Dementia is a group of neurodegenerative disorders that affect human cognition and behaviors, such as decline in memory, impairment in reasoning and language ability, and changes in personality and behavior (McKhann et al., 2011). Alzheimer’s disease (AD) is the main cause of dementia, accounting for approximately 60–80% of cases (Alzheimer’s Association, 2022). Due to the aging of our society, the prevalence of AD dementia has been increasing continuously, representing a major issue for public welfare (Alzheimer’s Association, 2022). The progressive accumulation of amyloid-beta (Aβ) plaques and neurofibrillary tau tangles represent bio-pathological structures in the AD brain (Sperling et al., 2011; Jack et al., 2018). Notably, the pathophysiological process of AD may begin many years before the onset of clinical symptoms (Bateman et al., 2012; Villemagne et al., 2013). At the preclinical stage, cognitive abilities are not yet impaired, but evidence of cortical Aβ deposition may already be present; this is considered the most upstream process in the pathological cascade of AD (Jack, et al., 2013). By the time dementia clinically manifests, brain damage due to synaptic loss and neuronal dysfunction has become irreversible (Kumar et al., 2022). To date, there is no cure for AD or any type of dementia (Kumar et al., 2022). Therefore, detecting cognitive impairment due to AD at its early stages is an urgent prerequisite for therapeutic treatment to slow or halt disease progression before major brain damage occurs (Sperling et al., 2011). The key to this success is the discovery of reliable biomarkers that can distinguish preclinical and prodromal AD from cognitively normal (CN) aging.

Clinically, AD dementia is detected by comprehensive evaluation procedures, which may include history assessment, neurological examinations, neuropsychological tests, and brain imaging (Bon et al., 2011). In addition, cerebrospinal fluid (CSF) Aβ and tau have been considered as core biomarkers of AD in clinical diagnosis and research framework (Blennow, 2017); the decrease of Aβ peptide (i.e., Aβ_42/40) and the increase of tau protein (i.e., threonine P-tau₁₈₁) are commonly reported as potential CSF indicators of AD dementia (Andreasen et al., 1999; Ashton et al., 2022). Recently, research attempt to detect AD dementia using fluid Aβ and tau obtained from plasma or serum as alternative approaches (Lee et al., 2019). Similar to CSF, plasma Aβ and tau protein concentrations have been found in AD with similar trends. However, these findings suffer from inconsistent results across cognitive stages and low diagnostic accuracy (Song et al., 2011; Feinkohl et al., 2020). Other imaging techniques such as magnetic resonance imaging (MRI) and/or positron emission tomography (PET) can be alternatively used to provide practical information on the biological construct of AD pathology as well as the structure and functions of the brain. Nevertheless, they suffer from several limitations that restrict accessibility, such as being costly, invasive (e.g., radiotracers are injected to measure the levels of Aβ in the human brain on PET scan), relying on specializer’s expertise, and time-consuming testing procedures (Delso et al., 2015). The need for a simpler and more accessible brain imaging method has led to an increase in the use of electroencephalography (EEG) to screen AD stages.

Electroencephalography is a method to record the electrical activities of cortical pyramidal neurons in the brain (Monllor et al., 2021). When there is a large number of neurons fired synchronously, sufficient postsynaptic potentials can be produced and thus detectable from the scalp of the brain. The analysis of EEG data therefore can reflect the normal brain functions and abnormal neural activities such as in AD. This scalp EEG method is not only non-invasive, relatively safe, and fast but also inexpensive, widely available, and allows repeated measurements on a vast number of high-risk elderly individuals (Tsolaki et al., 2014). Additionally, variables obtained from EEG variants [i.e., event-related potential (ERP)] can overcome the cultural and educational limitations that may otherwise restrict the feasibility of conducting neuropsychological tests (Polich and Corey-Bloom, 2005). ERPs are very small voltages occurred on the EEG signal as responses to specific events or stimuli (Blackwood and Muir, 1990). These voltage peaks on the ERP signals are time-locked to specific events that reflect the sensation, perception, and cognition of the human brain. ERPs therefore can provide physiological and pathological components related to human cognitive ability.

Numerous studies have validated the feasibility and reliability of resting-state EEG and ERP in relation to cognitive impairment due to AD (Cassani et al., 2018; Paitel et al., 2021; Tarawneh et al., 2021). Notably, EEG spectral measures, synchronization patterns, and ERP latencies have served as potential indicators to screen individuals with mild cognitive impairment (MCI) and AD dementia (Garcés et al., 2013; Paitel et al., 2021; Sedghizadeh et al., 2022). EEG spectral analysis captures the fluctuations of the electrical brain signals in the form of frequency with respect to their amplitude (Monllor et al., 2021). In EEG spectral measures, the frequency components (e.g., peak frequency) reflect the speed of neural oscillation and power density features (e.g., spectrum triangular index) reveal the cortical neural synchronization which are often seen in neurodegenerative patients (Garcés et al., 2013; Paitel et al., 2021; Sedghizadeh et al., 2022). While EEG avoids any form of stimuli, ERP takes the reflection of the brain into account when interacting with a particular stimulus (Blackwood and Muir, 1990). A delayed response on ERP signals demonstrates a deterioration in the cognitive ability to process and recognize the stimuli. Consequently, a slowing of oscillatory brain activity, a reduction in regional brain synchronization and complexity, and a prolonged peak latency are often exhibited in individuals with MCI and AD dementia compared to CN persons (Garcés et al., 2013; Paitel et al., 2021; Sedghizadeh et al., 2022).

As documented in previous studies, the slowing of brain oscillations has been captured by EEG rhythm across the AD continuum and is proportional to the progression of AD (Gouw et al., 2017; Cassani et al., 2018). Decreasing high frequency band power (i.e., alpha and beta), increasing low frequency band power (i.e., delta and theta), increasing the ratio between lower and higher frequency (i.e., theta-to-alpha ratio), and shifting of median frequency toward the lower band are quantitative manifestations of EEG slowing (Gouw et al., 2017; Cassani et al., 2018). Nevertheless, EEG slowing patterns were mostly reported in the dementia stages of AD rather than in the preclinical AD process. A few studies have suggested that although relative alpha power starts to decrease from the preclinical AD stage, alpha peak power does not shift until the prodromal stage of AD (López-Sanz et al., 2016). From the prodromal AD stage, these slower brain oscillations continuously increase along with the severity of cognitive impairment on the AD spectrum (Jafari et al., 2020). Notably, the brain’s slowing pattern might be also influenced by the existence of neurodegeneration and the individuals’ particular demographic characteristics (Gaubert et al., 2019; Babiloni et al., 2020). For example, the EEG slowing pattern has emerged in MCI (locally at temporal areas) and AD persons such that the power of low frequencies increases (Meghdadi et al., 2021), but in preclinical AD individuals with neurodegeneration, a decrease in low frequency oscillations is exhibited compared with those without neurodegeneration (Gaubert et al., 2019). People who exhibit subjective memory complaints and have a higher educational level might also have higher alpha power than those with lower educational levels (Babiloni et al., 2020).

Similarly, ERP latency has been used to reflect the temporal dynamics of brain activity in cognitively impaired individuals. In people with AD pathology, prolonged ERP latencies have been reported as a useful indicator for monitoring or even predicting the progression of age-related cognitive decline (Paitel et al., 2021).

In regards to cortical synchronization, the decrease in brain synchrony and complexity is likely due to reductions in cortical connections and neural network communication (Sun et al., 2020; Ranasinghe et al., 2021). A recent study demonstrated a significant decrease in functional connectivity in individuals with preclinical stages of AD within the medial parietal lobe (precuneus area particularly) in the default mode network using magnetoencephalography (Nakamura et al., 2018). The synchronization indexed by Granger causality matrices and local and global network connecting efficiency were also reduced in prodromal AD compared to the controls (Franciotti et al., 2019). On the other hand, subjective cognitive decline in elderly people with neurodegeneration demonstrated higher spectral entropy and complexity than those without neurodegeneration (Gaubert et al., 2019). These few studies on the preclinical and prodromal stages of AD suggest that EEG slowing, brain synchronization, and ERP delay in these very early stages of the AD continuum might still not be well understood. Therefore, more studies are needed to elucidate changes in EEG features, especially those that are correlated with amyloid beta deposition and neurodegeneration status.

Notably, these abovementioned findings were mostly manifested by EEG signals obtained from multichannel devices. However, with recent advancements in hardware and signal processing, few-channel EEG devices have increasingly been employed to examine neuroelectrical signals in research (Choi et al., 2019, 2020; Yi et al., 2019; Doan et al., 2021). Compared to multichannel EEG techniques, the simplicity of few-channel EEG systems can improve their practicality and portability while maintaining data quality (Gargiulo et al., 2010; Rogers et al., 2016). For example, a two-channel system such as prefrontal EEG techniques. Although the validity and reliability of prefrontal EEG in detecting cognitive impairment due to AD have been confirmed (Choi et al., 2019, 2020; Yi et al., 2019, 2021; Park et al., 2020), more studies employing this technique in the very early stages of the AD continuum are needed to rapidly enhance its practicality in clinical environments.

However, few-channel EEG studies on the early stages of AD that consider neurodegeneration information in addition to Aβ are scarce. Therefore, this study aimed to investigate whether the impact of amyloid load and neurodegeneration could be explained by prefrontal EEG and ERP markers. To evaluate the changes in prefrontal EEG/ERP metrics, we divided the subjects into four groups depending on their amyloid status and neurodegeneration status. The four groups were (1) prodromal AD (pAD) individuals who were amyloid positive and neurodegeneration positive (A+N+), corresponding to the MCI stage due to AD; (2) asymptomatic AD (aAD) individuals who were amyloid positive and neurodegeneration negative (A+N−); (3) MCI(-) individuals who were amyloid negative and neurodegeneration positive (A−N+), corresponding to “suspected non-Alzheimer’s pathologic change with MCI; and (4) cognitive normal (CN) individuals who were amyloid negative and neurodegeneration negative (A−N−). Amyloid-positive groups, pAD and aAD, belong to the Alzheimer’s disease continuum. We hypothesized that both amyloid-positive individuals would present specific EEG/ERP differences from CN or MCI subjects.

This study revealed relationships between prefrontal resting-state EEG/ERP variables in the very early stages of AD pathology as well as in non-AD MCI conditions. Overall, we found that the pAD group exhibited a shift of MDF toward a lower frequency and a delay in the PPL compared to the CN group (Tables 2, 3). Although the p-value did not reach the significance threshold, individuals with aAD also exhibited a trend of reduced brain synchrony indicated by a decrease in STI compared with CN individuals (Table 2). Additionally, the non-AD MCI group demonstrated a pattern of longer PPLs with respect to the CN group, faster EEG rhythms than the pAD group, and lower synchronization than the aAD individuals. These results were found after controlling for potential confounders such as age, sex, and educational level.

Previous studies have reported significant changes in EEG/ERP features in individuals with cognitive impairment due to AD. Slower EEG rhythms, a loss of signal synchronization, and a delay in stimulus- response time are commonly found, especially in stages of the AD spectrum that are associated with dementia (Jeong, 2004; Dauwels et al., 2011; Howe et al., 2014; Fruehwirt et al., 2019). Our findings in these features are consistent with these studies. Nevertheless, these patterns were not homogeneously exhibited in the preclinical AD stage and people with non-AD MCI (Gaubert et al., 2019; Meghdadi et al., 2021).

First of all, in terms of MDF, we found that the pAD group exhibited significant differences with the CN individuals, while MCI(-) had no significant relationship with MDF when compared with the CN individuals. Additionally, although the aAD group showed a significant association with MDF in Table 3, MDF did not significantly differ between aAD and CN as shown in Table 2. These results suggested that slowing of brain oscillations which was indicated by a reduction of MDF did not occur in the aAD stage or MCI(-) but was only present in a later stage of pAD when neurodegeneration appeared. Meghdadi et al. (2021) reported a similar slowing in EEG oscillation in the AD group but not strongly significant in the MCI individuals by an increase of spectral power at the slower frequencies (Delta, Theta) on a twenty-channel EEG device. The other studies also reported slower EEG patterns in MCI and AD groups from the multi-channel EEG techniques (Czigler et al., 2008; Ya et al., 2015). The slowing of brain oscillations is thought to be caused by the loss of cholinergic innervation, which is indicated by the reductions in cholinergic-monoaminergic interactions and atrophy of cholinergic nuclei (Dringenberg, 2000; Schumacher et al., 2020). This cholinergic denervation mechanism remains active in the early stages, such as MCI (Mesulam et al., 2004; Teipel et al., 2011), and even before the onset of clinical symptoms due to AD (Beach et al., 2000). Nonetheless, this suggested that EEG slowing may be an AD-specific pattern and can only be observed from the pAD stage of the continuum.

Previous studies have demonstrated a decrease in prefrontal MDF in eyes-closed resting-state EEG in cognitively healthy participants (Choi et al., 2020) that positively correlated with the MMSE score (Choi et al., 2019). In this study, MDF exhibited a significant positive correlation with the MMSE score upon controlling for the potential covariates such as age, sex, and education years as shown in Supplementary Figure 2. In addition, this MDF reduction has been found in patients experiencing a decline of brain functions, such as individuals with postoperative delirium or those exposed to methyl bromide. For instance, the postoperative delirium group showed a lower preoperative MDF than the non-delirium group; surprisingly, a sharp decrease in prefrontal MDF was observed after surgery under general anesthesia in both groups, and this effect remained even after 1 month (Kim et al., 2022). Similarly, a significant decrease was observed in prefrontal MDF after work exposure to methyl bromide, which is a fumigant that is widely used for killing pests (Park et al., 2020). These previous studies consistently pointed out that prefrontal MDF reduction is related to a decline in brain functions. Similar to anesthetics and pesticides, amyloid plaque deposition is neurotoxic, which may cause MDF lowering in the group with a high amyloid load.

On the other hand, we found that the aAD group, who had Aβ deposition but had not yet shown neurodegeneration damage, showed a trend of a reduction in EEG signal synchronization compared to the CN (Table 2) and MCI(-) persons (Table 4). Relationships between neuro-desynchronization and the reduction in neuronal communication and cortical connections have also been previously reported (Sun et al., 2020; Ranasinghe et al., 2021). However, this reduction did not remain significant in pAD individuals, as shown by insignificant ORs and mean differences from the CN and MCI groups (Tables 2, 4). A recent study showed that functional connectivity across long-term neural networks decreases (becoming more abnormal) with the progression of AD neurodegeneration (Bokde et al., 2009). If the functional connectivity in the resting state network is strong, synchronization with enhanced temporal phase modulation occurs, which induces an increase in spectral power of EEG activity around the synchronized frequency and consequently increases the STI value. In this study, the aAD group showed a reduced STI compared to the non-AD MCI patients, and this association remained significant after adjusting for age, sex, and education levels, as shown in Table 4. A previous study reported that patients with mild AD featured a prominent reduction in the likelihood of synchronization of alpha rhythms in fronto-parietal regions compared to non-Alzheimer’s dementia patients (vascular dementia). They suggested that mild AD patients were characterized by the fronto-parietal coupling of dominant alpha rhythms (Babiloni et al., 2004). This report is in alignment with the STI results in this study in the aAD group when compared with the MCI(-) individuals. Although the aAD group exhibited a lower mean STI than the CN group, the difference was not statistically significant which might be due to the small sample size. Therefore, future studies with larger sample sizes are needed to understand the change in EEG synchrony in the early stages of the AD continuum.

Furthermore, the PPL was found to be related to pAD and MCI(-). We found a significantly prolonged PPL in pAD and MCI(-) patients, as indicated by higher mean values and odds ratios that were greater than one. The prolonged peak latency was suggested to be caused by a reduction in cognitive processing ability requiring participants to need more time to process sensory information. Nevertheless, this result suggested that this delay in the sensory function of the brain might not represent AD-specific changes but rather a general cognitive decline. ERP responses to auditory stimuli contain discriminative information that predicts which MCI patients are likely to progress to more severe cognitive impairment (Bennys et al., 2011), and ERP variables of cognitive function are increasingly affected in longitudinal studies on MCI and AD patients (Lai et al., 2010; Papaliagkas et al., 2011). Moreover, people with the causative genes of AD such as amyloid precursor protein (APP) and presenilin 1 (PSEN1) showed significant ERP changes even before the onset of AD-related behavioral abnormalities (Golob et al., 2009; Quiroz et al., 2011). Golob et al.’s (2009) showed significantly delayed latencies for the early ERP components among familial AD mutation carriers. In an ERP, the P200 peak that occurs at approximately two hundred milliseconds after stimulus onset is considered to reflect exogenous sensory attention (Katada et al., 2004). Delayed P200 latency has been found in patients with sporadic or familial AD and other abnormalities (Martinelli et al., 1996; Missonnier et al., 2007), indicating that early sensory-cognitive processes might be compromised in AD. We were also able to confirm that the PPL marker, corresponding to the latency of P200 in auditory ERP, became delayed with AD progression, as shown in Tables 2, 3. This is consistent with our previous result; PPL showed a negative correlation with the MMSE score (Supplementary Figure 2; Doan et al., 2021) and was delayed in the dementia group compared to the non-dementia group (Doan et al., 2021).

There have been recent studies on the predictive utility of EEG/ERP variables for dementia. Compared with other imaging modalities, EEG/ERP can promptly detect subtle changes in brain function, especially those indicating early-stage cognitive decline, due to its high temporal resolution. While most previous studies used multichannel EEG techniques, this study used a simple and wearable EEG band with two prefrontal channels at Fp1 and Fp2. A previous study reported that prefrontal EEG/ERP variables outperformed MMSE scores in predicting dementia; combining both measures can increase prediction accuracy (Doan et al., 2021). Therefore, prefrontal EEG/ERP has wide potential applications as a non-invasive, wearable, and cost-effective screening tool for cognitive impairment. To the best of our knowledge, however, there have been no studies on the predictive utility of prefrontal EEG/ERP variables in the very early stages of the AD continuum. Although data analysis from EEG has already led to the proposal of putative biomarkers for MCI and AD, studies on cognitively normal Alzheimer’s or non-Alzheimer’s MCI were rare. Our study is the first report on the EEG/ERP variable changes in the preclinical and prodromal stages due to AD.

The EEG variables of MDF and STI and the ERP variable of PPL used in the current study indicate the functional degeneration corresponding to the functional slowing and reduced connection in the resting state network (RSN) and task positive network (TPN), respectively. Those variables have low dependence on absolute signal amplitude, thus not sensitive to scalp thickness or contact resistance, and low dependence on delta and gamma bands, thus not sensitive to movement or muscle contraction (Definitions and detailed calculation methods for MDF, STI, and PPL are described in detail in “section 2.4. EEG/ERP variables and computation,” in which low dependences on signal amplitude and on delta/gamma bands can be easily inferred). Therefore, MDF, STI, and PPL variables have high reproducibility and robustness against artifacts and scalp morphology variabilities between individuals, which is advantageous to practicality in real clinical usage.

Meanwhile, MDF decrease has been reported in various disease conditions such as aging, cognitive decline, toxic exposure, and postoperative delirium among others. It implies that MDF decrease and probably STI and PPL changes may not be AD-specific but reflects brain functional changes occurred by AD progression, which is a limitation of EEG variables in general.

Several limitations of this study need to be addressed. Mainly, the number of participants for each group was relatively small, thus limiting the uses of more advanced EEG/ERP features as well as the generalizations of these results. A larger cohort can help validate these findings as well as additionally elucidate the relationship between EEG/ERP variables and predementia stages of the AD spectrum. Additionally, we were not able to observe and monitor the participants over time. Longitudinal studies may help provide additional insight into how these EEG/ERP features change along with the progression of AD dementia.

This study employed two resting-state EEG variables and one sensory ERP marker obtained from the two prefrontal EEG channels to examine the associations with predementia stages of AD and non-AD MCI. We found that a slower intrinsic EEG oscillation was associated with MCI due to AD (prodromal AD), whereas a prolonged ERP latency was more likely to be related to global cognitive impairment rather than AD-specific impairment. Additionally, a trend of reduced brain synchronization was observed in the preclinical stage of AD. These findings provided additional insight into how EEG/ERP features change in the very early stages of the AD continuum and suggested that prefrontal EEG/ERP variables can serve as potential markers for screening early stages of cognitive impairment due to AD.

The datasets presented in this article are not readily available because of ethical and access restrictions. Requests to access the datasets should be directed to the corresponding authors.

The study protocol was approved by the Institutional Review Board of Chonnam University Hospital (CNUH-2019-279). The patients/participants provided their written informed consent to participate in this study.

JC led the manuscript preparation. BK analyzed the data and wrote the manuscript. DD and JP handled the manuscript qualification and reviewed the manuscript. JC and WC extracted the relevant EEG/ERP variables and controlled the data quality. KL designed and led the project, obtained the institutional review board approval, and managed the data. JK supervised and wrote the manuscript. All authors revised and approved the contents of the manuscript, contributed to the article, and approved the submitted version.