The oscillatory mechanisms of working memory maintenance

Abstract

Working memory (WM) is a cognitive process which allows for maintenance of information that is no longer perceived. Although theoretical models have recognized that working memory involves interactions across cell assemblies in multiple brain areas, the exact neural mechanisms which support this process remain unknown. In this thesis I investigate the neural dynamics in the human hippocampus, the ventral, dorsal and frontal cortex as well as the long-range network connectivity across these brain areas to understand how such a distributed network allows for maintenance of various information pieces in WM. The results described here support a model in which working memory relies on dynamic interactions across frequencies (the cross-frequency coupling, CFC) in a distributed network of cortical areas coordinated by the prefrontal cortex. In particular, maintenance of information during a delay period selectively involves the hippocampus, dorsal and ventral visual stream as well as the prefrontal cortex each of which represents different features.<br /> The hippocampus contributes to this large network specifically by representing multiple items in working memory. In two independent experiments I observed that the low-frequency activity (a marker of neural inhibition) was linearly reduced across memory loads. Importantly, the hippocampus showed very prominent low-frequency power during maintenance of a single item suggesting that during this condition the neural processing was strongly inhibited. In turn, the broadband gamma activity was linearly increasing as a function of memory load. This pattern of results may be interpreted as reflecting an increased involvement of the hippocampus in representing longer sequences. Importantly, the low-frequency decrease was not static but fluctuated periodically between two different modes. One of the modes was characterized by the load-dependent power decreases and reduced cross-frequency coupling (memory activation mode) whereas the other mode was reflected by the load-independent high levels of power and increased coupling strength (load-independent mode). Crucially, these modes were temporally organized by the phase of an endogenous delta rhythm forming a “hierarchy of oscillations”. This periodicity was essential for the successful performance. Finally, during the memory activation mode the WM capacity limit was inter-individually correlated with the peak frequency change as predicted by the multiplexing model of WM. All these effects were subsequently replicated in an independent dataset. These results suggest that the hippocampus is involved in WM maintenance showing periodic fluctuations between two different oscillatory modes. Parameters of the hippocampal iEEG signal correlate with individual WM capacity, specifically during the memory activation mode.<br /> The ventral and dorsal visual stream each contributes to the distributed WM network by representing configuration and spatial information, respectively. Specifically, the alpha power in the ventral visual stream was decreased during maintenance of face identities. In turn, the alpha power was desynchronized in the dorsal visual stream while participants were maintaining face orientations. This shows that the alpha power double dissociates between the feature specific networks in the ventral and dorsal visual stream. These effects are further interpreted as reflecting selective involvement of the dorsal and ventral visual pathway depending on the maintained features. Importantly, each of the visual streams was selectively synchronized with the prefrontal cortex depending on the memory condition and the alpha power. This corroborates a central prediction from the gating by inhibition model which assumes that the increased alpha power serves as the mechanism for gating of information by inhibiting task redundant pathways. Moreover, during maintenance of information the phase of alpha modulated the amplitude of high-frequency activity both in the dorsal and ventral visual stream. Additionally,the low-frequency phase in the prefrontal cortex modulated high-frequency activity both in the dorsal and ventral visual stream. These results suggest that both the dorsal and ventral visual streams are selectively involved during maintenance of distinct features (i.e. face orientation and identity, respectively). They also indicate that the prefrontal cortex selectively gets synchronized with the visual regions depending on the alpha power in that region and the maintained feature. Finally, the activity in the prefrontal cortex influences processing across long distance as evident from changes in the phase synchrony with the visual cortical areas and by modulating gamma power in the visual cortical regions.<br /> It is also noted that the ventrolateral prefrontal cortex (vlPFC) contains information regarding abstract rules (i.e. response mapping). In particular, using a multivariate decoding approach I found that the local field potentials recorded from the vlPFC dissociate between different types of responses. At the same time I observed no evidence for the load-dependent or stimulus-specific changes in that brain region. The null effect should be treated with caution. Nevertheless, the current results suggest that the vlPFC may contribute to working memory by processing of abstract rules such as a mapping between the stimulus and the response. Furthermore, I found that the alpha power dependent duty cycle in the vlPFC constrains the duration of the gamma burst which has been suggested as a mechanism for neural inhibition. This finding is important because such a property of the alpha activity has never been observed in a brain region other than the primary sensory cortex.<br /> Together, the results presented in this thesis support a model according to which the working memory is a complex and highly dynamic process engaging hierarchies of oscillations across multiple cortical regions. In particular, the hippocampus is important for the multi-item WM. The dorsal and ventral visual streams are relevant for distinct visual features. Finally, the prefrontal cortex represents abstract rules and influences processing in other cortical regions likely providing a top down control over these regions.