By many current theories, we accomplish control over behavior by using the prefrontal cortex to “bias” the competitive dynamics playing out in the rest of the brain. By some models, this bias is positive – it helps the goal-relevant representations win the competition. By other models, the bias is also negative – it can help the goal-irrelevant representations lose the competition. Regardless, this “prefrontal biasing” is usually considered in terms of the amount of activity in a particular area (higher when that area is under a positive bias, and lower when under a negative bias). But Gazzaley et al’s excellent 2005 JoCN article demonstrates that this kind of top-down modulation can also have effects on the speed of neural processing, where positive and negative biasing actually have opposite effects on the latency of some well known ERP components.
To demonstrate this, Gazzaley et al asked 11 subjects to participate in a two session experiment (one in the fMRI scanner and one with electrodes pasted to their heads) in which they saw sequences of pictures of faces and places, in random order. These randomized sequences were preceded by instructions to pay attention to one or the other type of picture while ignoring the other, to pay attention to both, or to just passively view the pictures. A subsequent memory test showed that people were better able to recognize the pictures they’d been told to remember than those they’d passively viewed, and that this advantage was reduced when subjects had been asked to remember both types of stimuli relative to just one. This pattern is consistent with the idea that enhancement of task-relevant information is a resource-limited process that is less effective when it needs to be used more. However, the authors didn’t report any reduction in recognizing the to-be-ignored stimuli – perhaps indicating that negative biasing either didn’t occur or wasn’t effective.
On the other hand, the fMRI results told a different story – one that includes some putative suppression effects. Although fMRI has poor temporal resolution, its excellent spatial resolution allowed the authors to show that neural activity in face- and place-selective regions of the cortex (the FFA and PPA, respectively) was increased when the respective stimulus type was to-be-attended, and decreased when the respective stimulus type was to-be-ignored, relative to the case where both stimulus types were to be passively viewed (the latter “suppression” effect was significant only for the place-selective region, but was in the right direction for the face-stimuli too). Similarly, when subjects had been instructed to remember both stimulus types, this increase in neural activity was reduced (at least for the place-selective region; they didn’t report results from the face-selective region), again indicating that top-down enhancement is resource demanding. These effects demonstrate that there is a robust top-down enhancement effect, and also suggest some suppression effect, on the magnitude of neural activity as measured by fMRI.
Gazzaley et al’s ERP results take this a step farther: the authors found that the peak latency of a electrical potential characteristic of face processing was earlier when face stimuli were to-be-attended, and later when face stimuli were to-be-ignored, relative to cases where face stimuli were to be passively viewed. In addition, this apparent enhancement of the speed of face processing was reduced when subjects had been instructed to remember both stimulus types, again suggesting that top-down enhancement is resource-limited.
In their discussion, the authors note that such latency shifts have not been observed in single-cell recording studies, suggesting that the top-down effects may be visible only at the level of the local field potential or the resulting ERP. They also note that the latency shifts have not been observed in other selective attention studies, and suggest that this may have to do with task difficulty or that the effect is specific to memory encoding. I found this a little speculative, given that face-selective can show up in either latency or amplitude, and observing one instead of the other is not necessarily informative (some have suggested that these two measures might trade-off).
In a concluding paragraph, the authors note that their effects pertain to the sites of control (i.e., face- and place-selective neural phenomena) and not to the source of that control (i.e., the prefrontal or parietal cortex). Gazzaley et al suggest that multivariate pattern recognition techniques could be used to identify such sources, which might be taken to imply that traditional averaging techniques used in this paper failed to identify those sources.
In summary, Gazzaley et al have collected compelling evidence that goals can be used to positively bias both the magnitude of activity in posterior cortex and the speed with which that processing reaches its peak amplitude. They’ve also shown some negative effects, such that the magnitude and speed of activity in regions selective for the to-be-ignored items can be reduced relative to passively viewing them, but this is a little less clear. While the enhancement effect is solid and robust, the suppression effect was a little less robust (not significant in the FFA), and did not result in any behavioral suppression effect (at least, not one that was reported).
One possibility here is that passive viewing and ignoring yield identical behavior because in both cases subjects are simply thinking about other things as they see the to-be-ignored or to-be-viewed stimuli. One might get a reduction in visual processing of to-be-ignored stimuli if subjects simply closed their eyes or focused on a different part of the display, and this need not reflect true top-down suppression. If something like this had occurred, it would be visible in the ERP data, because ERP’s are exquisitely sensitive to blinking and saccades. Indeed, a relatively large proportion of subjects were excluded from ERP analysis due to excessive oculomotor activity (4/18), perhaps suggesting that just such a strategy was used by some subjects. Even those subjects who were not excluded for that reason nonetheless had very few included trials, due to the detection of oculomotor artifacts (or alpha waves, sometimes thought to reflect a cognitively “idle” state) on 66% of trials (on average; 160/240). It seems to me that this strategy is an alternative explanation not only for the neural “suppression” phenomena observed here but also for the apparent lack of a behavioral suppression effect in this particular study.