How would an ideal behavioral method for cognitive enhancement actually affect the brain? Perhaps cognitive enhancement would be accompanied by more activity in the prefrontal cortex, indicating more successful engagement of control – or perhaps by less, indicating more efficient processing? Perhaps it would be accompanied by a transition from prefrontal activation to parietal activation, suggesting more automatic processing of task information. Perhaps it would make representations in prefrontal cortex more abstract and generalizable, or perhaps it would cause representations there to become more specific and tuned to the particular demands of difficult tasks. These are just a few of many possibilities, and they may all be correct depending on what kinds of training we’re talking about.
For example, tactile discrimination training yields large-scale cortical reorganization only for those areas of skin that have been trained (demonstrating highly specific training effects). However, the corresponding receptive fields for those areas of skin enlarge in sensory cortex, and neural recordings show earlier, higher amplitude, more rapidly oscillating, and more temporally focused activation in response to tactile stimulation. Similar effects were also observed in an auditory discrimination paradigm.
Of course, cortical representations are only one part of the network that is involved in such sensory training. Although subcortical reorganization has been observed after practice, it appears to occur only when cortical reorganization has also occurred. On the other hand, cortical reorganization can occur in the absence of subcortical reorganization.
Some argue that changes in the temporal profile of neuronal responses may be particularly important in training effects. Based on the often-demonstrated principles of Hebbian learning (in which spatiotemporal correlations between environmental stimuli lead to stronger neuronal associations between their representations), it is self-evident that precise temporal tuning would be an important aspect of neural computation. On the other hand, changes in receptive field size have shown small or outright inconsistent relationships with changes in sensory discrimination performance (e.g., Westheimer, 1979; Polly et al 1999; this vs. this; also compare this and this)
Some argue that there is a discrepancy in the effect of practice on neural representations, with practice on sensory tasks being associated with increased activation and practice on cognitive tasks being associated with decreased activation. This discrepancy parallels one in the behavioral domain, in which sensory and motor training rarely shows transfer to other tasks, but higher-level cognitive training usually does.
But even in so-called “cognitive” tasks, activation can be changed at the tonic or at the phasic level, and sometimes in different directions. For example, Kelley et al demonstrated tonic activation of the frontoparietal network may decrease with practice in a cognitive task while its phasic activation may actually increase. In addition, these practice-related changes occurred in the absence of any behavioral manifestations, indicating that similar behavioral performance can reflect multiple underlying causes. Kelley et al suggested that practice reduced the resources required for tonic working memory demand, thereby allowing for greater phasic use of those resources in accord with momentary task demands. (This claim is also consistent with work showing executive training-related reductions in many areas of the frontoparietal task network, and increases in those frontoparietal areas involved in conflict processing [e.g., anterior cingulate]).
Using another complex “cognitive” task known as the n-back, Erickson et al demonstrated another motif in the training literature, that of cortical redistribution. Erickson et al note that much previous work shows training on dual task situations can transfer to untrained tasks and that these gains can last months after training, but that these effects sometimes seem to be driven by decreases in neural activation, sometimes by increases in neural activation, and sometimes by shifts in the regions of activation. To assess which among these phenomena might be occuring during high-level cognitive training, Erickson et al. split 31 healthy adult subjects into control and experimental groups undergoing both pre- and post-tests on a simple dual task paradigm inside an MRI scanner; only the experimental group underwent five 1-hour sessions of speeded reaction time training with adaptive feedback (subjects were encouraged to reach the 63rd percentile of their reaction times during the previous session). Relative to the untrained control group, the trained subjects:
1) showed greater reductions in reaction times between pre- and post-test, but only in the conditions involving dual tasking or task-switching,
2) showed greater accuracy improvements between pre- and post-test, but only in the dual-task condition;
3) showed greater reductions in activation of the right inferior frontal gyrus and bilateral superior parietal lobe between pre- and post-test, with the change in inferior frontal gyrus activation limited to the dual task and task-switching conditions
4) showed activation of middle frontal gyrus in the dual-task condition, only at post-test
5) showed performance improvements across all task conditions that were correlated with reductions in right inferior frontal gyrus activity, performance improvements in the task switching and dual tasking conditions which were correlated with increases in dorsolateral frontal cortex, and performance improvements in the dual tasking condtiion which were correlated with decreases in superior parietal activity.
Thus, Erickson et al demonstrate heterogenous plasticity in high-level cognition or “executive control,” since improvements were limited to those task conditions which requires executive control, and since different regions of frontal and parietal cortex showed different patterns of training- and performance-related change, including both decreases and increases in activity in the inferior and dorsal lateral frontal cortex, respectively.
This pattern of training-related redistribution of neural activity is characteristic of many tasks. For example, Shadmehr & Holcomb showed a frontal-to-parietal-&-cerebellar shift in neural activity on a novel motor task. Similarly, using 4 weeks of n-back training, Hempel et al showed early practice-related increases in activation which ultimately translated into decreased activation, interpreted to reflect the “consolidation of performance gains.”
Kelley, Foxe & Garavan also report preliminary results of n-back training in which training improved accuracy and reaction time in all trial types (match, nonmatch, and lure trials), but particularly for lure trials – precisely those trials which predict fluid intelligence (both via lure performance and lure-related hemodynamic response). Additionally, subjects showed transfer of training effects to two untrained tasks, the spatial 4-back and a verbal Sternberg paradigm.
Thus it seems that in high-level cognitive training paradigms, neural activity may undergo multiple transformations as a result of training – even in the absence of any (measurable) behavioral improvements! To the extent that the tasks involve sensory or motor skill learning, those regions of the brain may undergo changes in temporal tuning profiles as well as expansion of the corresponding cortical maps. To the extent that the tasks involve high-level cognitive processes like attention and working memory, the corresponding fronto-parietal network may reduce its tonic activation, reflecting increased neural efficiency. Conversely, transient task demands may actually yield larger phasic responses despite (or perhaps because of) the lowered tonic activity. (I speculate that both of these phenomena may reflect the same mechanism observed in sensory tasks – tighter temporal tuning profiles – but may be associated with decreased rather than increased activity partly due to methodological differences in neuroimaging of sensory and cognitive tasks). Finally, activation may undergo a frontal to posterior redistribution or shift over practice, reflecting the use of more specialized or dedicated processes as the practiced task becomes more familiar.