Neural mechanisms underlying working memory : computational and neuroimaging studies

Abstract

The performance on various cognitive tasks, from language to selective attention and guidance of future actions depends on working memory (WM), the ability to hold and manipulate limited items of information for a period of up to a few seconds. During childhood development, WM capacity, the number of items one can maintain in WM, increases. However, the neural correlates of WM capacity, distractibility and maturational processes underlying WM development are still unclear. The present work addresses these issues by the integration of computational modeling, functional magnetic resonance imaging (fMRI) and behavioral methods.

In the first study we used distracting visual stimuli in order to identify cellular mechanisms that account for the observed behavioral decrease in mnemonic accuracy as a function of the spatial distance to distractors. The computational model suggests that independently of the cellular and synaptic properties, increased neuronal firing rates accounted for higher mnemonic accuracy and resistance against distractors.

In the second study we performed fMRI experiments on adults and children to monitor brain activity during a WM task. We isolated the delay-related activity and analyzed group differences and the distractor influence both behaviorally and in terms of changed brain activity. Accompanying the higher WM capacity and lower distractibility of adults, the fMRI study showed higher brain activity in middle frontal gyrus and intraparietal cortex in adults compared to children during the delay periods of WM tasks.

In a subsequent study we addressed the cellular changes during WM development. The study combined a computational analysis with FMRI in order to establish putative maturation processes governing developmental changes in brain activity. We found that the increase in activity together with higher resistance against distractions could be explained by stronger connectivity between network areas.

The final study addressed the limited storage capacity of previous WM models. Implementing structural and connectivity changes likely to occur during WM development in a biophysical WM model we have obtained multiple-item storage capacity similar to human WM performance. Furthermore, by using fMRI we found that the informationactivity curve predicted by the model corresponds to that in the human posterior parietal cortex during performance of WM tasks.

In conclusion, in the context of neural networks dominated by reverberatory synaptic input, our studies demonstrate the correlation between a higher WM capacity, resistance to distractors, mnemonic accuracy, BOLD response and average firing rate of the neuronal populations responsible for the memory maintenance.