4. DISCUSSION
By utilizing a semantic memory paradigm that did not require explicit word retrieval, we were able to identify different patterns of EEG oscillations between normal aging versus pathological aging. When comparing YA and HA groups, no significant behavioral differences were seen but three within-group condition effects were identified on EEG measures: differences in (1) theta ERS and (2) alpha ERD between conditions were observed in YA but not in HA after 500 ms post-stimulus onset, and (3) a difference in high beta band ERS between conditions in the HA group but not in YA group during 400-500 ms post-stimulus onset. When individuals with MCI were compared to HA, they performed worse than HA across conditions accompanied by two EEG effects: (1) a difference in low beta ERS between conditions was observed in the HA group but not in the MCI group during 400-500 ms post-stimulus onset and (2) a difference in high beta ERS between conditions, divergent in pattern, was noted between MCI and HA between 224-261 ms post-stimulus onset.
Our behavioral data showed preserved accuracy and RT in the HA group compared to the YA group consistent with findings from our prior studies and the literature supporting relatively preserved access to semantic memory with normal aging, (Chiang et al., 2014; Verhaegen & Poncelet, 2013; Rönnlund et al., 2015). However, differences between YA and HA groups observed on EEG suggest early age-related changes emerge between conditions during 400 and 500 ms post-stimulus onset. The HA group showed differences in high beta between retrieval and non-retrieval conditions relative to the YA group. High beta has been associated with retrieval of semantic information (Slotnick et al., 2002; Ferree et al., 2009; Hart et al., 2013; Lewis & Bastiaansen, 2015; Weiss & Mueller, 2012). Differences across conditions observed in the HA group suggest recruitment of additional neural resources, mediated by high beta activity, to support varying demands related to semantic processing between the two conditions. In contrast to HA, the YA group showed differences across conditions after 500 ms post-stimulus onset in theta and alpha frequency power. In particular, consistent with our prior findings within similar time windows (Chiang et al., 2016; Ferree et al., 2009; Hart et al., 2013), YA demonstrated increased alpha ERD during retrieval compared to non-retrieval conditions, which might be linked to semantic integration of features which lead to object memory activation. In the subsequent time window (628 to 701 ms post-stimulus onset), YA had higher theta ERS during non-retrieval compared to retrieval conditions, which might be related to enhanced search for potential feature integration prior to making a final judgment (Yes/No response on button push). In HA, the loss of such differentiation in theta and alpha responses (as shown in Supplemental Figures 2 & 5) might suggest age-related neurophysiological alterations on both retrieval and non-retrieval conditions as processing related to the task continues. However, these neural alterations in HA at later time frames do not seem to be enough to impact task performance, which also implicates that the earlier high beta differential effects in HA may be a compensatory mechanism to support semantic memory function. Given that both retrieval and non-retrieval conditions elicited similar magnitude of changes in neural responses (i.e., alpha ERD and theta ERS) in older adults, this finding is consistent with the literature that suggests dedifferentiation and reduced segregation in neural activity in healthy aging (Koen & Rugg, 2019; Martin et al., 2022).
Our task was able to differentiate the HA group from individuals with MCI on behavioral task performance as well as EEG measures. Consistent with our hypothesis, lower accuracy was noted in the MCI group compared to the healthy older adult group, suggesting impaired semantic memory performance, which has been reported in our prior studies and the literature (Chiang et al., 2015; Kraut et al., 2007; Dudas et al., 2005; Joubert et al., 2021). We also found greater low beta ERS in retrieval compared to non-retrieval conditions in the HA group but not in the MCI group in earlier time windows (< 500 ms post stimulus onset). These findings suggest that while the HA group differentially allocates neural resources, mediated by low beta, across the two conditions, these differentiations are lost in the MCI group. Interestingly, within the first 300 ms post-stimulus onset, both groups showed differences in high beta power across conditions, albeit different in patterns. While the HA group had higher high beta power in retrieval compared to non-retrieval conditions, an opposite direction was observed in the MCI group with non-retrieval greater than retrieval conditions. It is unclear what this reverse pattern indicates, but these findings suggest that HA and MCI groups may use different strategies to retrieve or process semantic information during the task. Nevertheless, the MCI group shows patterns of neural oscillations that can be differentiated from the HA group on the SORT task and these alterations may be indicative of early neural decompensation, as suggested in prior studies (Woodard et al., 2009; Barbeau et al., 2012; Pistono et al., 2021).
Overall, with age, there is a shift in differentiation across conditions (retrieval vs non-retrieval) from slower frequencies (alpha, theta) to faster frequencies (low and high beta) on the SORT paradigm. The transition from slower to faster brain oscillatory responses may be a sign of aging in general. While both normal and pathological aging effects were specifically identified in high beta power, the YA and HA groups differed in the time frame of 444-481 ms post stimulus onset, which is later than the time frame of 224-261 ms which differentiates the HA and MCI groups. It is possible that alterations in high beta in the earlier time window is a sign of pathological aging, while high beta in the later time window reflects normal aging. Given the evidence that integration of semantic memory was thought to be associated with high beta frequency (25-30 Hz) activity from both the cortex and thalamus during the SORT task (Slotnick et al., 2002; Hart et al., 2013), it is plausible that differential changes in cortico-thalamo-cortical synchronization (Crosson, 2021) may underlie the changes we found in high beta frequency in normal versus pathological aging, though during an earlier time frame. It has been shown that changes in thalamic function and cortico-thalamo-cortical connectivity can occur in normal aging (Goldstone et al., 2018; Fama & Sullivan, 2015) as well as in MCI (Hahn et al., 2016; Zhou et al., 2013; Aggleton et al., 2016). Future work that combines EEG and MRI techniques would be useful to explore these possibilities.
Behavioral measures of semantic memory (e.g., semantic/category fluency) have already shown promising predictive value in risk of progression to MCI and dementia (Sutin et al., 2019; Gustavson et al., 2020). EEG markers of semantic memory will provide additional insights given that neural changes may precede behavioral changes in pathological aging (Olichney et al., 2002; Jack et al., 2013). An important implication of our study is that these brain oscillatory responses can potentially be used as outcome measures to assess efficacy of interventions (e.g., cognitive training, pharmacological treatment) in order to gain a mechanistic view. Furthermore, with the advent of non-invasive brain stimulation methods, these oscillatory brain responses may serve as targets that can be modulated directly by electrical stimulation to stabilize compensatory processes or to interfere pathological processes, and thus to maintain or even improve cognitive function in aging populations (Grover et al., 2022).