Tải bản đầy đủ - 0 (trang)


Tải bản đầy đủ - 0trang

type are fully specified by an external cue. However, these inferences are limited in

at least two respects. First, it is possible to isolate preparatory activity by directly

comparing trials with and without a preparatory component, other factors being

equal. In other words, one needs to assume that movement preparation is a standalone cognitive module, indifferent to the selection and execution components of

the sensorimotor process. But response selection appears to be significantly influenced by the possibility of preparing a response before a trigger cue.137,138 We have

already discussed how this issue might have confounded a series of studies on rate

effects in the motor cortex. In the context of motor preparation, it is possible to

overcome this limitation by isolating specific delay-related activity, while accounting

for selection and execution components of the sensorimotor process.139–141 A second

point that deserves to be mentioned concerns the nature of the information processes

implemented by frontal regions during the transformation of sensory stimuli into

motor responses. Although it might be important to define which regions are implicated in movement preparation, neuroimaging studies have usually avoided addressing the crucial question of how a given cerebral region contributes to the preparatory


A few notable exceptions to this consideration come from fMRI studies trying

to investigate the dynamics of the BOLD signal to gather temporal information from

the pattern of hemodynamic responses evoked by a given motor task. The rationale

behind this approach is to extract the sequence of neural events occurring during a

given motor task in order to map different cerebral regions onto different stages of

a given cognitive process. The study of Wildgruber et al.142 was one of the first to

address this issue, in the context of self-generated movements known to engage

mesial motor cortical regions earlier than lateral central regions. Their results showed

consistent temporal precedence of the onset of the BOLD response in a mesial ROI

(putative SMA) as compared to a lateral ROI (putative M1). However, these data

do not allow one to infer that the temporal offset is neural in nature. It might equally

well be the case that mesial and lateral regions have different neurovascular coupling

properties. This potential confound was considered in a follow-up study by Weilke

et al.143 by analyzing responses to two motor tasks, namely self-generated movements

and externally triggered movements. The authors found a temporal shift of the BOLD

response between the rostral portion of SMA and M1 of 2000 msec during the selfgenerated movements compared to only 700 msec during the externally triggered

movements. In an elegant study by Menon et al.,144 the authors used intersubject

variability in reaction times to dissociate neural from vascular delays in the BOLD

responses measured across visual and motor brain regions. By correlating the difference in fMRI response onset of pairs of regions (visual cortex–supplementary

motor area; supplementary motor area–primary motor cortex) with the reaction times

on a subject-by-subject basis, the authors showed that reaction time differences could

be predicted by BOLD delays between SMA and M1, but not between V1 and SMA.

In other words, the authors localized the source of visuomotor processing delays to

the motor portion of the sensorimotor chain bringing visual information to the motor

cortex. However, one could argue that the observations of these reports143,144 crucially

depend on how BOLD delays are measured. In both studies, the authors fitted a

linear regression to the initial uprising portion of the BOLD response. The intercept

Copyright © 2005 CRC Press LLC

of this regressed line with “zero intensity” was taken as the onset point of the BOLD

response. Therefore, this measure of response onset depends crucially on defining

a stable baseline. For primary sensory or motor regions it is conceivable to define

baseline as the absence of sensory stimuli or motor responses, but this criterion

would not be appropriate for higher order cerebral regions. There is a further

difficulty with this approach, namely, how to disentangle changes in response magnitude from changes in response latency. The study by Weilke et al.,143 as well as

other studies using “time-resolved” fMRI,145,146 deals with this issue simply by

scaling BOLD responses of different areas or conditions to the unit range, thus

avoiding the task of effectively accounting for changes in response latency induced

by changes in response magnitude. An elegant alternative approach to this problem

was suggested by Henson et al.147: explicitly estimating response latency via the

ratio of two basis functions used to fit BOLD responses in the General Linear Model;

namely, the inverse of the ratio between a “canonical” hemodynamic response

function148 and its partial derivative with respect to time (temporal derivative).

To summarize, the studies on the contribution of M1 to movement preparation

reviewed here agree in suggesting that this cortical region is mainly involved in the

executive stages of the sensorimotor chain. This role seems to fit into the more general

perspective of the organization of the parieto-frontal system, with parietal areas

involved in evaluating the potential motor significance of sensory stimuli,141,149 frontal

areas involved in preparing movements as a function of their probability,150,151 and

central regions focused on executing the actual movement.114,139


Although the available neuroimaging data on motor imagery and movement preparation might suggest that the contribution of M1 to sensorimotor tasks is limited to

movement performance, one should not neglect that neural responses are dynamic

in nature and vary over time. Accordingly, it could be argued that, during overlearned situations, the contribution of M1 to cognitive aspects of sensorimotor tasks

is reduced to a minimum. However, this scenario might not be true in the context

of learning. For instance, it has been shown that motor cortex contributions to the

performance of a given task appear to change dramatically as a function of learning.152–154 Following 5 weeks of daily practice in the performance of a thumb-finger

opposition sequence, Karni et al.152 reported an increase in the number of task-related

voxels along the precentral gyrus and the anterior bank of the central sulcus. However, there were no differences in the actual signal intensity measured during performance of a trained and an untrained sequence. This result is quite puzzling, given

that the cortical point spread function of vascular signals related to neural activity

has been estimated at around 4 mm,155 i.e., for small voxels (<4 mm) a change in

signal intensity should result in a change in signal extent and vice versa. Furthermore,

the findings of Karni et al.152 appeared to be in conflict with subsequent reports. For

instance, De Weerd et al.156 report a reduction in the number of responsive M1 voxels

following extensive practice in motor sequence learning, while Muller et al.157 report

changes in premotor but not in motor cortex during proficient performance of a

sequence of finger movements. Irrespective of these conflicting findings, it is relevant

Copyright © 2005 CRC Press LLC

to emphasize that, in order to ascribe a crucial role to motor cortex, it is essential

to disentangle neural responses genuinely associated with learning from other timedependent phenomena like habituation, fatigue, and motor adjustments during action

repetitions.158 For instance, both Karni et al.152 and De Weerd et al.156 used sequences

of thumb–finger opposition movements, but it remains unclear whether the motor

components of this motorically complex task can be adequately characterized by

error rate and performance speed alone. Other studies of motor learning have relied

on simpler finger flexions,154,159–162 but this procedure, per se, is obviously not

sufficient to guarantee an appropriate level of control.

For instance, in Toni et al.,161 motor sequence learning was compared to a passive

visual condition, thus preventing a distinction between time-dependent and motor

learning-dependent changes in neural responses. This potential confound was explicitly addressed in Toni et al.162 by comparing brain activity during performance of

two visuomotor tasks, one learned before and the other during the scanning session.

This approach allowed the authors to assess learning-related effects not confounded

by behavioral effects, since the mean reaction times in the two conditions did not

change differentially as a function of time, despite a strong time-dependent decrease

common to both conditions. There were no specific learning-related changes in motor

cortex. This finding was confirmed in other related studies,159,163 although it should

be emphasized that the focus of these papers was on learning visuomotor associations, rather than motor skills. In this latter respect, Ramnani et al.164 have studied

the learning of an extremely well-controlled motor response, namely the eye-blink

reflex. The authors reported specific learning-related increases in the BOLD signal

in the ventral sector of the precentral gyrus, in the region containing a motor

representation of the face.164 This result is in agreement with previous PET studies

concerned with motor skill learning as assessed by the serial reaction time task.154,165

In summary, there appear to be contributions of M1 to motor learning, reflecting

genuine changes in neuronal processing rather than spurious byproducts of changes

in motor output. However, this does not imply that M1 plays a general role in motor

learning, as documented by the studies on the acquisition of novel sensorimotor




In this chapter, we have reviewed some of the findings on M1 function obtained by

functional neuroimaging in humans. We have also reviewed some of the basic

experimental approaches that functional neuroimaging can take: mapping, measuring

stimulus-response functions or context-dependent modulations, and analyzing timeresolved response sequences. Naturally, this chapter is not exhaustive. Resting state

fluctuations, pharmacological manipulations, and analyses of functional or effective

connectivities were not covered or were barely touched upon, though they offer

interesting prospects. Yet, in addition to providing some sort of overview, we hope

this chapter can help achieve a better assessment of the strengths and limitations of

magnetic resonance as a tool in the neurosciences in general and in the study of

Copyright © 2005 CRC Press LLC

motor control in particular. Sensorimotor function is associated with a distributed

neural substrate, and the fact that fMRI readily covers the entire brain is helpful in

this respect. Highly focalized research techniques are hypothesis-driven, at least in

terms of the location they target, and this impairs the potential for new discoveries.

In that sense, even if fuzzy, a picture of activity in the entire brain may help to

generate and then test novel hypotheses, apart from offering plausibility controls for

ongoing studies. Furthermore, the fact that the contrast agent exploited in the BOLD

contrast is endogenous and thus permanently present permits a true neurophysiological recording that can go beyond evoked responses.

However, the spatiotemporal response function, i.e., the dispersion of the fMRI

response in time and space, poses the most relevant limitation for this type of

recording. In other words, fMRI hits hard biological limits, not technical ones, even

though technical difficulties are abundant and not yet always fully mastered. Studies

in the visual system with fMRI have established that dedicated acquisition and

analysis techniques can resolve much smaller functional cortical units than in the

currently available motor studies discussed here.166 This increase in spatial resolution

is of interest because, in contrast to earlier neuroimaging techniques, fMRI experiments readily generate highly significant findings in single subjects.

Many of the topics discussed in the previous sections illustrated that one of the

major shortcomings of functional neuroimaging studies still lies in the uncertainties

of anatomical labeling. Each brain is different, but previous neuroimaging techniques

required normalizing the data into a common standard stereotactic space so as to

perform averaging of voxel-based signals from roughly homologous brain areas

across subjects. These group analyses then had sufficient statistical power and the

advantage of ensuring some degree of generality in terms of volume coverage and

intersubject variability. Yet, the price paid for this procedure was at the level of

anatomical analysis. Even if a spatial normalizing technique incorporates nonlinear

algorithms that warp one gyrification pattern rather well into another, the correspondence of actual brain areas becomes blurred by these procedures, and accordingly

probabilistic atlases are the closest one can get to reality in this setting. In the

previous sections, it has become obvious that such maps can indeed be helpful in

tentatively assigning fMRI responses to certain areas, but often enough, even probabilistic statements leave painful uncertainties as to which areas we are obtaining

effects from.

But what defines an area as charted in an atlas? The set of neuroanatomical

criteria range from cyto- and myeloarchitectonic features to densities and laminar

distributions of receptors and other neurochemical markers.167 In the case of M1,

recent detailed analyses have demonstrated a considerable degree of variability both

between different brains and within individual brains, i.e., between hemispheres.168

Moreover, similar methods have rather recently unveiled the fact that, regarding

Brodmann area 4, we are actually dealing with two architectonically distinct areas

instead of one.169 If form follows function, we must also assume different response

properties of these two areas, and we have discussed some of the evidence from

functional neuroimaging that this may indeed be the case. However, these conclusions were based on relating functional findings from one or several brains to a

database formed from many other and thus different brains. The desideratum at this

Copyright © 2005 CRC Press LLC

Tài liệu bạn tìm kiếm đã sẵn sàng tải về


Tải bản đầy đủ ngay(0 tr)