Abstract

It has been suggested that the cortico-striatal system might play a crucial role in learning behavioural plans of action. We have tested this hypothesis by studying the dynamics of functional coupling among the neural elements of cortico-striatal circuitry. Human cerebral activity was measured with functional magnetic resonance imaging (fMRI) during the learning of an associative visuomotor task. Structural equation modelling of regional fMRI time-series was used to characterize learning-related changes in effective connectivity. We report that learning to associate visual instructions with motor responses significantly altered cortico-striatal functional couplings. Specific learning-related increases of effective connectivity were found in temporo-striatal and fronto-striatal circuits. Connectivity among portions of the frontal cortex decreased as a function of learning. Temporo-frontal and parieto-frontal couplings were not altered during learning. We infer that novel visuomotor associations are established through the enhancement of specific cortico-striatal circuits, rather than through the alteration of direct temporo-frontal or parieto-frontal connectivity.

Introduction

Primates can rely on a rich behavioural repertoire because of their ability to combine sensory stimuli with motor responses according to associative rules (Wise and Murray, 2000). The flexibility endowed with the arbitrary combination of stimuli and responses can be compared with the stereotypical performance of spatially or temporally congruent sensorimotor associations, such as reaching for a spatial target or a blink in anticipation of a conditioned stimulus. Accordingly, these arbitrary context-dependent sensorimotor associations need to be learned by combining stimulus–response mappings with their outcomes.

Learning processes have been considered as an emergent (or aggregate) property of a neural network, rather than an effect of local variations of neuronal properties alone (McIntosh, 2000; Martin et al., 2000). Therefore, learning must be accounted for, not just in terms of synaptic plasticity, but also in terms of the information flow across specific cerebral networks. Recently, cortico-striatal loops have emerged as especially relevant, not only for control of ongoing movements (DeLong et al., 1984), but also for learning to solve sensorimotor problems (Canavan et al., 1989; Passingham et al., 1998; Jog et al., 1999; Toni and Passingham, 1999). It has been argued that the striatum can combine cortical representations of sensory contexts and motor acts with mesencephalic reward signals, thus conveying behavioural context into cortical ensembles as a function of learning (Dominey et al., 1995; Houk and Wise, 1995; Dominey and Boussaoud, 1997; Graybiel, 1998; Hikosaka et al., 1999; Schultz, 2000). In particular, Hikosaka (Hikosaka, 1993) has detailed how the caudate could contribute to the learning of motor sequences (i.e. by selecting a specific response in the context of a previous movement), while Passingham (Passingham, 1993) has suggested a more general involvement of the striatum in sensorimotor learning (i.e. by selecting a specific response given a specific sensory context).

However, while electrophysiological, lesion and imaging studies have confirmed the pivotal role of cortico-striatal activity in the gradual learning of stimulus-to-response mapping (Canavan et al., 1989; Tremblay et al., 1998; Toni and Passingham, 1999; Murray et al., 2000; Toni et al., 2001), the learning-related dynamics of functional coupling among the neural elements of such circuits have not been investigated. Furthermore, previous studies have not directly compared, at a system level, the relevance of cortico-cortical pathways and cortico-striatal loops in learning sensorimotor associations.

Structural equation modelling can reveal the spatio-temporal distribution of learning-related changes in local interactions between brain regions (Buchel and Friston, 1997). The power of this approach is clearly dependent on a plausible and constrained model of relationships between the elements of the network under analysis. Neuroimaging provides the possibility of building these constraints on direct measurements of neural activity (Buchel and Friston, 2000) and human neuroanatomy, although the lack of a detailed knowledge of anatomical connectivity in the human brain makes it necessary to rely on extrapolations from non-human primates, hints from human neurological reports, or post-mortem studies (Rademacher et al., 2001).

Here, we have applied structural equation modelling to an anatomically grounded (Passingham et al., 1998, 2000) and functionally characterized (Toni and Passingham, 1999; Toni et al., 2001) cerebral network, in order to test the hypothesis that cortico-striatal interactions play a critical role in learning-specific and arbitrary visuomotor associations (Passingham et al., 1998). Functional magnetic resonance imaging (fMRI) was used to measure human brain activity during performance of two tasks requiring the association of visual patterns with motor responses. Both tasks were learned by trial and error, either before (visuomotor control, VMC) or during (visuomotor learning, VML) the scanning session. In a previous paper we have detailed the behavioural and neural correlates of learning such a task, i.e. learning to make decisions on the basis of associative rules (Toni et al., 2001). Those results indicated that both performance and regional neural effects measured during the visuomotor learning task smoothly converged onto the values of the steady-state control condition. The analysis performed in Toni et al. (Toni et al., 2001) revealed the temporal dynamic of activity of different brain areas as a function of learning. However, that approach could not expose the anatomical route through which visual instructions affect the motor system. Here we report results from a novel and independent analysis of the same data set. This analysis focuses on regionally specific modulations of effective connectivity (Friston, 1994) as a function of learning visuomotor associations.

Materials and Methods

Task and Imaging

Details of the experimental procedures have previously been described (Toni et al., 2001). In brief, 600 fMRI images sensitive to the blood oxygenation level dependent (BOLD) contrast (T2*-weighted echo-planar sequence, TE/TR 40 ms/4.7 s, 48 transverse slices, voxel size 3 3 3 mm) were acquired for each of seven subjects during the alternated performance of a VML and a closely matched VMC, intermixed with a baseline period (Fig. 1B). In VML, the subjects had to learn, during the scanning session, the correct association between four visual patterns and the appropriate finger movements (Fig. 1A). The learning occurred by trial and error. In VMC, the subjects had to retrieve, during the scanning session, a previously learned association between another set of patterns and the finger movements (Fig. 1C). The patterns were presented for 150 ms every 1570 ± 100 ms over blocks of 28.3 s. The random jitter prevented the subjects from learning a precise rhythm of response during the experiment. On each trial, immediately after the movement, a visual feedback stimulus (green or red filled squares on a black background) was presented for 50 ms. This stimulus informed the subjects whether the movement was correct (green square) or not (red square). The stimulus– response associations were kept constant across the experiment. Each condition was performed in blocks of 18 successive trials (28.3 s, equivalent to six whole-brain images; Fig. 1A,C). A scanning session consisted of 100 blocks, i.e. 25 blocks for each test condition and 50 baseline blocks (Fig. 1B).

The VML task was practised for three blocks and the VMC task for 93 blocks during a training session, outside the scanner. In the VML task, the patterns associated with the movement differed between training and scanning sessions, although both were built from the same prototypical structures. This allowed the subjects to become familiar with the task requirements, while avoiding learning the specific visuomotor associations of the VML task performed during the scanning session. The subjects were given two further blocks of training in the scanner, just before the beginning of the scanning session, with the same stimuli used during the training session outside the scanner.

Data Set

The data were analysed with SPM99 (www.fil.ion.ucl.ac.uk/spm). A general linear model considered the voxel-wise effects of tasks (main effect of task: VMLM, VMCM) and the changes in these effects across scans (task × time interactions: VMLT, VMCT). Task × time interactions were modelled using a set of polynomial basis functions (1st and 2nd order). The main effects and task × time interactions represented orthogonal covariates of a multiple regression analysis. A statistical parametric map of the F-statistic (SPM{F}) for the group data was generated for the general linear model described above. Voxels considered in the structural equation model were selected from this SPM{F}. We report the results of a fixed-effect group analysis. The sensitivity of this type of analysis is higher than the sensitivity achieved by a random-effect analysis, thus allowing for the detection of subtle effects (Friston et al., 1999). However, the inferences drawn from this analysis are about the presence of an effect in these subjects during these scanning sessions.

Structural Equation Modelling

Structural equation modelling evaluates the effects of an experimental manipulation by minimizing the difference between the observed covariance matrix and that implied by a pre-specified structural (i.e. anatomical) model (Maruyama, 1998). As emphasized in the Introduction, the power of this technique depends heavily on the biological and statistical plausibility of the underlying model. In neuroimaging, the model selection process is constrained (and thus facilitated) by the necessity to comply with known neuroanatomy. In this study, we have taken further advantage of these neuroanatomical constraints: we have exploited the spatial information derived from an independent analysis of the regional distribution of learning-related activity in this particular data set (Toni et al., 2001). This procedure has allowed us to target our anatomical model onto the precise anatomical features of our sample. Our model is illustrated in Figure 2A,B. It consists of 10 volumes of interest (VOIs). Six of these VOIs were centred on stereotactic coordinates revealed by an independent analysis of the same data set (Toni et al., 2001). Four VOIs were included in the model on the basis of their expected involvement in task performance, either on the basis of lesion data — dorsal premotor cortex (Passingham, 1985) — or anatomical plausibility (left and right primary visual cortex, left motor cortex). Each of these four VOIs was centred on the stereotactic coordinates of the local SPM{F} maximum detected in the relevant anatomical region (i.e. left and right calcarine fissure for primary visual cortex; dorsal precentral gyrus for dorsal premotor cortex; dorsal bank of the central sulcus for primary motor cortex). Each of the 10 VOIs was defined as a sphere of radius 7 mm (corresponding to the FWHM of the SPM{F}), centred on the coordinates reported in Figure 2B, including all voxels that exceeded the statistical threshold of P < 0.001 (SPM{F}, uncorrected for multiple comparisons), masked by the group average of segmented grey-matter mean T2* images. The latter procedures ensured that only grey-matter voxels contributed to each VOI. Each VOI was represented by a time-series given by the first principal component of its adjusted BOLD signal (Buchel and Friston, 1997). This procedure ensured that each of these time-series accounted for most of the variance expressed by the supra-threshold grey-matter voxels of the respective VOI. The connectivity matrix was defined on the basis of human lesion data and studies on fibre tracts from other primates. Details for each connection are provided in Figure 2A and Table 1.

Learning-related effects on task-related covariance between two VOIs were modelled using interaction terms (Buchel and Friston, 1997). Only linear temporal effects were considered, to ensure sensitivity and computational stability. The model included the task × time interaction and VOI × task × time interactions (Fig. 2C). For each pair of connected VOIs, the modulation of connectivity induced by regionally specific learning-related effects (VOI × task × time interaction; Fig. 2Ciii) was assessed over and above the direct influences of the source areas onto the target area (Fig. 2Ci) and the time-dependent changes within regions induced by the task (task × time interaction; Fig. 2Cii). The task × time covariates were convolved by a canonical haemodynamic response function (Friston et al., 1998) and multiplied by the activity of each VOI to form the interaction or moderator variables. Therefore, the stationary couplings were assessed from time-series that included learning, control and baseline conditions, within VOIs showing significant time-differential changes in BOLD signal between the visuomotor learning and control epochs. Conversely, the regionally specific learning-related changes of connectivity (VOI × task × time interactions), being the product of a task × time covariate times VOI time-series, were assessed from time-series reflecting time-varying covariance during the visuomotor learning epochs.

The physiological (i.e. overall inter-regional connectivity; Fig. 2Ci), cognitive (i.e. learning-related changes in connectivity within regions = task × time interaction; Fig. 2Cii) and psycho-physiological interaction variables (i.e. regionally and learning-specific changes in connectivity = VOI × task × time interactions; Fig. 2Ciii) were normalized to a mean of zero and unit standard deviation.

Structural equation modelling was performed by means of a previously published implementation (Buchel and Friston, 1997) within the framework of the SPM99 imaging analysis package using an iterative maximum likelihood algorithm (Higham, 1993) to estimate covariances that best predicted the observed variance–covariance structure of the empirical data. Residual variance was modelled using auto-connections at each node in the model. Statistical inferences about the path coefficients were based on the comparison of a free model (in which all parameters were unconstrained) with constrained models in which a given anatomical or modulatory connection was forced to be zero. The difference in goodness of fit between free and constrained models was expressed as χ2 (with degrees of freedom determined by the number of constraints). Under the null hypothesis that the zeroed connection does not significantly contribute to fitting the model to the observed covariance structure of the data, the free and constrained models do not differ in goodness of fit. The statistical threshold (α) of P < 0.05 was adopted.

Structural equation modelling allowed us to assess path coefficients for every connection in the model (Figs 2A and 4A). These path coefficients describe the relationship between a given pair of VOIs, having taken into account the contribution of the other variables included in the model. In other words, path coefficients are partialized and standardized indexes of correlation among the variables of the model (Maruyama, 1998). Note that the biological relevance of high-order interaction terms, as with learning-dependent effects modulating steady-state, task-related, inter-regional couplings (Fig. 2C), needs to be seen in a hierarchical perspective, rather than in simple terms of absolute effect size. The direction of the interaction term is also relevant. Positive path coefficients represent an increase of the effect of a source area on a target area as learning progresses. An alternative interpretation is that the activity of a source area increases the extent to which learning influences the activity of a target area.

Results

Subjects’ behavioural data indicated that the imaging experiment covered the learning process, from the initial stages at chance level until an almost error-free performance towards the end of the scanning session (Fig. 3A). The subjects’ performance of the control task was virtually error-free. These behavioural results indicate that the VCL and VCM conditions can be considered as legitimate correlates of the learning and retrieval aspects of the visuomotor task.

The anatomical model (Fig. 2A), supported by previous data on the spatial distribution of regional responses with specific learning-related activity (Toni et al., 2001), was further constrained by masking each VOI with the grey-matter distribution of the subjects (Fig. 2B).

There were strong stationary couplings (i.e. steady-state, learning-independent, inter-regional influences; Fig. 4A) between dorsal precentral cortex and the rostral bank of the central sulcus (0.51), between the inferior frontal gyrus and the anterior part of the inferior frontal sulcus (0.73 and 0.69, respectively) and between the middle temporal gyrus and inferior frontal gyrus (0.34). The sizes of these path coefficients are comparable with previous neuroimaging reports on steady-state, inter-regional connectivity (Buchel and Friston, 1997). Weaker couplings were found between the left calcarine fissure and the middle temporal gyrus (0.04) and in the connections of the anterior striatum with the middle temporal gyrus, the inferior frontal gyrus, the anterior part of the inferior frontal sulcus and the intraparietal sulcus (0.01, 0.10, 0.07 and –0.08, respectively). We also considered the time-dependent changes within regions related to task performance (Fig. 4B). The effects associated with these changes were weaker than the stationary couplings, but generally stronger than the regionally specific, learning-related changes of connectivity (Fig. 4).

For each pair of connected VOIs, the regionally specific, learning-related changes of connectivity (VOI × task × time interaction; Fig. 2Ciii) were assessed over and above the direct influences of the source areas onto the target area (Fig. 2Ci) and the time-modulated changes within regions related to task performance (task × time interaction; Fig. 2Cii). The VOI × task × time interactions are higher-order modulatory effects, rather than directly related to the inter-regional connectivity induced by the task. The relative magnitudes of the modulatory and modulated effects need to be seen in the same perspective. Learning the visuomotor associative task significantly changed the coupling between specific sub-sets of regions in our structural model (Fig. 4C). Two learning-related pathways emerged. First, as a function of learning, effective connectivity significantly increased between the calcarine fissure and the middle temporal gyrus (+0.04), from there to the anterior striatum (+0.04) and then onto the dorsal precentral gyrus (+0.03). The afferences from the anterior part of the inferior frontal sulcus to the anterior striatum also showed a positive learning-related effect (+0.05). Second, as a function of learning, effective connectivity significantly decreased along a path leading from the anterior part of the inferior frontal sulcus to the inferior frontal gyrus (–0.04), to the opercular precentral region (–0.01). The other paths of the structural model did not show significant learning-related modulation of connectivity. In particular, there were no changes in effective connectivity centred on the intraparietal VOI.

In order to take into account the variability between the individual subjects of our group, we performed a confirmatory random effect analysis on the seven significant VOI × task × time interactions observed in the main group analysis. The path coefficients extracted from subject-specific time-series were considered as dependent variables of one-sample t-tests to assess the null hypothesis that they had a mean of zero. Each higher-order interaction detected in the fixed effect group analysis was confirmed in the random effect analysis. Comparing subject-specific structural equation modelling revealed significant (P < 0.05) changes in effective connectivity between the caudate and dorsal premotor cortex [t(6) = 1.98]. Strong trends to significance (P = 0.06) were found for the learning-related changes in connectivity between calcarine fissure and middle temporal gyrus [t(6) = 1.78] and between inferior frontal sulcus and caudate nucleus [t(6) = 1.73]. A weak trend to significance (P = 0.1) was found in the random effect analysis applied to the learning-related inter-regional couplings between middle temporal gyrus and the caudate [t(6) = 1.34]. Finally, the subject-specific path coefficients between inferior frontal gyrus and inferior frontal sulcus, and between inferior frontal sulcus and opercular precentral region did not reach statistical significance. Overall, the results of the random effect analyses do not contradict the results of the main fixed effect analyses.

Discussion

In this experiment, we have isolated specific learning-related changes of effective connectivity within a complex cerebral network involved in acquiring novel visuomotor associations. The decomposition of covariance by structural equation modelling analysis ensured that fluctuations in the common inputs to source and target areas of the specific structural model used here could not account for the learning-related modulations of connectivity. Our results showed that fronto-striatal and temporo-striatal connectivity was selectively enhanced as a function of learning (Fig. 4C). Conversely, connectivity between portions of the frontal cortex was reduced as learning progressed. Moreover, direct temporo-frontal or parieto-frontal connectivity was unaffected by the learning challenge, despite high stationary couplings between the activities of these cortical regions (Fig. 4A). In the following sections, we discuss these results in the context of a system-level account of how visual information can affect the motor system.

Cerebral Visual-to-Motor Pathways

Our model allowed for visual information to be relayed to motor regions through several independent cortico-cortical and cortico-striato-cortical pathways (Fig. 2A). The results indicate that establishing and retaining novel associations between visual stimuli and motor responses does not modulate direct corticocortical visuomotor pathways, namely temporo-prefrontal, parieto-prefrontal, or parieto-premotor connections. [Direct temporo-premotor projections have not been considered, given reports of their absence (Seltzer and Pandya, 1989; Distler et al., 1993; Boussaoud et al., 1995).]

Conversely, our data are consistent with the hypotheses of others (Hikosaka, 1993; Passingham, 1993) on the involvement of cortico-striatal circuits during the generation of novel context-dependent visuomotor associations. Our results emphasize the integrative role of the striatum in binding together cortical sensory representation [anterior temporal cortex (Miller et al., 1993)] with action selection [anterior prefrontal cortex (Rowe et al., 2000)] in order to bias motor programs [dorsal premotor cortex (Weinrich and Wise, 1982)]. The concomitant reduction of inter-regional frontal connectivity suggests that the creation of new cortico-striatal ensembles reduces the computational demands of the sensorimotor transformation, allowing prefrontal resources to react to new events. A dynamic account of these findings would suggest that prefrontal regions could be involved in the rapid creation of task-dependent arbitrary associations (Asaad et al., 1998; Wallis et al., 2001), while gradually promoting the interaction of temporal and premotor regions through the striatum in order to consolidate the associative rules (Passingham et al., 1998). Although it is unclear whether there is microstructural convergence of cortical afferences at the striatal level (Selemon and Goldman-Rakic, 1985; Percheron and Filion, 1991), alternative binding mechanisms are also compatible with our system-level analysis (Stern et al., 1998).

These results complement the evidence provided by interference (Passingham, 1985; Petrides, 1985; Canavan et al., 1989; Eacott and Gaffan, 1992; Kurata and Hoffman, 1994; Murray et al., 2000; Wang et al., 2000) and single unit studies (Mitz et al., 1991; Tremblay et al., 1998) on the neural basis of learning visuomotor associations. While those studies have shown the necessary and specific contribution of a series of cortical and subcortical regions to the acquisition of novel conditional responses, here we have assessed how these regions modify their couplings as a function of learning.

Our study complements previous reports addressing the spatial distribution of visuomotor activity in humans (Deiber et al., 1997; Grafton et al., 1998; Toni and Passingham, 1999) or recent studies showing modulations of effective connectivity during sensory associative learning (Buchel and Friston, 1997; Buchel et al., 1999; Fletcher et al., 1999; McIntosh et al., 1999). There are also methodological differences. Here we have assessed modulatory changes of effective connectivity that are not only learning-related, but also regionally specific. This specificity has been achieved by directly contrasting the covariance explained by regionally and learning-related changes in connectivity (VOI × task × time; Fig. 2Ciii), against the covariance associated with the steady-state, inter-regional connectivity (Fig. 2Ci) and the changes within regions driven as a function of time by task performance (Fig. 2Cii). This novel approach has allowed us to test whether specific and directed changes in inter-regional connectivity are related to a specific visuomotor learning process, in the context of a complex and intact human cerebral network.

Effective Connectivity

In this study we have been able to incorporate the specific functional anatomies of our group of subjects within a general model designed to compare the contribution of cortico-striatal and cortico-cortical circuits in learning novel visuomotor associations. Accordingly, the anatomical model we tested maintains the system-level perspective intrinsic in whole-brain imaging. For instance, it is obviously inaccurate to describe cortico-striatal loops without their thalamic relay (Kemp and Powell, 1971). However, building a plausible model requires the weighting of anatomical exactness against computational stability and interpretability (McIntosh and Gonzales-Lima, 1994; Horwitz et al., 2000). Therefore, we have focused on a network which has been repeatedly shown to be essential for learning arbitrary visuomotor associations (Passingham, 1985; Canavan et al., 1989; Eacott and Gaffan, 1992; Hasegawa et al., 1998; Toni and Passingham, 1999; Murray et al., 2000; Toni et al., 2001). While the choice of this particular network constrains the scope of our inferences, it is sufficient for testing our hypothesis on the critical involvement of the striatum in learning specific and arbitrary visuomotor associations (Passingham, 1993; Passingham et al., 1998).

In the present study, all effects were included in a single model, with moderator variables indicating continuous learning-related modulations of connectivity between two areas. In the context of learning studies, this procedure focuses on gradual changes of BOLD signal across repeats of a given condition, rather than on a comparison between ‘early’ and ‘late’ learning stages (Buchel et al., 1999), or ‘high’ and ‘low’ activity in a modulated region (Buchel and Friston, 1997). Furthermore, this approach emphasizes the modulation of experimentally induced, task-related, inter-regional covariances, rather than their absolute values during a specific task (McIntosh and Gonzales-Lima, 1994; Nyberg et al., 1996). Our analysis was not suited to detect temporally non-linear changes of connectivity, for instance changes occurring over just a few minutes of our scanning window. Therefore, it is not possible to exclude the possibility that short-lasting interactions between inferior temporal and ventral prefrontal cortex might occur in the very initial stages of the learning process. The necessity of these temporal-prefrontal interactions has been suggested by lesion studies (Gaffan and Harrison, 1988; Eacott and Gaffan, 1992; Murray et al., 2000), although it is not clear whether the crucial interference occurred over a cortico-cortical or a cortico-striato-cortical pathway.

The regionally specific, learning-related changes in connectivity might appear of modest magnitude with respect to the overall steady-state connectivity (cf. Fig. 4A,C). However, the learning-related changes in connectivity (Fig. 2Ciii) constitute higher-order interactions, affecting the direct task-related inter-regional couplings. The relative magnitudes of the modulatory and modulated effects (Fig. 4) need to be seen in this hierarchical framework.

Segregation and Integration

The present analysis of learning-related modulations of effective connectivity complements a previous study on the same data set concerned with the spatial distribution of regional responses showing specific learning-related activity (Toni et al., 2001). The two studies addressed orthogonal issues, as inter-regional couplings and intra-regional changes of neurovascular activity represent two independent metrics of neural signal (Horwitz et al., 2000). Some of the responses found in these two studies provide an empirical example of such independence: the learning-related changes in the inferior frontal gyrus and in the anterior part of the inferior frontal sulcus showed an increase in BOLD signal and a decrease in their mutual connectivity (Fig. 5). This finding complements earlier work (Buchel et al., 1999), where the opposite pattern of results was found between areas of the dorsal and ventral visual stream in the context of a sensory associative learning task. Another example of dissociations between intra-regional activity and inter-regional connectivity was found in the dorsal precentral cortex. The functional coupling between striatal and dorsal precentral regions increased as a function of learning (Fig. 4C). However, several independent reports failed to find significant learning-related changes of local neurovascular activity in the dorsal precentral region (Deiber et al., 1997; Toni and Passingham, 1999; Toni et al., 2001). The present result further qualifies the involvement of premotor cortex in the learning of visuomotor associations as specifically related to its connectivity with the striatum (Kurata and Hoffman, 1994). Finally, our results converge with other studies based on functional segregation and interference approaches (Rushworth et al., 1997; Toni and Passingham, 1999; Toni et al., 2001) in suggesting a non-essential role of the posterior parietal region in learning to decide between alternative courses of action on the basis of an external context (Passingham and Toni, 2001).

Conclusions

Recent assessments of the functions of the basal ganglia have emphasized their role in learning behavioural plans of action (Hikosaka, 1993; Passingham, 1993; Graybiel, 1995). Here, we have provided evidence for a pivotal role of the striatum in connecting temporal, prefrontal and premotor activities during the learning of novel visuomotor associations.

Table 1

Connectivity matrix — list of references

Figure 1.

Task setup. Stimuli used in the visuomotor learning task (A) and control task (C), the associated finger responses and the feedback stimuli (dark grey = red; light grey = green). The numbers indicate trials within each experimental block. (B) Experimental time-course: black = visuomotor learning task; grey = visuomotor control task; white = baseline. The numbers indicate blocks of trials within each scanning session.

Figure 1.

Task setup. Stimuli used in the visuomotor learning task (A) and control task (C), the associated finger responses and the feedback stimuli (dark grey = red; light grey = green). The numbers indicate trials within each experimental block. (B) Experimental time-course: black = visuomotor learning task; grey = visuomotor control task; white = baseline. The numbers indicate blocks of trials within each scanning session.

Figure 3.

Behavioural data. (A) Differential median percentage number of errors (±SEM). (B) Differential mean response times (±SEM).

Figure 3.

Behavioural data. (A) Differential median percentage number of errors (±SEM). (B) Differential mean response times (±SEM).

Figure 4.

Inter-regional and learning-related connectivity — results. (A) Task-related inter-regional effective connectivity. The paths coefficients are derived from a decomposition of the inter-regional covariance, assessed over the whole fMRI time-series. Path coefficients represent the response of the target area to a unitary change of activity in the source area (in units of standard deviation), whilst the other variables in the model are held constant. (B) Significant (P < 0.05) time-dependent changes of task-related couplings within regions (increases, blue dashed lines; decreases, magenta dashed lines). (C) Significant (P < 0.05) regionally specific, learning-related modulations of inter-regional couplings. Specific cortico-striatal connections showed an increase of the effect of a source area on a target area as learning progresses (green dotted lines); other frontal connections displayed a reduced influence of a source area on the activity of a target area (red dotted lines).

Figure 4.

Inter-regional and learning-related connectivity — results. (A) Task-related inter-regional effective connectivity. The paths coefficients are derived from a decomposition of the inter-regional covariance, assessed over the whole fMRI time-series. Path coefficients represent the response of the target area to a unitary change of activity in the source area (in units of standard deviation), whilst the other variables in the model are held constant. (B) Significant (P < 0.05) time-dependent changes of task-related couplings within regions (increases, blue dashed lines; decreases, magenta dashed lines). (C) Significant (P < 0.05) regionally specific, learning-related modulations of inter-regional couplings. Specific cortico-striatal connections showed an increase of the effect of a source area on a target area as learning progresses (green dotted lines); other frontal connections displayed a reduced influence of a source area on the activity of a target area (red dotted lines).

Figure 5.

Learning-related modulations of regional effects and connectivity. Time-course of learning-related changes in local activity and inter-regional connectivity. The curve in blue represents block-by-block group-averaged normalized signals (median, SE) from the first principal component of a VOI over the inferior frontal gyrus (centre: –52, 40, –4; radius: 7 mm), time-smoothed through a moving average with a Gaussian kernel extending over four blocks. The curve in red represent block-by-block group averaged normalized covariance (median, SE) between the previous VOI and a second VOI over the anterior part of the inferior frontal sulcus (centre: –46, 46, 12; radius: 7 mm). The local signal (in blue) increases as a function of time, whereas the corresponding inter-regional covariance (in red) shows a decrease as a function of time.

Figure 5.

Learning-related modulations of regional effects and connectivity. Time-course of learning-related changes in local activity and inter-regional connectivity. The curve in blue represents block-by-block group-averaged normalized signals (median, SE) from the first principal component of a VOI over the inferior frontal gyrus (centre: –52, 40, –4; radius: 7 mm), time-smoothed through a moving average with a Gaussian kernel extending over four blocks. The curve in red represent block-by-block group averaged normalized covariance (median, SE) between the previous VOI and a second VOI over the anterior part of the inferior frontal sulcus (centre: –46, 46, 12; radius: 7 mm). The local signal (in blue) increases as a function of time, whereas the corresponding inter-regional covariance (in red) shows a decrease as a function of time.

We are very grateful to K. Friston and C. Büchel for providing the SPM-structural equation modelling software. We wish to thank R. Kötter for consultation with CoCoMac (www.cocomac.org). I.T. was supported by the JRAC (National Hospital, Queen Square, London), the Hermann von Helmholtz Gemeinschaft and the F.C. Donders Centre for Cognitive Neuroimaging. J.R. and R.E.P. were supported by the Wellcome Trust. K.E.S. was supported by the Forschungszentrum Juelich. A copy of the Matlab script used to specify the structural equation model tested in this paper is available on request (ivan.toni@fcdonders.kun.nl).

References

Alexander GE, DeLong MR, Strick PL (
1986
) Parallel organization of functionally segregated circuits linking basal ganglia and cortex.
Annu Rev Neurosci
 
9
:
357
–381.
Asaad WF, Rainer G, Miller EK (
1998
) Neural activity in the primate prefrontal cortex during associative learning.
Neuron
 
21
:
1399
–1407.
Barbas H, Pandya DN (
1987
) Architecture and frontal cortical connections of the premotor cortex (area 6) in the rhesus monkey.
J Comp Neurol
 
256
:
211
–228.
Barbas H, Pandya DN (
1989
) Architecture and intrinsic connections of the prefrontal cortex in the rhesus monkey.
J Comp Neurol
 
286
:
353
–375.
Boussaoud D, di Pellegrino G, Wise SP (
1995
) Frontal lobe mechanisms subserving vision-for-action versus vision-for-perception.
Behav Brain Res
 
72
:
1
–15.
Buchel C, Friston KJ (
1997
) Modulation of connectivity in visual pathways by attention: cortical interactions evaluated with structural equation modelling and fMRI.
Cereb Cortex
 
7
:
768
–778.
Buchel C, Friston K (
2000
) Assessing interactions among neuronal systems using functional neuroimaging.
Neural Netw
 
13
:
871
–882.
Buchel C, Coull JT, Friston KJ (
1999
) The predictive value of changes in effective connectivity for human learning.
Science
 
283
:
1538
–1541.
Canavan AG, Nixon PD, Passingham RE (
1989
) Motor learning in monkeys (Macaca fascicularis) with lesions in motor thalamus.
Exp Brain Res
 
77
:
113
–126.
Cavada C, Goldman-Rakic PS (
1989
) Posterior parietal cortex in rhesus monkey: II. Evidence for segregated corticocortical networks linking sensory and limbic areas with the frontal lobe.
J Comp Neurol
 
287
:
422
–445.
Cheng K, Saleem KS, Tanaka K (
1997
) Organization of corticostriatal and corticoamygdalar projections arising from the anterior inferotemporal area TE of the macaque monkey: a Phaseolus vulgaris leucoagglutinin study.
J Neurosci
 
17
:
7902
–7925.
Deiber MP, Wise SP, Honda M, Catalan MJ, Grafman J, Hallett M (
1997
) Frontal and parietal networks for conditional motor learning: a positron emission tomography study.
J Neurophysiol
 
78
:
977
–991.
DeLong MR, Alexander GE, Georgopoulos AP, Crutcher MD, Mitchell SJ, Richardson RT (
1984
) Role of basal ganglia in limb movements.
Hum Neurobiol
 
2
:
235
–244.
Distler C, Boussaoud D, Desimone R, Ungerleider LG (
1993
) Cortical connections of inferior temporal area TEO in macaque monkeys.
J Comp Neurol
 
334
:
125
–150.
Dominey PF, Boussaoud D (
1997
) Encoding behavioral context in recurrent networks of the frontostriatal system: a simulation study.
Brain Res Cogn Brain Res
 
6
:
53
–65.
Dominey P, Arbib MA, Joseph JP (
1995
) A model of corticostriatal plasticity for learning oculomotor associations and sequences.
J Cogn Neurosci
 
7
:
311
–336.
Dum RP, Strick PL (
1997
) Cortical inputs to the digit representations in the primary motor cortex and the dorsal premotor area of the cebus monkey.
Soc Neurosci Abs
 
23
:
502
.
Eacott MJ, Gaffan D (
1992
) Inferotemporal-frontal disconnections: the uncinate fascicle and visual associative learning in monkeys.
Eur J Neurosci
 
4
:
1320
–1332.
Felleman DJ, Van Essen DC (
1991
) Distributed hierarchical processing in the primate cerebral cortex.
Cereb Cortex
 
1
:
1
–47.
Fletcher P, Buchel C, Josephs O, Friston K, Dolan R (
1999
) Learning-related neuronal responses in prefrontal cortex studied with functional neuroimaging.
Cereb Cortex
 
9
:
168
–178.
Friston KJ (
1994
) Functional and effective connectivity in neuroimaging: a synthesis.
Hum Brain Mapp
 
2
:
56
–78.
Friston KJ, Holmes AP, Worsley KJ (
1999
) How many subjects constitute a study?
Neuroimage
 
10
:
1
–5.
Friston KJ, Josephs O, Rees G, Turner R (
1998
) Nonlinear event-related responses in fMRI.
Magn Reson Med
 
39
:
41
–52.
Gaffan D, Harrison S (
1988
) Inferotemporal-frontal disconnection and fornix transection in visuomotor conditional learning by monkeys.
Behav Brain Res
 
31
:
149
–163.
Grafton ST, Fagg AH, Arbib MA (
1998
) Dorsal premotor cortex and conditional movement selection: a PET functional mapping study.
J Neurophysiol
 
79
:
1092
–1097.
Graybiel AM (
1995
) Building action repertoires: memory and learning functions of the basal ganglia.
Curr Opin Neurobiol
 
5
:
733
–741.
Graybiel AM (
1998
) The basal ganglia and chunking of action repertoires.
Neurobiol Learn Mem
 
70
:
119
–136.
Hasegawa I, Fukushima T, Ihara T, Miyashita Y (
1998
) Callosal window between prefrontal cortices: cognitive interaction to retrieve long-term memory.
Science
 
281
:
814
–818.
Higham NJ (
1993
) Optimization by direct search in matrix computations.
SIAM J Matrix Anal Applic
 
14
:
317
–333.
Hikosaka O (1993) Role of basal ganglia in motor learning: a hypothesis. In: Brain mechanisms of perception and memory (Ono M, Squire LR, Raichle ME, Perrett DI, Fukuda H, eds), pp. 497–513. Oxford: Oxford University Press.
Hikosaka O, Nakahara H, Rand MK, Sakai K, Lu X, Nakamura K, Miyachi S, Doya K (
1999
) Parallel neural networks for learning sequential procedures.
Trends Neurosci
 
22
:
464
–471.
Horwitz B, Friston KJ, Taylor JG (
2000
) Neural modeling and functional brain imaging: an overview.
Neural Netw
 
13
:
829
–846.
Houk JC, Wise SP (
1995
) Distributed modular architectures linking basal ganglia, cerebellum, and cerebral cortex: their role in planning and controlling action.
Cereb Cortex
 
5
:
95
–110.
Ilinsky IA, Kultas-Ilinsky K (
1987
) Sagittal cytoarchitectonic maps of the Macaca mulatta thalamus with a revised nomenclature of the motor-related nuclei validated by observations on their connectivity.
J Comp Neurol
 
262
:
331
–364.
Jog MS, Kubota Y, Connolly CI, Hillegaart V, Graybiel AM (
1999
) Building neural representations of habits.
Science
 
286
:
1745
–1749.
Kemp JM, Powell TP (
1971
) The connexions of the striatum and globus pallidus: synthesis and speculation.
Philos Trans R Soc Lond B Biol Sci
 
262
:
441
–457.
Kurata K, Hoffman DS (
1994
) Differential effects of muscimol microinjection into dorsal and ventral aspects of the premotor cortex of monkeys.
J Neurophysiol
 
71
:
1151
–1164.
McGuire PK, Bates JF, Goldman-Rakic PS (
1991
) Interhemispheric integration: I. Symmetry and convergence of the corticocortical connections of the left and the right principal sulcus (PS) and the left and the right supplementary motor area (SMA) in the rhesus monkey.
Cereb Cortex
 
1
:
390
–407.
McIntosh AR (
2000
) Towards a network theory of cognition.
Neural Netw
 
13
:
861
–870.
McIntosh AR, Gonzales-Lima F (
1994
) Structural equation modelling and its application to network analysis in functional brain imaging.
Hum Brain Mapp
 
2
:
2
–22.
McIntosh AR, Rajah MN, Lobaugh NJ (
1999
) Interactions of prefrontal cortex in relation to awareness in sensory learning.
Science
 
284
:
1531
–1533 [published erratum, Science (1999) 285:1493].
Martin SJ, Grimwood PD, Morris RG (
2000
) Synaptic plasticity and memory: an evaluation of the hypothesis.
Annu Rev Neurosci
 
23
:
649
–711.
Maruyama GM (1998) Basics of structural equation modeling. London: Sage.
Matelli M, Camarda R, Glickstein M, Rizzolatti G (
1986
) Afferent and efferent projections of the inferior area 6 in the macaque monkey.
J Comp Neurol
 
251
:
281
–298.
Matelli M, Govoni P, Galletti C, Kutz DF, Luppino G (
1998
) Superior area 6 afferents from the superior parietal lobule in the macaque monkey.
J Comp Neurol
 
402
:
327
–352.
Miller EK, Li L, Desimone R (
1993
) Activity of neurons in anterior inferior temporal cortex during a short-term memory task.
J Neurosci
 
13
:
1460
–1478.
Mitz AR, Godschalk M, Wise SP (
1991
) Learning-dependent neuronal activity in the premotor cortex: activity during the acquisition of conditional motor associations.
J Neurosci
 
11
:
1855
–1872.
Murray EA, Bussey TJ, Wise SP (
2000
) Role of prefrontal cortex in a network for arbitrary visuomotor mapping.
Exp Brain Res
 
133
:
114
–129.
Nyberg L, McIntosh AR, Cabeza R, Nilsson LG, Houle S, Habib R, Tulving E (
1996
) Network analysis of positron emission tomography regional cerebral blood flow data: ensemble inhibition during episodic memory retrieval.
J Neurosci
 
16
:
3753
–3759.
Pandya DN, Yeterian EH (1998) Comparison of prefrontal architecture and connections. In: The prefrontal cortex (Roberts AC, Robbins TW, Weiskrantz L, eds), pp. 51–55. Oxford: Oxford University Press.
Passingham RE (
1985
) Premotor cortex: sensory cues and movement.
Behav Brain Res
 
18
:
175
–185.
Passingham RE (1993) The frontal lobes and voluntary action. Oxford: Oxford University Press.
Passingham RE, Toni I (
2001
) Contrasting the dorsal and ventral visual systems: guidance of movement versus decision making.
Neuroimage
 
14
:
S125
–S131.
Passingham RE, Toni I, Schluter N, Rushworth MF (
1998
) How do visual instructions influence the motor system?
Novartis Found Symp
 
218
:
129
–141.
Passingham RE, Toni I, Rushworth MF (
2000
) Specialisation within the prefrontal cortex: the ventral prefrontal cortex and associative learning.
Exp Brain Res
 
133
:
103
–113.
Percheron G, Filion M (
1991
) Parallel processing in the basal ganglia: up to a point.
Trends Neurosci
 
14
:
55
–59.
Petrides M (
1985
) Deficits on conditional associative-learning tasks after frontal- and temporal-lobe lesions in man.
Neuropsychologia
 
23
:
601
–614.
Petrides M, Pandya DN (
1999
) Dorsolateral prefrontal cortex: comparative cytoarchitectonic analysis in the human and the macaque brain and corticocortical connection patterns.
Eur J Neurosci
 
11
:
1011
–1036.
Rademacher J, Burgel U, Geyer S, Schormann T, Schleicher A, Freund HJ, Zilles K (
2001
) Variability and asymmetry in the human precentral motor system: a cytoarchitectonic and myeloarchitectonic brain mapping study.
Brain
 
124
:
2232
–2258.
Rowe JB, Toni I, Josephs O, Frackowiak RS, Passingham RE (
2000
) The prefrontal cortex: response selection or maintenance within working memory?
Science
 
288
:
1656
–1660.
Rushworth MF, Nixon PD, Passingham RE (
1997
) Parietal cortex and movement. I. Movement selection and reaching.
Exp Brain Res
 
117
:
292
–310.
Schultz W (
2000
) Multiple reward signals in the brain.
Nat Rev Neurosci
 
1
:
199
–207.
Selemon LD, Goldman-Rakic PS (
1985
) Longitudinal topography and interdigitation of corticostriatal projections in the rhesus monkey.
J Neurosci
 
5
:
776
–794.
Seltzer B, Pandya DN (
1989
) Frontal lobe connections of the superior temporal sulcus in the rhesus monkey.
J Comp Neurol
 
281
:
97
–113.
Stern EA, Jaeger D, Wilson CJ (
1998
) Membrane potential synchrony of simultaneously recorded striatal spiny neurons in vivo.
Nature
 
394
:
475
–478.
Toni I, Passingham RE (
1999
) Prefrontal-basal ganglia pathways are involved in the learning of arbitrary visuomotor associations: a PET study.
Exp Brain Res
 
127
:
19
–32.
Toni I, Ramnani N, Josephs O, Ashburner J, Passingham RE (
2001
) Learning arbitrary visuo-motor associations: temporal dynamic of brain activity.
Neuroimage
 
14
:
1048
–1057.
Tremblay L, Hollerman JR, Schultz W (
1998
) Modifications of reward expectation-related neuronal activity during learning in primate striatum.
J Neurophysiol
 
80
:
964
–977.
Van Hoesen GW, Yeterian EH, Lavizzo-Mourey R (
1981
) Widespread corticostriate projections from temporal cortex of the rhesus monkey.
J Comp Neurol
 
199
:
205
–219.
Wallis JD, Anderson KC, Miller EK (
2001
) Single neurons in prefrontal cortex encode abstract rules.
Nature
 
411
:
953
–956.
Wang M, Zhang H, Li BM (
2000
) Deficit in conditional visuomotor learning by local infusion of bicuculline into the ventral prefrontal cortex in monkeys.
Eur J Neurosci
 
12
:
3787
–3796.
Webster MJ, Bachevalier J, Ungerleider LG (
1994
) Connections of inferior temporal areas TEO and TE with parietal and frontal cortex in macaque monkeys.
Cereb Cortex
 
4
:
470
–483.
Weinrich M, Wise SP (
1982
) The premotor cortex of the monkey.
J Neurosci
 
2
:
1329
–1345.
Wise SP, Murray EA (
2000
) Arbitrary associations between antecedents and actions.
Trends Neurosci
 
23
:
271
–276.
Yeterian EH, Pandya DN (
1993
) Striatal connections of the parietal association cortices in rhesus monkeys.
J Comp Neurol
 
332
:
175
–197.