Multispectral Brain Connectivity During Visual Attention Distinguishes Controlled from Uncontrolled Hypertension

Abstract

Hypertension-related changes in brain function place individuals at higher risk for cognitive impairment and Alzheimer’s disease. The existing functional neuroimaging literature has identified important neural and behavioral differences between normotensive and hypertensive individuals. However, previously-used methods (i.e., magnetic resonance imaging, functional near-infrared spectroscopy) rely on neurovascular coupling, which is a useful but indirect measure of neuronal activity. Furthermore, most studies fail to distinguish between controlled and uncontrolled hypertensive individuals, who exhibit significant behavioral and clinical differences. To partially remedy this gap in the literature, we used magnetoencephalography (MEG) to directly examine neuronal activity that is invariant to neurovascular coupling changes induced by hypertension. Our study included 52 participants (19 healthy controls, 15 controlled hypertensives, 18 uncontrolled hypertensives), who completed a modified Flanker attention task during MEG. We identified significant oscillatory neural responses in two frequencies (alpha: 8–14 Hz, gamma: 48–60 Hz) for imaging and used grand-averaged images to determine seeds for whole-brain connectivity analysis. We then conducted Fisher-Z tests for each pair of groups, using the relationship between the neural connectivity and behavioral attention effects. This highlighted a distributed network of regions associated with cognitive control and selective attention, including frontal-occipital and interhemispheric occipital connections. Importantly, the inferior frontal cortex exhibited a unique neuro-behavioral relationship that distinguished the uncontrolled hypertensive group from the controlled hypertensive and normotensive groups. This is the first investigation of hypertension using MEG and identifies critical whole-brain connectivity differences based on hypertension profiles.

Graphical Abstract

The relationship between magnetoencephalography-based connectivity and behavioral (reaction time) measures collected during a flanker task were examined in healthy controls, adults with controlled hypertension, and those with uncontrolled hypertension. Spectrally-specific neuro-behavioral profiles of the dorsolateral prefrontal cortex distinguished the uncontrolled hypertension group from healthy control and controlled hypertensive groups.

Introduction

Hypertension affects approximately 1.4 billion people worldwide and is one of the leading contributors to all-cause mortality, with 7.5 million annual deaths (World Health Organization 2022; Mills, Stefanescu, and He 2020). Current recommendations for hypertension management include a combination of medications (e.g., diuretics, ACE inhibitors) and lifestyle adjustment, including diet and sleep modifications (Al-Makki et al. 2022; Mancia et al. 2013). Though treatment can be effective and reduces mortality, there remains a large portion of individuals with uncontrolled hypertension (UHT) where blood pressure reduction is not achieved either due to a lack of patient compliance and/or knowledge, or despite adherence to three or more antihypertensive drugs (Sarafidis Pantelis A. and Bakris George L. 2008; Egan et al. 2011; Canoy et al. 2022; Brunström and Carlberg 2018; Noubiap et al. 2019; Morgado et al. 2010; Messerli, Williams, and Ritz 2007). There are important differences in clinical outcomes between individuals with controlled hypertension (CHT) and those with UHT, as the risk of cardiovascular disease (CVD) and other vascular complications scales with increasing blood pressure(Whelton et al. 2020). Importantly, hypertension-related changes such as inflammation and oxidative stress promote cerebral small vessel disease (CSVD), leading to structural and functional changes in the brain (e.g., hemorrhages, decreased neurovascular coupling) and concomitant cognitive impairment (Liu et al. 2018; Jiménez-Balado et al. 2019; Pires et al. 2013; Girouard and Iadecola 2006). Specifically, Muela et al. report that Mini-Mental State Examination and Montreal Cognitive Assessment scores in adults scaled with hypertension severity in multiple cognitive domains and hypertension is associated with an increased risk of dementia and Alzheimer’s disease (Muela et al. 2017; Iadecola et al. 2016).

Previous literature using magnetic resonance imaging (MRI) has focused on white matter hyperintensities and cerebral blood flow to probe neurovascular and hemodynamic changes in individuals with hypertension (Zhang et al. 2022; Carnevale et al. 2020; Jiménez-Balado et al. 2019). These hypertension-associated changes have been shown to induce vascular cognitive impairment (VCI), including mild cognitive impairment and full dementia and impact various domains of cognition including visual processing and selective attention (Madden and Blumenthal 1998; Iadecola et al. 2016; O’Brien et al. 2003). However, most of these studies have focused on structural or hemodynamic changes (i.e., BOLD) and do not examine the potential impact of small vessel damage on the functional human connectome, despite the significance of network integration in healthy brain function (Craddock, Tungaraza, and Milham 2015; Rogers et al. 2007; Yeo et al. 2011; Sepulcre et al. 2010). Since these tools do not directly measure the activity of neuronal populations, the few notable MRI or functional near-infrared spectroscopy (fNIRS) studies that have examined connectivity have had only a moderate impact due to our limited understanding of neurovascular coupling (i.e., linear, nonlinear) (Mackert et al. 2008; Bu et al. 2018; Carnevale et al. 2020; X. Li et al. 2015; Feng et al. 2020). While studies using electroencephalography (EEG) have directly measured neural differences in hypertension, EEG literature is sparse and volume conduction limits the interpretation of the neural generators of the signal observed at the level of the scalp (Montagnese et al. 2013; Roberts and Walker 1954). Furthermore, these neuroimaging studies have focused primarily on comparing healthy controls and hypertensive individuals, which are often categorized as a homogeneous group (i.e., not divided in CHT and UHT). This precludes the identification of more subtle, but potentially important neural differences that may exist among adults with CHT and UHT, given the known behavioral and clinical corollaries (Qian et al. 2016; Unmuessig et al. 2016; Jung et al. 2017; Muela et al. 2017; Iyer et al. 2010; Iadecola and Gottesman 2019).

To partially remedy this gap in the literature, we use magnetoencephalography (MEG) to examine selective attention differences between healthy control (HC) participants, individuals with CHT, and individuals with UHT. MEG is a growing neuroimaging technique with excellent temporal (~1 millisecond) and spatial (~3–5 millimeter) resolution that noninvasively quantifies the magnetic fields emanating from post-dendritic potentials of neural populations that contribute to neural oscillatory activity (Baillet 2017; Wilson et al. 2016). Importantly for studies of hypertension, the measured signals are immune to neurovascular coupling changes and provide a direct measure of neural activity. These oscillations underlie a wide range of cognitive functions and have previously been used in a wide range of clinical populations, including human immunodeficiency virus (HIV), substance use, diabetes, and Alzheimer’s (Schantell et al. 2022; Arif et al. 2020; Springer et al. 2021; Lew et al. 2018; Lew, Salimian, and Wilson 2021; Embury et al. 2018). Specifically, alpha and gamma oscillatory activity and connectivity are critical elements underlying successful selective attention processes, with previous literature showing sensitivity of these neural responses to developmental, aging, and pathophysiological changes (Taylor et al. 2021; Lew et al. 2018; Arif et al. 2023; Wiesman and Wilson 2020; Son et al. 2023; McDermott et al. 2017; Proskovec et al. 2018). As such, we use MEG to investigate long-range differences in spectrally-specific (e.g., alpha, gamma) functional connectivity and their relationships to behavior between HC, CHT, and UHT individuals using a modified Flanker task designed to probe selective attention (George Bush and Shin 2006; G. Bush et al. 2003; Wiesman et al. 2020; Wiesman and Wilson 2020; Eriksen and Eriksen 1974). To date, this is the first investigation of hypertension using MEG. Examining differences in behavior and neural activity across healthy and hypertensive groups will help better characterize the impact of hypertension on attention function.

Methods

Ethical Approval

This work conformed to the standards of the Declaration of Helsinki, except for registration in a database. The study protocol was reviewed and approved by the University of Nebraska Medical Center’s Institutional Review Board (ethics approval reference number: 20-30-XP). A full description of the study was provided to all participants followed by written informed consent, prior to administration of the study protocol.

Participants

A total of fifty-two adults (mean age: 51.72 years; 30 males) between the ages of 34 and 65 years-old were recruited, including 19 healthy controls (HC), 15 individuals with controlled hypertension (CHT), and 18 individuals with uncontrolled hypertension (UHT). Three blood pressure readings were taken at least 40 minutes apart and the average systolic blood pressure (SBP) and diastolic blood pressure (DBP) values were used to separate our sample into three groups. HC participants did not have formal diagnoses of hypertension and had blood pressure readings of < 140SBP and < 90DBP, CHT individuals had diagnoses of hypertension and blood pressure readings of < 140SBP or < 80DBP, and UHT individuals had diagnoses of hypertension and blood pressure > 140 SBP or DBP > 80. If the SBP and DBP readings were in different categories, the higher of the two categories was used. Exclusion criteria included individuals with medical illness affecting central nervous system function (except hypertension), history of significant head trauma, current substance abuse, a diagnosed neurological or psychiatric disorder, or metal implants (e.g., retainer, pacemaker) that would negatively affect MEG data acquisition or be an MRI safety concern. All participants were screened for eligibility through a six-question screening form administered by trained research assistants. In the CHT group, participants had been diagnosed with hypertension for an average of 11.94 (SD = 11.52) years, while UHT participants had been diagnosed with hypertension for an average of 12.67 (SD = 17.01) years. There was no significant difference in hypertension diagnosis duration (Mann-Whitney U Test, p = .719). Furthermore, cognitive function was determined using the fully adjusted cognition composite T-scores from all seven tests in the NIH Toolbox (population average = 50, SD = 10), with an average in this sample of 52.78 (SD = 9.44). No difference in T-scores were found by group (ANOVA, p = .645). Given that severe or long-lasting phenotypes of hypertension have wide-ranging physiological consequences, we note that none of the participants had any diagnosed ophthalmic conditions, two participants had prior heart attacks, and three participants had type II diabetes insipidus.

Experimental Paradigm and Stimuli

Each participant was seated in a magnetically shielded room and completed a modified number-based version of the Flanker task to probe cognitive interference(Eriksen and Eriksen 1974). Participants were shown a centrally-presented fixation cross for 2200 ± 200 ms, followed by three equally spaced, horizontally centered numeric stimuli (i.e., digits ranging from 0 to 3) for 1500 ms. For each trial, two of these numbers were identical (task irrelevant) and the third was different (task relevant). Participants were asked to indicate the “odd-number out” by pressing the button corresponding to its numerical identity (i.e., index = 1, middle = 2, ring = 3), with the congruent condition reflecting no interference (i.e., 1 0 0/0 2 0/0 0 3) and the incongruent condition reflecting cognitive interference (e.g., 1 2 2/1 2 1/3 2 3/2 2 3; Figure 1). Trial types and responses were pseudorandomized, such that no condition or response was repeated more than twice in a row. Each participant completed 100 trials of each condition, for a total of 200 trials. Custom visual stimuli were programmed in Matlab (Mathworks, Inc.) using Psychophysics Toolbox V3 (Brainard 1997) and back-projected onto a nonmagnetic screen.

Figure 1. Flanker task paradigm and reaction time differences by group.

Each trial consisted of a fixation period for a randomly varied interval of 2000–2400ms, followed by a row of three equally spaced integers between 0 and 3. Participants were instructed to indicate with their right hand the “odd-number-out” using a button pad, with the index, middle, and ring fingers corresponding to the integers 1, 2, and 3, respectively. Significant reaction time differences were found between the HC (Healthy Control), CHT (Controlled Hypertensive), and UHT (Uncontrolled Hypertensive) groups. ** = p < .005, * = p < .05.

Behavioral Data Analysis

For each participant, incorrect and no-response trials were removed prior to calculating the reaction time for each condition. Accuracy per condition was computed as the percentage of correct trials / total trials. On the single participant level, a standard single-trial reaction time threshold of three standard deviations was used to remove outlier trials. Mean reaction times were subsequently calculated for each participant and condition. Reaction time and accuracy data were then analyzed using a two-way analysis of covariance (ANCOVA) to identify conditional effects. All behavioral analyses were implemented in SPSS v.25.

MEG Data Acquisition

All MEG data were collected using a 306-sensor VectorView MEG system (Helsinki, Finland) sampled at 1kHz with an acquisition bandwidth of .1–330 Hz. The system included 204 planar gradiometers and 102 magnetometers, but we focused on the planar gradiometers in this investigation. During data collection, study participants were monitored in real-time via audio-visual stream from within the magnetically shielded room. Raw MEG data were noise corrected for head motion and extraneous noise using the signal space separation method with a temporal extension (Taulu and Simola 2006).

Structural MRI Processing and MEG Coregistration

Prior to MEG acquisition, four head position indicators (HPIs) were attached the participant’s head and localized in 3-D space with three fiducial points and the scalp surface (per participant) using a 3-D digitizer (Fastrak 3SF0002, Polhemus Navigator Sciences, Colchester, VT, USA). During data acquisition, a current with a known, unique frequency label (e.g., 322 Hz) was fed to each of the four HPIs, allowing us to measure and track each of the coils in space using the MEG array. This allowed the location of each HPI in reference to the sensors to be known throughout the recording session, which enabled us to correct for motion and transform the MEG measurements into a common coordinate system. With this coordinate system, each participant’s MEG data were coregistered with individual high-resolution T1-weighted structural MRI data, prior to source space analysis. For each participant, MEG data were aligned parallel to the anterior and posterior commissures for co-registration, then transformed into standardized space using BESA MRI (Version 2.0) prior to source-space analysis.

MEG Preprocessing and Time-Frequency Transformation

Cardiac and ocular artifacts were identified in the raw MEG recordings and removed using signal-space projection (SSP), a method that assumes that magnetic field distributions generated by brain sources have spatial distributions sufficiently different from those generated by non-brain sources. Removal of these artifacts was accounted for during source reconstruction (Uusitalo and Ilmoniemi 1997). In addition, line noise (60Hz) was removed from the data. The continuous MEG time series was epoched into 1500 ms segments, extending from 500 ms prior to visual stimulus presentation to 1000 ms after the stimulus. The baseline period was defined as the −500 to 0 ms window, while the active period ranged from 0 to 1000 ms. Epochs containing remaining artifacts were rejected per participant using a fixed threshold method, supplemented with visual inspection. A two-way ANCOVA was conducted to compare the number of epochs available for subsequent analyses across the three groups.

Artifact-free epochs were transformed into the time-frequency domain using complex demodulation (Hoechstetter et al. 2004; Kovach and Gander 2016). The resulting spectral power estimations per sensor were averaged across trials to generate mean spectral density plots. The data were then normalized per time frequency bin using the respective bin’s baseline power, which was calculated as the mean power during the −500 to 0 ms time period. These normalized spectral power plots were then examined statistically to determine windows of interest for subsequent source analysis.

MEG Sensor-Level Statistics

Each data point in the spectrogram was initially evaluated using a mass univariate approach based on the general linear model. To reduce the risk of false positive results while maintaining reasonable sensitivity, a two-stage procedure was followed to control for Type-1 error. In the first stage, paired-sample t-tests against baseline were conducted on each data point, and the output spectrogram of t-values was thresholded at p < .05 to define time-frequency bins containing potentially significant oscillatory deviations across all participants. In stage two, time-frequency bins that survived the p < .05 threshold were clustered with the temporally and/or spectrally neighboring bins that were also above the threshold (p < .05), and a cluster value was derived by summing the t-values of all data points in the cluster. Nonparametric cluster-based permutation testing was then conducted using a Monte Carlo approach to randomly sample participants and reassign their active versus baseline data before recomputing the cluster sum values, which were eventually used to build a null distribution based on 10,000 permutations. The significance level of the observed clusters (from stage one) was then tested directly using this distribution (Ernst 2004; Maris and Oostenveld 2007). Based on these analyses, the time-frequency windows that contained statistically significant oscillatory events across all participants and conditions were subjected to a beamforming analysis.

MEG Source Imaging and Statistics

Neural responses were imaged using the dynamic imaging of coherent sources (DICS) beamformer, which applies spatial filters in the time-frequency domain to calculate source power per voxel for the entire volume (Gross et al. 2001). Each task condition was imaged separately per participant for the statistically defined time-frequency bins (see Results). The single images were derived from the cross-spectral densities of all combinations of MEG gradiometers averaged over the time-frequency range of interest, and the solution of the forward problem for each location on a grid specified by input voxel space. Following convention, we computed noise-normalized source power for each voxel per participant using active (i.e., task) and passive (i.e., baseline) periods of equal duration and bandwidth at a resolution of 4 mm isotropic (Hillebrand et al. 2005). Such images are typically referred to as pseudo-t maps, with units (pseudo-t) that reflect noise-normalized power differences (i.e., active versus passive) per voxel. After imaging the neural responses, grand-average maps were computed by averaging each time-frequency specific neural response map across conditions and participants. MEG pre-processing and imaging used the BESA Research (version 7.0) software.

Functional Connectivity

The peak voxels from the grand averaged beamformer images were used as seed voxels for the calculation of a coherence beamformer using the DICS approach (Gross et al. 2001). The congruent and incongruent coherence maps were computed separately, and whole-brain interference maps were computed by subtracting the congruent from the incongruent coherence maps. These maps were then corrected for age and response power at the individual level to help dissociate the potential group-wise differences that may be due to differences in age or response power, and not functional connectivity. Following this adjustment, the resulting maps were subjected to whole-brain voxel-wise correlations with behavioral interference effects (i.e., incongruent RT - congruent RT) for each of the HC, CHT, and UHT groups separately. We then compared the resulting whole brain neuro-behavioral statistical maps among groups by using a Fisher Z transformation. In sum, we conducted correlational analyses separately for each group (factor with three levels: HC, CHT, UHT), with age, seed voxel power, and target voxel power as continuous covariates of no interest, followed by Fisher r-to-z transformations to compare correlation coefficients across independent samples (i.e., groups). To account for multiple comparisons, a significance threshold of p < .005 and cluster threshold of at least 5 contiguous voxels were used. We did not analyze peaks within 4 cm of the seed region, to account for volume conduction effects in neurophysiological analyses(Brookes et al. 2011). Furthermore, peaks identified within motor regions were not further analyzed, as the cognitive control aspects of task completion were the focus of this investigation. All whole-brain statistical analyses were conducted with customized scripts in MATLAB and additional statistical analyses were conducted in IBM SPSS v.25.

Results

Behavioral Results and Demographics

Following all exclusions and MEG data quality assurance, 52 adults remained in the sample, including 19 HC individuals (mean age: 47.32 years; age range: 34–62 years; mean SBP/DBP: 123/78; min-max SBP/DBP: 97–136 / 60–90), 15 CHT participants (mean age: 52.73 years; age range: 39–65 years; mean SBP/DBP: 130/81; min-max SBP/DBP: 115–139 / 69–88), and 18 UHT individuals (mean age: 55.06 years, age range: 36–63 years; mean SBP/DBP: 153/95; min-max SBP/DBP: 136–184 / 83–119). Furthermore, we tested for group differences in demographic variables (i.e., age, sex, BMI) that may have contributed to neuro-behavioral changes. There was a significant difference in age between the CHT and UHT groups (p = .002) and thus we included age as a covariate of no interest in our analyses. No significant group differences were found for sex (p = .406) or BMI (mean HC = 28.03, mean CHT = 31.31, mean UHT = 29.44; p = .276). Overall, participants performed well on the task with mean accuracies of 92.02% (SD = 6.82%) and 91.77% (SD=6.71%), and mean response times of 680.33 ms (SD = 97.87 ms) and 804.44 ms (SD = 111.33 ms) in the congruent and incongruent conditions, respectively. A two-way analysis of covariance (ANCOVA) revealed no significant differences in accuracy by condition or group.

A separate two-way ANCOVA of reaction time controlling for age revealed significant group-by-condition interaction effects (F(2, 48) = 7.182, p = .002), such that the reaction time interference effects (i.e., incongruent - congruent) were larger in both the CHT (p = .002) and UHT (p = .000) groups, relative to the healthy control group (Figure 1). Furthermore, the incongruent response time in the UHT group was greater than that of the HC group (p = .012), but was not significantly different from that of the CHT group. In addition, there was a main effect of condition (F(1, 48) = 9.010, p = .004), such that the reaction time in the incongruent condition was greater than that of the congruent condition, across all groups, controlling for age (Figure 1B).

MEG Sensor-Level Results

Sensor-level spectrograms were statistically examined using non-parametric permutation testing to derive the precise time-frequency bins for follow up beamforming analyses (all p’s < .05, corrected). These analyses indicated significant clusters of sustained decreases (i.e., desynchronization) in alpha (8–14 Hz) activity that began 250 ms after stimulus onset and continued until 650 ms (p < 0.05, corrected), as well as beta (16–22 Hz) activity during the same temporal window. There were also significant clusters of sustained gamma (48–60 Hz) increases with an early response from 150–300 ms and a late response from 400–600 ms after stimulus onset (p < 0.05). To determine the brain regions generating these oscillatory responses, we imaged these significant time-frequency bins and a baseline period of equal bandwidth and duration using a beamformer, and then grand-averaged the resulting maps across both conditions and all participants.

MEG Beamforming and Source-Level Results

The grand-averaged maps across both conditions showed alpha oscillations in the bilateral occipital cortices, as well as gamma oscillations in the occipital cortices for both the early and late responses (Figure 2). Since the gamma responses were highly similar, the early gamma window was used for subsequent analyses. For beta activity, oscillations in the contralateral motor cortex were identified, but were not further examined as somato-motor processing was not the aim of this study.

Figure 2. Grand-averaged time-frequency spectrograms during Flanker task performance for the early gamma (48–60 Hz, 150–300ms), late gamma (48–60 Hz, 400–600ms), beta (16–22 Hz, 250–650ms), and alpha (8–14 Hz, 250–650ms) responses, with the corresponding grand-averaged beamformer images to the right.

Left) All signal power is represented as a percent difference from the baseline period, with the color bar on the right. Time is denoted in ms on the x-axis, and frequency is denoted in Hz on the y-axis. Significant time-frequency windows used for beamforming are shown with dashed white boxes. Each spectrogram represents group-averaged data from one gradiometer sensor that was representative of the neural responses seen in multiple sensors near the peak response. Right) Grand-averaged beamformer images for each time-frequency bin, with responses in pseudo-t values calculated as statistically significant differences from baseline. Bottom) The distribution of activity across sensors for each time–frequency window is shown, as well as the representative sensor (circled in black) used to produce the spectrograms.

Alpha Differences in Neuro-Behavioral Relationships

The peak alpha voxels in the bilateral occipital cortices were used as the seeds for whole-brain voxel-wise cortico-cortical coherence analysis. Results utilizing the peak from the left occipital cortex indicated a significant difference between the HC and CHT groups in the right lateral occipital cortex (z = 3.575, p < .001), such that in the HC group stronger flanker interference effects on occipital connectivity were associated with smaller behavioral interference effects (r = −.607, p = .013), while in the CHT group greater flanker interference differences in connectivity were associated with slower reaction times (r = .651, p = .012). Furthermore, the HC and CHT groups differed in the left midcingulate cortex (z = 3.874, p < .001), such that stronger flanker interference connectivity effects in this region were correlated with faster reaction times (r = −.498, p = .050), while no significant relationship was found in CHT (r = .060, p = .839; Figure 3).

Figure 3. Scatterplots and regression lines showing the relationship between the flanker reaction time interference effect and connectivity interference effect (i.e., Incongruent – Congruent) in the alpha frequency.

Connectivity profiles using either the left occipital seed (left) or the right occipital seed (right), with corresponding connectivity interference effects by group. The x-axis indicates the percent difference in connectivity (i.e., incongruent – congruent) and the y-axis indicates the difference in reaction time (ms).

Unsurprisingly, results utilizing the peak from the right occipital cortex indicated the same significant differences between the HC and CHT groups in the left occipital cortex (z = 4.463, p < .001) and the left midcingulate cortex (z = 4.126, p <.001). In both peaks, HC with a stronger flanker interference connectivity effect had smaller behavioral interference effects (left occipital: r = −.518, p = .040; left midcingulate: r = −.762, p = .001). In the CHT group greater flanker interference differences in connectivity were associated with slower reaction times in the left occipital cortex (r = .750, p = .002), while no significant relationship was found in the left midcingulate (r = .401, p = .155). Furthermore, there was a significant difference between the HC and UHT groups in the left inferotemporal cortex (z = 3.586, p < .001), such that in the HC group stronger flanker interference connectivity effects were associated with larger behavioral interference effects (r = .791, p <.001), while no significant relationship was found in UHT (r = −.316, p = .251). Finally, there was a significant difference between the CHT and UHT groups in the superior temporal region (z = 3.572, p < .001), such that in the CHT group stronger flanker interference connectivity effects were associated with larger behavioral interference effects (r = .531, p = .050), while in the UHT group stronger flanker interference connectivity effects were associated with smaller behavioral interference effects (r = −.531, p = .042). There was also a significant difference between the CHT and UHT groups in the primary motor area (z = 3.541, p < .001), such that in the UHT group stronger interference connectivity effects were associated with smaller behavioral interference effects (r = −.497, p = .050), while no significant relationship was found in CHT (r = .301, p = .275). However, caution is warranted with these latter findings as analyses time-locked to movement onset are more ideal for studying motor-related responses and connectivity.

Gamma Differences in Neuro-Behavioral Relationships

The peak gamma voxels were also used for whole-brain voxel-wise cortico-cortical coherence analysis. Results using the peak from the left occipital cortex indicated a difference between HC and CHT groups in the right middle frontal gyrus (z = 4.019, p < .001), such that in the HC group stronger flanker interference connectivity effects were associated with smaller behavioral interference effects (r = −.614, p = .007), while in the CHT group greater flanker interference differences in connectivity were associated with slower reaction times (r = .726, p = .005). In addition, the HC and UHT groups differed in the left inferior frontal gyrus (z = 4.665, p < .001), such that stronger flanker interference connectivity effects in the UHT group were associated with smaller behavioral interference effects (r = −.783, p < .001), while HC showed the opposite effect (r = .527, p = .024). Furthermore, differences between the CHT and UHT groups were identified in the same region of the left inferior frontal gyrus (z = 4.224, p < .001), such that those in the CHT group exhibited a relationship similar to HC where stronger flanker interference connectivity effects were associated with larger behavioral interference effects (r = .564, p = .045). Finally, the CHT and UHT groups differed in the right middle occipital cortex (z = 4.224, p < .001), such that greater flanker interference differences in connectivity were associated with larger flanker reaction time costs (r=.748, p=.003) in the CHT group, while those in the UHT showed no relationship (r = −.356, p = .160; Figure 4).

Figure 4. Scatterplots and regression lines showing the relationship between reaction time difference and connectivity difference (i.e., Incongruent – Congruent) in the gamma frequency.

Connectivity profiles using the left occipital seed are shown in the left column, while connectivity profiles using the right occipital seed are shown in the right column. The x-axis indicates the percent difference in connectivity (i.e., incongruent – congruent) and the y-axis indicates the difference in reaction time (ms).

Results utilizing the peak from the right occipital cortex indicated very similar effects in the right middle frontal gyrus (z = 4.164, p < .001), such that stronger flanker interference connectivity effects were associated with larger behavioral interference effects in the CHT group (r = .803, p = .001), while the opposite was true in UHT individuals (r = −.559, p = .020). Furthermore, there was a significant difference in the left inferior frontal gyrus between the HC and UHT groups (z = 4.789, p < .001), such that in the UHT group stronger flanker interference connectivity effects were associated with weaker behavioral interference effects (r = −.777, p < .001). In the CHT group, stronger flanker interference connectivity effects were associated with stronger behavioral interference effects (r = .748, p = .003), while in the HC group there was no significant relationship (r = .480, p = .060). In addition, there were significant differences between UHT and CHT in the right superior temporal gyrus (z = 3.755, p < .001), such that stronger flanker interference connectivity effects were associated with smaller behavioral interference effects (r = −.761, p = .003) in the CHT group, while no significant relationship was found in the UHT group (r = .440, p = .077).

Discussion

In this study, we used MEG to characterize behavioral and neural differences between HC, CHT, and UHT individuals during performance of a modified Flanker task. Our main findings were group differences in how alpha and gamma band connectivity were related to behavior in a distributed network of regions throughout the brain that are known to be critical for attention and cognitive control, including the occipital, temporal, and prefrontal cortices (Hong et al. 2017; Hahn, Ross, and Stein 2006; Miller 2000; Forstmann, van den Wildenberg, and Ridderinkhof 2008; Ramezanpour and Fallah 2022; Sani et al. 2021). Our novel findings contribute to the existing neuroimaging literature as the first MEG study to probe and identify group-wise neuro-behavioral differences between healthy control individuals and individuals with hypertension.

We found significant behavioral differences between our groups, such that the reaction time interference effect was more pronounced in both the controlled (p = .002) and uncontrolled (p < .001) hypertensive groups relative to the healthy control group. In other words, those with hypertension are negatively impacted to a greater degree by distracting information. Notably, there was no significant difference in the reaction time interference effect between the hypertensive groups. This suggests that the cognitive control mechanisms underlying successful completion of the Flanker task were negatively impacted in both hypertensive groups relative to healthy controls, which aligns with the existing literature (A. P. Shapiro et al. 1982; de Menezes et al. 2021; Iadecola et al. 2016; Lopez et al. 2003).

To examine how neural recruitment underlying cognitive control relates to behavior, we then conducted a whole-brain correlation between the connectivity interference maps (i.e., incongruent – congruent) and reaction time interference. Overall, we observed group-wise multispectral connectivity differences in both basic visual processing regions (i.e., interhemispheric occipital connections) and higher-order regions, including the temporal and frontal cortices (both intra- and interhemispheric). In the alpha band, the HC and UHT groups differed in the inferior temporal cortex, while the superior temporal cortex showed a dissociation between CHT and UHT groups. These findings suggest that there are changes in neural processing of visual attentional information in the temporal cortex that are unique to the UHT group and that may distinguish them from both HC and CHT individuals. Of note, these regions of the temporal cortex are strongly implicated in attentional control, memory, and overall cognitive function, and have previously been found to undergo structural and functional changes (i.e., cortical thickness, functional connectivity) in hypertensive patients (Sani et al. 2021; Ramezanpour and Fallah 2022; K. Shapiro, Hillstrom, and Husain 2002; W. Li, Yue, and Xiao 2021; Feng et al. 2020).

In the gamma band, we observed a dissociation in the inferior frontal gyrus such that the relationships between neural connectivity and behavioral differences were similar between HC and CHT groups, while the relationship in the UHT group was in the opposite direction. The prefrontal cortex is commonly associated with the top-down control of selective attention and is part of several strongly connected networks (e.g., dorsal attention) that are engaged during a variety of cognitive control tasks (Prado, Carp, and Weissman 2011; Peterson et al. 2002; Badre and Wagner 2004; Goghari and MacDonald 2009; Miller 2000; Killanin et al. 2020; McDermott et al. 2017; Heinrichs-Graham and Wilson 2015). Given that these differences were found in the exact same voxel (i.e., region) of the brain, this neural dissociation by group indicates that uncontrolled hypertension has a unique impact on neural processing of visual information in this region that directly affects behavior. In addition, the inferior frontal cortex has previously been shown to be modulated by hypertension, indicating that findings using hemodynamic methods (Raz et al. 2005; Feng et al. 2020) likely reflect neural changes and are not an artifact of alterations in neurovascular coupling. Taken together with the behavioral data, this suggests that while behavioral differences may not be apparent between the two hypertensive groups, there are specific neural signatures that differ and these may lead to behavioral differences in more severe cases of uncontrolled hypertension, but future studies are needed to directly probe this possibility.

Before closing, it is important to note some limitations of this study. We focused our investigation on the spatiotemporal dynamics underlying performance in those with hypertension, though we divided our sample into groups and did not assess hypertension as a continuous variable. Such an approach could lead to new insight and future studies should consider hypertension as a continuous variable. In addition, information regarding antihypertensive medication timing (e.g., morning or evening dose) and adherence were not available for analyses. Future analyses should incorporate these measures to better disambiguate the impact of antihypertensive medications and hypertension status. Finally, the neural mechanisms required for the successful planning and execution of the button press should also be explored in future analyses using imaging windows determined from time-frequency analyses that are time-locked to movement onset (and not stimulus onset, as in the current study). These analyses will also allow for the exploration of potential visual-motor changes that are present in the context of hypertension and cognitive interference. Overall, our study utilized the excellent temporal and spatial precision of MEG to identify multispectral connectivity differences during performance of an established cognitive control task between healthy controls and patients with controlled and uncontrolled hypertension. Connectivity and reaction time relationships in the gamma band in the inferior frontal gyrus uniquely distinguished uncontrolled hypertensive individuals from HC and CHT. Our findings support the importance of studying controlled and uncontrolled hypertensive individuals separately to index the impact of unmanaged hypertension on brain and cognitive function.

Key Points.

• Structural and functional changes in brain circuitry scale with hypertension severity and increase the risk of cognitive impairment and Alzheimer’s disease.

• We harness the excellent spatiotemporal precision of magnetoencephalography (MEG) to directly quantify dynamic functional connectivity in healthy control, controlled hypertensive, and uncontrolled hypertensive groups during a flanker task.

• In the first MEG study of hypertension, we show that there are neuro-behavioral relationships that distinguish the uncontrolled hypertensive group from healthy and controlled hypertensive group in the prefrontal cortex.

• These results provide novel insights into the differential impact of hypertension on brain dynamics underlying selective attention.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author, Dr. Tony W. Wilson, upon reasonable request.