Elsevier

NeuroImage

Volume 37, Issue 1, 1 August 2007, Pages 343-360
NeuroImage

The cognitive control network: Integrated cortical regions with dissociable functions

https://doi.org/10.1016/j.neuroimage.2007.03.071Get rights and content

Abstract

Consensus across hundreds of published studies indicates that the same cortical regions are involved in many forms of cognitive control. Using functional magnetic resonance imaging (fMRI), we found that these coactive regions form a functionally connected cognitive control network (CCN). Network status was identified by convergent methods, including: high inter-regional correlations during rest and task performance, consistently higher correlations within the CCN than the rest of cortex, co-activation in a visual search task, and mutual sensitivity to decision difficulty. Regions within the CCN include anterior cingulate cortex/pre-supplementary motor area (ACC/pSMA), dorsolateral prefrontal cortex (DLPFC), inferior frontal junction (IFJ), anterior insular cortex (AIC), dorsal pre-motor cortex (dPMC), and posterior parietal cortex (PPC). We used a novel visual line search task which included periods when the probe stimuli were occluded but subjects had to maintain and update working memory in preparation for the sudden appearance of a probe stimulus. The six CCN regions operated as a tightly coupled network during the ‘non-occluded’ portions of this task, with all regions responding to probe events. In contrast, the network was differentiated during occluded search. DLPFC, not ACC/pSMA, was involved in target memory maintenance when probes were absent, while both regions became active in preparation for difficult probes at the end of each occluded period. This approach illustrates one way in which a neuronal network can be identified, its high functional connectivity established, and its components dissociated in order to better understand the interactive and specialized internal mechanisms of that network.

Introduction

An act as common as searching for a friend in a crowd can involve many cognitive control processes. Such processes include, among others, target working memory (keeping in mind what your friend looks like), attention to stimuli (viewing the individuals in the crowd), target–stimulus comparison (deciding if the viewed individual is your friend), response preparation (determining what to do if you see your friend), and response initiation (initiating the planned motor response to get your friend's attention). These same components are involved in a large variety of tasks, both in daily life and in the cognitive control neuroimaging literature. This is the case because any novel or conflict-laden task requires a set of instructions/intentions to be dynamically converted to stimulus–response (S–R) associations for proper task performance. As illustrated above, this conversion can involve a number of processes.

A collection of neural components of similar number to the above collection of cognitive components is consistently involved in a large variety of cognitive control tasks as well (see Cabeza and Nyberg, 2000, Duncan and Owen, 2000, Schneider and Chein, 2003, Wager et al., 2004, Brass et al., 2005, Chein and Schneider, 2005, Dosenbach et al., 2006). These neural components include dorsolateral prefrontal cortex (DLPFC), anterior cingulate cortex/pre-supplementary motor area (ACC/pSMA), dorsal premotor cortex (dPMC), anterior insular cortex (AIC), inferior frontal junction (IFJ), and posterior parietal cortex (PPC). The present study examines the proposal that this set of regions forms a cognitive control network (CCN) of anatomically distinct component processing brain regions that interact in a tightly coupled fashion to implement cognitive control in a variety of task contexts.

The trouble with tightly coupled networks is that their components can be difficult to clearly dissociate, and the CCN is no exception. As mentioned above, most cognitive control tasks involve the same set of component processes. Additionally, most of these processes have to occur in the short temporal window between stimulus and response (∼ 500 ms typically). This makes it difficult to separate task processes with a technique such as functional magnetic resonance imaging (fMRI) which has a multi-second response function. Also, many of these processes (e.g., working memory and target–stimulus comparison) must interact in order for information from one process to influence the other. The highly interactive nature of these processes makes component separation problematic, though factorial analysis (e.g., showing one region has greater sensitivity to difficulty relative to another) can be used to separate these components in some cases.

The present study employed an alternative approach in which the S–R processes were separated from the initiation, maintenance, and preparation of the CCN for task performance. This allowed working memory initiation and maintenance to be observed separately from response preparation, and all three processes to be observed separately from the many components of S–R processing. A mixed blocked/event-related task design, which can be used to separate estimates of sustained from transient neural processes (Visscher et al., 2003), was used to accomplish these component separations.

Duncan and Owen (2000) pointed out that two of the regions mentioned above, DLPFC and ACC/pSMA, have especially highly correlated activation patterns across a large variety of task demands across a large number of studies. There are very few claims of having separated these regions despite the many studies reporting activity in them. MacDonald et al. (2000) have made one of the strongest claims to this effect. They found that cues indicating hard vs. cues indicating easy upcoming probes showed a larger response in DLPFC (not ACC/pSMA), while comparing hard vs. easy probes showed a larger response in ACC/pSMA (not DLPFC). This result has not been supported by several recent studies looking at preparatory cues (see Luks et al., 2002, Brown and Braver, 2005, Luks et al., 2007, Schumacher et al., 2007). For instance, Schumacher et al. (2007) found that DLPFC activity was increased by both cue and probe difficulty.

A number of studies have shown that DLPFC responds to working memory demands while ACC/pSMA does not (e.g., Barch et al., 1997, Manoach et al., 1997, Smith et al., 1998, Ranganath and D'Esposito, 2001). This is perhaps the most consistently characterized difference between the two regions, and yet there are a number of studies that have shown ACC/pSMA activity during working memory delays (see Wager and Smith, 2003 for review). For instance, Petit et al. (1998) showed sustained activity within ACC/pSMA during both spatial and face working memory maintenance.

We hypothesized that ACC/pSMA is involved during working memory delays due to its involvement in preparatory processes, while DLPFC is involved during working memory delays due to active maintenance of task goal information. We hypothesized that both regions are involved in S–R processing.

In order to test these hypotheses, we used a modified visual search task, the occluded target switching paradigm, to separate target maintenance and preparatory activity from S–R processing activity. The basic task involved presentation of a line orientation every second with the goal being to press a button whenever the target orientation appeared. In this modified task, target maintenance was separated from probe processing using ‘occluded’ periods in which very few probes occurred and the main task demand was to actively maintain the present target in memory (“non-switch trials”; see Fig. 1A). In another occluded condition, targets were switched internally (i.e., working memory was updated) every 4 s between two highly familiar line orientations (i.e., “/” and “\”), such that the main task demand was not only to maintain the current target but also to update the current target periodically (“target switch trials”; see Fig. 1B).

We expected DLPFC, not ACC/pSMA, to be active throughout both of these occluded periods. We expected ACC/pSMA to be active as well during the final third of the target switching occluded period, since subjects were much more likely to prepare for an upcoming probe during this time. Preparation was more likely toward the end of the occluded period because there was a much higher probability of a probe at the end relative to the rest of each occluded period (∼ 10× increased chance; learned through practice with the task). We expected that less preparation-related activity would be present during the non-switching condition since S–R mappings remained consistent throughout each of these trials, while in the target switching condition S–R mappings were variable, causing an increased need for control (Schneider and Shiffrin, 1977) and thus increasing the need for preparation for these more demanding probes.

DLPFC and ACC/pSMA are not the only regions that are coactive across a large variety of cognitive control task demands. As mentioned above, a consensus has emerged that a set of six regions (DLPFC, ACC/pSMA, dPMC, IFJ, AIC, and PPC) is involved in core cognitive control functions. We hypothesized that this set of cognitive control regions forms a functionally connected network for implementing cognitive control.

In accordance with the functional integration of these regions, as well as its proposed role in cognitive control, we predicted that the entire network would be active during S–R processing and many/all areas of the network would show sensitivity to S–R processing difficulty. We used a varied mapping manipulation (see Schneider and Shiffrin, 1977) in which subjects searched for either a 45° or 135° oriented line (i.e., “/” or “\”) among serially presented probes consisting of 0°, 90°, 45°, or 135° (i.e., “—”, “|”, “/”, or “\”). There were three types of probe S–R mappings: distracters, targets, and foils. Pure distracters were probes (0° or 90°) that never required a response and were consistently irrelevant to the task throughout the experiment. Distracters required very little cognitive control since once recognized they could be easily ignored. Target and foil probes were selected from a common pool of line stimuli. Therefore, the current foil probe was previously a target and vice versa. To illustrate: assume in trial 1 the subject searched for “\” and on trial 2 for “/”. On trial 2 the appearance of “–” would be a distracter, “/” would be the target, and “\” the foil. Trial 1 had built up the association that “\” was the target, but on trial two this association was not only irrelevant but also conflicting with the correct S–R process.

The conflict engendered by this S–R inconsistency manipulation was of the most demanding type (negative priming; see Malley and Strayer, 1995). We hypothesized that the entire set of regions, the CCN, would respond to these high conflict situations because they are the most demanding for cognitive control processes. This finding would support the central role of these regions in cognitive control, and it would also support their hypothesized functional unity.

Inter-regional functional connectivity was assessed in order to firmly establish the functional unity of the CCN. The proposal that the CCN is a highly integrated functional network predicts (1) that the network has high internal functional connectivity. Further, it predicts (2) that this internal functional connectivity is higher than CCN connectivity to non-CCN regions, and also (3) higher than average connectivity between all cortical regions. An additional prediction inherent in the proposed role of this network (i.e., monitoring and controlling much of cortex) is (4) that CCN connectivity to non-CCN regions is higher than the average connectivity between all cortical regions.

Resting state correlations in fMRI signals were used to assess functional connectivity in order to rule out connectivity patterns tied to any specific task. The fluctuations underlying these correlations likely originate from spontaneous slow-wave electrical activity across cortex. Golanov et al. (1994) showed that such spontaneous electrical activity occurs across cortex in anaesthetized rats, and is followed by increases in regional cerebral blood flow lasting ∼ 12 s. Such increases in blood flow are known to increase the blood oxygen level dependent (BOLD) signal, as measured by fMRI (Ogawa et al., 1992). Furthermore, Golanov et al. found that these spontaneous waves of electrical activity and the corresponding blood flow changes (R = 0.94 between these events) occurred at ∼ 0.1 Hz, which is approximately the same rate as resting state BOLD fluctuations in humans (Cordes et al., 2001). These slow BOLD fluctuations are correlated between functionally connected regions (Xiong et al., 1999) likely because when two regions are anatomically connected with significant synaptic weights these connections do not ‘turn off’ when not being used for task performance. Instead, any spontaneous neural firing in one region will likely cause an increase in action potential propagation across even long-distance axonal connections. Thus, once potential confounding signals are dealt with (see Materials and methods), resting state BOLD correlations likely reflect functional connectivity as utilized by task performance. Supporting this conclusion, Cordes et al. (2000) found very similar patterns across cortex when comparing fMRI functional activity (using motor, language, and visual tasks) and cortex-wide resting state correlations. BOLD correlations between well-known functionally and anatomically connected regions (e.g., LIP and FEF) have been found in anaesthetized monkeys as well (Vincent et al., 2007).

Section snippets

Subjects

We included nine right-handed subjects (7 male, 2 female), aged 19 to 42 in the study. These subjects were recruited from the University of Pittsburgh and surrounding area. Subjects were excluded if they had any medical, neurological, or psychiatric illness, any contraindications for MRI scans, or were left-handed. All subjects gave informed consent.

Cognitive task

The basic task was to detect a target line orientation by comparing it to each serially presented stimulus (see Fig. 1). The target could be one of

Behavioral results

Subjects were correct 98% of the time on average (SD: 1.4%). Accuracy dropped during the occluded conditions, but remained high at 86% (SD: 10.9%). This shows that subjects were compliant during the occluded periods, in which they had to maintain or switch the target stimulus during long blank periods with infrequent probes.

Reaction time data are only available for target trials since these were the only trials with a button response (there were not enough false alarms for inclusion in the

Discussion

In the present study, we hypothesized that a set of cortical regions consistently co-active during cognitive control tasks forms a network (the CCN). We used resting state correlations to characterize and compare functional connectivity within this proposed network. Results showed that the CCN regions were better correlated within the network than to regions outside the network, indicating that calling this set of regions a ‘network’ is meaningful in the context of the network that constitutes

Acknowledgments

This work was funded by a grant from the Office of Naval Research (N00014-01-10677). The authors wish to thank Eliezer Kanal, Scott Kurdilla, and Becca Smith for technical support.

References (52)

  • T.L. Luks et al.

    Evidence for anterior cingulate cortex involvement in monitoring preparatory attentional set

    NeuroImage

    (2002)
  • T.L. Luks et al.

    Preparatory allocation of attention and adjustments in conflict processing

    NeuroImage

    (2007)
  • T. Nichols et al.

    Valid conjunction inference with the minimum statistic

    NeuroImage

    (2005)
  • C. Ranganath et al.

    Medial temporal lobe activity associated with active maintenance of novel information

    Neuron

    (2001)
  • W. Schneider et al.

    Controlled and automatic processing: behavior, theory, and biological mechanisms

    Cogn. Sci.

    (2003)
  • E.H. Schumacher et al.

    Selection and maintenance of stimulus-response rules during preparation and performance of a spatial choice-reaction task

    Brain Res.

    (2007)
  • K.M. Visscher et al.

    Mixed blocked/event-related designs separate transient and sustained activity in fMRI

    NeuroImage

    (2003)
  • T.D. Wager et al.

    Neuroimaging studies of shifting attention: a meta-analysis

    NeuroImage

    (2004)
  • K. Bornhovd et al.

    Painful stimuli evoke different stimulus-response functions in the amygdala, prefrontal, insula and somatosensory cortex: a single-trial fMRI study

    Brain

    (2002)
  • M.M. Botvinick et al.

    Conflict monitoring and cognitive control

    Psychol. Rev.

    (2001)
  • T.S. Braver et al.

    Anterior cingulate cortex and response conflict: effects of frequency, inhibition and errors

    Cereb. Cortex

    (2001)
  • J.W. Brown et al.

    Learned predictions of error likelihood in the anterior cingulate cortex

    Science

    (2005)
  • R. Cabeza et al.

    Imaging cognition II: an empirical review of 275 PET and fMRI studies

    J. Cogn. Neurosci.

    (2000)
  • C. Constantinidis et al.

    Neuronal activity in posterior parietal area 7a during the delay periods of a spatial memory task

    J. Neurophysiol.

    (1996)
  • D. Cordes et al.

    Mapping functionally related regions of brain with functional connectivity MR imaging

    AJNR Am. J. Neuroradiol.

    (2000)
  • D. Cordes et al.

    Frequencies contributing to functional connectivity in the cerebral cortex in resting-state data

    AJNR Am. J. Neuroradiol.

    (2001)
  • Cited by (785)

    View all citing articles on Scopus
    View full text