Segmental specificity in belly dance mimics primal trunk locomotor patterns

Belly dance provides a novel approach for studying spinal cord neural circuits. New evidence suggests that primitive locomotor circuits may be conserved in humans. Erector spinae activation patterns during the hip shimmy at different tempos are similar to those observed in salamander walking and swimming. As movement frequency increases, a sequential pattern similar to lamprey swimming emerges, suggesting that primal involuntary control mechanisms dominate in fast lateral rhythmic spine undulations even in humans.

Abstract

Belly dance was used to investigate control of rhythmic undulating trunk movements in humans. Activation patterns in lumbar erector spinae muscles were recorded using surface electromyography at four segmental levels spanning T10 to L4. Muscle activation patterns for movement tempos of 2 Hz, 3 Hz, and as fast as possible (up to 6 Hz) were compared to test the hypothesis that frequency modulates muscle timing, causing pattern changes analogous to gait transitions. Groups of trained and untrained female subjects were compared to test the hypothesis that experience modifies muscle coordination patterns and the capacity for selective motion of spinal segments. Three distinct coordination patterns were observed. An ipsilateral simultaneous pattern (S) and a diagonal synergy (D) dominated at lower frequencies. The S pattern was selected most often by novices and resembled the standing wave of activation underlying the alternating lateral trunk bending in salamander trotting. At 2 Hz, most trained subjects selected the D pattern, suggesting a greater capacity for segmental specificity compared with untrained subjects. At 3–4 Hz, there emerged an asynchronous pattern (A) analogous to the rostral-caudal traveling wave in salamander and lamprey swimming. The neural networks and mechanisms identified in primitive vertebrates, such as chains of coupled oscillators and segmental crossed inhibitory connections, could explain the patterns observed in this study in humans. Training allows modification of these patterns, possibly through improved capacity for selectively exciting or inhibiting segmental pattern generators.

NEW & NOTEWORTHY Belly dance provides a novel approach for studying spinal cord neural circuits. New evidence suggests that primitive locomotor circuits may be conserved in humans. Erector spinae activation patterns during the hip shimmy at different tempos are similar to those observed in salamander walking and swimming. As movement frequency increases, a sequential pattern similar to lamprey swimming emerges, suggesting that primal involuntary control mechanisms dominate in fast lateral rhythmic spine undulations even in humans.

fundamental to the coordination of human movement is the control of the body axis. Trunk control as it relates to meeting challenges to stability imposed by upright, bipedal posture has been previously investigated (Henry et al. 1998; Küng et al. 2009; Oddsson and Thorstensson 1987; Vernazza-Martin et al. 2006). However, the trunk’s multifunctionality extends beyond its role in facilitating static balance to include dynamic stability during locomotion (Bouisset and Do 2008; Cappellini et al. 2010; Carlson et al. 1988; Ceccato et al. 2009; Thorstensson et al. 1982, 1984), steering and head/trunk orientation (Courtine et al. 2006; Mouchnino et al. 1992; Schmid et al. 2005), configuration changes to negotiate environmental obstacles, and facilitate focal tasks (Frank and Earl 1990; Ivanenko et al. 2005), and expression of emotions or aesthetic principles as in dance (Bartenieff 1965; Daprati et al. 2009; Dell 1977; Gross et al. 2012; Nugent et al. 2012; Riskind and Gotay 1982). In the absence of pathology, the hierarchical control structure of the central nervous system (CNS) allows for seamless integration of voluntary control and automatic postural adjustments. Whereas control of trunk movements during gait, target-directed movements of the upper limb and automatic postural responses to loss of balance have been extensively studied, studies of simple rhythmic movements of the trunk (e.g., cyclic flexion/extension) as goals in themselves are limited (Andersson et al. 1996; Carlson et al. 1988; Nugent et al. 2012; Oddsson and Thorstensson 1986, 1987). Little has been published on the possible mechanisms of control in skilled rhythmic trunk movements, such as those employed in dance (Nugent et al. 2012; Or 2006). The common finding among studies of isolated trunk movements is task-dependent specificity of activation at different segmental levels and in different compartments of epaxial spine extensor muscles, which is predicted from functional anatomy (Macintosh and Bogduk 1987). Several studies have investigated lumbar extensor patterns in gait by recording electromyography (EMG) at a single segmental level (Cappellini et al. 2010; Carlson et al. 1988; Ivanenko et al. 2004; Saunders et al. 2005; Thorstensson et al. 1982). However, recording at only a single level may not be sufficient to characterize complex coordination patterns such as those associated with intricate undulations of the spine employed in dance.

Several studies have shown that trunk movements during locomotion and other types of rhythmic trunk motion in humans involve differential timing in paraspinal muscles at different segmental levels (Anders et al. 2007; Ceccato et al. 2009; de Sèze et al. 2008; Ivanenko et al. 2006; Nugent et al. 2012; Winter 1991). De Sèze et al. (2008) recorded erector spinae (ES) activity from four electrodes at spinal levels C7, T3, T12, and L4 in male subjects during forward and backward walking as well as crawling on hands and knees, revealing a complex rostral-caudal sequence of epaxial muscle patterning in these tasks, whereas an ascending wave characterized arm swinging as well as crawling on all fours with the hands on a treadmill, but with the knees stationary on the floor. Ceccato et al. (2009) found the same pattern in both walking and gait initiation. This descending metachronal wave closely resembles the traveling wave of activation recorded in lamprey and salamander swimming, whereas a caudal-rostral wave produces backward swimming (Delvolvè et al. 1997; Grillner 1985). These rhythmically generated spine oscillations are considered models of the earliest form of vertebrate locomotion. In interpreting the functional significance of this patterning in humans, de Sèze et al. (2008) and Falgairolle et al. (2006) proposed that the coordination mechanism behind the lateral undulation patterns that evolved to propel aquatic animals continues to be exploited in humans for the coordination between trunk and legs to maintain dynamic stability in locomotion and possibly other rhythmic movements. Among the variety of human trunk movements, belly dance may involve movements most similar to the lateral trunk undulations of primitive vertebrates.

A previous study (Nugent et al. 2012) used belly dance movements as a paradigm for exploring functional and neurological specificity in the lumbar ES during voluntary, segmental spine motion. Differences in the activation timing of adjacent, ipsilateral muscle compartments above and below the level of the third lumbar vertebrae during 0.5- and 1-Hz isolated rhythmic pelvic motion were taken as evidence for segmental control of lumbar extensor muscles. Both novice and trained dancers were able to sequentially control lumbar extensor muscles at different vertebral levels. The direction of activation sequencing (rostral-caudal or caudal-rostral) switched with a change in movement direction. We concluded that the separate innervation of these compartments corresponds to distinct biomechanical functions of different portions of the lumbar erector spinae, and that this capacity for neuromuscular specificity is innate and hard-wired.

To further investigate the structure of the paraspinal control system with the objective of elucidating the control mechanisms, we focused on the belly dance hip shimmy, which can be performed at various tempos up to ~6 Hz, with performance at higher frequencies being associated with extensive practice. We hypothesized that there would be a pattern transition in the activity of lumbar spine extensors as the shimmy frequency increased. Patterns should differ with respect to the relative peak activation timing at different ES levels and can be characterized accordingly. Furthermore, we expected that the characteristics of each pattern and the relative likelihood of a specific pattern occurring would be dependent on the amount of training.

MATERIALS AND METHODS

Subjects.

Fourteen novice (NOV) subjects and 16 trained (TRD) subjects were recruited. TRD subjects were defined as those having engaged in 3 or more years of regular belly dance training and practice. NOV subjects had no prior belly dance training. All participating subjects were free from injuries or medical conditions that might impair physical performance. Novices were selected such that their ages at the time of data collection were matched to the ages at which TRD subjects had initiated their belly dance training. The study was approved by the McGill Research Ethics Board, and all participants gave written informed consent. Care was taken to select participants with a low-to-normal body mass index (BMI) to maximize signals recorded with surface electromyography (sEMG) (De la Barrera and Milner 1994). Subjects whose data set did not include a minimum of five trials of useable data for each movement condition were not included in the analysis. Reasons for data exclusion were loss of signal due to excessive perspiration (1 subject), incomplete data sets (2 subjects), and low signal amplitude due to high skinfold thicknesses (3 subjects). After exclusions, 12 NOV and 12 TRD subjects remained. Subject demographics are provided in Table 1. NOV age range was 18 to 32 yr (25 ± 5 yr), and TRD age range was 27 to 57 yr (39 ± 9 yr), with age at start of training ranging from 16 to 33 yr (26 ± 5 yr). For the TRD group, amount of training ranged from 3.5 to 32 yr (12.4 ± 9.21 yr). The first author was a professional belly dancer and one of the subjects.

Table 1.

Information for subjects and trials: subject demographics and trial information

							No. of Useable trials/Actual Mean Frequency, Hz
ID	Age, yr	Age at Start of Training	Training, yr	Height, cm	Weight, kg	BMI, kg/m2	2 Hz		3 Hz		FAP
Trained subjects
T1	39	28	11	172.0	52.6	17.8	9	2.07	10	2.55	8	5.49
T2	37	31	6	157.5	53.5	21.6	10	2.01	9	2.82	10	5.38
T3	28	24	3.5	156.0	53.9	22.1	9	2.00	10	2.99	10	5.37
T4	41	33	8	158.5	55.2	22.0	10	2.03	10	2.65	10	3.94
T6	36	30	6.5	173.0	66.5	22.2	9	2.00	10	3.50	10	4.06
T8	50	17	30	166.0	56.0	20.3	10	2.01	10	3.42	10	6.31
T9	41	27	14	168.5	65.4	23.0	10	2.01	10	3.14	10	4.65
T10	41	30	11	175.5	64.0	20.8	10	2.01	10	3.12	10	5.17
T11	29	24	5	168.5	52.9	18.6	10	2.00	10	3.18	10	3.60
T13	38	27	11	164.5	54.0	20.0	10	2.00	10	2.97	10	5.37
T14	57	25	32	164.6	61.5	22.7	7	2.34	10	2.81	9	4.94
T16	27	16	11	147.3	40.8	18.8	10	2.00	10	3.02	10	5.38
Mean	38.7	26.0	12.4	164.3	56.4	20.8		2.0		3.0		5.0
SD	8.71	5.24	9.21	8.18	7.13	1.74		0.10		0.29		0.78
Novice subjects
N2*	32		0	173.0	56.7	18.9	6	1.99	6	2.69	6	3.48
N3	29		0	166.4	64.2	23.2	10	2.04	10	2.64	10	4.12
N4	29		0	175.5	62.5	20.3	10	2.02	10	4.16	10	4.79
N5	24		0	174.5	64.4	21.1	10	1.96	10	2.99	10	3.35
N7	20		0	160.0	55.7	21.8	10	2.00	10	3.15	10	4.15
N8	30		0	168.3	49.6	17.5	9	1.93	9	3.09	10	3.26
N9	24		0	170.0	52.3	18.1	10	1.99	10	3.01	10	4.00
N10	26		0	172.7	59.0	19.8	10	2.02	10	3.12	10	3.95
N11	18		0	156.5	45.0	18.4	10	2.31	10	2.89	10	3.68
N12	19		0	156.5	45.5	18.6	10	2.01	10	3.03	10	3.96
N13	18		0	159.0	51.6	20.4	10	2.00	10	3.16	10	3.41
N14	31		0.2	165.1	50.8	18.6	10	2.20	10	3.19	10	4.19
Mean	25.0		0	166.5	54.8	19.7		2.0		3.1		3.9
SD	5.26			7.02	6.77	1.69		0.11		0.38		0.44

First column provides the subject identification number (ID). The next five columns list demographic information (BMI, body mass index). The last six columns list the number of trials used in statistical analyses for each movement condition (2 Hz, n = 12 for trained subjects and 12 novices; 3 Hz, n = 12 trained subjects and 12 novices; and FAP, n = 12 trained subjects and 12 novices) and the mean movement frequencies by subject.

^*Only 6 trials of each condition were collected for subject N2.

Procedure.

To probe the effects of training (TRG) and movement frequency on coordination pattern and to test the hypothesis of a frequency threshold for pattern change, three target frequencies for the hip shimmy were chosen as movement conditions (MC): 2 and 3 Hz, performed to a digital metronome, and as fast as possible (FAP) with no auditory stimulus. Control in the 2-Hz shimmy was assumed to be voluntary, being slow enough for cycle-by-cycle control and easily entrained with the metronome. We presumed further that the FAP shimmy, depending on individual ability, would reach frequencies too high for cycle-by-cycle voluntary control, thus requiring some other or additional control mode(s) for generation and maintenance. The 3-Hz MC was chosen to test the possibility of a threshold frequency for switching from one control mode to another, since performing at this tempo was difficult and could lead to switching to either a higher or lower frequency. For the FAP condition, the maximum achievable frequency was expected to vary widely from subject to subject, depending on individual ability and body mechanics. A functional stance width for performing these movements was established for each subject and marked on the platform with masking tape. Feet were aligned with the tape markings to standardize width of base of support across trials.

For each MC, subjects performed three practice trials while watching a videotaped demonstration (recorded with the metronome for 2- and 3-Hz MC). Verbal cues and a brief demonstration on how to perform each MC were given before the practice trials and during rest periods between trials, as necessary. For the NOV group, this served to cue subjects on the necessary joint motions; for the TRD group, it served to constrain the options of subjects’ extensive movement repertoires to the task requirements. The three practice trials were followed by a block of 10 acquisition trials of the same condition. Practice and acquisition trial durations for the 2-Hz, 3-Hz, and FAP conditions were 25, 15, and 10 s, respectively, consisting of continuous, cyclic repetition alternating with ~10–20 s of rest (standing in place) between trials. Data collection thus yielded 400–500 cycles per task condition for a total of 1,200–1,500 cycles per subject.

Trial blocks by MC were ordered to alternate between more and less tiring tasks to prevent fatigue effects in muscle signals. Overall, level of physical exertion would be considered low to moderate. Because optimal performance was desired, positive transfer from one task to another was not a concern; rather, the order of progressive difficulty was intended to enhance performance. An additional justification for task order was the concern that a random ordering of MCs within trial blocks would cause confusion, degrading performance, and increasing intertrial variability. The MC order was 2 Hz (easiest), followed by FAP, then two conditions unrelated to the present study, then the 3-Hz condition, followed by two more tasks unrelated to the present study. In all, trials for 7–9 different MCs were performed over a period of 60–75 min. Approximately halfway through the session, subjects were allowed a sitting rest break for ~5 min.

Kinematics.

Kinematics were recorded using a Vicon Motion Systems (Oxford, UK) six-camera setup and the standard Plug-in Gait (PiG) full-body reflective marker set, including upper body and arms. Kinematic data were sampled at 200 Hz. To define specific rotations of the trunk at different spinal levels, which the standard PiG marker set cannot capture, two custom marker frames were constructed, each with three reflective markers defining a plane and placed at approximately the first (TH1) and twelfth (TH12) thoracic levels (Fig. 1A). These were molded to fit each subject and affixed to the body with double-sided tape. The pelvic segment (PELV) was defined using the four PiG pelvic markers placed on the anterior and posterior superior iliac spines. The subject was oriented such that lateral displacement to the right was in the positive x direction, forward sagittal displacement was in the positive y direction, and upward vertical displacement was in the positive z direction. Subjects were instructed to keep the head level, facing forward, with arms held slightly out to the sides during trials.

Fig. 1.

Placements for EMG electrodes and kinematic markers. A: in addition to the standard VICON full-body marker set from which the pelvic segment (PELV) was derived, subjects were fitted with marker frames at the first thoracic (TH1) and twelfth thoracic (TH12) segments. Three reflective markers on each frame and 3 of the pelvic markers defined the 3 planes for calculating relative segmental rotation angles. B: EMG recording sites for ES muscles, bilaterally from longissiumus thoracis at T10, L1, and L4 and from the iliocostalis lumborum midway between L2 and L3. [Photos used by permission, Marilee Nugent.]

Electromyography.

Muscle activity was recorded using the Delsys Bagnoli (Natick, MA) 16-channel sEMG system integrated with Vicon for synchronization with kinematics. Electrodes are bipolar with a 10-mm interelectrode distance. EMG was collected bilaterally from the trunk and unilaterally from the right leg, at 16 muscle sites sampled at 2 kHz. The EMG of eight of these muscle sites was analyzed in this study.

In a previous study on compartmental specificity in slow, medium-amplitude belly dance movements (Nugent et al. 2012), erector spinae (ES) activity was recorded unilaterally from eight electrodes spanning the right lumbar region [rostrocaudally from the first (L1) to the fifth (L5) lumbar vertebral levels and mediolaterally from multifidus to iliocostalis lumborum (IL)] following the electrode placement grid of Macintosh and Bogduk (1987). Because no difference in timing was found within each grouping of four electrode sites above and below the level of L3, a subset of these sites was chosen for the present study. Electrodes were placed bilaterally on the ES: longissimus thoracis (LT) at L1 and L4 levels and over the IL approximately midway between the L2 and L3 levels (Fig. 1B). To determine whether recording across a wider range of vertebral levels during the shimmy would reveal even greater segmental specificity, as in the traveling wave found by de Sèze et al. (2008), two more electrodes were placed bilaterally on LT at the level of the tenth thoracic vertebra (T10). At this vertebral level, the LT is covered only by aponeurosis as opposed to the fibers of the trapezius (de Sèze and Cazalets 2008). Other researchers have similarly recorded from the ES at levels spanning T9–T12 during locomotion (Cappellini et al. 2006; Ceccato et al. 2009; de Sèze et al. 2008; Ivanenko et al. 2006).

Data processing.

After preprocessing of kinematic data with NEXUS software, all EMG and kinematic data were processed using custom MATLAB script (R2010b; The MathWorks, Natick, MA). Motion data were low-pass filtered at 8 Hz. Rotation angles were calculated for each trunk segment (TH1, TH12, and PELV) around each rotational axis (x, sagittal rotation; y, frontal rotation; z, transverse rotation) relative to the initial stationary position of the subject. For each trial, mean cycle period (T) was calculated by first applying the MATLAB function “findpeaks” to the pelvic frame angle PELVy to identify frontal plane rotation maxima, after which the time intervals between peaks in the trial were averaged. Mean frequency for each MC for each subject was calculated as 1/T.

Raw EMG data for all shimmy trials were demeaned and plotted for visual inspection. Start and end points of each trial were manually determined to include only activity representing consistent performance of task relevant motion. In the majority of trials, this amounted to 0.5 to 2 s less than the trial length. Full-wave rectified EMG was low-pass filtered at 8 Hz using a digital third-order Butterworth filter. The MATLAB function “filtfilt” was used to avoid introducing time lags. Relative timing of peak activation between muscle sites was derived separately for each subject in each MC. For each subject-by-MC, the electrode channel corresponding to the ES signal with the best signal quality was selected as the reference for cross-correlation with each of the other seven EMG channels using the MATLAB function “xcorr.” Cross-correlations between EMG signals were performed for each given trial of 10–25 s in duration, and then the mean lag and standard deviation (SD) of each were averaged across the trials. The number of trials per subject and per condition included in the analyses are listed in Table 1. Outliers were identified as lag values smaller or larger than 2 SD from the mean. These amounted to 6% of the total number of right-side ES lag values. The rhythmic nature of the movements constrained lags to be less than one cycle period, which explains why the criterion for outliers was 2 SD rather than the usual 3 SD.

The relative latency between the EMG recorded by the reference electrode and each of the other electrodes was defined as the lag at the peak of the cross-correlation function. The set of relative latencies for each trial was then normalized to percentages of the mean movement cycle period (%T) of the trial. Lag%T for each muscle was represented as the percentage of mean cycle period relative to rL4 by subtracting Lag%T for rL4 from all others. Thus Lag%T for rL4 was always 0%. Positive values corresponded to activation preceding that at rL4 (lead), and negative values, to activation following rL4 (lag). The Lag%T values for each electrode site were then averaged across trials by condition and subject. Mean lags for each signal were calculated from not fewer than six trials per subject. For simplicity in determining the intersegmental coordination patterns of ES compartments, only right-side values for Lag%T were used in the analyses, since muscle activation was symmetrical on the right and left sides, differing by one half cycle.

Statistical analyses.

All statistical analyses were performed using MATLAB or SPSS. All confidence limits were set at 95% (P < 0.05). Subject means for each electrode site were first calculated across the trials for each MC before pooled means and SD were determined across subjects. To test the hypothesis that pattern changes with frequency (first hypothesis), the 2-Hz pattern was compared with the 3-Hz pattern and the 3-Hz pattern was compared with the FAP pattern. We calculated the Euclidean distance between Lag%T values (square root of the sum of squared differences) at corresponding ES locations (excluding rL4, which was always 0%) for each subject for 2 Hz and 3 Hz and for 3 Hz and FAP. Thus, for each comparison, there was a set of 24 values, corresponding to the Euclidean difference between two MCs for each of the 24 subjects. Each set of differences was tested for significance with respect to a selected median value of 13%T using a Wilcoxon signed-rank test. The test median was based on the following rationale. A difference of 25%T in the Lag%T value (a quarter cycle) at any electrode location would represent a substantial shift in the relative timing because it is equivalent to the relative time difference between minimum and maximum rate of shortening during muscle contraction in a rhythmic movement. We reasoned that a shift of even half this magnitude, or a 13%T overall difference in the Euclidean distance between groups, would be functionally significant.

To test for a difference in timing patterns between TRG groups (second hypothesis), each NOV subject was paired with every TRD subject and the Euclidean distance between Lag%T values at corresponding ES locations was calculated for each of the three MCs. This generated a set of 432 distances (12 NOV × 12 TRD × 3 MCs). The hypothesized median Lag%T value of 13%T was used as the test statistic.

In classifying activation patterns we considered timing differences observed in lateral shifting movements of the pelvis (belly dance hip slide) where we found a timing difference of 50%T, a diagonal pattern synergy pattern during opposing lateral motion in upper and lower trunk segments. Also, in our previous study (Nugent et al. 2012), a difference of 22–28%T was found between L1 and L5 in pelvic isolation movements in the majority of the subjects, although some subjects produced simultaneous activation. Thus we expected ES latencies to vary between ~0%T (simultaneous) and ~50%T (diagonal). The findings of Ceccato et al. (2009) and de Sèze et al. (2008) also suggested that an intermediate phase-shifted pattern could be expected.

To investigate both the degree of segmental control specificity in the ES muscles during rhythmic lateral spine undulations at different tempos, and to more comprehensively characterize differences among activation patterns, we performed nonparametric Friedman’s two-way ANOVA by ranks on T10–L4 %T for each classified pattern separately, with TRG groups pooled. The values for each right-side ES electrode position (rL4, rL2/3, rL1, rT10) were evaluated as repeated measures with stepwise post hoc comparisons between electrodes/segmental levels. Significance was adjusted for multiple post hoc comparisons. Any missing %T values were filled with column means. These amounted to 11 of 272 data points (subject means by electrode/MC), or 4% of the total right-side ES interelectrode latency values, none of which were rT10 or rL4 values, being either rL2/3 or rL1.

RESULTS

The activity in the ES showed clear bilateral antiphasic patterns with peak activation at each segmental level occurring at one-half cycle delays between opposite sides the spine. Bursting at each electrode location occurred at the same frequency as the movement frequency. Fig. 2 shows the muscle activation pattern (8 traces at bottom) for both right- and left-side ES at all segmental levels and the associated segmental rotations for the FAP MC (1 trained subject, 1 trial, ~4.5 movement cycles). At the right of each EMG trace, the relative peak activation timing as a percentage of the movement cycle is shown, ordered relative to rT10 peak activity as defining the beginning of the cycle. These data show a clearly sequentially phase-shifted pattern of activity ipsilaterally from T10 to L4 levels with left-right 50%T difference at each segmental level. The top three traces of Fig. 2 show that the rotation amplitudes in the frontal plane for the three segments are similarly phase shifted.

Fig. 2.

Sequential activation of the ES during the FAP shimmy: example from one TRD subject (T10) for one trial averaging 5.0 Hz movement frequency (4–5 movement cycles shown). Bottom: 8 traces representing EMG signals corresponding to right-side ES (r; black traces) and left-side ES (l; gray traces) for the 4 segmental levels: tenth thoracic (T10), first lumbar (L1), midway between the second and third lumbar (L2/3), and fourth lumbar (L4). Top: frontal plane rotation angles for PELV, TH12, and TH1 segments. Clockwise (CW) rotation corresponds to right hip elevation, left shoulder depression; counterclockwise (CCW) rotation corresponds to left hip elevation, right shoulder depression.

Figure 3 compares the relative timing across right-side ES segmental levels for NOV and TRD groups by MC. Relative timing appears to differ both by training group and MC. The relative timing above and below L2 tends to be greater in the TRD group and greater with decreasing movement frequency. Figure 4 provides a similar depiction of ES timing by MC, separately for each subject (Fig. 4A, NOV subjects, and Fig. 4B, TRD subjects). Comparing these two figures, we see that the large variance in the mean latency values of pooled data is due primarily to the between-subjects range and variability of pattern selection.

Fig. 3.

Relative timing in the ES comparing the effects of MC and TRG. Mean lags have been shifted on the x-axis such that rT10 peak activity occurs at 0% of the movement cycle. Error bars equal 1 SD for each electrode mean, representing the between-subject/within-group variability. Where no bars are visible, 1 SD is smaller than the width of the data symbol. n = 12 for each TRG group for each MC.

Fig. 4.

Relative timing in the ES comparing the effect of MC for each subject individually. A: NOV group. B: TRD group. Mean lags have been shifted on x-axis such that rT10 peak activity occurs at 0% of the movement cycle. Error bars equal 1 SD for each electrode mean, representing the within-subject variability. Where no bars are visible, 1 SD is smaller than the size of the data symbol. Missing data points resulted from technical problems with the signal or lack of phasic bursting at the site recorded.

Effect of movement frequency on ES pattern.

The first hypothesis predicts that the CNS modifies the relative timing of ES muscle activation at different segmental levels as it increases the frequency of a rhythmic spinal movement. Table 2 lists the means and SD for T10–L4 timing latencies by TRG group and MC. For the TRD group, the mean timing difference between ipsilateral rostral and caudal electrode locations in both the 2-Hz and 3-Hz MC is about one-half the cycle period. The 3-Hz pattern shows greater variability in timing (SD = 12.5%T) compared with the 2-Hz condition (SD = 5.0%T) and thus might represent a threshold between two timing patterns, since increasing variability with increasing movement frequency has been found to precede spontaneous gait and multi limb pattern changes (Jeka et al. 1993). The mean for the FAP condition is less than 30%T. Like the 3-Hz condition, this pattern has a large variance (SD = 16.2%T).

Table 2.

Mean T10–L4%T by MC and TRG

	TRD					NOV
MC	n	Mean	SD	Mean −SD	Mean +SD	n	Mean	SD	Mean −SD	Mean +SD
2 Hz	12	57.6*	5.0	52.6	62.6	12	16.8	26.3	−9.6	43.1
3 Hz	12	43.5	12.5	31.0	56.0	12	26.6	20.6	6.0	47.2
FAP	12	23.7	16.2	7.5	39.9	12	21.1	17.8	3.3	38.8

Descriptives of T10–L4%T for each movement frequency condition (MC) include n, mean, and SD, as well as the mean plus and minus SD, by training (TRG) group to show the difference in specificity between the two. TRD, trained subjects; NOV, novices.

^*Outliers removed: −2.17, 0.23.

Results from the Wilcoxon signed-rank test for differences in timing patterns, represented as the Euclidean distance between MCs, showed that the Euclidian distance across ES levels was not significantly different between 2-Hz and 3-Hz MCs, since it was equal to the hypothesized median of 13%T: median = 13%T (P = 0.157, 1-tailed, N = 24). However, the Euclidian distance across ES levels was different for 3-Hz and FAP MCs: median = 17%T (P = 0.034, 1-tailed, N = 24). Thus, for pooled TRG groups, there was no clear change in pattern between the 2-Hz and 3-Hz movement frequencies, but the effect of frequency on pattern was confirmed in the comparison between the 3-Hz and FAP conditions.

Figure 5A shows box plots of T10–L4 Lag%T by MC and TRG groups. Whereas the TRD group shows a clear trend of decreasing relative timing of ES activation with increasing frequency, the picture is less clear for the NOV group, whose T10–L4 timing differences range from zero to one-half of the cycle period for every MC. This is consistent with expectations that new learners employ a wider range of strategies in performance attempts, whereas experts show both high within- and between-subject consistency in the performance of skilled movement. What is noteworthy is that each subject maintained a consistent pattern within each MC across the 10 trials. That is, for a given MC and subject, there was no pattern switching across trials. This is indicated by the small within-subject variance at each ES level, as depicted in Fig. 4.

Fig. 5.

Descriptive plots of activation timing differences between ipsilateral ES recording locations. A: mean T10–L4%T values by MC and TRG group. Box edges denote 25th and 75th percentiles. Horizontal lines inside boxes denote medians. Whiskers denote extreme values; crosses denote outliers. B: differences between NOV and TRD groups in the distribution of T10–L4%T values (MCs pooled). C: overall effect of TRG on means for each right ES segment (MCs pooled).

Effect of training on ES pattern.

The second hypothesis predicts that the preferred timing pattern will depend on the amount of training such that patterns requiring more specific control will be observed more frequently for TRD than NOV subjects. Results from the Wilcoxon signed-rank test for differences in pattern between TRG groups showed the pattern of relative timing across ES levels was not the same for TRD and NOV subjects. The Euclidean distance between the two groups was greater than the test median of 13%T: median = 31%T (P < 0.0005, 1-tailed, N = 432). Figure 5B shows the frequency distributions of T10–L4 Lag%T values for NOV and TRD groups. Note the multimodal characteristic, which reflects the separation of T10–L4 relative timing between simultaneous and nonsynchronous ipsilateral activity. The NOV group shows a greater tendency toward ipsilateral simultaneous ES activation than the TRD group, with values around 0% occurring most frequently. Asynchronous timing dominates in the TRD group with most lags in the 45%–65% range. Figure 5C illustrates the overall lower lag values for NOV compared with TRD groups. The asynchronous relative timing requires more specificity of control than the simultaneous activation of muscles at different segmental levels, suggesting that highly trained performers employ muscle activation patterns that require specific control more often than novices.

Timing patterns by MC and classification of activation patterns.

From visual examination of EMG there appeared to be three distinct rostrocaudal timing patterns. Figure 6 shows two sets of EMG traces recorded from NOV subject N2, depicting typical examples of the simultaneous (S) bursting pattern across ipsilateral ES levels in two different shimmy frequency conditions: 2 Hz and FAP (mean frequency 3.5 Hz). The respective Lag%T values are given to the right of each EMG trace. For the 2-Hz MC, the S pattern was seen in 8 of 12 NOV subjects but in only 2 of 12 TRD subjects. In contrast, the majority of the TRD subjects (10) showed a strict antiphasic pattern between ipsilateral ES T10 and ES L4. Examples from two TRD subjects (T10, T8) performing the 2-Hz and 3-Hz MCs, respectively, are shown in Fig. 7. In this pattern, ES activity at more rostral levels is antiphasic (antagonistic) to ipsilateral activity at more caudal levels. By virtue of the half-cycle antiphasic activity occurring at each vertebral level, taking right- and left-side ES together, this ipsilateral antiphasic coordination represents a diagonal (D) synergy between contralateral ES T10 and L4. Only 4 of the 12 NOV subjects showed this D pattern for the 2-Hz MC. Both the S and D bursting patterns were also seen in the 3-Hz and FAP MCs; however, a third, asynchronous (A) pattern emerged as the dominant one in the FAP condition. Examples from one NOV and one TRD subject in Fig. 8 (see also Fig. 2) clearly show incremental phase shifts in burst activity along a rostral-caudal direction in ipsilateral segments of the ES. This A pattern also occurred in the 3-Hz condition, but never in the 2-Hz MC. The frontal plane rotation amplitudes for segments TH1, TH12, and PELV are shown for each of the EMG examples in Figs. 6–8. Strictly antiphasic rotation between TH1 and PELV seen for the S pattern (Fig. 6, top) would suggest that the spine describes one continuous c-curve between these vertebral levels, consistent with the idea of simultaneous activation along spine lateral flexors from ES T10 to L4. Reduced amplitude of TH1 motion with no particular phase relationship to PELV (Fig. 7A, top) or a clear antiphasic relationship between TH12 and PELV (greater angle between these 2 segments; Fig. 7B, top) appeared to correspond to the D pattern. The ES T10 activation synchronized with the contralateral L4 burst likely provides the counter moment needed to stabilize the shoulder girdle/upper spine and prevent ipsilateral thoracic lateral flexion, possibly creating an additional curve inflection point for an s-shaped spinal curvature to produce the segmental motion. The kinematics corresponding to the A pattern (Fig. 8, top) illustrate a clear phase shift in the rotational movement between segments in a rostral-caudal sequence.

Fig. 6.

EMG and segment kinematics for the simultaneous (S) pattern for single trials of 1 NOV subject (N2) for the 2-Hz MC (A) and the FAP MC (B). Bottom: 4 traces representing EMG for right ES. Top: frontal plane rotation angles for PELV, TH12, and TH1 segments. Clockwise (CW) rotation corresponds to right hip elevation, left shoulder depression; counterclockwise (CCW) rotation corresponds to left hip elevation, right shoulder depression.

Fig. 7.

EMG and segment kinematics for the diagonal (D) pattern for single trials of 2 TRD subjects: subject T10, 2-Hz MC, left-side ES (A) and subject T8, 3-Hz MC (actual frequency 3.3 Hz), right-side ES (B). Bottom: 4 traces representing EMG for each ES level. Top: frontal plane rotation angles for PELV, TH12, and TH1 segments. Clockwise (CW) rotation corresponds to right hip elevation, left shoulder depression; counterclockwise (CCW) rotation corresponds to left hip elevation, right shoulder depression.

Fig. 8.

EMG and segment kinematics for the asynchronous (A) pattern for single trials of the FAP MC for NOV subject N5 at 3.3 Hz (A) and TRD subject T10 at 5.0 Hz (B). Top: frontal plane rotation angles for PELV, TH12, and TH1 segments. Bottom: 4 traces representing EMG for right ES. Clockwise (CW) rotation corresponds to right hip elevation, left shoulder depression; counterclockwise (CCW) rotation corresponds to left hip elevation, right shoulder depression.

With confirmation of a change in pattern with movement frequency from the comparisons between MCs, and the indication of pattern differences coming largely from the TRD group, the descriptive statistics for the mean T10–L4%T lags by MC for the TRD group were used to reclassify all subject × condition latency sets as S, D, or A patterns. Note that in the TRD 2-Hz condition, there are two extreme outliers (Fig. 5A), with values around zero, which were not included in the calculation of the mean and SD. The ipsilateral pattern of activation was considered simultaneous if T10–L4%T was less than the TRD mean − 1 SD for the FAP condition, 8%T. The asynchronous pattern was defined as T10–L4%T being greater than 8%T but less than 31%T (mean − 1 SD for the 3-Hz condition). The right-left diagonal coordination pattern was identified where T10–L4%T was greater than 31%T and L2/3–L4%T was greater than 8%T to meet the requirement that activation above and below L3 was one-half cycle out of phase. There were two cases in which T10–L4% was greater than 31%T (N5: 42%T and T9: 41%T) but with L2/3%T values being extreme outliers (34%T for N5, 32%T for T9) compared with the mean/SD for L2/3 (1.11/1.97). Consequently, these two cases were classified as A patterns. The D pattern was further subdivided into one group that showed synchrony of L1 with caudal (L4, L2/3) and one with more rostral (T10) ES levels: D-ros and D-cau, respectively.

Table 3 lists the means and SD for T10–L4%T once all cases were reclassified by pattern type. Figure 9 depicts the relative timing at ES segmental levels for each pattern (TRG pooled), illustrating more clearly than in Fig. 3 (by MC) the different possible ES coordination patterns for producing the shimmy oscillations. Results from the Friedman’s ANOVA tests across right-side ES levels by pattern illustrate the different characteristics of each pattern with regard to relative timing between segmental levels of the ES. There was no difference in timing across the four vertebral levels in the S pattern [χ²(3,19) = 0.778 (4 TRD, 15 NOV), P = 0.855]. Overall difference in peak activation timing of ipsilateral levels of the ES spanning T10–L4 was significantly different from zero for both diagonal patterns [D-cau pattern: χ²(3,15) = 16.33 (7 TRD, 8 NOV), P = 0.000; D-ros pattern: χ²(3,15) = 22.47 (12 TRD, 3 NOV), P = 0.000]. These findings are self-evident given that the two ipsilateral but antiphasic activation groupings in the D pattern correspond to functional specificity in opposite halves of the movement cycle. Overall timing difference across ES levels was also significantly different from zero in the A pattern [χ²(3,20) = 56.70 (11 TRD, 9 NOV), P = 0.000].

Table 3.

Mean T10–L4%T by pattern

				Pattern Occurrence			TRD/NOV			Frequency, Hz
Pattern	Criteria	Mean	SD	n	n TRD	n NOV	2Hz	3Hz	FAP	Mean	Min	Max
S	T10–L4%T < 8	1.2	3.1	19	4	15	2/8	0/3	2/4	2.9	2.0	5.4
D	T10–L4%T > 31 and L2/3 < 8	50.7	8.4	33	21	12	10/4	9/6	2/2	2.7	2.0	4.2
A	8 ≤ T10–L4%T ≤ 31	21.9	11.1	20	11	9	0/0	3/3	8/6	4.2	2.6	6.3
N				72	36	36	24	24	24

The criteria listed are the cutoff values for T10–L4%T from which cases were classified as S, D, or A patterns. Means and SD for each pattern are from pooled TRG groups, which results in SD that are even smaller than those for the MCs of the TRD group only. Summary of pattern occurrence by TRG and task condition highlight were the specific differences or similarities between TRG groups. Most frequently occurring patterns by TRG and MC are highlighted in bold. Mean, minimum, and maximum movement frequencies are indicated for each pattern.

Fig. 9.

Relative timing between right-side ES segments for the simultaneous (S), diagonal (D), and asynchronous (A) patterns. Cases in the D pattern group were further subdivided dependent on whether rL1 was synchronous with rostral (D-ros) vs. caudal (D-cau) segments. Mean lags have been shifted on the x-axis such that rT10 peak activity occurs at 0% of the movement cycle. Error bars represent 1 SD. Where none are visible, 1 SD is smaller than the size of the data symbol.

The results of the multiple comparisons between rES levels by pattern are summarized in Table 4. For the two D patterns, these confirm the strict synergistic/antagonistic functional distinctions in ipsilateral ES at different segmental levels. For the two D patterns and the A pattern, T10 timing was significantly different from that at L4 (P < 0.05), which distinguishes these three from the S pattern. For the D-cau pattern, right rL4-rL2/3-rL1 formed one group with simultaneously activity, all having a significantly different latency from peak activation at rT10. In the D-ros pattern, the two groupings of activation were rL4-rL2/3 vs. rL1-rT10. In the A pattern, rL1 timing was not different from that at rT10, and rL4 was not different from that at rL2/3. However, rL1 and rL2/3 peak activation timing was significantly different (P < 0.05). Thus the A pattern shows the same distribution of significant differences in timing across the ES levels as in the D-ros pattern, but with all of the activity occurring in the same half cycle, suggesting even greater segmental specificity. Overall, the results suggest a capacity for ipsilateral segmental specificity of activation timing in the ES at three levels spanning T10 and L4.

Table 4.

Lag %T pairwise comparisons by pattern

	S	D-ros	D-cau	A
Overall P	0.855	0.000†	0.000†	0.000†

	Significance	Adjust Significance
T10–L4		0.000†	0.004	0.000†
T10–L1		1.000	0.000†	0.165
T10–L2/3		0.000†	0.000†	0.000†
L1–L2/3		0.000†	1.000	0.029*
L1–L4		0.011*	1.000	0.000†
L2/3–L4		1.000	1.000	0.300
N	19	15	15	20

Results of Friedman’s ANOVA: overall timing difference between right-side T10 and L4 by pattern and post hoc tests between ES levels.

^*Significance at P < 0.05.

^†Significance at P < 0.001.

Table 3 also highlights the variation of patterns across the MCs for each TRG group. Once again, we see that the S pattern dominated the 2-Hz MC for the NOV group (8 of 12), whereas the D pattern was the preferred coordination for the TRD group (10 of 12). For the 3-Hz condition, the D pattern was the most utilized activation pattern for both groups (TRD: 9 of 12; NOV: 6 of 12). The remaining three TRD subjects utilized the A pattern for 3 Hz, which suggests that this frequency may be a threshold for pattern change. Five subjects displayed no pattern changes with frequency/movement condition: N2 utilized the S pattern throughout conditions, and subjects N7, N11, T4, and T11 utilized the D pattern throughout.

Maximum achievable movement frequency in the hip shimmy was significantly higher for the TRD group, with the average being 1.09 Hz faster than that of the NOV group (P < 0.0005). Maximum frequency (SD) for the NOV group was 3.90 (0.45) Hz and ranged from 3.26 to 4.79 Hz. For the TRD group, maximum frequency was 4.95 (0.77) Hz and ranged from 3.60 to 6.31 Hz. Table 3 lists the mean, minimum, and maximum movement frequencies for each pattern. Pattern changes from S or D to A occurred anywhere from 2.6 Hz (minimum for A pattern) to 4.2 Hz (mean for A pattern and maximum for D pattern). At frequencies of 4 Hz and above, the D and S patterns were not observed, suggesting that at higher frequencies, the asynchronous pattern is the only possible one and may be selected involuntarily.

Figure 10 shows the linear regressions for the relationship between right-side T10–L4 latencies (in ms) and movement frequency, plotted and grouped by pattern category. At 2 Hz there are two clusters of values around 0%T and 45–55%T, respectively, showing again that, at the lowest movement frequency, subjects selected one of two possible coordination patterns. An interaction between training and movement frequency can be seen. Two opposite trends in latency scaling occur with increasing spine oscillation frequency: T10–L4 timing difference for the simultaneous pattern increases slightly with frequency, whereas that for the D pattern decreases sharply with decreasing T (increasing frequency), switching to an asynchronous pattern. Both trends converge around a 20- to 50-ms timing difference between ipsilateral T10 and L4 ES levels once oscillation frequency of the spine reaches ~4 Hz. Kruskal-Wallis tests confirmed that timing difference between ipsilateral T10 and L4 ES was not equivalent across patterns [χ²(2,72) = 57.42, P = 0.000]. Results of post hoc pairwise comparisons for each combination of patterns were all significant [S vs. A: χ²(1,39) = −20.00, P = 0.002; S vs. D: χ²(1,49) = −43.66, P = 0.000; A vs. D: χ²(1,50) = 23.66, P = 0.002].

Fig. 10.

Relationship between ipsilateral T10–L4 relative timing (in ms) and movement frequency with TRG groups pooled. Each data point corresponds to an MC mean for 1 subject (3 data points per subject). Simultaneous (S) pattern denoted by crosses, diagonal (D) pattern by circles, and asynchronous (A) pattern by squares.

DISCUSSION

The present study used voluntary segmentally specific trunk oscillations from belly dance to explore coordination and control of spine extensor muscles in rhythmic lateral spine undulations. Different patterns were identified by differences in the relative timing of ES muscle activation across four vertebral levels spanning T10 to L4. Changes in erector spinae activation timing patterns with increasing frequency confirmed the hypothesis that different control mechanisms underlie rhythmic undulating spine motion at different tempos in humans. Differences between novices and trained performers in the likelihood of engaging a particular activation timing pattern occurred at the lower frequencies, supporting the hypothesis that these mechanisms are modifiable by training.

Three distinct muscle activation patterns were identified: ipsilateral simultaneous bursting across segmental levels of erector spinae, with alternation between right and left sides every half cycle (S pattern); a diagonal synergy seen as simultaneous contralateral activation of the ES at T10 and L4 levels, alternating with the opposite diagonal synergy every half cycle (D pattern); and nonsynchronous, sequential activation in a rostral-caudal direction along ipsilateral ES segments (A pattern). The S and D patterns are analogous to standing waves of activation, whereas the A pattern resembles a metachronal traveling wave of activation. There was a general trend for the activation pattern to change with movement frequency, as well as an interaction between the amount of training and the pattern most frequently adopted.

Pattern A, which involved the greatest degree of compartmental specificity, seen as incrementally phase shifted activation along segments, emerged between 3 and 4 Hz. At the lowest shimmy frequency (2 Hz), the S and D patterns were equally likely, but there was an influence of training on pattern selection, with S being more likely for NOV subjects and D occurring more frequently in TRD subjects. As movement frequency increased, the T10–L4 Lag%T increased for the S pattern but decreased for the D pattern as both converged toward the A pattern. At the higher frequencies, the effect of training was manifested as differences in maximum movement frequency, with TRD achieving faster tempos.

The A and S patterns closely resembled those described by de Sèze et al. (2008) in their investigation of axial muscle patterning in humans during various forms of locomotion. From recordings at four spinal levels of the ES (C7, T3, T12, L4), they observed a traveling descending wave during forward walking, backward walking, and crawling on hands and knees. In forward and backward walking, the phase shift in burst onset increased linearly with distance between electrode sites (ES muscles) in a rostrocaudal direction. The increasing phase shift in peak burst activation that we found in our A pattern mirrors this finding. In contrast, when de Sèze et al. (2008) recorded activity in amble walking (homolateral arm/leg moving in the same direction), ES activation showed no significant change in burst onset timing across ES levels, similar to our finding in the S pattern of no difference in peak activation timing across the ipsilateral ES segmental levels. De Sèze et al. (2008) compared their findings to the gait-dependent patterns in salamanders (Delvolvè et al. 1997). Trotting over land involves ipsilateral simultaneous activation of axial muscles along segments between the fore- and hindlimb girdles, a right-left antiphasic standing wave of activation analogous to the S pattern. In swimming, axial muscle activation switches to a rostral-caudal sequential wave analogous to the A pattern. The S and A spine extensor patterns observed in the present study may involve pattern generators that produce an analogous phase shift between gaits.

In the lamprey (Grillner 1985), the capacity for rhythmic motor output is distributed along the cord, with each segment activating a particular motoneuron pool. Intersegmental coordination is achieved via mutual excitation of serially coupled networks. This segment-to-segment excitation occurs at a constant phase lag. Left-right coordination of antiphasic activity is afforded through reciprocal inhibitory connections between corresponding units at each segmental level. This functional arrangement allows for the type of propagating sequential wave of paraspinal muscle activation along the spine that we observed in the A pattern at the higher shimmy frequencies.

This same overall organization is presumed to exist in the salamander, however, with additional circuitry for controlling the legs and coordinating their motion with that of the trunk. Ijspeert (2001) used simulations and robotic salamanders to show that tonic drive to the spinal oscillators produces the swimming pattern, with increased level of drive producing faster swimming. If the drive to the leg oscillators dominates that of the spinal oscillators, the spinal pattern is entrained to the leg pattern, which results in the characteristic side-bending pattern of the trunk (ipsilateral synchronous activation in axial muscles between limb girdles). Ijspeert noted that during very fast trotting, the spinal activation pattern of salamanders returns to the undulatory pattern of swimming. This is similar to the switch we observed from the S to the A pattern once spine oscillations reached 3–4 Hz.

Using localized NMDA excitation to stimulate fictive locomotor patterns in in vitro lamprey spinal cord preparations, Matsushima and Grillner (1992) showed that increasing a segment’s excitability causes it to become the leading segment, which also determines the cycle duration. Adjacent segments are entrained by the one with the higher excitability at a fixed phase lag. Furthermore, the magnitude of the phase lag between segments is determined by the difference in excitability between the leading segment and adjacent ones: the greater the difference in excitability, the greater the phase lag between segments.

Recent studies using in vitro spinal preparations from neonatal rats suggested that similar central mechanisms may underlie the generation of rhythmic axial muscle patterns in mammals (Beliez et al. 2015; Falgairolle and Cazalets 2007). These researchers recorded from the ventral roots at six segmental levels (T4, T6, T10, T12, L2, L5) during fictive locomotion. They identified three spinal regions capable of independent generation of bilateral alternating motoneuron activity: thoracic (axial muscles), lumbar (axial and hindlimb muscles), and sacral (tail). In addition, activity was metachronal, traveling caudorostrally with a constant intersegmental phase lag. The authors suggested that the neuronal networks that produce sequential activation patterns in the axial muscles of primitive vertebrates are highly conserved in mammalian locomotor circuits. These include both crossed connections and long and short propriospinal connections to produce the undulating spine movements that allow coordination between trunk and limbs during locomotion.

There is some evidence that trunk muscle activation patterns similar to those we observed may be innate. Felt et al. (2012) described slow, cyclic sideways bending movements recorded in human embryos. This simple and stereotyped motion is the only movement pattern from weeks 5.5 to 7 of development. The authors propose that these patterns reflect the earliest developmental features of neural networks. Similar findings in a variety of animal species suggest their evolutionary conservation. Dominici et al. (2011) hypothesized a developmental trajectory for locomotion that begins with motor primitive patterns are that retained and modified through experience: during the first few months, basic motor patterns emerge via additions to existing circuitry and tuning through the increasing influence of descending and sensory signals. These researchers elicited a stepping response in newborns, recording kinematics and EMG of leg and trunk muscles, including ES at the L2 level. Their results showed striking similarities between neonatal humans and rats, and among toddlers, monkeys, cats, rats, and guinea fowl. Their findings are presented as evidence for the phylogenetic conservation of locomotor patterns across vertebrates as inborn motor primitives that are modified and augmented rather than discarded and replaced. In light of these findings, it may be plausible that the belly dance shimmy involves exploitation of such patterns, possibly in the form of learning to selectively activate them instead of more dominant patterns.

Pattern selection in axial muscles during human rhythmic spine motion is related to training and movement frequency. At lower frequencies, training and experience influence pattern selection. The S pattern does not appear to require independent control of trunk muscles, unlike the D pattern, which may require learning to disrupt innate global trunk lateral flexor patterns to create a new pattern where muscles and/or muscle compartments at different segmental levels can be specifically controlled. The undulatory locomotor patterns of the lamprey are modifiable for different behaviors such as burrowing or mating (Grillner 1985), which require different relative timing of muscle activation. The independent control of parts of the neural network of the lamprey spinal cord that produce such timing differences may provide insight into human control of trunk muscles.

Training appears to be indispensable in achieving the highest movement frequency, where experience is related to the ability of trained dancers to modulate muscle bursts at higher frequencies to achieve higher hip shimmy frequencies than novices. Acquired skill may also include learning to inhibit unwanted activation such as concomitant increased drive to limbs, which may interfere with or overshadow the emergence of primal axial pattern generation.