlapping intervals. The maximum of these averaged values was then

used as a normalization factor for the matching EMG measurement [Kristiansen et al, 2015b].

For the extraction of muscle synergies, a nonnegative matrix factorization was applied

to the concatenation of the 21 bench press cycles [Oliveira et al, 2014] using the Lee and

Seung algorithm [Lee and Seung, 1999] in line with previous studies [Hug et al, 2011;

Kristiansen et al, 2015b; Torres-Oviedo and Ting, 2007; Turpin et al, 2011a]. Nonnegative

matrix factorization decomposes the initial matrix E into two multiplication matrices (W and

C). E is a p-by-n matrix of the normalized EMG envelopes, where p is the number of muscles

(13) and n is the number of time points (2100). n is calculated as 21 bench press cycles times

100 time points per bench press cycle. W is a p-by-s matrix (s is number of synergies), and

consists of the muscle synergy vectors. Muscle synergy vectors represent the relative

weighting of each muscle within each synergy. C is an s-by-n matrix, and represents the

synergy activation coefficient. Synergy activation coefficients represent the recruitment of

the muscle synergy over time. The number of synergies chosen for further analysis is

dependent on the variance accounted for (VAF). In line with previous work [Kristiansen et al,

2015b; Torres-Oviedo and Ting, 2007], the smallest number of synergies that provided a total

VAF 90% for all participants was accepted.

After finding the smallest number of muscle synergies that described 90% of the

VAF, the muscle synergies extracted from each participant were functionally sorted [Torres-

Oviedo and Ting, 2007]. Functional sorting is necessary as the order of the muscle synergies

7
may be swapped among participants following the application of the non-negative matrix

factorization.

To evaluate the between-day reliability of the extracted muscle synergy vectors and

synergy activation coefficients, we performed a cross-validation analysis similar to that

performed by Frre and Hug [Frre and Hug, 2012] and Muceli et al [Muceli et al, 2010]. In

this iterative procedure [1], the muscle synergy vectors extracted in the first test session were

recomputed, using the fixed synergy activation coefficients from the second test session.

[1]

when C is fixed, W is being recomputed, n is the iteration counter, and i and j are indices

corresponding to the row and column elements of the matrices. This iteration process, not to

be confused with the one previously described for sorting of muscle synergies, continues until

the reconstruction error converges. and were initially randomized in equation [1]

and [2], respectively. T is the matrix transpose. A similar procedure was then performed to

re-compute the muscle synergy vectors extracted in the second test session using the fixed

synergy activation coefficients from the first test session. The new recomputed muscle

synergy vectors were then compared to their original versions in each day using correlation

analysis. A similar procedure [2] as described above, was carried out for the synergy

activation coefficients, as they were recomputed, using the fixed muscle synergy vector from

the opposite day.

[2]

The new recomputed synergy activation coefficients were then compared with their

original versions in each day. For this purpose the cross-correlation function was used. The

magnitude of the correlation coefficients across the procedures was considered for evaluation

8
of the between-day reliability of muscle synergy vectors and synergy activation coefficients.

To evaluate the between-day reliability of total VAF, we first determined total VAF for the

first and second test session. The total VAF of each test session was then recomputed using

the procedure described above, by fixing either the muscle synergy vector (termed VAFFix_W)

or the synergy activation coefficient (termed VAFFix_C) of the opposite day. A substantial

drop in total VAF would indicate low reliability across test sessions. The same procedure was

carried out for VAF of each of the 13 individual muscles (termed VAFMuscle).

Figure 1 near here

2.5. Statistical analysis

As muscle synergy vectors and synergy activation coefficients are not single scalar quantities,

the use of a conventional intraclass correlation coefficient (ICC) or standard error of the

measurement (SEM) to evaluate reliability was not directly applicable. In order to be able to

compare components across test sessions, we therefore calculated Pearsons correlation

coefficient (r) for muscle synergy vectors, and the maximum of the cross-correlation function

(rmax) for synergy activation coefficients of each subject. To compare original components

and recomputed components, r-values were calculated between the recomputed component of

the first test session and the original component of the first test session and between the

recomputed component of the second test session and the original component of the second

test session for each subject. This yielded a total of four r-values per subject. We then tested

the r-values for systematic bias across test sessions, by calculating the effect of trials, using a

single-factor, within-subjects repeated measures analysis of variance (RM-ANOVA).

Further, the relative and absolute reliability across test sessions of the r-values

obtained when comparing the original and the recomputed components were then evaluated

using ICC and SEM. The relative reliability was evaluated by calculating a 2-way fixed

ICC3,1 and the absolute reliability was evaluated by calculating SEM. As a measure of the

9
minimal difference needed to be considered real, we calculated a 95% confidence interval

(CI). The same statistics were applied to VAFFix_W, VAFFix_C, and VAFMuscle. r-values were

interpreted using categories previously suggested in which an r-value of 0.10-0.30 is

considered weak, 0.31-0.50 is considered moderate, 0.51-0.70 is strong, and 0.71-1.00 is very

strong [Martnez-Valencia et al, 2013]. ICC3,1-values were interpreted using the categories

proposed previously in which an ICC3,1 of 0.00-0.20 is considered poor, 0.21-0.40 is fair,

0.41-0.60 is moderate, 0.61-0.80 is substantial, and 0.81-1.00 is almost perfect [Landis and

Koch, 1977]. SPSS Version 22.0 statistics software (IBM Corp; Armonk, NY, USA) was

used for all statistical analyses. Statistical significance was accepted at p 0.05. Results are

presented as meanSD, unless otherwise indicated.

Figure 2 near here

3. Results

Two muscle synergies caused the total VAF to be 90% for all participants (Figure 1). The

synergy activation coefficients together with the muscle synergy vectors for the first and

second test session are depicted in Figure 2. Muscle synergy 1 mainly involved the eccentric

phase of the bench press cycle while muscle synergy 2 mainly involved the concentric phase.

For seven subjects, however, one muscle synergy was sufficient for the VAF to be 90%, and

VAFMuscle to be 75%, in both test sessions. The muscle synergy vector and synergy

activation coefficient for these subjects in the first and second test session are depicted in

Figure 3.

The correlations between the first and second test session were strong (0.58 and 0.62)

for muscle synergy vectors and very strong (0.84 and 0.89) for synergy activation coefficients

(Table 1). However, for both muscle synergy vectors, two subjects displayed negative

correlation values.

10
 Comparing the original components to the recomputed components for the first and

second test session, muscle synergy vectors and synergy activation coefficients showed a

very strong correlation in all instances (0.74-0.88) (Table 2). For muscle synergy vectors,

ICC3,1-values were almost perfect (0.85 and 0.95), and SEM calculated on the r-values were

0.10 and 0.16 (Table 2). For synergy activation coefficients, ICC3,1-values were substantial

(0.70) and almost perfect (0.90), and SEM calculated on the rmaxvalues were 0.06 in both

cases (Table 2).

When fixing the synergy activation coefficient, VAFFix_C was 92.41.9 and 92.72.7

for the first and second test session, respectively. When fixing the muscle synergy vector,

VAFFix_W was 84.910.0 and 86.87.2 for the first and second test session, respectively. The

ICC3,1-values of VAFFix_C and VAFFix_W were moderate (0.59) and almost perfect (0.86),

respectively. The SEM was in all cases low (0-0.03%). For a few muscles (AD, TBMFix_C,

and SOLFix_C), the VAFMuscle exhibited systematic bias (p0.05) between the first and second

test session. The ICC3,1-values ranged from poor (0.06) to almost perfect (0.87) while the

SEM was in general low and ranged from 0-3.07%.

Figure 3 near here

4. Discussion

In the present study, we evaluated between-day reliability of muscle synergies, extracted by

applying a nonnegative matrix factorization algorithm to EMG data collected during bench

press. In line with our hypothesis, we found that the extracted muscle synergies which

described bench press, were stable and reliable across days.

The between-day reliability of muscle synergy vectors and synergy activation

coefficients were strong and very strong, respectively. As muscle synergy vectors were

slightly less reliable, this may indicate that the relative weighting of each of the muscles

within each synergy is subject to small variations between the first and second test session.

11
This could be explained by the redundancy or abundancy of the musculoskeletal system,

whereby a movement can be performed using different muscle recruitment options [Latash et

al, 2002]. This can also be explained by the experimental task since biomechanically

constrained task have been recently reported to reduce the accuracy of estimated synergies

[Steele et al, 2015]. Another possible explanation, might be related purely to computation, as

the cross correlation of synergy activation coefficients are calculated from 100 samples,

while only 13 samples are used for muscle synergy vectors. Due to this difference, a larger

estimation variance is expected in muscle synergy vectors compared to synergy activation

coefficients [Mak, 2004]. Further, it is possible that small variations in the placement of

EMG electrodes among sessions could in part explain why muscle synergy vectors were

slightly less reliable. From a neurophysiological point of view, muscle synergies are thought

to be of neural origin located at the spinal level, thus being controlled by the afferent systems

and the motor cortical areas [Bizzi and Cheung, 2013]. Within this framework, some believe

that muscle activity is the result of a few temporal activation components distributed to

various muscles [Ivanenko et al, 2006]. This is further advocating the higher reliability of the

synergy activation coefficients over the muscle synergy vectors. Despite this, the reliability of

the muscle synergy vectors were still strong and indicated good between-day reliability and

the synergy activation coefficients displayed very strong correlations between the first and

second test session. In addition, muscle redundancy could also explain why negative

correlation values were obtained in two out of 21 cases for muscle synergy vector 1 (Table

1). This may have caused some inter-trial variation in the motor patterns, in line with a

previous study [Torres-Oviedo and Ting, 2010].

In the present experimental setup, a metronome guided the tempo of the execution of

the task, which may explain the very strong correlation between synergy activation

coefficients. The strong to very strong correlations of muscle synergy vectors and synergy

12
activation coefficients suggest that the same basic motor output patterns were responsible for

the muscle coordination in the first and second test session. It therefore seems that at least for

trained individuals, the control of bench press is stable. This indicates that extracting muscle

synergies from EMG data using nonnegative matrix factorization is a reliable method for

studying muscle coordination during bench press.

In addition to evaluating the reliability of muscle synergy vectors and synergy

activation coefficients using Pearsons r, we also performed a cross validation analysis of the

extracted muscle synergies in which we recomputed each of the muscle synergy vectors and

synergy activation coefficients. This procedure was done to verify the within-subject

reliability of the extracted muscle synergies, meaning that if different muscle coordination is

used in the first and second test session, the correlation of the recomputed component would

be weak. However, the very strong correlation of the r-values, the high ICC3,1 values and the

low SEM values further indicates that the within-subject reliability of the extracted muscle

synergies across test sessions was high. This has important implications for the use of

nonnegative matrix factorization as an assessment tool, as these data for the first time provide

information on the robustness of the low dimensional structure of muscle coordination across

days.

When we recomputed VAF by fixing either the muscle synergy vectors or the synergy

activation coefficients of the opposite day, the mean values of VAF Fix_C for both test sessions

were still above the 90% threshold (92.7% and 92.4%), indicating that only a minor part of

the variability was not accounted for. For VAFFix_W, mean values decreased to just below the

90% threshold (86.8% and 84.9%). This decrease may be the result of the muscle synergy

vectors being slightly less reliable than synergy activation coefficients. When comparing first

and second test session with respect to values of VAFFix_W and VAFFix_C, ICC3,1 values were

almost perfect for VAFFix_W, combined with a very low SEM-value of 0.03%. ICC3,1 values

13
were merely moderate for VAFFix_C, despite the negligible SEM value of 0.00%. The

reason for the latter could be that the ICC value may be small in case of low between-subject

variability. This may in fact be the case for VAFFix_C based on the almost identical mean

values (92.4% and 92.7%) and the very low SD values (1.9% and 2.7%) [Weir, 2005]. This

point is further underlined when looking at VAF for individual muscles. Thus, PM, PMFix_C,

ADFix_C, LDFix_C, GML, SOL, VL, and VM all display identical mean VAFMuscle values in the

first and second test session, with very low SD and SEM values. Yet, the ICC3,1 only ranged

from poor to moderate. When calculating VAFMuscle, 75% of the variability must be

accounted for to adequately reconstruct muscle data vectors [Torres-Oviedo and Ting, 2007].

This was the case for almost all versions of VAFMuscle, even the ones that were recalculated,

with the exception of ADFix_C, BBFix_C, TBLFix_C, TBMFix_C, and RFFix_C (5 out of 39 versions

of VAFMuscle). This implies that the reconstruction of single muscle data vectors was in most

cases adequate. And even when fixing either the muscle synergy vector or the synergy

activation coefficient of the opposite test session, the VAF Muscle would in most cases still

satisfy the 75% threshold for adequate reconstruction. In summary, we found a very small

drop in VAFFix_W (9.0% and 11.3%) and VAFFix_C (3.2% and 3.9%) compared to mean total

VAF, a moderate to almost perfect ICC3,1 and very low SEM-values for VAFFix_W and

VAFFix_C, in addition to the fact that versions of VAFMuscle in most cases were above the 75%

threshold. This indicates that by using a fixed synergy component from another test session, a

very large portion of the variability in the dataset is still accounted for following

recomputation. This further confirmed the high between-day reliability of muscle synergy

vectors and synergy activation coefficients.

As we were aware that some of the subjects displayed VAF-values above 90%, using

just one synergy component, we applied the commonly used criteria for VAF Muscle in which a

VAFMuscle above 75% is required for adequate reconstruction [Torres-Oviedo and Ting,

14
2007]. Only 8 participants out of 21 (=38%) met this criterion using one synergy component.

For 7 of these 8 subjects, the criterion was met at both the first and second test session. In

addition to this, the extraction of two synergy components fits nicely with the functional role

of the muscles and the phases of the task. In combination, this made us consider that two

synergy components would mainly be required for an adequate description of the task. For

the seven subjects requiring only one muscle synergy, the muscle synergy vector and synergy

activation coefficient may reflect a merge of the two originally extracted muscle synergies as

shown in Figure 3. It has previously been shown, that muscle synergies extracted during

walking in patients with post-stroke hemiparesis were merged together, resulting in impaired

walking performance [Clark et al, 2010]. It is possible that the merge of two muscle synergy

components into one, is the result of these subjects not being capable of differentially

activating the two muscle synergies. One possible explanation is that the subjects are

consistently coping against the weight in both the eccentric as well as the concentric phase.

This consistent application of muscle activation is illustrated by the synergy activation

coefficient in Figure 3, in which the difference between maximum and minimum activation is

small. For the rest of the subjects, however, it seems that muscle activation is predominantly

applied at a later stage of the eccentric phase. This can be seen by the well-defined peak of

activation occurring just before the 50th time point in synergy activation coefficient 1 in

Figure 2. This peak most likely represents the shift between the eccentric and the concentric

phase. This could point towards a more confident strategy in which the muscle activation is

minimized, thus leaving the control and stabilization of the weight to the very last moment of

the eccentric phase. In an attempt to explain why 7 out of 21 subjects exhibit only one muscle

synergy, we tried to establish a link between these particular subjects and variables such as

absolute strength, relative strength, age, body mass and training experience. However, this

was not possible, and we are not sure of the reason causing discrepancy in number of muscle

15
synergies among subjects. One explanation is that altered supraspinal drive and sensory-

motor control may have affected the modular control in some of the subjects, as previously

discussed [Clark et al, 2010].Limitations of this study include that ipsilateral, rather than

bilateral, EMG data was obtained. Bilateral recording of EMG data would have yielded

further data, to support the study conclusions. Moreover, we opted for a VAF-criterion that

could accommodate all the subjects and thus we did not separate the subjects in two groups

with one and two muscle synergies for the main analysis. Further, there is an ongoing debate

as to whether muscle synergies are functional entities produced by the nervous system or

simply a result of the biomechanical constraints of the motor task [Steele et al, 2015]. The

design of the present study does not allow for any speculations on this matter. Regardless, the

reliability of the application of nonnegative matrix factorization to derive synergy

components during a functional task remains important. A perspective of the present study is

that it appears sound to use nonnegative matrix factorization algorithm for quantification of

changes in muscle coordination over time, e.g. training intervention studies [Kristiansen et al,

2015a]. Further, biomechanical studies including 3D kinematics may help to delineate the

relationship between the number of muscle synergies and the movement kinematics.

In conclusion, the between-day reliability of muscle synergies was evaluated for the

first time in bench press. The correlations of muscle synergy vectors and synergy activation

coefficients between two test sessions were strong to very strong, and further analysis

indicated that ICC3,1 values ranged from substantial to almost perfect combined with low

SEM values. The present findings suggest that the same general structure of the muscle

coordination was present across days. This demonstrates the consistency of the strategy used

by the nervous system to control movements like bench press.

16
Acknowledgements

The study was partly supported by grants from the Ministry of Culture Committee on Sports

Research in Denmark and the Danish Rheumatism Association. ER Equipment is thanked for

loan of the lifting equipment.

Conflict of interest

The authors declare no competing financial, consultant, institutional or other conflicts of

interests.

References

Bernstein N. The Coordination and Regulation of Movements. New York: Pergamon Press,
1967.

Bizzi E, Cheung VCK. The neural origin of muscle synergies. Frontiers in Computational
Neuroscience. 2013;7(Article 51):DOI: 10.3389/fncom.2013.00051.

Cappellini G, Ivanenko YP, Poppele RE, Lacquaniti F. Motor patterns in human walking and
running. Journal of neurophysiology. 2006;95(6):3426-3437.

Clark DJ, Ting LH, Zajac FE, Neptune RR, Kautz SA. Merging of healthy motor modules
predicts reduced locomotor performance and muscle coordination complexity post-stroke.
Journal of neurophysiology. 2010;103:844-857.

D'Avella A, Bizzi E. Shared and specific muscle synergies in natural motor behaviors.
Proceedings of the National Academy of Sciences of the United States of America.
2005;102(8):3076-3081.

Dorel S, Drouet J, Couturier A, Champoux Y, Hug F. Changes of pedaling technique and

muscle coordination during an exhaustive exercise. Medicine and science in sports and
exercise. 2009;41(6):1277-1286.

Frre J, Hug F. Between-subject variability of muscle synergies during a complex motor skill.
Frontiers in Computational Neuroscience. 2012;6:99(Article 99):DOI:
10.3389/fncom.2012.00099.

Hermens HJ, Freriks B, Disselhorst-Klug C, Rau G. Development of recommendations for

SEMG sensors and sensor placement procedures. Journal of Electromyography and
Kinesiology. 2000;10(5):361-374.

Hug F. Can muscle coordination be precisely studied by surface electromyography? Journal

of Electromyography and Kinesiology. 2010;21(1):1-12.

17
Hug F, Turpin NA, Couturier A, Dorel S. Consistency of muscle synergies during pedaling
across different mechanical constraints. Journal of neurophysiology. 2011;106(1):91-103.

Hug F, Turpin NA, Guvel A, Dorel S. Is interindividual variability of EMG patterns in

trained cyclists related to different muscle synergies? Journal of applied physiology.
2010;108(6):1727-1736.

Ivanenko YP, Poppele RE, Lacquaniti F. Motor control programs and walking.
Neuroscientist. 2006;12(4):339-348.

Ivanenko YP, Poppele RE, Lacquaniti F. Five basic muscle activation patterns account for
muscle activity during human locomotion. The Journal of physiology. 2004;556(1):267-282.

Kottner J, Audig L, Brorson S, Donner A, Gajewski BJ, Hrbjartsson A, Roberts C, Shoukri

M, Streiner DL. Guidelines for reporting reliability and agreement studies (GRRAS) were
proposed. Journal of Clinical Epidemiology. 2011;64(1):96-106.

Krishnamoorthy V, Latash ML, Scholz JP, Zatsiorsky VM. Muscle synergies during shifts of
the center of pressure by standing persons. Experimental Brain Research. 2003;152(3):281-
292.

Kristiansen M, Samani A, Madeleine P, Hansen EA. Effects of five weeks of bench press
training on muscle synergies - A randomized controlled study. Journal of Strength &
Conditioning Research. 2015a;DOI Number: 10.1519/JSC.0000000000001282.

Kristiansen M, Madeleine P, Hansen EA, Samani A. Inter-subject variability of muscle

synergies during bench press in power lifters and untrained individuals. Scandinavian Journal
of Medicine and Science in Sports. 2015b;25(1):89-97.

Lacquaniti F, Ivanenko YP, Zago M. Development of human locomotion. Current opinion in

neurobiology. 2012;22:822-828.

Landis JR, Koch GG. The measurement of observer agreement for categorical data.
Biometrics. 1977;33(1):159-174.

Latash ML, Scholz JP, Schner G. Motor control strategies revealed in the structure of motor
variability. Exercise and sport sciences reviews. 2002;30(1):26-31.

Lee DD, Seung HS. Learning the parts of objects by non-negative matrix factorization.
Nature. 1999;401(6755):788-791.

Lehman GJ, MacMillan B, MacIntyre I, Chivers M, Fluter M. Shoulder muscle EMG activity
during push up variations on and off a Swiss ball. Dynamic Medicine. 2006;5:7(art. no.
7):DOI: 10.1186/1476-5918-5-7.

MacLellan MJ, Ivanenko YP, Massaad F, Bruijn SM, Duysens J, Lacquaniti F. Muscle
activation patterns are bilaterally linked during split-belt treadmill walking in humans.
Journal of neurophysiology. 2014;111:1541-1552.

18
Mak TK. Estimating variances for all sample sizes by the bootstrap. Computational Statistics
and Data Analysis. 2004;46(3):459-467.

Martnez-Valencia MA, Linthorne NP, Alcaraz PE. Effect of lower body explosive power on
sprint time in a sled-towing exercise. Science and Sports. 2013;28(6):e175-e178.

Muceli S, Boye AT, D'Avella A, Farina D. Identifying representative synergy matrices for
describing muscular activation patterns during multidirectional reaching in the horizontal
plane. Journal of neurophysiology. 2010;103(3):1532-1542.

Muceli S, Falla D, Farina D. Reorganization of muscle synergies during multidirectional

reaching in the horizontal plane with experimental muscle pain. Journal of neurophysiology.
2014;111(8):1615-1615-1630.

Oliveira AS, Gizzi L, Farina D, Kersting UG. Motor modules of human locomotion:
Influence of EMG averaging, concatenation, and number of step cycles. Frontiers in Human
Neuroscience. 2014;8(MAY).

Steele KM, Tresch MC, Perreault EJ. Consequences of biomechanically constrained tasks in
the design and interpretation of synergy analyses. Journal of neurophysiology.
2015;113(7):2102-2113.

Torres-Oviedo G, Ting LH. Muscle synergies characterizing human postural responses.

Journal of neurophysiology. 2007;98(4):2144-2156.

Torres-Oviedo G, Macpherson JM, Ting LH. Muscle synergy organization is robust across a
variety of postural perturbations. Journal of neurophysiology. 2006;96(3):1530-1546.

Torres-Oviedo G, Ting LH. Subject-specific muscle synergies in human balance control are
consistent across different biomechanical contexts. Journal of neurophysiology.
2010;103(6):3084-3098.

Turpin NA, Guvel A, Durand S, Hug F. Effect of power output on muscle coordination
during rowing. European journal of applied physiology. 2011a;111(12):3017-3029.

Turpin NA, Guvel A, Durand S, Hug F. No evidence of expertise-related changes in muscle

synergies during rowing. Journal of Electromyography and Kinesiology. 2011b;21(6):1030-
1040.

Walter SD, Eliasziw M, Donner A. Sample size and optimal designs for reliability studies.
Statistics in medicine. 1998;17(1):101-110.

Weir JP. Quantifying test-retest reliability using the intraclass correlation coefficient and the
SEM. Journal of Strength and Conditioning Research. 2005;19(1):231-240.

19
Figure captions
Figure 1. The percentage of variance accounted for (VAF, %) depicted as a function of the

number of the original extracted muscle synergies for the first and second test session. Values

are presented as mean25th and 75th percentiles.

Figure 2. Synergy activation coefficients and muscle synergy vectors of the first (A) and

second test session (B). Muscle synergy 1 mainly involved the eccentric phase (1. half) of the

bench press cycle, while muscle synergy 2 mainly involved the concentric phase (2. half) of

the bench press cycle. For muscle synergy vectors, the black bars represent group mean (with

SD bars). The thin gray bars represent individual muscle synergy vectors. For synergy

activation coefficients, the thick black line represents the group mean, while the thin gray

lines represent individual synergy activation coefficients. PM=pectoralis major, AD=anterior

deltoideus, BB=biceps brachii, TBL=triceps brachii lateral head, TBM=triceps brachii medial

head, LD=latissimus dorsi, ES=erector spinae, RF=rectus femoris, BF=biceps femoris,

GML=gastrocnemius lateral head, SOL=soleus, VL=vastus lateralis, and VM=vastus

medialis. N = 21.

Figure 3. Synergy activation coefficient and muscle synergy vector of the first (A) and

second (B) test session for the seven subjects who achieved a VAF value above 90% using

only one muscle synergy. The muscle synergy vector and synergy activation coefficient

reflect a merge of the two muscle synergy components shown in Figure 2. For explanation of

black and gray bars and lines, and muscle acronyms, see Figure 2 caption. N = 7.

20
21
22
Table 1. Comparison of muscle synergies across test sessions.

________________________________________________________________________________________________________________

Muscle synergy vector 1 Muscle synergy vector 2

_____________________ ______________________

First vs second test session (r) 0.580.42 [-0.59 0.97] 0.620.41 [-0.56 0.95]

________________________________________________________________________________________________________________

Synergy activation coefficient 1 Synergy activation coefficient 2

_____________________________ _____________________________

First vs second test session (rmax) 0.840.22 [0.32 0.99] 0.890.13 [0.62 0.99]

________________________________________________________________________________________________________________

Correlation coefficients (r) of muscle synergy vectors and cross-correlation coefficients (rmax) of synergy activation coefficients. Values are
presented as means SD [min-max].

23
Table 2. Comparison of original and recomputed components of muscle synergies consisting of muscle synergy vectors and synergy
activation coefficients.

_______________________________________________________________________________________________________ _________

Muscle synergy vector 1 Muscle synergy vector 2

____________________ ____________________

Recomputed component vs original component

First test session (r) 0.740.46 [-0.59 - 1.00] 0.750.46 [-0.67 1.00]

Second test session (r) 0.750.41 [-0.51 0.99] 0.820.36 [-0.22 1.00]

Effect of trials (p-value) 0.797 0.163

ICC3,1 0.95 0.85

SEM (r) 0.10 0.16

CI95% 0.25 0.41

________________________________________________________________________________________________________________

Synergy activation coefficient 1 Synergy activation coefficient 2

___________________________ ___________________________

Recomputed component vs original component

First test session (rmax) 0.860.22 [0.28 - 1.00] 0.750.46 [-0.67 1.00]

Second test session (rmax) 0.880.19 [0.38 - 1.00] 0.820.36 [-0.22 1.00]

Effect of trials (p-value) 0.359 0.338

ICC3,1 0.90 0.70

SEM (rmax) 0.06 0.06

CI95% 0.18 0.16

________________________________________________________________________________________________________________

r and rmax,-values are presented as means SD [min-max]. ICC3,1 = 2-way fixed intraclass correlation coefficient. SEM = Standard error of
measurement. CI = Confidence interval.

24
Author biography

Mathias Kristiansen

Mathias Kristiansen earned a MS degree in sports science at Aalborg University in 2012 and received
his PhD in Sports Science in 2015 from Aalborg University, Denmark. He is currently employed at the
Department of Health Science and Technology at Aalborg University, Denmark. His main areas of
research are neural adaptations to resistance training.

Afshin Samani

Afshin Samani received his PhD in Biomedical Engineering and Science in 2010 from Aalborg
University, Denmark. He is currently employed as an associate professor in sports science and
ergonomics at the Department of Health Science and Technology at Aalborg University, Denmark. He
is co-director of the laboratory for Ergonomics and Work-related Disorders. His specific research field
is focused on methods of quantification of work exposures and risk factors for the development of
musculoskeletal disorders and interactions between muscle pain and motor control in computer
users.

Ernst Albin Hansen

Ernst Albin Hansen was born in Roskilde, Denmark, on 6 November 1969. He earned a MS degree in
sports science at University of Copenhagen in 1997 and a PhD degree entitled Is the freely chosen
pedal rate optimal during cycling? at University of Southern Denmark in 2003. He has been
associate professor at SMI, Department of Health Science and Technology, Aalborg University since
2010 and defended a DSc degree in 2015 entitled On voluntary rhythmic leg movement behaviour
and control during pedalling. He is currently director of Motor Behaviour and Performance
Laboratory as well as member of the Research Interest Group of Physical Activity and Human
Performance. He is board member of the Danish Biomechanical Society and the Danish University
Extension. His scientific production currently includes more than 40 articles published in peer-
reviewed journals. His research areas include human rhythmic motor behaviour and control as well
as strength training and sports performance.

Pascal Madeleine

Pascal Madeleine was born in was born in Toulouse, France, in 1969. He received his DrSc degree and
PhD from Aalborg University, Denmark. He is currently employed as a Professor in Sports and
Ergonomics at SMI, Department of Health Science and Technology at Aalborg University, Denmark.
He is head of the research interest group within Physical Activity and Human Performance and co-
director of the laboratory for Ergonomics and Work-related Disorders. He has published more 160
peer reviewed scientific journal publications and book chapters. His main area of research interests
are the development and application of novel methods and technologies in Ergonomics and Sports.