Spatial mapping of posture-dependent resistance to passive displacement of the hypertonic arm post-stroke

Introduction

Muscles in the post-stroke arm commonly demonstrate abnormal reflexes that result in increased position- and velocity-dependent resistance to movement (i.e. spastic hypertonus: [21, 25]. Spastic hypertonia is understood to reflect systematic reductions in stretch reflex thresholds [16, 24], decreased range of regulation of these stretch reflex thresholds [23, 47], as well as altered non-reflex phenomena such as abnormalities in the intrinsic mechanical properties of spastic muscles and altered viscoelastic properties of passive tissues [28, 43, 49]. Importantly, systematic reduction in stretch reflex threshold could lead to significant increase in stretch reflex excitability [17] and agonist/antagonist coactivation in some regions of the workspace [22, 30], which could lead to complex, posture-dependent and potentially time-varying joint impedances in the hemiparetic arm. Because the feedforward control of goal-directed movements relies on accurate predictions of limb impedance [38], spatial and temporal complexity of joint impedance may be a significant contributor to impairment of movement coordination post-stroke [20, 40, 48]. Although a velocity-dependent increase in muscle tone (spasticity is frequently assessed clinically and has been quantified in single joints such as the elbow and wrist [15, 23, 26, 44], quantitative assessments of multi-joint control deficits have been rare (see [30, 36]) and sometimes rely on techniques such as kinematic and/or inverse dynamics analyses, which can be insensitive to the presence and effects of abnormal muscle coactivations (e.g., [18, 33]).

The goal of this study was to develop a reliable approach for measuring the mechanical consequences of abnormal neuromuscular mechanisms as a function of hand location in the reachable workspace in the hemiparetic arm post-stroke. We are motivated in this work by a growing body of experimental evidence supporting the idea that the neural mechanisms contributing to the control of limb posture and movement can be differentially compromised by stroke [19, 27, 40], see also [8, 37, 48] and by other neuromotor disorders (cf. [10, 11]), and by the belief that greater understanding of the mechanisms contributing to sensorimotor deficits will eventually lead to improved efficacy of therapeutic interventions [13]. Our work builds on prior studies that examined transient mechanical and electromyographic responses to passive displacements of the wrist or elbow to quantitatively assess post-stroke spasticity (cf. [15, 23, 26, 44, 47]) and spastic dystonia [46]. In one example, Schmit and colleagues used a motorized device to passively flex and extend the elbow of hemiparetic stroke survivors over a range of speeds ranging from slow (6º/s) to fast (90º/s) [44]. Their goal was to assess the reliability of three different biomechanical correlates of spasticity, which they isolated from other aspects of spastic hypertonia associated with dystonia, contracture, and increased joint stiffness. They did so by subtracting the torque response to the slowest displacement from responses to faster displacements, leaving only reflex torque. This approach is effective because the stiffness of the passive tissues about the elbow joint are largely velocity insensitive [9]. Of the three biomechanical measures considered—peak torque, peak joint stiffness, and onset angle of reflex torque responses—peak torque values measured during displacements at 90°/s were most reliable on repeated measures in a single testing session (> 80% reliability), and most highly correlated with clinical assessments of spasticity (Ashworth Scale). In another example, Mirbhageri and colleagues [29] used system identification techniques to quantify the contributions of reflex and intrinsic (i.e., non-reflex) stiffness to total elbow stiffness at several different elbow angles in the paretic and nonparetic arm of chronic hemiparetic stroke survivors. Each position was examined under passive conditions in the range of full elbow flexion to full elbow extension. They reported that intrinsic and reflex stiffness both contributed strongly to net joint torque, that the effects were significantly larger in the paretic than in the non-paretic elbow muscles, and that these differences increased with the increasing joint angle indicating position dependence. Although these prior studies suggest that manifestations of spasticity and hypertonia may vary in complex ways across the reachable workspace after stroke, a more comprehensive approach to quantifying mechanical expressions of spasticity and hypertonia across the workspace has yet to be described.

We designed a set of experiments using a two-joint, planar robot to measure the dynamic and quasistatic mechanical and electromyographic responses to controlled displacements of the upper extremity at several locations in the arm’s workspace. Subjects were instructed to relax as the robot moved their hand sequentially between target locations spanning the reachable workspace at speeds ranging from very slow to fast. Trajectories were selected such that movements were largely limited to either the shoulder or elbow joint, but not both. The robot stabilized the hand at the target for at least 20 s following the end of each movement. We analyzed the time series of horizontal planar hand forces and electromyographic (EMG) activations for selected arm muscles to determine the duration of phasic, velocity-dependent resistance to stretch, to characterize the spatial topography of tonic, position-dependent hand forces throughout the workspace, and to characterize the muscle activations that give rise to these postural bias forces. The resulting data demonstrate that the robotic assessment of posture-dependent bias forces was repeatable across days, that the phasic component of these stroke-related forces lasted no more than 2 s after the end of limb re-positioning, and that these bias forces were partly neuromuscular in origin (not merely due to passive tissue resistance to stretch) such that elevated “resting” EMG activations exhibited posture-dependence in some muscles, but posture-invariance in others. We expect that quantitative evaluation of posture-dependent bias forces may facilitate future assessments of stroke's impact on the interaction between the control of upper extremity posture and movement.

Methods

A convenience sample of ten unilateral, hemiparetic survivors of stroke (HS; aged 43–62 years) and nine neurologically intact control subjects (NI; age-range matched: 43–61 years) gave informed consent to participate in this study (see Appendix 1). All procedures were approved by Marquette University’s Office of Research Compliance in accord with the Declaration of Helsinki. All HS were recruited from the pool of hemiparetic stroke outpatients of Medical College of Wisconsin and the Milwaukee VA Medical Center and all were in the chronic stage of recovery (between 2- and 28-years post-stroke). Exclusion criteria for HS included: inability to give informed consent, inability to follow 2-step directions, history of tendon transfer in the affected limb, physical dimensions prohibiting appropriate interaction with the robotic assessment system (Fig. 1) despite reasonable adjustment attempts (e.g. abdominal intrusion into the robot’s workspace or an inability to see the feedback display after chair height adjustment), use of aminoglycoside antibiotics, curare-like agents, or other agents that might interfere with neuromuscular transmission, botulinum toxin treatment within the previous 8 months, and/or shoulder pain in the test position of 75 to 90° abduction. These exclusion criteria were designed to ensure participant safety and to eliminate confounding factors that could compromise the measurement and interpretation of abnormal hand forces arising from abnormal neuromuscular responses to imposed hand displacements. The presence of contracture or shoulder subluxation did not exclude subjects from participating unless it limited their ability to perform the experiments comfortably. NI control subjects were required to have no history of neurological disorder. All subjects were able to achieve all test positions without discomfort. Three of the NI subjects participated in three experimental sessions with at least one week separating each session. The remaining six NI subjects participated in a single session. Seven HS participated in two experimental sessions separated in time by at least one week. The remaining HS participated in three experimental sessions typically separated by at least one week. Each session lasted ~ 2 h.

Clinical assessments—Prior to experimentation, HS participated in a series of clinical assessments. The upper extremity portion of the Fugl-Meyer assessment of motor performance (FM_UE) was used to assess impairment in the ability of each HS to move the more affected arm [7]. Scoring for the FM_UE is on a 0–66 point scale, which is based on the subject’s ability to move both within and outside of muscle synergies. Increasing values reflect greater motor control. The Modified Ashworth Scale (MAS) was used to assess hypertonia in the flexors and extensors at the shoulder, elbow, and wrist. Scoring for the MAS is on a 0–4-point scale with 0 indicating no increase in tone and a 4 indicating that the more-affected limb is rigid in flexion or extension. Within each limb, MAS scores were averaged across the three tested joints to obtain an overall estimate of upper extremity spasticity (range: 0–4; see also [48]). Passive range of motion of the elbow and shoulder in the workspace plane was measured using a goniometer both with and without the robotic test apparatus. Handedness was assessed for all subjects using the Edinburgh Handedness Inventory [31]. HS were asked to use pre-stroke preferences to guide their answers. Appendix 1 provides a description of subject characteristics for all subjects. Within the group of survivors, FM_UE scores ranged from 19 to 53, reflecting a broad spectrum of functional motor impairment within our subject pool.

Experimental procedures—Subjects were seated in an adjustable high-backed chair fixed in front of an actuated, 2-joint robotic manipulator designed to move in a horizontal plane (see [39]). Subjects were strapped into the chair using a seatbelt-style chest harness that minimized trunk motion. Efforts were taken to situate subjects in the same location and orientation relative to the robot from one testing day to the next. An opaque screen mounted 1 cm above the plane of hand motion occluded vision of the arm, hand, and robot arm. The subject’s wrist was splinted at 0° flexion and fixed at the wrist to the robot’s handle with Velcro straps. The arm was supported against gravity (between 75 and 90° abduction angle) using a lightweight, chair-mounted support. Arm segment lengths and passive range of motion at the shoulder and elbow were used to determine each individual’s reachable workspace in the horizontal plane. A minimum of 18 spatial sample locations (targets) spanned the full range of shoulder and elbow flexion and extension (Fig. 1B and C) except in subjects with reachable workspaces exceeding the robot’s workspace. In those cases, the sample locations spanned the intersection of the subject's range of motion and that of the planar robot. For each subject, detailed data analyses were performed at three targets defined as center locations (red dots labeled C1–C3 in Fig. 1B and C) and at six targets defined as boundary locations (blue dots labeled B1–B6 in Fig. 1B and C). The remaining points were used to construct spatial maps of posture-dependent interface forces and muscle activations as described below. Subjects were verbally coached throughout the experiment to “relax” at all times and not to intervene as the robot generated passive movements of the arm through its workspace. To reinforce our verbal coaching, the word RELAX was projected onto the occluding screen throughout the experimental session.

The robot monitored hand position using rotational encoders (A25SB17P180C06E1CN; Gurley Instruments Inc., Troy, NY) mounted on the motor shafts (M-605-A Goldline; Kollmorgen, Inc. Northampton, MA). The location of the handle could be resolved within 0.038 mm throughout the operating workspace. A 16-bit data acquisition board (PCI-6031E DAQ; National Instruments Inc., Austin, TX) sampled analog force data from a load cell (85M35A-I40-A-200N12; JR3 Inc., Woodland, CA) that was mounted immediately under the handle. The robot was programmed to generate stiff control of hand position; data collection and robot control were performed using the Matlab XPC toolbox (Mathworks, Inc., Natick, MA) at a rate of 1000 samples per second (see also [14]).

The robot enforced transitions between targets along paths designed so that motion occurred predominantly at one joint (the focal joint: either the shoulder or elbow) while the non-focal joint remained essentially still. Figure 1B and C show elbow and shoulder trajectories (dashed and solid purple arcs, respectively) along which only the focal joint was flexed or extended. Desired hand paths were computed in real-time using a previously published approach [34] based on anthropometric measurements of upper and forearm limb segment lengths and localization of the participant’s shoulder relative to the origin of the robot’s workspace. Hand trajectories had bell-shaped speed profiles computed such that rotation of the focal joint had peak angular velocities that were slow (6°/s), medium (30°/s), or fast (90°/s). In contrast to step- or ramp-shaped velocity profiles, bell-shaped profiles achieve smoother transitions between one location and another, imitating the smooth changes in arm position applied during manual testing of muscle resistance by clinicians [24]. After the end of transition (EOT), the robot held the hand at the target location for at least 20 s (the holding period). Targets C1–C3 were visited from all four directions (elbow extension, elbow flexion, shoulder extension, and shoulder flexion) in pseudorandom order. Targets at a workspace boundary (B1-B6) were visited from at least two directions in pseudorandom order. In total, 105 transitions were used to visit all target from all desired directions at the three different speeds (also pseudorandomized across trials).

Data Collection and Analysis—Our experimental setup allowed us to synchronously record instantaneous hand position and force as well as surface electromyograms (EMG) from: shoulder horizontal adductors pectoralis major (PECS) and anterior deltoid (ADL), shoulder horizontal abductor posterior deltoid (PDL), elbow flexors biceps (short and long heads: BICS and BICL) and brachioradialis (BRD), as well as elbow extensors triceps (lateral and long heads: TRILT and TRILG). Raw EMGs were amplified × 1000 (Myosystem 1200, Noraxon, Inc. Scottsdale, AZ) and band-pass filtered between 10 and 500 Hz in hardware prior to digitization. 60 Hz and 120 Hz artifacts were eliminated in post-processing using zero-phase, 4th-order Butterworth notch filters. Residual offsets were removed from the digitized EMGs prior to rectification and low-pass filtering at 4 Hz with a zero-phase, 4th-order Butterworth filter. All data (kinematics, forces, and EMGs) were streamed to disk during the experiments for subsequent off-line processing.

To facilitate comparison of EMGs across the study population, each subject’s muscle activations were normalized by the peak value of the rectified and filtered activation recorded from that muscle during a series of 5 s duration maximum voluntary isometric contractions (MVICs). MVICs were performed with the hand stabilized in the center of the arm’s horizontal planar workspace using a rigid handle fixed to the robot frame. Each subject performed 12 MVIC trials before the experiment: 3 each of maximal isometric elbow flexion, elbow extension, shoulder horizontal abduction and adduction (cf. [38, 45]). Peak EMG activation for each muscle was defined as the largest EMG value for that muscle within any MVIC trial after signal processing as described in the preceding paragraph. We also recorded 5 s of quiet resting activation in the same limb configuration prior to MVIC recording after approximately 5 min of rest, wherein subjects were encouraged to “relax”.

A primary goal of our analysis was to differentiate velocity-dependent resistance to stretch post-stroke from position-dependent, steady-state effects. To this end, we filtered the kinematic and kinetic time series off-line using a zero lag, 4th order, 10 Hz Butterworth low-pass filter and analyzed the hand’s force vector at nine points in time including the moment of peak hand speed (V_MAX), the moment of target acquisition (End of Transition or EOT, defined as the time when hand velocity first dropped below 15% of its peak value), as well as at 1, 2, 3, 5, 10, 15, and 20 s after EOT. We also analyzed the magnitude of rectified and filtered EMG signals at the same time points.

Statistical testing—We performed two main analyses. The first analyzed interface forces at the robot's handle to quantify velocity-dependent resistance to passive motion of the arm. We controlled for posture-dependencies to determine the earliest point in time after EOT beyond which hand force ceases to demonstrate velocity-dependence. This was done to better isolate posture-dependent effects in the second analysis described below. As we will show, hand forces measured at the end of the holding period displayed no systematic dependence on movement speed, and so we regarded hand forces measured at EOT + 20 s as representative of asymptotic values. We subtracted these values from hand forces measured at other time points in the same trial before performing a six-way, repeated measures, general linear model ANOVA examining how changes in measured hand forces (relative to EOT + 20 s) varied by subject group (HS or NI), movement speed (6°/s, 30°/s, 90°/s), sampling instant (time of peak transit velocity, EOT, and 1, 2, 3, 5, and 15 s after EOT), movement direction (elbow flexion, elbow extension, shoulder flexion, shoulder extension), workspace location (center vs. boundary, wherein the center locations were collapsed across C1–C3 and the boundary locations were collapsed across B1-B6) and testing day (1–3). Post-hoc ANOVA and Tukey t-tests were performed to examine significant main and interaction effects.

The second analysis characterized the spatial topography of tonic, position-dependent hand forces as a function of workspace location at the earliest time point wherein hand forces neither depended on transport velocity nor differed from the asymptotic values. To do so, we plotted raw hand force vectors as a function of workspace location for each subject. These plots were then co-registered with contemporaneous maps of normalized EMG activations to determine whether the observed hand forces might be partly neuromuscular in origin. Specifically, we evaluated the extent to which shoulder and elbow muscle activations varied as a function of joint angle along the solid (PECS, ADL, and PDL) and dashed lines (BICS, BICL, BRD, TRILT, and TRILG) in Fig. 1B and C.

Data processing and statistical testing were carried out within the Matlab (Mathworks, Inc., Natick, MA) and Minitab (Minitab Inc., State College, PA) computing environments. Effects were considered statistically significant at the α = 0.05 level. Specifically, we applied a Bonferroni correction to the 20 statistical tests performed on the hand force data, which yielded an individual test significance threshold of p = 0.0025. Despite consistency in the mechanical recordings, the EMG data demonstrated considerable variation across stroke survivors. Consequently, we did not correct the EMG analyses for multiple comparisons (i.e. we accepted statistical significance at p = 0.05).

Results

All subjects were alert throughout each experimental session even though they were not asked to engage in any task-dependent activation aside from relaxing. Robot-generated hand movements were always smooth, having unimodal, bell-shaped velocity profiles (Fig. 2A). Passive displacement of the hand modulated EMG activation in HS subjects (especially in stretched muscles) but rarely did so in NI control subjects. For example, stretching shoulder or elbow muscles in the hemiparetic arm caused EMG activation in some muscles to increase during movement (Fig. 2A, left), as might be expected due to spastic hypertonia (i.e. [41]. Interestingly, elevated muscle activation could remain active long after motion had ceased in some regions of the workspace, as might be expected due to a decrease in the stretch reflex threshold post-stroke [23], although other explanations are possible. In some muscles, tonic EMG activation after passive displacement of the hand appeared to depend on the final limb configuration (limb posture). Persistence of muscle activation was never observed in NI control subjects during or after passive movement (cf. Figure 2A, right).

Elevated EMG activations in HS subjects were not an artifact due to poor signal transduction or improper EMG normalization in these subjects. Because EMG normalization with respect to MVIC could overestimate voluntary muscular effort if the signals were corrupted by environmental noise, we visually verified that EMGs recorded from each muscle were of high quality by plotting the EMG signal power spectrograms from MVIC trials for each muscle and compared it to spectrograms obtained during quiet rest. In all cases, MVIC EMG signal power was distributed across the frequency range in a unimodal pattern characteristic of high-quality surface EMG recordings [35] (Fig. 2B, black). By contrast, EMG signal power was lower during rest in the center of the workspace (Fig. 2B, blue). MVIC EMG exceeded quiet resting signal power by a factor greater than 100 in all cases (≫ 20 dB signal to noise ratio). Thus, normalization with respect to MVIC yielded a high-quality assessment of relative voluntary effort. Note that while resting EMG power was uniformly negligible at all frequencies in NI subjects, it commonly displayed measurable, tonic activation post-stroke with a spectral signature of quality surface EMGs rather than broadband environmental noise (e.g. Figure 2B, left; purple).

We sought to characterize how mechanical resistance to passive joint motion varied with movement velocity (i.e., spastic hypertonia; [21]. Visualization of raw hand forces from a selected HS subject (Fig. 3A) demonstrated that hand force magnitude could vary systematically by transport speed during movement, that variations in steady-state hand force were not systematic during the hold phase that followed, and that hand forces changed very little from EOT + 20 s onward. Despite efforts to minimize trunk motion, some trial-by-trial variability in hand force undoubtedly arose from subjects shifting in their seat during the ~ 1.5-h testing period. Because the effect of these infrequent postural shifts would be to alter the bias forces recorded at the handle throughout the trial, we reasoned that greater sensitivity in subsequent analyses of velocity-dependence would be achieved by removing the effects of postural shifts (i.e. by analyzing hand forces relative to their asymptotic values). Indeed, exploratory three-way repeated measures ANOVA found that hand force magnitude at EOT + 20 s post-stroke did not vary systematically with transport speed (F_(3,207) = 1.40, p = 0.25). We therefore aligned the raw hand force profiles in time with respect to EOT and subtracted asymptotic values on a trial-by-trial basis (Fig. 3B) prior to evaluating how subject group, workspace location and other factors influence the transient effect of movement on reaction forces at the hand.

Discussion and conclusions

We used a planar robot to assess the spatial and temporal topography of position-dependent hand forces and electromyographic activations that arise during and after passive movement of the arm after stroke. Survivors of hemiparetic stroke (HS) and neurologically intact (NI) control subjects were instructed to relax as the robot repositioned the hand at several different testing locations within a horizontal planar slice through the hemiparetic arm's passive range of motion. The robot was programmed such that movements to each testing location could occur from at least two different directions and at three different speeds ranging from very slow (6°/s) to fast (90°/s). The robot held the hand at the testing location for a minimum of 20 s after each transition. We recorded hand force and electromyographic activations in selected muscles spanning the shoulder and elbow joints throughout the transition and during the holding period that followed. All of the HS and a subset of the NI subjects returned to the lab on subsequent days to repeat the testing procedures. We found that hand interaction forces were relatively small at all speeds and sample times in NI subjects, whereas they varied systematically by transport speed during and shortly after movement in HS subjects. Observation of velocity-dependence after movement had ceased indicates that hand forces were not solely due to passive tissue viscoelasticity post-stroke, but rather implicated the presence of abnormal stretch reflexes. In the HS group, spastic responses (i.e. velocity-dependent resistance to stretch) diminished within 2 s after the end of transition (EOT). Hand forces changed very little from EOT + 2 s to the end of the sampling period (EOT + 20 s), exhibiting dependence on limb posture but no systematic dependence on movement speed and/or movement direction. Although each HS subject displayed a unique field of hand forces across the arm’s reachable workspace after movement ceased, the magnitude of steady-state hand force was greater near the outer boundaries of the workspace than in the center of the workspace for the HS group but not the NI group. In the HS group, hand forces were stronger on one side of the reachable workspace and pointed towards an equilibrium posture located in the approximate center of the workspace. These results were repeatable across testing days. We consider the steady-state hand forces to reflect abnormal postural biases because they depend on history of stroke and limb configuration but not on the speed nor direction of passive movements bringing the hand to the different spatial sampling locations.

Stroke-related postural bias forces were partly neuromuscular in origin because they were accompanied by persistently elevated EMG activation levels in most muscles spanning the shoulder and elbow joints (Fig. 7). During and shortly after movement, passive translation of the hemiparetic arm led to increased EMG activations during movement consistent with spastic hypertonia (e.g., [41]. In some muscles, elevated muscle activation could remain active long after motion had ceased (cf., Fig. 2A, left), as might be expected due to a decrease in the stretch reflex threshold post-stroke [23, 24]. These neuromuscular responses to passive displacement of the hand were persistent in the sense that the complex spatial patterns of EMG activations observed at EOT + 2 s were also observed at EOT + 20 s. The specific details as to which muscles exhibited increased activations varied from person to person, although some general trends were apparent. Across subjects in the HS group, muscles that exhibited posture dependent activations at EOT + 2 s also did so at EOT + 20 s (BRD, BICL). Muscles that exhibited elevated activation throughout the workspace at EOT + 2 s also did so at EOT + 20 s (TRILG, PECS). Muscles that did not exhibit elevated activations at EOT + 2 s also failed to do so at EOT + 20 s (BICS, TRILT, ADL, PDL). Persistence of upper extremity muscle activity (sustained spontaneous discharge) has also been observed after voluntary isometric contractions in the hemiparetic arm after stroke [3]. By contrast, persistence of muscle activation was never observed in NI control subjects during or after passive movement (Figs. 6B and 7).

Author contributions

Conceived of study: CG and RS; referred participants: JM; collected data: PKM, MB, TS; analyzed data: PKM, MB, RS; interpreted data: PKM, MB, TS, JM, CG, RS; drafted manuscript: PKM, MB, CG, RS; revised manuscript: PKM, MB, TS, JM, CG, RS; All authors read and approved the final manuscript.

Funding

This work was supported in part by Grants from the National Institutes of Health NICHD R01HD053727 and R15HD093086.

Availability of data and materials

The datasets generated and/or analyzed during the current study are not publicly available due to potential HIPPA concerns but are available from the corresponding author on reasonable request.

Declarations