PLoS BIOLOGY

Disruption of State Estimation

in the Human Lateral Cerebellum
R. Chris Miall1*, Lars O. D. Christensen2, Owen Cain1, James Stanley1
1 School of Psychology, University of Birmingham, Birmingham, United Kingdom, 2 Department of Experimental Psychology, University of Oxford, Oxford, United Kingdom

The cerebellum has been proposed to be a crucial component in the state estimation process that combines
information from motor efferent and sensory afferent signals to produce a representation of the current state of the
motor system. Such a state estimate of the moving human arm would be expected to be used when the arm is rapidly
and skillfully reaching to a target. We now report the effects of transcranial magnetic stimulation (TMS) over the
ipsilateral cerebellum as healthy humans were made to interrupt a slow voluntary movement to rapidly reach towards
a visually defined target. Errors in the initial direction and in the final finger position of this reach-to-target movement
were significantly higher for cerebellar stimulation than they were in control conditions. The average directional errors
in the cerebellar TMS condition were consistent with the reaching movements being planned and initiated from an
estimated hand position that was 138 ms out of date. We suggest that these results demonstrate that the cerebellum is
responsible for estimating the hand position over this time interval and that TMS disrupts this state estimate.
Citation: Miall RC, Christensen LOD, Cain O, Stanley J (2007) Disruption of state estimation in the human lateral cerebellum. PLoS Biol 5(11): e316. doi:10.1371/journal.pbio.
0050316

[7,11]. A forward model receives efferent copies of the motor

Introduction
commands and also receives sensory inputs that describe the
The central nervous system (CNS) can never know exactly motor state. The output of the model is a prediction of the
the current state of the peripheral motor apparatus—the sensory consequences of the motor command, i.e., a
limbs and muscles that are under CNS control—because of prediction of the change in motor state. State estimation
unavoidable delays in conduction of sensory afferent signals must be a predictive process because of central delays in
from the periphery, as well as in their central neural processing of the motor command, in peripheral conduction
processing. Hence the sensed state of the system (the set of of the efferent signal, and in neuromuscular excitation-
variables including limb segment positions and velocities that contraction coupling. Hence the true motor state of the
capture its behaviour) always lags behind its true state [1]. motor periphery lags behind the central (CNS) changes in
These delays vary with the sensory modality but can be motor commands. The state estimation process is therefore
substantial, and estimates of the delay involved in using visual inseparably coupled to the process of forward modelling [11].
feedback to control and correct ongoing movements vary Forward modelling has been proposed to be a key function
from about 100–300 ms [2–5]. In addition, any physiological of the cerebellum [7,12,13], and the cerebellum has been
sensor will have some inaccuracies, compounded by neural specifically linked to state estimation [14,15], possibly in
noise, that lead to errors in the measurements. Furthermore, conjunction with the superior parietal cortex [4,16–18]. The
the parameters that the CNS might aim to control, such as the cerebellum receives appropriate ascending proprioceptive
position or velocity of the peripheral motor system, are often inputs and the efferent copies of descending motor com-
hidden from the CNS by indirect relationships between these mands, and it outputs to cortical and brain stem motor nuclei
peripheral variables (muscle lengths or joint angles) and the [12]. It also has the necessary adaptive mechanisms to support
sensory encoders. For example, vertebrate joint angles are this hypothesised role, because the forward model predic-
encoded mainly in information carried by muscle spindles, tions must be refined and maintained by experience-based
which can only provide a mixed signal that is proportional to motor learning [11,12]. However, to date, there has been no
muscle length and its rate of change. To measure and control direct experimental evidence of this cerebellar contribution
the kinematics of a movement requires decoding these to state estimation; indirect evidence has been derived from
afferent signals to estimate joint angles from muscle lengths. brain imaging [19–22] and from studies of cerebellar patients
In addition, combining an independent prediction of the with chronic lesions [23–25].
state of the peripheral apparatus with afferent measurements
of its state can provide an estimate that is more accurate than
Academic Editor: James Ashe, University of Minnesota, United States of America
that of either predictor or sensors alone [6]. For these various
Received July 11, 2007; Accepted September 28, 2007; Published November 27,
reasons, it is widely assumed that the brain generates an 2007
estimate of the true state of the peripheral motor system, by
Copyright: Ó 2007 Miall et al. This is an open-access article distributed under the
integration of the latest afferent sensory information with an terms of the Creative Commons Attribution License, which permits unrestricted
efferent copy of motor commands using prior knowledge of use, distribution, and reproduction in any medium, provided the original author
the relationships between efferent signals and the subsequent and source are credited.
sensory reafference [7–10]. The process of translating an Abbreviations: CNS, central nervous system; PPC, posterior parietal cortex; TMS,
efferent copy of a motor command into predicted sensory transcranial magnetic stimulation
reafference is encapsulated by the idea of a forward model * To whom correspondence should be addressed. E-mail: r.c.miall@bham.ac.uk

PLoS Biology | www.plosbiology.org 2733 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

Author Summary stimulate neck muscles, the brachial plexus, muscles in the
neck or shoulder, and is sufficiently loud that it can provide a
Motor control depends on the brain’s awareness of the current state startling stimulus affecting speed of movement onset. We
of the body. Knowing the current position and movement of the have used a series of control conditions to separate non-
arm, for example, allows one to reach rapidly and accurately towards specific effects from a specific change in initial movement
a target. However, sensory information reaches the brain only after a direction and in terminal error, which were seen only with
short delay, and the arm may already be in motion. Therefore, it has cerebellar TMS.
been proposed that the brain must calculate a ‘‘state estimate’’—by
combining sensory information about the last known position of the
arm with predictions of its responses to recent movement Results
commands—which it uses to accurately plan and control a reaching
movement. To test this idea, we used transcranial magnetic Participants viewed a virtual image of a static target in
stimulation to briefly disrupt several separate areas in the brain as three-dimensional (3-D) space ahead of them, and started
participants reached to a target. We show that stimulation over the each trial by lifting their right index finger from a start key
cerebellum caused reaching errors consistent with movements and moving steadily towards their right (Figure 1A). Liquid
planned on the arm’s position about 140 ms previously, whereas crystal device (LCD) goggles blocked the view of their hand
stimulation of other brain areas did not disrupt reaching direction. and of the target as soon at the start key was released. An
These results add weight to the hypothesis that the cerebellum auditory go cue, 500-1500 ms after trial onset, instructed
predicts the state of the motor system. This hypothesis can explain them to make a rapid upwards- and leftwards- pointing
the loss of movement control experienced by cerebellar patients
movement to the virtual target. Their index finger had
and supports computational theories that the cerebellum is a
predictive model of the motor system. typically moved laterally 10–40 cm from its original position
when the go cue was delivered (Figure 1B). Final positional
errors on control trials were small (Figure 1B) and averaged
4.2 cm across all conditions. Thus, participants were normally
A loss of state estimation would lead to inaccuracies in able to compensate for their initial lateral arm motion and
motor control, because control signals would be based on reach the target despite the lack of visual feedback. Vision
out-of-date information. Thus a rapid reaching action made was allowed after the reach-to-target motion was complete,
without state estimation of the moving hand would tend to avoiding any slow drifting of accuracy across trials. However,
overshoot its target, because information that the desired on a random 50% of trials in each block, TMS was delivered
target had been reached would only arrive at the CNS after within their reaction time after the auditory go cue, in order
the hand had passed beyond. This would result in movement to disrupt the planning and initiation of the reach-to-target
errors analogous to the hypometria of cerebellar patients movement. Reaction times for control trials without TMS
[12]. State estimation is also important for the synchronous averaged 265 ms (discussed later), but were reduced to 170 ms
and coordinated activation of different motor effectors. If the during TMS trials; the three TMS pulses were delivered at 50,
future state of one effector can be predicted, then control 100, and 150 ms during this interval.
signals to the other can be issued to produce simultaneous
actions, which are a key feature of coordinated action. Cerebellar TMS Increases Final Error
Without these predictions, the two effectors could only be The short train of three TMS pulses delivered over the
controlled reactively [23,24], after measurement of the lateral cerebellum caused a significant within-subject increase
outcome of each command. The loss of coordination and in mean error for TMS trials compared with non-TMS trials.
asynchrony of joint actions that would be expected from a In our initial experiments, we tested eight participants with
failure of state estimation are again similar to the poorly TMS over the right ipsilateral cerebellum, the contralateral
coordinated and ataxic movements of cerebellar subjects (left) motor cortex, and the ipsilateral neck, using separate
[12,26,27]. Thus, there is theoretical and experimental recording sessions separated by at least one day. The mean
evidence to suggest that the cerebellum is involved in state increase in end-point errors with the cerebellar stimulation
estimation. To date, we are aware of no studies that have site was 36% (2.26 cm 6 0.37 standard error of the mean
directly tested this hypothesis by experimental disruption of [SEM], n ¼ 8, t(7) ¼ 5.72, p , 0.0001), and was significantly
the cerebellum. higher than the other two conditions (repeated measures
So, to further test the hypothesis that the human analysis of variance [ANOVA], F(2,14) ¼ 4.468, p ¼ 0.032).
cerebellum is involved in the generation of a state estimate, However, this reduction in pointing accuracy could have
we have used transcranial magnetic stimulation (TMS) over been a nonspecific effect of the TMS stimulation, which can
the ipsilateral cerebellum during voluntary arm actions to be uncomfortable and even startling. To include other
briefly disturb its function. We used a task in which humans control conditions, we then expanded our cerebellar test
were required to make a slow, lateral, but untargeted group to a total of 32 participants, testing each participant in
movement with their arm before being suddenly cued to this main condition of interest as well as in one or more other
make a rapid pointing movement towards a static target. conditions. Because the extra participants were not tested in
Accurate reaching in these circumstances requires up-to-date all other stimulation conditions, the following analyses are
knowledge of the arm’s moving position at the moment of the reported as between-group comparisons.
go cue. Any failure to estimate the arm’s initial state caused With this expanded dataset (Figure 2A), the increase in
by the cerebellar TMS should be evident as inaccurate mean terminal errors in cerebellar TMS trials compared to
movement. However, because of its location, the human non-TMS trials was reduced from 36% to 23.7% (or 1.71 cm
cerebellum is difficult to stimulate with transcranial coils, and 6 0.144 SEM, n ¼ 32, t(31) ¼ 3.80, p , 0.001). However, this
TMS targeted at the lateral cerebellum can also directly TMS-induced error was still significantly higher when

PLoS Biology | www.plosbiology.org 2734 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

Figure 1. The Experimental Task and Typical Single-Participant Data

The experimental task (A), individual trial data (B and C), and session averaged data (D and E) (n ¼ 30 trials) from one typical participant as TMS were
applied over the right lateral cerebellum. (B and D) show the finger trajectory viewed from behind and from the right of the subject (C and E). In all
panels, TMS trials are plotted in red and non-TMS trials are in blue. Clockwise rotation in (B and D) is defined as increasing azimuth angle; clockwise
rotation in (C and E) is defined as decreasing elevation angle.
doi:10.1371/journal.pbio.0050316.g001

compared with stimulation lower on the neck (1.12 cm, n ¼ the other four conditions were all significant, p , 0.025). The
11), or over the hotspot in the primary motor cortex for difference from stimulation over the hand area of the
inducing visible twitches in the first dorsal interosseous contralateral motor cortex was smallest (p ¼ 0.025); the other
muscle in the hand (1.23 cm, n ¼ 21), and higher than when four control conditions were not significantly separable from
startling auditory clicks were presented either using the TMS each other (p . 0.27)]. Thus, whereas each of these sites
coil over the ear or using ear phones without TMS (1.18 cm, n induced some increase in end-point error, presumably due to
¼ 4 and 7, respectively). It was also higher than with the nonspecific effects of the stimulation, the effects caused
stimulation over the contralateral posterior parietal cortex by stimulation over the ipsilateral cerebellum were most
(1.07 cm, n ¼ 12), which was targeted using the coordinates of pronounced and statistically reliable
the P3 electrode in the 10–20 electroencephalogram (EEG) The increase in error was partly due to a 14% increase in
electrode positioning scheme [28]. A one-way ANOVA with end-point variability across trials. However, the RMS end-
five conditions (cerebellum, neck, startle, parietal, and motor point standard deviation measured across all three dimen-
cortical stimulation) was significant (F(4,78) ¼ 3.79, p ¼ 0.007, sions was not significantly different for any of the five
and post-hoc comparisons of the cerebellar condition with conditions (p . 0.103). There was also a significant end-point

PLoS Biology | www.plosbiology.org 2735 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

increase in overall amplitude of the reach-to-target move-

ment did not reach statistical significance (p ¼ 0.14).

Baseline Performance Differences

Our analysis compares within-subject average errors across
sessions including 30 TMS and 30 non-TMS trials. To
estimate the expected level of difference between these
means due to random sampling from a distribution of
variable movements, we analysed the training data for 18
participants before TMS was applied. In training, the TMS
machine was placed behind the participant and was triggered
exactly as in the test sessions, so that its activation was audible
to the participants but had no direct effect. Trials were then
grouped by TMS activation versus nonactivation. As ex-
Figure 2. TMS-Induced Difference in Mean End-Point Error
Each bar is the group mean difference for TMS versus non-TMS trials (þ1
pected, there were no significant differences between move-
SEM). TMS was applied over the cerebellum during rightwards and ment trajectories, reaction times, or peak velocities of the
leftwards movement (CBR, n ¼ 32, CBL, n ¼ 13) and when stationary (STR, movements. The mean terminal errors differed by 0.40 cm
n ¼ 9). Control conditions included during startle trials (STL, n ¼ 11),
stimulation of the ipsilateral neck (NK, n ¼ 10), the hand area of (60.1 cm SEM, n ¼ 18) and the average spatial separation of
contralateral primary motor cortex (M1, n ¼ 20), and the contralateral the mean end positions in the two data sets was 0.8 cm (60.09
posterior parietal cortex (PPC, n ¼ 12). cm SEM, n ¼ 18). This suggests that random sampling of any
doi:10.1371/journal.pbio.0050316.g002
one of our datasets would produce differences representing
about half of that seen in our control conditions (Figure 2),
positional bias for cerebellar stimulation, as the TMS trials and less than a quarter of the effect seen for TMS over the
ended on average 1.0 cm above and slightly behind the non- cerebellum. Moreover, these figures (50% and 25%) are
TMS trials (Figure 3A). This corresponds with a small conservative, based on 18 training sets compared with the
hypermetric overshoot and a small directional error; the total sample of 32 for the cerebellum and about 10–12 for

Figure 3. Group Mean Trajectories

Group mean trajectories (A) for TMS trials (red) and non-TMS trials (blue) applied over the cerebellum (n ¼ 32). (B) Results from startling TMS or auditory
trials, without cerebellar disruption (n ¼ 11). In both panels, the curved path followed from bottom left to right is during the pre-cue period. Shortly
after the go cue and TMS, a rapid reach-to-target towards the upper left target position is made. The 3-D inset figures show an expanded view of the
reach-to-target initiation. Black dots mark the position on the non-TMS mean trajectory (blue line) from which a similar angular deviation between start
and maximum velocity would be found as seen in the TMS trials.
doi:10.1371/journal.pbio.0050316.g003

PLoS Biology | www.plosbiology.org 2736 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

significantly different from the other control conditions

(Figure 2). Hence the TMS-induced effect is specific to those
conditions in which the initial state of the arm is dynamically
changing, when its true state must be estimated,

Cerebellar TMS Causes Initial Aiming Error

The duration of the cued reach-to-target movement was
about 725 ms (723.5 ms with TMS over the cerebellum, 725.9
ms without), allowing time for an initial error in the onset of
the reach to be corrected during its execution. Hence,
although they are significant, the final errors reported above
may only reflect a small part of the disruption caused by the
TMS. We therefore measured the angular deviation in the
initiation of the reach-to-target movement. Individual trials
started from different positions (Figure 1B), so we measured
for each trial the angular difference between two lines—one
joining target position to the hand position at the start of the
reach-to-target, and one joining the starting hand position to
its position at maximum velocity—and we compared this
angle within participants across TMS versus non-TMS trials
(see Methods). Because the hand was travelling in a predom-
inantly rightwards direction at cue onset (with the average
speed in that direction 760% greater than upwards, and
490% greater than forwards), we expect the errors to be most
prominent in azimuth angle (Figure 1B)
For stimulation over the cerebellum, the 5.138 (60.488 SEM)
difference in azimuth was highly significant (Figure 4A, one-
Figure 4. TMS-Induced Difference in Mean Azimuth (A) and Elevation sample t-test, t(32)¼10.58, p , 0.0001). The azimuth angle
Angles (B) differences between groups were also highly significant [one-
Each bar is the group mean difference for TMS versus non-TMS trials (þ1 way ANOVA (F(4,78) ¼ 5.43, p ¼ 0.001; post-hoc LSD t-test
SEM); see Figure 2. (A) Positive azimuth angles are defined as clockwise comparisons showed the cerebellar stimulation condition
rotations in the frontal plane (see Figure 1B and 1D). (B) Negative differed from all others (p , 0.013); the other four control
elevation angles are defined as clockwise rotations in the sagittal plane
(Figure 1C and 1E). conditions were not significantly separable from each other (p
doi:10.1371/journal.pbio.0050316.g004 . 0.45)]. In all five conditions, the changes in elevation angle
between stimulated and normal trials were not significantly
other conditions. The larger datasets would be less affected by different from each other [Figure 4B, one-way ANOVA
(F(4,78) ¼ 1.11, p ¼ 0.36; post-hoc LSD t-test comparisons
random sampling.
showed no significant differences (p . 0.22), except between
TMS Effect Is Exposed by Dynamic State Change cerebellar and parietal stimulation, which approached
TMS took place during the reaction time between the go significance (p ¼ 0.06, uncorrected)]. Hence TMS over the
cue and the start of the reach-to-target movement, while the cerebellum induced a significant change in the initial
hand was being actively moved towards the right. Thus, the direction of the targeted reach, which was partly but not
TMS-induced error for the cerebellar stimulation condition fully corrected by the end of the movement (Figure 3A).
is, we hypothesize, due to the disruption of the state Testing the differences in the training datasets, where TMS
estimation process within the ipsilateral cerebellum and was distant from the head and thus ineffective, we found that
would affect the estimation during the current rightwards the mean azimuth angle differences between the two sets of
movement. If true, then the effect should not be seen if the trials was 0.38, which is approximately 6% of the difference
recorded for cerebellar TMS, and not significantly different
need for state estimation was minimized. We therefore
from zero (one-sample t-tests, p ¼ 0.7, n ¼ 18).
compared TMS stimulation at the same cerebellar location
but with the participant holding their arm stationary at the Direction Specificity
moment of cue onset. In this control condition, the starting The direction of the initial pointing error (a clockwise
button was shifted 20 cm laterally, to be coincident with the deviation relative to non-TMS trials) suggests that the
mean start position of the hand in other conditions, and the reaching movement towards the target was inaccurately
participant was instructed to lift the finger from the start planned. We distinguish two possible reasons. One is that
button but to then remain stationary until the auditory go there could be a direction-specific effect due to mislocation
cue. Hence the starting position was known and static prior of the arm at the initiation of the targeted reach, during the
to cue onset. A reach-to-target from this fixed position would rightwards movement from the start key. If the reach-to-
not require renewed state-estimation because the state was target movement was planned based on out-of-date informa-
constant and up to date throughout the reaction time period. tion, i.e., on an estimated start position leftwards of the actual
End point errors were significantly lower in this condition position of the hand, then the movement direction would be
than with cerebellar stimulation (0.92 cm, n ¼ 9) and were not rotated clockwise. To test this hypothesis more directly, we

PLoS Biology | www.plosbiology.org 2737 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

significant correlation between change in azimuth angle

and change in x error during leftwards movement condition
(p ¼ 0.061, Bonferroni adjusted); this is not the axis in which
changes in azimuth angle would be most prominent, given the
near-vertical plane of movements (Figure 1).
Thus we interpret this as further evidence that the initial
direction was inaccurate and that some but not all of this
error was corrected during the reach. We find no evidence
that the final position was mislocated and that the initial
angles were altered to reach this final location.

Estimating the Internal Error in Hand State

To estimate the hand state used to plan the reach-to-target
action, we could backtrack along the mean trajectory of non-
TMS trials, participant by participant, to find a point at which
the angle towards the maximum velocity position was equal
to the mean angular deviation seen for that participant in
TMS trials. In other words, by assuming that the angular error
in aiming to the target was due to a failure of the cerebellar
state estimation of the hand position, during the initial slow
movement, we found that the prior position of the hand—
before the go cue—at which the angular difference between
Figure 5. Group Mean Trajectories for TMS Trials (Red) and Non-TMS start and maximum velocity points would be the same as was
Trials (Blue) Applied over the Cerebellum (n ¼ 13) found between TMS and non-TMS trials. This position was
Solid lines indicate stimulation during initial rightwards movement; 3.97 cm (6 1.01 SEM) leftwards of the actual reach-to-target
dotted lines show stimulation during initial leftwards movement. The
deviation between between TMS and non-TMS trajectories at the start of start position (black dot, Figure 3A). Converting from
the reach towards the final target is reversed between the two positional differences along their mean path between TMS
conditions, while final errors are similar. The insert at top right is the and non-TMS trials, and given the mean hand speed at the
terminal portion of the trajectories, rotated into the frontal plane. This
emphasises the greater overshoot in the z-axis for rightwards TMS trials
cue onset, measured subject-by-subject, the distance of 3.97
(red solid lines) compared to leftwards TMS trials (red dotted lines), cm suggests that the reach-to-target was planned based on the
which mainly overshot in depth (x-axis). hand’s position 138 ms (619 ms SEM) previously. This ‘‘state-
doi:10.1371/journal.pbio.0050316.g005 estimation interval’’ is significantly greater than zero (one-
sample t-test, t(18) ¼ 7.33, p , 0.0001) and is also greater than
include one additional group of participants, who were tested the change in reaction time caused by the TMS (one-tailed
with TMS over the cerebellum during movements made from paired-sample t-test, t(18) ¼ 2.011, p ¼ 0.030; see ‘‘nonspecific
a far-right position, such that the arm at the go cue was, on effects’’ below). For the group tested making initially left-
average, at the same position, but was moving leftwards wards arm movement, the ‘‘state-estimation interval’’ was
(Figure 5). This group showed end-point error amplitudes almost exactly the same, 134 ms (617 ms SEM). This
(mean 1.88 cm, n ¼ 13) inseparable from the original prediction interval cannot be calculated for the stationary-
cerebellar group (1.71 cm, n ¼ 32), and as before, significantly start condition, because backtracking on a stationary trajec-
greater than all the other control conditions (Figure 2, one- tory is not possible.
way ANOVA, F(5,69) ¼ 3.57, p ¼ 0.006, post-hoc t-tests p , Other Nonspecific Stimulation Effects
0.005). The mean azimuth angle was 2.848, in other words, The velocity profiles of the reach-to-target movements
rotated counterclockwise, and hence statistically different were significantly altered by cerebellar TMS (Figure 6A–6C).
from all other conditions (Figure 4A, p , 0.025); the mean The effect was to reduce the reaction time for the rapid
elevation angle did not differ from other conditions (Figure reach-to-target movement by about 80 ms and to increase the
4B, p . 0.19). peak velocity by about 15%. For the main condition of
The second possibility is that the TMS caused the interest, with TMS over the lateral cerebellum during
participants to mislocate the target, and they were therefore ongoing movement, the mean reaction time for TMS versus
accurately reaching to the wrong location. We can dismiss non-TMS trials was 172.5 ms versus 265.9 ms. TMS-induced
this with the stationary start position data. In this case, there startle effects have been reported previously [29] because of
was very little azimuth error (1.748, not significantly different the noise and cutaneous stimulation. TMS over the cerebel-
from all other control conditions), and the end point errors lum using the large double-cone coils does cause noticeable
were actually the smallest observed. In addition, we corre- auditory stimulation, as well as cutaneous and muscular
lated the directional errors in azimuth and elevation for both stimulation. Very similar effects on the movement profiles
groups (cerebellar TMS during rightwards and leftwards were seen in a control group (Figure 6D) tested with TMS
movements) against the end-point errors in the x-, y-, and z- stimulation over the right ear (n ¼ 4) or with startling auditory
axis (Figure 1). We hypothesize that if TMS caused target white noise bursts played over headphones (n ¼ 7), confirming
mislocation, then the initial angular deviations would be that the reduction in reaction time and increase in peak
correlated with the final positional errors. However, we found velocity is likely to be due to a startle response. For this startle
no evidence to support this: there was only one near- control group (n ¼ 11), the mean reaction time was 175.2 ms

PLoS Biology | www.plosbiology.org 2738 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

Figure 6. Group Speed Profiles for TMS Trials (Red) and Non-TMS Trials (Blue)
Each panel shows the group average speed profile (61SEM) for 800 ms after cue onset. The time of TMS stimulation is indicated by the three arrows.
(A–D) Startle effects: TMS over cerebellum during rightward movement (A: n ¼ 32); during leftward movement (B: n ¼ 13), or with a stationary start
position (C: note zero intial velocity; n ¼ 11) leads to a reduced reaction time and increased peak speed very similar to that induced by startle
stimulation (D: with TMS over the ear, n ¼ 4, or with sound stimuli, n ¼ 7). (E–G) Control stimulation sites with TMS over the ipsilateral neck (E: n ¼ 10),
contralateral hand area of motor cortex (F: n ¼ 20) or contralateral posterior parietal cortex (G: n ¼ 12); the startle effects are smaller with slighter
reduction of reaction time and increase of peak speed.
doi:10.1371/journal.pbio.0050316.g006

for startle trials, and 260.7 ms for nonstartle trials. Consid- Discussion
erably weaker effects were seen for TMS stimulation over the
neck (Figure 6E), the motor cortex (Figure 6F), or posterior We have shown that a brief train of TMS over the lateral
parietal cortex (Figure 6G). The reduced reaction time seen cerebellum, applied during the reaction time to initiate a
for motor cortical TMS has also seen in other TMS experi- rapid reaching movement towards a remembered target,
ments [30,31] and is attributed to an alerting but nonstartling resulted in a directional deviation of the reaching movement
effect of the TMS stimulus. and in increased positional error. We suggest that the TMS
Time to maximum velocity after the onset of the reach-to- temporarily blocked the contribution of the lateral cerebel-
target movement was similar (247 ms versus 258 ms for lum to state estimation. As a result, the reaching movements
cerebellar TMS versus non-TMS trials and 227 ms versus 250 were planned based on the residual, out-of-date knowledge of
ms for startle versus nonstartle trials). Overall movement the previous state of the arm. Thus, we propose that during
duration was nonsignificantly shorter in the TMS or startle rightwards arm movements the arm was estimated to be
trials. For the cerebellar group, mean duration was 723 ms for further leftwards than its true position, while during left-
TMS trials and 726 ms for non-TMS trials. For the startle wards movement it was estimated to be further rightwards.
group, mean duration was 693 ms versus 726 ms. Thus, there These errors lead to clockwise and counterclockwise direc-
was a subtle increase in the deceleration phase of the tional deviation of the initial movement towards the target.
movement. When the arm was stationary at movement onset and hence

PLoS Biology | www.plosbiology.org 2739 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

the state estimate was unchanging, then the direction of the ent behavioural strategies. They found that functional
reaching movements was largely unaffected by the stimula- activation changes in the lateral cerebellum were better
tion. explained by state estimation than by timing [45], consistent
Our results therefore provide further evidence for a with the present results.
contribution by the cerebellum to state estimation and, by Our estimates of the angular deviations imply that the
inference, to forward modelling the sensory consequences of rapid reaching movements were planned on information that
action [12,22,32,33]. These forward model predictions, com- is about 138 ms out of date, an interval that is in the
bined with independent sensory information from the appropriate order of magnitude of sensory reafference
periphery, provide an optimal estimate of the current state [2,3,5]. From this result, we would predict that cerebellar
of the arm. However, there is considerable debate about the pathology would lead to movement control based on the
role of the cerebellum in motor control [14,20,34,35] and arm’s state about 138 ms less advanced than actual. This
other functions, such as timing, motor learning, predictive would result in direction- and speed-specific deviations in the
control, and inverse dynamic modelling, have all been initial segment of any rapid movement, especially those made
proposed. We believe the present data are best understood during ongoing action, as we have shown here. It would also
in the context of state estimation. First, predictive control lead to hypermetria, as the state estimate would be disturbed
[36,37] is very closely related to state estimation, because the throughout the reaching movement, rather than just at its
consequences of motor commands must be predicted in initiation. Thus, even the final stages of reaching to a target
order to update and control subsequent actions. Hence state would be affected more obviously than we have seen in our
estimation is a subset of predictive control [1]. Next, the experiments, where, we suspect, the TMS-induced disruption
cerebellum appears important in timing, especially during of state estimate was brief and may have largely recovered by
discrete motor tasks [38,39]. The timing component of our the end of the reaching action. The consequences of
task is minimal as movements were initiated after a random experimentally delaying visual feedback, which effectively
interval, eliminating a timing strategy and participants were makes the state estimate inaccurate with respect to actual
instructed to reach as fast as possible after the go cue. feedback, are similar to that of cerebellar inactivation [2,46].
Furthermore, the target was static, and so the directional Delayed state estimation would therefore explain loss of
differences in reaching behaviour in TMS versus non-TMS coordination and ataxia.
trials are difficult to justify by timing alone. The state estimation process is likely to be iterative [8], with
Another important postulated function is in inverse the current state being updated by an optimised weighting of
dynamic modelling, with a functional role for the cerebellum afferent proprioceptive and visual information and by
in generating motor commands rather than in predicting efferent motor commands. An iterative calculation is
their outcome [7,33,40]. Again, the pattern of results we have optimal, because previous state estimates, even if inaccurate,
shown is not easily fit by this hypothesis. In particular, the provide an additional source of information to be used in the
reach-to-target from a stationary position was not different new estimate. This is particularly true for physiological
from control conditions, whereas disruption of an inverse systems in which the state cannot change instantly; there
model would affect all movement for which the model was must be a strong correlation between previous state and new
used. These data also argue against an explanation that the states. We cannot tell from the present results to whether the
TMS impaired control of interjoint dynamics. Failure to 138 ms interval reflects ‘‘freezing’’ of the state estimate, until
compensate for interjoint dynamics, which has been pro- the cerebellum generates a new estimate, or whether it
posed as an explanation for cerebellar ataxia [27], might reflects the fall-back use of out-of-date proprioceptive and
contribute to our results, as limb dynamics would be different visual information. Possible methods to address this question
for the movements made from left and right. However, the would involve using longer trains of TMS, to stretch the time
initial movements were relatively slow (mean velocity 27.9 cm/ that it was perturbed, or to use repetitive TMS to induce a
s, Figure 6) and so these dynamic effects would be small in temporary ‘‘virtual lesion’’, coupled with adaptation to
comparison to their effects during the much faster reach-to- delayed visual feedback.
target action. Indeed, the reach-to-target made from the However another important issue is whether the state
static position was as rapid (peak velocities of about 100 cm/s, estimate is localized entirely within the cerebellum or is
Figure 6C) as in other conditions and, we assume, was fast distributed across this and other areas. One obvious
enough to expose weak interjoint coordination. Its trajectory candidate, given its well-documented role in spatial repre-
was not different from the control conditions. But whether sentations, is the posterior parietal cortex [47–49]. It has also
these effects contribute more subtly to our overall results will been proposed as a locus of the state estimate [16,17,50] and
require further work, perhaps using faster baseline move- has been recently implicated in the sense of agency and the
ments so that both the dynamic effects and the positional mental representation of [51,52]; agency is also dependent on
misestimation are larger. forward modelling [53]. Our data argue against the possibility
It is also thought that state estimation is required to that the state estimations are generated exclusively by the
minimize timing differences between effectors during coor- parietal cortex, because of the disruption caused by cer-
dinated actions [41,42]. Loss of coordination is one of the ebellar TMS, but it seems plausible that both areas are
cardinal symptoms of cerebellar dysfunction [43] and is involved. One possibility is that the parietal cortex maintains
evident as gross ataxia as well as in more subtle measures such a body representation or body schema [50,54] that is updated
as the failure to coordinate grip and lift forces during object during movements [16]. We would argue that this update is
manipulation [23,24,44]. Recently, Deidrichsen et al. tested calculated by the cerebellum.
coordination of a reach and a button press and were able to Anatomical connections to the cerebellum are consistent
separate experimentally time-dependent and state-depend- with this, as it receives powerful projections from posterior

PLoS Biology | www.plosbiology.org 2740 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

parietal cortex [55], which may hold a representation of the cortex at a site at which we have previously caused disruption
current state estimate [16], as well as from cortical motor of visually guided action [70], and at which TMS is known to
areas, sending an efferent copy of descending commands [56]. affect cerebellar–cerebral projections leading to measurable
It also receives visual and proprioceptive afferents, although changes in motor cortical excitation in the contralateral hand
in our task, visual feedback was blocked during the action. area [69]. One important control condition was therefore to
The output of the cerebellar processing, which we propose test the same TMS protocol applied directly to the contrala-
constitutes an estimation of the change in the motor state teral motor cortex, to rule out indirect effects of the
caused by the efferent signals, may then return to posterior cerebellar stimulation at this remote site. Motor cortical
parietal cortex [57–60] to update its representation, or be TMS did raise terminal errors lead to some directional
directed to motor areas to contribute to the control of the deviation of the reaching movement, but both of these effects
actions [56]. In this framework, one might expect that TMS of were of a significantly smaller magnitude than those seen
posterior parietal cortex would also disrupt reach-to-target after cerebellar stimulation (Figures 2 and 3).
actions in this task. Targeting the superior parietal cortex At the same time, the cerebellar TMS stimulation caused a
using the P3 electroencephalogram electrode coordinates did noticeable change in movement kinematics, with a significant
not cause significant effects. However, this negative result reduction in movement onset latency and an increase in peak
should be taken with caution, because we may have missed a velocity. Similar but weaker effects were also generated by
critical locus within the posterior parietal cortex (PPC) stimulation over the neck, at a site 3 cm below the cerebellar
responsible for maintaining the state representation. stimulation site. We reproduced the cerebellar effect on
The PPC also contributes to the representation of target reaction times using a startling stimulus that did not involve
positions [61], and hence it is important to distinguish loss-of- functional TMS, either by using one wing of the double-cone
state estimation from errors in localizing the remembered coil placed over the participants’ right ear to induce the noise
target. This might be a result of TMS-perturbed input to the and possible auditory nerve stimulation caused by the TMS
PPC, or it may be possibly due to a cerebellar role in target stimulation over the cerebellum or neck, or by playing loud
localization. However, we saw no evidence for mislocalization white noise bursts through headphones without any active
of the target in any of our conditions. In particular, we found TMS. Thus TMS aimed at the lateral cerebellum can startle
direction-specific changes in initial movement direction with the participant and lead to changes in the velocity profile of
cerebellar TMS that were uncorrelated with changes in end the movement, regardless of its effect on the cerebellum.
position of the reach-to-target movement. Moreover, these However, even though these control conditions could induce
effects were significantly different from the errors caused by a similar magnitude shift in reaction time and increase in
cerebellar TMS when the hand was initially stationary, peak velocity, they induced neither the terminal errors nor
although one might expect any effect of target mislocaliza- the initial directional errors that were caused by cerebellar
tion to be common across all these three conditions. There is TMS. This confirms that the initial directional deviation and
limited published evidence of a role for the cerebellum in the final positional errors were not a result of the startle
localization of a visual target [62], and several opposing effect. Furthermore, we also tested cerebellar stimulation
results [63–65]. Thus, it seems unlikely that the changes in from a static starting position, a condition that minimizes the
direction were a sign of movements planned towards a need for a dynamic update of the state estimation. Again, we
perturbed position, saw change in movement onset and velocity attributable to
However, comparing the TMS data from the stationary startle, but we saw no directional deviation or terminal error.
start condition to the two active movement conditions does It is also possible that TMS applied over the lateral
raise another concern, because it is well known that the TMS cerebellum could cause movement errors due to direct
thresholds in motor cortex are lower during active movement stimulation of muscles in the neck, which might lead to
than during rest. We did not test thresholds for activation shoulder or upper arm deviation or cause arm movement by
over the cerebellum; however, we note that the arm was not at stimulation of the brachial plexus [69]. We discounted both
rest in this static position but was actively held in the air just possibilities by testing TMS stimulation over the neck at a site
above the start key; this is a motor task in which the more likely to activate the brachial plexus, and that
cerebellum is actively engaged [66–68]. Others have used stimulation generated visible twitches in the neck muscles
single-shock stimulation levels of 55% with the same double- but without inducing the directional or terminal errors. TMS
cone coil and have seen brief changes in excitability of the at the level used (45% of machine output) is unlikely to cause
contralateral motor cortex that are consistent with activation cortico-spinal stimulation [69,71]. Another control involved
of the cerebellar cortex [69]; we have seen the same effects on measuring the total deviation of the hand that was held static
muscle-evoked potential (MEP) amplitude (unpublished over the start position, without any active reaching task,
results) using triplets of 20-Hz pulses as used in the present during TMS of the cerebellum. This would expose any
experiments, even at stimulation levels of as low as 35% and involuntary hand motion induced either by the TMS,
with the arm genuinely at rest. So although it is possible that including activation of the cortico-spinal collaterals [71], or
our TMS protocol was less effective during the stationary by its startling effect. On TMS trials (three participants), the
condition, we do not think this likely to have influenced these index finger was briefly deviated laterally by less than 5 mm,
results. and within 200 ms had returned to within 1 mm of its initial
position, within the normal reaction time period (266 ms).
Control Experiments and Other Considerations Hence our results are unlikely to be due to TMS-induced
The TMS pulse train is expected to lead to a temporary peripheral effects.
disruption of the neural processing in underlying target A final methodological consideration is that the signifi-
tissue. We targeted the hand area of the ipsilateral cerebellar cantly reduced reaction times seen after cerebellar TMS

PLoS Biology | www.plosbiology.org 2741 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

mean that the reach-to-target starts from a position near to positioned against the scalp, and another 60 trials collected, with
the mean hand position at which the angular difference active TMS on half the trials.
Nine participants were also tested in a condition is which the start
between TMS and non-TMS trials is equal (Figure 3A). In key was moved 20 cm laterally, to the average position at which the
other words, one could argue that the TMS has merely shifted reach to the target started (Figure 1B). Subjects were instructed to lift
the mean start position leftwards in accordance with the the index finger off the start key but to remain stationary until the go
cue signalled the rapid reach to the target. All other aspects of the
reduction in reaction time, and has not affected the internal task remained the same.
state estimate of the hand. However, the interval of 138 ms TMS. Repetitive stimulation was delivered as three biphasic pulses
estimated from comparing directional errors on TMS versus triggered at 20 Hz (50 ms) by the experimental control computer, at
45% of machine output, using a Magstim Rapid (Magstim Co.). For
non-TMS trials is significantly greater than the change in stimulation of the lateral cerebellum (n ¼ 32), a 90-mm radius double-
reaction time (93 ms). Furthermore, reduction in reaction cone coil was centred 3 cm lateral and 1 cm below the inion [69,70]. In
time alone should not necessarily cause a directional error. If this position, one wing of the coil normally overlaps the participant’s
right ear. Ear plugs were provided.
the reach-to-target movements on TMS trials were planned To test the effects of TMS noise and its possible stimulation of the
using an accurate state estimate, then their initial direction right ear, the orientation of the double cone coil was reversed in four
should be towards the target, despite their reduced latency. participants so that one coil surrounded the right ear while the other
was approximately normal to the scalp. Biphasic stimulation was set
This result was clearly seen for the startle trials (Figure 3B), in at 45% of the machine output. These data were combined with that
which the reaction times were advanced by 85 ms, but the of a group of seven participants in which the TMS trigger pulses were
initial direction was unchanged. Thus the startling stimulus used to trigger brief white noise bursts (100 dB, 20-ms duration),
played through binaural headphones. This white noise was suffi-
does not affect the movement direction, whereas cerebellar ciently loud to evoke observable reflexive blinks in all participants,
TMS does, and this dissociates the effects of reduction in while remaining within safety limits. Comparison of the data from the
reaction time from the loss of state estimation of the hand. two groups (ear TMS versus auditory stimulation) revealed no
In conclusion, we suggest that these results indicate that the significant differences, and the two datasets were combined.
For stimulation of the neck (n ¼ 11) a flat, 70-mm radius figure-of-
lateral cerebellum is responsible for estimating the true state eight coil was used, with the coil centre 3 cm below the site used for
of the peripheral motor system over a short time interval. We cerebellar stimulation (3 cm lateral and 4 cm below the inion).
assume this estimation is based on forward modelling of the Stimulator output was set at 45% of machine output.
For stimulation of the motor cortex (n ¼ 20) the flat, 70-mm radius
expected consequences of outgoing motor commands and figure-of-eight coil was positioned at a site where an observable
that the updated estimation is sent from the cerebellum to twitch of the right first dorsal interoseus muscle was seen. Stimulator
cerebral areas responsible for planning and controlling the output was set at the resting threshold. For posterior parietal cortex,
all participants (n ¼ 12) were also tested with M1 stimulation and the
reaching action. These experiments do not tell us how the same stimulator intensity was used. The P3 electrode position was
cerebellum generates these signals, whether the TMS protocol measured using standard landmarks.
has any influence on cerebellar learning, or whether state Data analysis. Index finger trajectories were analysed in Matlab
version R2007a. A Polhemus Fastrak receiver was taped above the
estimates are topographically organized in the cerebellar right index finger, and before the experiment began, each participant
cortex or nuclei. Experiments using methods with finer held the index finger stationary in the position of the virtual target,
spatial resolution than TMS will be needed to address these under full vision. The recorded marker position was then taken as the
target position in all subsequent analysis, accounting for the 1–1.5 cm
important questions. positional offset of the marker from the index finger pad. Finger
position was recorded in three axes at 120 Hz; angular rotations of
the hand that would invalidate this positional offset were minimal
Materials and Methods and estimated at less than 1 mm; the relative difference between TMS
Participants. Forty-five right-handed participants (age range 22–48 and non-TMS trials is less than 10% of this (0.1 mm).
y, 13 male) received TMS, after providing informed written consent, TMS artefacts. TMS magnetic pulses can generate a significant
and with approval from the Central Office for Research Ethics one-sample (8 ms) artefact in the Polhemus motion tracking data,
Committees. Two of these were authors of this article: RCM and which uses magnetic field technology. These artefacts were detected
LODC. Seven participants (age range 22–50, 6 male) were tested with in the first-differenced time series data and removed by interpolation
auditory stimulation, including the authors RCM and JS. across neighbouring data points.
Tasks. Participants sat at a table with their head supported by a To assess the impact of these artefacts, we recorded the apparent
marker position of a static marker placed at the average start position
chin rest and wearing Plato LCD goggles (Translucent Tech). A TMS
of the reach-to-target movement, with the TMS coil placed in
coil was held in position using a Magic-Arm (Adaptivation). The
approximate similar position as when testing a participant. Artefact
position of the right index finger was recorded using a Polhemus removal was successful and the apparent residual motion of the
Fastrak at 120 Hz. Trials were timed by a computer running under marker was under 0.1 cm or 1 cm/s. The duration of the artefact was
DOS. Each trial began with the index finger depressing a start key on also restricted mainly to within the typical reaction time, so any
the table top in front of the right shoulder. Cued by a set of three residual error did not affect analysis of the reach trajectory. Testing
rising tones at 500-ms intervals, the participant was required to with the marker attached to a participant’s finger held stationary
release the start key on the third tone, and to begin to move the right while TMS was applied to the cerebellum (three participants, 20 trials
hand towards the right side. The Plato goggles were switched to each) showed that stimulation of the cerebellum caused minor finger
opaque as soon as the start key was released. Early or late release of motion that was recovered within 200 ms of TMS termination.
the key led to the trial being aborted. At a uniform random time 500– The cleaned positional data were then low-pass filtered (8th order
1,500 ms after the key was released, the onset of a fourth continuous zero phase 7.5 Hz Butterworth filter). Low-pass filtered index finger
tone cued the participant to make a rapid forwards and upwards trajectories were differentiated to velocity, and the tangential speed
reach to place the index finger on the position of a virtual target was averaged across all TMS (n ¼ 30) and non-TMS trials (n ¼ 30) per
image, reflecting a 1-cm target in a mirror. The target was subject for each condition. The subject mean velocities were then
approximately 28 cm above and 15 cm in front of the start key. averaged across the subject group.
One second after this final go cue, the Plato goggles were switched to Mean speed and jerk (second derivative of speed) profiles were
transparent, allowing terminal vision of the static finger and virtual examined before and after the removal of the TMS-artefacts in the
target. The subject then returned to the start key at their own pace. Polhemus data to confirm that the artefact removal was effective. All
On each session, participants were given 60 practice trials, on a trials (TMS and non-TMS) were processed and filtered identically.
random 50% of which a series of three TMS clicks were heard, at 50, Detection of terminal and directional error. The following steps
100, and 150 ms after the onset of the go cue. During training, the coil were taken for analysis of each trial. First, the time-point of
was held about 1 m from the head; training data were only recorded maximum velocity was detected between the go cue and the end of
for 18 participants. Immediately after the practise, the TMS coil was recording (open circles, Figure 1B and 1C). Termination of the reach-

PLoS Biology | www.plosbiology.org 2742 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

Estimation of predictive interval. To estimate the positional offset

that corresponded to the directional error measured, the angle
between the start position and the position of maximum velocity was
calculated for the mean trajectory of all non-TMS trials (n ¼ 30) for
each subject in each condition. Trajectories were spatially re-sampled
before averaging (e.g., Figure 1D and 1E). Then, by iteratively
recalculating this angle for each data position before the start
position, we found the first position at which the angular difference
exceeded the mean angular difference between TMS and non-TMS
trials (black dots, Figure 3). The distance along the mean trajectory
between this position and the start position was found. To estimate
the time interval that corresponded to this spatial offset, the mean
velocity of the hand was found at cue onset, for each subject. Dividing
the estimated offset, subject-by-subject by the mean velocity, we
estimated the time interval of the state estimation.
To test the sensitivity of this analysis to the arbitrary choice of the
point of maximum velocity as a reference position, we repeated the
above analysis choosing instead five time points in 50-ms steps from
50–250 ms after the start of the reach-to-target movement; the
Figure 7. Sensitivity of the Calculation of Estimation Interval on the maximum velocity was normally reached at about 250 ms (Figure 6).
Reference Points Chosen to Measure Movement Deviation The first 50 ms estimate was significantly different from all others
(Figure 7, Bonferroni adjusted p , 0.001, paired t-tests); the other
The five data points are the mean estimation interval (6 SEM, n ¼ 32) four estimates did not significantly differ (Bonferroni adjusted p .
calculated at fixed times after movement onset. The dashed line is the 0.3), and did not differ from the estimate based on the maximal
mean estimate (61 SEM) calculated from the point of maximum velocity
velocity (dashed lines, Figure 7). Hence, we are satisfied that the
in each trial.
results we present here are largely insensitive to the precise reference
doi:10.1371/journal.pbio.0050316.g007
point chosen within the reach-to-target trajectory at which we assess
movement direction, and may somewhat underestimate the magni-
to-target was then taken as the time point at which movement tude of the time interval. Only when this reference position is close to
velocity fell below 5% of the maximum (filled circles and trial the initiation point of the movement (50 ms, or 6 data sample) was
number, Figure 1B and 1C). the estimate lower than that calculate from the point of maximum
Reach-to-target movement onset was detected as the point of velocity. Figure 7 also makes clear that the highest estimate of the
maximum curvature between the go cue and the point of maximum interval (161 6 11.9 ms) was found using a reference point 150 ms
velocity (asterisks, Figure 1B and 1C). To find this, each trajectory into the movement, and this may then indicate that this (150–160 ms)
from go cue to termination was spatially re-sampled to 100 uniformly is indeed the best estimate; additional experimental work will be
spaced points, the rate of change of these spatial positions then required to get independence evidence of this, however.
recorded as curvature, and the maximum curvature spatial position
found. The time point of original data sample the within the original
time series closest to this spatial position was then recorded as the Acknowledgments
time point at which the reach movement initiated.
End point error was measured as the Cartesian distance of the Author contributions. RCM designed the experiments, analysed the
finger from the target. Directional errors were measured as the data, and wrote the paper. LODC, OC, and JS performed the
azimuth or elevation differenced between lines joining the start point experiments.
and target, versus the start point and maximum velocity point for Funding. This work was supported by the Wellcome Trust and the
each trial. Mean angular differences in azimuth and elevation Alfred Benson Foundation.
between all TMS (n ¼ 30) and all non-TMS trials (n ¼ 30) were Competing interests. The authors have declared that no competing
calculated per subject for each experimental session. interests exist.

References 14. Paulin MG (2005) Evolution of the cerebellum as a neuronal machine for
1. Wolpert DM, Flanagan JR (2001) Motor prediction. Curr Biol 11: R729– Bayesian state estimation. J Neural Eng 2: S219–S234.
R732. 15. Paulin MG, Hoffman LF, Assad C (2001) A model of cerebellar
2. Miall RC, Weir DJ, Stein JF (1985) Visuomotor tracking with delayed visual computations for dynamical state estimation. Auton Robots 11: 279–284.
feedback. Neurosci 16: 511–520. 16. Wolpert DM, Goodbody SJ, Husain M (1998) Maintaining internal
3. Saunders JA, Knill DC (2003) Humans use continuous visual feedback from representations: the role of the human superior parietal lobe. Nat Neurosci
the hand to control fast reaching movements. Exp Brain Res 152: 341–352. 1: 529–533.
4. Desmurget M, Grafton S (2000) Forward modeling allows feedback control 17. Desmurget M, Epstein CM, Turner RS, Prablanc C, Alexander GE, et al.
for fast reaching movements. Trends Cogn Sci 4: 423–431. (1999) Role of the posterior parietal cortex in updating reaching move-
5. Saunders JA, Knill DC (2005) Humans use continuous visual feedback from ments to a visual target. Nat Neurosci 2: 563–567.
the hand to control both the direction and distance of pointing 18. Buneo CA, Andersen RA (2006) The posterior parietal cortex: Sensor-
movements. Exp Brain Res 162: 458–473. imotor interface for the planning and online control of visually guided
6. Vaziri S, Diedrichsen J, Shadmehr R (2006) Why does the brain predict movements. Neuropsychologia 44: 2594–2606
sensory consequences of oculomotor commands? Optimal integration of 19. Desmurget M, Grea H, Grethe JS, Prablanc C, Alexander GE, et al. (2001)
the predicted and the actual sensory feedback. J Neurosci 26: 4188–4197. Functional anatomy of nonvisual feedback loops during reaching: a
7. Wolpert DM, Miall RC, Kawato M (1998) Internal models in the cerebellum. positron emission tomography study. J Neurosci 21: 2919–2928.
Trends Cogn Sci 2: 338–347. 20. Ito M (2002) Historical review of the significance of the cerebellum and the
8. Wolpert DM, Ghahramani Z (2000) Computational principles of movement role of Purkinje cells in motor learning. Ann N Y Acad Sci 978: 273–288.
neuroscience. Nat Neurosci 3: 1212–1217. 21. Miall RC, Jenkinson EW (2005) Functional imaging of changes in cerebellar
9. Wolpert DM, Ghahramani Z, Jordan MI (1995) An internal model for activity related to learning during a novel eye-hand tracking task. Exp
sensorimotor control. Science 269: 1880–1882. Brain Res 166: 170–183.
10. Ariff G, Donchin O, Nanayakkara T, Shadmehr R (2002) A real-time state 22. Blakemore SJ, Frith CD, Wolpert DM (2001) The cerebellum is involved in
predictor in motor control: study of saccadic eye movements during predicting the sensory consequences of action. Neuroreport 12: 1879–1884.
unseen reaching movements. J Neurosci 22: 7721–7729. 23. Muller F, Dichgans J (1994) Dyscoordination of pinch and lift forces during
11. Miall RC, Wolpert DM (1996) Forward models for physiological motor grasp in patients with cerebellar lesions. Exp Brain Res 101: 485–492.
control. Neural Netw 9: 1265–1279. 24. Nowak DA, Hermsdorfer J, Marquardt C, Fuchs HH (2002) Grip and load
12. Miall RC, Weir DJ, Wolpert DM, Stein JF (1993) Is the cerebellum a Smith force coupling during discrete vertical arm movements with a grasped
Predictor? J Motor Behav 25: 203–216. object in cerebellar atrophy. Exp Brain Res 145: 28–39.
13. Paulin MG (1989) A Kalman filter theory of the cerebellum. In: Arbib MA, 25. Nowak DA, Timmann D, Hermsdorfer J (2007) Dexterity in cerebellar
Amari S, editors. Dynamic interations in neural networks: Models and data. agenesis. Neuropsychol 45: 696–703.
New York: Springer-Verlag. pp. 239–260. 26. Sanguineti V, Morasso PG, Baratto L, Brichetto G, Luigi MG, et al. (2003)

PLoS Biology | www.plosbiology.org 2743 November 2007 | Volume 5 | Issue 11 | e316

 State Estimation in the Cerebellum

Cerebellar ataxia: Quantitative assessment and cybernetic interpretation. 50. Sirigu A, Duhamel JR, Cohen L, Pillon B, Dubois B, et al. (1996) The mental
Hum Mov Sci 22: 189–205. representation of hand movements after parietal cortex damage. Science
27. Bastian AJ, Martin TA, Keating JG, Thach WT (1996) Cerebellar ataxia: 273: 1564–1568.
Abnormal control of interaction torques across multiple joints. J Neuro- 51. MacDonald PA, Paus T (2003) The role of parietal cortex in awareness of
physiol 76: 492–509. self-generated movements: a transcranial magnetic stimulation study.
28. Herwig U, Satrapi P, Schonfeldt-Lecuona C (2003) Using the international Cereb Cortex 13: 962–967.
10–20 EEG system for positioning of transcranial magnetic stimulation. 52. Sirigu A, Daprati E, Pradat-Diehl P, Franck N, Jeannerod M (1999)
Brain Topogr 16: 95–99. Perception of self-generated movement following left parietal lesion. Brain
29. Leocani L, Cohen LG, Wassermann EM, Ikoma K, Hallett M (2000) Human 122: 1867–1874.
corticospinal excitability evaluated with transcranial magnetic stimulation 53. Frith CD, Blakemore SJ, Wolpert DM (2000) Abnormalities in the awareness
during different reaction time paradigms. Brain 123: 1161–1173. and control of action. Philos Trans R Soc Lond (Biol) 355: 1771–1788.
30. Sawaki L, Okita T, Fujiwara M, Mizuno K (1999) Specific and non-specific 54. Critchley M (1953) Disorders of the body image. In: Critchley M, editors.
effects of transcranial magnetic stimulation on simple and go/no-go The parietal lobe. New York: Hafner Press. pp. 225–255.
reaction time. Exp Brain Res 127: 402–408. 55. Stein JF, Glickstein M (1992) The role of the cerebellum in the visual
31. Pascual-Leone A, Valls-Sole J, Wassermann EM, Brasil-Neto J, Cohen LG, et guidance of movement. Physiol Rev 72: 967–1017.
al. (1992) Effects of focal transcranial magnetic stimulation on simple 56. Kelly RM, Strick PL (2003) Cerebellar loops with motor cortex and
reaction time to acoustic, visual and somatosensory stimuli. Brain 115: prefrontal cortex of a nonhuman primate. J Neurosci 23: 8432–8444.
1045–1059. 57. Amino Y, Kyuhou S, Matsuzaki R, Gemba H (2001) Cerebello-thalamo-
32. Ito M (1970) Neurophysiological aspects of the cerebellar motor control cortical projections to the posterior parietal cortex in the macaque
system. Int J Neurol 7: 162–176. monkey. Neurosci Lettr 309: 29–32.
33. Kawato M, Gomi H (1992) A computational model of four regions of the 58. Schmahmann JD, Pandya DN (1990) Anatomical investigation of projec-
cerebellum based on feedback-error-learning. Biol Cybern 68: 95–103. tions from thalamus to posterior parietal cortex in the rhesus monkey: a
34. Nixon PD (2003) The role of the cerebellum in preparing responses to WGA-HRP and fluorescent tracer study. J Comp Neurol 295: 299–326.
predictable sensory events. Cerebellum 2: 114–122. 59. Clower DM, West RA, Lynch JC, Strick PL (2001) The inferior parietal
35. Ohyama T, Nores WL, Murphy M, Mauk MD (2003) What the cerebellum lobule is the target of output from the superior colliculus, hippocampus,
computes. Trends Neurosci 26: 222–227. and cerebellum. J Neurosci 21: 6283–6291.
36. Bastian AJ (2006) Learning to predict the future: the cerebellum adapts 60. Clower DM, Dum RP, Strick PL (2005) Basal ganglia and cerebellar inputs
feedforward movement control. Curr Opin Neurobiol 16: 645–649. to ‘AIP’. Cereb Cortex 15: 913–920.
37. Vilis T, Hore J (1980) Central neural mechanisms contributing to cerebellar 61. Colby CL, Goldberg ME (1999) Space and attention in parietal cortex.
tremor produced by limb perterbations. J Neurophysiol 43: 279–291. Annu Rev Neurosci 22: 319–49.
38. Ivry RB (1996) The representation of temporal information in perception 62. Kitazawa S, Kimura M, Yin P-B (1998) Cerebellar complex spikes encode
and motor control. Curr Opin Neurobiol 6: 851–857. both destinations and errors in arm movements. Nature 392: 494–497.
39. Spencer RM, Ivry RB, Zelaznik HN (2005) Role of the cerebellum in 63. Nixon PD, Passingham RE (1999) The cerebellum and cognition: cerebellar
movements: control of timing or movement transitions? Exp Brain Res 161: lesions do not impair spatial working memory or visual associative learning
383–396. in monkeys. Eur J Neurosci 11: 4070–4080.
40. Kawato M (1999) Internal models for motor control and trajectory 64. Golla H, Thier P, Haarmeier T (2005) Disturbed overt but normal covert
planning. Curr Opin Neurobiol 9: 718–727. shifts of attention in adult cerebellar patients. Brain 128: 1525–1535
41. Miall RC (1998) The cerebellum, predictive control and motor coordina- 65. Diedrichsen J, Hashambhoy Y, Rane T, Shadmehr R (2005) Neural
tion. Novartis Found Symp 218: 272–290. correlates of reach errors. J Neurosci 25: 9919–9931.
42. Miall RC, Reckess GZ, Imamizu H (2001) The cerebellum coordinates eye 66. Vaillancourt DE, Thulborn KR, Corcos DM (2003) Neural basis for the
and hand tracking movements. Nat Neurosci 4: 638–644. processes that underlie visually guided and internally guided force control
43. Holmes G (1939) The cerebellum of man. Brain 62: 1–30. in humans. J Neurophysiol 90: 3330–3340.
44. Monzee J, Drew T, Smith AM (2004) Effects of muscimol inactivation of the 67. Suminski AJ, Rao SM, Mosier KM, Scheidt RA (2007) Neural and
cerebellar nuclei on precision grip. J Neurophysiol 91: 1240–1249. electromyographic correlates of wrist posture control. J Neurophysiol 97:
45. Diedrichsen J, Criscimagna-Hemminger SE, Shadmehr R (2007) Dissociat- 1527–1545.
ing timing and coordination as functions of the cerebellum. J Neurosci 27: 68. Monzee J, Smith AM (2004) Responses of cerebellar interpositus neurons to
6291–6301. predictable perturbations applied to an object held in a precision grip. J
46. Miall RC, Weir DJ, Stein JF (1987) Visuo-motor tracking during reversible Neurophysiol 91: 1230–1239.
inactivation of the cerebellum. Exp Brain Res 65: 455–464. 69. Werhahn KJ, Taylor J, Ridding M, Meyer BU, Rothwell JC (1996) Effect of
47. Buneo CA, Andersen RA (2006) The posterior parietal cortex: sensorimo- transcranial magnetic stimulation over the cerebellum on the excitability
tor interface for the planning and online control of visually guided of human motor cortex. Electroenceph Clin Neurophysiol 101: 58–66.
movements. Neuropsychologia 44: 2594–2606. 70. Miall RC, Christensen LO (2004) The effect of rTMS over the cerebellum in
48. Pellijeff A, Bonilha L, Morgan PS, McKenzie K, Jackson SR (2006) Parietal normal human volunteers on peg-board movement performance. Neurosci
updating of limb posture: an event-related fMRI study. Neuropsychologia Lett 371: 185–189.
44: 2685–2690. 71. Lai M, Baker MR, Fisher KM, Baker SN (2005) Inhibition of motor cortex
49. Blakemore SJ, Sirigu A (2003) Action prediction in the cerebellum and in following magnetic stimulation over the contralateral occiput may have a
the parietal lobe. Exp Brain Res 153: 239–245. corticospinal origin. Soc Neurosci 179.14.

PLoS Biology | www.plosbiology.org 2744 November 2007 | Volume 5 | Issue 11 | e316