rTMS of dlPFC Dissociates Maintenance and Manipulation Processes

Active maintenance of information is clearly critical to working memory (WM), but also important are the attentional or executive functions required to manipulate the information that is being actively maintained. Evidence reviewed in these two posts supports the idea that at least two such mechanisms are at work in a variety of WM tasks. Fascinating new evidence - from Brad Postle's new article in the October issue of the Journal of Cognitive Neuroscience - suggests the two mechanisms might be maintenance and manipulation operations.

In their experiment, Postle et al. scanned 12 subjects with fMRI while they performed the following task. Subjects were presented with 2 or 5 letters, and were told to either mentally alphabetize or simply remember the letters that were presented. After an 8 second delay, a single letter was presented, along with a digit; this digit indicated the serial position of the letter for "simply remember" trials, and the alphabetical position for the "alphabetize" trials. Subjects had to indicate whether the digit correctly represented the placement of the accompanying letter, but only on 50% of the trials was this letter-digit pair correct.

Further methodological details are in italics:

The authors identified regions that showed significantly different activation between the 3 and 5 letter trials, and also between the "simply remember" and the "alphabetize" trials. Thus, they isolated regions that were sensitive to either load (3 vs 5) or to manipulation (simply remember vs. alphabetize) operations within each subject, and mapped this activity onto an anatomical scan. The authors then targeted the regions sensitive to manipulation with repetitive transcranial magnetic stimulation (rTMS) during a task where each of 144 trials was either a "simply remember" or a "alphabetize" trial for displays consisting of 5 letters. During the last 6 seconds of the 8-second delay period between stimulus presentation and the letter-digit probe, rTMS was applied to regions within the middle frontal gyrus (MFG) of the dlPFC and to the superior parietal lobe (SPL) that showed alphabetization sensitivity, as identified through real-time infrared stereotaxy, coregistered with the anatomical MRI scans done on each individual previously (an impressive technical accomplishment).

The fMRI results showed that regions that are sensitive to manipulation are different from those sensitive to load. rTMS of manipulation-sensitive regions in dlPFC significantly affected alphabetization accuracy, but not the "simply remember" task, in comparison to stimulation of a control region. In contrast, rTMS of the manipulation-sensitive regions in SPL affected both alphabetization and retention accuracy - which seems to contradict the fMRI evidence showing these processes were spatially dissociable.*

All in all, this evidence should not be interpreted to suggest that dlPFC regions are important primarily for manipulation of information, and not for simple retention of that information over time, because there's one large caveat here: if maintenance-sensitive regions of dlPFC had undergone rTMS, perhaps we'd see the opposite dissociation. Or maybe we'd see no dissociation at all (i.e., performance might decrease on both "simply-remember" and "alphabetization" tasks) if dlPFC is also orchestrating maintenance - a function that is probably even more important for manipulation than simply-remember tasks.

What we can say, based on this study, is that some regions of dlPFC are more important when manipulation demands are highest. Disruption of activity in those regions leads to performance deficits in tasks with high manipulation demands, and not necessarily in those with moderate maintenance demands. This is a very interesting conclusion, because it begins to identify possible neural substrates of these two distinct sources of variance in WM tasks.

Postle has recently made clear that he doesn't believe PFC contributes to the active maintenance of information, but instead that active maintenance is subserved by more posterior and thus modality-specific regions. But according to this view, it's not clear why any regions of PFC should be load sensitive (as some were in this experiment), nor would one expect rTMS of load-sensitive regions in dlPFC to affect load performance. Neither of these issues is discussed in this paper, unfortunately.

It is also worth mentioning that to some extent the emphasis on dlPFC may be unwarranted, given that the largest mean BOLD increase in alphabetization tasks was actually in the superior parietal lobe. In contrast, the largest mean BOLD increase in the high load condition was in right inferior frontal gyrus (ventrolateral prefrontal cortex, or vlPFC), a region also implicated in selection of items from memory. It may be that what I'm calling "active maintenance" is actually accomplished by vlPFC, while dlPFC is involved in strategic processing of that information (such as involved in task switching, alphabetization, etc).

The beautiful image at the start of this article illustrates the combined anatomical, fMRI, and infrared stereotaxy information from one subject. In this image, the white blobs indicate the regions of dlPFC showing sensitivity to alphabetization (increased manipulation), the orange dot shows the focus of rTMS, and the orange line shows the "the maximal energy vector of the rTMS-induced magnetic field."

* - The authors suggest this contradiction is more apparent than real, and can be attributed to the poor spatial resolution of rTMS in a region like the SPL, where retention-related areas may be closer to manipulation related areas than in PFC.