Spatiotemporal oscillatory dynamics during the encoding and maintenance phases of a visual working memory task

Abstract

Many electrophysiology studies have examined neural oscillatory activity during the encoding, maintenance, and/or retrieval phases of various working memory tasks. Together, these studies have helped illuminate the underlying neural dynamics, although much remains to be discovered and some findings have not replicated in subsequent work. In this study, we examined the oscillatory dynamics that serve visual working memory operations using high-density magnetoencephalography (MEG) and advanced time-frequency and beamforming methodology. Specifically, we recorded healthy adults while they performed a high-load, Sternberg-type working memory task, and focused on the encoding and maintenance phases. We found significant 9–16 Hz desynchronizations in the bilateral occipital cortices, left dorsolateral prefrontal cortex (DLPFC), and left superior temporal areas throughout the encoding phase. Our analysis of the dynamics showed that the left DLPFC and superior temporal desynchronization became stronger as a function of time during the encoding period, and was sustained throughout most of the maintenance phase until sharply decreasing in the milliseconds preceding retrieval. In contrast, desynchronization in occipital areas became weaker as a function of time during encoding and eventually evolved into a strong synchronization during the maintenance period, consistent with previous studies. These results provide clear evidence of dynamic network-level processes during the encoding and maintenance phases of working memory, and support the notion of a dynamic pattern of functionally-discrete subprocesses within each working memory phase. The presence of such dynamic oscillatory networks may be a potential source of inconsistent findings in this literature, as neural activity within these networks changes dramatically with time.

Introduction

Working memory refers to the short-term storage and manipulation of information for immediate cognitive processing. Conceptually speaking, a stimulus can be loaded or “encoded” into working memory, stored and refreshed through some form of rehearsal or “maintenance” of the memory trace, and ultimately utilized or “retrieved” to perform a goal-oriented action; the information is then discarded or allowed to fully “decay.” There are a number of different neurocognitive models of working memory [e.g., (Baddeley, 1992, Baddeley, 2000, Baddeley et al., 2011, D'Esposito, 2007)], but all support the existence of encoding, maintenance, and retrieval phases as being necessary for successful working memory retention.

Many functional magnetic resonance imaging (fMRI) and positron-emission tomography (PET) investigations have examined working memory circuits in healthy adults [e.g., (Cabeza and Nyberg, 2000, Rottschy et al., 2012)]. These studies have highlighted a group of brain regions that include the dorsolateral prefrontal cortex (DLPFC), parietal and occipital regions, and the left supramarginal gyrus as critical to working memory function (Cabeza and Nyberg, 2000, Rottschy et al., 2012). While these PET and fMRI studies have provided valuable information on the specific brain regions that are active during working memory operations, delineating the brain areas that are active during each phase (i.e., encoding, maintenance, retrieval) of working memory processing has been more difficult due to the temporal limitations of these imaging modalities. However, recent neurophysiological investigations using magnetoencephalography (MEG) and electroencephalography (EEG), which possess excellent temporal resolution, have begun to fill this void by clarifying the spatiotemporal dynamics of working memory processing [e.g., (Bonnefond and Jensen, 2012, Bonnefond and Jensen, 2013, Brookes et al., 2011, Honkanen et al., 2014, Jensen et al., 2002, Jensen and Tesche, 2002, Jiang et al., 2015, Morgan et al., 2011, Palva et al., 2010a, Palva et al., 2011, Palva et al., 2010b, Roux and Uhlhaas, 2014, Roux et al., 2012, Sauseng et al., 2009, Spitzer et al., 2013)]. While some findings in the MEG/EEG working memory literature have been variable, much of this variability can be attributed to task and stimulus design features (e.g., Sternberg versus N-back, auditory versus visual, verbal versus spatial). For example, neurophysiological studies of working memory have used auditory (van Dijk et al., 2010, Nolden et al., 2013), visuospatial (Jensen et al., 2002, Jensen et al., 2007, Jensen and Tesche, 2002, Palva et al., 2010a, Palva et al., 2011, Palva et al., 2010b), vibrotactile (Spitzer et al., 2013), and language-based stimuli (Brookes et al., 2011). Slight variations in task design were also the norm in fMRI and PET studies, and many of the inconsistencies in this literature are likely attributable to this factor (Cabeza and Nyberg, 2000, Rottschy et al., 2012).

Jensen and Tesche (2002) recorded MEG on healthy younger adults during a visual Sternberg task (Sternberg, 1969) and found that theta (4–7 Hz) activity in the frontal cortices parametrically increased as a function of memory load during the memory maintenance phase, becoming stronger in amplitude from the 1-item condition up to the 7-item condition (Jensen & Tesche, 2002). Similar results have been reported by Gevins, Smith, McEvoy, and Yu (1997) and more recently by Scheeringa et al. (2009). Likewise, an MEG study by Brookes et al. (2011) also found an increase in frontal theta activity relative to the baseline, and a positive correlation between task difficulty and frontal theta. However, it is important to note that Brookes et al. did not distinguish between the encoding, maintenance, and retrieval phases so it is unclear which cognitive processes were best reflected in their images of oscillatory activity. In a related EEG study, Jensen et al. (2002) found that alpha (8–14 Hz) activity in the central parieto-occipital areas increased as a function of memory load during the maintenance phase, being weakest in the 2-item condition and strongest in the 6-item condition (Jensen et al., 2002). This pattern of increased alpha has been replicated and expanded in many recent studies, and findings of increased alpha during the maintenance phase appear to be especially robust (Bonnefond and Jensen, 2012, Bonnefond and Jensen, 2013, Jiang et al., 2015, Tuladhar et al., 2007). In contrast to these studies of working memory maintenance, Palva et al. (2011) reported that the amplitude of theta-alpha, high alpha, beta, and gamma activity were significantly reduced during the maintenance phase relative to the baseline phase in most brain areas, and in no regions was neuronal activity (at any frequency) significantly stronger during the maintenance phase relative to the baseline (Palva et al., 2011). Furthermore, they found that only high-alpha, beta, and gamma–frequency activity was positively correlated with memory load in the prefrontal cortices; that is, suppression of activity in these frequency bands became weaker as memory load increased (Palva et al., 2011). These and other discrepancies between studies may be attributable to not only focusing on different temporal phases of working memory (i.e., interpretations of what encompasses encoding, maintenance, or retrieval), but also focusing on distinct temporal periods within each phase across studies.

In the current study, we examined the dynamic distributed processes that serve working memory during a visual task that used single-letter stimuli that were simultaneously presented. Specifically, we focused on neural dynamics within the encoding and maintenance phases of working memory processing, where information is loaded, retained, and rehearsed. Prior EEG/MEG studies have not focused on the changing oscillatory dynamics within the encoding and maintenance periods specifically, which may lead to greater insight into the source of inconsistencies in this literature. To this end, we utilized high-density MEG to record healthy adult participants while they performed a modified, high-load, Sternberg-type working memory task. We hypothesized that participants would show relatively continuous neuronal activity across the encoding and maintenance periods in the left prefrontal and supramarginal regions, along with a dynamic response pattern in the occipital cortices, as function in this brain region switches from encoding stimuli to protecting active representations from incoming interference.