Conceptual Binding: Integrated Visual Cues Reduce Processing Costs in Bimanual Movements

J Neurophysiol 102: 302 311, 2009. First published April 15, 2009; doi:10.1152/jn.91090.2008. Conceptual Binding: Integrated Visual Cues Reduce Processing Costs in Bimanual Movements N. Wenderoth, 1 M. Van Dooren, 1 A. Vandebroek, 1 J. De Vos, 1 S. Vangheluwe, 1 C. M. Stinear, 2,3 W. D. Byblow, 2 and S. P. Swinnen 1 1 Motor Control Lab, Department of Biomedical Kinesiology, Katholieke Universiteit Leuven, Leuven, Belgium; and 2 Movement Neuroscience Laboratory, Department of Sport and Exercise Science and 3 Department of Medicine, University of Auckland, Auckland, New Zealand Submitted 30 September 2008; accepted in final form 13 April 2009 Wenderoth N, Van Dooren M, Vandebroek A, De Vos J, Vangheluwe S, Stinear CM, Byblow WD, Swinnen SP. Conceptual binding: integrated visual cues reduce processing costs in bimanual movements. J Neurophysiol 102: 302 311, 2009. First published April 15, 2009; doi:10.1152/jn.91090.2008. In discrete reaction time (RT) tasks, it has been shown that nonsymmetric bimanual movements are initiated slower than symmetric movements in response to symbolic cues. By contrast, no such RT differences are found in response to direct cues ( direct cue effect ). Here, we report three experiments showing that the direct cue effect generalizes to rhythmical bimanual movements and that RT cost depends on different cue features: 1) symbolic versus direct or 2) integrated (i.e., action of both hands is indicated as one entity) versus dissociated (i.e., action of each hand is indicated separately). Our main finding was that dissociated symbolic cues were most likely processed serially, resulting in the longest RTs, which were substantially reduced with integrated symbolic cues. However, extra RT costs for switching to nonsymmetrical bimanual movements were overcome only when the integrated cues were direct. We conclude that computational resources might have been exceeded when the response needs to be determined for each hand separately, but not when a common response for both hands is selected. This supports the idea that bimanual control benefits from conceptual binding. INTRODUCTION Principles of bimanual coordination have often been studied by means of rhythmical movement tasks. The most robust finding revealed by this research paradigm is that mirror-symmetric bimanual movements (in-phase) are executed with higher accuracy and consistency than nonsymmetric movements (antiphase) (for reviews see Donchin and Cardoso De Oliveira 2004; Kelso 1995; Swinnen 2002; Swinnen and Wenderoth 2004). This difference in pattern stability becomes particularly prominent when the system is challenged, such as when movements have to be performed in parallel with a secondary task or at higher frequencies that often induce an involuntary switch from antiphase to in-phase (Kelso 1984; Monno et al. 2002; Peper et al. 1995; Temprado et al. 2001). Similarly, planned switches between patterns are performed faster from the less stable antiphase to the more stable in-phase pattern (to IP) than vice versa, i.e., from in-phase to antiphase (to AP) (Byblow et al. 2000; Scholz and Kelso 1990). Also for more complex bimanual tasks, such as moving the left and right arm rhythmically along different trajectories in space, performance is best when the trajectories are symmetric with Address for reprint requests and other correspondence: N. Wenderoth, FABER, KU Leuven, Tervuursevest 101, 3001 Heverlee (Leuven), Belgium (E-mail: nici.wenderoth@faber.kuleuven.be). respect to the body midline (e.g., when both arms perform vertical movements), whereas performance decreases and strong interference emerges for nonsymmetrical trajectories (e.g., one arm moves vertically and the other horizontally) (Swinnen et al. 1997, 1998, 2002; Wenderoth et al. 2004, 2005, 2006). It has been argued that the advantage of symmetric bimanual actions arises from a spontaneous coupling tendency between homologous muscles (Donchin and Cardoso De Oliveira 2004; Serrien et al. 2002, 2003), which is assumed to emerge from transcallosal connections between motor areas of the left and right hemispheres (Carson 2005; Eliassen et al. 2000; Franz et al. 1996). It was hypothesized that limitations in the control of asymmetrical movements can be attributed to intermanual interference arising from such neuronal hard-wiring in the form of transcallosal connections between motor cortices. We will refer to these neural-motor constraints on bimanual coordination as the motor outflow hypothesis. An alternative view has recently been put forward, arguing that bimanual interference does not primarily result from motor programming or implementation but rather depends on the central processes upstream of the associated motor activity (Albert et al. 2007; Diedrichsen et al. 2001, 2003; Hazeltine et al. 2003; Kunde and Weigelt 2005; Kunde et al. 2009; Mechsner et al. 2001; Weigelt 2007; Weigelt et al. 2006, 2007). This central processing hypothesis was strongly supported by results from reaction time (RT) paradigms where discrete bimanual actions are cued using different visual stimuli. Diedrichsen et al. (2001) showed that bimanual interference (as indicated by longer RTs for initiating nonsymmetric than symmetric movements) was abolished when the targets were cued directly by flashing light-emitting diodes at the end position of each hand, compared with symbolic cues, where letters indicated the movement direction of each hand. It was argued that interference arises at a cognitive level rather than the motor programming level (Albert et al. 2007; Diedrichsen et al. 2001, 2003; Hazeltine et al. 2003; Kunde and Weigelt 2005; Kunde et al. 2009; Weigelt 2007; Weigelt et al. 2006). Here, we extended this line of research in several ways. First, in experiments 1 and 2, we investigated whether the findings for discrete bimanual aiming tasks generalize to rhythmic bimanual movements, which have been used in the majority of experiments arguing in favor of the motor outflow hypothesis. We used a rhythmic circle-drawing task, where subjects performed a bimanual pattern until a visual cue instructed them to switch to a new pattern. We compared performance when response selection was completed long before 302 0022-3077/09 $8.00 Copyright 2009 The American Physiological Society www.jn.org

INTEGRATED VISUAL CUES PROMOTE CONCEPTUAL BINDING 303 the switch (experiment 1) to conditions where subjects had to select responses immediately as indicated by the cue (experiment 2). Based on the central processing hypothesis, one would predict that RT differences for switches to IP versus to AP would be found only when the bimanual pattern was selected in response to symbolic switch cues. By contrast, the motor outflow hypothesis predicts that switches to IP are always initiated faster than those to AP. Second, based on the results of experiment 2, we performed an additional experiment asking which specific cue characteristics benefit bimanual switching. It has been argued that symbolic versus direct cues represent a functional dichotomy, such that symbolic cues induce extra RT costs for nonsymmetric compared with symmetric movements, whereas no such differences are found with direct cues. Here we will adopt this terminology in a slightly modified way, whereby symbolic cues consist of letters or other abstract symbols that are arbitrarily mapped to movements. Direct cues relate intuitively to the controlled-movement parameters here direction of the circle-drawing movements (note that this differs from previous studies on aiming movements, where the term direct cues was used for physical targets in space). We hypothesize that processing costs for bimanual movements are additionally influenced by another cue feature i.e., whether the required bimanual response is cued as an integrated entity or as two dissociated components. Integrated cues might consist of two congruent substimuli or represent one common goal for both hands (e.g., a bimanual pattern that needs to be initiated), whereas dissociated cues instruct the left hand independently from the right hand. In particular, the effect of symbolic cues might largely differ when integrated versus dissociated information is provided, such as when symmetric actions are triggered by congruent substimuli (same letter for both hands) and nonsymmetric actions by incongruent substimuli (different letters for the left and right hands) (see e.g., Diedrichsen et al. 2001; Hazeltine et al. 2003; Spijkers and Heuer 1995; Spijkers et al. 1997; Weigelt et al. 2007). It has been argued that voluntary movements are planned as a whole and are executed only after features are bound together (Hommel et al. 2001). Here, we investigated this binding hypothesis and determined whether RTs are shortened when a required bimanual action is symbolized in an integrated way compared with a dissociated way. Because it is difficult to define direct but truly dissociated visual cues we performed two experiments to disentangle the influence of the different cue features. In experiment 2, subjects produced identical bimanual motor responses triggered by either direct integrated cues or by dissociated cues that varied systematically according to their direct versus symbolic nature. In experiment 3, all cues were symbolic, but conveyed either integrated or dissociated information. We hypothesized that an advantage in terms of reaction time and degree of interlimb interference would be accrued from integrated visual cues that facilitate the conceptual binding of the left- and right-hand action into a common movement plan. METHODS Participants Twelve subjects (8 men, 4 women) with a mean age of 24 3yr participated in experiment 1. Sixteen different subjects (3 men, 13 women) with a mean age of 25 4 yr participated in experiment 2. Data from one subject were removed due to an exceptionally high error rate (41.9%) while performing the task. Twelve new subjects (3 men, 9 women) with a mean age of 24 4 yr participated in experiment 3. All subjects were right-handed (Oldfield 1971), naïve with respect to the task, and had normal or corrected-to-normal vision. None had a history of neurological or psychiatric disease or exhibited overt sensorimotor deficits. The study was approved by the local ethical committee of KU Leuven and subjects gave written informed consent in accordance with the Helsinki Declaration. Experiment 1 PROCEDURE AND SWITCHING STIMULI. Participants were seated in front of two joysticks such that their elbows were resting at a table, the forearms were nearly vertical, and the wrists were used mainly to produce circular joystick movements (Fig. 1A). Virtually frictionless joysticks were made in-house using high-precision, optical shaft encoders (spatial resolution 0.18 ; Hewlett-Packard, Penang, Malaysia) that were mounted at orthogonal axes to measure two-dimensional displacements within a horizontal range of 170 and a vertical range of 130. Additionally, subjects looked at a computer monitor, which was used to provide visual stimuli. After becoming familiar with the joysticks, subjects were instructed to perform rhythmical circle-drawing movements to the beat of an auditory metronome (Korg DTM-12, Tokyo, Japan) such that the hands moved either in-phase (i.e., simultaneous inward or outward cycling) or antiphase (i.e., simultaneous rightward or leftward cycling). The critical cycling frequency for antiphase movements was determined (operationally defined as the lowest frequency imposed by the metronome that evoked phase wandering or spontaneous transitions to in-phase) by performing a series of leftward and rightward cycling trials, each lasting 12 s. Between trials, the frequency was increased by steps of 0.0667 Hz. The main experiment was performed at 75% of the individual critical frequency (Byblow et al. 2000), to ensure that both patterns were performed without inducing involuntary phase transitions. Each trial started with an instruction cue, showing both the first and the second bimanual patterns, each represented by two arrows indicating the cycling direction of the left and right hands (Fig. 1B). Subjects had to initiate and maintain the first pattern for a randomized interval of 4 6 s. Then, a neutral switching stimulus appeared for 2 s (green square), indicating that subjects had to switch to the second bimanual pattern (Fig. 1, C and D). This second pattern was maintained for 6 8 s such that one trial had a total duration of 12 s. Thus subjects could select the required motor response well in advance. A series of 32 trials was executed in succession, forming one block. Subjects performed five blocks in total and, between blocks, short breaks of about 5 min were provided. The switches required the subjects to reverse cycling direction of 1) the left hand, 2) the right hand, 3) both hands, or 4) no hand (catch trials). Additionally, transitions required a switch toward either an in-phase or an antiphase pattern. All these factors were counterbalanced in the experiment, such that each transition mode/final pattern combination was performed 20 times. Before each block, subjects were instructed to switch with the correct hand as fast as possible and to preserve a fluent cycling movement with the nonswitching hand to the beat of the metronome. DATA ANALYSIS AND STATISTICS. Data analysis was performed using software written in-house in Matlab 5.3 (The MathWorks, Natick, MA). First, a continuous estimate of the angular velocity was determined for each hand by d /dt with arctan (y/x), where x and y describe the mean corrected values for horizontal and vertical joystick displacements, respectively (Fig. 1, C and D). was filtered by a second-order Butterworth low-pass filter with a cutoff frequency of 5 Hz. The reaction time (RT) was determined as the time interval

304 WENDEROTH ET AL. A B C 4 angular velocity (rad/s) rightward leftward -4 0 left hand between the appearance of the stimulus and the first zero-crossing of, indicating that subjects reversed cycling direction (Fig. 1C). For trials requiring only one hand to switch, contralateral disruptions were identified when the cycle direction of the nonswitching hand was involuntarily reversed for 100 ms, together with the direction reversal of the switching hand (Fig. 1D) (Byblow et al. 2000). A trial was classified as an error when 1) the wrong hand was switched, 2) the RT was 200 ms or 1, ms, or 3) the left and right RTs differed by 250 ms when both hands were switched. Statistical analyses were conducted with Statistica 8 (StatSoft, Tulsa, OK). Data were pooled such that in- and outward cycling was combined for in-phase movements, whereas left- and rightward cycling combined for antiphase movements. RT data were subjected to a repeated-measures ANOVA, with the within-subject factors Mode (one hand, both hands), Final Pattern (in-phase, antiphase), and Hand (left, right). Percentages of contralateral disruptions (%ContraDisr) were analyzed for within-subject factors Final Pattern (in-phase, antiphase) and Hand (left, right). Significance level was set to 0.05. Significant interactions were explored using Tukey s honestly significant difference. Experiment 2 RT right hand -1-0.5 0 0.5 1 1.5 2 Time (s) first pattern D second pattern -1-0.5 0 0.5 1 1.5 2 Time (s) PROCEDURE AND SWITCHING STIMULI. The main question of experiment 2 was how different visual cues influence switching behavior. Four types of switching stimuli were used (Fig. 2, A and B). In the abstract condition (Abst), the switch was indicated by the combination of a small/large triangle and a small/large circle. In the side condition (Side), two colored squares appeared on either side of the screen. A red square indicated the hand to switch. The side-direction condition (Side-Dir) was the same as the Side condition, with the addition of an arrow indicating the new cycling direction(s) of the switching hand(s) overlaid on the red square(s). Finally in the target pattern condition (Target) two arrows directly indicated the new cycling pattern of the hands (Fig. 2B). Note that the first three conditions (Fig. 2A) provide dissociated information, whereas the Target condition (Fig. 2B) provides integrated information. For the dissociated cues, the level of directness increased gradually from the Abst to the Side-Dir condition. Subjects performed a series of five blocks for each stimulus condition, randomized across subjects. Each trial started with an instruction stimulus indicating one of four possible bimanual patterns by displaying the words IN, OUT, LEFT, or RIGHT. Subjects had to initiate and maintain the instructed pattern until a switching stimulus indicated the required switch to the new pattern. For the Abst condition, the association between the symbols and the required action was randomized across subjects and subjects learned this association before the Abst condition block was performed. During this training session, subjects held the joysticks in a neutral position and made a brisk movement with either the left hand, right hand, or both hands, indicating whether they knew the meaning of the Abst cues. They executed four training blocks before and one test block after the switching experiment, whereby each symbol was displayed 10 times per block. DATA ANALYSIS AND STATISTICS. A trial was classified as an error when 1) the wrong hand was switched, 2) the RT was 300 ms or 2,000 ms, or 3) the left and right RTs differed by 250 ms when both hands were switched. RT data were analyzed by a repeated-measures ANOVA with the within factors Cue (Abst, Side, Side-Dir, Target), Final Pattern (inphase, antiphase), Mode (one hand, both hands), and Hand (left, right). The %ContraDisr was subjected to a repeated-measures ANOVA with within-subject factors Cue, Final Pattern, and Hand. In RESULTS, we will focus on our main research question and thus report only main effects and higher interactions containing the factor Cue. However, complete statistics are reported in the supplemental data. 1 Experiment 3 contradisr FIG. 1. A: general task setup showing the position of the subjects using the joysticks. B: exemplary cues used in experiment 1 indicating the transition from the 1st to the 2nd bimanual pattern. Each pattern is indicated by a pair of arrows representing the required cycle direction of the left hand and right hand, respectively. The cues indicate a switch with the right hand (top; see exemplary behavioral response in C and B) and with both hands (bottom). C: time course of angular velocity for a representative righthand switch. Time is synchronized to stimulus appearance and reaction time (RT) was determined on the basis of the zero-crossing of angular velocity (i.e., when the right hand reversed cycling direction). D: right-hand switch producing a contralateral disruption (contradisr) as indicated by the brief reversal of cycling direction of the left hand (black line). PROCEDURE AND SWITCHING STIMULI. The general experimental setup and procedures were identical to those of experiment 2. Each 1 The online version of this article contains supplemental data.

INTEGRATED VISUAL CUES PROMOTE CONCEPTUAL BINDING 305 A Abstract Side Side Direction B Target C Letter Word Switch Both Switch left Switch right No switch FIG. 2. Cues used in experiments 2 and 3. A: dissociated cues of experiment 2: For the Abst cues, the association between the symbol and the required action was learned prior to the main experiment. For the Side and Side-Dir cues, the red square indicates which hand has to reverse direction. B: integrated cues from experiment 2: Target cues indicate the required bimanual pattern by a pair of arrows. The Target cues here indicate (from top to bottom) inward, outward, rightward, and leftward cycling of the 2 hands. C: Letter and Word cues from experiment 3: the Letter cues indicate the required cycling direction (L, left; R, right) such that the left letter indicates the left-hand action and the right letter the right-hand action. The Word cues indicate the required bimanual pattern. Note that font, size, and vertical position of the letters were constantly varied to prevent perceptual stimulus response association from forming. trial started with a nontextual instruction stimulus consisting of two arrows indicating either inward, outward, leftward, or rightward cycling. Three different types of switch cues were used (Fig. 2C). The horizontal letter cues (horletters) were R (for rightward cycling) or L (for leftward cycling). The leftmost and rightmost letters indicated the cycling direction of the left hand and right hand, respectively. In the vertical letter condition (vertletters), letters were displayed vertically. Subjects were instructed that the upper letter indicated the movement direction of the right hand, whereas the lower letter indicated the movement direction of the left hand. The Word cues displayed the required bimanual pattern as a single entity: IN for inward cycling, OU for outward cycling, LE for leftward cycling, and RI for rightward cycling. Letter cues provided directional information of each hand in a dissociated manner, whereas Word cues integrated direction information for both hands together. To prevent subjects from associating cues with a motor response, instead of interpreting their meaning, we constantly varied font, size, and vertical position of the letters shown in both cue types. Subjects performed a series of five blocks for each stimulus condition, randomized across subjects. DATA ANALYSIS AND STATISTICS. Computations of RT and %ContraDisr as well as the statistical analysis by means of a repeatedmeasures ANOVA were analogous to those of experiment 2. RESULTS Experiment 1 For this group, 75% of the individual critical frequency was 1.53 0.25 Hz. Errors occurred on 14% of trials (one hand: 17%; both hands: 10%) and these were removed from the analysis. Switches to IP were initiated faster than switches to AP when stimulus response selection requirements were minimal, irrespective of whether this switch was executed with one hand or both hands, as indicated by a significant main effect of Final Pattern [F(1,11) 4.94, P 0.05] but not of Mode [F(1,11) 3.05, P 0.1] (Fig. 3). Statistics revealed a significant Final Pattern Hand interaction [F(1,11) 6.08, P 0.05]: switches to AP were slower when executed with the left hand (507 40 ms) compared with the right hand (476 25 ms; P 0.05), but for switches to IP, no left right differences were observed (left: 461 39 ms; right: 468 38 ms; P 0.9). RT of the right hand differed only slightly between switches to IP and to AP (P 0.8). There were no other significant effects or interactions (all P 0.09). There were significantly fewer %ContraDis when the left hand reversed direction than when the right hand reversed direction (Left: 10%; Right: 23.0%) [Hand main effect: F(1,11) 32.4, P 0.001]. Experiment 2 In this group, 75% of the individual critical threshold corresponded to a mean movement velocity of 1.61 0.3 Hz (range 1.13 to 2 Hz). In total, 17.9% of the trials were removed from further analysis because of errors. The occurrence of errors depended on the experimental conditions and the lowest error rate was found for Target cues (Abst: 31%; Side-Dir: 25%; Side: 15%; Target: 7%) and for switches with both hands RT (ms) 550 450 400 one hand towards in-phase towards anti-phase both hands FIG. 3. RT results of experiment 1. Mean RT results for switches toward the in-phase (gray) and antiphase pattern (black). Vertical bars indicate the SEs.

306 WENDEROTH ET AL. (one hand: 23%; both hands: 11%). Additional control analyses confirmed that subjects memorized the symbols of the Abst condition successfully and that plateau performance was reached before the symbolic cues were used in the context of the switch task (see supplemental data). The most interesting result was the significant Cue Mode Final Pattern interaction (Fig. 4) (statistics summarized in Table 1). Single-hand reversals (Fig. 4, left) were significantly faster when switching to IP (gray) than when switching to AP (black) but only for the Abst, Side, and Side-Dir cues. For the Target cues, RT differed only moderately between switches to AP and to IP. Switches with both hands were generally faster than switches with one hand, but for both-hand switches no significant RT differences between the two final patterns were found, irrespective of the cue condition (Fig. 4, right). Additionally it can be seen that RT costs increased gradually for the Side-Dir, Side, and Abst conditions, which were most pronounced for one-hand switches than for both-hand switches, as further confirmed by post hoc tests (significant differences between all cue types for one-hand switches, P 0.05). The average occurrence of contralateral disruptions was significantly influenced by the cueing condition (Table 2, Fig. 5). Disruptions were observed in 30% of the trials in the Abst, Side, and Side-Dir conditions, but only in about 10% of the trials in the Target condition (Cue main effect). The cue condition interacted with the required final pattern, such that only for the Abst, Side, and Side-Dir cues were substantially fewer disruptions observed when subjects switched to IP than to AP. By contrast, for the Target condition, the occurrence of contralateral disruptions was equally infrequent for switches to IP and to AP. Finally, contralateral disruptions occurred more often when the dominant right hand (31 10%) than when the nondominant left hand had to reverse cycling direction (24 9%) (Hand main effect). Experiment 3 In this group, 75% of the individual critical threshold corresponded to a mean movement velocity of 1.38 0.2 Hz (range 1.2 1.8 Hz). In total, 19% of the trials were removed from further analysis because of errors, which were lowest for the Word cues (Word: 10%; horletter: 23%; vertletter: 23%) and for switches with both hands (one hand: 22%; both hands: 16%). There was an RT advantage for Word cues compared with horletter and vertletter cues (see Table 1, Fig. 6). Additionally, the final pattern influenced RT differentially for the three cue types, confirmed by a significant Cue Final Pattern interaction: for Word cues, RT was faster for switches to IP TABLE 1. Summary of the ANOVA results on the RT data of experiments 2 and 3 Experiment 2 Experiment 3 df F P df F P Cue (C) 3, 42 70.5 0.001 2, 22 31.0 0.000 Mode (M) 1, 14 20.6 0.001 1, 11 3.1 0.105 Final pattern (FP) 1, 14 9.5 0.008 1, 11 2.8 0.120 Hand (H) 1, 14 0.1 0.822 1, 11 0.0 0.898 C M 3, 42 5.1 0.004 2, 22 2.3 0.120 C FP 3, 42 5.8 0.002 2, 22 11.0 0.001 C H 3, 42 2.9 0.047 2, 22 9.2 0.001 C M FP 3, 42 3.7 0.019 2, 22 1.6 0.220 C M H 3, 42 1.9 0.140 2, 22 2.1 0.144 C FP H 3, 42 0.7 0.554 2, 22 5.3 0.013 C M FP H 3, 42 0.8 0.516 2, 22 8.6 0.002 than to AP. In contrast, RT was faster for switches to AP than to IP for the horletter cues and differed only slightly for the vertletter cues. This is in marked contrast to all previous findings with other cue types and is most likely related to the (in)congruent substimuli forming the horletter and vertletter cues (see DISCUSSION). HorLetter switches were faster with the left hand than with the right hand (Fig. 7A), whereas the reverse effect of hand was found for the two other cue conditions. Even though a left-hand RT advantage was demonstrated previously for discrete aiming (Carson et al. 1995; Chua et al. 1992), it was highly unexpected for our task because the previous experiments indicated that the dominant right hand typically reacts faster than the nondominant left hand (see experiment 2 and Byblow et al. 2000). This surprising result gave rise to the hypothesis that subjects read and processed the two letters from left to right, which is common practice in the Western world. Interestingly, the left right difference was reversed (i.e., right-hand switches were faster than left-hand switches) for the vertletter condition, where the upper letter indicated the right-hand direction and the lower letter, the left-hand direction. This was further confirmed by significant three- and four-way interactions that can best be understood in light of the highly significant Cue Final Pattern Mode Hand interaction (P 0.005). This interaction was mainly driven by left right differences in RT between the horletter and vertletter cues most pronounced for unimanual switches to IP (horletter, left hand: 713 78 ms; horletter, right hand: 885 196 ms; vertletter, left hand: 812 96 ms; vertletter, right hand: 758 148 ms). The occurrence of contralateral disruptions did not differ between the cue types. However, more contralateral disruptions were found for switches to AP than to IP (significant Final Pattern effect) and also for switches made with the right RT (ms) 1300 1100 900 700 ** * 1300 towards in-phase 1100 towards anti-phase 900 700 Sym Side Side-Dir Target Sym Side Side-Dir Target one hand only both hands FIG. 4. Mean RT for experiment 2 is shown for each cue condition, when only one hand had to switch direction (left) and when both hands had to switch direction (right). Vertical bars indicate the SE. Significant differences between switches toward the in-phase pattern (gray) and the antiphase pattern (black) are symbolized by *P 0.05 and **P 0.01.

INTEGRATED VISUAL CUES PROMOTE CONCEPTUAL BINDING 307 TABLE 2. Summary of the ANOVA results on the percentage of contralateral disruptions in experiments 2 and 3 hand compared with the left hand (significant Hand effect) (Table 2, Fig. 8). DISCUSSION Experiment 1 Experiment 2 Experiment 3 df F P df F P Cue (C) 3, 42 15.2 0.001 1, 11 2.4 0.116 Final pattern (FP) 1, 14 29.9 0.001 1, 11 8.4 0.014 Hand (H) 1, 14 6.9 0.020 1, 11 10.3 0.008 C FP 3, 42 5.7 0.002 1, 11 0.0 0.993 C H 3, 42 1.3 0.291 1, 11 1.8 0.183 C FP H 3, 42 0.2 0.873 1, 11 0.4 0.669 The RT data are consistent with several studies showing that the in-phase pattern is more stable than the antiphase pattern and that voluntary and involuntary transitions occur more rapidly when switching toward the in-phase pattern (Byblow et al. 1999; Scholz and Kelso 1990). In our experiment, both the start and the final pattern were indicated several seconds before the actual switch maneuver was cued. Thus it is unlikely that the observed differences in RT result from stimulus interpretation and response selection processes because subjects could preplan their response early in each trial. Instead, the RT data support the motor outflow hypothesis that differences between symmetric and nonsymmetric movements arise at the motor preparation and/or execution level. By contrast, the central processing hypothesis predicted that switches to IP versus those to AP would not differ when response selection was completed beforehand. Interestingly, RT was particularly long when the nondominant hand had to break away from the preferred IP and had to switch to the less-preferred AP pattern. This result is in excellent agreement with experimental as well as modeling studies, indicating that, in right-handers, the coupling from the right to the left hand is stronger than vice versa (Byblow et al. 1995, 1999, 2000; de Poel et al. 2006, 2007, 2008; Viviani et al. 1998; Walter and Swinnen 1990). Further support for the dominating role of the right hand was revealed by contralateral disruptions that occurred more frequently when the right compared with the left hand reversed cycling direction. Experiment 2 Subjects responded to cues indicating either the hand required to reverse cycling direction or the resulting bimanual pattern to be achieved. Thus in contrast to experiment 1, the RT data reflect not only the motor response, but also cognitive processes involved in response selection. Accordingly, RTs were significantly slower than those in experiment 1 (478 83 ms), even for the fastest Target condition (647 82 ms) (independent t-test: t 5.23, P 0.001). Experiment 2 revealed two main results. First, switches cued by the integrated Target stimuli were clearly faster than switches indicated by the dissociated cues (Abst, Side, Side-Dir). Second, differences in intermanual interference for switches to AP, compared with IP, disappeared with Target cues but not with the other cues. Target cues differed from the other cue types since the final movement direction of both left and right hands was indicated in an integrated manner. Conversely, all other cues provided dissociated information, since only the action of the switching hand was indicated. This effect of integrated versus dissociated information can probably be observed most clearly when comparing the Target and the Side-Dir cues for switches with one hand. Both provide the required cycling direction, i.e., direct information for the switching hand, but only Target cues integrate this information with the cycling direction of the nonswitching hand. With Side-Dir cues, processing costs were generally higher and RT was slower for switches to AP than that to IP. This effect cannot be explained by stimulus congruency or other secondary cue effects because identical stimuli were used to indicate a switch to either pattern. Instead, this result suggests that in the Side-Dir condition the new bimanual pattern was initiated only after the response of the switching hand had been bound to the movement of the nonswitching hand. We propose that the integrated Target cues facilitated this binding process as evidenced by lower RT costs. Similar results have been obtained for more complex tasks, where bimanual movements were most stable when the resulting visual feedback could be conceptualized as one unified pattern or gestalt (Franz et al. 2001; Mechsner et al. 2001). The RT findings for switching with one hand were consistent with measures of intermanual interference (%ContraDis), which was lowest for the Target condition compared with all other cue types. For non-target cues there was a clear advantage for switches to IP compared with switches to AP, consistent with previous studies (Byblow et al. 2000). The results for the Target condition are in excellent agreement with the central processing hypothesis (Diedrichsen et al. 2001; Weigelt et al. 2007). RT was generally faster for switches with both hands than switches with one hand, which differs from discrete tasks where RT is usually faster for unimanual than bimanual re- %ContraDisr 60 40 20 0 * * Abst towards in-phase towards anti-phase Side Side-Dir Target FIG. 5. Occurrence of contralateral disruptions in experiment 2: the mean percentage of contralateral disruptions (%ContraDisr) is shown for each cue condition. Vertical bars indicate the SE. Note that this parameter can be determined only for the one hand switches. Significant differences between switches toward in-phase pattern (gray) and switches toward the antiphase pattern (black) are symbolized by *P 0.05.

308 WENDEROTH ET AL. 1000 towards in-phase towards anti-phase 1000 RT (ms) 750 750 FIG. 6. Mean RT results of experiment 3 are shown using the same conventions as in Fig. 4. horletters vertletters Word horletters vertletters Word sponses (Obhi and Haggard 2004). This difference can best be explained by an advantage in response selection because only for switches with both hands could the same rule be applied, treating the effectors as an integrated entity. Our findings extend previous results in three ways. First, direct and integrated cues abolish the symmetry advantage in a similar fashion for rhythmical as for discrete bimanual tasks (Diedrichsen et al. 2001). Second, the beneficial effect of direct cues generalizes to movement parameters such as cycling direction, extending previous studies demonstrating direct cue effects when movement end positions were indicated (see also General considerations). Third, the Side-Dir, Side, and Abst cues differed along a continuum with respect to the processing steps required to complete the switching task: the Side-Dir condition indicated directly which hand to switch and also the new cycling direction. The Side condition indicated only which hand to switch, although this was cued in a relatively direct way. The Abst condition required subjects to associate the meaning of the random symbols to identify which hand had to be switched. The gradually increasing processing costs between the three conditions are reflected by the RTs that generally increased from the Side-Dir to the Abst condition. In summary, experiments 1 and 2 revealed that RT reflected different processes depending on the task context: when subjects had sufficient time to select the required bimanual response as in experiment 1, RT was faster for switches to IP than those to AP in accordance to the motor outflow hypothesis. By contrast, when subjects had to respond as fast as possible to the appearance of a cue, RT strongly reflected cognitive processing associated with stimulus response mapping in accordance with the central processing hypothesis. RT (ms) 900 700 one hand only horletters vertletters Word left hand right hand FIG. 7. Mean RT results of experiment 3 are shown for switches with one hand only when either the left hand or the right hand had to reverse cycling direction. Vertical bars indicate the SE. both hands This latter result prompted the question: Which specific cue characteristics are most beneficial for bimanual responses selected and executed under time pressure? Our results suggest that cueing bimanual actions as an integrated entity speeds up stimulus response mapping by facilitating a unifying conceptualization for bimanual movements. Experiment 3 investigated whether integrated cues are still advantageous compared with dissociated cues when only symbolic stimuli are used. Experiment 3 There was an RT advantage for the integrated Word cues compared with horletter and verletter cues. For the letter-cue types there was an advantage depending on whether the substimuli were congruent (L-L, R-R) or incongruent (L-R, R-L). Cues displaying congruent substimuli (i.e., L-L, R-R) made switching to AP equally fast as, or even faster than, switching to IP. This is unique for this cue type compared with all others. This type of stimulus congruency has been shown to be beneficial for bimanual RTs, irrespective of whether subjects have to perform symmetric or nonsymmetric bimanual movements (Diedrichsen et al. 2003; Weigelt et al. 2007). Here we argue that the L-L and R-R cues were naturally integrated by the subjects because the same rule could be applied to both hands such that both effectors were controlled as an integrated entity. However, there was still a general RT advantage for Word cues, underscoring the beneficial effect of integrated cues in this bimanual switching task. Comparing the results from the horletter and the vertletters condition, RT was systematically faster for the hand cued first. This finding might be indicative of a serial effect such that subjects processed the dissociated cues either one after the other or that processing the first cue slowed down the processing of the second cue. One has to keep in mind though, that the use of letters might have contributed to this effect because all our subjects were used to process textual cues from left to right and top to bottom. This interaction between left right hand switches and cue types (horletters vs. vertletters) was reflected only in RT data but not in %ContraDisr, where the switching with the nondominant hand always produced fewer disruptions. In our series of experiments, it is the first time that the results for RT and %ContraDisr were differently influenced by the cue type, indicating that the two parameters might reflect distinct underlying processes. Even though RT was generally faster for the Word than the letter cues, the Word cues did not eliminate differences in RT between switches to IP and to AP. This is an

INTEGRATED VISUAL CUES PROMOTE CONCEPTUAL BINDING 309 A %ContraDisr 50 25 0 towards in-phase towards anti-phase B 50 25 0 left hand right hand horletters vertletters Word horletters vertletters Word FIG. 8. Occurrence of contralateral disruptions in experiment 3. A: mean %ContraDisr is shown depending on the cue type as well as the final pattern using the same conventions as in Fig. 5. B: mean %ContraDisr is shown as a function of cue type and which hand had to switch. interesting finding since, like Target cues in the previous experiment, Word cues provided integrated information for both hands. General considerations Experiments 1 and 2 tested whether a bimanual RT task requiring cyclical movements is constrained by motor requirements, as predicted by the motor outflow hypothesis, or rather by processes upstream from motor programming and implementation, as proposed by the central processing hypothesis. Our results indicate that the task context determined whether bimanual RT reflected predominantly neural-motor or cognitive principles: RT for switches to IP was significantly faster than switches to AP when subjects had sufficient time to prepare their response (experiment 1, Fig. 3), supporting the motor outflow hypothesis (Heuer and Klein 2006; Kelso 1995; Swinnen 2002; Swinnen and Wenderoth 2004). This rules out a purely cognitive origin, as hypothesized previously (Diedrichsen et al. 2001; Mechsner et al. 2001). However, when subjects responded to the cues under time pressure, RT depended strongly on the characteristics of the cues. In this condition, i.e., when stimulus response mapping was involved (experiments 2 and 3), direct integrated cues abolished RT differences between switches to IP and to AP (Target condition) (experiment 2, Fig. 4). This indicates that the central processing hypothesis also holds for continuous bimanual movements, which are traditionally considered as being strongly constrained by neural-motor principles. Experiments 2 and 3 revealed another novel result in that integrated cues promoted conceptual binding of bimanual movements such that both hands were controlled as an integrated entity. RT was substantially shorter for integrated than that for dissociated cues and RT costs for dissociated cues resulted most likely from serial effects when the stimulus response mapping had to be determined for the two hands (experiment 3, Fig. 7). However, interlimb interference for switches to AP was abolished only with direct, but not symbolic, integrated cues (experiment 2, Fig. 3; experiment 3, Fig. 6). In the following sections we discuss that both dissociated versus integrated cues and symbolic versus direct cues are processed differently and speculate about the underlying neural correlates. Processing of dissociated versus integrated cues Here we presented experimental evidence showing that dissociated symbolic cues are probably processed serially for the left hand and right hand (experiment 3, Fig. 7), inducing high RT costs. By contrast, RT was substantially shorter when bimanual actions are symbolized by integrated cues, such that one combined response of the hands is selected (experiment 3, Fig. 6). These behavioral results suggest that computational resources might have been insufficient to determine two different stimulus response mappings at the same time. This is in accordance with functional imaging and patient work indicating that stimulus response mapping is determined within a parieto-premotor network that is lateralized to the left hemisphere (Diedrichsen et al. 2006; for review see Rushworth et al. 2003). In the context of bimanual movements, the left-hemispheric specialization might form a bottleneck for assigning different stimulus response mapping to the two hands (Ivry et al. 2004). In accordance with previous work, some features seem to be integrated naturally according to perceptual grouping mechanisms or congruency effects of the substimuli (Weigelt et al. 2007). For example, for the dissociated letter condition, RT was substantially faster for identical letters (LL or RR) than for different letters (LR or RL). This resulted in faster switches to AP (associated with the LL/RR cue) than to IP (associated with the LR/RL cue), indicating that the RT results were rather dominated by the cue characteristics than the final bimanual pattern. The integrated versus dissociated letter cues had no effect on the occurrence of contralateral disruptions, which were always more frequent for switches to AP than to IP (experiment 3, Fig. 8). Also, right-hand switches were more likely to produce contralateral disruptions, irrespective of whether the horletter cues or the vertletter cues were used, unlike RT that differed significantly between the two letter conditions. In other words, RT was highly sensitive to the cue characteristics, but %ContrDisr was not. It is tempting to speculate that these two behavioral parameters reflect different processes during visuomotor transformations, such that RT is influenced by abstract stimulus response selection, whereas %ContraDisr is mainly reflective of processes related to motor implementation. The premotor cortex is a likely neural correlate for controlling this aspect of the task because it is closely related to movement implementation and selective inhibition in relation to the behavioral context (Andersen and Buneo 2002; Cavina-Pratesi et al. 2006; Coxon et al. 2009; Hesse et al. 2006; Isoda and Hikosaka 2007; Majdandzic et al. 2007; Rushworth et al. 2001, 2003; Thoenissen et al. 2002). Processing of symbolic versus direct cues RT was faster with integrated cues than dissociated cues, but intermanual interference for switches to AP was abolished only when the integrated cues were direct (Target condition, exper-

310 WENDEROTH ET AL. iment 2, Fig. 3), rather than symbolic (Word condition, experiment 3, Fig. 6). Intermanual interference persisted even for integrated cues when textual processing was required (experiment 3, even though the Word cues provided the same amount of integrated information as the Target cues) and when the switch was indicated by a neutral stimulus (experiment 1, even though the required pattern was indicated at the beginning of the trial, by the same Target cues as in experiment 2). Thus the direct cue effect emerges only when the controlled movement parameter is directly provided by the cue and available on-line in the visual system. These specific behavioral characteristics support the hypothesis that intermanual interference is abolished because movements to direct cues are guided by the visual dorsal stream controlling both hands in parallel (Diedrichsen et al. 2004; Ivry et al. 2004) and in a nearly automatic manner without the subject s conscious awareness (Fang and He 2005; Pisella et al. 2000; Prablanc and Martin 1992; Schindler et al. 2004). Parallel control might arise from processing visual cues simultaneously in extrastriatal areas of both hemispheres, which represent not only visual information but also associated motor responses (Caminiti et al. 1998; Mirabella et al. 2007; Schenk et al. 2005). This view is in agreement with the %ContraDisr, which was generally higher for switches to AP than to IP, except when Target cues were used and disruptions were equally infrequent for switches to AP and to IP. Even though the exact mechanisms underlying contralateral disruptions are not yet clear, we hypothesize that it reflects de- and recoupling of the two limbs, coming at a cost when the default coupling of homologous muscles needs to be overcome. In the direct (Target) condition, this extra cost was abolished, probably because the left and right limbs were controlled in parallel by each hemisphere, resulting in relative independence. In summary, the present results suggest that the Target cues of experiment 2 were processed in an automatic and nearly parallel manner. This implied that intermanual interference for switches to AP was abolished, as shown by the RT data and percentage of contralateral disruptions. It is tempting to speculate that dissociated and integrated cues are processed along different neural pathways: the former might activate a left hemispheric parieto-premotor network, whereas the latter might rely on bilateral dorsal stream areas speeding up stimulus response mapping (see also Ivry et al. 2004). However, other methods, such as functional imaging or a virtual lesion approach by means of transcranial magnetic stimulation, are needed to test this working hypothesis. Summary and conclusions We have shown that bimanual constraints can emerge at both cognitive and motor levels of the information processing stream, depending on the task. Our data suggest that dissociated cues are processed more serially, resulting in longer RTs. This serial RT cost is substantially reduced when integrated cues are used to indicate the required bimanual action as an entity, supporting the idea of conceptual binding. However, intermanual interference for nonsymmetrical bimanual movements is overcome only when the integrated cues directly indicate the to-be-controlled movement parameter. This indicates that the visual features of the cues influence perception action coupling during bimanual movements. These results are not only theoretically relevant in the context of cognition action interfaces but might also be of practical importance either in promoting fast responses under time pressure when navigating through complex environments or in rehabilitation settings when cues are used to improve performance of neurological patients (Nieuwboer 2008). Our findings suggest that direct, integrated cues might be most beneficial. GRANTS This work was supported by Flanders Fund for Scientific Research Project Grants G.0577.06, G.0593.08, and G.0749.09; the Research Fund of Katholieke Universiteit (K. U.) Leuven, Belgium Contract Grants OT/07/073 and CREA/07/037; Interuniversity Attraction Poles Program Grant P6/29 from the Belgian federal government; and K. U. Leuven Research Fund fellowship F/04/029 to W. Byblow. REFERENCES Albert NB, Weigelt M, Hazeltine E, Ivry RB. Target selection during bimanual reaching to direct cues is unaffected by the perceptual similarity of the targets. J Exp Psychol Hum Percept Perform 33: 1107 1116, 2007. Andersen RA, Buneo CA. Intentional maps in posterior parietal cortex. Annu Rev Neurosci 25: 189 220, 2002. Byblow WD, Chua R, Goodman D. Asymmetries in coupling dynamics of perception and action. J Mot Behav 27: 123 137, 1995. Byblow WD, Lewis GN, Stinear JW, Austin NJ, Lynch M. The subdominant hand increases in the efficacy of voluntary alterations in bimanual coordination. Exp Brain Res 131: 366 374, 2000. Byblow WD, Summers JJ, Semjen A, Wuyts IJ, Carson RG. Spontaneous and intentional pattern switching in a multisegmental bimanual coordination task. Motor Control 3: 372 393, 1999. Caminiti R, Ferraina S, Mayer AB. Visuomotor transformations: early cortical mechanisms of reaching. Curr Opin Neurobiol 8: 753 761, 1998. Carson RG. Neural pathways mediating bilateral interactions between the upper limbs. Brain Res Brain Res Rev 49: 641 662, 2005. Carson RG, Chua R, Goodman D, Byblow WD, Elliott D. The preparation of aiming movements. Brain Cogn 28: 133 154, 1995. Cavina-Pratesi C, Valyear KF, Culham JC, Kohler S, Obhi SS, Marzi CA, Goodale MA. Dissociating arbitrary stimulus-response mapping from movement planning during preparatory period: evidence from event-related functional magnetic resonance imaging. J Neurosci 26: 2704 2713, 2006. Chua R, Carson RG, Goodman D, Elliott D. Asymmetries in the spatial localization of transformed targets. Brain Cogn 20: 227 235, 1992. Coxon JP, Stinear CM, Byblow WD. Stop and go: the neural basis of selective movement prevention. J Cogn Neurosci 21: 1193 1203, 2009. de Poel HJ, Peper CL, Beek PJ. Intentional switches between bimanual coordination patterns are primarily effectuated by the nondominant hand. Motor Control 10: 7 23, 2006. de Poel HJ, Peper CL, Beek PJ. Handedness-related asymmetry in coupling strength in bimanual coordination: furthering theory and evidence. Acta Psychol (Amst) 124: 209 237, 2007. de Poel HJ, Peper CL, Beek PJ. Laterally focused attention modulates asymmetric coupling in rhythmic interlimb coordination. Psychol Res 72: 123 137, 2008. Diedrichsen J, Grafton S, Albert N, Hazeltine E, Ivry RB. Goal-selection and movement-related conflict during bimanual reaching movements. Cereb Cortex 16: 1729 1738, 2006. Diedrichsen J, Hazeltine E, Kennerley SW, Ivry RB. Moving to directly cued locations abolishes spatial interference during bimanual actions. Psych Sci 12: 493 498, 2001. Diedrichsen J, Ivry RB, Hazeltine E, Kennerley S, Cohen A. Bimanual interference associated with the selection of target locations. J Exp Psychol Hum Percept Perform 29: 64 77, 2003. Diedrichsen J, Nambisan R, Kennerley SW, Ivry RB. Independent on-line control of the two hands during bimanual reaching. Eur J Neurosci 19: 1643 1652, 2004. Donchin O, Cardoso De Oliveira S. Electrophysiological approaches to bimanual coordination in primates. In: Neuro-behavioral Determinants of Interlimb Coordination: A Multidisciplinary Approach, edited by Swinnen SP, Duysens J. Boston, MA: Kluwer Academic, 2004, p. 131 153. Eliassen JC, Baynes K, Gazzaniga MS. Anterior and posterior callosal contributions to simultaneous bimanual movements of the hands and fingers. Brain 123: 2501 2511, 2000.