Reaching and Grasping Movements in Parkinson’s Disease: A Review

Distortion of the speed-accuracy tradeoff

As Flowers and others pointed out [7, 8], if PD patients in a reaching task exploit corrective mechanisms in order not to fall short of the target point, this way lengthening the movement times, then hypokinesia can explain bradykinesia, the latter becoming a secondary phenomenon. On the other hand, not all the patients show corrective submovements [19, 25], and bradykinesia has been reported also in non-visually guided movements [26] (see Fig. 2B) and for small target distances [27]. Now, as defined by the current criteria of the Movement Disorders Society (MDS), PD bradykinesia consists of a progressive decrement in speed or amplitude as the movement is continued [28]. In this definition, bradykinesia and hypokinesia seem to be actually combined. This is because PD patients seem unable to perform repetitive or alternating finger or hand movements of normal amplitude at normal speed, thus both signs are generally present on medical examination, although not always simultaneously [28]. Under the hypothesis of reduced muscular scaling [21], the bradykinesia and hypokinesia must be indeed seen as two facets of the same underlying pathophysiological issue. In fact, when assessing upper limb motor performances by means of clinical motor scales, such as UPDRS-III (Unified Parkinson’s Disease Rating Scale, motor section [6]) and the MBRS (Modified Bradykinesia Rating Scale [29]), amplitude impairments appear to be worse and more prevalent than speed or rhythm impairments [30], even though clinical raters generally give greater weight to amplitude than to speed or rhythm [31]. Dopaminergic treatment seems to predominantly improve the bradykinesia [30], while deep brain stimulation (DBS) at the STN improves the speed and amplitude measurements, affects less the rhythm, and does not affect the decrements of speed and amplitude [32]. To note, the appearance and prevalence of one of the two signs considerably depend on the motor task and its requirements [33–35].

In two consecutive studies, Sheridan and collea-gues [36, 37] demonstrated that excessive movement variability is a major determinant of bradykinesia. They suggested that hypokinesia and bradykinesia are not primary deficits, but rather a consequence of adaptation to this inherent variability. In other words, rather than being simply an inability to produce adequate forces, the parkinsonian pathological behavior can result from the fact that patients are not able to properly program these forces and thus control their degree of accuracy at normal pace, so they must trade off speed for accuracy in a dysfunctional manner, going disproportionately slow to compensate for likely endpoint errors. This can be called “compensation hypothesis” [8, 36–39].

More recently, Doan and colleagues [40] showed that accuracy constraints, imposed through the level of water fill of a glass to grasp and drink from, led to the appearance of different control strategies. In the high accuracy constraint condition, healthy subjects increased the peak acceleration and peak deceleration of reaching, as well as the time to peaks, so that movement time remained constant between conditions. Instead, PD patients off medication decreased the peak velocity, peak acceleration and peak deceleration, while increasing the duration of the planning phase, with an increased number of movement units and variability of the reach phase duration. Thus, non-medicated PD patients did not modulate their motor response to keep the overall reach duration consistent, but instead tended to satisfy the accuracy constraint by spending more time and resources in planning to reach precisely, in line with the compensation hypothesis. The authors speculated that the increased duration observed for the premotor phase implies that the high accuracy constraint imposes greater cognitive commitment among the off-medication PD participants, also coherently with attentional interference observed in this population [41]. PD patients on medication did not show significant differences with healthy controls, nor great modification of kinematics and number or duration of movement units with the accuracy level. This indicates that dopamine replacement may not only restore movement timing, but also improve selection and execution of the initial feedforward motor plan, regardless of implicit task constraints [40]. Interestingly, the effects of dopaminergic medication on movement time and on peak velocity are quite persistent, while the effects on early kinematics, i.e., peak acceleration, rapidly deteriorate after drug consumption, perhaps due to a progressive decline in movement programming capabilities [42].

If bradykinesia is an active strategy to compensate for deficiencies in planning, manifested in difficulties in developing precise movement trajectories at high speed, it is reasonable to expect that PD patients show prolonged deceleration phases associated with visual guidance. Flash and colleagues [43] reported asymmetrical velocity profiles with a short initial accelerative phase, followed by a prolonged interval composed of alternating decelerative and accelerative phases, while Castiello and colleagues [44] reported a significant decrease of time spent for decelerating after dopaminergic medication. In the work of Rand et al. [45], when movement accuracy was constrained, the number of zero crossings of the acceleration curve was accentuated and the deceleration phase became longer and more variable. In contrast, Phillips and colleagues [46] reported that patients spent more time in the accelerative phase of the movement even when moving at their preferred speed, revealing inefficiencies in the rate of force production for the required accuracy, rather than increased caution and visual guidance. This leads to more submovements and zero-crossings of the acceleration function when moving at a comfortable pace, and more endpoint errors when moving at a sustained pace. Authors proposed that an altered synchronization of motor signal outflows from BG to supplementary motor area (SMA) may cause problems with maintenance and continuation of behavioral output observed in PD [46].

At this point, it is reasonable to wonder whether bradykinesia is still present when there is little requirement for movement precision. Teasdale and colleagues [33] early analyzed the temporal structure of the movement organization rather than the spatial accuracy, and showed that PD patients can actually vary the duration of their movements to comply with temporal constraints. Without any accuracy requirement, i.e., performing an initial 20° elbow flexion in minimum time without stopping at that precise location, bradykinesia was still evident. However, patients increased their speed to a greater extent (15% for patients against 9% for controls, on average) when passing from a condition in which they should move “as fast as possible” to a condition with an imposed duration. Given also the previous results of Sanes [39] that PD patients performed as quickly and accurately as elderly subjects when the task had a low index of difficulty, Teasdale and colleagues suggested that the task requirements and the experimental context may influence PD patients’ performances to a degree that was underestimated before. On the other hand, even though patients were able to modulate the time to reach the peak force when requested to go faster as well as the healthy subjects, this peak occurred later in time both absolutely and relatively to their own movement duration, as also found later in [46]. This implies that the temporal structure of movements is altered in PD and the slowness cannot be simply accounted for by a deficit in motor programming. Since slower, not faster, movements were characterized by a greater number of EMG bursts, Teasdale and colleagues proposed that, rather than problems in matching the force to the goal distance, PD patients may have lost ability in controlling and sustaining muscle activation due to abnormalities in recruitment of motor units [33]. Corcos et al. [47] reported that some patients off medication had a reduced ability to generate rapid extensor contractions and that measurements of muscle strength and force relaxation correlated with changes in clinical status. At least in advanced stages, PD may therefore determine a difficulty in deploying muscle force rapidly, which may represent the primary cause of the manifested bradykinesia.

Bradykinesia as deficit in motor motivation

In 2007, Mazzoni et al. [26] presented an influential study on reaching movements in PD that opened new perspectives on the discussion about bradykinesia. Patients were asked to make a certain minimum number of reaching movements within a predefined range of speed, without vision of their moving hand. Variable levels of difficulty were obtained by changing the required speed and the target distance. Patients showed movement trajectories and velocity profiles similar to controls. But, despite they were capable of moving at high speeds with preserved endpoint accuracy, patients manifested an increased sensitivity to the experimental condition difficulty, in particular the energetic demand of the task, which led them to select slower movements more often. Thus, bradykinesia seemed to be caused by a distortion in the mechanism of speed selection attributable to an abnormally perceived effort, rather than by an anomaly in the speed–accuracy trade-off. This PD patients’ sensitivity to the energetic cost of the movement was depicted not as a motor weakness but as a motivational deficit. Authors suggested that the level of tonic dopamine from the substantia nigra to the neostriatum, which is abnormally low in PD, might signal an implicit motivation value to the motor centers, this way regulating the likelihood of spending energies to move fast. Deficiency in motor motivation would then determine a shift in the balance between the perceived reward of accomplishing the task and the energetic cost required to do it. This is in line with the models of optimal feedback control that predict movement trajectories on the basis of an optimization of cost functions [48]. Shadmehr and Krakauer [49] proposed that the BG help generate both the expected costs of motor commands, related to the energy expenditure and to the amount of noise in the produced forces (which results in variability and thus greater difficulty in controlling movements), and the expected rewards of internally predicted sensory states. This framework is also consistent with findings that pathologies of the BG affect the conversion of monetary incentives into isometric force [50], movement speed [51] and initiation times [52].

Experimental data on animals, such as mice and primates, have shown that the level of direct involvement of BG in movement execution varies among experiments and species, but BG activity always tracks and predicts kinematic parameters [53]. In particular, animal BG seem to regulate a sort of “movement gain” or “movement vigor”, i.e., the rate of movement or any speed covariate, according to the context or task demands [53–56]. Moisello and colleagues [57] provided behavioral evidence on humans that is the balance between the outputs of direct and indirect pathways in BG that specifically determines a modulation of kinematics of voluntary movements. They showed that the flexibility of kinematics selection is overall reduced in BG diseases, but the pattern of impairment is opposite: when the direct pathway is affected (PD), patients select low-effort movements, or fail to elicit higher muscle force levels as normal subjects do in unpredictable target conditions, whereas when the indirect pathway is affected (Huntington’s disease), patients do not exploit the option of reducing the effort in predictable target conditions, e.g., when the targets are presented in a predefined order.

The works of Mazzoni et al. [26] and Moisello et al. [57] demonstrated in humans that deficits in BG function lead to movements of reduced vigor when the subjects have to move at instructed velocity. This left open the question of whether patients, if free to choose their own appropriate speed, will still move more slowly or will adopt different control strategies. Baraduc and colleagues [27] tried to answer this question with a free choice, variable-amplitude operant task, in which healthy subjects, PD patients on and PD patients off STN-DBS had to move as quickly and precisely as possible a handle towards several targets on a transversal line, without vision of their moving hand. The authors used a modelling approach based on optimal feedback control theory [48] to test if the organization of movements in the PD group was ultimately similar to that of healthy controls. Furthermore, they wanted to test if movements of off-DBS patients and those of on-DBS patients could share the same kinematic properties with just a global change in velocity levels. The endpoint variance, the peak velocity and peak acceleration and their times were all significantly affected by PD, while DBS restored the peak kinematics but not the endpoint variability. For all groups, movement time and endpoint variability increased with movement amplitude, in contrast with a minimum-variance model [58] according to which movement duration is selected as the minimum time that allows to minimize the final endpoint variance required by the task. Instead, data were coherent with a model based on the minimization of neuromuscular costs while satisfying the constraint of reaching the measured endpoint [59]: movement time was reproduced as the duration of minimum-cost movements that respected the experimentally derived dynamic range of motor commands, i.e., range from baseline to the maximum population activity in the motoneurons. This computed motor range could describe both healthy and PD subjects’ trajectories and was mostly constant across movement amplitudes in all participants. The acceleration at 100 ms from the start of the movement was modulated by target distance for controls and patients on DBS, but not for patients off DBS, who showed a plateau. Since the initial acceleration conveys information about the motor planning and is directly proportional to the rate of rise of muscle force, these results demonstrated that the differences in performance between controls and patients were not linked to different motor control strategies but could be accounted for by an abnormally narrow motor range available for PD patients. Therefore, it seems that bradykinesia may be not due to an underestimation of the intensity of necessary impulsive force [21], but to a limitation in the mechanism that produces the motor commands. Baraduc and colleagues suggested that the narrow motor range in PD may be caused either by agonist–antagonist coactivation during movement, due to abnormal supraspinal regulation of spinal networks, or by a lack of central motor activation imposing an internal limit on the expendable neuronal effort. But, given that upper limb bradykinesia cannot fully be accounted for by excessive antagonist activity [24], the central limit appears to be the better candidate to explain all the reported results, and the observed changes in the available motor range among groups and with DBS/levodopa intervention can be interpreted as changes in the level of implicit motor motivation [27].

In the frame of the theory of motor motivation, PD patients would show efficient performances when the energetic cost of the movement was very low, since this condition would augment their poor motor motivation. Gepshtein and colleagues [60] tried to test this hypothesis by using an ideal planner framework as an individual benchmark of performance, that is measuring each participant’s performance against his or her individual precision of movement. They engaged PD patients in a monetarily rewarded task of pointing towards targets on a monitor screen. The targets were placed above and below the starting position in order to vary the energetic cost of the movement as a consequence of gravity. Penalty regions were defined in the vicinities of the rewarding target in the direction of movement or in the orthogonal direction. The ideal planner was defined individually by modelling optimal performance, i.e., modelling the choice of the aim point that maximizes the expected net gain in the face of the estimated individual endpoint variability. When the movements were assisted by gravity, PD patients performed similarly to their optimal planner and also to age-matched controls. When the movements were against gravity, both the healthy and PD participants showed inferior scores with respect to their optimal planner, but patients’ performance was significantly worse than elderly controls. These results confirmed that patients are able to plan pointing movements as effectively as control subjects, but they are more sensitive to the biomechanical cost of the movement. Therefore, the energetic cost of action sets a limit for efficient planning and this effect is mediated by tonic levels of dopamine in the dorsal striatum, which determine a threshold of energetic motor cost beyond which it is not possible to optimize individual motor behavior [60].

Kinesia paradoxa and effect of sensory cues

An interesting aspect of movement execution is that PD patients can exceed their response speed in the context of urgent situations, so much as restoring their performances [61]. This phenomenon is called “kinesia paradoxa” and may be seen in line with the existence of an implicit motor motivation. Indeed, Majsak and colleagues [62] demonstrated that PD patients are able to exceed their self-determined maximal speed of reaching when asked to reach and grasp a ball in motion, which acted as a pressing, externally timed cue. This enhancement of performance with respect to the stationary condition came with preserved accuracy. Therefore, PD patients do not seem to have lost ability in moving fast overall. Their slowness appears to result more from an inability to internally maximize the movement speed, or from a distorted speed selection on a motivational basis [26], rather than a compensation for hazard. Now, the contextual variation of movement speed may be a general property of the motor system, being also present in healthy subjects [63]. This means that the normal speed of reaching may be determined not only by the accuracy constraint of the task but also by the internal motivation, and that modulation of movement time can overcome usual constraints in a situation of urgency [9].

In a successive study of Majsak and colleagues [64], PD patients slowed down and reduced the grasp components (slower velocity of hand opening and closing, smaller maximal aperture, longer time to maximal aperture) in both stationary and moving ball conditions, while the visual motion cues and external temporal constraints had a greater beneficial effect on performances of reaching than on grasping. This suggests that bradykinesia might be a possible compensation for poor prehensile abilities. However, the phenomenon of kinesia paradoxa has been documented also for tasks that do not require fine motor control [61]. Bieńkiewicz and colleagues [65] were able to study the ability of PD patients to catch a moving object through a metal disk attached to the hand, thus removing the prehensile component. Also, they let the temporal dynamics of the imminent task be uncertain and not predefined, thus participants could not fully prepare the subsequent action in advance and use the cueing for timing the initiation, but mostly rely on online movement control. Not only the ability of PD patients to successfully intercept the object was preserved, but also the kinematics of their reaching was adapted to the speed of the moving ball in different cueing situations, demonstrating the intact effect of urgency on the performances of patients.

Paradoxical movements can be related to the general, beneficial effect of sensory stimuli on PD patients’ performances, especially the visual ones [66–68]. Glickstein and Stein [61] noted that the stimuli that often elicit such paradoxical movements are indeed similar to those that relay visual information through the pontine nuclei to the cerebellum for visuomotor control. In a behavioral study, Schenk et al. [69] found a much smaller GPi-DBS effect on the kinematics of externally rather than internally guided movements in PD patients involved in a standard prehension task as well as in a grasp-only task. These results confirmed the limited involvement of BG in the control of externally timed movements. In particular, as clarified below in “Abnormal movement-related brain activity”, this might be due to the fact that movements driven by external cues are primarily controlled by lateral circuits involving the premotor cortex, which are less affected by the PD. This can account for the clinical efficacy of external stimuli in improving movement freezing [70]. In [71], Bieńkiewicz and colleagues used a travelling light display reproducing the real hand motion of a healthy subject as a kinematic guide for the patient. They observed that participants moved significantly faster when the same movement was triggered by the display compared to the self-paced condition, i.e., moving as fast as possible without a guide, and the speed of the display had a significant effect on the movement time. This is in line with the results of Majsak et al. [62] and the others concerning kinesia paradoxa. Bieńkiewicz and colleagues [65] suggested that such improvements in the motor behavior are modulated by changes in the dynamics of the ongoing sensory information: the temporal information of external sensory cues may help bypass the damaged habitual control loops in the BG (caudal putamen) and allow a direct involvement of the preserved goal-directed control ones (STN, GPe) [72], thus enhancing the motor performances. Bek and colleagues [73] recently demonstrated that, in PD patients as well as healthy subjects, movement duration can be modulated to a greater extent by adding to a simple visual cue a moving human hand to imitate. This was explained as a major focus on the dynamics rather than on simply the endpoint, when the kinematic information of the hand was available. Furthermore, imitation of trajectory and visual tracking improved when following a human hand (biological stimulus), compared to following a moving shape (non-biological stimulus) with the same kinematic profile. However, in the work of Castiello et al. [74] significant facilitation effects by human action observation took place for PD patients only when the action matched their actual performing capacities, giving hints as to BG involvement in representation of observed actions.

Hypokinesia and defective proprioception-based control

During non-visually guided arm movements, PD patients tend to make larger spatial errors than healthy controls [8, 25, 43, 75]. This can be related to the impossibility to exploit corrections, even though secondary trajectory adjustments are possible also in absence of visual feedback, by relying on proprioceptive sense [76, 77]. Klockgether and Dichgans [25] reported that even though patients showed consistent undershooting when the vision of the hand was occluded, their movements were both fast and accurate when the target disappeared after the starting signal. Thus, that population of patients could correctly perceive and memorize target position and potentially use this information to successfully program the movement. Indeed, the spatial working memory (or visuospatial mental imagery) seems to be intact in patients performing three-dimensional pointing to remembered target location in the dark [78], even though the degree of hypometria is generally greater in memory-guided movements [79].

If there is not a basic motor planning deficit causing hypokinesia, then it is plausible that the shortfall is in the afferent signal from the periphery and that the extreme reliance on visually-based compensatory mechanisms is the consequence of proprioceptive deficits [75, 80]. Interestingly, PD patients show reduced sensitivity for detecting changes in limb position both on proximal [81] and distal joints [82]. They also need significantly larger limb displacements to detect passive joint motion [83]. These kinesthetic deficiencies often correlate with disease severity and duration [81], but they can appear very early and even precede the motor signs (for a review, see [84]). However, microneurographic recordings of muscle spindles [85] and experiments on illusion of limb displacement by muscle vibration [86] have shown that if proprioception is disturbed in PD, it is not due to peripheral malfunctions. Therefore, an alternative explanation is an anomalous central processing of this afferent signal and hypokinesia may result from an abnormal sensorimotor integration [3, 25, 87–90]. Indeed, PD patients seem to be selectively impaired in conditions that require integration of arm proprioceptive feedback with visual (or spatial memory) information for achieving accurate movements [90]. Their accuracy of reaching seems not to degrade when the starting location of the hand is visible, thus minimizing the so-called “localization bias” [91], while dramatic endpoint errors are likely when starting in the dark [8, 25]. These deficits in the processing of proprioceptive information may occur already at very early stages of PD, while those in the use of visual feedback may develop progressively in later stages [92]. Furthermore, when vision is occluded, a reduced spatio-temporal coupling between the pointing component and the postural adjustments component (control over center-of-mass displacements) emerges in PD [93], revealing an involvement of BG in the integration of proprioceptive information for posture-movement coordination.

Another interesting occurrence is that PD patients usually speak in a low voice, thinking they are speaking normally [94, 95], just as well as they think they are matching the target in goal-directed arm movements when they are actually undershooting [25]. In [87], Moore proposed that during slow, closed-loop arm movements the corollary discharges (efferent copies) that the brain uses to internally estimate the amount of force produced are diminished in PD with respect to the kinesthetic, i.e., proprioceptive, feedback from the periphery. This mismatch would cause an overestimation of the distance covered by the arm and induce the motor centers to dampen the commands towards the musculoskeletal system, resulting in hypokinetic movements. However, in a study by Klockgether et al. [96] PD patients were systematically hypometric with respect to control subjects also in passive movements, not involving internal corollary discharges. This result can be accounted for by an altered processing of kinesthetic information in the BG-related networks [90, 96]. Demirci et al. [88] found that PD patients underestimate more than healthy subjects the motion of their fingers during passive displacement, in contrast with Moore’s model and with the previous findings of overestimated hand movements [8, 25, 96, 97]. Authors suggested that kinesthesia is reduced in PD, which can account for the underestimation, but also the corollary discharges are diminished, so that the error signal is nil and the final motor output is always decreased with respect to the task requirements. This “tuned-down” sensorimotor apparatus explains the discrepancy between the impression of correct motion that patients have and their actual performances, which is reset when exteroceptive information is available [88].

Online modulation of movement trajectory has indeed appeared to be influenced by BG activity [98, 99]. When initial proprioceptive errors are imposed through force perturbations, BG-damaged patients seem not to be able to adjust their ongoing arm movements effectively and smoothly, which suggests an involvement of BG in the online feedback control [99, 100]. However, in experiments of Desmurget and colleagues [101] no substantial deficit was found in PD patients when asked to accurately point towards a target whose location jumped during ocular saccade, i.e., when vision is suppressed (double-step reaching task). In this case, all subjects responded with a smooth adjustment of the trajectory to reach the new location, indicating an effective, subtle modulation of the ongoing motor command. The only important difference between groups was the delay with which patients reacted to the target jump with respect to healthy subjects. Instead, the authors found an abnormal result in the PD group when participants were asked to point towards targets jumping at the individual movement onset, thus requiring them to exploit discrete corrective submovements. PD patients were dramatically late or even unable to employ a secondary adjustment before completing the initial response. Most of them did not show the overlapping of subcomponents typically seen in the velocity profiles of healthy subjects [102, 103]. These results revealed, first, that BG may not be critically involved in non-visual feedback loops involving a smooth modulation of the ongoing motor command (see also [104, 105]). Interestingly, dopamine medication did not consistently normalize performances to the level of controls in several proprioception-based tasks [75, 81, 106–108], suggesting that disruption of the dopaminergic circuits may be not, at least primarily, responsible for all the proprioceptive processing deficits of PD patients. Second, from the work of Desmurget and colleagues [101] it emerges that the BG circuits can instead importantly contribute to the generation of discrete corrective submovements, as also shown in sequential and simultaneous movement tasks [109–113], especially in situations where the movement sequence is not known in advance [114]. Given that the motor plant is non-linear, in order to produce successive movements the brain must estimate the real time position and velocity of the limb by deploying a forward model of the arm dynamics [115]. Therefore, the reported deficits on iterative online control suggest that the BG are involved in forward modelling [101], or they may have a key role in a sort of contextual motor decision making, controlling the timing and the necessity of corrective submovements [116], which is coherent with the concept of BG as a “gate” for action release [117].

It is important to note, however, that gross non-corrective submovements, i.e., submovements not performed to increase accuracy and corresponding to reversal in the trajectory, can emerge in adults in the process of motion termination, i.e., during final antagonist co-activation near the target. Instead, fine non-corrective submovements, corresponding to zero-crossing of the acceleration or the jerk profiles, may be seen as fluctuations in movement velocity emerging from mechanical and neural sources of motion variability [118]. When accuracy constraints are more stringent, these trajectory irregularities are more pronounced. Indeed, the decreasing of speed due to the speed-accuracy tradeoff may come along with a less smooth trajectory, because maintaining low velocities requires a fine regulation of motor unit activations for the steady production of muscle force [119], complicated by aging effects on motor output variability [120]. The presence of such submovements was demonstrated also in the PD population [121], but a specific deficiency was shown in smooth motion termination, also coherently with findings of BG involvement in movement deceleration and stopping [122]. PD patients produced, instead, fine submovements at least as frequently as controls: the majority of their submovements were not corrective but rather due to irregular fluctuations of velocity [121], perhaps exacerbated by the mentioned anomalies in muscular activation [23, 123, 124], especially at low speeds [33].

The fact that conscious perception of target jump interferes with automatic motor control in PD [101] has been recently contested in [125]. Indeed, when the target jump is imposed at the movement onset, the akinesia in limb motion inevitably prolongs the period from target appearance to target displacement. This may disproportionately impact the performances of patients, since their long preparations do not favor quick re-modulation of the trajectory [126]. By using large target displacements during initial saccadic fixation to target, but still consciously perceived by the subjects, Merritt et al. [125] ensured equivalent onsets of target displacement for PD and controls and demonstrated that automatic trajectory modifications were equivalent in both groups and were unaffected by dopaminergic therapy. This questioned the idea that BG actually mediate the online motor control. Interestingly, the PD group did not increase the movement duration in jump trials with respect to stationary trials as much as age-matched healthy controls did. This suggested that aging-related deficiencies in online motor control may be masked by PD-related dysfunctions of dorsal striatum in suppressing automaticity of behavioral responses (i.e., inhibitory control).

The magnitude and variance of perceptual errors about body configuration depend on movement duration: in healthy adults, longer movements result in a stabilization of these errors, because of a more accurate and certain estimate of hand position when the duration increases [127]. Koop et al. [89] found that, instead, PD patients’ perceptual errors peaked at longer times than controls and did not significantly decrease for prolonged movements, while their variance continued to increase with duration. The absence of a stabilization of the perceptual errors informed authors of an abnormal proprioceptive feedback and/or an abnormal internal predictive model. Their proposal depicted hypokinesia as an underestimation of the amplitude of movement needed and may be seen in line with the theory of inadequate scaling of the motor commands to the effort required by the task [21]. Now, when the confounding effect of initial hand localization bias is removed from the error measurement, the movement extent error and variability can still result increased in patients with respect to control subjects, without a concomitant increase in direction error and variability ([25, 128], but see [43]). Desmurget and colleagues [129] demonstrated that behavioral tasks requiring amplitude planning produce increased activation in the BG network. Changes in the BG activity as a function of the movement extent and/or velocity have been observed in other studies [130, 131]. This suggests that BG participate in the planning of movement amplitude (or its covariates, such as effort), by modulating cortical activity that prepares the instruction to move and/or by appropriate peri-movement tuning of the spinal interneuron systems [128, 129], as also suggested by Baraduc et al. [27].

Hypokinesia may be hence a result of a deficit in pre-movement planning, independent of proprioceptive online processing. Behavioral data have indeed suggested that the estimation of the current hand location during reaching movements is mainly based on the efferent signal and little on proprioceptive feedback [104]. On the other hand, as we have seen, the vast majority of studies showed that PD patients have poor accuracy only when they have to rely on proprioceptive information. Mongeon and colleagues [80] investigated whether this inaccuracy depends on an impaired sensory processing in the phase of movement planning or in the phase of online movement control. Subjects were asked to point in the 3D space towards visually memorized targets, with or without vision of the arm, and targets defined proprioceptively by active or passive reference movements. There were no significant differences in the movement duration, magnitude and time-to-peak velocity between groups (controls, PD on, PD off medication state). Final spatial errors were larger in patients than in controls only for proprioceptively-defined targets, especially for medicated patients. While initial spatial distances between hand position at 150 ms and target position were similar, the percentage of change in hand position relative to the target between time of peak velocity and time of movement endpoint was lower for patients, i.e., they compensated for spatial distance relatively less during the deceleration phase. This happened mostly in proprioceptive-based movements. More specifically, PD patients’ hand position at peak velocity was systematically higher on the vertical axis, indicating that they tended to first raise the hand at the elevation of the target in the acceleration phase and then reach out in the target’s direction in the deceleration phase. This strategy could have been employed to simplify the control of the multijoint movement, as a compensation for their proprioceptive deficits. Taken together, these results suggest that the PD patients’ lower endpoint accuracy cannot be primarily accounted for by a different planning strategy as much as by an ineffective movement guidance in conditions of absence of visual feedback. As the authors pointed out, since the results on the sensorimotor integration in PD are somewhat divergent, the contribution of BG in the online control of movements might be context-dependent. Consistent with this view, neurophysiological studies in humans have shown that BG activity can be correlated with the number of corrective submovements only during the execution of a motor task in which a proprioceptive error is induced, but not when visually guided online control is required [132]. Thus, as mentioned above, BG may be selectively involved in submovement generation mechanisms based on proprioceptive information. Coherently, proprioceptive localization errors can be reduced with STN-DBS, even though at the cost of increased endpoint variability [133].

Even in full-vision condition, PD patients’ performance becomes fully accurate only when both the target and the hand are simultaneously visible [134]. If PD patients extremely rely on visual guidance to compensate for impaired proprioceptive integration, then their gaze should be persistently aligned with their hand in pointing towards objects. Lukos and colleagues [135] observed that during the first phase of a full vision reach-to-grasp the eye-hand positions correlation was higher in PD patients and the minimum horizontal and vertical distances separating the gaze position and hand position on the screen were significantly smaller compared to controls. Patients also made numerous short saccades between the thumb and the index finger during the movement, demonstrating that they almost tracked their hands with their gaze. Even when the vision of the hand was blocked, patients continued to look at the screen in a similar fashion. All these results suggest an abnormal eye-hand coordination, i.e., patients over-rely on visual feedback of their hands instead of employing feed-forward anticipatory mechanisms. Prediction of the movement outcome for anticipatory control may be deterred by the greater variability in the initial force impulse [136].

Abnormal movement-related brain activity

To which extent can we relate the aforementioned behavioral dysfunctions to pathological neural activity? A disproportionate inhibition signal from the BG to the thalamus may cause a failure to reach cortico-thalamic activation thresholds to initiate the prepared motor program, which may largely contribute to the slowness observed in movement initiation and execution of PD patients [145]. Indeed, while the transmission of the motor commands from the cortex to downstream motor centers appears to be intact in PD [154], the pre-movement cortical excitability increases more slowly than normal [155]. This slowness in reaching motor excitability thresholds has been demonstrated in EEG studies measuring the so-called Bereitschaftspotential (or “readiness” potential) that rises over a widespread area of the scalp just before the onset of voluntary movements, thus reflecting preparatory activity. Libet et al. [156] proposed that the early component of this potential, which starts about 1 or 2 s before the EMG-detected movement onset, reflects the preparation to act soon, while the late component, with an onset of about 500 ms before movement, is associated with an endogenous urge or an intention to act. While this latter component is significantly higher for the self-initiated than externally triggered (e.g., via visual or auditory cues) movements, the former one does not differ between the two types of movement [157]. Dick et al. [2] showed that in PD patients off medication this premotor potential has a normal peak amplitude, but it is reduced 650 ms before the peak, that is 1-2 seconds before movement, especially in midline electrodes. As a net effect, the rise of the readiness potential results steeper in the PD population, even though attentional strategies may help increase early-stage activity [153]. In successive studies [153, 157, 158], with respect to normal subjects, PD patients showed a general under-activation of the medial cortical motor circuits (e.g., linking BG to SMA) for self-initiated hand movements, and an extra-activation of the dorsolateral areas (e.g., premotor area and dorsolateral prefrontal cortex) when external cues were given. The inadequate BG activation of the SMA accounts for the reduction of the early EEG component, which can be related to impairment in the preparation of proper movement commands [156]. Interestingly, in healthy subjects the readiness potential is much larger when performing a choice rather than a simple RT task [159], which can be modelled considering an extra-activation of the SMA to perform additional processing necessary for the choice, a mechanism probably lacking in PD patients [160]. Indeed, Filipovic et al. [161] found that PD patients with smaller difference in response latency between choice RT and simple RT tasks have smaller amplitudes of the readiness potential, presumably caused by reduced movement pre-programming capacity. Furthermore, the medial circuit has been associated with the ability to use predictive models to internally program action sequences with or without cues [158, 162], an ability impaired in PD [109, 158]. Instead, the patients’ overactivation in the later component of the potential, nearer to the movement onset, reflects compensatory activity of the lateral premotor and parietal areas [2]. Hyperactivation of premotor and cerebellar areas for defective BG was confirmed by successive fMRI studies [163–167].

The parkinsonian slow build-up of the motor excitability reflects a delay from the preparation of the motor plan to its expression in the motor cortical areas, which delays movement initiation. Coherently, event-related desynchronization patterns in the alpha and beta bands preceding voluntary movements are abnormal in PD. In particular, the appearance of the uncoupling of population activity, necessary for movement initiation, is significantly delayed, at least for the contralateral hemisphere, revealing a programming deficit that accounts for the akinetic and bradykinetic behaviors [168–171]. Also, longer and higher beta bursts have been recently reported [172]. Torrecillos et al. [173] measured the local field potentials (LFP) from the STN of PD patients under medication and executing visually-cued reaching task with a joystick. At single trial level, within-subject correlations between pre-movement beta bursts and their movement performance revealed that, according to their timing and amplitude, these bursts can reduce the speed of the forthcoming movement and prolong the RT. In particular, the trials in which a beta burst in the contralateral STN occurred roughly 650-500 ms before the movement onset were associated with a reduced peak velocity and this effect was related only to the burst amplitude and not to the mean beta power in each trial. Moreover, the presence of the burst in a 200 ms window after the “go” cue was associated with increased RT. In a recent LFP recording study, Tinkhauser and colleagues [174] added that slowest movements appear to be those that are preceded by multiple bursts of beta activity occurring at short time intervals within a trial and spatially overlapping between the two STNs.

These electrophysiological studies shed new light on the involvement of BG nuclei in reaching and grasping. BG seem to release the cortex from idling rhythms in beta frequencies and let necessary motor channels be favored and become coherent at higher “functional” frequencies to achieve a specific movement [175, 176]. Indeed, Brown et al. [177] revealed that populations of neurons in the subthalamic nucleus and internal pallidum are coupled at 20–30 Hz in the off-medication state, whereas in the on-medication state they oscillate at 60–70 Hz. Power attenuation of the alpha and beta rhythms can indeed be restored with dopaminergic medication, an effect that correlates with improvements in bradykinesia [178]. But dopamine deficiency in PD apparently results not only in excessive oscillations but also increased inter-regional coherence at beta band frequencies [179], suggesting a role of striatal dopamine in preserving subcortical structures from excessive cortical entrapment and cross hemispheric coupling that would prevent fine tuning of movements [180].

The oscillatory synchronization in the STN is then modulated by context and task demands [181]. In particular, this nucleus seems to play a selective role in response inhibition. Kühn and colleagues [170] recorded the LFP of the STN in PD patients involved in RT tasks in which an imperative cue signaled whether to move or not to move. In “no-go” trials the pre-movement desynchronization was reduced with respect to the “go” trials. In the late post-movement phase, they found an increase in beta power, i.e., post-movement synchronization, that was higher and a few hundred milliseconds shorter in latency for the “no-go” trials than the “go” trials. Thus, the STN is involved in the preparation also of externally-paced movements, other than self-paced movements, and the degree of synchronization in its beta band activity is associated with the suppression or facilitation of movement programming and initiation. Coherently, DBS of STN can partially restore the motor strategy preparation and the inhibitory control in PD patients [182]. Furthermore, other LFP recordings in PD patients revealed that ballistic fast movements of the elbow joint cause synchronized STN activity in the alpha range [183], and that also changes in alpha band oscillatory patterns can be associated with reactive control [184]. Alterations in the oscillatory activity of the BG in PD also affect the gamma and high frequency bands (for a review, see [185]). Bradykinesia of the upper limb can be indeed predicted from the beta and gamma activities of the primary motor and premotor cortices, and the STN [32]. In the paper by Joundi et al. [186] a synchronization in the gamma range in PD patients’ STN developed during the reaching movement and was especially pronounced in response to cues signaling an upcoming fast reach than to cues indicating slow or normal reaches. Instead, only the timing of beta-band desynchronization depended on task demands, and not its extent. This suggests functionally segregated roles for beta and gamma frequencies in motor preparation and execution.

Future perspectives

The simultaneous presence of issues that directly stem from cortico-BG malfunctions and issues indirectly stemming from attempts to compensate these malfunctions makes for a complex picture of reach and grasp in PD. This led to the development of progressively more elegant experimental designs trying to disentangle the different components. Future behavioral studies should address the still open questions, such as those concerning the origins of the distortion in the speed selection. This particular issue might be tackled by struggling to reduce between- and within-group differences related to the management of accuracy constraints, for instance defining individual indices of task difficulty along the dimensions of effort and accuracy on the basis of each subject’s performance levels, in the wake of the work by Gepshtein and colleagues [60]. Other challenges will be trying to separate more clearly planning and preparation from execution problems, and to elucidate the role of the BG in forward modelling and in online feedback control. This will also help resolve the controversies around the relative proportion of deceleration and acceleration in the movement profiles of patients for the same task. In the attempt to compare results with previous literature, the disease stage, inter-individual variability as well as any confound of attentional interferences and potential cognitive impairments should be taken into account.

A key element for future investigations in the field will be to complement observational studies with perturbation studies involving cortical stimulations. Recording studies have already allowed researchers to associate behavioral dysfunctions with anomalies of brain activity. For instance, EEG analysis has been helpful in identifying predictors of bradykinetic signs (and their improvement), as activity variations mainly in the delta and gamma bands have been associated with distinct parameters of movement execution. In addition to this kind of studies, correlations may be established by direct (invasive or not) intervention on the interested circuits, as stimulation experiments have shown. Stimulation studies might have an even more relevant role in the future, thanks to the possibility of exerting safe and reversible perturbations while recording neural activity. This recently allowed Muthuraman and colleagues [32] to relate DBS-induced changes in speed, amplitude and rhythm of patients’ hand movements to the evolution of motor, premotor and STN activities individually, by analyzing the subnetworks’ preferred frequency of oscillation with the EEG. Furthermore, the recent technique of adaptive DBS [230], i.e., stimulation automatically adjusted and tailored to the patient’s clinical state, may provide also further insights into the movement pathophysiology. The stimulation can be triggered by non-neural, movement biomarkers, for example computed by processing signals coming from wearable sensors monitoring the upper limb joints’ configuration. By comparing on- and off-stimulation states, more reliable associations between the local circuitry and affected spatio-temporal parameters would be obtained. Importantly, these biomarkers evolve with the course of the disease and can powerfully discriminate among different motor phenotypes of PD, facilitating the tailoring of the stimulation protocol to the specific patient.

Once pathological behaviors have been mapped into some neural substrates, novel hypotheses on dysfunctional mechanisms can be tested through computational modelling studies. In face of the obvious limitations in biological realism, virtual simulations provide extreme flexibility, since pathological states can be simulated at different levels of the system, locally or globally, and different assumptions can be first translated into both circuital and functional model features and then verified in the emerging behavior of the simulated network. On one side, purely functional models will be useful for efficiently embedding a BG unit in a brain-inspired architecture and simulating motor control and learning mechanisms in an artificial agent [243, 244], as well as reproducing overall characteristics of the disease in silico. On the other side, neural networks modelling [245, 246] can capture micro-structural features which sound essential for studying the dynamical properties of the altered circuits. Spiking neural networks, in particular, can prove whether kinematic anomalies arise from alterations in the dynamics and activity patterns of specific BG nuclei or pathways, eventually localizing them. Perturbations can be modelled as well: in this context, the work of Neumann and colleagues [220] is a meaningful example of how the integration of behavioral test, stimulation experiment and computational validation can bring about new useful insights both for neuroscientific research and for DBS treatment. We envisage that this integration may be the right key to solve the open debates on the reaching and grasping anomalies of parkinsonian patients, leading to new knowledge and more complete understanding of the disease and of the human motor control in general. Moreover, it might acquire clinical relevance in the future since computational modelling will enable to find a set of individual motor parameters that are fundamental for outlining a motor control profile of the PD patient. Together with the refinement of motor assessment techniques necessary for clinical investigations of more complex aspects of movements, this effort will finally provide quantitative tools to establish predictors of disease trajectories as well as reliably monitor the effects of both invasive and non-invasive therapies.