Control-related systems in the human brain
Jonathan Power, Steve Petersen
Current Opinion in Neurobiology 2013 Apr; 23(2):223-8
This paper reviews a decade or so of task control studies in the Petersen lab spanning roughly 2003-2013. The review opens with studies by Nico Dosenbach examining the mixed block/event-related design, and then follows links from those studies to my own resting state work. More task control studies have since emerged from the Petersen lab, so the paper is a little out of date (see the recent and interesting papers by Joe Dubis or Maital Neda).
I think I'll only add 2 small points to this paper.
The first point is historical. Steve has been interested in attention (now called task control) for decades, since his work with Mike Posner in the 1980s. Since the early days of neuroimaging, Steve has had ideas about what task control signals might look like (as reviewed in the paper). The key to defining task-general control regions is to search for such signals over many different task sets. By the mid-1990s, WashU had collected enough different PET experiments that it seemed feasible to attempt to identify task-general control regions. This led to a now-famous meta-analysis of PET data. That analysis more or less completely failed to identify control regions. You can hear the disappointment in the title of the paper: "Common blood flow changes across visual tasks: I. Increases in subcortical structures and cerebellum but not non-visual cortex.". On the other hand, anyone who knows the history of the default network recognizes the companion paper: "Common blood flow changes across visual tasks: II. Decreases in cerebral cortex". Gordon Schulman was first author on those studies, Steve was senior. A bust from a task control perspective, but a huge step for cognitive neuroscience. Once enough fMRI tasks had been collected, however, Steve re-approached the issue, now with the considerable added benefit of event-related modeling, which led to the material discussed in this review.
The second point has to do with interpreting task-general activity. The "core" task control regions (bilateral anterior insulae and dorsal anterior cingulate cortex) discussed in this review were identified because (particular kinds of) activity was detected at those locations over many different kinds of tasks (see the interesting review by Steve Nelson for a literature-mining variant of this identification). One feature of these regions, which we only realized as part of my resting state work, was that these regions are also relatively densely populated with functional systems - many different resting state signals are found in close proximity at those locations. Spatial precision is difficult in meta-analyses due to subject variability, across-subject averaging, and combining disparate datasets. It seems possible that one contributor to task-general effects, in locations populated by several systems, is labeling contributions from these different systems as a general effect. I don't know if this is actually a real issue with these studies, but it's a caveat worth keeping in mind.