Visuospatial working memory involves encoding, temporary maintenance, and retrieval of visual and spatial information (Baddeley, 1992; Ma et al., 2014) and is essential for everyday cognitive function. Increased familiarity with a given task through practice improves working memory performance (Crone et al. 2006; Miotto et al., 2006; Schneiders et al., 2011; Kundu et al., 2013; Constantinidis and Klingberg, 2016; Thompson et al., 2016; Wang et al., 2019). The utilization of visuospatial working memory involves a distributed network of brain regions that work together to encode, maintain, and retrieve information. This network includes the lower- and higher-order visual areas, medial-temporal regions, dorsolateral prefrontal cortices, and the posterior inferior-frontal gyrus (pIFG) (Smith and Jonides, 1999; Constantinidis and Klingberg, 2016; Wu and Buckley, 2022). Converging evidence for the involvement of these brain regions in visuospatial working memory has been provided by functional MRI (fMRI) (LaBar et al., 1999; Pochon et al., 2001; Kwon et al., 2002; Krasnow et al., 2003; Croizé et al., 2004; Suchan et al., 2006; Ganis et al., 2007; Schmidt et al., 2007; Srimal and Curtis, 2008; Edin et al., 2009; Michels et al., 2010; Christophel and Haynes, 2014; Vetter et al., 2014; Zumer et al., 2014; Darki and Klingberg, 2015; Schmidt and Blankenburg, 2018; Yaple et al., 2019; Henderson et al., 2022), electrophysiology (Croizé et al., 2004; Vogel and Machizawa, 2004; Sauseng et al., 2005; Agam and Sekuler, 2007; Axmacher et al., 2008; Michels et al., 2010; Reinhart et al., 2012; Roux et al., 2012; Lozano-Soldevilla et al., 2014; Zumer et al., 2014; Johnson et al., 2018a; Sato et al., 2018; Reinhart and Nguyen, 2019; Goddard et al., 2022; Pavlov and Kotchoubey, 2022), and lesion-to-deficit studies (Ferreira et al., 1998; Hillary et al., 2006; Olson et al., 2006; Chase et al., 2008; Kas et al., 2011; Jeneson et al., 2012; Bowren et al., 2020). The initial processing of visuospatial information is carried out by lower- and higher-order visual areas in the occipital and temporal lobes within 200 ms after stimulus onset (Vogel and Machizawa, 2004; Agam and Sekuler, 2007; Reinhart and Nguyen, 2019; Peylo et al., 2022). Investigators have further highlighted the causal role of the medial temporal lobe in visuospatial working memory encoding and maintenance, in addition to the formation and retrieval of long-term memory (Olson et al., 2006; Axmacher et al., 2008; Jeneson et al., 2012; Suthana et al., 2015; Wu and Buckley, 2022). Meanwhile, the dorsolateral prefrontal cortices and pIFG are suggested to play a critical role in the maintenance and manipulation of mental representations of visuospatial information (Ferreira et al., 1998; Hillary et al., 2006, Chase et al., 2008; Johnson et al., 2017; Davoudi et al., 2021; Parto Dezfouli et al., 2021). fMRI studies have clarified the spatial extent of functional connectivity modulations during visuospatial working memory tasks, as defined by time-specific co-activation (or co-deactivation) in two distinct cortical regions, and the network showing such task-related enhancement of functional connectivity involved extensive regions, including the prefrontal and visual cortices of each hemisphere (Toepper et al., 2014; Elton and Gao, 2015; Shine et al., 2015; Galeano Weber et al., 2017; O'Connell and Basak, 2018; Di and Biswal, 2019; Finc et al., 2020; Lugtmeijer et al., 2023). However, the temporal dynamics of functionally-connected neural networks underlying visuospatial working memory, along with their supporting white matter tracts, are poorly understood. It is also unclear how these dynamics adapt as humans become familiarized with a task across successive trials and how these changes contribute to improvement of working memory performance.
This study aimed to investigate the dynamic pattern of neuronal activity conveyed through defined white matter tracts that support visuospatial working memory processes. To achieve this, we utilized a ‘dynamic tractography’ technique that incorporates intracranial EEG (iEEG) recording and diffusion-weighted imaging (DWI) tractography (Kitazawa et al., 2023; Ono et al., 2023). As detailed in the methods section below, for example, we defined functional connectivity modulation in relation to memory load as the simultaneous presence of memory load effects on neural activity within two brain regions and the existence of direct DWI streamlines connecting them. To quantify memory load- and task familiarity-dependent effects on neural activity, we measured event-related high-gamma amplitude at 70–110 Hz, a surrogate marker of neural activation with excellent signal fidelity and temporal resolution (Crone et al., 2006; Nir et al., 2007; Ball et al., 2009; Burke et al., 2014; Buzsáki and Schomburg, 2015; Rich et al., 2017; Sonoda et al., 2022). Event-related high-gamma augmentation is tightly correlated with increased firing rate on single neuron recordings (Mukamel et al., 2005; Ray et al., 2008; Rich and Wallis, 2017; Leszczyński et al., 2020), increased hemodynamic activation on fMRI (Nir et al., 2007; Harvey et al., 2013; Kunii et al., 2013; Hill et al., 2021), and increased glucose metabolism on positron emission tomography (PET) (Nishida et al., 2008), and correlates with behavioral changes induced by direct cortical stimulation (Arya et al., 2018). We previously found that event-related high-gamma amplitude modulation was a better predictor of cognitive outcomes after cortical resection than event-related amplitude modulation of low-frequency bands (Sonoda et al., 2022). In the present study, we created animations illustrating when and where cortical high-gamma activity and functional connectivity through white matter tracts were modulated under varying levels of memory load and task familiarity. We tested the following hypotheses. First, we expected that brain regions supporting memory encoding and maintenance would exhibit high-gamma amplitude enhancement as memory load increased: specifically the visual cortex, medial temporal region, dorsolateral prefrontal cortices, and pIFG (Olson et al., 2006; Agam and Sekuler, 2007; Jeneson et al., 2012; Reinhart and Nguyen, 2019; Peylo et al., 2022; Wu and Buckley, 2022). Second, we predicted that specific brain regions, such as the dorsolateral prefrontal cortices or pIFG, would exhibit high-gamma enhancement as task familiarity increased from trial to trial (Ferreira et al., 1998; Hillary et al., 2006; Chase et al., 2008). We conducted a trial-by-trial analysis and examined whether the high-gamma amplitudes in the regions showing task familiarity-dependent enhancement were predictive of successful recall.
Why is there a need to employ intracranial EEG for evaluating the neural dynamics during visuospatial working memory tasks? Artifacts from the temporal and ocular muscles pose challenges to the quantitative measurement of event-related high-gamma activity via scalp EEG and magnetoencephalography (MEG). As such, many of the prior non-invasive studies on visuospatial working memory have largely reported on modulations of low-frequency band cortical signals instead. For example, noninvasive electrophysiology studies report augmentation of alpha and beta amplitudes during working memory tasks (Sauseng et al., 2005; Mazaheri and Jensen, 2008; van Dijk et al., 2010; Reinhart et al., 2012), while others report alpha and beta attenuation (Proskovec et al., 2018). It has been hypothesized that augmentation of event-related alpha/beta amplitudes during working memory tasks is indicative of disengagement of underlying cortical modules (Hanslmayr et al., 2016; Johnson et al., 2020; Yin et al., 2020). A tight inverse correlation has been reported between high-gamma and low-frequency band amplitudes in the primary sensory and motor cortices during spontaneous body movements (Crone et al., 2006; Ono et al., 2023). To enhance our understanding of working memory task-related modulations of low-frequency band activity, we explored the correlation between task-related amplitude modulations of high-gamma and low-frequency bands in each region of interest (ROI). We examined the possibility of an inverse relationship, in which the attenuation of high-gamma amplitudes would be associated with the augmentation of low-frequency band amplitudes at a given moment, and vice versa. We expected that the current study might offer insights into the potential mechanisms underlying modulations of low-frequency band activities during working memory tasks.