Introduction

Cognitive decline is a defining feature of Alzheimer’s disease (AD) and mild cognitive impairment (MCI), but it is also commonly seen in patients with diabetes, Parkinson’s disease, HIV infection, and many other conditions and in its milder form is also a feature of healthy aging [32, 67, 70, 72]. Over five million Americans are currently living with AD spectrum conditions (i.e., AD/MCI), and this is expected to increase to about 13.8 million by the year 2050 [1, 2, 30, 44]. In light of these ominous numbers, understanding the neural pathophysiology that gives rise to the associated cognitive deficits is of critical importance. The severity of AD is known to be associated with the accumulation of beta-amyloid proteins and tau neurofibrillary tangles in the brain [33], with both leading to synaptic damage and neurodegeneration. The gradual buildup of plaques and tangles is known to begin well before significant symptoms emerge and continues for decades [40]. Some of the earliest symptoms include subtle declines in memory and language abilities [12].

While the incidence and prevalence of AD is on the rise, it is currently outpaced by diabetes mellitus, which affects more than 38 million people in the USA [11, 14, 45]. Cognitive function and diabetes are generally seen as having a bidirectional relationship, as diabetes can lead to cognitive decline and such impairments can affect one’s ability to properly manage the disease, including glucose monitoring and medication adherence, which when poorly managed leads to more severe brain dysfunction over time [35, 37]. Further, multiple diabetes-related complications can increase the risk of developing cognitive decline and AD, including vascular damage due to poor management of diabetes, heart disease, and other factors [12, 52]. There is also a known association between the frequency of abnormal glucose levels, including episodes of both hypoglycemia and hyperglycemia, and risk for AD and related dementias [49, 61]. Previous work has shown that higher blood glucose levels (i.e., higher A1C) are significantly associated with AD at both the diabetes and prediabetes stages [46]. Similar to the bidirectional relationship with cognitive dysfunction, poor glycemic control may lead to sleep disruptions and poor sleep is in turn associated with altered glycemic control, insulin resistance, and other health problems, including increased risk for AD [7, 23, 52]. Further, prior work has shown that middle-aged adults with diabetes exhibit cognitive decline over 20 years at a much higher rate than those without diabetes [76], leading some to suggest that AD could be referred to as type 3 diabetes [34].

Many recent studies of AD-related pathophysiology have relied on magnetoencephalographic (MEG) imaging, which provides a direct measure of ongoing neuronal activity with millisecond temporal resolution and good spatial precision. Several features of MEG make it a particularly attractive choice for studies of AD, aging, and/or diabetes, including the fact that it is noninvasive, silent, non-claustrophobic, and a direct measure of neural activity that is not dependent on intact vasculature [3, 47]. The technique is also applicable to resting-state neural activity, which is often the most feasible study design in populations with AD, although it should be noted that extensive MEG work in AD has also leveraged task-based designs (see [9] for a review). Some of the key findings from resting-state MEG studies in AD include altered levels of slow wave delta and theta activity compared to demographically matched controls [22, 42, 66, 68, 69]. Several of these studies have further shown that these neural aberrations are coupled to impaired performance on neuropsychological scales [50, 68, 69], with the regions involved corresponding to the known functional organization of the cortex. Recent work has also demonstrated colocalization of altered spontaneous delta and elevated amyloid accumulation in AD [51], and that AD-related aberrations in delta also colocalize with dopaminergic, serotonergic, and mu-opioid neurotransmitter systems [66]. Overall, these data have given rise to the perspective that pathological changes in slow wave activity are coupled to risk factors for AD and may be critical to the progression of AD symptomatology, although additional work is needed to fully understand the implications and establish the specificity for AD spectrum disorders.

In the current study, we evaluate whether the strength of spontaneous neural activity during the resting-state is differentially associated with glycemic control in patients on the AD spectrum (ADS; i.e., MCI or AD) relative to healthy aging controls. All participants completed a resting-state MEG recording and the strength of spontaneous neural activity within each canonical band (theta, alpha, etc.) was estimated per vertex. All participants also completed a comprehensive neuropsychological battery and a hemoglobin A1C (A1C) blood test to quantify glycemic control. Our primary hypotheses were that participants on the ADS would exhibit significantly altered relationships between A1C levels and slow wave spontaneous activity in prefrontal cortices, and that this would correlate with measures of higher-order cognition in patients on the ADS.

Methods

Participants

A total of 80 participants were selected for this study from a larger ongoing project. Participants on the ADS were selected based on completion of a MEG resting-state recording, blood test, neuropsychological assessment, and an amyloid-positive positron-emission tomography (PET) scan. A total of 37 biomarker-positive adults met this criterion and were unaffected by the exclusionary criteria listed below. Cognitively normal controls were selected based on normal neuropsychological performance, no amnestic or cognitive complaints, and completion of a MEG resting-state recording and blood test. To best approximate the ADS cohort in terms of age and sex, we selected the oldest 43 control participants who remained eligible after consideration of the exclusion criteria. Demographic and cognitive characteristics of both groups are described in Table 1. All patients on the ADS were diagnosed with either MCI or AD, and were confirmed biomarker positive using Florbetapir 18F PET, as determined by a fellowship-trained neuroradiologist specializing in nuclear radiology (CMJ). All patients who were taking an acetylcholine esterase inhibitor to improve symptoms of cognitive impairment were stable on their medication for at least a month prior to study participation. No patients were taking, or had previously taken, any Aβ-targeted medications. Exclusionary criteria included any neurological or medical illness affecting the central nervous system (except MCI/AD), chronic psychiatric disorders (e.g., schizophrenia, autism, bipolar disorder), history of head trauma, current substance use disorder, and the standard exclusion criteria for MEG/MRI (e.g., ferromagnetic implants).

Table 1 Descriptive statistics by group
Full size table

Note that while we excluded chronic psychiatric conditions, we allowed those that often remit (e.g., depression, anxiety), as such conditions are very common and including these participants makes the sample more representative of the population under study, particularly in studies of AD. Multiple large-scale studies, including the Human Connectome Project in Aging, have taken a similar approach and allowed such common conditions. In the current sample, two patients in the ADS group and zero controls endorsed anxiety on their demographic form, while six patients and three controls endorsed depression. In addition, all participants except one control completed the Beck Depression Inventory (BDI), which indicated that one control and five patients in the ADS group had mild or moderate symptoms, one ADS patient had severe depression symptoms, and all other participants fell into the minimal or no depression symptoms range. Importantly, we recomputed our main analyses without the one patient with severe depression symptoms and the results were unchanged, so we left this participant’s data in for all analyses. Medication-wise, 10 controls and 19 patients on the ADS were taking depression/anxiety medications, but caution is warranted as SSRIs are frequently prescribed to ADS patients to help manage behavior regardless of depression. Thus, these medications may not reflect depression treatment and our BDI data support that interpretation.

Ethics

The Institutional Review Board at the Boys Town National Research Hospital approved the study and all research protocols complied with the Declaration of Helsinki. Written informed consent was obtained from each participant after a full description of the study. Patient study partners, or informants, were also frequently present for participants on the ADS to ensure their comfort over the course of the study, and as such, informed consent was obtained from each partner/informant as well. For patients whose capacity to consent was questionable, informed assent was obtained from the research participant in addition to informed consent from a legally authorized representative.

Blood collection and hemoglobin A1C processing

Participants underwent a panel of blood tests according to the standards of care described by the American Diabetes Association. These tests included glycated hemoglobin (A1C), creatinine, glucose, aspartate transaminase, alanine transaminase, albumin/creatinine, thyroid-stimulating hormone, and B12. Herein, we focus on the A1C measures given our hypotheses. A1C measures the approximate average of an individual’s blood glucose value over the span of the last three months. Of note, a person who does not have diabetes will generally have an A1C under 5.7%, while those over 6.5% most likely have diabetes mellitus. Individuals falling in the 5.7–6.5% range are generally characterized as having prediabetes [17]. Patients with diabetes are considered to be well-controlled, via medical or lifestyle interventions, with an A1C under 7%.

MEG data acquisition

MEG recordings took place in a two-layer magnetically-shielded VACOSHIELD room (Vacuumschmelze, Hanau, Germany). Participants were seated in a nonmagnetic chair within the room, with their head positioned within the sensor array of a 306-sensor MEGIN TRIUX Neo MEG system (MEGIN, Helsinki, Finland), equipped with 204 planar gradiometers and 102 magnetometers. Neuromagnetic responses were sampled continuously at 1 kHz with an acquisition bandwidth of 0.1–330 Hz. The same instrument was used across all MEG recordings. Participants were instructed to rest with their eyes open and fixate on a centrally-located crosshair for 5 min. They were monitored by a real-time audio video feed from inside the shielded room throughout MEG data acquisition.

Structural MRI acquisition and MEG-MRI coregistration

Individual structural MRI (sMRI) data were obtained with a 3 T Siemens Prisma MRI scanner using a 32-channel head coil with the following parameters: TR: 2400 ms; TE: 2.05 ms; field of view: 256 mm; matrix: 256 × 256; slice thickness: 1 mm with no gap; voxel size: 1.0 × 1.0 × 1.0 mm; acquisition plane: sagittal; flip angle: 8 degrees. All sMRI data were acquired and segmented with the computational anatomy toolbox (CAT12 v12.6; [28]. Structural MRI data were aligned parallel to the anterior and posterior commissures and transformed into standardized space (i.e., MNI ICBM152).

Prior to MEG acquisition, five coils were attached to the participants’ heads and localized, together with three fiducial points and the scalp surface, using a 3-D digitizer (Fastrak, Polhemus Navigator Sciences, Colchester, VT, USA). Once the participant was positioned for MEG, an electrical current with a unique frequency label (e.g., 322 Hz) was fed to each of the coils. This induced a measurable magnetic field and allowed each coil to be localized in reference to the sensors throughout the recording session. Since coil locations were also known in head coordinates, all MEG measurements could be transformed into a common coordinate system. With this coordinate system, each participant’s MEG data were co-registered with their structural MRI prior to source space analyses using Brainstorm.

MEG data pre-processing

Each MEG dataset was individually corrected for head motion and subjected to noise reduction using the signal space separation method with a temporal extension (tSSS; correlation limit: 0.950; correlation window duration: 6 s; Taulu & Simola, 2006). Noise-reduced MEG data underwent standard data preprocessing procedures using the Brainstorm software [52, 60]. A high-pass filter of 1 Hz, low-pass filter of 200 Hz, and notch filters at 60 Hz and its harmonics were applied. Datasets with excessive dental or temporal muscle artifacts were low-pass filtered at 60 Hz; thus, analyses focused on frequency bands lower than this (i.e., delta, theta, alpha, and beta). Cardiac and ocular artifacts (e.g., blinks and eye movements) were identified in the raw MEG data and removed using an adaptive signal-space projection approach, which was subsequently accounted for during source reconstruction [31, 62]. Data were then divided into four-second epochs for detection and rejection of bad segments of data. Amplitude and gradient metrics for each epoch were computed, and epochs containing outlier values were rejected using an individualized fixed threshold method, supplemented with visual inspection. Individually determined thresholds were used in order to account for variance across participants in the distance between the brain and the MEG sensor array, as well as other sources of variance. Individual thresholds were determined based on the signal distribution for both amplitude and gradient. The number of accepted epochs was not different between ADS and HC participants (t[78] =  − 0.17, p = 0.864). Linear regression revealed there was not a direct effect of hemoglobin A1C on trial count (t[76] =  − 0.30, p = 0.763), or a group-by-A1C interaction (t[76] =  − 0.24, p = 0.815).

MEG source imaging and frequency power maps

Source imaging methods and processing largely followed the analysis pipeline outlined in previous publications [50, 71]. Source analysis of neuromagnetic fields used an overlapping-spheres forward model, unconstrained to the cortical surface. This approach models a single sphere per sensor, which included 204 gradiometers, minus any channels previously marked as “bad” for each participant. A linearly constrained minimum variance (LCMV) beamformer implemented in Brainstorm was used to spatially filter the epoched data based on the data covariance computed from the resting-state recording. Using these source estimates, we then estimated the power of cortical activity in each of four canonical frequency bands: delta (2–4 Hz), theta (5–7 Hz), alpha (8–12 Hz), and beta (15–29 Hz).

Welch’s method was used for estimating power spectrum densities (PSD) per 4 s epoch, with 1-s sliding Hamming windows overlapping at 50%. To create relative maps, we standardized the raw PSD values of each epoch at each frequency bin to the total power across the frequency spectrum, thus these maps are measured in percentages. We then averaged the PSD maps across epochs to obtain one relative PSD map per participant. Finally, the norm of the three unconstrained orientations per location were then projected onto a common MNI ICBM152 template surface.

Neuropsychological assessment

All participants underwent an extensive battery of neuropsychological tests in collaboration with a clinical neuropsychologist specializing in memory disorders (PEM-W). Scores with a skewed distribution were transformed to normalize the distribution based on skewness and kurtosis values. Raw scores for each participant were then converted to Z-scores using the mean and standard deviation (SD) of the control population. Following this, scores on each assessment were averaged in order to create domain scores: learning (Hopkins Verbal Learning Test–Revised [HVLT-R] Learning Trials 1–3 [6], Wechsler Memory Scale 4th Edition [WMS-IV] Logical Memory I [64]), memory (WMS-IV Logical Memory II [64], HVLT-R Delayed Recall [6]), attention (Wechsler Adult Intelligence Scale [WAIS-IV] Digit Span Forward, Backward, and Sequencing [63]), processing speed (WAIS-IV Coding [63], Trail Making Test Part A [29], Stroop Color and Word Subtests [13], executive function (Trail Making Test Part B [29] and Stroop Interference [13] and language (Boston Naming Test, Phonemic Verbal Fluency [FAS], Semantic Verbal Fluency [Animals] (R. K. [29]) cognitive domains. General cognitive status was measured using the Mini-Mental State Examination (MMSE; [25] and clinical staging was based on the Quick Dementia Rating System (QDRS; [27].

Statistical analysis and visualization

Whole-brain statistical analyses were performed on the source-imaged MEG data using a general linear approach in SPM12. To account for non-uniform spatial autocorrelation in the data, avoid assumptions of parametric modeling, and avoid selecting arbitrary cluster-forming thresholds, threshold-free cluster enhancement (TFCE; E = 1.0, H = 2.0; 5000 permutations; [54] was performed, with multiple comparisons correction set to cluster-wise pFWE < 0.05. Clusters surviving at this threshold were used to create logical masks that were applied to the original statistical contrasts (i.e., vertex-wise F-values) for visualization in Brainstorm. Initial tests investigated the interaction between group (controls vs. ADS) and hemoglobin A1C on resting-state power in each frequency band, controlling for age, sex, and days between MEG and A1C collection. Data were extracted from the vertex exhibiting the strongest relationship in each significant cluster and plotted using ggplot2 [65] for interpretation of directional effects and further statistical analyses.

In addition, we wanted to determine the potential mediating effects of spontaneous neural activity on the relationship between hemoglobin A1C and neuropsychological performance, specifically in the patients on the ADS. To this end, we used structural equation models (SEM) controlling for age, sex, education, and day between collections. In order to determine the significance of indirect effects, we examined the 95% confidence intervals (CI) of bias-corrected bootstrapped CIs based on 1,000 bootstrapped samples [16]. This method provides a robust estimate of asymmetrical mediation effects [26], as traditional tests of indirect effects (e.g., Sobel test) often violate the assumption of normality. All mediation analyses were conducted with full information maximum likelihood estimation for missing data using the lavaan package in R (V4.4.1).

Results

Demographic information for the 80 participants is reported in Table 1. The two groups significantly differed in age, with participants in the ADS group being slightly older than the controls (ADS: 73.23 years, Controls: 69.77 years, p = 0.038), but were otherwise demographically similar. The two groups were also comparable in A1C scores (ADS: 5.74%, Controls: 5.85%, p = 0.410) and the number of participants scoring in the range for prediabetes (ADS: 19 participants, Controls: 17 participants, p = 0.369) and diabetes (ADS: 5 participants, Controls: 6 participants, p = 0.955). As expected, the ADS group scored significantly worse than the controls on the MMSE (ADS: 24.14, Controls: 29.17, p < 0.001) and all seven domains of neuropsychological function (see Table 1). Regarding the composition of the ADS group, clinical staging based on the QDRS indicated that 15 participants were in the MCI range (QDRS scores: 1.5–5.5, mean: 3.57), 20 were in the mild/moderate dementia range (QDRS scores: 6.0–18, mean: 8.25), one was in the normal range, and one participant was missing QDRS scores. Importantly, the patient in the normal range was impaired on multiple more comprehensive measures, and their normal score on the QDRS likely reflected potentially biased responses from their informant. As expected, those at less severe clinical stages exhibited higher mean MMSE scores (MCI: 24.60, AD: 23.55) and scores on each of the seven neuropsychological domains (learning: MCI: 37.05, AD: 34.26; memory: MCI: 30.33, AD: 25.97; attention: MCI: 46.88, AD: 45.55; processing speed: MCI: 43.76, AD: 35.75; executive function: MCI: 43.40, AD: 38.28; language: MCI: 45.71, AD: 43.25; motor: MCI: 43.17, AD: 37.03). Importantly, all of our primary analyses collapsed across severity levels and focused on the ADS and control groups.

Hemoglobin A1C and disease status have interactive effects on resting-state neural activity

Significant interactions between Hemoglobin A1C and disease status (i.e., ADS vs controls) were identified in the delta (Fig. 1) and alpha (Fig. 2) frequency bands. The interactive effect on delta power extended across frontotemporal and inferior parietal cortices, with the strongest effects in the left superior frontal gyrus (Fpeak = 10.55, pFWE < 0.001) and the right superior frontal gyrus (Fpeak = 9.23, pFWE = 0.035). This effect was such that, in the ADS group, delta power decreased with higher hemoglobin A1C, while the opposite pattern of delta power increases with higher hemoglobin A1C was observed in the control group. Interactive effects on alpha power were identified primarily in inferior frontal and temporal regions, with the peak effects in the left frontotemporal cortex and posterior middle temporal gyrus (Fpeak = 6.96, pFWE = 0.001). In contrast to the delta band, this interaction was such that alpha power decreased with higher hemoglobin A1C in the control group, with the opposite pattern being observed in the ADS group.

Fig. 1
Fig. 1
Full size image

Group-by-hemoglobin A1C interactions on spontaneous delta activity. Whole-brain ANCOVA, controlling for age, sex, and days between blood collection time and MEG, on spontaneous delta (δ, 2–4 Hz) activity. Scatter plots display relative power values (%) extracted from the peak voxel of each cluster (left hemisphere on the left, right hemisphere peak on the right). Significance of the relationship between hemoglobin A1C and neural activity is shown in each scatterplot. The HC group is represented in green and the ADS group in red. The 3D brain maps denote significant clusters where interactions were identified based on a cluster-level pFWE <  0.05 threshold. F-values across each cluster are indicated by the color scale bar. HC: healthy controls; ADS: Alzheimer’s disease spectrum

Fig. 2
Fig. 2
Full size image

Group-by-hemoglobin A1C interactions on spontaneous alpha activity. Whole-brain ANCOVA, controlling for age, sex, and days between blood collection time and MEG, on spontaneous alpha (α, 8–12 Hz) activity. Scatter plots display relative power values (%) extracted from the peak voxel. Significance of the relationship between hemoglobin A1C and neural activity is represented within the scatterplot. The control group is represented in green and the ADS group in red. The 3D brain map shows significant regions identified based on a cluster-level pFWE <  0.05 threshold. F-values are indicated by the color scale bar. HC: healthy controls; ADS: Alzheimer’s disease spectrum

Resting state delta mediates the relationship between hemoglobin A1C and attention in ADS

We used mediation models to test if hemoglobin A1C-related delta and alpha activity (i.e., Figs. 1 and 2) mediated the relationship between hemoglobin A1C and domain-specific neuropsychological performance in patients on the ADS (Fig. 3). We found a significant indirect effect of A1C predicting attention scores via delta power in the left superior frontal gyrus (βindirect = 0.16, 95% CI [0.01, 0.75]). There was not a direct effect of A1C on attention scores (β =  − 0.04, p = 0.868), but A1C did significantly predict delta activity (β =  − 0.44, p = 0.004). Further, the effect of delta activity on attention scores was trending (β =  − 0.36, p = 0.098), such that weaker delta activity was related to better attention performance. No mediation effects were observed for alpha activity.

Fig. 3
Fig. 3
Full size image

Hemoglobin A1C and delta activity predict attention in patients on the Alzheimer’s disease spectrum. Potential indirect effects of delta activity mediating the relationship between A1C and attention were tested using structural equation modeling. The model indicated a mediating effect of delta activity in the left superior frontal gyrus (SFG) on the relationship between hemoglobin A1C and attentional performance. Age, education, sex, and days between sample collection time and MEG were included as covariates of no interest. Each path denotes standardized coefficients. **p <  0.01. †p <  0.10

Discussion

In this study, we examined whether the strength of spontaneous neural activity during the resting state was differentially related to glycemic control in patients on the ADS and cognitively normal controls. To this end, participants completed a resting-state MEG recording and a blood draw to derive their level of glycemic control (i.e., A1C). We then imaged spontaneous cortical activity in each of the canonical frequency bands (e.g., alpha) using the MEG data and determined that there were significant interaction effects between A1C and disease status (i.e., ADS vs. control) in the delta and alpha bands. The brain regions implicated in the delta frequency interaction extended across a broad swath of frontotemporal and inferior parietal cortices, with the strongest effects in the bilateral superior frontal regions where delta power significantly decreased with increasing A1C in patients on the ADS, while controls showed a slight trend in the opposite direction as delta power tended to increase with increasing A1C. In contrast, the interaction effect in the alpha range extended across inferior frontal and temporal cortices, peaking in the left frontotemporal cortex and posterior middle temporal gyrus, and primarily reflected decreases in alpha power with higher hemoglobin A1C in the control group, with the opposite pattern (nonsignificant) being observed in the ADS group. Finally, we found that delta activity in the left superior frontal gyrus mediated the effect of A1C on neuropsychological performance in the attention domain in patients on the ADS. Below, we discuss the implications of these important findings.

Our most interesting findings in this study were likely the differential relationship between glycemic control and spontaneous delta activity in bilateral frontotemporal cortices. Abnormal delta activity has been frequently reported in those with AD and/or MCI [5, 24, 36, 43, 53, 66, 68, 69], and thus, we were not surprised to see that delta levels in critical prefrontal brain regions were significantly linked to glycemic control. Related work examining AD risk factors has also shown that sleep efficiency is differentially related to spontaneous delta power in individuals on the ADS compared to controls [50, 51], which is broadly consistent with our findings linking glycemic control (i.e., another risk factor) with cortical delta power. In fact, the group-wise directionality differences observed in the current study between A1C and delta power were remarkably similar to those observed between sleep and delta power in prior work. In the current study, we propose that the nonsignificant increases in delta with higher A1C in controls reflects the common finding that negative health factors (and emergent neurological conditions) are associated with pathological increases in low-frequency activity (i.e., neural slowing). Conversely, in patients on the ADS, the sharp decreases in delta power with increasing A1C may reflect a more severe outcome, whereby the increased neural stress leads to more extensive synaptic damage and overall reduced cortical activity. A similar interaction involving poor sleep and delta power may also be playing out in patients on the ADS [50, 51]. In our study, the interaction effect was strongest in the left superior frontal gyrus and follow-up structural equation modeling showed that the strength of delta activity in this region mediated the relationship between glycemic control and attention performance in patients on the ADS. Such findings are broadly consistent with work showing that alterations in spontaneous neural activity are intricately linked to cognitive deficits, including studies in AD, MCI, and HIV-related cognitive impairment [8, 10, 39, 48, 58, 59, 73, 74]. Regarding mechanisms, some have suggested that elevated levels of neuroinflammatory activity may lead to synaptic damage and altered spontaneous activity [15, 41, 56, 57, 75], although most of this work has focused on neuroHIV and further studies examining AD/MCI are needed.

Another critical finding was the glycemic control by group interaction observed in the alpha range in the inferior frontal and temporal cortices. This effect was strongest in the left frontotemporal cortex and posterior middle temporal region, although it was significant across an extended cortical region. Unlike the interaction observed for delta, which was largely driven by the decrease in power with increasing A1C in the ADS group, the alpha interaction primarily reflected sharp decreases in alpha power with increasing A1C (i.e., reduced glycemic control) in the control group, with the ADS group showing a limited trend in the opposite direction. While many studies focusing on ADS conditions have reported altered alpha activity (e.g., [66, 68], a notable feature of this finding is that alpha aberrations have also been widely reported in the diabetes literature [19, 20] and, in some cases, have been directly linked to glycemic control levels [18, 21]. Further, several studies have found aberrant alpha responses in the context of adults with hypertension, which in some cases were more aberrant in those with uncontrolled versus controlled hypertension [4, 55]. Whether this overall pattern of findings reflects that alpha activity is affected by multiple risk factors for AD (i.e., diabetes, glycemic variability, and hypertension) or a common pathway toward cognitive dysfunction remains unclear and should be a focus of future work. Finally, regarding the opposite pattern observed in the ADS group, it is important to note that the simple slope was not significant and thus caution is warranted in interpreting these data in isolation, particularly since alpha was not related to cognitive functioning in our ADS group.

Before closing, it is important to acknowledge some limitations of this study. First, we focused on resting-state cortical activity, which has the advantage of being immune to task performance differences but the weakness of less experimental control. While task-related neuroimaging studies in patients with AD are often difficult due to impaired performance, they are certainly possible [9], and future work should attempt to link glycemic control to task-based neurophysiological activity. That said, studies have shown that MEG resting-state measures of both delta and alpha activity are highly reliable over a three-year period in healthy aging adults [38], which may suggest that these findings remain robust despite the limited level of experimental control. Second, the level of variability in glycemic control was only modest in our sample and future work should aim for a broader range of glycemic control in their cohort. We expect that the effects reported here would be more robust in such a sample. Third, we included participants who endorsed depression and/or anxiety in our study, with the rationale that allowing these conditions makes the sample more representative of the population under study. While only a few participants in each group endorsed such conditions, these conditions were more frequent in our patient group, which is not surprising since depression is common in those on the ADS. Further, more patients in the ADS group were receiving medications for their depression and/or anxiety than controls, although caution is warranted as SSRIs are often prescribed to patients on the ADS to manage behavioral conditions other than depression. While we conducted supplementary analyses to ensure our results were not affected by severe depression, we cannot totally guarantee that medication and/or depression symptoms did not affect our results and future studies should examine larger samples and/or consider excluding those with depression and/or related conditions to better isolate the effects of interest. Fourth, our study collapsed across the differing levels of dementia severity and focused on the general effects of ADS conditions. We did not conduct exploratory analyses dividing the ADS group into MCI and AD subgroups because the vast majority of our patients clustered around the QDRS cutoff value for MCI versus mild to moderate dementia, and comparing groups with almost the same QDRS values could yield misleading results. Future studies of ADS conditions using larger samples should consider dividing their sample by severity and conducting exploratory analyses to assess whether relationships with A1C change as the disease progresses. Finally, while our participants in the ADS group were all biomarker-positive, our controls did not undergo biomarker testing and thus we cannot confirm they were negative. Nonetheless, all controls underwent the MMSE and a comprehensive battery of neuropsychological assessments and scored in the normal range on both. Thus, while we would expect biomarker negativity in all of our controls, confirmation would have strengthened the study and future work should aim for this.

In conclusion, we found that spontaneous delta and alpha activity across widespread cortical regions that included key prefrontal and parietal areas were differentially related to glycemic control in patients on the ADS compared with controls, with delta levels in the left superior frontal cortices mediating the relationship between A1C and attention performance in those on the ADS. Whether these findings reflect different effects due to disease-related neurological changes (e.g., cortical atrophy) or that key risk factors for AD have a different impact in those who develop the condition versus those who do not remains to be determined and should be a focus of future work.