- Email: [email protected]
Co-ordination within and between verbal and visuospatial working memory: network modulation and anterior frontal recruitment
NeuroImage 20 (2003) 1298 –1308
www.elsevier.com/locate/ynimg
Co-ordination within and between verbal and visuospatial working memory: network modulation and anterior frontal recruitment A. Ku¨bler,a K. Murphy,a J. Kaufman,b E.A. Stein,b,1 and H. Garavana,b,* b
a Department of Psychology and Institute of Neuroscience, Trinity College Dublin, Dublin 2, Ireland Department of Psychiatry and Behavioral Medicine, Medical College of Wisconsin, Milwaukee, WI 53226, USA
Received 31 March 2003; revised 17 June 2003; accepted 30 June 2003
Abstract Attention switching between items being stored and manipulated in working memory (WM) is proposed to be an elementary executive function. Experiment 1 reveals a similar attentional limitation within and between verbal and visuospatial WM and identifies a supramodal switching process required for switching between WM items. By using functional magnetic resonance imaging, Experiment 2 investigated brain activation correlates of parametrically varied attention switching within and between these two WM modalities. Attention switching activation was broadly distributed, was quite similar across the three conditions, and, in almost all areas, increased with increasing switching demand, indicating that attention switching recruits and modulates the entire WM network. Dorsolateral prefrontal cortex was implicated in both within- and between-modality attention switching, but no significant activation was found in ventrolateral areas, supporting dorsal-ventral process models of prefrontal organization. A functional dissociation between anterior frontal and dorsolateral prefrontal cortex was found with the former being more activated when switching attention between modalities was required. The data challenge the notion of an anatomically separate attention switching executive function, but suggest that anterior frontal areas are recruited for the additional demand of coordinating the verbal and visuospatial WM slave systems. © 2003 Elsevier Inc. All rights reserved.
Introduction According to current models of working memory, two separable systems for verbal and visuospatial information processing are controlled by a central executive, which allocates attentional resources according to the task requirements (Cocchini et al., 2002; Cowan, 1993; Shallice and Burgess, 1996). Following Baddeley and Hitch (1974), verbal information is processed within the phonological loop and visuospatial information in the visuospatial sketch pad (Baddeley and Hitch, 1974). Tasks which engage verbal working memory have been shown to activate a widespread network in the brain including dorsolateral prefrontal cortex (DLPFC), Broca’s area, premotor and supplementary motor
* Corresponding author: Fax: ⫹353-1-671-2006. E-mail address: [email protected] (H. Garavan). 1 Current address: National Institute on Drug Abuse-IRP, Neuroimaging Research Branch, 5500 Nathan Shock Drive, Baltimore, MD 21224, USA. 1053-8119/$ – see front matter © 2003 Elsevier Inc. All rights reserved. doi:10.1016/S1053-8119(03)00400-2
area, anterior cingulate, parietal cortex and cerebellum (Jonides et al., 1998). For example, Paulesu and colleagues (Paulesu et al., 1993) localized the phonological store to the left supramarginal gyrus and the articulatory rehearsal process to Broca’s area. As in verbal working memory, tasks addressing visuospatial working memory activate multiple, widely distributed regions in parietal, motor, and prefrontal cortices (Haxby et al., 2000). While the exact role of the prefrontal cortices in working memory remains unclear, a number of distinct models have been proposed. According to one domain-specific model, the prefrontal cortex is functionally subdivided into a dorsolateral region for spatial information and a ventrolateral region for object information (Courtney et al., 1996; Goldman-Rakic, 1996). An alternative model argues instead for domain specificity across hemispheres, wherein a left midfrontal area is specialized for object or verbal (non-spatial) working memory and right inferior and superior frontal areas for spatial working memory (Smith et al., 1996; Ungerleider et al., 1998). Yet a third process-specific model
A. Ku¨bler et al. / NeuroImage 20 (2003) 1298 –1308
proposes that ventral and dorsolateral PFC are engaged for specific working memory processes such as the maintenance, retrieval and on-line manipulation of working memory representations (D’Esposito, et al., 2000; Kessels, et al., 2000; Owen, 1997; Owen et al., 1996; Wagner et al., 2001). It is proposed that ventral prefrontal areas are involved in the maintenance of information in working memory, but if this information has to be manipulated or shielded against distraction, activity increases in DLPFC only. Indeed, the nature of the manipulation-related processes ascribed to DLPFC may be nonmnemonic (Collette et al., 1999; D’Esposito et al., 2000; Postle, et al., 1999). Dorsolateral prefrontal cortex appears critical for executive functioning (Smith et al., 1998), as robust activation has been observed here in many studies that engage a variety of working memory and executive functions (Chein and Fiez, 2001; Collette et al., 1999; Diwadkar et al., 2000; Haxby et al., 2000; Rypma et al., 2002; Salmon et al., 1996). It would be erroneous, however, to conclude that executive functioning is solely restricted to prefrontal areas. Studies addressing executive functioning typically find widespread activation in parietal, premotor, cingulate, temporal, occipital and cerebellar areas (Collette et al., 1999; Le et al., 1998; Owen et al., 1996; Petrides et al., 1993; Raye et al., 2002; Salmon et al., 1996). Isolating specific executive functions within general working memory functioning can prove challenging as executive functions constitute a major and integral part of normal working memory. Following Garavan and colleagues (Garavan, 1998; Garavan et al., 2000), we used a task designed to isolate the allocation of attentional resources within working memory while holding constant on-line storage and rehearsal demands. The task probed executive functions by isolating switches of attention between items residing in working memory. We used a counting task, which requires verbal updating and should therefore engage verbal working memory, a spatial location updating task which should engage the visuospatial sketchpad and a between-modalities task, in which one count and one location had to be updated. Previous research has shown that switching between two counts in working memory results in a sizeable time cost (Garavan, 1998). This time cost persists even after intensive task practice, which indicates that people do not have simultaneous access to all items currently in working memory, consistent with Cowan’s model of a focus of attention within working memory (Cowan, 1988). We propose that switching attention between items in working memory is a basic and elementary executive function. Thus, we hypothesised that we would find the same switching costs as in verbal WM when switching between items in visuospatial WM or when switching between the two WM modalities. A behavioural study tested these hypotheses (Experiment 1) prior to functional imaging. To isolate functional activation associated with attention switching, we chose a parametric manipulation of executive
1299
demands instead of the more common subtraction method. We have previously demonstrated (Garavan et al., 2000) that activation associated with attention switching within verbal working memory was not restricted to prefrontal areas, but included large areas in left parietal, left cerebellar, occipital, temporal, and subcortical areas. With the present study, we attempted to determine if attention switching within visuospatial working memory would activate similar or distinct modality specific cortical areas, and whether this activation would be restricted to the prefrontal lobes. Based upon previous results (Garavan et al., 2000; Halgren et al., 2002; LaBar et al., 1999) and the behavioural data from Experiment 1, we expected activation in a frontal-parietal network in all tasks. As the tasks involved similar on-line manipulation demands (switching between and updating representations within working memory) but did so within distinct modalities (verbal and visuospatial), we were also interested whether our tasks would activate PFC according to a domain- or processing-specific model. Little is known about the dynamics between the different components of working memory. Koechlin and colleagues (Koechlin et al., 1999) demonstrated that, in addition to DLPFC, the frontopolar cortex (BA10) was activated when a major task goal had to be kept in mind while performing subgoals of a task. Therefore, based on a hierarchical model of working memory functioning, one might propose that frontopolar cortex could be activated when attention switching has to be coordinated between verbal and visuospatial working memory.
Experiment 1 Tasks and design Twenty-five subjects (20 female, mean age 21 ⫾ 4.5, age range 17–35) completed the attention switching tasks. All gave informed consent, which was approved by the institutional review board of the Psychology department in Trinity College Dublin. Three switching tasks were employed (Fig. 1). Two addressed attention switching within the verbal or visuospatial modalities. The verbal task required participants to keep a count of how many red and how many blue squares were presented and to report the results at the end of each trial. There were between 12 and 20 stimuli per trial and, following practice, subjects completed 25 trials. Stimulus presentation was self-paced; subjects pressed the spacebar to clear the screen of the current stimulus and present the next. Two event types are noted. A switch occurred when either a red square followed a blue square or vice versa, thereby requiring subjects to switch between the red and blue counts being maintained in working memory. A nonswitch occurred when two red squares or two blue squares were presented in succession thereby requiring subjects to update the same count twice without having to
1300
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 1. Schematic of the three tasks. (A) In the verbal task, blue and red squares had to be counted. (B) In the visuospatial task, the locations of a blue and red dot had to be updated. (C) In the combined task squares had to be counted and the location of the dot had to be updated. Storage and rehearsal demands were equal in all tasks and the manipulation of interest was the switch between the count or location representations.
switch between counts. In total, there were 192 switch and 183 nonswitch events. Comparison of the switch and nonswitch response times provides a measure of the switching cost. The visuospatial working memory locations task, designed to parallel the counting task, required subjects to maintain and update two locations. The locations were one red and one blue dot presented in a 2 ⫻ 2 matrix. At the start of each trial, subjects were presented with the starting locations of the dots (i.e., which cell each dot occupied), which were randomly chosen, and during each trial subjects were presented with a sequence of single arrows. The arrows were either red or blue and subjects were required to move the corresponding dot in the direction of the coloured arrow (e.g., a blue arrow pointing right instructs subjects to move the blue dot one cell to the right). The 2 ⫻ 2 matrix was presented only at the beginning of the trial so subjects had to maintain and update the dot locations in working memory. As in the verbal task, the arrow presentation was self-paced and switching time could be calculated by comparing successive updates of the same dot with successive updates of the two dots as the latter requires a switch between the current locations of the dots. The stimulus train was based on that used for the verbal task. That is, switches and nonswitches between the locations were in identical positions to those in the verbal task’s stimulus train. As above, subjects completed 25 trials following practice. The third task addressed attention switching between the verbal and visuospatial WM modality. In this combined task participants were required to maintain and update one count and one location through presentation of blue squares and red arrows. Other details were the same as for the preceding tasks. Order of the three tasks was counterbalanced across subjects.
Results Subjects performed well on the tasks with an average of 5.4% errors (i.e., in which either a count or location was reported incorrectly at the end of a trial). Only correct trials were included in the subsequent response time analyses from which one subject with incomplete data was dropped. A 3 (TASK: verbal, visuospatial, combined) ⫻ 2 (SWITCH: switch, nonswitch) ANOVA revealed a significant task effect [F(2,46) ⫽ 10.2, p ⱕ 0.0002], a significant switch effect [F(1,23) ⫽ 117.0, p ⱕ 0.0001] and a significant interaction [F(2,46) ⫽ 14.3, p ⱕ 0.0001]. As shown in Fig. 2 and confirmed by t-tests, substantial switching costs were present for each task. An ANOVA comparing the switching costs in the three tasks was significant [F(2.46) ⫽ 14.3, p ⱕ 0.0001] (verbal ⬎ visuospatial ⬎ combined) as were all pairwise comparisons bar visuospatial vs combined [t(23) ⫽ 1.92, p ⱕ 0.07]. In the combined task, switching from the
Fig. 2. Mean reaction time (RT) for switches and nonswitches for all three tasks. Reaction time for switches was significantly slower than for nonswitches in all tasks. Error bars are SEM.
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 3. Top panel shows non-significant correlation between nonswitches in the verbal and the visuospatial task. Switch costs in both tasks (bottom panel) were correlated. Lines show linear trend.
count to the location was not significantly different than switching in the opposite direction [t(24) ⫽ 0.28, p ⱕ 0.78]. Correlational analyses between the verbal and visuospatial tasks revealed that while the nonswitch response times were not related (r ⫽ .25, p ⱕ 0.24) the switching costs in the two tasks were related (r ⫽ .55, p ⱕ 0.006) (see Fig. 3). The switching costs in the within modality tasks were also related to the switching costs in the between modalities combined task (verbal and combined, r ⫽ .42, p ⱕ 0.04; visuospatial and combined, r ⫽ .39, p ⱕ 0.06).
Discussion The results have confirmed an attentional limitation in both verbal and visuospatial working memory consistent with Garavan (1998). Subjects do not have simultaneous and immediate access to more than one working memory item at a time as demonstrated by switching costs when they are required to switch between these items. This limitation was also present when maintaining two items in different working memory modalities as revealed by the sizeable switching costs when switching between the verbal count and visuospatial location. These results suggest that the attentional limitation is a general, supramodal one that tran-
1301
scends specific working memory slave systems. This conclusion is supported by the correlational results. The nonswitch response times of the verbal and visuospatial tasks (i.e., the time involved in the updating operations required of the tasks) were not correlated, providing evidence that different processes were involved in performing the verbal and visuospatial updates. This result is consistent with the substantial body of evidence showing the separateness of the two working memory slave systems (Baddeley and Hitch, 1974; Cocchini et al., 2002; Farmer et al., 1986). However, a significant correlation between the verbal and visuospatial switching costs was observed. These results suggest that while one’s ability to update counts was not related to one’s ability to update locations, one’s ability to switch between counts was related to one’s ability to switch between locations. Similarly, one’s ability to switch between items within each working memory modality was also correlated with one’s ability to switch between items in different modalities. The results of this study suggest that the switching manipulation engages attentional or central executive abilities within working memory that are distinct from modalityspecific abilities, thereby providing a rationale for using the attention switching task as a measure of executive functioning. These behavioural results suggest that the cognitive processes involved in attention switching are quite similar both within and between working memory slave systems and therefore, similar cortical involvement might be anticipated. Experiment 2 was designed to address this neuroanatomical question.
Experiment 2 Tasks and design Sixteen right-handed subjects participated in Experiment 2 (12 female, mean ⫾ SD age: 23.5 ⫾ 3.93, age range 18 –33). All gave informed consent, which was approved by the institutional review board of the Medical College of Wisconsin. The same three attention switching tasks as used in Experiment 1 were modified for imaging to determine the functional neuroanatomy of attention switching within verbal and visuospatial WM and between the two modalities. For imaging purposes, presentation of items was not selfpaced. In the verbal task each stimulus (for the imaging study stimuli were circles rather than squares) was presented for 1400 ms and successive circles were separated by a 100 ms fixation cross, the purpose of which was to clearly delineate successive presentations of the circles. In the visuospatial task, red and blue arrows indicating in which direction the dot should mentally be moved, were presented for 1400 ms and successive arrows were separated by a 100 ms fixation cross. The details for the combined task were the same as for the preceding tasks. To report the final location of a dot, participants used a
1302
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
4-button piano keypad; the keys corresponded to four response options presented on the screen. To report the number of circles, participants were presented with four numbers and had to choose the correct number by pressing the corresponding key of the keypad. Subjects were given 8 sec in which to make their responses. All subjects took part in one session comprising 6 runs with each run containing 9 trials. There were two runs (18 trials) for each of the three tasks. Presentation of tasks was counterbalanced. Trials varied in length (11–16 circles, arrows or circles and arrows) and in switching frequency: The 18 trials were comprised of six “High (H)” (5– 8 switches), six “Medium (M)” (2– 4 switches), and six “Low (L)” (1 switch) switching demand trials. The sequence of trials was HLMLLMLLM for run 1 and HMMHHLHMH for run 2 in all tasks. Trials were preceded by a 4 sec fixation cross. A rest period of 22 sec was provided after every third trial. Rest periods of 22 and 26 sec were included at the start and at the end of each run, respectively. At the end of the rest period a change in the fixation cross signalled the start of the next trial. In total, the experiment lasted approximately 45 minutes. fMRI parameters Nineteen contiguous 7 mm sagittal slices covering the entire brain were collected using a blipped gradient-echo, echo-planar pulse sequence (TE ⫽ 40 ms; TR ⫽ 2000 ms; FOV ⫽ 24 cm; 64 ⫻ 64 matrix; 3.75 ⫻ 3.75 mm in-plane resolution). All scanning was conducted on a 1.5 T GE Signa scanner equipped with a 30.5 cm i.d., three-axis local gradient coil and an endcapped quadrature birdcage radiofrequency head-coil (Wong et al., 1992). Foam padding was used to limit head movements comfortably within the coil. High-resolution (SPGR) anatomic images were acquired prior to functional imaging to allow subsequent anatomical localisation and normalisation of functional activation. Stimuli were back-projected onto a screen at the subject’s feet and were viewed with the aid of prism glasses attached to the inside of the radio-frequency head-coil. fMRI analysis All data processing was conducted with the software package AFNI (Cox, 1996). Time shifting, using Fourier interpolation to adjust for differences in slice acquisition times, 3D motion correction, and edge detection algorithms were first applied to the functional data. For each subject, the 2 runs per task were concatenated to produce one continuous dataset per task. The average percentage change in signal for all trials of each switching demand and task was calculated relative to the average signal during the rest periods. The average signal produced during the performance of each trial was based on only those images acquired during the updating portion of each trial (images acquired while the subject reported the final results or during the brief pre-trial periods were modelled with separate nuisance covariates). These percent-change scores, nine per
voxel per subject, served as the basic unit of analysis and are subsequently referred to as ‘activation.’ Activation maps were converted to a standard stereotaxic coordinate system (Talairach and Tourneaux, 1988) and spatially blurred using a 4.2 mm full-width-at-half-maximum isotropic Gaussian filter. Monte Carlo simulations revealed that a voxelwise threshold (p ⱕ 0.001) combined with a minimum cluster size criterion (164 l) resulted in a 0.01 false positive level for a cluster of activation and a posterior voxelwise threshold of p ⱕ 10⫺6. This threshold was used for all subsequent t-tests and ANOVAs. Basic task activation maps for each level of switching demand within each task were identified with one-sample t-tests against the null hypotheses of no change in activation. To identify areas that differed as a function of switching demand within each task a two-way, repeated measures, voxelwise ANOVA was performed with switching demand as a fixed factor and subjects as a random factor. A voxel was regarded as showing switching effects if it was significant in both the ANOVA and any of the task t-test maps. Cortical areas that increased in activation with switching demand were interpreted as subserving the attention switching executive function. To test for task specific activation and interactions between task and attention switching demand, a 3 (TASK) ⫻ 3 (SWITCHING DEMAND) voxelwise ANOVA with subjects as a random factor was also performed. Finally, functionally-defined region-of-interest (ROI) analyses were performed on bilateral DLPFC and on two separate clusters in anterior frontal cortex; a-prioridefined ROI analyses was performed on bilateral VLPFC. Performance analyses The number of correct reports of counts or locations, or both, allowing the subjects to score a maximum of two points per trial, determined accuracy. A second, more conservative measure scored each trial correct only if both reports were accurate, thereby ensuring that subjects had completed the attention switching required in each trial. Both measures, however, yielded identical results and only the first will be reported.
Results Task performance Two subjects were discarded from the performance and functional analyses because performance data revealed more incorrect than correct reports. A 3 (TASK) ⫻ 3 (SWITCHING DEMAND), repeated-measures ANOVA revealed main effects of switching demand (F2/26 ⫽ 8.191, p ⱕ .01) and task (F2/26 ⫽ 3.359, p ⱕ .05) and a significant interaction (F4/24 ⫽ 3.001, p ⱕ .05) (Fig. 4). Differences in performance, dependent upon the switching demands, were in the expected direction, but not all pairwise contrasts were significant. Analysis of simple main effects revealed that in
A. Ku¨bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 4. Performance measured as correct trial reports (final counts or final locations). Bars indicate the mean percentage of correct reports per task and switching demands. Error bars are SEM.
the verbal task subjects performed significantly better in low switching demand trials compared to medium (p ⬍ .05) and high (p ⬍ .05), which in turn were not different from each other. In the visuospatial task, no significant differences in switching demand were found and in the combined task subjects performed significantly worse in high switching demand trials compared to Medium (p ⬍ .01), but performed equally well in low and medium trials. In the medium level of switching demand, subjects performed best in the combined task (verbal p ⬍ .01, visuospatial p ⬍ .05); in low and high trials performance did not differ across tasks. All trials were included in the functional analysis, since previous data showed that trials in which errors in counting were made to be comparable to error-free trials in terms of switching costs (Garavan, 1998). In the present data, this was borne out by the nature of the errors in the incorrect trials. A total of 205 incorrect trials (13.6%) were seen across all subjects and tasks. Of these, on only 34 trials (16%) were both reports incorrect. On the incorrect counts trials, 92.4% of the errors consisted of the reported count being ⫾1 from the true count. The visuospatial task did not provide for a comparable analysis because of the limited number of locations within the 2 ⫻ 2 matrix. Although in this study we did not systematically assess participants’ strategy to update the count of circles and the location of dots, we know from previous studies that participants count subvocally when updating the circle counts (Garavan, 1998). For the visuospatial task, subjects reported anecdotally that they “moved the dot in their head,” indicating a visuospatial strategy. The lack of correlation in the nonswitch trials of the verbal and visuospatial tasks in Experiment 1 also indicates that participants used different strategies to accomplish the tasks. Functional activation Brain areas that increased in activation as a function of attention switching demand (i.e., High ⬎ Medium ⬎ Low), within each task were not restricted to the prefrontal lobes (voxelwise repeated measures ANOVA with switching de-
1303
mand as fixed factor and subjects as random factor calculated for each task separately). Instead, attention switching was associated with bilateral activation changes in a broadly distributed working memory network that included DLPFC, cingulate gyrus, basal ganglia, premotor areas, thalamus, parietal lobules, precuneus, temporal and occipital lobes, and cerebellum (Fig. 5). A combination of the attention switching activation maps of each task (OR map) revealed large overlap and yielded only a few separate clusters in prefrontal cortex. For this reason we restricted the ROI analysis to the prefrontal cortex. In general, brain areas involved in within-modality switching overlapped or fell adjacent to areas involved in between-modality switching (Fig. 5). The voxelwise 3 (TASK) ⫻ 3 (SWITCHING DEMAND) repeated-measures ANOVA with subjects as random factor on the activation maps yielded a main effect of switching demand for almost all activated areas (Fig. 5). A main effect of task was found in left precuneus, left inferior temporal gyrus, and left middle and superior occipital gyri. Pairwise comparisons within these regions revealed that the highest activation was associated with the visuospatial task. Activation levels in these areas were intermediate in the combined task and were not significantly greater than zero during the verbal task. No interaction between task and switching demand was found. Functionally-defined ROI analyses One exception to the largely overlapping activations of the attention switching activation maps were two separable bilateral anterior frontal clusters of activation (middle frontal gyrus and BA10, centre-of-mass: right x ⫽ ⫺41, y ⫽ ⫺41, z ⫽ 15, vol ⫽ 173 l; left x ⫽ 34, y ⫽ ⫺45, z ⫽ 18, vol ⫽ 119 l) which were only present in the betweenmodality switching condition (voxelwise ANOVA; Fig. 5 most anterior slice). However, a 2 (HEMISPHERE) ⫻ 3 (TASK) ⫻ 3 (SWITCHING DEMAND) repeated-measures ANOVA performed on mean activation within these functionally defined clusters revealed a significant main effect of switching demand (F2/26 ⫽ 28.64, p ⬍ .001; posthoc pairwise comparisons: High ⬎ Medium ⬎ Low, p ⬍ .05), but no main effect of task. A second, functionally-defined ROI analysis was performed on the bilateral DLPFC activation (Fig. 5 second most anterior slice). A discrete cluster of activation existed within the right hemisphere DLPFC (middle frontal gyrus and BA9, centre-of-mass: x ⫽ ⫺43, y ⫽ ⫺23, z ⫽ 29, vol ⫽ 496 l) and its rostral and caudal boundaries were used to demarcate the rostral and caudal boundaries of a more extensive left DLPFC activation (x ⫽ 26, y ⫽ ⫺22, z ⫽ 28, vol ⫽ 2186 l) which was contiguous with more posterior activations. A 2 (HEMISPHERE) ⫻ 3 (TASK) ⫻ 3 (SWITCHING DEMAND) repeated-measures ANOVA performed on mean activation within these clusters revealed a significant main effect of switching demand (F2/26 ⫽
1304
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 5. Areas showing a significant main effect of switching demand within each task are presented in an OR map. Blue: activation in the verbal task only, red: visuospatial only, green: combined only, yellow: overlapping areas. The activated areas were bilateral and comprised superior, middle, inferior and medial frontal gyri; anterior cingulate, cingulate, and posterior cingulate gyri; insula; basal ganglia; pre- and postcentral gyri; thalamus; inferior and superior parietal lobules, cuneus and precuneus, middle temporal and middle occipital gyri; fusiform gyrus; cerebellum—all bilateral.
70.39, p ⬍ .001; post-hoc pairwise comparisons: High ⬎ Medium ⬎ Low, p ⬍ .001), but no other main effects or interactions. In the voxelwise analyses, no significant activation was found in VLPFC. However, in light of our interest in the functional organization of prefrontal cortex, spherical VLPFC ROIs were defined by the centre-of-mass coordinates provided by Owen and colleagues (Owen et al., 1996) (x ⫽ ⫹/⫺47, y ⫽ ⫺22, z ⫽ 6; volume ⫽ 500 l to approximate the size of the right DLPFC cluster). Results revealed no significant activation (t-tests vs 0) within these ROIs. Larger clusters of 1000 l volume were also tested but yielded identical results, confirming the lack of activation in VLPFC. Prefrontal functional dissociations A 2 (REGION) ⫻ 2 (HEMISPHERE) ⫻ 3 (TASK) ⫻ 3 (SWITCHING DEMAND) repeated-measures ANOVA was performed on the bilateral anterior frontal and DLPFC functionally defined ROIs and revealed a significant main effect of switching demand (F2/26 ⫽ 67.58, p ⬍ .001; posthoc pairwise comparisons: High ⬎ Medium ⬎ Low, p ⬍ .05) and a significant interaction between region and task (F2/26 ⫽ 3.94, p ⬍ .05); posthoc pairwise comparisons were not significant. All other interactions and main effects were
not significant. Although only significant in the combined compared to the verbal task during high switching demands (p ⬍ .05 posthoc pairwise comparisons), activation in anterior frontal cortex tended to be higher for between-modality switching compared to within-modality switching (Fig. 6).
General discussion The same executive function—attention switching— when applied in the phonological, the visuospatial and combined working memory modalities activated similar and broadly distributed areas of the brain that have repeatedly been found to subserve working memory (D’Esposito et al., 1998; D’Esposito and Grossman, 1996; Haxby et al., 2000). Moreover, certain functional dissociations were found within the frontal lobes, with anterior frontal and dorsolateral cortex being differentially involved in within- and between-modality switching. Further, while dorsolateral prefrontal cortex was implicated in both within- and betweenmodality attention switching, no activation was found in ventrolateral areas for this function. These findings have implications for our understanding of the frontal lobes and how executive processes might be instantiated in the brain.
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 6. (A) Interaction between region and task. Anterior frontal cortex was most activated during performance of the combined task whereas highest DLPFC activation was associated with the visuospatial task (not significant). (B and C) Mean activation in functionally-defined regions of interest collapsed over hemisphere in anterior frontal cortex and dorsolateral prefrontal cortex. Error bars are SEM.
Brain correlates of attention switching within working memory The allocation of attentional resources within working memory was not accomplished solely by prefrontal cortex. Instead, activation in the full working memory network, including extensive parietal and premotor areas, varied as a function of the tasks’ attention switching demands. Others have also implicated posterior brain areas in similar cognitive processes. For example, Kimberg and colleagues (Kimberg et al., 2000) found regions in the superior parietal lobule and in the right DLPFC that were involved in switching between letter and number tasks. Comparing switch to nonswitch trials showed more neural activity in frontal, parietal, occipital and subcortical areas before and during the processing of switch trials. A possible involvement of right posterior parietal areas (BA40) in attention switching between representations within verbal working memory has
1305
also been reported (Jonides et al., 1998). Furthermore, Dove and colleagues (Dove, et al., 2000) could not identify any cortical areas that were exclusively activated during switching trials in an unpredictable task switching paradigm. Instead, using event-related fMRI, a widespread network, including bilateral lateral prefrontal, premotor cortex and insula, left intraparietal sulcus, the SMA/pre-SMA region and the cuneus and precuneus, was found to be involved in task switching. Parietal cortex is often considered the seat of storage and maintenance processes in verbal and spatial working memory tasks (Bunge et al., 2000; Hartley and Speer, 2000; Honey et al., 2000; Jonides et al., 1998; Paulesu et al., 1993; Salmon et al., 1996; Smith et al., 1998). For example, Postle and colleagues (Postle et al., 1999) reported a dissociation of mnemonic and executive processes within working memory with the former located in the inferior parietal lobule (BA40) and the latter represented in DLPFC. However, others have attributed an involvement in executive control to parietal cortices. For example, Le and colleagues (Le et al., 1998) found robust activation in the superior posterior parietal lobule, cuneus and precuneus during attention switching between two targets and proposed that posterior parietal cortices compute spatial codes involved in binding of features for an attended stimulus. Sohn and colleagues (Sohn, et al., 2000) presented their subjects with an alternating task switching paradigm in which 50% of the trials were cued. The authors concluded that superior posterior parietal cortex may be involved in endogenous goal-directed preparation for a task set, whereas other areas including inferior posterior parietal cortex may be responsible for stimulus-driven completion of adopting a task set. The present data demonstrated widespread parietal activation that was sensitive to switching demand but was not task specific, suggesting that parietal lobe contributions to working memory go beyond stimulus representation, storage and maintenance. Instead, we suggest that parietal cortices may be necessary for the allocation of attentional resources within working memory and may contribute, therefore, to executive control. The present results have found extensive cortical areas, not just parietal or prefrontal, associated with the tasks’ executive demands. This would suggest that attention switching might be reflected as a dynamic property of the larger task-activated working memory network rather than as a specific locus of activation. Additional data on this attention switching task suggests that switching times include a retrieval mechanism such that switching to an item in working memory entails the re-activation of the new item (Voigt and Hagendorf, 2002). Furthermore, switching away from an item in working memory may entail the suppression of that old item (Voigt and Hagendorf, 2002) as has been observed for switching between tasks (Mayr and Kliegl, 2000). These processes, re-activation and suppression of the contents of working memory may explain, in part, why such widespread functional activation of the working memory
1306
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
network is observed. However, such a conclusion must be reconciled with the pre-eminent role typically given to the frontal lobes for executive functioning (D’Esposito et al., 2000; Owen et al., 1998; Smith and Jonides, 1999; Wagner et al., 2001) and the prefrontal dissociations, discussed below, observed for different types of attention switching. One possibility is that executive functioning might be divided into two components. The first, associated with the prefrontal lobes, may select and initiate the function to be performed in accordance with the task rules and goals that are stored there. This is consistent with the aimless or inappropriate behaviour and behavioural inertia often evident in frontal patients (Anderson, et al., 2000; Manes et al., 2002; Zalla et al., 2001). The posterior brain regions, on the other hand, may be central to the enactment of the switch through the re-activation and suppression of the working memory contents, as described above (Bunge et al., 2002). Thus, storage and reactivation of information in parietal areas may be the actual means by which attention switching occurs. Such a conceptualisation, while still ascribing a representation storage role to posterior areas posits that dynamic changes within and between those representations may be the “machinery” by which the switch is accomplished. This conceptualisation permits for prefrontal-posterior dissociations while also explaining why widespread activation patterns might underlie the attention switching function in tasks such as these. It may, however, also call into question a firm distinction between a storage and an executive function. Functional specificity within the prefrontal cortex All tasks produced activation within dorsolateral areas that varied with switching demand, but no activation was found within ventrolateral areas thereby confirming the importance of bilateral DLPFC for executive control in a working memory task. These results are consistent with a processing-specific organization of the PFC with DLPFC being activated when representations in working memory, irrespective of modality, have to be manipulated (Barde and Thompson-Schill, 2002; Kessels et al., 2000; Owen et al., 1996; Owen et al., 1998; Wagner et al., 2001). The greater, but not significant, DLPFC activation for the visuospatial task might be interpreted as consistent with a domain-specific organization of the PFC (Goldman-Rakic, 1996; Gruber and von Cramon, 2001). However, a domain-specific approach would have to explain why activation of DLPFC was not higher in the combined task with its visuospatial component, compared to the verbal task and why DLPFC was activated in all tasks, including the verbal task which was completely non-spatial (Nystrom et al., 2000; Owen et al., 1998; Postle et al., 2000). The data support a functional dissociation between anterior frontal and dorsolateral cortex for manipulating infor-
mation within and between the working memory slave systems. The results suggest that anterior frontal cortex may be recruited when coordination between the two slave systems of working memory is necessary. Koechlin and colleagues (Koechlin et al., 1999) report recruitment of frontopolar cortex (BA10) in a task involving the cognitive process of branching, which refers to the process of allocating resources when attention has to be alternated between two concurrent activities. The authors concluded that frontopolar cortex is recruited when main goals have to be kept in mind during processing of subgoals. Although more anterior than our anterior frontal cluster the Koechlin et al. findings may be relevant to our results. Behavioural data in our study indicate that subjects performed better during between modality switching compared to within modality switching (significant for the medium switching demand). This is in agreement with the results of Experiment 1, which revealed less time cost in between modality switch trials compared to within modality switch trials. This effect may be explained by less interference between attentional resources. Duncan and colleagues (Duncan et al., 1997) have shown that response accuracy is reduced when attention is divided within but not between sensory modalities. The authors presented their subjects with two visual and two auditory streams in which they had to detect targets. Detecting a target in one stream led to sustained reduction of the ability to identify the second target in the other stream. No such effect was seen in the combined presentation of one visual and one auditory stream. The authors concluded that visual attention to a target did not interfere with the auditory attention to another. Since in our study anterior frontal areas were most activated in the combined task one might speculate that they were additionally recruited because coordination of attentional resources between the two slave systems of working memory was necessary. Incorporating these findings with Koechlin’s, it may be the case that anterior frontal cortex is recruited when two sufficiently distinct tasks must be actively maintained. It is noteworthy that anterior frontal areas were more activated during the combined task, which was also the easiest to perform, suggesting that this additional activation cannot simply be attributed to task difficulty. No support for a hemispheric organisation of PFC was found. It has previously been suggested that domain specificity across hemispheres is more pronounced than between dorsal and ventral regions (Ungerleider et al., 1998). With the exception of the involvement of left visual areas in the visuospatial task, the present results found no hemispheric differences in activation in relation to the task requirements. No hemispheric differences in dorsal, ventral or anterior frontal PFC were found. Such a failure to find hemispheric stimulus specificity was also reported by Nystrom and colleagues (Nystrom et al., 2000) who presented their subjects with verbal, spatial and non-verbal and non-spatial stimuli in an n-back task.
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Conclusion Two patterns of brain activation changes were found for attention switching and the dynamics of coordinating the working memory slave systems. First, increased activation of the entire task-related network was observed for switching both within and between modalities. Second, additional recruitment of anterior frontal regions into this network was observed for switching between modalities. However, it should be noted that activation in the anterior frontal areas, although sub-threshold for the within-modality tasks, did modulate with switching demand in these tasks. Consequently, while we have proposed differences in the executive function roles of the prefrontal and posterior areas, these results do challenge the notion of rigid anatomically separable modules within working memory. In contrast, Smith and Jonides (Smith and Jonides, 1999) argue for a clear distinction between short-term storage and executive processes as a main organizational principle of the PFC. As we held constant online storage and rehearsal demands, areas that were exclusively involved in these working memory components should not have increased in activation with increased switching demand. Furthermore, even those areas that showed a main effect of task, including the posterior areas that were specific to the visuospatial task, also showed switching demand effects. Independent of task and modality, attention switching activated a widespread frontoparietal network, consistent with a trend away from a strict localization of psychological processes to cortical areas to the view that multiple brain areas are involved in cognitive functioning to different degrees (Carpenter et al., 2000; Haxby et al., 2000). Acknowledgments The assistance of Catherine Fassbender and Eoin Mooney is gratefully acknowledged. This study is supported by NIDA grant RO1 DA14100 and GCRC M01 RR00058. References Anderson, S.W., Damasio, H., Tranel, D., Damasio, A.R., 2000. Long-term sequelae of prefrontal cortex damage acquired in early childhood. Developmental Psychology 18, 281–296. Baddeley, A., Hitch, G.J., 1974. Working memory, in: Bower, G.H. (Ed.), The Psychology of Learning and Motivation, Vol. 8, Academic Press, New York, pp. 47– 89. Barde, L.H.F., Thompson-Schill, S.L., 2002. Models of functional organization of the lateral prefrontal cortex in verbal working memory: evidence in favor of the process model. Journal of Cognitive Neuroscience 14, 1054 –1063. Bunge, S.A., Hazeltine, E., Scanlon, M.D., Rosen, A.C., Gabrieli, J.D.E., 2002. Dissociable contribution of prefrontal and parietal cortices to response selection. Neuroimage 17, 1562–1571. Bunge, S.A., Klingberg, T., Jacobsen, R.B., Gabrieli, J.D.E., 2000. A resource model of the neural basis of executive working memory.
1307
Proceedings of the National Academy of Sciences of the United States of America 97, 3573–3578. Carpenter, P.A., Just, M.A., Reichle, E.D., 2000. Working memory and executive function: evidence from neuroimaging. Current Opinion in Neurobiology 10, 195–199. Chein, J.M., Fiez, J.A., 2001. Dissociation of verbal working memory system components using a delayed serial recall task. Cerebral Cortex 11, 1003–1014. Cocchini, G., Logie, R.H., Della Sala, S., MacPherson, S.E., Baddeley, A.D., 2002. Concurrent performance of two memory tasks: evidence for domain-specific working memory systems. Memory and Cognition 30 (7), 1086 –1095. Collette, F., Salmon, E., van der Linden, M., Chicherio, C., Belleville, S., Degueldre, C., Delfiore, G., Franck, G., 1999. Regional brain activity during tasks devoted to the central executive of working memory. Cognitive Brain Research 7, 411– 417. Courtney, S.M., Ungerleider, L.G., Keil, K., Haxby, J.V., 1996. Object and spatial visual working memory activate separate neural systems in human cortex. Cerebral Cortex 6, 39 – 49. Cowan, N., 1988. Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human informationprocessing system. Psychological Bulletin 104, 163–191. Cowan, N., 1993. Activation, attention, and short-term memory. Memory and Cognition 21 (2), 162–167. Cox, R.W., 1996. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Computers and Biomedical Research 29, 162–173. D’Esposito, M., Aguirre, G.K., Zarahn, E., Ballard, D., Shin, R.K., Lease, J., 1998. Functional MRI studies of spatial and nonspatial working memory. Cognitive Brain Research 7, 1–13. D’Esposito, M., Grossman, M., 1996. The physiological basis of ececutive function and working memory. The Neuroscientist 2, 345–352. D’Esposito, M., Postle, B.R., Rypma, B., 2000. Prefrontal cortical contributions to working memory: evidence from event-related fMRI studies. Experimental Brain Research 133, 3–11. Diwadkar, V.A., Carpenter, P., Just, M.A., 2000. Collaborative activity between parietal and dorso-lateral prefrontal cortex in dynamic spatial working memory revealed by fMRI. Neuroimage 12, 85–99. Dove, A., Pollman, S., Schubert, T., Wiggins, C.J., von Cramon, D.Y., 2000. Prefrontal cortex activation in task switching: an event-related fMRI study. Cognitive Brain Research 9, 103–109. Duncan, J., Martens, S., Ward, R., 1997. Restricted attentional capacity within but not between sensory modalities. Nature 387, 808 – 810. Farmer, E.W., Berman, J.V.F., Fletcher, Y.L., 1986. Evidence for a visuospatial scratch-pad in working memory. The Quarterly Journal of Experimental Psychology 38A, 675– 688. Garavan, H., 1998. Serial attention within working memory. Memory and Cognition 26, 263–276. Garavan, H., Ross, T.J., Li, S.J., Stein, E.A., 2000. A parametric manipulation of central executive functioning. Cerebral Cortex 10 (6), 585– 592. Goldman-Rakic. 1996. The prefrontal landscape: implications of functional architecture for understanding human mentation and the central executive. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 351, 1445–1453. Gruber, O., von Cramon, D.Y., 2001. Domain-specific distribution of working memory processes along human prefrontal and parietal cortices: a functional magnetic resonance imaging study. Neuroscience Letters 297, 29 –32. Halgren, E., Boujon, C., Clarke, J., Wang, C., Chauvel, P., 2002. Rapid distributed fronto-parieto-occipital processing stages during working memory in humans. Cerebral Cortex 12, 710 –728. Hartley, A.A., Speer, N.K., 2000. Locating and fractionating working memory using functional neuroimaging: storage, maintenance, and executive functions. Microscopy Research and Technique 51, 45–53.
1308
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Haxby, J.V., Petit, L., Ungerleider, L.G., Courtney, S.M., 2000. Distinguishing the functional roles of multiple regions in distributed neural systems for visual working memory. Neuroimage 11, 380 –391. Honey, G.D., Bullmore, E.T., Sharma, T., 2000. Prolonged reaction time to a verbal working memory task predicts increased power of posterior parietal cortical activation. Neuroimage 12, 495–503. Jonides, J., Schumacher, E.H., Smith, E.E., Koeppe, R.A., Awh, E., Reuter-Lorenz, P.A., Marshuetz, C., Willis, C.R., 1998. The role of parietal cortex in verbal working memory. Journal of Neuroscience 18, 5026 –5034. Kessels, R.P., Postma, A., Wijnalda, E.M., de Haan, E.H., 2000. Frontallobe involvement in spatial memory: evidence from PET, fMRI, and lesion studies. Neuropsychology Review 10 (2), 101–113. Kimberg, D.Y., Aguirre, G.K., D’Esposito, M., 2000. Modulation of taskrelated neural activity in task-switching: an fMRI study. Cognitive Brain Research 10, 189 –196. Koechlin, E., Basso, G., Pietrini, P., Panzer, S., Grafman, J., 1999. The role of the anterior prefrontal cortex in human cognition. Nature 399, 148 –151. LaBar, K.S., Gitelman, D.R., Parrish, T.B., Mesulam, M.-M., 1999. Neuroanatomic overlap of working memory and spatial attention networks: a functional MRI comparison within subjects. Neuroimage 10, 695– 704. Le, T.H., Pardo, J.V., Hu, X., 1998. 4 T-fMRI study of nonspatial shifting of selective attention: cerebellar and parietal contributions. Journal of Neurophysiology 79, 1535–1548. Manes, F., Sahakian, B., Clarke, L., Rogers, R., Antoun, N., Aitken, M., Robbins, T., 2002. Decision-making processes following damage to the prefrontal cortex. Brain 125, 624 – 639. Mayr, U., Kliegl, R., 2000. Task-set switching and long-term memory retrieval. Journal of Experimental Psychology. Learning, Memory, and Cognition 26 (5), 1124 –1140. Nystrom, L.E., Braver, T.S., Sabb, F.W., Delgado, M.R., Noll, D.C., Cohen, J., 2000. Working memory for letters, shapes, and locations: fMRI evidence against stimulus-based regional organization in human prefrontal cortex. Neuroimage 11, 424 – 446. Owen, A.M., 1997. The functional organisation of working memory processes within human lateral frontal cortex: the contribution of functional neuroimaging. European Journal of Neuroscience 9, 1329 –1339. Owen, A.M., Evans, A.C., Petrides, M., 1996. Evidence for a two-stage model of spatial working memory processing within the lateral frontal cortex: a positron emission tomography study. Cerebral Cortex 6, 31–38. Owen, A.M., Stern, C.E., Look, R.B., Tracey, I., Rosen, B.R., Petrides, M., 1998. Functional organization of spatial and nonspatial working memory processing within the human lateral frontal cortex. Proceedings of the National Academy of Sciences of the United States of America 95, 7721–7726. Paulesu, E., Frith, C.D., Frackowiak, R.S., 1993. The neural correlates of the verbal component of working memory. Nature 362 (6418), 342– 345. Petrides, M., Alivisatos, B., Evans, A.C., Meyer, E., 1993. Dissociation of human mid-dorsolateral from posterior dorsolateral frontal cortex in
memory processing. Proceedings of the National Academy of Sciences of the United States of America 90, 873– 877. Postle, B.R., Berger, J.S., D’Esposito, M., 1999. Functional neuroanatomical double dissociation of mnemonic and executive control processes contributing to working memory performance. Proceedings of the National Academy of Sciences of the United States of America 96, 12959 –12964. Postle, B.R., Stern, C.E., Rosen, B.R., Corkin, S., 2000. An fMRI investigation of cortical contributions to spatial and nonspatial visual working memory. Neuroimage 11, 409 – 423. Raye, C.L., Johnson, M.K., Mitchell, K.J., Reeder, J.A., Greene, E.J., 2002. Neuroimaging a single thought: dorsolateral PFC activity associated with refreshing just-activated information. Neuroimage 15, 447– 453. Rypma, B., Berger, J.S., D’Esposito, M., 2002. The influence of working memory demand and subject performance on prefrontal cortical activity. Journal of Cognitive Neuroscience 14, 721–731. Salmon, E., van der Linden, M., Collette, F., Delfiore, G., Maquet, P., Degueldre, C., Luxen, A., Franck, G., 1996. Regional brain activity during working memory tasks. Brain 119, 1617–1625. Shallice, T., Burgess, P., 1996. The domain of supervisory processes and temporal organization of behaviour. Philosophical Transactions of the Royal Society of London. Series B. Biological Sciences 351 (1346), 1405–1411 discussion 1411–1412. Smith, E.E., Jonides, J., 1999. Storage and executive processes in the frontal lobes. Science 283, 1657–1661. Smith, E.E., Jonides, J., Koeppe, R.A., 1996. Dissociating verbal and spatial working memory using PET. Cerebral Cortex 6, 11–20. Smith, E.E., Jonides, J., Marshuetz, C., Koeppe, A.C., 1998. Components of verbal working memory: evidence from neuroimaging. Proceedings of the National Academy of Sciences of the United States of America 95, 876 – 882. Sohn, M.-H., Ursu, S., Anderson, J.R., Stenger, V.A., Carter, C.S., 2000. The role of prefrontal cortex and posterior parietal cortex in task switching. Proceedings of the National Academy of Sciences of the United States of America 97, 13448 –13453. Talairach, J., Tourneaux, P., 1988. Co-planar Stereotaxic Atlas of the Human Brain. Thieme Medical, New York. Ungerleider, L.G., Courtney, S.M., Haxby, J.V., 1998. A neural system for human visual working memory. Proceedings of the National Academy of Sciences of the United States of America 95, 883– 890. Voigt, S., Hagendorf, H., 2002. The role of task context for component processes in focus switching. Psychologische Beitraege 44, 248 –274. Wagner, A.D., Maril, A., Bjork, R.A., Schacter, D.L., 2001. Prefrontal contributions to executive control: fMRI evidence for functional distinctions within lateral prefrontal cortex. Neuroimage 14, 1337–1347. Wong, E.C., Boskamp, E., Hyde, J.S., 1992. A volume optimized quadratureelliptical endcap birdcage brain coil. Paper presented at the Eleventh Annual Scientific Meeting, Society for Magnetic Resonance Medicine, Berlin. Zalla, T., Plassiart, C., Pillon, B., Grafman, J., Sirigu, A., 2001. Action planning in a virtual context after prefrontal cortex damage. Neuropsychologia 39, 759 –770.
www.elsevier.com/locate/ynimg
Co-ordination within and between verbal and visuospatial working memory: network modulation and anterior frontal recruitment A. Ku¨bler,a K. Murphy,a J. Kaufman,b E.A. Stein,b,1 and H. Garavana,b,* b
a Department of Psychology and Institute of Neuroscience, Trinity College Dublin, Dublin 2, Ireland Department of Psychiatry and Behavioral Medicine, Medical College of Wisconsin, Milwaukee, WI 53226, USA
Received 31 March 2003; revised 17 June 2003; accepted 30 June 2003
Abstract Attention switching between items being stored and manipulated in working memory (WM) is proposed to be an elementary executive function. Experiment 1 reveals a similar attentional limitation within and between verbal and visuospatial WM and identifies a supramodal switching process required for switching between WM items. By using functional magnetic resonance imaging, Experiment 2 investigated brain activation correlates of parametrically varied attention switching within and between these two WM modalities. Attention switching activation was broadly distributed, was quite similar across the three conditions, and, in almost all areas, increased with increasing switching demand, indicating that attention switching recruits and modulates the entire WM network. Dorsolateral prefrontal cortex was implicated in both within- and between-modality attention switching, but no significant activation was found in ventrolateral areas, supporting dorsal-ventral process models of prefrontal organization. A functional dissociation between anterior frontal and dorsolateral prefrontal cortex was found with the former being more activated when switching attention between modalities was required. The data challenge the notion of an anatomically separate attention switching executive function, but suggest that anterior frontal areas are recruited for the additional demand of coordinating the verbal and visuospatial WM slave systems. © 2003 Elsevier Inc. All rights reserved.
Introduction According to current models of working memory, two separable systems for verbal and visuospatial information processing are controlled by a central executive, which allocates attentional resources according to the task requirements (Cocchini et al., 2002; Cowan, 1993; Shallice and Burgess, 1996). Following Baddeley and Hitch (1974), verbal information is processed within the phonological loop and visuospatial information in the visuospatial sketch pad (Baddeley and Hitch, 1974). Tasks which engage verbal working memory have been shown to activate a widespread network in the brain including dorsolateral prefrontal cortex (DLPFC), Broca’s area, premotor and supplementary motor
* Corresponding author: Fax: ⫹353-1-671-2006. E-mail address: [email protected] (H. Garavan). 1 Current address: National Institute on Drug Abuse-IRP, Neuroimaging Research Branch, 5500 Nathan Shock Drive, Baltimore, MD 21224, USA. 1053-8119/$ – see front matter © 2003 Elsevier Inc. All rights reserved. doi:10.1016/S1053-8119(03)00400-2
area, anterior cingulate, parietal cortex and cerebellum (Jonides et al., 1998). For example, Paulesu and colleagues (Paulesu et al., 1993) localized the phonological store to the left supramarginal gyrus and the articulatory rehearsal process to Broca’s area. As in verbal working memory, tasks addressing visuospatial working memory activate multiple, widely distributed regions in parietal, motor, and prefrontal cortices (Haxby et al., 2000). While the exact role of the prefrontal cortices in working memory remains unclear, a number of distinct models have been proposed. According to one domain-specific model, the prefrontal cortex is functionally subdivided into a dorsolateral region for spatial information and a ventrolateral region for object information (Courtney et al., 1996; Goldman-Rakic, 1996). An alternative model argues instead for domain specificity across hemispheres, wherein a left midfrontal area is specialized for object or verbal (non-spatial) working memory and right inferior and superior frontal areas for spatial working memory (Smith et al., 1996; Ungerleider et al., 1998). Yet a third process-specific model
A. Ku¨bler et al. / NeuroImage 20 (2003) 1298 –1308
proposes that ventral and dorsolateral PFC are engaged for specific working memory processes such as the maintenance, retrieval and on-line manipulation of working memory representations (D’Esposito, et al., 2000; Kessels, et al., 2000; Owen, 1997; Owen et al., 1996; Wagner et al., 2001). It is proposed that ventral prefrontal areas are involved in the maintenance of information in working memory, but if this information has to be manipulated or shielded against distraction, activity increases in DLPFC only. Indeed, the nature of the manipulation-related processes ascribed to DLPFC may be nonmnemonic (Collette et al., 1999; D’Esposito et al., 2000; Postle, et al., 1999). Dorsolateral prefrontal cortex appears critical for executive functioning (Smith et al., 1998), as robust activation has been observed here in many studies that engage a variety of working memory and executive functions (Chein and Fiez, 2001; Collette et al., 1999; Diwadkar et al., 2000; Haxby et al., 2000; Rypma et al., 2002; Salmon et al., 1996). It would be erroneous, however, to conclude that executive functioning is solely restricted to prefrontal areas. Studies addressing executive functioning typically find widespread activation in parietal, premotor, cingulate, temporal, occipital and cerebellar areas (Collette et al., 1999; Le et al., 1998; Owen et al., 1996; Petrides et al., 1993; Raye et al., 2002; Salmon et al., 1996). Isolating specific executive functions within general working memory functioning can prove challenging as executive functions constitute a major and integral part of normal working memory. Following Garavan and colleagues (Garavan, 1998; Garavan et al., 2000), we used a task designed to isolate the allocation of attentional resources within working memory while holding constant on-line storage and rehearsal demands. The task probed executive functions by isolating switches of attention between items residing in working memory. We used a counting task, which requires verbal updating and should therefore engage verbal working memory, a spatial location updating task which should engage the visuospatial sketchpad and a between-modalities task, in which one count and one location had to be updated. Previous research has shown that switching between two counts in working memory results in a sizeable time cost (Garavan, 1998). This time cost persists even after intensive task practice, which indicates that people do not have simultaneous access to all items currently in working memory, consistent with Cowan’s model of a focus of attention within working memory (Cowan, 1988). We propose that switching attention between items in working memory is a basic and elementary executive function. Thus, we hypothesised that we would find the same switching costs as in verbal WM when switching between items in visuospatial WM or when switching between the two WM modalities. A behavioural study tested these hypotheses (Experiment 1) prior to functional imaging. To isolate functional activation associated with attention switching, we chose a parametric manipulation of executive
1299
demands instead of the more common subtraction method. We have previously demonstrated (Garavan et al., 2000) that activation associated with attention switching within verbal working memory was not restricted to prefrontal areas, but included large areas in left parietal, left cerebellar, occipital, temporal, and subcortical areas. With the present study, we attempted to determine if attention switching within visuospatial working memory would activate similar or distinct modality specific cortical areas, and whether this activation would be restricted to the prefrontal lobes. Based upon previous results (Garavan et al., 2000; Halgren et al., 2002; LaBar et al., 1999) and the behavioural data from Experiment 1, we expected activation in a frontal-parietal network in all tasks. As the tasks involved similar on-line manipulation demands (switching between and updating representations within working memory) but did so within distinct modalities (verbal and visuospatial), we were also interested whether our tasks would activate PFC according to a domain- or processing-specific model. Little is known about the dynamics between the different components of working memory. Koechlin and colleagues (Koechlin et al., 1999) demonstrated that, in addition to DLPFC, the frontopolar cortex (BA10) was activated when a major task goal had to be kept in mind while performing subgoals of a task. Therefore, based on a hierarchical model of working memory functioning, one might propose that frontopolar cortex could be activated when attention switching has to be coordinated between verbal and visuospatial working memory.
Experiment 1 Tasks and design Twenty-five subjects (20 female, mean age 21 ⫾ 4.5, age range 17–35) completed the attention switching tasks. All gave informed consent, which was approved by the institutional review board of the Psychology department in Trinity College Dublin. Three switching tasks were employed (Fig. 1). Two addressed attention switching within the verbal or visuospatial modalities. The verbal task required participants to keep a count of how many red and how many blue squares were presented and to report the results at the end of each trial. There were between 12 and 20 stimuli per trial and, following practice, subjects completed 25 trials. Stimulus presentation was self-paced; subjects pressed the spacebar to clear the screen of the current stimulus and present the next. Two event types are noted. A switch occurred when either a red square followed a blue square or vice versa, thereby requiring subjects to switch between the red and blue counts being maintained in working memory. A nonswitch occurred when two red squares or two blue squares were presented in succession thereby requiring subjects to update the same count twice without having to
1300
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 1. Schematic of the three tasks. (A) In the verbal task, blue and red squares had to be counted. (B) In the visuospatial task, the locations of a blue and red dot had to be updated. (C) In the combined task squares had to be counted and the location of the dot had to be updated. Storage and rehearsal demands were equal in all tasks and the manipulation of interest was the switch between the count or location representations.
switch between counts. In total, there were 192 switch and 183 nonswitch events. Comparison of the switch and nonswitch response times provides a measure of the switching cost. The visuospatial working memory locations task, designed to parallel the counting task, required subjects to maintain and update two locations. The locations were one red and one blue dot presented in a 2 ⫻ 2 matrix. At the start of each trial, subjects were presented with the starting locations of the dots (i.e., which cell each dot occupied), which were randomly chosen, and during each trial subjects were presented with a sequence of single arrows. The arrows were either red or blue and subjects were required to move the corresponding dot in the direction of the coloured arrow (e.g., a blue arrow pointing right instructs subjects to move the blue dot one cell to the right). The 2 ⫻ 2 matrix was presented only at the beginning of the trial so subjects had to maintain and update the dot locations in working memory. As in the verbal task, the arrow presentation was self-paced and switching time could be calculated by comparing successive updates of the same dot with successive updates of the two dots as the latter requires a switch between the current locations of the dots. The stimulus train was based on that used for the verbal task. That is, switches and nonswitches between the locations were in identical positions to those in the verbal task’s stimulus train. As above, subjects completed 25 trials following practice. The third task addressed attention switching between the verbal and visuospatial WM modality. In this combined task participants were required to maintain and update one count and one location through presentation of blue squares and red arrows. Other details were the same as for the preceding tasks. Order of the three tasks was counterbalanced across subjects.
Results Subjects performed well on the tasks with an average of 5.4% errors (i.e., in which either a count or location was reported incorrectly at the end of a trial). Only correct trials were included in the subsequent response time analyses from which one subject with incomplete data was dropped. A 3 (TASK: verbal, visuospatial, combined) ⫻ 2 (SWITCH: switch, nonswitch) ANOVA revealed a significant task effect [F(2,46) ⫽ 10.2, p ⱕ 0.0002], a significant switch effect [F(1,23) ⫽ 117.0, p ⱕ 0.0001] and a significant interaction [F(2,46) ⫽ 14.3, p ⱕ 0.0001]. As shown in Fig. 2 and confirmed by t-tests, substantial switching costs were present for each task. An ANOVA comparing the switching costs in the three tasks was significant [F(2.46) ⫽ 14.3, p ⱕ 0.0001] (verbal ⬎ visuospatial ⬎ combined) as were all pairwise comparisons bar visuospatial vs combined [t(23) ⫽ 1.92, p ⱕ 0.07]. In the combined task, switching from the
Fig. 2. Mean reaction time (RT) for switches and nonswitches for all three tasks. Reaction time for switches was significantly slower than for nonswitches in all tasks. Error bars are SEM.
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 3. Top panel shows non-significant correlation between nonswitches in the verbal and the visuospatial task. Switch costs in both tasks (bottom panel) were correlated. Lines show linear trend.
count to the location was not significantly different than switching in the opposite direction [t(24) ⫽ 0.28, p ⱕ 0.78]. Correlational analyses between the verbal and visuospatial tasks revealed that while the nonswitch response times were not related (r ⫽ .25, p ⱕ 0.24) the switching costs in the two tasks were related (r ⫽ .55, p ⱕ 0.006) (see Fig. 3). The switching costs in the within modality tasks were also related to the switching costs in the between modalities combined task (verbal and combined, r ⫽ .42, p ⱕ 0.04; visuospatial and combined, r ⫽ .39, p ⱕ 0.06).
Discussion The results have confirmed an attentional limitation in both verbal and visuospatial working memory consistent with Garavan (1998). Subjects do not have simultaneous and immediate access to more than one working memory item at a time as demonstrated by switching costs when they are required to switch between these items. This limitation was also present when maintaining two items in different working memory modalities as revealed by the sizeable switching costs when switching between the verbal count and visuospatial location. These results suggest that the attentional limitation is a general, supramodal one that tran-
1301
scends specific working memory slave systems. This conclusion is supported by the correlational results. The nonswitch response times of the verbal and visuospatial tasks (i.e., the time involved in the updating operations required of the tasks) were not correlated, providing evidence that different processes were involved in performing the verbal and visuospatial updates. This result is consistent with the substantial body of evidence showing the separateness of the two working memory slave systems (Baddeley and Hitch, 1974; Cocchini et al., 2002; Farmer et al., 1986). However, a significant correlation between the verbal and visuospatial switching costs was observed. These results suggest that while one’s ability to update counts was not related to one’s ability to update locations, one’s ability to switch between counts was related to one’s ability to switch between locations. Similarly, one’s ability to switch between items within each working memory modality was also correlated with one’s ability to switch between items in different modalities. The results of this study suggest that the switching manipulation engages attentional or central executive abilities within working memory that are distinct from modalityspecific abilities, thereby providing a rationale for using the attention switching task as a measure of executive functioning. These behavioural results suggest that the cognitive processes involved in attention switching are quite similar both within and between working memory slave systems and therefore, similar cortical involvement might be anticipated. Experiment 2 was designed to address this neuroanatomical question.
Experiment 2 Tasks and design Sixteen right-handed subjects participated in Experiment 2 (12 female, mean ⫾ SD age: 23.5 ⫾ 3.93, age range 18 –33). All gave informed consent, which was approved by the institutional review board of the Medical College of Wisconsin. The same three attention switching tasks as used in Experiment 1 were modified for imaging to determine the functional neuroanatomy of attention switching within verbal and visuospatial WM and between the two modalities. For imaging purposes, presentation of items was not selfpaced. In the verbal task each stimulus (for the imaging study stimuli were circles rather than squares) was presented for 1400 ms and successive circles were separated by a 100 ms fixation cross, the purpose of which was to clearly delineate successive presentations of the circles. In the visuospatial task, red and blue arrows indicating in which direction the dot should mentally be moved, were presented for 1400 ms and successive arrows were separated by a 100 ms fixation cross. The details for the combined task were the same as for the preceding tasks. To report the final location of a dot, participants used a
1302
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
4-button piano keypad; the keys corresponded to four response options presented on the screen. To report the number of circles, participants were presented with four numbers and had to choose the correct number by pressing the corresponding key of the keypad. Subjects were given 8 sec in which to make their responses. All subjects took part in one session comprising 6 runs with each run containing 9 trials. There were two runs (18 trials) for each of the three tasks. Presentation of tasks was counterbalanced. Trials varied in length (11–16 circles, arrows or circles and arrows) and in switching frequency: The 18 trials were comprised of six “High (H)” (5– 8 switches), six “Medium (M)” (2– 4 switches), and six “Low (L)” (1 switch) switching demand trials. The sequence of trials was HLMLLMLLM for run 1 and HMMHHLHMH for run 2 in all tasks. Trials were preceded by a 4 sec fixation cross. A rest period of 22 sec was provided after every third trial. Rest periods of 22 and 26 sec were included at the start and at the end of each run, respectively. At the end of the rest period a change in the fixation cross signalled the start of the next trial. In total, the experiment lasted approximately 45 minutes. fMRI parameters Nineteen contiguous 7 mm sagittal slices covering the entire brain were collected using a blipped gradient-echo, echo-planar pulse sequence (TE ⫽ 40 ms; TR ⫽ 2000 ms; FOV ⫽ 24 cm; 64 ⫻ 64 matrix; 3.75 ⫻ 3.75 mm in-plane resolution). All scanning was conducted on a 1.5 T GE Signa scanner equipped with a 30.5 cm i.d., three-axis local gradient coil and an endcapped quadrature birdcage radiofrequency head-coil (Wong et al., 1992). Foam padding was used to limit head movements comfortably within the coil. High-resolution (SPGR) anatomic images were acquired prior to functional imaging to allow subsequent anatomical localisation and normalisation of functional activation. Stimuli were back-projected onto a screen at the subject’s feet and were viewed with the aid of prism glasses attached to the inside of the radio-frequency head-coil. fMRI analysis All data processing was conducted with the software package AFNI (Cox, 1996). Time shifting, using Fourier interpolation to adjust for differences in slice acquisition times, 3D motion correction, and edge detection algorithms were first applied to the functional data. For each subject, the 2 runs per task were concatenated to produce one continuous dataset per task. The average percentage change in signal for all trials of each switching demand and task was calculated relative to the average signal during the rest periods. The average signal produced during the performance of each trial was based on only those images acquired during the updating portion of each trial (images acquired while the subject reported the final results or during the brief pre-trial periods were modelled with separate nuisance covariates). These percent-change scores, nine per
voxel per subject, served as the basic unit of analysis and are subsequently referred to as ‘activation.’ Activation maps were converted to a standard stereotaxic coordinate system (Talairach and Tourneaux, 1988) and spatially blurred using a 4.2 mm full-width-at-half-maximum isotropic Gaussian filter. Monte Carlo simulations revealed that a voxelwise threshold (p ⱕ 0.001) combined with a minimum cluster size criterion (164 l) resulted in a 0.01 false positive level for a cluster of activation and a posterior voxelwise threshold of p ⱕ 10⫺6. This threshold was used for all subsequent t-tests and ANOVAs. Basic task activation maps for each level of switching demand within each task were identified with one-sample t-tests against the null hypotheses of no change in activation. To identify areas that differed as a function of switching demand within each task a two-way, repeated measures, voxelwise ANOVA was performed with switching demand as a fixed factor and subjects as a random factor. A voxel was regarded as showing switching effects if it was significant in both the ANOVA and any of the task t-test maps. Cortical areas that increased in activation with switching demand were interpreted as subserving the attention switching executive function. To test for task specific activation and interactions between task and attention switching demand, a 3 (TASK) ⫻ 3 (SWITCHING DEMAND) voxelwise ANOVA with subjects as a random factor was also performed. Finally, functionally-defined region-of-interest (ROI) analyses were performed on bilateral DLPFC and on two separate clusters in anterior frontal cortex; a-prioridefined ROI analyses was performed on bilateral VLPFC. Performance analyses The number of correct reports of counts or locations, or both, allowing the subjects to score a maximum of two points per trial, determined accuracy. A second, more conservative measure scored each trial correct only if both reports were accurate, thereby ensuring that subjects had completed the attention switching required in each trial. Both measures, however, yielded identical results and only the first will be reported.
Results Task performance Two subjects were discarded from the performance and functional analyses because performance data revealed more incorrect than correct reports. A 3 (TASK) ⫻ 3 (SWITCHING DEMAND), repeated-measures ANOVA revealed main effects of switching demand (F2/26 ⫽ 8.191, p ⱕ .01) and task (F2/26 ⫽ 3.359, p ⱕ .05) and a significant interaction (F4/24 ⫽ 3.001, p ⱕ .05) (Fig. 4). Differences in performance, dependent upon the switching demands, were in the expected direction, but not all pairwise contrasts were significant. Analysis of simple main effects revealed that in
A. Ku¨bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 4. Performance measured as correct trial reports (final counts or final locations). Bars indicate the mean percentage of correct reports per task and switching demands. Error bars are SEM.
the verbal task subjects performed significantly better in low switching demand trials compared to medium (p ⬍ .05) and high (p ⬍ .05), which in turn were not different from each other. In the visuospatial task, no significant differences in switching demand were found and in the combined task subjects performed significantly worse in high switching demand trials compared to Medium (p ⬍ .01), but performed equally well in low and medium trials. In the medium level of switching demand, subjects performed best in the combined task (verbal p ⬍ .01, visuospatial p ⬍ .05); in low and high trials performance did not differ across tasks. All trials were included in the functional analysis, since previous data showed that trials in which errors in counting were made to be comparable to error-free trials in terms of switching costs (Garavan, 1998). In the present data, this was borne out by the nature of the errors in the incorrect trials. A total of 205 incorrect trials (13.6%) were seen across all subjects and tasks. Of these, on only 34 trials (16%) were both reports incorrect. On the incorrect counts trials, 92.4% of the errors consisted of the reported count being ⫾1 from the true count. The visuospatial task did not provide for a comparable analysis because of the limited number of locations within the 2 ⫻ 2 matrix. Although in this study we did not systematically assess participants’ strategy to update the count of circles and the location of dots, we know from previous studies that participants count subvocally when updating the circle counts (Garavan, 1998). For the visuospatial task, subjects reported anecdotally that they “moved the dot in their head,” indicating a visuospatial strategy. The lack of correlation in the nonswitch trials of the verbal and visuospatial tasks in Experiment 1 also indicates that participants used different strategies to accomplish the tasks. Functional activation Brain areas that increased in activation as a function of attention switching demand (i.e., High ⬎ Medium ⬎ Low), within each task were not restricted to the prefrontal lobes (voxelwise repeated measures ANOVA with switching de-
1303
mand as fixed factor and subjects as random factor calculated for each task separately). Instead, attention switching was associated with bilateral activation changes in a broadly distributed working memory network that included DLPFC, cingulate gyrus, basal ganglia, premotor areas, thalamus, parietal lobules, precuneus, temporal and occipital lobes, and cerebellum (Fig. 5). A combination of the attention switching activation maps of each task (OR map) revealed large overlap and yielded only a few separate clusters in prefrontal cortex. For this reason we restricted the ROI analysis to the prefrontal cortex. In general, brain areas involved in within-modality switching overlapped or fell adjacent to areas involved in between-modality switching (Fig. 5). The voxelwise 3 (TASK) ⫻ 3 (SWITCHING DEMAND) repeated-measures ANOVA with subjects as random factor on the activation maps yielded a main effect of switching demand for almost all activated areas (Fig. 5). A main effect of task was found in left precuneus, left inferior temporal gyrus, and left middle and superior occipital gyri. Pairwise comparisons within these regions revealed that the highest activation was associated with the visuospatial task. Activation levels in these areas were intermediate in the combined task and were not significantly greater than zero during the verbal task. No interaction between task and switching demand was found. Functionally-defined ROI analyses One exception to the largely overlapping activations of the attention switching activation maps were two separable bilateral anterior frontal clusters of activation (middle frontal gyrus and BA10, centre-of-mass: right x ⫽ ⫺41, y ⫽ ⫺41, z ⫽ 15, vol ⫽ 173 l; left x ⫽ 34, y ⫽ ⫺45, z ⫽ 18, vol ⫽ 119 l) which were only present in the betweenmodality switching condition (voxelwise ANOVA; Fig. 5 most anterior slice). However, a 2 (HEMISPHERE) ⫻ 3 (TASK) ⫻ 3 (SWITCHING DEMAND) repeated-measures ANOVA performed on mean activation within these functionally defined clusters revealed a significant main effect of switching demand (F2/26 ⫽ 28.64, p ⬍ .001; posthoc pairwise comparisons: High ⬎ Medium ⬎ Low, p ⬍ .05), but no main effect of task. A second, functionally-defined ROI analysis was performed on the bilateral DLPFC activation (Fig. 5 second most anterior slice). A discrete cluster of activation existed within the right hemisphere DLPFC (middle frontal gyrus and BA9, centre-of-mass: x ⫽ ⫺43, y ⫽ ⫺23, z ⫽ 29, vol ⫽ 496 l) and its rostral and caudal boundaries were used to demarcate the rostral and caudal boundaries of a more extensive left DLPFC activation (x ⫽ 26, y ⫽ ⫺22, z ⫽ 28, vol ⫽ 2186 l) which was contiguous with more posterior activations. A 2 (HEMISPHERE) ⫻ 3 (TASK) ⫻ 3 (SWITCHING DEMAND) repeated-measures ANOVA performed on mean activation within these clusters revealed a significant main effect of switching demand (F2/26 ⫽
1304
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 5. Areas showing a significant main effect of switching demand within each task are presented in an OR map. Blue: activation in the verbal task only, red: visuospatial only, green: combined only, yellow: overlapping areas. The activated areas were bilateral and comprised superior, middle, inferior and medial frontal gyri; anterior cingulate, cingulate, and posterior cingulate gyri; insula; basal ganglia; pre- and postcentral gyri; thalamus; inferior and superior parietal lobules, cuneus and precuneus, middle temporal and middle occipital gyri; fusiform gyrus; cerebellum—all bilateral.
70.39, p ⬍ .001; post-hoc pairwise comparisons: High ⬎ Medium ⬎ Low, p ⬍ .001), but no other main effects or interactions. In the voxelwise analyses, no significant activation was found in VLPFC. However, in light of our interest in the functional organization of prefrontal cortex, spherical VLPFC ROIs were defined by the centre-of-mass coordinates provided by Owen and colleagues (Owen et al., 1996) (x ⫽ ⫹/⫺47, y ⫽ ⫺22, z ⫽ 6; volume ⫽ 500 l to approximate the size of the right DLPFC cluster). Results revealed no significant activation (t-tests vs 0) within these ROIs. Larger clusters of 1000 l volume were also tested but yielded identical results, confirming the lack of activation in VLPFC. Prefrontal functional dissociations A 2 (REGION) ⫻ 2 (HEMISPHERE) ⫻ 3 (TASK) ⫻ 3 (SWITCHING DEMAND) repeated-measures ANOVA was performed on the bilateral anterior frontal and DLPFC functionally defined ROIs and revealed a significant main effect of switching demand (F2/26 ⫽ 67.58, p ⬍ .001; posthoc pairwise comparisons: High ⬎ Medium ⬎ Low, p ⬍ .05) and a significant interaction between region and task (F2/26 ⫽ 3.94, p ⬍ .05); posthoc pairwise comparisons were not significant. All other interactions and main effects were
not significant. Although only significant in the combined compared to the verbal task during high switching demands (p ⬍ .05 posthoc pairwise comparisons), activation in anterior frontal cortex tended to be higher for between-modality switching compared to within-modality switching (Fig. 6).
General discussion The same executive function—attention switching— when applied in the phonological, the visuospatial and combined working memory modalities activated similar and broadly distributed areas of the brain that have repeatedly been found to subserve working memory (D’Esposito et al., 1998; D’Esposito and Grossman, 1996; Haxby et al., 2000). Moreover, certain functional dissociations were found within the frontal lobes, with anterior frontal and dorsolateral cortex being differentially involved in within- and between-modality switching. Further, while dorsolateral prefrontal cortex was implicated in both within- and betweenmodality attention switching, no activation was found in ventrolateral areas for this function. These findings have implications for our understanding of the frontal lobes and how executive processes might be instantiated in the brain.
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Fig. 6. (A) Interaction between region and task. Anterior frontal cortex was most activated during performance of the combined task whereas highest DLPFC activation was associated with the visuospatial task (not significant). (B and C) Mean activation in functionally-defined regions of interest collapsed over hemisphere in anterior frontal cortex and dorsolateral prefrontal cortex. Error bars are SEM.
Brain correlates of attention switching within working memory The allocation of attentional resources within working memory was not accomplished solely by prefrontal cortex. Instead, activation in the full working memory network, including extensive parietal and premotor areas, varied as a function of the tasks’ attention switching demands. Others have also implicated posterior brain areas in similar cognitive processes. For example, Kimberg and colleagues (Kimberg et al., 2000) found regions in the superior parietal lobule and in the right DLPFC that were involved in switching between letter and number tasks. Comparing switch to nonswitch trials showed more neural activity in frontal, parietal, occipital and subcortical areas before and during the processing of switch trials. A possible involvement of right posterior parietal areas (BA40) in attention switching between representations within verbal working memory has
1305
also been reported (Jonides et al., 1998). Furthermore, Dove and colleagues (Dove, et al., 2000) could not identify any cortical areas that were exclusively activated during switching trials in an unpredictable task switching paradigm. Instead, using event-related fMRI, a widespread network, including bilateral lateral prefrontal, premotor cortex and insula, left intraparietal sulcus, the SMA/pre-SMA region and the cuneus and precuneus, was found to be involved in task switching. Parietal cortex is often considered the seat of storage and maintenance processes in verbal and spatial working memory tasks (Bunge et al., 2000; Hartley and Speer, 2000; Honey et al., 2000; Jonides et al., 1998; Paulesu et al., 1993; Salmon et al., 1996; Smith et al., 1998). For example, Postle and colleagues (Postle et al., 1999) reported a dissociation of mnemonic and executive processes within working memory with the former located in the inferior parietal lobule (BA40) and the latter represented in DLPFC. However, others have attributed an involvement in executive control to parietal cortices. For example, Le and colleagues (Le et al., 1998) found robust activation in the superior posterior parietal lobule, cuneus and precuneus during attention switching between two targets and proposed that posterior parietal cortices compute spatial codes involved in binding of features for an attended stimulus. Sohn and colleagues (Sohn, et al., 2000) presented their subjects with an alternating task switching paradigm in which 50% of the trials were cued. The authors concluded that superior posterior parietal cortex may be involved in endogenous goal-directed preparation for a task set, whereas other areas including inferior posterior parietal cortex may be responsible for stimulus-driven completion of adopting a task set. The present data demonstrated widespread parietal activation that was sensitive to switching demand but was not task specific, suggesting that parietal lobe contributions to working memory go beyond stimulus representation, storage and maintenance. Instead, we suggest that parietal cortices may be necessary for the allocation of attentional resources within working memory and may contribute, therefore, to executive control. The present results have found extensive cortical areas, not just parietal or prefrontal, associated with the tasks’ executive demands. This would suggest that attention switching might be reflected as a dynamic property of the larger task-activated working memory network rather than as a specific locus of activation. Additional data on this attention switching task suggests that switching times include a retrieval mechanism such that switching to an item in working memory entails the re-activation of the new item (Voigt and Hagendorf, 2002). Furthermore, switching away from an item in working memory may entail the suppression of that old item (Voigt and Hagendorf, 2002) as has been observed for switching between tasks (Mayr and Kliegl, 2000). These processes, re-activation and suppression of the contents of working memory may explain, in part, why such widespread functional activation of the working memory
1306
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
network is observed. However, such a conclusion must be reconciled with the pre-eminent role typically given to the frontal lobes for executive functioning (D’Esposito et al., 2000; Owen et al., 1998; Smith and Jonides, 1999; Wagner et al., 2001) and the prefrontal dissociations, discussed below, observed for different types of attention switching. One possibility is that executive functioning might be divided into two components. The first, associated with the prefrontal lobes, may select and initiate the function to be performed in accordance with the task rules and goals that are stored there. This is consistent with the aimless or inappropriate behaviour and behavioural inertia often evident in frontal patients (Anderson, et al., 2000; Manes et al., 2002; Zalla et al., 2001). The posterior brain regions, on the other hand, may be central to the enactment of the switch through the re-activation and suppression of the working memory contents, as described above (Bunge et al., 2002). Thus, storage and reactivation of information in parietal areas may be the actual means by which attention switching occurs. Such a conceptualisation, while still ascribing a representation storage role to posterior areas posits that dynamic changes within and between those representations may be the “machinery” by which the switch is accomplished. This conceptualisation permits for prefrontal-posterior dissociations while also explaining why widespread activation patterns might underlie the attention switching function in tasks such as these. It may, however, also call into question a firm distinction between a storage and an executive function. Functional specificity within the prefrontal cortex All tasks produced activation within dorsolateral areas that varied with switching demand, but no activation was found within ventrolateral areas thereby confirming the importance of bilateral DLPFC for executive control in a working memory task. These results are consistent with a processing-specific organization of the PFC with DLPFC being activated when representations in working memory, irrespective of modality, have to be manipulated (Barde and Thompson-Schill, 2002; Kessels et al., 2000; Owen et al., 1996; Owen et al., 1998; Wagner et al., 2001). The greater, but not significant, DLPFC activation for the visuospatial task might be interpreted as consistent with a domain-specific organization of the PFC (Goldman-Rakic, 1996; Gruber and von Cramon, 2001). However, a domain-specific approach would have to explain why activation of DLPFC was not higher in the combined task with its visuospatial component, compared to the verbal task and why DLPFC was activated in all tasks, including the verbal task which was completely non-spatial (Nystrom et al., 2000; Owen et al., 1998; Postle et al., 2000). The data support a functional dissociation between anterior frontal and dorsolateral cortex for manipulating infor-
mation within and between the working memory slave systems. The results suggest that anterior frontal cortex may be recruited when coordination between the two slave systems of working memory is necessary. Koechlin and colleagues (Koechlin et al., 1999) report recruitment of frontopolar cortex (BA10) in a task involving the cognitive process of branching, which refers to the process of allocating resources when attention has to be alternated between two concurrent activities. The authors concluded that frontopolar cortex is recruited when main goals have to be kept in mind during processing of subgoals. Although more anterior than our anterior frontal cluster the Koechlin et al. findings may be relevant to our results. Behavioural data in our study indicate that subjects performed better during between modality switching compared to within modality switching (significant for the medium switching demand). This is in agreement with the results of Experiment 1, which revealed less time cost in between modality switch trials compared to within modality switch trials. This effect may be explained by less interference between attentional resources. Duncan and colleagues (Duncan et al., 1997) have shown that response accuracy is reduced when attention is divided within but not between sensory modalities. The authors presented their subjects with two visual and two auditory streams in which they had to detect targets. Detecting a target in one stream led to sustained reduction of the ability to identify the second target in the other stream. No such effect was seen in the combined presentation of one visual and one auditory stream. The authors concluded that visual attention to a target did not interfere with the auditory attention to another. Since in our study anterior frontal areas were most activated in the combined task one might speculate that they were additionally recruited because coordination of attentional resources between the two slave systems of working memory was necessary. Incorporating these findings with Koechlin’s, it may be the case that anterior frontal cortex is recruited when two sufficiently distinct tasks must be actively maintained. It is noteworthy that anterior frontal areas were more activated during the combined task, which was also the easiest to perform, suggesting that this additional activation cannot simply be attributed to task difficulty. No support for a hemispheric organisation of PFC was found. It has previously been suggested that domain specificity across hemispheres is more pronounced than between dorsal and ventral regions (Ungerleider et al., 1998). With the exception of the involvement of left visual areas in the visuospatial task, the present results found no hemispheric differences in activation in relation to the task requirements. No hemispheric differences in dorsal, ventral or anterior frontal PFC were found. Such a failure to find hemispheric stimulus specificity was also reported by Nystrom and colleagues (Nystrom et al., 2000) who presented their subjects with verbal, spatial and non-verbal and non-spatial stimuli in an n-back task.
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Conclusion Two patterns of brain activation changes were found for attention switching and the dynamics of coordinating the working memory slave systems. First, increased activation of the entire task-related network was observed for switching both within and between modalities. Second, additional recruitment of anterior frontal regions into this network was observed for switching between modalities. However, it should be noted that activation in the anterior frontal areas, although sub-threshold for the within-modality tasks, did modulate with switching demand in these tasks. Consequently, while we have proposed differences in the executive function roles of the prefrontal and posterior areas, these results do challenge the notion of rigid anatomically separable modules within working memory. In contrast, Smith and Jonides (Smith and Jonides, 1999) argue for a clear distinction between short-term storage and executive processes as a main organizational principle of the PFC. As we held constant online storage and rehearsal demands, areas that were exclusively involved in these working memory components should not have increased in activation with increased switching demand. Furthermore, even those areas that showed a main effect of task, including the posterior areas that were specific to the visuospatial task, also showed switching demand effects. Independent of task and modality, attention switching activated a widespread frontoparietal network, consistent with a trend away from a strict localization of psychological processes to cortical areas to the view that multiple brain areas are involved in cognitive functioning to different degrees (Carpenter et al., 2000; Haxby et al., 2000). Acknowledgments The assistance of Catherine Fassbender and Eoin Mooney is gratefully acknowledged. This study is supported by NIDA grant RO1 DA14100 and GCRC M01 RR00058. References Anderson, S.W., Damasio, H., Tranel, D., Damasio, A.R., 2000. Long-term sequelae of prefrontal cortex damage acquired in early childhood. Developmental Psychology 18, 281–296. Baddeley, A., Hitch, G.J., 1974. Working memory, in: Bower, G.H. (Ed.), The Psychology of Learning and Motivation, Vol. 8, Academic Press, New York, pp. 47– 89. Barde, L.H.F., Thompson-Schill, S.L., 2002. Models of functional organization of the lateral prefrontal cortex in verbal working memory: evidence in favor of the process model. Journal of Cognitive Neuroscience 14, 1054 –1063. Bunge, S.A., Hazeltine, E., Scanlon, M.D., Rosen, A.C., Gabrieli, J.D.E., 2002. Dissociable contribution of prefrontal and parietal cortices to response selection. Neuroimage 17, 1562–1571. Bunge, S.A., Klingberg, T., Jacobsen, R.B., Gabrieli, J.D.E., 2000. A resource model of the neural basis of executive working memory.
1307
Proceedings of the National Academy of Sciences of the United States of America 97, 3573–3578. Carpenter, P.A., Just, M.A., Reichle, E.D., 2000. Working memory and executive function: evidence from neuroimaging. Current Opinion in Neurobiology 10, 195–199. Chein, J.M., Fiez, J.A., 2001. Dissociation of verbal working memory system components using a delayed serial recall task. Cerebral Cortex 11, 1003–1014. Cocchini, G., Logie, R.H., Della Sala, S., MacPherson, S.E., Baddeley, A.D., 2002. Concurrent performance of two memory tasks: evidence for domain-specific working memory systems. Memory and Cognition 30 (7), 1086 –1095. Collette, F., Salmon, E., van der Linden, M., Chicherio, C., Belleville, S., Degueldre, C., Delfiore, G., Franck, G., 1999. Regional brain activity during tasks devoted to the central executive of working memory. Cognitive Brain Research 7, 411– 417. Courtney, S.M., Ungerleider, L.G., Keil, K., Haxby, J.V., 1996. Object and spatial visual working memory activate separate neural systems in human cortex. Cerebral Cortex 6, 39 – 49. Cowan, N., 1988. Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human informationprocessing system. Psychological Bulletin 104, 163–191. Cowan, N., 1993. Activation, attention, and short-term memory. Memory and Cognition 21 (2), 162–167. Cox, R.W., 1996. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Computers and Biomedical Research 29, 162–173. D’Esposito, M., Aguirre, G.K., Zarahn, E., Ballard, D., Shin, R.K., Lease, J., 1998. Functional MRI studies of spatial and nonspatial working memory. Cognitive Brain Research 7, 1–13. D’Esposito, M., Grossman, M., 1996. The physiological basis of ececutive function and working memory. The Neuroscientist 2, 345–352. D’Esposito, M., Postle, B.R., Rypma, B., 2000. Prefrontal cortical contributions to working memory: evidence from event-related fMRI studies. Experimental Brain Research 133, 3–11. Diwadkar, V.A., Carpenter, P., Just, M.A., 2000. Collaborative activity between parietal and dorso-lateral prefrontal cortex in dynamic spatial working memory revealed by fMRI. Neuroimage 12, 85–99. Dove, A., Pollman, S., Schubert, T., Wiggins, C.J., von Cramon, D.Y., 2000. Prefrontal cortex activation in task switching: an event-related fMRI study. Cognitive Brain Research 9, 103–109. Duncan, J., Martens, S., Ward, R., 1997. Restricted attentional capacity within but not between sensory modalities. Nature 387, 808 – 810. Farmer, E.W., Berman, J.V.F., Fletcher, Y.L., 1986. Evidence for a visuospatial scratch-pad in working memory. The Quarterly Journal of Experimental Psychology 38A, 675– 688. Garavan, H., 1998. Serial attention within working memory. Memory and Cognition 26, 263–276. Garavan, H., Ross, T.J., Li, S.J., Stein, E.A., 2000. A parametric manipulation of central executive functioning. Cerebral Cortex 10 (6), 585– 592. Goldman-Rakic. 1996. The prefrontal landscape: implications of functional architecture for understanding human mentation and the central executive. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 351, 1445–1453. Gruber, O., von Cramon, D.Y., 2001. Domain-specific distribution of working memory processes along human prefrontal and parietal cortices: a functional magnetic resonance imaging study. Neuroscience Letters 297, 29 –32. Halgren, E., Boujon, C., Clarke, J., Wang, C., Chauvel, P., 2002. Rapid distributed fronto-parieto-occipital processing stages during working memory in humans. Cerebral Cortex 12, 710 –728. Hartley, A.A., Speer, N.K., 2000. Locating and fractionating working memory using functional neuroimaging: storage, maintenance, and executive functions. Microscopy Research and Technique 51, 45–53.
1308
A. Ku¨ bler et al. / NeuroImage 20 (2003) 1298 –1308
Haxby, J.V., Petit, L., Ungerleider, L.G., Courtney, S.M., 2000. Distinguishing the functional roles of multiple regions in distributed neural systems for visual working memory. Neuroimage 11, 380 –391. Honey, G.D., Bullmore, E.T., Sharma, T., 2000. Prolonged reaction time to a verbal working memory task predicts increased power of posterior parietal cortical activation. Neuroimage 12, 495–503. Jonides, J., Schumacher, E.H., Smith, E.E., Koeppe, R.A., Awh, E., Reuter-Lorenz, P.A., Marshuetz, C., Willis, C.R., 1998. The role of parietal cortex in verbal working memory. Journal of Neuroscience 18, 5026 –5034. Kessels, R.P., Postma, A., Wijnalda, E.M., de Haan, E.H., 2000. Frontallobe involvement in spatial memory: evidence from PET, fMRI, and lesion studies. Neuropsychology Review 10 (2), 101–113. Kimberg, D.Y., Aguirre, G.K., D’Esposito, M., 2000. Modulation of taskrelated neural activity in task-switching: an fMRI study. Cognitive Brain Research 10, 189 –196. Koechlin, E., Basso, G., Pietrini, P., Panzer, S., Grafman, J., 1999. The role of the anterior prefrontal cortex in human cognition. Nature 399, 148 –151. LaBar, K.S., Gitelman, D.R., Parrish, T.B., Mesulam, M.-M., 1999. Neuroanatomic overlap of working memory and spatial attention networks: a functional MRI comparison within subjects. Neuroimage 10, 695– 704. Le, T.H., Pardo, J.V., Hu, X., 1998. 4 T-fMRI study of nonspatial shifting of selective attention: cerebellar and parietal contributions. Journal of Neurophysiology 79, 1535–1548. Manes, F., Sahakian, B., Clarke, L., Rogers, R., Antoun, N., Aitken, M., Robbins, T., 2002. Decision-making processes following damage to the prefrontal cortex. Brain 125, 624 – 639. Mayr, U., Kliegl, R., 2000. Task-set switching and long-term memory retrieval. Journal of Experimental Psychology. Learning, Memory, and Cognition 26 (5), 1124 –1140. Nystrom, L.E., Braver, T.S., Sabb, F.W., Delgado, M.R., Noll, D.C., Cohen, J., 2000. Working memory for letters, shapes, and locations: fMRI evidence against stimulus-based regional organization in human prefrontal cortex. Neuroimage 11, 424 – 446. Owen, A.M., 1997. The functional organisation of working memory processes within human lateral frontal cortex: the contribution of functional neuroimaging. European Journal of Neuroscience 9, 1329 –1339. Owen, A.M., Evans, A.C., Petrides, M., 1996. Evidence for a two-stage model of spatial working memory processing within the lateral frontal cortex: a positron emission tomography study. Cerebral Cortex 6, 31–38. Owen, A.M., Stern, C.E., Look, R.B., Tracey, I., Rosen, B.R., Petrides, M., 1998. Functional organization of spatial and nonspatial working memory processing within the human lateral frontal cortex. Proceedings of the National Academy of Sciences of the United States of America 95, 7721–7726. Paulesu, E., Frith, C.D., Frackowiak, R.S., 1993. The neural correlates of the verbal component of working memory. Nature 362 (6418), 342– 345. Petrides, M., Alivisatos, B., Evans, A.C., Meyer, E., 1993. Dissociation of human mid-dorsolateral from posterior dorsolateral frontal cortex in
memory processing. Proceedings of the National Academy of Sciences of the United States of America 90, 873– 877. Postle, B.R., Berger, J.S., D’Esposito, M., 1999. Functional neuroanatomical double dissociation of mnemonic and executive control processes contributing to working memory performance. Proceedings of the National Academy of Sciences of the United States of America 96, 12959 –12964. Postle, B.R., Stern, C.E., Rosen, B.R., Corkin, S., 2000. An fMRI investigation of cortical contributions to spatial and nonspatial visual working memory. Neuroimage 11, 409 – 423. Raye, C.L., Johnson, M.K., Mitchell, K.J., Reeder, J.A., Greene, E.J., 2002. Neuroimaging a single thought: dorsolateral PFC activity associated with refreshing just-activated information. Neuroimage 15, 447– 453. Rypma, B., Berger, J.S., D’Esposito, M., 2002. The influence of working memory demand and subject performance on prefrontal cortical activity. Journal of Cognitive Neuroscience 14, 721–731. Salmon, E., van der Linden, M., Collette, F., Delfiore, G., Maquet, P., Degueldre, C., Luxen, A., Franck, G., 1996. Regional brain activity during working memory tasks. Brain 119, 1617–1625. Shallice, T., Burgess, P., 1996. The domain of supervisory processes and temporal organization of behaviour. Philosophical Transactions of the Royal Society of London. Series B. Biological Sciences 351 (1346), 1405–1411 discussion 1411–1412. Smith, E.E., Jonides, J., 1999. Storage and executive processes in the frontal lobes. Science 283, 1657–1661. Smith, E.E., Jonides, J., Koeppe, R.A., 1996. Dissociating verbal and spatial working memory using PET. Cerebral Cortex 6, 11–20. Smith, E.E., Jonides, J., Marshuetz, C., Koeppe, A.C., 1998. Components of verbal working memory: evidence from neuroimaging. Proceedings of the National Academy of Sciences of the United States of America 95, 876 – 882. Sohn, M.-H., Ursu, S., Anderson, J.R., Stenger, V.A., Carter, C.S., 2000. The role of prefrontal cortex and posterior parietal cortex in task switching. Proceedings of the National Academy of Sciences of the United States of America 97, 13448 –13453. Talairach, J., Tourneaux, P., 1988. Co-planar Stereotaxic Atlas of the Human Brain. Thieme Medical, New York. Ungerleider, L.G., Courtney, S.M., Haxby, J.V., 1998. A neural system for human visual working memory. Proceedings of the National Academy of Sciences of the United States of America 95, 883– 890. Voigt, S., Hagendorf, H., 2002. The role of task context for component processes in focus switching. Psychologische Beitraege 44, 248 –274. Wagner, A.D., Maril, A., Bjork, R.A., Schacter, D.L., 2001. Prefrontal contributions to executive control: fMRI evidence for functional distinctions within lateral prefrontal cortex. Neuroimage 14, 1337–1347. Wong, E.C., Boskamp, E., Hyde, J.S., 1992. A volume optimized quadratureelliptical endcap birdcage brain coil. Paper presented at the Eleventh Annual Scientific Meeting, Society for Magnetic Resonance Medicine, Berlin. Zalla, T., Plassiart, C., Pillon, B., Grafman, J., Sirigu, A., 2001. Action planning in a virtual context after prefrontal cortex damage. Neuropsychologia 39, 759 –770.