Switching and Maintenance: Evidence for Distinct Mechanisms?

Normal children - and adult patients with frontal damage - frequently have difficulty changing their responses to stimuli when the correct response changes. This difficulty is often considered an inability to switch between rules, but might result not so much from an inability to switch as from an inability to represent the stimuli as having two possible responses in the first place (i.e., to represent the stimuli as "bivalent"). Supporting this distinction is a new article in the Journal of Neuroscience that claims to distinguish the networks supporting "bivalent" representation from those involved in response switching per se, and how they change with age.

To do this, Crone et al. provided 54 subjects, aged 8 to 25, with a complex task while they were inside an fMRI scanner. Each trial consisted of a cue stimulus, then a probe stimulus, and finally a subject's response. For example, a subject might see a cue image of an arrow, which was always followed by either a picture of a car (to which a "right" keypress was required), or by a picture of a flower (to whcih a "left" keypress was required). Thus, if subjects saw an "arrow" they might rehearse something like "flower-left car-right." However, the other two types of cue stimuli were more complex. If subjects saw a "circle" cue image followed by a tree probe image, they were to respond to the tree with a left keypress, and if the circle was followed by a house probe image, they were to respond to the house with a right keypress. But just the opposite keypresses were required if the cue image was a triangle.

Thus, the task has two basic trial types: bivalent trials in which the stimuli are the same but the responses are different (cued by triangles or circles) and univalent trials in which both the stimuli and the responses are the same (preceeded by the arrow cues). Similarly, some trials were rule repeat trials relative to the previous trial (i.e., the same cue was present twice in a row), whereas other trials were rule switch trials (i.e., a different cue was presented on the previous trials). Finally, the trials were presented in grouped fashion or in "mixed" fashion (where rule repeat and rule switch trials were intermixed) or "blocked" (consisting of only rule repeat trials).

The structure of the task, as described above, allowed Crone et al. to measure "rule representation" ability by comparing reaction times (RTs) on bivalent trials with RTs on univalent trials. Likewise, "switching" ability can be measured by compariing RTs on switch trials relative to those on rule-repeat trials. The same analyses can be done on outright errors and fMRI data as well, for similar indices of rule representation and switching processes. For the purposes of analysis, the participants were split into 3 age groups: 8-12 year olds, 13-17 year olds, and 18-25 year olds.

Not surprisingly, older participants were faster and more accurate at both switching and rule representation. All three age groups were significantly different in rule representation ability, whereas only the youngest and the oldest groups differed in switching ability (with the middle age group somewhere in between). The authors interpret this to show that switching ability matures earlier that rule representation, and thus follows a different developmental trajectory.

There is at least one significant problem with this interpretation: two measures might appear to show differences simply because of differences in their sensitivity. In other words, the set switching measures might simply be easier than rule representation measures, despite indexing the same underlying cognitive process. If this were true, the same pattern of results could be expected, but one wouldn't claim this shows distinct developmental trajectories for the two measures!

Slightly more interesting data comes from the neuroimaging analyses, which showed that only supplementary motor cortex (a region in medial prefrontal cortex; SMA) was more active for bivalent than univalent stimulus types in the blocked task, whereas both SMA, ventrolateral prefrontal cortex (vlPFC) and superior parietal cortex all showed this effect when trial types were presented in a "mixed" fashion. Although Crone et al. argue that this supports a role for SMA in rule representation, it is difficult to rule out that SMA is merely sensitive to amount of effort.

Among adults, vlPFC activity was stronger for bivalent than univalent trials, regardless of whether the rules had switched. In contrast, both adolescents and children showed stronger vlPFC activity on switch trials, with children showing more vlPFC activation on univalent switch trials than bivalent switch trials, and with adolescents showing the reverse trend. To summarize this rather confusing pattern of resutls, vlPFC appeared to be involved in rule representation in adults and set switching in the younger age groups, suggesting that a) age influences task strategy, b) that age influences the type of computations performed by vlPFC, or c) some combination of these two possibilities. The differences between adolescents and children are even harder to explain, but could be spurious and driven by high error rates of children in bivalent switch trials (~17%). vlPFC is generally thought to be involved in simple active maintenance processes, which might be common to both switching and rule representation.

All age groups showed stronger SMA activation for switch trials relative to repeat trials, but the adults and adolescents only showed this effect for bivalent stimuli. Again, this result could be due to sensitivity or floor effects: univalent trials (regardless of whether they involve rule repeats or rule-switches) are likely much easier for adolescents and adults than for children, and so may show less differences in activation.

Unlike the other age groups, children also showed an effect within rule repetition trials, in that SMA activity was stronger for bivalent than univalent rule-repetition trials. Like the evidence reviewed in the previous paragraph, this too could result from sensitivity differences, in that rule-repetition trials are always very easy for the older age groups, and so bi- or univalence would have little impact. The results for superior parietal cortex were less interesting, in that all age groups showed more activity in this region on bivalent relative to univalent trials, and on switch relative to repetition trials.

The authors conclude that rule representation and set switching follow distinct developmental trajectories and rely on distinct neural networks. While it's conceivable that this is true, this study does not present unequivocal evidence for that view. For me, this data underscores several difficulties in conducting cross-sectional neuroimaging research: strategy and effort differences between groups are difficult to control, and activation differences therefore need not reflect unitary cognitive processes. Nonetheless, this study has been cited as demonstrating a difference between set switching and rule representation when in fact a variety of other interpretations are possible.

There are yet other complications to this research as well. Some might argue that the bivalent "switch" trials are not really switch trials at all, since only the specific stimulus-response mappings and not the stimuli or responses themselves are switched. This is more commonly termed "reversal" and is in fact neurally dissociable from task switching proper. Secondly, Crone et al. have confounded the effects of changing task cue (in this case, circle, square or arrow) with the effects of changing task stimulus-response mappings. Some work has shown that changing the cue but repeating the task can account for a large portion of the reaction time slowing on trials where the cue and the stimulus-response mappings change, suggesting that these two factors need to be distinguished both in behavioral and neuroimaging research. Finally, there are additional reaction time costs on switch trials where responses repeat relative to the previous trial, which were not controlled in this study.

Computational models of developmental change in task switching performance have demonstrated that change in active maintenance ability alone is sufficient to explain a vast array data. This underscores the need to critically examine any paper claiming to show distinct mechanisms for processes involved in task switching. Accordingly, tomorrow's post will review a new paper in Developmental Neuropsychology which follows in the tradition of the Crone et al paper reviewed above, by presenting evidence for distinct switching and maintenance mechanisms.

Related Posts:
Task-Switching: A Role for Inferior Parietal Cortex
Localizing Executive Functions in Prefrontal Cortex
Backward Inhibition: Evidence and Possible Mechanisms

More like this

Cognitive theories of "executive function" vary greatly in the number of distinct cognitive processes they propose to subserve the goal-directed coordination of behavior. Some theories suggest that strong active maintenance of information, and a way of "updating" the information that is maintained…
Though anatomically heterogenous, the human prefrontal cortex seems to perform a rather general function: it actively maintains context representations to guide and control behavior. What, then, is the reason for the anatomical diversity within this region of the brain? Some theories suggest…
"Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child's?" - Alan Turing (Computing Machinery, p456) One of the defining features of childhood cognition is "behaving without thinking." Not surprisingly, developmental…
How does memory help to accomplish moment-to-moment goal-directed action? Classic accounts, such as Baddeley's working memory model, suggest that there are separate storage and processing ("executive") mechanisms, whereas newer accounts (proposed by a variety of researchers) propose that storage…