Interference resolution: Insights from a meta-analysis of neuroimaging tasks

TOR D. WAGER 0 0 Columbia University , New York, New York A quantitative meta-analysis was performed on 47 neuroimaging studies involving tasks purported to require the resolution of interference. The tasks included the Stroop, flanker, go/no-go, stimulus-response compatibility, Simon, and stop signal tasks. Peak density-based analyses of these combined tasks reveal that the anterior cingulate cortex, dorsolateral prefrontal cortex, inferior frontal gyrus, posterior parietal cortex, and anterior insula may be important sites for the detection and/or resolution of interference. Individual task analyses reveal differential patterns of activation among the tasks. We propose that the drawing of distinctions among the processing stages at which interference may be resolved may explain regional activation differences. Our analyses suggest that resolution processes acting upon stimulus encoding, response selection, and response execution may recruit different neural regions. - The need to select information among competing alternatives is ubiquitous. Oftentimes, successful cognition depends on the ability to focus resources on goal-relevant information while filtering out or inhibiting irrelevant information. How selective attention operates and whether and how irrelevant information is inhibited or otherwise filtered out has been a major focus of research since the inception of experimental psychology. For the past 15 years, cognitive neuroscientists have used neuroimaging to uncover the brain mechanisms underlying the processes responsible for handling irrelevant information. Much of this research has used variants of classic cognitive interference resolution tasks, each different in its superficial characteristics but sharing the common requirement to resolve conflict. What have we learned from this large corpus of data? Examining the multitude of studies focusing on interference resolution tells an extremely varied story. Figure 1A shows a plot of the peaks of activation of 47 studies that purport to examine interference resolution (see the studies listed in Table 1). Ostensibly, there appears to be little consistency in these data. Several factors may be contributing to the massive interstudy variance. First, Figure 1A includes activations from different tasks, subjects, equipment, scanning parameters, and statistical analyses. If we constrain our focus to just one task, however, the activations do not appear to be much more consistent. Figure 1B shows the activations arising just from the Stroop task (Stroop, 1935), and these do not appear any more orderly. Indeed, the variability among the reported peaks across all interference resolution tasks corroborates behavioral findings that correlations in performance among different interference resolution tasks are low (Kramer, Humphrey, Larish, Logan, & Strayer, 1994; Shilling, Chetwynd, & Rabbitt, 2002). Indeed, even simple changes in task parameters appear to produce very different results (e.g., de Zubicaray, Andrew, Zelaya, Williams, & Dumanoir, 2000; MacLeod, 1991). It seems clear that understanding interference resolution will take deeper analytic methods that interrogate possible strategic and mechanistic differences. Some researchers have attempted to examine the neural signatures of various interference resolution tasks within the same subjects to uncover whether any consistency can be found (Fan, Flombaum, McCandliss, Thomas, & Posner, 2003; Liu, Banich, Jacobson, & Tanabe, 2004; Peterson et al., 2002; Wager et al., 2005). These efforts have revealed that activations in different tasks overlap in a number of regions but that there are also regions unique to one task or another. What underlies these commonalities and differences? At this point, there have been a sufficient number of studies of interference resolution to begin to answer these questions. Here, we will attempt to sift through the inter Figure 1. (A) Peaks from the 47 studies included in the meta-analysis, plotted in a single brain. (B) Peaks from the studies in which the Stroop task was used.1 study variance in the interference resolution literature and pick out the consistencies among studies and tasks. In addition to trying to uncover the neural basis of interference resolution, we shall also consider why variations in tasks and task parameters may lead to separable patterns of neural activation. Although the meta-analytic methods used here preclude us from drawing strong conclusions about interference resolution (because they rely on reported peak coordinates from previous studies), they allow us to begin to form hypotheses that further investigations can either confirm or deny (e.g., Fox, Laird, & Lancaster, 2005). Study Selection For our analyses, we included six tasks that have been prominent in the interference resolution literature: the go/ no-go task, flanker task, Stroop task, stimulusresponse compatibility (SRC) task, Simon task, and stop signal task (all described below). Studies were included only if they reported peaks of activation in standardized coordinate space (Talairach or MNI). Notably absent are tasks used to examine the resolution of proactive interference (e.g., Jonides, Smith, Marshuetz, Koeppe, & Reuter-Lorenz, 1998), since a review of these data has already been published (Jonides & Nee, 2006). Furthermore, we do not include the antisaccade task in this mix, because models of this task are already at the single-unit level and our coarse techniques of analysis would be unable to inform this literature further (Munoz & Everling, 2004). We included neuroimaging studies in which either PET or f MRI was used between 1990 and 2005 and in which normal, healthy, young adults were examined.2 Although we recognize that there may be differences between blocked and event-related designs in terms of neural activations, there were insufficient studies to examine each separately. Therefore, we have combined both types of designs in our analyses. Forty-seven studies met our criteria and are listed in Table 1. When possible, we restricted our analyses to correct trials only. Tasks Go/no-go. In the go/no-go task, subjects are required to respond to one stimulus (e.g., the letter Y) but to withhold a response to another stimulus (X). Responses are labeled go trials, whereas trials on which a response is to be withheld are called no-go trials. It has been argued that as the number of go trials preceding a no-go trial increases, a greater prepotent tendency to respond is formed (de Zubicaray et al., 2000; Durston, Thomas, Worden, Yang, & Casey, 2002; Durston, Thomas, Yang, et al., 2002; Rubia et al., 2001). This prepotent response must be resolved in order to perform properly on no-go trials. Our analyses included contrasts of no-go versus go responses. Flanker. The flanker task requires a subject to attend to a centrally fixated stimulus while ignoring flanking stimuli (Eriksen & Eriksen, 1974). In a paradigmatic case, the central (...truncated)