Split-arm swinging: the effect of arm swinging manipulation on interlimb coordination during walking

Control mechanisms for four-limb coordination in human locomotion are not fully known. To study the influence of arm swinging (AS) on bilateral coordination of the lower limbs during walking, we introduced a split-AS paradigm in young, healthy adults. AS manipulations caused deterioration in the anti-phased stepping pattern and impacted the AS amplitudes for the contralateral arm, suggesting that lower limb coordination is markedly influenced by the rhythmic AS during walking.

Abstract

Human locomotion is defined by bilateral coordination of gait (BCG) and shared features with the fore-hindlimb coordination of quadrupeds. The objective of the present study is to explore the influence of arm swinging (AS) on BCG. Sixteen young, healthy individuals (eight women; eight right motor-dominant, eight left-motor dominant) participated. Participants performed 10 walking trials (2 min). In each of the trials AS was unilaterally manipulated (e.g., arm restriction, weight on the wrist), bilaterally manipulated, or not manipulated. The order of trials was random. Walking trials were performed on a treadmill. Gait kinematics were recorded by a motion capture system. Using feedback-controlled belt speed allowed the participants to walk at a self-determined gait speed. Effects of the manipulations were assessed by AS amplitudes and the phase coordination index (PCI), which quantifies the left-right anti-phased stepping pattern. Most of the AS manipulations caused an increase in PCI values (i.e., reduced lower limb coordination). Unilateral AS manipulation had a reciprocal effect on the AS amplitude of the other arm such that, for example, over-swinging of the right arm led to a decrease in the AS amplitude of the left arm. Side of motor dominance was not found to have a significant impact on PCI and AS amplitude. The present findings suggest that lower limb BCG is markedly influenced by the rhythmic AS during walking. It may thus be important for gait rehabilitation programs targeting BCG to take AS into account.

NEW & NOTEWORTHY Control mechanisms for four-limb coordination in human locomotion are not fully known. To study the influence of arm swinging (AS) on bilateral coordination of the lower limbs during walking, we introduced a split-AS paradigm in young, healthy adults. AS manipulations caused deterioration in the anti-phased stepping pattern and impacted the AS amplitudes for the contralateral arm, suggesting that lower limb coordination is markedly influenced by the rhythmic AS during walking.

human walking is characterized by a specific pattern of bilateral coordination of gait (BCG). The legs oscillate in a reciprocal manner (i.e., anti-phased left-right stepping), with symmetric step lengths and step times (Finley et al. 2013; Reisman et al. 2005). Experimental evidence suggests that interlimb coordination during human walking, characterized by consistent phasing between the rhythmic activity of each of the limbs, is an important factor in healthy and stable locomotion (Dietz and Berger 1984; Roerdink et al. 2007; Wu et al. 2009). Interlimb coordination during walking involves all four limbs and appears to be gait speed dependent (Donker et al. 2001; Mirelman et al. 2015; Plotnik et al. 2013).

It has been suggested that interlimb coordination of human gait evolved from quadruped locomotion and that arm movements and stepping are controlled by spinal central pattern generators (CPGs) (Dietz 2002a, 2002b, 2011; Dietz et al. 2001; MacLellan et al. 2012, 2013a). Unlike quadruped gait in animals, rhythmic movements of the upper limbs are not critical for enabling effective locomotion in humans (Nakakubo et al. 2014), yet they contribute to minimizing energy consumption, as well as optimizing stability and neural performance (Arellano and Kram 2014; Bruijn et al. 2010; Ledebt 2000). Arm swinging (AS) during walking is characterized by upper interlimb and ipsilateral lower limb anti-phase coordination (Dietz 2002a, 2011). Functionally, AS is often considered a mechanism for counteracting free vertical moments (i.e., torque about the body’s vertical axis) due to the swinging of lower extremities and thus allows forward progression by balancing the angular moments (Ballesteros et al. 1965; Braune and Fischer 1895; Collins et al. 2009; Elftman 1939; Hogue 1969; Meyns et al. 2013).

Whether AS during gait is mostly active or passive is a question that has yet to be elucidated (Collins et al. 2009; Kubo et al. 2004; Kuhtz-Buschbeck and Jing 2012; Pontzer et al. 2009). The “passive theory” is supported by the argument that the trunk and shoulders act primarily as elastic linkages between the pelvis, shoulder girdle, and arms (Bruijn et al. 2010; Goudriaan et al. 2014; Meyns et al. 2013). However, arm electromyography signals and clinical observation [e.g., AS asymmetry in patients with Parkinson’s disease (PD)] indicate that arm movements are more than an “epiphenomenon” of passive motion. An integrative approach may thus suggest that AS can be viewed as an integral part of human bipedal gait, arising mainly from passive movements which are stabilized by active muscle control as part of the neural control of locomotion (Kuhtz-Buschbeck et al. 2008; Meyns et al. 2013).

Neuronal substrates enabling quadrupedal coordination of gait have been described in animal studies (see, e.g., Grillner 2006 and Kiehn 2006 for reviews). Results from human studies implicitly support the existence of similar organizational schemes for human gait involving coordination among the four limbs. For example, studies performed on split-belt treadmills with walking speed dictated differentially, demonstrated contralateral effects on leg muscle activation patterns in response to the manipulation of gait speed for the contralateral leg (Dietz et al. 1994). The effect of split-belt walking on AS also supports the active quadrupedal coordination theory (Dietz 2002a, 2011). A handful of studies employing AS manipulations focused mainly on the biomechanical effects rather than coordination (Ballesteros et al. 1965; Collins et al. 2009; Hogue 1969; Stephenson et al. 2009). Based on the above, Fig. 1A gives a speculative heuristic model of quadrupedal control in humans.

Fig. 1.

Factors influencing quadrapedal coordination during human walking. The general layout of this figure is inspired from graphic and conceptual models depicting interlimb coordination in animals (for example, by Ijspeert et al. 2007, Fig. 1; by Maes and Abourachid 2013, Fig. 5B; by Grillner 2006, Fig. 8). Direct electrophysiological confirmation for most of the model components hardly exists for humans (e.g., Barthelemy and Nielsen 2010). A: during walking, supraspinal (i.e., cortex, cerebellum, basal ganglia, brain stem) input provides ongoing activation of locomotion-related CPGs in each of the limbs. This “higher” control also interferes with the CPGs’ “automatic” activity to alter and adapt locomotion patterns, or to execute voluntary motor activity. These CPGs are also under the influence of direct (e.g., proprioceptive) or indirect (e.g., visual) sensory input (sensory input can also eventuate in change of supraspinal control) (O’Connor and Donelan 2012). Unit CPGs, here, a group of individual CPGs generating synergic compound rhythmic activations of different muscles around different joints within a limb (denoted by green circles with red sine symbol inside), have phase relationships with any other unit CPGs (denoted by thin green lines; two upper limb-contralateral lower limb coordination lines are not shown). Lower limb coordination has left-right anti-phased stepping pattern. B: speculative mechanism for impact of arm swinging manipulation on lower limb BCG. When one of the arms is intentionally subjected to excessive swinging (e.g., the right arm), the amplitude of the rhythmic activation of its unit CPGs is increased (bolded red sine symbol). Potentially, the altered proprioceptive input from this arm also affects the function of this unit CPG. We anticipate that the increased amplitude rhythmic activation of the unit CPGs of the right arm, in this example, will affect the rhythmic activation of the unit CPGs of the right leg, with which it has phase relationship. Consequently, we hypothesize that lower limb coordination will be altered. Contributions to these alterations may also arise from the need for the participant to concentrate (i.e., to engage simultaneous cognitive resources) to produce excessive arm swinging. As a consequence, BCG may worsen in terms of the ability to maintain the left-right anti-phased stepping pattern. Note: black frames in A turn red in B to symbolize that alterations are expected in these factors. Thin green lines in A turn bold black in B, when coordination changes between the respective limbs are anticipated. The gray circle on the left shoulder represents the effect on AS of the contralateral arm (see hypotheses in the text).

Understanding the role of AS in gait coordination has significant clinical importance, as individuals with neurological disorders (e.g., poststroke, PD, multiple sclerosis) demonstrate gait asymmetry (GA), impaired BCG (Kalron and Achiron 2014; Meijer et al. 2011; Plotnik et al. 2007, 2014) and an increased risk of falling (Fasano and Plotnik 2012; Giladi and Nieuwboer 2008; Plotnik et al. 2005, 2008, 2011).

The objective of the present study is to introduce a new paradigm using split-AS to explore the influence of AS on BCG. We hypothesize that unilateral manipulation of AS will impact contralateral AS [similar to the dynamics between the legs in split-belt treadmill (TM) studies, e.g., Mawase et al. 2013, Mohammadi et al. 2015], and that upper limb swinging manipulations will influence coordination between the lower limbs.

Figure 1 heuristically lays out the theoretical framework for the latter hypothesis. Presumably, the control of rhythmic movements generated by one or several CPGs is influenced by direct and indirect sensory and motor inputs modulated by supraspinal regions (i.e., cortex, cerebellum, basal ganglia, brain stem; Fig. 1A). Manipulation of AS via one or both arms affects sensory control inputs to the CPGs responsible for executing the rhythmic movements of the manipulated arm(s). Under the assumption that a network of CPGs orchestrates the coordinated rhythmic movements of the various limbs (Ijspeert et al. 2007), modulation of CPG activity related to AS will affect coordination between all interconnected CPGs, including coordination between the lower limbs (Fig. 1B). This hypothesis is also true for conditions in which AS of both arms is manipulated simultaneously. In these cases, arm-leg coordination is altered on both sides, and it is hypothesized that these alterations do not “cancel each other out,” but rather exacerbate interleg discoordination (see also Table A1 in appendix a).

METHODS

Design Rationale

There was no a priori hypothesis regarding the effect of arm dominance. However, in hand-tapping paradigms, the dominant hand has been reported to tap more rapidly and rhythmically (Hubel et al. 2013; Peters 1980). We, therefore, decided to control for this factor by 1) recruiting one-half left- and one-half right-hand dominant participants; 2) performing unilateral manipulations on each arm (see below); and 3) incorporating dominance within our statistical models (see below).

Participants

Sixteen healthy participants between the ages of 18 and 40 yr participated in the study. One-half of the participants exhibited left motor dominance, and one-half right motor dominance, as defined by sidedness questionnaires (Elias et al. 1998; Papadatou-Pastou et al. 2013), and within these groups there were equal numbers of men and women. Inclusion criteria were as follows: 1) normal joint range of motion; 2) limb muscle strength rated 5/5 on the Medical Research Council scale for muscle strength (Paternostro-Sluga et al. 2008); 3) intact sensation in four limbs; 4) 20/20 vision (corrected or uncorrected); 5) self-declared ability to participate in gait trials for about 1 h; and 6) able to understand written or verbal study instructions. Exclusion criteria were as follows: previous or current medical conditions (e.g., orthopedic, rheumatic, neurological, cardiovascular) that might have affected gait or balance. The experimental protocol was approved by the Human Studies Committee of the Sheba Medical Center, and all participants provided written, informed consent before entering the study.

Apparatus

Figure 2 depicts the experimental setup. A split-belt TM (R-Mill, ForceLink) placed in a virtual reality facility (V-Gait, Motek Medical) was used. In the present study, no virtual environment was projected on the large monitor. The tied belts mode was used (i.e., always same speed for both TM belts). A motion capture system (Vicon, Oxford, UK) recorded kinematic data from an array of passive markers attached to the walking participant’s body (41 markers). The sampling rate was 120 Hz. TM speed was regulated in a self-paced mode by a built-in controller algorithm described in depth elsewhere (Bohannon 1997; Sloot et al. 2014).

Fig. 2.

Depiction of the V-Gait system. In the present study, no virtual environment was projected on the large monitor. A cross was presented in the middle of the screen, roughly at the height level of the participant, and the participants were asked to maintain fixation on the cross while walking.

Experimental Procedure

Physical evaluation.

Before execution of the gait protocol, participants completed a basic physical examination, demographic questionnaire, anthropometric measurements, and two sidedness questionnaires: one for the upper limbs, including 10 items from a previously described questionnaire, and one for the lower limbs, which also includes 10 questions (Elias et al. 1998; Papadatou-Pastou et al. 2013).

Gait protocol.

Before the beginning of the session, each participant was hooked to a safety harness, hanging from the ceiling above the TM. The safety harness did not provide body weight support to the participants. Gait trials began with walking practice on the TM in a self-paced mode (Plotnik et al. 2015). Data collection started once the participant reported that he/she felt comfortable with this TM mode.

The experimental conditions were as follows:

•   1.Normal walking (i.e., without any manipulation; “baseline”).
•   2.Walking with weight (0.45 kg) on wrist of dominant arm.
•   3.Walking with weight (0.45 kg) on wrist of nondominant arm.
•   4.Walking with weights (0.45 kg each) on both wrists.
•   5.Walking with dominant arm restrained. Restraint was achieved by tying a wristband to the safety harness at hip level.
•   6.Walking with nondominant arm restrained.
•   7.Walking with both arms restrained.
•   8.Walking with excessive AS of dominant arm (participant was instructed to increase AS).
•   9.Walking with excessive AS of nondominant arm.
• 10.Walking with excessive AS of both arms.

To eliminate cross-condition order effects, condition order was randomized. Each condition lasted 90 s, measured from when the participant reached a “steady-state” speed. Details on the steady-state speed values achieved in each condition are provided in appendix b. The start of the steady-state period was determined by a real-time algorithm that identified the time point at which the standard deviation of the TM speed became <0.07 m/s for 12 s (sample-by-sample moving window; TM speed sampling rate was 120 Hz).

Data handling.

During offline analyses, the final 85 s of the steady-state walking period were used. Data quality, i.e., uninterrupted capturing of the markers’ data allowed us to obtain gait parameters from 75–80 gait cycles in 151 of the 160 trials and from only 18–73 gait cycles in the remaining nine trials.

Time series describing the spatial position of the following body markers were extracted: heel, toe, elbow and shoulder, bilaterally. These were imported by a MATLAB graphical user interface that was built to semiautomatically detect the time points of the heel strike (HS) and toe off of each foot within a gait cycle.

The use of force plate (FP) data is usually considered the “gold standard” for detecting HSs. Specifically, a HS is considered to have occurred when the vertical component of the ground reaction force crosses a threshold of 20 N (Wells and Winter 1980). As we used a dual-belt treadmill, extracting HS times using this method might be compromised whenever the participant steps with the left leg on the right belt and vice versa. Therefore, we chose to extract HS time from the vertical axis of the heel marker. We validated this procedure against the conventional procedure of obtaining HS times from the FPs and found only a 9-ms difference. The full validation process is described in appendix c.

From the elbow and the shoulder markers, we calculated the AS amplitude for each gait cycle according to the following formula:

$$\text{AS\hspace{0.17em}amplitude}=\left[X_{\text{ELf}}–X_{\text{SHf}}\right]–\left[X_{\text{ELb}}–X_{\text{SHb}}\right]$$ (1)

where X_Elf is the anterior-posterior coordinate of the elbow marker when the hand is in its forward-most position; X_SHf is the anterior-posterior coordinate of the shoulder marker when the hand is in its forward-most position; and X_ELb and X_SHb are the coordinates of the corresponding markers when the hand is in its rear-most position.

primary outcome measures.

Along with AS amplitudes, we focused on the gait parameters characterizing left-right stepping coordination and symmetry:

• 1.Phase coordination index (PCI): The coordination of left-right stepping was assessed using this measure. A full description and derivation of the PCI metric is detailed elsewhere (Meijer et al. 2011; Plotnik et al. 2007). Briefly, PCI is a metric that combines the accuracy and consistency of stepping phase generation with respect to the value of 180°, which represents the ideal anti-phased left-right stepping. Lower PCI values reflect more consistent and accurate phase generation, while higher values indicate more impaired BCG (Krasovsky and Levin 2009; Plotnik et al. 2008, 2013).
• 2.Gait asymmetry (GA): This measures the relative difference between left and right leg swing times (Krasovsky and Levin 2009; Plotnik et al. 2005; Yogev et al. 2007). Based on prior studies of AS during split-belt walking (e.g., MacLellan et al. 2013b), we hypothesize that unilateral but not bilateral manipulations of AS will affect GA. With unilateral manipulation of AS, we assume that asymmetric upper limb swinging will cause a “limp-like” epiphenomenon in the lower limbs to counterbalance the rhythmic rise in angular momentum.

secondary outcome measures.

The secondary outcome measures are mean gait speed and gait variability (as measured by the coefficient of variation of mean stride time) during the steady-state period.

Statistical Analysis

To evaluate the study hypotheses that AS manipulation impacts BCG (as measured by PCI) and GA, separate general linear (mixed) (GLMs) models with repeated measures were applied to PCI and GA data. To control for motor dominance, we used a statistical model that had 10 × 2 levels: 10 Condition levels (within participants) and 2 Dominance levels (between participants). Significant effects were followed up by post hoc contrast analyses comparing each manipulation condition with the baseline condition. Correction for multiple comparisons used the least significance difference method. One missing GA value was imputed by inserting the group mean for the relevant experimental condition (walking with nondominant arm restrained).

In the event that the full model revealed significant effects, we analyzed each type of manipulation (weights, restraining, excessive AS) separately, using GLM models to more specifically explore the role of side of motor dominance. These models had three Condition levels (manipulation on dominant arm, manipulation on nondominant arm, and manipulation on both arms). This statistical design was employed to evaluate the hypothesis that unilateral manipulation of AS impacts AS amplitude of the contralateral arm (see results).

Secondary outcomes were treated similarly, and any additional statistical procedures used are described in the results. A P value of ≤0.05 (two sided) was considered statistically significant. Analyses were performed using the SPSS software (SPSS version 21.0, IBM).

RESULTS

Table 1 summarizes the demographic and physical characteristics of the participants with right- and left-side motor dominance.

Table 1.

Demographic and physical characteristics of participants

	Left-Side Dominance		Right-Side Dominance
Characteristic	Mean ± SD	Range	Mean ± SD	Range	P Value*
Age, yr	31.19 ± 5.71	24.0–38.5	34.53 ± 5.01	25.50–40.25	0.175
Scores on hand-sidedness questionnaire	4.3 ± 0.4	3.7–5.0	1.2 ± 0.2	1.0–1.6	<0.001
Scores on feet-sidedness questionnaire	3.5 ± 0.6	2.2–4.0	2.1 ± 0.9	1.0–3.6	0.012
Height, cm	165.8 ± 9.4	151.5–181.0	170.6 ± 10.2	157.0–181.0	0.406
Weight, kg	64.8 ± 14.2	45.6–87.7	71.8 ± 15.2	44.7–89.8	0.389
BMI, kg/m2	23.3 ± 3.2	17.6–26.8	24.4 ± 3.2	18.1–27.4	0.502

Values are means ± SD and ranges; n = 16 participants (8 women). An equal number (i.e., 4) of participants with left-side motor dominance and right-side motor dominance were included in each sex group. BMI, body mass index. Values in bold are significantly different.

^*Paired t-test.

Effects of AS Manipulation on AS Amplitudes

Figure 3 depicts the mean values of AS amplitude during AS manipulation. Data from the excessive AS conditions were analyzed to test the effects of side of dominance (i.e., dominant side vs. nondominant side) and walking condition on the dependent variable of AS amplitude. We used repeated measures within participants (2 × 4) to explore two factors: Dominance (2 levels: dominant arm, nondominant arm) and Condition (4 levels: baseline; walking with excessive swinging of the dominant arm; walking with excessive swinging of the nondominant arm; walking with excessive swinging of both arms). The analysis showed a main effect of Condition (F_3,45 = 81.97; P < 0.001) but no effect of Dominance (F_1,15 = 0.28; P = 0.6). There was no interaction between the factors (F_3,15 = 2.32; P = 0.09). As expected, the “excessive swinging” arm had higher values of swinging amplitude, while swinging amplitudes for the “passive” arm decreased significantly (see Fig. 3 for details).

Fig. 3.

Arm swinging amplitudes during the experimental conditions. Data for the baseline, weights on both wrists, and excessive swinging of both arms conditions were combined as no difference was found between arms (see also Fig. 6). Values are means ± SE. Levels of statistically significantly differences from the baseline condition are indicated above the bars.

Similar repeated-measures analysis was performed for the four walking conditions: baseline, wearing weight on the wrist of the dominant arm, wearing weight on the wrist of the nondominant arm, and wearing weights on the wrists of both arms. The analysis showed only a main effect of Condition (F_3,45 = 7.41; P < 0.001), with no significant effects of Dominance (F_1,15 = 2.42; P = 0.141) and no significant interaction (F_3,15 = 1.12; P = 0.351).

From the data obtained in the restrained arm condition, we found that both the dominant and the nondominant AS amplitudes were increased when the contralateral arm was restrained (paired t-tests, t > 3.67; P < 0.003).

The Bilateral Relations between AS Amplitudes during the Symmetric and Split AS Conditions

We found that, at baseline, in the condition with weights on both wrists and in the “excessive swinging” condition, AS amplitudes were not statistically significantly different between dominant and nondominant arms (3 × paired t-tests, t ≤ 1.07; P ≥ 0.301). In Fig. 4A, AS amplitude values are plotted for these conditions. It can be seen across participants that AS is symmetric, as exhibited by the fact that AS amplitudes are roughly equally distributed on both sides of the diagonal unity line. Interestingly, at baseline, the left AS values were significantly higher than the right AS values (t = 2.80; P = 0.0136; paired t-test).

Fig. 4.

A: arm swinging amplitudes of the nondominant hand (NDH) are plotted against those of the dominant hand (DH) in all symmetric conditions (baseline, solid circles; excessive arm swinging of both arms, shaded circles; and weights on both wrists, X). B: the relative change (%) occurring during the unilateral excessive arm swinging conditions in the NDH are plotted vs. the changes occurring in the same conditions in the DH. It can be seen that the relative increases are highly correlated. Regression line Y = aX + b (solid line) is a good fit for all data points (solid circles), excluding an outlier (surrounded by dashed circle). Regression line coefficients are a = 0.6 and b = 18.1 (%) (R² = 0.79, P < 0.001). Data are also plotted for the relative passive decrease in arm swinging occurring in the arm contralateral to the “excessive swinging” arm (open circles). It can be appreciated that the relative passive decrease is correlated between the arms (Spearman’s ρ = 0.594; P = 0.015), and that there is a good linear regression fit for these data as well (Y = a′X + b′a′ = 0.6; b′ = −8.2; R² = 0.35, P = 0.017).

Figure 4B depicts the relative percent increase in AS for the “excessive swing” condition (compared with baseline) for the nondominant vs. dominant arm (solid circles). It can be seen that the relative increase in over-swinging of the nondominant arm strongly correlates with the relative increase in the over-swinging of the dominant arm. The mean values (±SE) of the relative increase of the dominant and the nondominant arms were 174 ± 36 and 150 ± 35%, respectively (t = 0.73; P = 0.478, paired t-test; see legend for more details).

Furthermore, it can be appreciated (Fig. 4B, open circles) that the relative decrease (i.e., percentage) in swing amplitude of the passive arms (i.e., those not in the “excessive swing” condition) is also correlated between the dominant and nondominant arms.

No correlation was found between the relative increase in AS amplitude of the nonrestrained dominant arm and the relative increase in the nonrestrained, nondominant arm (Spearman’s ρ = 0.121; P = 0.656). The mean values of the relative increases (±SE) were 37.0 ± 6.1 and 24.9 ± 7.1%, respectively (t = 1.43; P = 0.172, paired t-test).

Effect of AS Manipulation on BCG

An example of the effect of a split-AS manipulation on the antiphase stepping pattern is shown in Fig. 5. Data from one participant during baseline condition (left) and from the walking condition in which he was instructed to excessively swing his left arm (right) are depicted. It can be seen that, in the latter condition, the generated stepping phase values become more variable and more distant from the “ideal” 180° value compared with baseline. Accordingly, the PCI value is increased.

Fig. 5.

An example of arm swinging performance and stepping phases seen in one participant with left-side motor dominance. Two walking conditions are compared. During baseline walking, this participant exhibited slightly increased left arm swinging amplitudes and generated a relatively consistent anti-phased right-left stepping pattern. In the walking condition for which he was instructed to excessively swing his left arm (with no instructions regarding his right arm), it can be appreciated that, while left arm swinging amplitudes increased, the right arm swinging amplitudes decreased. Simultaneously, stepping phase generation was disrupted, as also reflected by the PCI values.

Figure 6 summarizes the effects of the various AS manipulations on BCG as expressed by PCI values. Each panel shows group mean values of PCI during AS manipulation on the dominant arm, the nondominant arm, and on both arms. For comparison, mean PCI value at baseline is shown.

Fig. 6.

Effects of arm swinging manipulations on PCI values. Mean PCI values in the different experimental conditions are shown. Values are means ± SE. *Significant difference from the baseline condition (P < 0.05).

For PCI, the statistical model revealed a significant effect for Condition (F_9,126 = 2.99; P = 0.003) but not for Dominance (F_1,14 = 0.56; P = 0.47). The Dominance × Condition interaction was not statistically significant (F_9,126 = 0.34; P = 0.96). Post hoc contrasts showed that baseline PCI value was significantly lower (P < 0.042) in comparison to six conditions: weight on nondominant arm, restraining of dominant arm, restraining of both arms, and all three excessive AS conditions. The conditions in which PCI was not significantly different from baseline were as follows: weight on dominant arm (P = 0.1), weights on both arms (P = 0.077), and restraining of nondominant arm (P = 0.14). With regard to differential effects of body weight and use of the same wrist weight for all participants (i.e., 0.45 kg), we found that post hoc results for comparisons between conditions with weights and baseline remained the same when body mass index was included in the model as a covariate.

The separate post hoc models ran for the three groups of conditions (i.e., weights on wrists, restraining of arms, and excessive AS) showed no significant intercondition differences for each of the three cases (P > 0.5).

Effect of AS Manipulation on GA and Gait Cycle Swing Times

For GA, the effect of Condition (F_9,126 = 1.77; P = 0.080) failed to reach significance, and there was no effect of Dominance (F_1,14 = 0.05; P = 0.826) or significant Dominance × Condition interaction (F_9,126 = 0.94; P = 0.492).

To assess the overall effect of AS on swing times generated by the legs during the gait cycle, we first calculated the average between left and right swing times for each condition. A GLM model using this dependent variable revealed a significant effect of Condition (F_9,135 = 5.87; P < 0.001). It is important to note that the actual differences were rather minute. For example, swing time increased by 0.019 ± 0.022 s in the “excessive swinging” condition (both arms) compared with baseline (i.e., ~5% change). A full description of the effect of AS manipulation on leg swing times is provided in appendix d.

Correlation Between PCI Values Under Different Conditions

Correlational analyses showed a significant relationship among PCI values in the following experimental conditions: baseline, wearing weight (×3 conditions), arm restraint (×3 conditions), and in the “excessive swinging” condition (Spearman’s ρ ≥ 0.597; P ≤ 0.015; for an example of a very high correlation, see Fig. 7; Spearman’s ρ = 0.956; P < 0.001). The relatively high correlations among PCI values in the various conditions is consistent with the high reliability of the PCI metric.

Fig. 7.

Correlation between PCI values when both arms are restrained vs. baseline (normal) walking. The robust impact of arm restraint in reducing bilateral coordination of stepping is reflected by the observation that the majority of the points are above the unity line (dotted diagonal line). For the three exceptional points that fall below the line, the mean relative decrease (±SE) in PCI was −3.1 ± 0.8%. For all other participants (n = 13), PCI increased (worse coordination). The mean relative increase (±SD) in PCI was 25.5 ± 3.7%. Short solid line represents the linear regression line: PCI_{both arms restrained} = A × PCI_baseline + B, where A = 1.10 and B = 0.33 (R = 0.91; P < 0.001).

Effect of AS Manipulation on Gait Speed and Gait Variability

For gait speed, the statistical model revealed no significant effect of Condition (F_9,126 = 1.30; P = 0.241). However, the Dominance effect was found to be significant (F_1,14 = 9.273; P = 0.009; for more on this finding, see appendix e). The Condition × Dominance interaction was not statistically significant (F_9,126 = 0.69; P = 0.419).

No significant effects were found for gait variability (Condition: F_9,126 = 1.33; P = 0.228; Dominance: F_1,14 = 0.91; P = 0.357; Condition × Dominance interaction: F_9,126 = 0.265; P = 0.98).

Sex Effect

Although we had no a priori hypothesis regarding sex, given that our convenience sample comprised an equal number of men and women, we ran post hoc analyses to examine potential sex effects. The results are presented and discussed in appendices f and g, respectively.

DISCUSSION

Summary of Findings

AS manipulations during walking influence the coordination between the lower limbs, as reflected by the PCI metric. In contrast, gait speed, asymmetry, and variability are not affected by AS manipulations. Manipulating AS amplitude in one arm leads to changes in AS amplitude in the contralateral arm. AS amplitudes of the left and right arm are symmetric during baseline walking and during conditions in which the same manipulation (e.g., restraining) is simultaneously applied to both arms. Results were similar for manipulation of the dominant or the nondominant hand.

To What Extent Do the Results Support the Study Hypotheses?

Our first hypothesis stated that “unilateral manipulation of AS will affect contralateral AS.” This hypothesis is supported by the current findings. When one arm is restrained, the contralateral arm swings excessively compared with baseline (Fig. 3). This is consistent with previous findings (Ford et al. 2007). The results also indicate that, when participants excessively over-swing one of their arms during walking, AS amplitudes of the contralateral arm decrease, but no effect exists in the contralateral nonmanipulated arm when weights are attached to the wrist.

Our second hypothesis that upper limb swinging manipulations will influence coordination between lower limbs is also supported by the present findings. Six out of the nine conditions showed a significant effect on PCI values (c.f., Fig. 6). For the other three conditions (c.f., results), PCI values were increased (as in the case of the six “significant” conditions). It might be the case that the sample size (n = 16) did not provide sufficient statistical power in these cases (0.14 ≥ P ≥ 0.07).

Interpretation of the Results: Effect of AS Manipulation on AS Amplitudes

In general, the effect on AS amplitude is opposite (contralateral) to the manipulation (e.g., the contralateral AS is decreased when one arm swings excessively or the AS of the unrestrained arm is increased in comparison to baseline walking, c.f., Fig. 3). This does not support the passive/mechanical AS theories. Based on these theories (e.g., Collins et al. 2009), the introduction of inappropriate counterrotations (by decreasing or increasing the usual AS) is expected to be accompanied by similar effects in the contralateral arm.

Thus we propose that the contralateral effect on AS is the result of neural processes related to various motor control inputs that affect gait-related multilimb rhythmic motor activity. According to our model (c.f., Fig. 1), both sensory input and supraspinal (e.g., cortical) input contribute to control of locomotion. Thus, when a participant focuses on increasing AS amplitude on one side during an excessive AS trial, an attention shift (i.e., supraspinal effect) may compromise contralateral AS, resulting in an amplitude decrease (Figs. 3–5).

While this assertion, i.e., that the participant’s perception drives AS amplitude effects during the excessive AS trials, is largely speculative, it is supported, in our opinion, by the following findings: 1) change in AS amplitude was correlated within participants between the dominant and the nondominant arms; 2) the synergistic reduction in AS amplitudes in the contralateral arms were also bilaterally intercorrelated. Put differently, the perceptual quantification of “excessive” swinging is consistent within a participant, irrespective of which arm is excessively swung, and, consequently, the relative reductions in AS amplitude of the contralateral arms are also intercorrelated within participants. Still, sensory feedback from the excessively swung arm to the contralateral arm unit CPG (c.f., Fig. 1) or reciprocal inhibition between the two arm unit CPGs cannot be ruled out.

Unilateral AS manipulation with weights did not significantly increase the AS amplitudes of the contralateral arm. Apparently, while effective in reducing AS amplitudes in the manipulated arm (Fig. 3), the relatively low load of the 0.45-kg weights (<1% of body weight) was not sufficient to drive a contralateral effect.

Interpretation of the Results: Effect of AS Manipulation on Lower Limb Gait Parameters

Among the lower limb gait parameters studied, PCI (reflecting BCG) and swing times were sensitive to AS manipulations. That PCI was found to be correlated within participants across the different conditions (Fig. 7) is consistent with its internal reliability. PCI directly assesses the coordination of anti-phased left-right stepping. It is, therefore, suggested that AS manipulation does not influence the rhythm of the lower limbs during walking, as gait variability was not affected.

The preserved gait symmetry during AS manipulations suggests that neither the ipsilateral nor the contralateral lower limb (relative to the manipulated side) is differentially affected in terms of their swing times, not concurring with our hypothesis about unilateral AS manipulations and their impact on GA. Yet the results also point to a bilaterally present small effect on lower limb swing times (see the results and discussion in appendices d and g, respectively).

Based on these results and in accord with the proposed heuristic model (Fig. 1), we speculate that the observed changes in upper limb amplitude affect the coordination between the lower limb unit CPG and the contralateral lower limb unit CPG. A small amplitude effect propagates bilaterally in a roughly equal manner (i.e., change in lower limb swing times). Therefore, it may be proposed that the present split-arm swing paradigm results “mirror” findings obtained with split-belt TM paradigms, as reported by MacLellan et al. (2013b): 1) amplitudes of the contralateral limb (relative to the manipulated side) are affected; 2) the level of symmetry between the function of the nonmanipulated pair of limbs is not affected; 3) subtle amplitude change in the swinging of the nonmanipulated pair of limbs is observed (see elaborated quantitative comparative analyses in reference to split-belt TM studies in appendix g).

In terms of the a priori model presented in Fig. 1, we suggest that the change in interleg coordination is likely manifested through ipsilateral neural connections between the upper and lower limbs (Fig. 1) (Grillner 2006; Kiehn 2006).

Regarding the minute changes observed in swing times, it is worth noting that leg swing times increased during the excessive AS conditions, decrease during the weights conditions, and remained fairly unchanged in the arm restraining conditions (c.f., Table D1 in appendix d). In reference to the model components depicted in Fig. 1, one might speculate that, once the upper limb unit CPGs are set to rhythmically activate AS in increased/decreased amplitudes, the change is transferred to the lower limbs, perhaps with the mediation of supraspinal overall involvement that shifts the amplitude change to a new “global” scaling set point, acting symmetrically on both legs. Such change is not observed in the restraining conditions, probably because in these conditions the restrained arm function is out of the spectrum of AS amplitudes (i.e., movement is “nulled”).

Clinical Implications

Utilizing physical therapy to induce neural plasticity and motor learning can improve functions in persons with severe neurological conditions [e.g., PD (Petzinger et al. 2013)]. In view of evidence that upper and lower limb movements influence one another during walking (Dietz 2002a; 2011), targeting AS may increase the efficacy of gait rehabilitation programs for several central neurological pathologies (Behrman and Harkema 2000; Meyns et al. 2011; Stephenson et al. 2009). In particular, normalizing coordination between limbs might improve gait performance in poststroke patients, as well as patients with PD (Dietz 2011) or cerebral palsy (Meyns et al. 2012). With regard to biomechanics, it appears useful to properly coordinate the limbs for normalization of angular momentum (Bruijn et al. 2008) and decreased energy expenditure (Collins et al. 2009). Even passive manipulation of the CPGs has been shown effective in improving gait performance (Lo et al. 2010; Ustinova et al. 2011). Therefore, we propose that, in addition to restoring normal AS amplitudes (Meyns et al. 2013), enhancing BCG (i.e., normalizing lower interlimb coordination) should become a target for potential therapeutic interventions, using AS as a “vehicle” to improve “pathological” gait in neurologically disabled individuals and reduce its adverse effects [i.e., falling (Fasano and Plotnik 2012; Giladi and Nieuwboer 2008; Plotnik et al. 2005, 2008, 2011].

Limitations

Studying the function of the presumed spinal neural network in human models requires intense, well-controlled observational behavioral trials, since direct electrophysiological methodologies (e.g., recording neural activity invasively) are not feasible. The present experimental protocol included several gait manipulations, but additional ones that might have provided useful additional information were not included in an effort to reduce the duration of the experiment. For example, the intentional excessive AS trials might have been complemented by intentional inhibited AS trials to more fully evaluate our contention about supraspinal involvement of coordination between limbs (see above).

In addition, this study was limited to young, healthy subjects. Thus caution is required in attempting to generalize to neurologically impaired individuals. For instance, in individuals with PD, characterized by asymmetric AS (Lewek et al. 2010), it would be difficult to ascertain, with regard to coordination between limbs, which of the observed phenomena is part of the pathology and which is an outcome of compensation processes that occur even in the presymptomatic stage of the disease.

Future Research Directions

The present findings, rather implicitly, provide general support for the notion that coordination of rhythmic arm and leg movements during walking is governed by quadrupedal locomotor neural network circuitry and that AS is not a “passive” consequence (Dietz 2002a, 2002b, 2011; Dietz et al. 2001; Falgairolle et al. 2006; Juvin et al. 2005; MacLellan et al. 2013b; Zehr et al. 2009). Yet specific aspects of coordination between limbs evident in the present study as well as earlier studies warrant further examination:

• 1)Some degree of asymmetry in AS during healthy gait is physiological. Significant differences between left and right AS amplitude have been reported in more than one-half of the extant gait trials in healthy subjects (Kuhtz-Buschbeck et al. 2008; Meyns et al. 2013). In agreement with these earlier studies, we noticed that the left arm tended to swing more than the right arm, regardless of arm dominance (data were not presented). Future studies may provide more insight on the laterality of brain functions and on asymmetric deteriorations in AS during the course of progressive neurodegenerative diseases (e.g., PD).
• 2)Additional clinically driven research might aim to reduce the muscle tone of spastic paralyzed arms in patients after stroke (Stephenson et al. 2009, 2010). Would AS improve following such a procedure and, if so, would gait coordination consequently improve?

This study proposes a novel approach, the split-AS concept, as a paradigm for researching limb coordination during human walking. Whether, and how, it can be implemented for clinical use warrants further research.

GRANTS

The study was funded in part by the Refael Rozin Research Fund.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

M.B., G.Z., and M.P. conceived and designed research; M.B. and M.P. performed experiments; M.B., A.B., and M.P. analyzed data; M.B., G.Z., and M.P. interpreted results of experiments; M.B. and M.P. prepared figures; M.B. and M.P. drafted manuscript; M.B., G.Z., A.B., A.F., and M.P. edited and revised manuscript; M.B., G.Z., A.B., A.F., and M.P. approved final version of manuscript.

APPENDIX A: HYPOTHESIZED PROCESSES FOR AFFECTING BCG

One of the two main hypotheses of this study is that upper limb swinging manipulations will influence coordination between the lower limbs. Figure 1 heuristically lays out the theoretical framework for this hypothesis (Fig. 1A) and exemplifies how, in the case of excessive AS, a potential effect on lower limb coordination will arise. Table A1 details similar processes for all types of AS manipulations used in the study.

APPENDIX B: ACTUAL TREADMILL SPEED FOR EACH EXPERIMENTAL CONDITION

As indicated in methods, TM speed was regulated in a self-paced mode by a built-in controller algorithm described in depth elsewhere (Bohannon 1997; Sloot et al. 2014). Analysis of an experimental condition began from the point at which the participant reached a steady state. The steady-state TM speeds for each experimental condition across all participants are shown in Table B1.

APPENDIX C: DETECTION OF HEEL STRIKE TIMES FROM THE HEEL MARKER: VALIDATION VS. THE FORCE PLATE “GOLD STANDARD”

We conducted a validation procedure to assess the difference between heel strike (HS) detection based on the positional data of the heel marker and detections obtained from force plate (FP) data. Briefly, five healthy young adults walked at their own comfortable pace on a split-belt TM (R-Mill, ForceLink) operated in self-paced mode (Plotnik et al. 2015). The TM was equipped with two FPs, one underneath each belt. Positional and FP data were extracted only for gait segments with no stepping of either leg on the contralateral belt. The number of gait cycles included varied from 88 to 125 per participant.

HS, using the kinematic data, was considered to have occurred when the vertical position of the heel marker was at its minimum. HS using the FP data was considered to have occurred when the vertical component of the ground reaction force crossed a threshold of 20 N. To assess time differences in detection, we defined two types of errors, true errors (TE) and absolute errors (AE), as follows:

$$\text{TE}\hspace{0.17em}=\hspace{0.17em}t{\text{(HS}}_{\text{kinematic}}\text{)}-t{\text{(HS}}_{\text{FP}}\text{)}$$ (C1)

$$\text{AE\hspace{0.17em}=}\hspace{0.17em}\left|\text{TE}\right|$$ (C2)

Across the five participants, the mean value of TE (± SD) was 9 ± 3 ms, and the mean value of AE was 23 ± 20 ms. These results are in agreement with offset errors reported in previous studies (Hreljac and Marshall 2000; Wells and Winter 1980; Zeni et al. 2008).

APPENDIX D: THE EFFECT OF AS MANIPULATION ON SWING TIMES OF THE LOWER LIMBS

Although there was no significant effect of AS manipulation on GA, as reflected by swing times, the data suggest a certain pattern (c.f., results). Table D1 shows the swing times for each condition across participants (averaged for both legs). We ran three additional analyses comparing lower limb swing times for baseline walking and the three conditions (i.e., dominant arm, nondominant arm, both arms) for each type of manipulation (i.e., weight, restraint, excessive swinging). For the weight conditions, we found a significant effect (F_3,45 = 3.21; P = 0.032), with baseline significantly different from weights on both arms (P = 0.03; see Table D1). For the restraint conditions, no effect was found (F_3,45 = 0.9; P = 0.448). Finally, for the excessive swinging conditions, we found a significant effect on lower limb swing times (F_3,45 = 3.92; P = 0.014). Post hoc contrasts revealed that baseline lower limb swing time values were significantly different from those for excessive swinging of the nondominant arm (P = 0.003).

APPENDIX E: THE RELATION BETWEEN GAIT SPEED AND ARM DOMINANCE

Gait speed was defined as a secondary outcome measure (see methods). The statistical model revealed that Dominance had a significant effect on gait speed (P = 0.009; see results). For this analysis, gait speed was averaged for each participant across all experimental conditions. Mean (±SD) for the participants with left arm dominance was 1.35 ± 0.16 m/s, and for participants with right arm dominance gait speed was 1.59 ± 0.20 m/s. To assess the magnitude of the dominance effect on gait speed, we ran bivariate correlations (Spearman’s) between gait speed and responses from the hand-sidedness questionnaire (higher values for left-side dominance; see Table 1). For the 16 participants, gait speed values appeared to be inversely related to the scores on the hand-sidedness questionnaire, but this correlation did not reach significance (Spearman’s ρ = −0.433; P = 0.094). Post hoc analyses within each sidedness group (n = 8 in each group) revealed no correlation between gait speed and responses from the hand-sidedness questionnaire (Spearman’s ρ ≤ |0.393|; P ≥ 0.335).

APPENDIX F: SEX EFFECT

The study sample comprised an equal number of men and women. We had no a priori hypothesis regarding sex. However, we ran post hoc analyses to explore potential sex effects. The statistical model had 10 × 2 levels: 10 Condition levels (within participants) and 2 Sex levels (between participants). For the PCI, the statistical model revealed significant effects of Condition (F_9,126 = 3.16; P = 0.002) and Sex (F_1,14 = 11.57; P = 0.004). The Sex × Condition interaction was not statistically significant (F_9,126 = 1.14; P = 0.34). To follow up on the significant effects, we analyzed each type of manipulation (weights, restraining, excessive AS) separately, using GLM models to more specifically explore the respective roles of arm dominance and sex. These models had 4 Condition levels (baseline, manipulation on dominant arm, manipulation on nondominant arm, and manipulation on both arms, within participants) and 2 Sex levels (between participants). Significant effects were followed up by post hoc contrasts among the four conditions using the least significance difference method. Table F1 shows the mean PCI values for each condition and results of the inferential statistical analyses.

APPENDIX G: THEORETICAL CONSIDERATIONS

Impact of Arm Swing Manipulation on Swing Times of the Lower Limbs

Table D1 depicts only small changes in the swing times of the lower limbs in response to AS manipulation. In split-belt TM studies, increased speed for one belt is accompanied by an increase in the amplitude (i.e., step length) of the leg on the fast belt and concomitant decrease in amplitude of the leg on the slower belt (MacLellan et al. 2013b; Reisman et al. 2005). Regarding the upper limbs, only at a relatively high ratio of differences between the belt speeds (e.g., 2:6) is a bilateral increase in arm swing amplitude (of ~50%) observed (MacLellan et al. 2013b). How are current “split AS” results interpreted within this context?

The obtained effect on the contralateral arm for excessive swinging and restrained conditions can be interpreted in a similar manner. Specifically, when one arm is over-swung, the contralateral arm shows reduced AS amplitude, and, conversely, when one arm is restrained, the contralateral arm shows increased AS amplitude. A similar parallel can be drawn to split-belt TM studies for the obtained effects on the lower limb swing times. Specifically, a bilateral ~5% increase in leg swing times was observed when one arm was excessively swung. From Fig. 3 it can be inferred that the ratio of split AS is 1:2.5, roughly equivalent to 2:6. However, while split-belt TM studies show an increase of ~50% in AS amplitudes, the current “split-AS” manipulation yielded only a 5% increase in lower limb swing times.

Sex Effect

Our exploratory analyses revealed a significant sex effect on BCG, as expressed by the PCI metric. This sex effect interacts inconsistently with the effect of the AS manipulation of the PCI values (compare statistical information of the top two sections of Table F1, with that of the bottom one). PCI is a relatively new metric [i.e., it was introduced approximately a decade ago (Plotnik et al. 2007)]. To the best of our knowledge, no sex effect has been reported using this or any other measure of interlimb coordination.

Sex differences in gait performance have been observed throughout the history of the study of human gait. Indeed sex can be accurately identified solely by observing data from markers attached to the body of the walking individual (Kozlowski and Cutting 1977). The sex differences reported are related to such gait measures as joint range of motion and have been shown to interact with age (e.g., Cronström et al. 2016; Kobayashi et al. 2016).

The handful of studies that have addressed gait coordination did not find differences between men and women (e.g., Chow and Stokic 2015). In contrast, sex effects have been reported for other gait measures (i.e., lower GA and regularity in men, only among elderly healthy participants and not in young healthy participants) (Kobayashi et al. 2016). The present results indicate an apparent sex effect for PCI, with men showing lower left-right stepping coordination. However, given the small number of participants, further research is necessary to elucidate this effect. For example, if men indeed have higher values of PCI, can this difference be due to biomechanics? If so, is there a relationship between PCI and body mass index? Additionally, we propose that deeper theoretical work is required to understand the mechanism underlying this presumed effect.

Table A1.

Hypothesized processes for affecting BCG in response to the experimental conditions

Condition*	Hypothesized Processes
Weight on dominant arm	Afferent feedback is different compared with the unloaded condition → altering muscle activation via lower reflexes and supraspinal input → the rhythmic activation of the unit CPGs of the arm will be affected → arm’s unit CPGs coordination with the ipsilateral leg’s unit CPGs will be affected → the rhythmic motor activation of the unit CPGs of the ipsilateral leg will be affected → the phase coordination between the legs will be affected.
Weight on nondominant arm	Similar process as in the case of “Weight on dominant arm.”
Weights on both arms	Within each side of the body, similar process as in the case of weight on a single arm occurs → arm-leg coordination is altered on both sides → alterations do not “cancel each other out” → exacerbation of interleg dis-coordination.
Dominant arm restrained	The movement of the arm is severely restrained (i.e., tied to the hip) → afferent feedback and potential intentional habituation of arm swinging commands severely impair the rhythmic pattern in the arm’s muscles → affect the rhythmic motor activation of the unit CPGs of the ipsilateral leg, with which it has phase relationship. Consequently, we hypothesize that lower limb coordination will be altered.
Nondominant arm restrained	Similar process as in the case of “Dominant arm restrained.”
Both arms restrained	Within each side of the body, similar process as in the case of restraining single arm occurs → arm-leg coordination is altered on both sides → alterations do not “cancel each other out” → exacerbation of interleg dis-coordination.
Excessive arm swing of the dominant arm	See detailed proposed mechanism in Fig. 1 legend.
Excessive arm swing of the nondominant arm	See detailed proposed mechanism in Fig. 1 legend.
Excessive arm swing of both arms	Within each side of the body, similar process as in the case of single arm’s excessive swinging occurs → arm-leg coordination is altered on both sides → alterations do not “cancel each other out” → exacerbation of interleg dis-coordination (greater dual task effect compared with single arm over-swinging; see Fig. 1 legend).

^*There was no a priori hypothesis regarding the effect of arm dominance; see methods.

Table B1.

Self-paced gait speed values in the various experimental conditions

Condition	Gait Speed, m/s
Normal walking (“baseline”)	1.43 ± 0.20
Weight on the wrist of the dominant arm	1.48 ± 0.23
Weight on the wrist of the nondominant arm	1.49 ± 0.20
Weights on both wrists	1.50 ± 0.24
Dominant arm restrained	1.44 ± 0.23
Nondominant arm restrained	1.46 ± 0.19
Both arms restrained	1.42 ± 0.22
Excessive AS of the dominant arm	1.52 ± 0.27
Excessive AS of the nondominant arm	1.47 ± 0.23
Excessive AS of both arms	1.50 ± 0.26

Values are means ± SD.

Table D1.

Swing time values: lower limbs

Condition	Swing Time, s
Normal walking (“baseline”)	0.387 ± 0.029
Weight on the wrist of the dominant arm	0.382 ± 0.029
Weight on the wrist of the nondominant arm	0.383 ± 0.031
Weights on both wrists	0.380 ± 0.030
Dominant arm restrained	0.385 ± 0.031
Nondominant arm restrained	0.388 ± 0.035
Both arms restrained	0.391 ± 0.034
Excessive AS of the dominant arm	0.397 ± 0.029
Excessive AS of the nondominant arm	0.406 ± 0.033
Excessive AS of both arms	0.398 ± 0.034

Values are means ± SD.

Table F1.

Mean PCI Values in different walking conditions

Condition	Men	Women
Walking with weights conditions*
Baseline	3.86 ± 1.18	3.08 ± 0.62
Weight on dominant arm	5.19 ± 2.21	2.91 ± 0.76
Weight on nondominant arm	4.71 ± 1.34	2.94 ± 0.69
Weights on both arms	4.81 ± 1.75	2.97 ± 0.73
Walking with the arm(s) restrained†
Baseline	3.86 ± 1.18	3.08 ± 0.62
Dominant arm restrained	5.01 ± 1.57	3.19 ± 0.63
Nondominant arm restrained	4.67 ± 1.63	2.94 ± 0.66
Both arms restrained	4.74 ± 1.21	3.66 ± 0.98
Excessive arm swinging conditions‡
Baseline	3.86 ± 1.18	3.08 ± 0.62
Dominant arm	5.36 ± 1.90	3.44 ± 0.49
Nondominant arm	5.59 ± 1.45	3.88 ± 0.77
Both arms	5.11 ± 1.53	3.81 ± 0.94

Values are means ± SD in %.

^*A 4 × 2 mixed GLM model showed a significant effect of Sex (F_1,14 = 8.14; P = 0.013) but not of Condition (F_3,42 = 2.59; P = 0.065). Condition × Sex interaction was significant (F_3,42 = 4.33; P = 0.010). To further explore the interaction, we ran a separate model for each sex and found a significant Condition effect for men (F_3,21 = 4.15; P = 0.019), with the baseline condition showing significantly lower PCI values than all three weight conditions (P < 0.037), but PCI did not differ among the individual weight conditions (P > 0.349). The Condition effect was not significant for the female group (F_3,21 = 0.33; P = 0.803).

^†A 4 × 2 mixed GLM model showed a significant effect of Sex (F_1,14 = 7.79; P = 0.014). There was also a main effect of Condition (F_3,42 = 5.26; P = 0.004), as well as a significant Condition × Sex interaction (F_3,42 = 4.36; P = 0.009). To further explore the interaction, we ran a separate model for each sex and found a significant effect of Condition for men (F_3,21 = 4.95; P = 0.009), with a significantly lower PCI for baseline relative to the three other conditions (P < 0.041) and no differences among the remaining conditions (P > 0.207). The effect of Condition was also significant for women (F_3,21 = 4.36; P = 0.015). Among women, PCI was higher when both arms were restrained compared with all other conditions (P < 0.043), which did not differ from one another (P > 0.393).

^‡A 4 × 2 mixed GLM model showed a significant effect of Sex (F_1,14 = 10.76; P = 0.005). There was also a main effect of Condition (F_3,42 = 5.23; P = 0.004), but the Condition × Sex interaction was not significant (F_3,42 = 1.09; P = 0.366). For both sexes, the significant Condition effect was attributable to the difference between baseline and the other conditions (P < 0.017); PCI did not differ among the remaining three conditions.