1. INTRODUCTION
Semantic memory is relatively preserved with normal cognitive aging but declines in early stages of Alzheimer’s disease and related dementia (AD-RD) including mild cognitive impairment (MCI) (Salmon et al., 1999; Gainotti et al., 2014; Kraut et al., 2006, 2007; Gustavson et al., 2020). Although cognitively healthy older adults experience some challenges with word retrieval (e.g., recalling names) and engage additional neural resources compared to younger controls (Shafto et al., 2010; Baciu et al., 2016; Haitas et al., 2021), their semantic knowledge remains intact (Verhaegen & Poncelet, 2013; Nicholas et al., 1985). In contrast, older adults with MCI and AD-RD have semantic memory deficits that may go beyond word retrieval issues and can involve degradation of semantic knowledge (Dudas et al., 2005; Joubert et al., 2021). With disease progression from MCI to AD-RD, there is more noticeable semantic memory deterioration (Pelgrim et al., 2021; Aschenbrenner et al., 2015; Almeida & Radanovic, 2022). The majority of this research has been conducted using naming tasks (e.g., confrontational naming, verbal fluency, associations). One of the major challenges in understanding the nature of alterations in semantic memory in MCI and AD-RD relates to difficulty in determining whether deterioration is in word retrieval, accessing semantic memory, or semantic knowledge per se .
A semantic memory task that examines access to semantic knowledge without requiring overt word retrieval is thus ideal to investigate changes in semantic memory retrieval associated with normal versus pathological aging. We designed the Semantic Object Retrieval Task (SORT), which examines how objects are retrieved via explicit evaluation of object-associated features (Kraut et al. 2002; Kraut et al., 2006, 2007). In the SORT paradigm, the features (e.g., “humps” and “desert”) normally facilitate retrieval of an object (e.g., “camel”). While participants are not required to overtly recall the word (e.g., “camel”), they are asked to judge whether the features bring to mind a particular object by pushing the ‘yes’ or ‘no’ button on the response box. Our task is distinct from tasks involving semantic priming, contextual constraints, etc. where participants are asked to judge semantic relatedness between stimuli or if stimuli share a category (probed as individual words/pictures or in the context of a sentence; Kiefer, 2001; Kutas & Federmeier, 2000; Wlotko et al.,2010). In contrast, we explicitly require participants to judge whether the stimuli evoke retrieval of a specific object and not just relatedness.
Electroencephalogram (EEG), with its millisecond temporal resolution and direct measure of post-synaptic neuronal activity, is a useful tool to study neural dynamics associated with semantic memory (Chiang et al., 2016, 2020). Previously, on EEG-based semantic object memory retrieval test (SORT), we observed differences in event-related potentials (ERPs) between (1) cognitively normal younger and older adults, and (2) individuals with MCI and cognitively normal older controls. A frontal ERP between 800-1000 ms differentiated retrieval from non-retrieval trials in cognitively normal older adults but not in younger adults, which was posited to relate to a more extensive search during the non-retrieval trials with aging (Chiang et al., 2014). A similar fronto-parietal potential between 950-1050 ms differentiated retrieval from non-retrieval trials in individuals with MCI but not in cognitively normal age- and education-matched controls (Chiang et al., 2015), suggesting a differentiation between normal and pathological aging in semantic search. This fronto-parietal activity was also found to be correlated with episodic memory performance. Those with worse performance on the logical memory subtest (immediate and delayed recall) of Wechsler Memory Scale- Third Edition (Wechsler, 1997) had a delayed fronto-parietal ERP effect, supporting the value of this task in differentiating normal and pathological aging. Despite the utility of ERPs for examining the temporal unfolding of neural activity, non phase-locked and time-locked activity is treated as noise during averaging of ERPs and thus information that could provide useful information about the underlying neurophysiological process is not included. The current study was motivated by the additional insights that time-frequency EEG analysis (Roach et al., 2008; Cavanagh & Frank, 2014) may provide about semantic memory related to the SORT task, and aimed at examining differences in neural dynamics that could inform how semantic memory functions differ from normal to pathological aging.
Prior EEG studies using time-frequency analysis have suggested functional roles of low and high beta (12-20 and 20-30 Hz, respectively) in lexical semantic retrieval (Bakker et al., 2015; Bastiaansen and Hagoort, 2006; Slotnick et al., 2002; Lewis & Bastiaansen, 2015; Weiss & Mueller, 2012). While theta (4-8 Hz) and alpha (8-12 Hz) oscillations have been linked to word or sentence processing (Bakker et al., 2015; Bastiaansen and Hagoort, 2006; Li & Yang, 2013; Shahin et al., 2009; Chiang et al., 2016). Also, the slower frequency bands (theta, alpha) are thought to be associated with domain-general processes such as cognitive control and attentional demands during semantic processing (Cavanagh & Frank, 2014; Klimesch, 2012). How these different frequency bands specifically relate to semantic memory processing is still under scrutiny; therefore, the current study examined oscillatory dynamics in all three frequency bands (theta, alpha, high/low beta) during the SORT task.
The current literature on EEG oscillations related to semantic memory mostly involves studies on young adults. There are a few EEG studies examining time-frequency responses directly elicited by semantic tasks in normal aging or pathological aging (e.g., MCI) (Markiewicz et al., 2021; Poulisse et al., 2020; Packard et al., 2020; Mazaheri et al., 2018), and none have compared such effects across normal and pathological aging. Since there is not sufficient literature to guide an adequately-informed a priori hypothesis to focus on specific frequency bands or time frames, we used a data-driven approach based on permutation tests to examine neural oscillations on the SORT task.
Our goals were to examine (1) normal aging effects by comparing younger to cognitively normal older adults and (2) pathological aging effects by comparing cognitively normal older adults to individuals with MCI. Our general hypothesis was that changes would be observable at the neural level with no differences in behavioral performance for normal aging effects. In contrast, we expected changes both in task performance and neural oscillations in older adults with MCI compared to their age- and education-matched cognitively normal controls.