You are on page 1of 10

Available online at www.sciencedirect.

com

ScienceDirect

Feedback control during voluntary motor actions


Stephen H Scott1,2,3, Tyler Cluff1, Catherine R Lowrey1 and
Tomohiko Takei1

Humans possess an impressive ability to generate goal- processing in a highly distributed network of cortical
oriented motor actions to move and interact with the and subcortical circuits.
environment. The planning and initiation of these body
movements is supported by highly distributed cortical and There have been several recent reviews to discuss the
subcortical circuits. Recent studies, inspired by advanced principles of feedback control and their application to
control theory, highlight similar sophistication when we make biological control [1,2], with several in-depth reviews
online corrections to counter small disturbances of the limb or focused on the importance of somatosensory feedback
altered visual feedback. Such goal-directed feedback is likely [3–7], visual feedback [8,9], or the integration of vision
generated by the same neural circuits associated with motor and somatosensory feedback during voluntary actions
planning and initiation. These common neural substrates afford [10]. Related topics of interest are the importance of
a highly responsive system to maintain goal-directed control sensory feedback in motor decisions [11] and the use
and rapidly select new motor actions as required to deftly move of dynamical systems theory to interpret the cortical
and interact in a complex world. control of movement [12].
Addresses
1
Centre for Neuroscience Studies, Queen’s University, Kingston, ON, Task-dependent corrective responses during
Canada
2
voluntary motor actions
Department of Biomedical and Molecular Sciences, Queen’s
Somatosensory feedback: upper limb
University, Kingston, ON, Canada
3
Department of Medicine, Queen’s University, Kingston, ON, Canada A common paradigm to explore how somatosensory
feedback influences voluntary control is to observe
Corresponding author: Scott, Stephen H (steve.scott@queensu.ca) how subjects respond to small mechanical perturbations
[5–7]. Recent studies have shown that corrective
responses are faster when the urgency to respond is
Current Opinion in Neurobiology 2015, 33:85–94 increased by manipulating the return time, or size and
This review comes from a themed issue on Motor circuits and action shape of the spatial goal [13,14]. Rapid corrective
Edited by Ole Kiehn and Mark Churchland
responses occur even for very small disturbances that
approach the natural variability of limb motion [15],
For a complete overview see the Issue and the Editorial
illustrating that feedback processes are omnipresent in
Available online 28th March 2015 voluntary motor actions. Sophisticated corrective
http://dx.doi.org/10.1016/j.conb.2015.03.006 responses are also observed when a muscle is unloaded
0959-4388/# 2015 Published by Elsevier Ltd. by a mechanical perturbation [16].

A striking example of the sophistication of human correc-


tive responses is the ability to quickly respond to a me-
chanical perturbation and re-route a movement to avoid
obstacles in the environment [17] (Figure 1a). The selec-
Introduction tion to move to the right or left side of the obstacle
Over the last two decades, most research on voluntary appeared to depend on the hand’s position at the time
motor control has focused on the planning and initiation the perturbation was applied. If the hand position was
of motor actions, with little emphasis on afferent feed- deviated slightly towards the left/right, the subject was
back from the moving limb. The ideas of optimal feed- more likely to go leftward/rightward around the obstacle.
back control and other advanced control theories have The rapid selection of movements to avoid an obstacle
refocused attention on the importance of sensory feed- suggests that multiple potential strategies can be prepared
back in biological control. The result is a renaissance in and launched as required during motor actions.
exploring the complex ways in which somatosensory and
visual feedback can influence ongoing motor actions. Detailed examination of electromyographic (EMG) ac-
Here we review studies demonstrating that small correc- tivity can provide an important clue as to the exact time
tive responses during voluntary motor actions possess the motor system can generate task-dependent feedback
most, if not all, of the complexities associated with motor responses. The fastest of these responses occurs from
planning and movement initiation. Further, we discuss 25 to 50 ms (short latency response, SLR). The SLR
how these sophisticated corrective responses reflect engages spinal circuits, is relatively small and considers

www.sciencedirect.com Current Opinion in Neurobiology 2015, 33:85–94


86 Motor circuits and action

Figure 1

(a) Obstacle Avoidance (b) Lateral Triceps


R1 R2 R3 EV
3 * * * *

ΔEMG [a.u.]
2

1
Load
On 0
Between obstacles
2 cm 0 100 200
Around obstacles
Time [ms]

(c) One Target Multiple Targets (d) Lateral Triceps


R1 R2 R3 EV
4 * *

ΔEMG [a.u.]
3

1
Load Load
On On 0

Correct to initial target 0 100 200


2 cm Correct to new target
Time [ms]
Current Opinion in Neurobiology

Rapid online decisions following mechanical perturbations. (a) Subjects made reaching movements from a start position (bottom) to an end target
(top) while avoiding obstacles (filled circles) in their workspace. In some trials, a perturbation deviated their hand directly towards an obstacle.
Subjects either corrected back between the obstacles (blue) or deviated around the obstacles (red). (b) Perturbation-related muscle activity
(DEMG relative to unperturbed trials) in which subjects moved between (blue) or around (red) the obstacles. Data are aligned to perturbation onset
(solid vertical line). R1 (20–45 ms), R2 (45–75 ms), and R3 (75–105 ms) denote standard time epochs of the muscle stretch response, and EV
indicates the early voluntary component (105–135 ms). Note that differences in muscle activity are evident in the R2 epoch (45–75 ms after
perturbation onset). (c) Hand trajectories from a representative subject when there is one goal (left) or multiple potential goals (right). (d)
Perturbation-evoked response of lateral triceps (mean  SEM). Differences in muscle activity were not evident until the R3 epoch (75–105 ms after
perturbation onset). *P < 0.05.
Adapted from Nashed et al. [17].

how much the muscle is lengthened as well as back- decisional processes [21]. In this study, subjects viewed a
ground muscle activity (i.e., ‘gain scaling’ [18–20]). In dynamic random dot motion display and were asked to
contrast, task-dependent responses are observed during indicate the direction of stimulus motion by moving to a
the long-latency time period (LLR; 50–105 ms post-per- left or right target. Leftward and rightward force pulses
turbation [1,10]). In the Nashed et al. study [17], the were randomly interleaved to probe the ongoing deci-
choice to redirect to the left or right around the obstacle sional processes. LLRs scaled with the strength of the
was reflected in EMG activity at 60 ms (Figure 1b; R2 stimulus (i.e., coherence) and duration of viewing, sug-
epoch). Variations of this task demonstrated that when gesting that LLRs track an evolving decision variable
subjects encountered a perturbation while reaching to based on accumulated sensory evidence. Collectively,
one of multiple goal targets, they sometimes switched to these findings highlight a striking parallel linking the
reaching towards a secondary goal target (Figure 1c). The computations involved in classic decision-making tasks
selection of this alternate goal target occurred a little later with online motor control [22].
at 75 ms after perturbation onset (Figure 1d; R3 epoch).
This 15 ms shift in timing suggests different brain path- Somatosensory feedback: lower limb
ways or processes may be engaged in launching a response In standing subjects, perturbations to the support surface
to attain a target versus selecting a new goal target during evoke short, medium and long latency responses in lower
an ongoing action (how to attain the selected goal versus limb muscles. Short and medium-latency bursts in lower
what goal to select). limb EMG are thought to involve spinal and brainstem
pathways, whereas LLRs are linked to cortical processing
Indeed, recent evidence indicates that the neural pro- (see [23] for review). Recent work highlights that sensory
cesses that generate LLRs are tightly coupled to ongoing feedback drives the continuous modulation of leg muscle

Current Opinion in Neurobiology 2015, 33:85–94 www.sciencedirect.com


Feedback control during voluntary motor actions Scott et al. 87

activity during dynamic balance control tasks [24,25]. perturbation) and phase of the gait cycle [38]. However, it
During trains of platform accelerations applied in multi- is difficult to determine the latency of these responses
ple directions, LLRs better reflected the control of a as EMG activity was averaged for 200 ms following
‘global’ task variable (center of body mass) than control the earliest onset of perturbation-evoked muscle activity,
of ‘local’ variables such as individual joint motion patterns which likely incorporates voluntary contributions.
[24]. Further work is necessary to establish how quickly
task-dependency from upper limb tasks is incorporated
Although corrective responses typically vary with proper- into corrective responses in the lower limbs.
ties of the task, there may be some functional differences
between upper and lower limb LLRs. For example, when Visual feedback
subjects controlled a virtual inverted pendulum with their The importance of visual feedback in voluntary control
foot, a reduction in the stability of the virtual load was has received a lot of attention in motor neuroscience.
associated with reductions in both the short [26] and Visuomotor corrections have been addressed with the
medium-latency response [27]. The authors argued that ‘double-step’ paradigm, in which the target unexpectedly
feedback gains may be reduced in favour of a muscle co- changes location before or during the course of movement
contraction strategy, which may increase joint stiffness (‘target jump’ [39–42]), or by displacing the hand-aligned
and the amplitude of the SLR to minimize displacement cursor (‘cursor jump’ [43]) during reaching (see [8,9]).
caused by the perturbation [26,27,28]. In contrast, com-
pliant (i.e., unstable) loads applied to the upper limb are Recently, Dimitriou and colleagues [44] used visual
associated with increased LLRs [29], suggesting different shifts of the target or hand-cursor to test whether the
control strategies may be engaged by certain upper and motor system is able to update the control policy during
lower limb motor tasks. However, it is difficult to inter- reaching. By perturbing hand-cursor feedback either si-
pret these discrepancies in feedback modulation because multaneously with or 100 ms after the target jump, the
subjects were instructed not to respond to the perturba- authors showed that feedback responses took 100 ms to
tion in both the upper and lower limb tasks (i.e., ‘do not consider the new target position (see also [45]). Visual
intervene’). Further work may be required to delineate feedback of target information is influenced by visual
the task-relevancy of this feedback modulation. attention, such that environmental distractors reduce the
amplitude and increase the latency of corrective
Somatosensory feedback: interlimb responses following a sudden shift in target position. In
Many behavioural tasks require coordination between contrast, there appears to be a more dedicated link for
limbs. For example, when the two hands control a single visual feedback of the limb as responses to cursor jump
cursor, which is presented at the spatial midpoint of the perturbations are less affected by environmental distrac-
hands, perturbations to one arm lead to fast, feedback- tors [46]. This dedicated binding mechanism for visual
driven responses in most muscles in the opposite arm feedback of the moving hand may be generated by the
within 75–100 ms [30,31]. Remarkably, participants can presence of faster somatosensory feedback used to esti-
independently learn and switch between control policies mate limb motion.
acquired for different force fields during unimanual and
bimanual movements [32,33], although reoptimization Feedback responses and learning
may be incomplete when switching between two control A key feature of the voluntary motor system is the ability
policies for unimanual arm movements [34]. Similarly, to learn new motor skills. Many studies have demonstrat-
perturbations applied to one leg during locomotion can ed that the voluntary motor system updates its motor
evoke rapid responses in muscles of the contralateral leg commands when exposed to novel environments that
in less than 75 ms [35]. Critically, these inter-limb alter the physical properties of our limbs (e.g., using
responses only occur when the contralateral limb is sports equipment, walking on sand or in water [47]). As
weight bearing (i.e., stance phase), and a stabilizing described above, numerous studies have highlighted
response is necessary to maintain standing balance. flexible, goal-directed feedback responses across a broad
range of voluntary motor actions. Thus, it seems reason-
Feedback responses are also highly coordinated between able that there is a link between motor learning and
the upper and lower limbs. Perturbations to standing corrective responses.
balance can generate muscle activity in the upper limbs
within 90 ms, which is faster than voluntary reaction We recently tested this hypothesis by examining how
times [36]. These rapid arm movements appear to be corrective responses were altered when subjects learned
influenced by the direction of the perturbation even on to make reaching movements in the presence of resistive
the first exposure [37]. Similarly, perturbations applied to elbow loads (velocity dependent [48]). In random trials,
a handle held by subjects during treadmill walking evoke mechanical perturbations to the elbow were used to probe
muscle responses in the upper and lower limbs that are LLRs during the learning phase. We found that LLRs
modulated by verbal instructions (resist or do not resist adapted during learning and scaled with the strength of

www.sciencedirect.com Current Opinion in Neurobiology 2015, 33:85–94


88 Motor circuits and action

the novel resistive loads applied at the elbow. These Figure 2


altered LLRs correlated with reaching errors during
learning, showing that subjects who adapted more to Neural control
the novel load (i.e., had smaller reach errors) displayed SMA
greater modulation of their LLRs in perturbation trials.
PMd
These results highlight an important link between the M1 S1 5 7
adaptation of feedforward and feedback control, and PMv

suggest a key part of motor adaptation is to adjust feed-


back responses when the physical properties of the limb
or environment are altered. In addition, LLRs compen- BG V1
sate for novel loads when a perturbation is applied before
movement, suggesting the adapted control policy is en-
gaged while preparing a voluntary action [49]. Finally,
both voluntary actions and corrective responses are C
adapted when the strength of the force field is consistent
across trials. In contrast, exposure to variable force fields
Motor behaviour
appears to only modify corrective responses [50], indicat-
ing the adaptation of voluntary actions and feedback
responses may be partly dissociated in some circum-
stances.
Spinal cord
A key challenge with feedback control is dealing with
sensory and motor delays. Recent work illustrates that
forward models might help remove this delay [51].
Further, this study demonstrated that state estimation Musculoskeletal mechanics
processes were altered on a trial-by-trial basis. These
rapid trial-by-trial updates in state estimation are broadly Current Opinion in Neurobiology
consistent with two-timescale models of motor learning
[52] and the recent suggestion that fast learning is associ- Neural basis of feedback processing during voluntary motor actions.
ated with forward models (i.e., state estimation) whereas Spinal feedback provides a fast default response for somatosensory
slow learning is related to inverse models (i.e., feedback feedback. However, goal-directed corrective responses probably
gains [53]). involve cortical and subcortical circuits already ascribed to feed-
forward planning and initiating motor actions (M1, primary motor
cortex; S1, primary somatosensory cortex; PMd, dorsal premotor
Corrective responses to visual cursor shifts also update cortex; PMv, ventral premotor cortex, SMA, supplementary motor
during force and kinematic adaptation [54–56]. Visuomo- area; 5, parietal area 5; 7, parietal area 7; V1, primary visual cortex;
tor corrective responses are modulated by the direction BG, basal ganglia; C, cerebellum).
Adapted from [99].
and structure of visual rotations [56], and selectively
account for the relevance of cursor-jump perturbations
[54]. Remarkably, corrective responses can be learned
independently based on the relevance of visual perturba-
tions towards the left or right side of the workspace. For has been assumed that co-contraction of antagonist mus-
example, subjects can learn to make vigorous corrective cles plays an important role in minimizing the effect of
responses when visual perturbations persist until the end disturbances. However, estimates for muscle stiffness
of movement on the left side of their workspace, but learn commonly used in the field are almost an order of magni-
to disregard transient perturbations on the right side of tude too high, as they include contributions from neural
their workspace [54]. These flexible visuomotor feedback processes [28]. The main benefit of co-con-
responses highlight the specificity of how the motor traction may be to increase the amplitude of spinal stretch
system evaluates and responds to sensory feedback dur- responses (i.e., short latency) that scale with the level of
ing ongoing actions. Collectively, the adaptation of cor- muscle activity, although the link between co-contraction
rective responses highlighted in recent studies establishes and spinal feedback gains may be altered in some situa-
they play an important role in the basic mechanisms of tions [26].
generating skilled actions.
Spinal feedback
Neural implementation of feedback control There is considerable debate regarding the relative con-
The voluntary motor system is hierarchically organized, tribution of spinal versus supraspinal feedback during
and correspondingly, corrective responses reflect contri- voluntary motor tasks, with some concluding spinal feed-
butions from many levels of this hierarchy (Figure 2). It back plays a dominant role [57] and others a minimal role

Current Opinion in Neurobiology 2015, 33:85–94 www.sciencedirect.com


Feedback control during voluntary motor actions Scott et al. 89

[6]. Neural networks inspired by spinal cord circuitry can Genetic tools have provided new insight that spinal
generate reaching movements or respond to perturbations interneurons send efference copies of descending motor
assuming simple descending motor signals [57]. Howev- commands to the cerebellum [66,67]. Azim et al. [67]
er, neural networks without spinal circuitry can also recently illustrated that ablation or modulation of the
generate good control [58]. In all cases, these models activity of V2a neurons that project back to the cerebel-
inevitably ignore a major problem in biological control — lum disturbed reaching behaviour in mice. Taken togeth-
sensory and motor delays. Such delays are devastating for er, the use of genetic and molecular techniques is helping
controlling fast movements or corrective responses [28], to shed new light on how the spinal cord contributes to
but can be mitigated by using state estimation to predict voluntary motor actions.
the sensory consequences of our actions [51]. The com-
plexity of state estimation during voluntary actions may Transcortical feedback
explain why goal-directed muscle responses are only It is generally accepted that transcortical feedback plays
observed at 60 ms when transcortical feedback can an important role in LLRs (for review see [1,3,5–7,10]).
contribute [1,6,10]. Further experimental work, notably The use of transcranial stimulation techniques has dem-
at the spinal level, is essential to resolve this issue, as only onstrated the involvement of M1 in modulating LLRs
a limited number of studies have recorded spinal neural related to knowledge of limb mechanics [68], and phase-
activity in non-human primates during single-joint motor dependent responses between the legs during locomotion
tasks (for review see [59]) or prehension tasks [60,61]. [35]. However, inhibition of motor cortex did not alter
task-dependent LLRs when subjects interacted with stiff
The development of novel molecular and genetic tech- versus compliant loads [69], which opposes previous
niques provides an exciting new approach to understand observations by this group [70]. It may be worthwhile
the function of spinal circuits during voluntary actions to re-examine this issue for tasks where subjects attain a
[62–64]. Fink et al. [63] selectively ablated GABAergic quantifiable goal following the perturbation rather than
interneurons in mice cervical spinal cord that mediate the commonly used ‘do not intervene’ instruction.
presynaptic inhibition of proprioceptive afferents. The
mice could still walk and maintain posture after selective Interest in exploring sensory feedback in M1 has
removal of these interneurons, but displayed pronounced returned. Studies have shown that neural activity is
forelimb oscillations during reaching (Figure 3). These broadly tuned to mechanical perturbations applied to
forelimb oscillations suggest that feedback gains were too the shoulder and elbow [71], and reflects knowledge of
high during movement, which could involve many factors multi-joint limb dynamics [68]. The prevalence of so-
from changes in muscle stiffness to an imbalance between matosensory feedback in M1 is also highlighted by its
spinal and supraspinal feedback processes [28,65]. ability to enhance the performance of Brain Machine
Interfaces generated from M1 activity [72].
Figure 3
The role of M1 in task-dependent feedback processing
was first demonstrated in the seminal experiments by
(a) (c)
Cutaneous
Reach trajectories Evarts and Tanji [73], in which monkeys were instructed
Pre-DT
Proprioceptor 10
Post-DT to either ‘push’ or ‘pull’ a lever triggered by a mechanical
perturbation (Figure 4a [73]). An updated version of this
vGluT1
task was recently performed where monkeys responded
z (mm)

GAD2
0 Pellet
to a mechanical perturbation by moving their hand to a
GAD1
Motor neuron spatial goal [74]. The perturbation either pushed the
(b)
monkey’s hand towards or away from the target, and a
background load was applied before the perturbation to
Camera –10 prevent subthreshold modulation of motoneuron pools.
0
20 10 0 5 y (mm) As in Evarts and Tanji [73], the initial perturbation
Camera x (mm)
response occurred as early as 20 ms and was largely
Current Opinion in Neurobiology
insensitive to the location of the spatial goal. This early,
non-specific response was followed by a task-dependent
Behavioural effect of ablation of presynaptic inhibition on primary
response beginning at 50 ms after perturbation onset
afferents. (a) Gad2-expressing interneurons, which form axo–axonic
contacts with primary afferent terminals, were the target of the genetic (Figure 4b).
manipulation and selectively ablated after the administration of
diphtheria toxin (DT) in mice spinal cord. (b) Mice reached for and The presence of an early non-specific response might
retrieved a food pellet through a narrow access window. (c) Before DT indicate that it takes time for M1 to generate task-
administration mice displayed smooth paw trajectories (pre-DT). In
contrast, after DT administration the reaches lost smoothness and
dependent responses. However, it may also reflect ongo-
exhibited pronounced oscillation (post-DT).Adapted from Fink ing postural control before the perturbation was applied,
et al. [63]. and that it takes time to disengage postural control

www.sciencedirect.com Current Opinion in Neurobiology 2015, 33:85–94


90 Motor circuits and action

Figure 4

(a) Evarts and Tanji (1976) (b) Pruszynski et al. (2014) (c) Omrani et al. (2014)

ins-push/ pert-pull
E L
Posture
60 OUT

Firing Rate (Hz)


40 IN
Movie
ins-pull/ pert-pull
20 Difference
Difference

–500 0 500 –40 0 40 80 120 –40 0 40 80 120


Time (ms) Time (ms) Time (ms)
Current Opinion in Neurobiology

Task-dependent modulation of perturbation responses in M1. (a) Perturbation response of a single pyramidal tract neuron (PTN) for lever ‘pull’
perturbation under different instructions. When the monkey was instructed to ‘push’ the lever (instruction-push/perturbation-pull condition, top),
the PTN showed clear sustained activity until monkey attained the target (thick dot in each row). When the monkey was instructed to ‘pull’ the
lever responding to the same ‘pull’ perturbation (instruction-pull/perturbation-pull condition, bottom), the PTN showed suppression of activity
following the initial excitation. (b) M1 population activity in a spatial target task (n = 104). Monkeys were trained to move to spatial targets
following a mechanical perturbation. The disturbance either moved the hand into (IN, blue) or away from the target (OUT, red). Although M1
neurons responded to the perturbation as early as 20 ms (E, early epoch), target-dependent responses (OUT–IN, gray) began at 50 ms (L, late
epoch). (c) M1 population response in posture and movie tasks (n = 129). M1 activities were recorded when the monkey responded to a
mechanical perturbation to maintain its arm position (posture task, red) or when it was watching a movie and not required to respond to the
perturbation (movie task, blue). Black arrow denotes population response to perturbation and gray arrow denotes task-dependent response. All
data are aligned to perturbation onset (solid vertical line).
Panels (a), (b) and (c) are adapted from [73], [74] and [75], respectively.

before generating a response to the spatial goal. Omrani There is growing evidence that transcortical feedback
et al. [75] tackled this question by comparing M1 per- involves cortical motor regions beyond primary motor
turbation responses when the monkey was engaged cortex. M1 along with dorsal premotor cortex (PMd) show
(positioning its hand at a spatial target) or not engaged a reduction in SEPs during movement [79]. Moreover,
in postural control (watching a movie, Omrani et al. [75]). SEPs in M1 and PMd are also reduced during the delay
The results again indicated the initial M1 response was period prior to movement onset. Inhibition of supple-
identical across behavioural contexts with task-depen- mentary motor area (SMA) using transcranial magnetic
dent responses beginning only 40 ms post-perturbation stimulation (TMS) diminishes the typical modulation of
(Figure 4c). A similar response pattern was observed in the LLR observed for a ‘resist’/do not resist’ paradigm.
M1 when a monkey was required to convert patterns of Reduction of LLR modulation following SMA inhibition
limb motion into the appropriate motor responses to deal suggests SMA may have a direct influence on the LLR
with limb dynamics [68]. This delay may reflect the need and/or that SMA ‘presets’ M1 activity according to task
to wait for multiple sources of sensory feedback before instruction prior to perturbation onset [80]. Finally, a
M1 expresses the appropriate motor response. recent study found that task-dependent LLRs were
absent in chronic stroke patients with unilateral cortical
Transcortical feedback beyond primary motor cortex lesions [81]. Importantly, although strokes were unilat-
A key somatosensory pathway to M1 is through primary eral, deficits in goal-directed LLRs were observed bilat-
somatosensory cortex (S1) where neurons respond to erally in both the affected and unaffected arms. Bilateral
either active or passive arm movements [76]. Move- impairments in LLRs suggest ipsilateral cortex plays a
ment-related activity in S1 can occur before movement direct or indirect role in task-dependent motor correc-
onset, suggesting that some activity in S1 reflects an tions. However, a follow-up study in healthy subjects
efference copy of descending motor commands and found that inhibition of ipsilateral M1 with a strong TMS
may arise from connections with M1 [77,78]. Transmis- pulse did not alter the amplitude of LLRs [69]. Further
sion of sensory information to S1 is reduced during but work is required to understand the role of ipsilateral
not prior to movement, as demonstrated by reductions in cortex in the generation of task-dependent LLRs in
sensory evoked potentials (SEPs) [79]. healthy and patient groups.

Current Opinion in Neurobiology 2015, 33:85–94 www.sciencedirect.com


Feedback control during voluntary motor actions Scott et al. 91

The involvement of cerebellum in LLRs was established responses share most, if not all, of the complexities
in the seminal studies by Strick [82] and Villis and Hore typically ascribed to voluntary control. Considerable work
[83]. Recent work confirms that cerebellar dysfunction is remains to understand how these feedback processes are
associated with altered scaling of the LLR with either generated by the brain and spinal cord. It is likely that the
reduced [84] or exaggerated [85] amplitude. Although role of different brain regions is similar for planning,
cerebellar damage impairs active compensation for inter- initiating and correcting motor actions, blurring the dis-
action torques during multi-joint arm movements [86], tinction between feed-forward and feedback control.
cerebellar patients appear to maintain some knowledge of
limb mechanics in the LLR [84]. Difficulty in dealing
Conflict of interests
with limb mechanics may be linked to deficits in estimat-
SHS is associated with BKIN Technologies, which com-
ing active limb motion [87,88] as knowledge of limb
mercializes the KINARM robot used in some studies
dynamics seems to be compromised by cerebellar damage
described in this review. TC, CRL and TT do not have
[89].
any conflicts of interest.
The basal ganglia may also modulate the transcortical
Acknowledgements
pathway, as a hallmark of stretch responses in patients
This work was supported by grants from the Natural Sciences and
with Parkinson’s disease (PD) is an exaggerated LLR to a Engineering Research Council of Canada (NSERC) and the Canadian
mechanical perturbation [90]. There is also evidence that Institutes of Health Research (CIHR). SHS is supported by a GSK-CIHR
PD patients lose the ability to scale responses to features Chair in Neuroscience. TC is supported by a Banting Postdoctoral
Fellowship.
of the task [91]. However, recent work with a non-human
primate model of PD found no overt change in M1
activity associated with the observed increase in reflex References and recommended reading
gain [92], which implicates subcortical circuits in goal- Papers of particular interest, published within the period of review,
have been highlighted as:
directed feedback responses.
 of special interest
 of outstanding interest
Neural basis of visual feedback
Visual feedback during voluntary motor actions is as- 1. Crevecoeur F, Cluff T, Scott SH: Computational approaches for
sumed to follow the dorsal visual stream towards PMd goal-directed movement planning and execution. The
Cognitive Neurosciences. MIT Press; 2014:: 461-475.
and M1 [9]. PMd appears to play an important role in
online visual feedback as TMS inhibition in this area 2. Roth E, Sponberg S, Cowan N: A comparative approach to
closed-loop computation. Curr Opin Neurobiol 2014, 25:54-62.
disrupts online corrections and reduces the ability to learn
3. Shemmell J, Krutky MA, Perreault EJ: Stretch sensitive reflexes
visuomotor rotations [93]. Sudden changes in target loca- as an adaptive mechanism for maintaining limb stability. Clin
tion at movement onset elicit corrective activity in PMd Neurophysiol 2010, 121:1680-1689.
and M1 [94]. The earliest change in activity following a 4. Diedrichsen J, Shadmehr R, Ivry RB: The coordination of
shift in target location occurs in PMd, followed by M1 and movement: optimal feedback control and beyond. Trends Cogn
Sci 2010, 14:31-39.
then only later in parietal area 5 [95]. Although parietal
area 5 responds after motor cortical regions during online 5. Pruszynski JA, Scott SH: Optimal feedback control and the
long-latency stretch response. Exp Brain Res 2012, 218:341-
adjustments, this cortical region appears to be important 359.
for online control as bilateral muscimol injections delays 6. Scott SH: The computational and neural basis of voluntary
corrective responses following a change in target position motor control and planning. Trends Cogn Sci 2012, 16:541-549.
[96]. 7. Pruszynski JA: Primary motor cortex and fast feedback
responses to mechanical perturbations: a primer on what we
know now and some suggestions on what we should find out
Recent evidence from patients with damage to cortical next. Front Integr Neurosci 2014, 8:72.
areas has delineated hemisphere-specific contributions to
8. Sarlegna FR, Mutha PK: The influence of visual target
the use of this visual feedback. For example, damage to information on the online control of movements. Vision Res
the right frontal lobe delays the onset of corrective 2014 http://dx.doi.org/10.1016/j.visres.2014.07.001.
responses [97,98], whereas left hemisphere damage 9. Archambault PS, Ferrari-Toniolo S, Caminiti R, Battaglia-Mayer A:
interferes with inter-limb coordination during rapid on- Visually-guided correction of hand reaching movements: the
neurophysiological bases in the cerebral cortex. Vision Res
line corrections [97]. 2014 http://dx.doi.org/10.1016/j.visres.2014.09.009.
10. Cluff T, Crevecoeur F, Scott SH: A perspective on multisensory
Conclusions integration and rapid perturbation responses. Vision Res 2014
A growing body of literature highlights the surprising http://dx.doi.org/10.1016/j.visres.2014.06.011.
sophistication of rapid corrective responses generated 11. Cisek P, Pastor-Bernier A: On the challenges and mechanisms
by visual and somatosensory feedback. Long-latency of embodied decisions. Philos Trans R Soc B Biol Sci 2014,
369:20130479.
stretch responses are highly context-dependent reflecting
12. Shenoy KV, Sahani M, Churchland MM: Cortical control of arm
the behavioural goal and many other factors essential for movements: a dynamical systems perspective. Annu Rev
moving and interacting in the world. These corrective Neurosci 2013, 36:337-359.

www.sciencedirect.com Current Opinion in Neurobiology 2015, 33:85–94


92 Motor circuits and action

13. Crevecoeur F, Kurtzer I, Bourke T, Scott SH: Feedback 31. Omrani M, Diedrichsen J, Scott SH: Rapid feedback corrections
responses rapidly scale with the urgency to correct for during a bimanual postural task. J Neurophysiol 2013, 109:
external perturbations. J Neurophysiol 2013, 110:1323-1332. 147-161.
14. Nashed JY, Crevecoeur F, Scott SH: Influence of the behavioral 32. White O, Diedrichsen J: Flexible switching of feedback control
goal and environmental obstacles on rapid feedback mechanisms allows for learning of different task dynamics.
responses. J Neurophysiol 2012, 108:999-1009. PLoS ONE 2013, 8:e54771.

15. Crevecoeur F, Kurtzer I, Scott SH: Fast corrective responses are 33. Nozaki D, Kurtzer I, Scott SH: Limited transfer of learning
evoked by perturbations approaching the natural variability of between unimanual and bimanual skills within the same limb.
posture and movement tasks. J Neurophysiol 2012, 107:2821- Nat Neurosci 2006, 9:1364-1366.
2832.
34. De Xivry J-JO: Trial-to-trial reoptimization of motor behavior
16. Nashed JY, Kurtzer IL, Scott SH: Context-dependent inhibition due to changes in task demands is limited. PLoS ONE 2013,
of unloaded muscles during the long-latency epoch. J 8:e66013.
Neurophysiol 2015, 113:192-202.
35. Stevenson AJT, Geertsen SS, Andersen JB, Sinkjær T, Nielsen JB,
17. Nashed JY, Crevecoeur F, Scott SH: Rapid online selection  Mrachacz-Kersting N: Interlimb communication to the knee
 between multiple motor plans. J Neurosci 2014, 34:1769-1780. flexors during walking in humans. J Physiol 2013, 591:4921-
This study used mechanical disturbances to show the motor system can 4935.
make online decisions to avoid obstacles in the environment or select This study shows that M1 is involved in regulating long-latency responses
amongst alternate behavioural goals. Decisions about how to reach a when the contralateral limb is perturbed during locomotion. Mechanical
target happen in 60 ms after a perturbation, whereas decisions to perturbations applied to the knee during walking evoked a response in
switch goals occur in 75 ms. contralateral leg muscles at 75 ms. Critically, this effect was task-
dependent as the response was seen only at a specific point in the gait
18. Bedingham W, Tatton WG: Dependence of EMG responses cycle (50% gait cycle, when contralateral limb is loaded in single stance).
evoked by imposed wrist displacements on pre-existing The authors used TMS and TES to show that the coordination of the reflex
activity in the stretched muscles. Can J Neurol Sci J Can Sci responses between the two limbs is dependent on cortical processing in
Neurol 1984, 11:272-280. M1.

19. Matthews PB: Observations on the automatic compensation of 36. McIlroy WE, Maki BE: Early activation of arm muscles follows
reflex gain on varying the pre-existing level of motor discharge external perturbation of upright stance. Neurosci Lett 1995,
in man. J Physiol 1986, 374:73-90. 184:177-180.

20. Pruszynski JA, Kurtzer I, Lillicrap TP, Scott SH: Temporal 37. Corbeil P, Bloem BR, van Meel M, Maki BE: Arm reactions
evolution of ‘‘automatic gain-scaling’’. J Neurophysiol 2009, evoked by the initial exposure to a small balance perturbation:
102:992-1003. a pilot study. Gait Posture 2013, 37:300-303.

21. Selen LPJ, Shadlen MN, Wolpert DM: Deliberation in the motor 38. Forero J, Misiaszek JE: Balance-corrective responses to
system: reflex gains track evolving evidence leading to a unexpected perturbations at the arms during treadmill
decision. J Neurosci 2012, 32:2276-2286. walking. J Neurophysiol 2014, 112:1790-1800.

22. Wolpert DM, Landy MS: Motor control is decision-making. Curr 39. Georgopoulos AP, Kalaska JF, Massey JT: Spatial trajectories
Opin Neurobiol 2012, 22:996-1003. and reaction times of aimed movements: effects of practice,
uncertainty, and change in target location. J Neurophysiol 1981,
23. Jacobs JV, Horak FB: Cortical control of postural responses. J 46:725-743.
Neural Transm 2007, 114:1339-1348.
40. Goodale MA, Pelisson D, Prablanc C: Large adjustments in
24. Safavynia SA, Ting LH: Long-latency muscle activity reflects visually guided reaching do not depend on vision of the hand
 continuous, delayed sensorimotor feedback of task-level and or perception of target displacement. Nature 1986, 320:748-
not joint-level error. J Neurophysiol 2013, 110:1278-1290. 750.
This study showed that long-latency responses of lower limb muscles 41. Pélisson D, Prablanc C, Goodale MA, Jeannerod M: Visual control
evoked by postural perturbations are modulated by task-level variables of reaching movements without vision of the limb. Exp Brain
(control of the center of body mass) rather than driven by feedback about Res 1986, 62:303-311.
individual joint displacements.
42. Prablanc C, Martin O: Automatic control during hand reaching
25. Safavynia SA, Ting LH: Sensorimotor feedback based on task- at undetected two-dimensional target displacements. J
relevant error robustly predicts temporal recruitment and Neurophysiol 1992, 67:455-469.
multidirectional tuning of muscle synergies. J Neurophysiol
2013, 109:31-45. 43. Sarlegna F, Blouin J, Bresciani J-P, Bourdin C, Vercher J-L,
Gauthier GM: Target and hand position information in the
26. Finley JM, Dhaher YY, Perreault EJ: Contributions of feed- online control of goal-directed arm movements. Exp Brain Res
forward and feedback strategies at the human ankle during 2003, 151:524-535.
control of unstable loads. Exp Brain Res 2012, 217:53-66.
44. Dimitriou M, Wolpert DM, Franklin DW: The temporal evolution of
27. Finley JM, Dhaher YY, Perreault EJ: Acceleration dependence  feedback gains rapidly update to task demands. J Neurosci
and task-specific modulation of short- and medium-latency 2013, 33:10898-10909.
reflexes in the ankle extensors. Physiol Rep 2013, 1:e00051. This study shows that unexpected changes in the movement goal require
100 ms for online corrections to be updated to the new location of the
28. Crevecoeur F, Scott SH: Beyond muscles stiffness: importance spatial goal.
 of state-estimation to account for very fast motor corrections.
PLoS Comput Biol 2014, 10:e1003869. 45. Yang L, Michaels JA, Pruszynski JA, Scott SH: Rapid motor
This study shows that state estimation is required to deal with sensory responses quickly integrate visuospatial task constraints. Exp
feedback delays to generate rapid upper limb corrections. This study also Brain Res 2011, 211:231-242.
shows that intrinsic muscle stiffness is insufficient for successful and
stable performance during movement. 46. Reichenbach A, Franklin DW, Zatka-Haas P, Diedrichsen J: A
 dedicated binding mechanism for the visual control of
29. Perreault EJ, Chen K, Trumbower RD, Lewis G: Interactions with movement. Curr Biol 2014, 24:780-785.
compliant loads alter stretch reflex gains but not This study outlines a dedicated mechanism for processing visual feed-
intermuscular coordination. J Neurophysiol 2008, 99:2101- back related to a moving limb. This binding mechanism for limb motion is
2113. not influenced by visual attention and may be generated by the presence
of faster somatosensory feedback to estimate limb motion.
30. Dimitriou M, Franklin DW, Wolpert DM: Task-dependent
coordination of rapid bimanual motor responses. J 47. Wolpert DM, Diedrichsen J, Flanagan JR: Principles of
Neurophysiol 2012, 107:890-901. sensorimotor learning. Nat Rev Neurosci 2011, 12:739-751.

Current Opinion in Neurobiology 2015, 33:85–94 www.sciencedirect.com


Feedback control during voluntary motor actions Scott et al. 93

48. Cluff T, Scott SH: Rapid feedback responses correlate with 66. Pivetta C, Esposito MS, Sigrist M, Arber S: Motor-circuit
 reach adaptation and properties of novel upper limb loads. J communication matrix from spinal cord to brainstem neurons
Neurosci 2013, 33:15903-15914. revealed by developmental origin. Cell 2014, 156:537-548.
This study highlights an important link between the adaptation of volun-
tary actions and online feedback responses to counter mechanical 67. Azim E, Jiang J, Alstermark B, Jessell TM: Skilled reaching relies
perturbations. This is the first study to show that motor adaptation alters  on a V2a propriospinal internal copy circuit. Nature 2014,
long-latency responses during reaching. 508:357-363.
This study shows that ablation of the V2a propriospinal (PN) neurons,
49. Ahmadi-Pajouh MA, Towhidkhah F, Shadmehr R: Preparing to which have dual projections to forelimb motoneurons and precerebeller
reach: selecting an adaptive long-latency feedback controller. neurons of the lateral reticular nucleus, disturbed reaching in mice. This is
J Neurosci 2012, 32:9537-9545. the first study to show that internal feedback via V2a PN is crucial for
voluntary motor control.
50. Yousif N, Diedrichsen J: Structural learning in feedforward and
feedback control. J Neurophysiol 2012, 108:2373-2382. 68. Pruszynski JA, Kurtzer I, Nashed JY, Omrani M, Brouwer B,
Scott SH: Primary motor cortex underlies multi-joint
51. Crevecoeur F, Scott SH: Priors engaged in long-latency integration for fast feedback control. Nature 2011, 478:387-390.
 responses to mechanical perturbations suggest a rapid
update in state estimation. PLoS Comput Biol 2013, 9:e1003177. 69. Fox J, Shemmell J: The ipsilateral motor cortex does not
This study shows that the motor system uses prior expectations (updated contribute to long-latency stretch reflex amplitude at the
from trial-to-trial) about the perturbation profile to extrapolate delayed wrist. Brain Behav 2014, 4:60-69.
sensory signals forward in time to estimate the present state of the limb.
This is the first study to show that the motor system exploits internal 70. Shemmell J, An JH, Perreault EJ: The differential role of motor
forward models for somatosensory feedback. cortex in stretch reflex modulation induced by changes in
environmental mechanics and verbal instruction. J Neurosci
52. Smith MA, Ghazizadeh A, Shadmehr R: Interacting adaptive 2009, 29:13255-13263.
processes with different timescales underlie short-term motor
learning. PLoS Biol 2006, 4:e179. 71. Herter TM, Korbel T, Scott SH: Comparison of neural responses
in primary motor cortex to transient and continuous loads
53. Yavari F, Towhidkhah F, Ahmadi-Pajouh MA: Are fast/slow during posture. J Neurophysiol 2009, 101:150-163.
process in motor adaptation and forward/inverse internal
model two sides of the same coin? Med Hypotheses 2013, 72. Suminski AJ, Tkach DC, Fagg AH, Hatsopoulos NG:
81:592-600. Incorporating feedback from multiple sensory modalities
enhances brain–machine interface control. J Neurosci 2010,
54. Franklin DW, Franklin S, Wolpert DM: Fractionation of the 30:16777-16787.
visuomotor feedback response to directions of movement and
perturbation. J Neurophysiol 2014, 112:2218-2233. 73. Evarts EV, Tanji J: Reflex and intended responses in motor
cortex pyramidal tract neurons of monkey. J Neurophysiol
55. Franklin S, Wolpert DM, Franklin DW: Visuomotor feedback 1976, 39:1069-1080.
gains upregulate during the learning of novel dynamics. J
Neurophysiol 2012, 108:467-478. 74. Pruszynski JA, Omrani M, Scott SH: Goal-dependent modulation
of fast feedback responses in primary motor cortex. J Neurosci
56. Telgen S, Parvin D, Diedrichsen J: Mirror reversal and visual 2014, 34:4608-4617.
rotation are learned and consolidated via separate
mechanisms: recalibrating or learning de novo? J Neurosci 75. Omrani M, Pruszynski JA, Murnaghan CD, Scott SH:
2014, 34:13768-13779.  Perturbation-evoked responses in primary motor cortex are
modulated by behavioral context. J Neurophysiol 2014,
57. Tsianos GA, Goodner J, Loeb GE: Useful properties of spinal 112:2985-3000.
circuits for learning and performing planar reaches. J Neural This study shows the initial response in M1 following a mechanical
Eng 2014, 11:056006. perturbation is present whether or not the monkey is not engaged in
an ongoing limb motor task. Thus a ‘default’ or ‘non-specific’ perturbation
58. Lillicrap TP, Scott SH: Preference distributions of primary motor response is always generated in M1 before it expresses task-related
cortex neurons reflect control solutions optimized for limb responses, suggesting that it takes time for the motor circuits to select the
biomechanics. Neuron 2013, 77:168-179. appropriate motor response following a limb disturbance.
59. Fetz EE, Perlmutter SI, Prut Y, Seki K, Votaw S: Roles of primate 76. London BM, Miller LE: Responses of somatosensory area
spinal interneurons in preparation and execution of voluntary 2 neurons to actively and passively generated limb
hand movement. Brain Res Rev 2002, 40:53-65. movements. J Neurophysiol 2013, 109:1505-1513.
60. Takei T, Seki K: Spinal interneurons facilitate coactivation of 77. Hill DN, Curtis JC, Moore JD, Kleinfeld D: Primary motor cortex
hand muscles during a precision grip task in monkeys. J reports efferent control of vibrissa motion on multiple
Neurosci 2010, 30:17041-17050. timescales. Neuron 2011, 72:344-356.

61. Takei T, Seki K: Spinal premotor interneurons mediate dynamic 78. Petreanu L, Gutnisky DA, Huber D, Xu N-L, O’Connor D, Tian L,
and static motor commands for precision grip in monkeys. J Looger L, Svoboda K: Activity in motor-sensory projections
Neurosci 2013, 33:8850-8860. reveals distributed coding in somatosensation. Nature 2012,
489:299-303.
62. Kinoshita M, Matsui R, Kato S, Hasegawa T, Kasahara H, Isa K,
Watakabe A, Yamamori T, Nishimura Y, Alstermark B et al.: 79. Seki K, Fetz EE: Gating of sensory input at spinal and cortical
Genetic dissection of the circuit for hand dexterity in primates. levels during preparation and execution of voluntary
Nature 2012, 487:235-238. movement. J Neurosci 2012, 32:890-902.

63. Fink AJP, Croce KR, Huang ZJ, Abbott LF, Jessell TM, Azim E: 80. Spieser L, Aubert S, Bonnard M: Involvement of SMAp in the
 Presynaptic inhibition of spinal sensory feedback ensures intention-related long latency stretch reflex modulation: a
smooth movement. Nature 2014, 509:43-48. TMS study. Neuroscience 2013, 246:329-341.
This study shows that ablation of presynaptic inhibition on primary
afferent terminals severely impairs smooth reaching in mice, suggesting 81. Trumbower RD, Finley JM, Shemmell JB, Honeycutt CF,
presynaptic inhibition contributes to feedback control. This is the first  Perreault EJ: Bilateral impairments in task-dependent
direct evidence to show the impact of primary afferent presynaptic modulation of the long-latency stretch reflex following stroke.
inhibition on the control of voluntary movement. Clin Neurophysiol 2013, 124:1373-1380.
This study shows that, unlike healthy subjects, stroke patients do not
64. Zhou K, Wolpert DM, De Zeeuw CI: Motor systems: reaching out modulate long-latency response by load type (stiff versus compliant). This
and grasping the molecular tools. Curr Biol 2014, 24:R269- demonstrates that cortical damage results in the loss of task-related
R271. flexibility of corrective responses. Surprisingly, the modulation of long-
latency responses was absent in both the paretic and non-paretic limb for
65. Scott SH, Crevecoeur F: Neuroscience: feedback throttled stroke patients demonstrating that unilateral cortical damage can lead to
down for smooth moves. Nature 2014, 509:38-39. bilateral sensorimotor impairments.

www.sciencedirect.com Current Opinion in Neurobiology 2015, 33:85–94


94 Motor circuits and action

82. Strick PL: The influence of motor preparation on the response This study used a non-human primate model of Parkinson’s disease (PD).
of cerebellar neurons to limb displacements. J Neurosci 1983, They replicated the exaggerated LLR which is a hallmark of reflex
3:2007-2020. responses in patients with PD. Despite larger reflex responses, cortical
recordings showed no evidence of a concomitant increase in M1 neuron
83. Vilis T, Hore J: Central neural mechanisms contributing to activity suggesting that increased reflex gain in PD is not directly caused
cerebellar tremor produced by limb perturbations. J by an overt increase in M1 activity.
Neurophysiol 1980, 43:279-291.
93. Lee J-H, van Donkelaar P: The human dorsal premotor cortex
84. Kurtzer I, Trautman P, Rasquinha RJ, Bhanpuri NH, Scott SH, generates on-line error corrections during sensorimotor
Bastian AJ: Cerebellar damage diminishes long-latency adaptation. J Neurosci 2006, 26:3330-3334.
responses to multijoint perturbations. J Neurophysiol 2013,
109:2228-2241. 94. Ames KC, Ryu SI, Shenoy KV: Neural dynamics of reaching
following incorrect or absent motor preparation. Neuron 2014,
85. Grimaldi G, Manto M: Anodal transcranial direct current 81:438-451.
stimulation (tDCS) decreases the amplitudes of long-latency
stretch reflexes in cerebellar ataxia. Ann Biomed Eng 2013, 95. Archambault PS, Ferrari-Toniolo S, Battaglia-Mayer A: Online
41:2437-2447. control of hand trajectory and evolution of motor intention in
the parietofrontal system. J Neurosci 2011, 31:742-752.
86. Bastian AJ, Martin TA, Keating JG, Thach WT: Cerebellar ataxia:
abnormal control of interaction torques across multiple joints. 96. Battaglia-Mayer A, Ferrari-Toniolo S, Visco-Comandini F,
J Neurophysiol 1996, 76:492-509. Archambault PS, Saberi-Moghadam S, Caminiti R: Impairment of
online control of hand and eye movements in a monkey model
87. Bhanpuri NH, Okamura AM, Bastian AJ: Predictive modeling by of optic ataxia. Cereb Cortex 2012 http://dx.doi.org/10.1093/
the cerebellum improves proprioception. J Neurosci 2013, cercor/bhs250.
33:14301-14306.
97. Schaefer SY, Mutha PK, Haaland KY, Sainburg RL: Hemispheric
88. Boisgontier MP, Swinnen SP: Proprioception in the cerebellum. specialization for movement control produces dissociable
Front Hum Neurosci 2014, 8:212. differences in online corrections after stroke. Cereb Cortex
89. Bhanpuri NH, Okamura AM, Bastian AJ: Predicting and 2012, 22:1407-1419.
correcting ataxia using a model of cerebellar function. Brain 98. Mutha PK, Stapp LH, Sainburg RL, Haaland KY: Frontal and
2014 http://dx.doi.org/10.1093/brain/awu115.  parietal cortex contributions to action modification. Cortex
90. Rothwell J, Obeso J, Traub M, Marsden C: The behaviour of the 2014, 57:38-50.
long-latency stretch reflex in patients with Parkinson’s This study localized specific cortical areas involved in corrective move-
disease. J Neurol Neurosurg Psychiatry 1983, 46:35-44. ments following a visual perturbation. Patients with focal lesions to right
and left frontal and parietal areas performed a reaching task with a visual
91. Bloem BR, Beckley DJ, Remler MP, Roos RAC, van Dijk JG: cursor jump: right frontal lobe damage was associated with delayed onset
Postural reflexes in Parkinson’s disease during ‘‘resist’’ and of corrective movements and left parietal damage decreased the accu-
‘‘yield’’ tasks. J Neurol Sci 1995, 129:109-119. racy of the correction suggesting unique, a hemisphere-specific roles for
these areas in action modification following visual perturbation.
92. Pasquereau B, Turner RS: Primary motor cortex of the
 Parkinsonian monkey: altered neuronal responses to muscle 99. Scott SH: Optimal feedback control and the neural basis of
stretch. Front Syst Neurosci 2013, 7:98. volitional motor control. Nat Rev Neurosci 2004, 5:534-546.

Current Opinion in Neurobiology 2015, 33:85–94 www.sciencedirect.com

You might also like