We tested the hypothesis that the frontoparietal control network would be flexibly engaged with either the default or dorsal attention network in support of goal-directed cognition. In a data-driven multivariate PLS analysis, we found that autobiographical planning, like imagining personal future events, engages the default network. Consistent with previous observations, we also observed activity in the dorsal attention network while subjects engaged in visuospatial planning (Baker et al., 1996
; Morris et al., 1993
; Newman et al., 2003
). Critically, both planning tasks engaged the frontoparietal control network. Task-related functional connectivity analyses indicated that the frontoparietal control network coupled its activity with the default network during autobiographical planning; furthermore, the frontoparietal control network coupled its activity with the dorsal attention network during visuospatial planning. These findings demonstrate that the frontoparietal control network can flexibly couple activity with the default or dorsal attention networks in support of goal-directed cognition.
The data-driven PLS results showed a pattern of task-related activity that was strikingly similar to previously observed intrinsic resting-state networks (Fox et al., 2005
; Vincent et al., 2008
; and Supplemental Fig. 3
). This similarity is all the more remarkable when considering that the same networks were identified in independent experimental runs, and using different analytic techniques to detect patterns of covarying BOLD signal at different frequency ranges. In the rsfcMRI analysis, we replicated the default, dorsal attention and frontoparietal control networks (Vincent et al., 2008
). We used these resting-state networks as a priori
ROIs to further interrogate the PLS results and quantify the degree and specificity of the task-related activation of these networks. We also performed complementary univariate ROI analyses to the same end. The PLS results showed that autobiographical planning significantly and reliably engaged both the default and frontoparietal control networks; additionally, visuospatial planning significantly and reliably engaged the dorsal attention and frontoparietal control networks (). Univariate analyses provided converging evidence for these observations, and were consistent whether we compared changes in BOLD signal relative to the counting condition or relative to trial onset (fixation): Percent BOLD signal significantly increased within the default and frontoparietal control networks for autobiographical planning, and within the dorsal attention network and frontoparietal control network for visuospatial planning (). Together, these findings confirm that goal-directed planning engages the frontoparietal control network, and is coactive with the default network or dorsal attention network, depending on task domain.
Results from the task-related functional connectivity analysis support the interpretation that the frontoparietal control network is not only co-active with the default or dorsal attention network, but couples its activity with each of these networks, depending on task domain. The temporal brain scores (i.e. composite measures of the neural activity at each timepoint) from the initial PLS analysis showed a high degree of correspondence in activity. Likewise, the BOLD activity extracted from the default and dorsal attention networks demonstrated a consistent temporal correlation pattern with frontoparietal control network activity during the autobiographical planning and Tower of London tasks, respectively. Both analyses demonstrated that neural activity within the default and frontoparietal control networks was coupled for autobiographical planning. Also, neural activity within the dorsal attention and frontoparietal control networks was coupled for visuospatial planning. This temporal correspondence is inconsistent with the alternative view that the default or dorsal attention networks and the frontoparietal control network act independently or sequentially without actually working directly together with the frontoparietal control network. If the networks were to independently modulate activity out of phase with one another over the course of individual trials, then this shifting would have to occur at a rate exceeding the fMRI sampling rate of 2500ms. While less parsimonious than the conclusion that network activity is coupled, the latter would be suggestive of network cooperation.
In a task-related functional connectivity analysis using seed PLS, we demonstrated a voxel-wise pattern of connectivity that is consistent with, and complementary to, the global measures discussed above. During autobiographical planning, we showed significant and reliable covarying patterns of activity within the default network (including the MPFC, PCC, lateral and medial temporal lobes, including the hippocampus, and the pIPL) with the frontoparietal control network (including the RLPFC, MFG, dACC, PCu, and the aIPL). During visuospatial planning, we showed significant and reliable covarying patterns of activity within the dorsal attention network (including the DLPFC, FEF, iPCS, MT+, and SPL) with the frontoparietal control network (including the RLPFC, MFG, dACC, PCu, and the aIPL). The extent of network connectivity, in its entirety, was quantitatively assessed and confirmed in an independent ROI analysis using the rsfcMRI maps (See and supplementary Figure 5
). Evidence of this network coupling was observed independent of the counting condition. Together, these data confirm that default and frontoparietal control regions of the brain can interact as a functional network. This observation is consistent with the behavior of dorsal attention and frontoparietal control regions interacting as a functional network, as observed here and elsewhere (e.g. Grady et al., in press
). The novel observation here, however, is that the frontoparietal control network may functionally couple with either the default or dorsal attention network, not just the dorsal attention network, in support of goal-directed cognition. A future challenge will be to examine node-to-node functional interactivity, within and across networks, modulated by task demands.
Additional analyses reduced the likelihood that task difficulty and RT differences could account for our findings. We addressed a potential confound of task difficulty by dividing the task data into easy and difficult trials and comparing task-related activity under easy and difficult conditions. If increased engagement of the frontoparietal control network during planning were due to increased difficulty during the planning tasks, relative to counting, we would expect increased activity within the frontoparietal control network for difficult versus easy autobiographical and visuospatial planning conditions. This effect was not observed, suggesting that frontoparietal control network activity was not modulated merely by task difficulty. Our categorization of autobiographical planning task difficulty was, however, based upon a post-scan self-report measure. Future work will be required to independently modulate and assess the effects of autobiographical planning difficulty.
Two additional analyses reduced the possibility that RT differences between autobiographical planning and counting could account for the pattern of network activity observed. First, significant engagement of the default and frontoparietal control networks by autobiographical planning was also observed relative to trial onset, independent of the counting control condition. Second, the block PLS analysis used to assess differences in difficulty between the planning conditions only included time points preceding the manual response, thereby eliminating potentially confounding RT differences. In addition to replicating the results using this method, we ruled out a differential RT explanation of differences in network activity between the conditions.
The neurocognitive function of the default network is not well characterized; yet the network is reliably engaged during spontaneous internal mentation (Andrews-Hanna et al., in press
), including self-referential (D'Argembeau et al., 2005
; Gusnard et al., 2001
), and autobiographical thoughts about the past and the future (Andrews-Hanna et al., in press
; Andrews-Hanna et al., 2010
; Spreng and Grady, 2010
; Spreng et al., 2009
). Task-related fMRI data suggest that the medial temporal lobes support recombination of elements of past experience into a simulation of a specific future event (Addis et al., 2009
; Addis et al., 2007
). Midline structures including MPFC and PCC that show increased activity during episodic simulation of future events have been linked specifically with both personal goal processing and self-referential thinking (D'Argembeau et al., 2010
; D'Argembeau et al., 2009
). The MPFC may be particularly sensitive to processing of self-promotion goals, which emphasize hopes, accomplishments, and advancement (Higgins, 1997
; Johnson et al., 2006
; Packer and Cunningham, 2009
). A remaining question is how imaginings supported by these components of the default network are integrated into goal-directed autobiographical plans in order to serve the adaptive function of guiding complex everyday behavior. Our findings suggest that co-activation of the default network with the frontoparietal control network occurs when individuals formulate autobiographical plans aimed at fulfilling an imagined goal state. The coactivation of the default network with the frontoparietal control network, however, is unlikely to be specific to autobiographical planning. Previous neuroimaging investigations of tasks that engage the default network (e.g. Spreng et al., 2009
) have placed low demands on control processes and are thus unlikely to robustly engage the frontoparietal control network. Conversely, introspective processes that require simultaneous goal-directed control of information will likely engage aspects of both the default and frontoparietal control networks, consistent with the current findings.
The Tower of London task predominantly engaged the dorsal attention network, in addition to the frontoparietal control network. Using PLS, we dissociated patterns of activity associated with this task between these two networks, which might support different aspects of task performance. Visuospatial planning places high demands on visuospatial attention, hence the observed robust activation of the dorsal attention network associated with externally driven attention. Visuospatial planning also taxes working memory and control processes that engage anterior prefrontal regions, including the RLPFC (Baker et al., 1996
; Wagner et al., 2006
). The Tower of London task was developed to assess goal-directed non-routine behavior that is not driven by immediate environmental stimuli (Shallice, 1982
). This aspect of the task may be common with autobiographical planning, as reflected by co-activation of the frontoparietal control network. Although we aimed to make the planning tasks as similar as possible with respect to perceptual characteristics and general task structure, further research will be needed to tease apart specific task components that differ between visuospatial and autobiographical planning, using a broader range of task manipulations.
By demonstrating that the frontoparietal control network is engaged during both autobiographical and visuospatial planning, this study provides new insight into its role in cognitive control and its access to domain specific information. The distributed frontoparietal control network is anatomically interposed between the default and dorsal attention networks. Vincent and colleagues (2008)
hypothesized that this network could serve to integrate information between the independent default and dorsal attention networks. In the present study, we compared tasks that are known to engage anticorrelated brain networks, but that both involved the goal-directed integration of information over time. This comparison provided a dynamic range of BOLD signal to detect dissociable and converging patterns of distributed activity. By demonstrating that the frontoparietal control network is actively engaged by two tasks that differentially rely upon either the default or dorsal attention network, we have provided evidence that the frontoparietal control network may flexibly gain access to information processed in either domain.
Recent work has clarified the role of various components of the frontoparietal control network in cognitive performance, including the role of RLPFC in cognitive control (Koechlin et al., 1999
), the hierarchical organization of control processes in lateral prefrontal cortex (Badre and D'Esposito, 2007
), and the role of the aIPL in the control of attention (Corbetta et al., 2008
) and memory (Cabeza et al., 2008
). Comparing tasks that vary in only a single feature, as in the foregoing studies, provides a means to identify subtle changes in brain activity attributable to that feature. This approach has been productive in determining precise functional roles of nodes within a broader network.
Complementing these previous findings, we demonstrate here that lateral prefrontal and parietal regions do not work in isolation to support control processes, but rather, function as an integrated network of regions in support of controlled cognitive performance. Diffusion tensor imaging provides anatomical evidence that the RLPFC and aIPL are structurally interconnected via the superior parietal fasciculus (van den Heuvel et al., 2009
). Neurophysiological techniques such as rsfcMRI suggest an even broader frontoparietal control network comprising multiple nodes associated with RLPFC and aIPL, including MSPFC, aIfO, dACC and PCu (Albert et al., 2009
; Damoiseaux et al., 2006
; Dosenbach et al., 2007
; Fox et al., 2006
; Smith et al., 2009
; Vincent et al., 2008
). The current study functionally identified this extended frontoparietal control network based on task-related activation during performance on two distinct planning tasks. The results provide novel evidence to support the hypothesis that the interposed nodes of the frontoparietal control network act as regional convergence zones that functionally interact with both default and dorsal attention regions during cognitive tasks. Further work is needed to delineate how the frontoparietal control network acts as a cortical mediator for crosstalk between internalized cognition and external attention in support of goal-directed cognition.