Exp Brain Res (2005) 164: 120–132 DOI 10.1007/s00221-005-2216-y

R ES E AR C H A RT I C L E

Isaac Kurtzer Æ Paul A. DiZio Æ James R. Lackner

Adaptation to a novel multi-force environment

Received: 26 April 2004 / Accepted: 15 December 2004 / Published online: 16 April 2005
Springer-Verlag 2005

Abstract Humans display accurate limb behavior when they move in familiar environments composed of many simultaneously-acting forces. Little is known about how multi-force environments are represented and whether this process partitions between the underlying force components, reflects the net forces present, or is cued to the force-context. We tested between these three main alternatives by examining how reaching movements adapt to a novel multi-force field composed of a velocity-dependent force and a constant force. These hypotheses were dissociated first by making the constant force larger and oppositely-oriented to the velocity-dependent force; thereby, the net force was always opposite the velocity-dependent component. Second, we tested adaptation with all novel forces removed to eliminate any potential cues for the force-context. In two experiments that used forces perpendicular or parallel to the forward movement direction, we found adaptation aftereffects consistent with a mechanism that partitioned the velocity-dependent component from the net force field. Specifically, we found aftereffects opposite the rightward or resistive velocity-dependent component of the multi-force field, even though the net force imposed was leftward or assistive, respectively. An additional experiment suggested that the velocity-dependent component is partitioned relative to the background load in a limb-based coordinate frame.

I. Kurtzer Æ P. A. DiZio Æ J. R. Lackner Ashton Graybiel Spatial Orientation Laboratory, Volen Center for Complex Systems, Brandeis University, 415 South St. Waltham, MA 02454, USA I. Kurtzer (&) Department of Anatomy and Cell Biology, Queen’s University, Kingston, ON, Canada, K7L 3N6 E-mail: [email protected] Tel.: +613-533-2600 Fax: +613-533-6880

Keywords Motor learning Æ Force partitioning Æ Reaching Æ Modularity

Introduction Accurate motor control is supported by neural mechanisms that adaptively anticipate the force requirements of the task. These anticipatory mechanisms are evident during the introduction, repetition, and removal of a novel movement-dependent force as a canonical pattern of movement disruption, return of accurate performance, and compensatory aftereffects (Lackner and DiZio 1992; Lackner and DiZio 1994; Shadmehr and Mussa-Ivaldi 1994). Such force adaptation paradigms have been intensively utilized over the past decade in examining adaptation of reaching movements across movement direction (Gandolfo et al 1996; Sainburg et al 1999; Thoroughman and Shadmehr 2000), movement speed (Goodbody and Wolpert 1998), limb configuration (Shadmehr and Mussa-Ivaldi 1994; Shadmehr and Moussavi 2000; Malfait et al 2002), in relation to visual and proprioceptive feedback (Lackner and DiZio 1994; Cohn et al 2000; DiZio and Lackner 2000), and transfer between limbs (DiZio and Lackner 1995; Wang and Sainburg 2004). These studies focused on adaptation to single force fields and demonstrated that adaptive force representations are encoded within a limb-based coordinate system dominated by proprioceptive inputs and displaying limited generalization over the tested dimensions. It remains unknown how force adaptation occurs within environments composed of multiple simultaneously acting forces, as is typical in natural settings that include the inertial dynamics of the limb, the background force of gravity, and forces associated with wielded objects and surrounding media. We considered three basic hypotheses for how multiforce environments could be adaptively represented. First, the nervous system could adaptively partition the net force into its underlying components, for example,

121

between the ‘‘static’’/gravity-related and ‘‘dynamic’’/ movement-related components. Second, the nervous system could adapt to the net force without regard to its underlying components. Lastly, the nervous system could utilize a context-dependent force representation that is engaged by some salient cue. The force partitioning hypothesis is the most likely candidate, as several studies suggest that ‘‘static’’/ gravity-related and ‘‘dynamic’’/movement-related forces are separately represented by the motor system. First, modeling studies demonstrate that such a partitioned organization could simplify the control of reaching over a range of movement speeds and background loads (Hollerbach and Flash 1982; Atkenson and Hollerbach 1985). Second, empirical studies report that unconstrained reaching trajectories tend to minimize the peak dynamic forces (Nishikawa et al 1999; Soechting et al 1995) and possess joint-torque and EMG patterns consistent with separate dynamic- and gravity-dependent neural drives (Flanders and Herrmann 1992; Gottlieb et al 1997). Lastly, altered-gravity studies do not report largely deranged movement patterns upon the rapid transitions to hypo- or hypergravity in parabolic flight. Instead, the accuracy of reaching (Fisk et al 1993), object lifting (Kingma et al 1999), and movements of the torso (Vernazza-Martin et al 2000) are near to those on Earth, suggesting that dynamic forces are planned independently of the forces that counteract gravity. However, these strongly suggestive studies are not definitive. Foremost, there is no clear understanding of how the background force conditions prior to movement onset are integrated into the patterning of force commands during movement. Second, studies of ‘‘normal performance’’ are arguably studies of highly overtrained behavior over a lifetime of interacting constraints. Third, behaviors observed under altered-gravity conditions include vestibular representations of the gravity force as well as motor and somatosensory sources (Lackner and DiZio 2000), and given its universal presence in terrestrial evolution, gravity might Fig. 1a–c Diagrams of force fields. a Velocity-dependent. b Constant. c Multi. The multi-force field is the combination of the velocity-dependent and constant force fields. X-axis: forward hand velocity ( mm/s); for clarity, only the forward component is indicated. Y-axis: imposed force (N). Experiment 1: +/
is leftwards/rightwards force; Experiment 2: +/
is assistive/resistive force. Gray shading for multi-force field indicates hand velocities above 600 mm/s threshold, where net force reverses sign

be associated with specialized adaptive mechanisms. Lastly, several authors have reported context-dependent force adaptation (Gandolfo et al 1996; Blakemore et al 1998; Wada et al 2003), although others have shown contextual ‘‘cueing’’ to be ineffective (Karniel and Mussa-Ivaldi 2002). Here we were able to predict categorically different aftereffects for the three alternative hypotheses by (1) programming a robot manipulandum to impose both a velocity-dependent force and a larger, oppositely-oriented constant force, and (2) by testing adaptation with both the velocity-dependent and constant forces removed (Figs. 1 and 2). We chose a velocity-dependent force because it has been widely used for inducing adaptation and a co-planar constant force since it is the simplest additional force and is reminiscent of the force of gravity. Importantly, the programmed constant force was larger and opposite to the velocity-dependent force throughout the movement, so the net force imposed was always opposite the underlying velocity-dependent component. Consider a forward movement in which we simultaneously impose one force that acts rightward proportional to the hand’s forward velocity and a larger force that acts leftward independently of hand motion. During a forward movement the summation of the two forces (the net force) would begin at a maximal leftward value at movement onset, decrease proportional to the hand’s forward velocity, and return to the same maximal leftward value at the movement’s completion. After removal of all the forces, each hypothesis predicts a categorically different aftereffect. A net force hypothesis predicts a rightward aftereffect opposite the net leftward force. In contrast, the force partitioning hypothesis predicts a leftward aftereffect opposite the rightward velocity-dependent component. Lastly, the context-dependent hypothesis predicts no aftereffect upon removing both forces, as any force-cue has also been removed. (As a first step, we will only focus on the potential adaptation to the underlying velocity-dependent component, since subjects are already known to adapt to velocity-dependent forces with a null force background.) This logic was utilized in two principal experiments with particular attention paid to the middle portion of the movement, where cumulative feedback effects and voluntary intervention is expected to be minimal (Shapiro et al 2002, 2004). The first experiment examined adaptation to a multi-force field applied lateral to the hand’s forward movement as described above.

122

Material and methods Subjects Twenty-four subjects (Experiment 1: n=8; Experiment 2: n=8; Experiment 3: n=8) from Brandeis University participated in three separate experiments. Subjects included both males (n=13) and females (n=11) ranging in age from 18 to 33 years. All were right-handed, neurologically normal, fluent English speakers, and naive to the purpose of the experiment. Subjects participated in one 90-min session and were compensated for their time. Before beginning, each gave their informed consent to the procedure approved by the Brandeis University IRB. Apparatus

Fig. 2 Outline of experiments. All experiments followed the same design with three force fields. Note the constant condition was always followed first by the multi or velocity-only condition. The baseline was comprised of the mean of all reaches in the no-force blocks, surrounded by a green rectangle. The initial effect for a force perturbation was comprised of the first reach (Experiments 1 and 3) or the mean of first two reaches (Experiment 2) occurring on the transition from a no-force block to a force field block, three black arrows for three force conditions. The average behavior during the last black with the force field (Final) is indicated by an oval, three ovals for three force conditions. The post aftereffect of the force field was comprised of the first reach (Experiments 1 and 3) or mean of first two reaches (Experiment 2) occurring on the transition from a force field block to a no-force block, three red arrows for three force conditions

The second experiment examined adaptation to a similar multi-force field applied parallel to the hand’s forward movement. A velocity-dependent resistive force was paired with a constant assistive force so that the net force was always assistive (decreasing proportional to the forward hand velocity). The adaptation patterns predicted by the different hypotheses were tested by examining the aftereffects when both forces were removed: net force hypothesis—a speed undershoot reflecting compensation of the assistive net force; force partitioning hypothesis—a speed overshoot reflecting compensation of the resistive force component; forcecontext hypothesis—no aftereffect as the context-cue is absent. In both experiments, subjects unambiguously exhibited trajectory aftereffects linked to the velocity-dependent component of force; rather than the net force, or the force-context alternatives. An additional experiment involving a velocity-dependent and position-dependent force suggests that the velocity-dependent component is partitioned relative to the background load within a limb-based coordinate frame.

Hand motion was recorded at the fingertip by a WATSMART or OPTOTRACK 3020 (Northern Digital, Waterloo, Ontario) motion detection system sampling at 200Hz while subjects reached with a PHANToM device (Sensable Devices, Cambridge, MA). This robotic device is lightweight, mobile in three dimensions, and was connected to a custom-molded cuff that encased the metacarpal region of the hand without obstructing the fingers. In Experiments 1 and 2, the PHANToM was programmed to deliver a constant force, a velocity-dependent force, or a multi-force field to the hand during forward reaching. Force is expressed in Newtons (N), hand velocity in m/s ð_xÞ; and viscosity (V) in Ns/m. During Experiment 1, the applied forces primarily acted lateral to the hand’s forward motion (Fig. 1a–c). The velocity-dependent force field acted rightward relative to the hand motion. Thereby, forward/backward movements induced right/left forces, while right/left movements induced backward/forward forces. We focused on the primary forward hand motion and the associated rightward force; note that the hand trajectories showed some changes in their forward motion (forward peak velocity and movement time) due to the secondary backward/forward forces, but these were quite variable and weak (p>0.05). The constant force was always a constant leftward force, so its magnitude and direction were independent of the hand motion. The multi-force field was the combination of these two forces. Multi - force Velocity - dependent force Constant force V x_ þf0;6gT N V x_ ; where V ¼f0;10;
10;0g f0;6gT N During Experiment 2 the applied forces acted parallel to the hand’s forward motion (Fig. 1a–c). The velocitydependent force field resisted the forward hand motion. The constant force assisted forward hand motion with a magnitude and direction independent of the hand

123

within the movement time. Subjects were informed that throughout the experiment they would encounter three force conditions whose presence or absence would be indicated after the first trial. This information was acMulti - force Velocity - dependent force Constant force curate and presented as such. Lastly, when the force V x_ þf6;0gT N V x_ ; where V ¼ f0;0;0;
10g f6;0gT N fields were removed, subjects were informed of the transition, re-stabilized their limb, and were instructed In Experiment 3, the robot applied a velocity-de- to ‘‘reach naturally, as if you had no prior experience pendent rightward force identical to that in Experi- with the force field’’. All subjects indicated that they ment 1 and a leftward position-dependent force that understood the instructions. generated an approximately constant torque at the elbow and shoulder; the multi-force field was the combination of these two forces. An approximately Experimental design constant joint torque was created by decreasing the lateral force as the limb’s forward distance, and hence All experiments followed the same design as outlined in moment arm, increased: Torque = Force · Moment Fig. 2. Each subject participated in a single session Arm. The position-dependent leftward force began where he or she reached in blocks of ten separated by a at a maximum value of 6 N and decreased with short rest period (1 min); a total of 210 reaches were the forward hand position at a rate depending on the made during the experiment. Four force conditions were initial distance between the hand and shoulder. The utilized: no-force, constant-only, velocity-only, and initial hand–shoulder forward distance was determined multi-force. All subjects performed 30 consecutive on a subject-by-subject basis (24–38 cm) such that the reaches in the no-force condition followed by 40 conlateral force at the hand decreased to 50–60% of its secutive reaches in the constant-only force condition. initial value at the final forward position of hand: x¢ is The velocity-only and multi-force conditions were prethe ratio of the initial hand–shoulder forward distance sented either third or fourth and balanced across subover the current hand–shoulder forward distance. motion. The multi-force field was the combination of these two forces.

Multi - load Velocity - dependent force Torque - Conserving force V x_ þP x’ V x_ ; where V ¼f0;10;
10;0g P x’,where P ¼f0;6g Procedure In all experiments, subjects reached forward to a single square target 5 mm across, impressed on a tabletop and along a line roughly in the parasagittal plane of their right shoulder. The target placement required a mediumsized movement ranging from 22 to 24 cm across subjects. Reaches were completed under full visual feedback in a well-lit room and always began with the entire arm not contacting the table so that subjects would have to actively stabilize the manipulandum against any forces applied before reaching. Our instructions were designed to allow unhurried and naturalistic reaching movements. Subjects were instructed to reach in a ‘‘single continuous movement’’ and ‘‘if you feel you are making a mistake do not stop, slow down, or ‘stiffen up’; rather continue towards the target as best you can’’. Reaches were selfinitiated and required to remain within a window of movement times, 800±100 ms. Therefore, the movements were moderately slow, but well within the range used to study motor adaptation (Lackner and DiZio 1994; Goodbody and Wolpert 1998). This feature allowed the peak forward hand velocities to remain under 600 mm/s so that the net force would always be in direction of the constant force component in Experiments 1 and 2. Subjects were given no strict criterion on their endpoint accuracy; they were simply told to land on target

jects. Two blocks of no-force trials separated all force field conditions. Analysis Position signals were low pass-filtered offline at 10Hz and again after each differentiation. Movement duration was measured using a 5% peak velocity criterion for the start and end. We examined both middle and terminal measures of the movements although we were primarily interested in the middle portion of the trajectory. Therefore, all hypotheses are evaluated with regard to the middle parts of the trajectory aftereffects. During Experiments 1 and 3 with the orthogonal perturbations, we examined the hand’s lateral position (perpendicular displacement from a line connecting the start and target) at the peak forward velocity and at the endpoint. In Experiment 2 with the parallel perturbations, we examined the hand’s peak forward velocity and the endpoint along the fore-aft axis. In all experiments, the critical information was the change from baseline during different learning periods of the constant, velocity-only, and multiforce conditions. Unless otherwise specified, reported measurements of the different learning periods are mean and standard error deviations from the baselines. Four learning periods were examined for each force condition and for each measure: baseline, initial, final, and post (Fig. 2). The baseline period included no-force

124

blocks before the perturbations and sufficiently after the perturbations to ensure complete re-adaptation to normal conditions, in other words 9–16 reaches beyond the first aftereffect (Lackner and DiZio 1994; Thoroughman and Shadmehr 2000). This period was considered the unbiased level of performance. The initial period included the first reach for Experiments 1 and 3 and the mean of the first two reaches for Experiment 2 upon introducing the force perturbation. We utilized the mean of the first two reaches for Experiment 2 because on-axis components of trajectories show considerable more variability than off-axis measures (Gordon et al 1994; Messier and Kalaska 1999). The final period was measured as the average final block performance to the force perturbation; this was taken as the most complete learning during the brief exposure with the perturbation. Lastly, the post period was comprised of the first reach for Experiments 1 and 3 and the mean of the first two reaches for Experiment 2 after removal of the force perturbation. The post period was taken to reflect the force representation underlying adaptation. Repeated measures ANOVAs examined the stabilities of the blocks that compose the baseline period (6 blocks · time) and the effect of each force field across the learning periods (4 periods · 1 force field). These were conducted separately for the middle and terminal measures. t-Tests compared baseline vs initial, initial vs final, and baseline vs post values to determine whether the initial reaches were perturbed from baseline, whether any adaptation occurred from the initial to final reaches with the force present, and whether any adaptive afterFig. 3a–i Summary of Experiment 1. a–c Rightward velocity-dependent force condition. d–f Leftward constant force condition. g–i Multi-force condition. Each row has a cartoon of the subject and robot-imposed force during an idealized bell-shaped velocity profile on the left; the middle panels show the initial (black) and post (red) hand trajectories of individual subjects, green trajectory is the group baseline; the right panels show the mean change in lateral position (mm) from baseline measured at the maximum forward velocity for the initial, final, and post periods; standard errors for experimental and baseline periods are indicated by error bars and width of green lines, respectively. Comparisons are baseline vs initial, initial vs final, and baseline vs post: * p