Decreased Cerebrovascular Reactivity in Mild Cognitive Impairment Phenotypes

Abstract

Background:

Cerebrovascular health plays an important role in cognitive health in older adults. Cerebrovascular reactivity (CVR), a measure of cerebrovascular health, changes in both normal and pathological aging, and is increasingly being conceptualized as contributory to cognitive decline. Interrogation of this process will yield new insights into cerebrovascular correlates of cognition and neurodegeneration.

Objective:

The current study examines CVR using advanced MRI in prodromal dementia states (amnestic and non-amnestic mild cognitive impairment phenotypes; aMCI and naMCI, respectively) and older adult controls.

Methods:

CVR was assessed in 41 subjects (20 controls, 11 aMCI, 10 naMCI) using multiband multi-echo breath-holding task functional magnetic resonance imaging. Imaging data were preprocessed and analyzed using AFNI. All participants also completed a battery of neuropsychological tests. T-tests and ANOVA/ANCOVA analyses were conducted to compare controls to MCI groups on CVR and cognitive metrics. Partial correlation analyses between CVR derived from regions-of-interest (ROIs) and different cognitive functions were conducted.

Results:

CVR was found to be significantly lower in aMCI and naMCI patients compared to controls. naMCI showed intermediate patterns between aMCI and controls (though aMCI and naMCI groups did not significantly differ). CVR of ROIs were positively correlated with neuropsychological measures of processing speed, executive functioning, and memory.

Conclusion:

The findings highlight regional CVR differences in MCI phenotypes compared to controls, where aMCI may have lower CVR than naMCI. Our results suggest possible cerebrovascular abnormalities associated with MCI phenotypes.

Keywords

Alzheimer’s disease cerebrovascular disease cerebrovascular reactivity dementia magnetic resonance imaging mild cognitive impairment

INTRODUCTION

Surmounting evidence connects cerebrovascular disease to the pathogenesis and cognitive symptomatology of Alzheimer’s disease (AD) and other neurodegenerative conditions [1 –3]. Declines in cerebral blood flow (CBF) are associated with age-related cognitive declines [1 –4], are observed at the earliest stages of pathological aging [3, 5], and are thought to play a role in neurodegenerative mechanisms (like those involved in AD) as well [6]. Declines in resting CBF, as well as declines in CBF response to neural activity (i.e., neurovascular coupling); however, both depend on the integrity of the underlying vasculature system [7]. Metrics that can more directly estimate cerebrovascular function are integral to understanding cognitive decline in normal and pathological aging. Cerebrovascular reactivity (CVR), defined as the CBF response to changes in arterial CO2, is a more direct estimate of the integrity of the underlying vascular system that has been linked to cognitive functions, normal aging, and pathological aging [8]. While early studies relied on gas inhalation to quantify CVR, recent work has demonstrated CVR can be measured successfully and reliably using a breath-holding (BH) task [9 –11].

Briefly, lower CVR has been reported across multiple different neurological samples (relative to control groups) [12] and declines in CVR have been associated with worse performance on cognitive measures [13, 14]. In healthy younger adult samples, findings regarding the relationship between CVR and cognitive functions are mixed [12, 13], suggesting CVR may emerge as a marker of poor cognitive functioning in pathological conditions. CVR differences have been observed in the early stages of pathological aging [7, 15], and CVR has been linked to risk factors associated with subsequent cognitive decline [3]. One study showed that young adults at risk for AD (e.g., homozygous for APOE ɛ4 allele) demonstrated reduced CVR compared to a healthy control group [16]. In AD, declines in CVR are observed in the early stages of AD and occur independent of amyloid-β and tau build-up (though also appear to be subsequently impacted by the presence of amyloid-β) [17]. Further, CVR is lower among those with mild cognitive impairment (MCI) compared to controls [18, 19]. Sur et al. (2020) found that CVR was significantly correlated with performance on a cognitive screener after controlling for AD pathological biomarkers. Importantly, one study found that CVR pathological values (lower values) increased the risk of conversion from MCI to AD [19].

While CVR has shown significant promise as a marker of interest for studying cognitive changes in older adults, with several studies to date showing positive correlations between CVR and cognitive test values [12 , 20–22], the research findings are inconclusive regarding CVR prediction of cognitive impairment in older adult cohorts [12, 13]. Perhaps, in part, this is due to notable variability in the methodological approach employed to measure CVR across studies, and studies that have examined the relationship between CVR and cognitive functions are often limited to cognitive screeners (e.g., use of the Mini-Mental Status Examination) in MCI and dementia cohorts [12]. Further, CVR is not uniform across the brain. There is preliminary support to suggest that there may be regional differences in CVR that emerge over the lifespan and that differentially relate to risk factors for pathological aging and age-related cognitive changes [13 , 24]. For instance, while increasing age was associated with lower CVR in temporal regions (relative to other lobes) [13, 23] in cognitively normal adults, having a greater number of vascular risk factors was associated with lower CVR in regions of the default mode network (DMN) relative to other brain regions examined [24]. Memory and attention scores have been linked to temporal CVR across the lifespan, whereas hippocampal CVR was associated with memory scores only in older adults [13]. Using longitudinal changes in CVR as a predictor of future cognitive decline, Peng et al. found that temporal and frontal lobe CVR were predictive of later declines in processing speed; while temporal, occipital, and parietal lobe CVR values were linked to later declines in memory [23]. While lobar differences in CVR have been informative, a more fine-grained assessment of regional CVR may better inform the relationship between cognition and pathological aging. Further, while previous work has considered risk factors associated with neurodegenerative conditions at the MCI and dementia stages, to our knowledge no studies have examined CVR across MCI clinical phenotypes.

In the current study, we add to the limited literature on CVR in prodromal dementia (i.e., MCI). Further, this is the first known study to provide preliminary evidence for CVR differences in MCI phenotypes (amnestic and non-amnestic) compared to older adult controls. Lastly, we examine correlations between CVR and neuropsychological measures of memory, language, processing speed, and executive functioning.

MATERIALS AND METHODS

Participants

All subjects provided written informed consent prior to participation in this study, which was approved by the local institutional review board and conducted in accordance with the Declaration of Helsinki. In total, 60 human participants were imaged for this study. Fifteen subjects were excluded because either BH imaging was not collected, they did not adequately perform the task, or they had poor data quality (i.e., excessive motion). Due to significant age differences between controls and MCI subjects, we removed all 55-year-old participants (4 controls were 55 years old). This resulted in 41 subjects who underwent further analyses (24 female, 17 male; age range 59–87 years; race: 40 white, 1 black): 20 cognitively normal control subjects (16 female, 4 male) and 21 patients (8 female and 13 male) diagnosed with MCI (11 amnestic MCI and 10 naMCI subjects) by a clinical neuropsychologist as part of routine clinical examination at a Midwestern US medical center. Exclusion criteria for all participants included a history of moderate to severe traumatic brain injury, brain tumor, severe psychiatric illness (i.e., schizophrenia and bipolar disorders), symptomatic stroke, intellectual disability, dementia/major neurocognitive disorder, reported history of color blindness,>3 alcoholic drinks per day, and age younger than 55 years. Inclusion criteria for controls included normal performance on a brief cognitive battery and no functional concerns. Inclusion criteria for MCI included MCI diagnosis made by a clinical neuropsychologist as part of a comprehensive neuropsychological evaluation and a documented MCI diagnosis in medical records by a neuropsychologist or memory disorders neurologist within 6 months or less of participants’ functional magnetic resonance imaging (fMRI) study visit. Medical history of vascular health conditions is summarized in Table 1. Controls and MCI groups were similar with respect to number of people with and without vascular health problems.

Table 1

Vascular Health History for Controls and MCI Participants

	Controls			MCI
Health Condition	No	Yes	% Medicated for Condition	No	Yes	% Medicated for Condition
Hypertension	15	5	60%	14	7	100%
Diabetes	20	0	n/a	19	2	100%
High Cholesterol	13	7	71%	14	7	100%
Heart Problems	16	4	75%	16	5	40%

Study procedure

MCI participants were recruited during their clinical visits with a memory disorders neurologist or clinical neuropsychologist and patient partners present during these visits were approached to recruit for potential controls (n = 14). Community controls were also recruited using flyers (n = 6). Neuropsychological test data were acquired from a routine clinical visit with a clinical neuropsychologist for the MCI group and at a study visit for the controls. Both groups completed an MRI visit. Study sessions were scheduled within days of each other unless otherwise needed to accommodate participants’ schedules, but never were separated by more than sixmonths.

Magnetic resonance imaging

A full suite of noninvasive anatomical, blood flow, and fMRI scans was collected using a 3T scanner (Signa Premier, GE Healthcare, Waukesha, WI) with a body transmit coil and a 32-channel NOVA (Nova Medical, Wilmington, MA) receive head coil. A 3D T1-weighted magnetization-prepared rapid acquisition with gradient echo (MPRAGE) anatomical image was acquired with TR/echo time (TE)=2200/2.8 ms, field of view (FOV)=24 cm, matrix size = 512×512×256, slice thickness = 0.5 mm, voxel size = 0.47×0.47×0.5 mm, and flip angle (FA)=8°. Subjects also underwent a BH task fMRI scan using an advanced multiband multi-echo (MBME) sequence and the following parameters: TR = 1000 ms, TE = 11,30,49 ms (three echoes), FOV = 24x24 cm, matrix size = 80×80, slice thickness = 3 mm (3×3×3 mm voxel size), 11 slices with multiband factor = 4 (44 total slices), FA = 60°, BW = 250 kHz, echo spacing = 0.51 ms, partial Fourier factor = 0.85, and in-plane acceleration with R = 2. The functional MBME scan lasted 320 s for a total of 320 volumes. The TEs for the MBME scan were set to the minimum possible.

A BH task was used to induce hypercapnia in participants to generate CVR, as described previously [9 –11]. The BH protocol was adapted from Cohen et al. and began with 72 s of paced breathing, followed by four cycles of 16 s of BH on expiration, 16 s of self-paced recovery breathing, and 24 s of paced breathing [10]. Scans ended with an additional 24 s of paced breathing. To control for respiration effects, paced breathing was implemented with cycles consisting of alternating 3 s inspiration and expiration blocks. The total scan time was 320 s. The BH paradigm is outlined in Supplementary Figure 1. Subjects received visual cues for all steps during the task. For the paced breathing a red bar (displayed in grayscale in Supplementary Figure 1) filled in or out during periods of inspiration or expiration, respectively. Subjects were notified the breath prior the BH that the BH was coming. Furthermore, to ensure compliance with the BH task, subjects’ breathing was monitored with respiratory bellows. Respiratory traces were manually inspected for all subjects and subjects without four clear breath holds at the correct timing were excluded from further analysis.

Clinical assessments

Control group: Control participants completed a medical history questionnaire and a brief cognitive battery which included a word list memory measure (Rey Auditory Verbal Learning Test, RAVLT) [25, 26], measurements of psychomotor processing speed and mental flexibility (Trail Making Test, Parts A and B) [27], and rapid word generation following letter and semantic cues (letter fluency and category fluency) [27]. Additionally, to further support that controls did not display behavioral changes concerning for dementia, the Quick Dementia Rating System (QDRS), a 10-item questionnaire was completed by an informant (scores range from 0 to 30 with higher scores representing greater cognitive and functional impairment) [28].

MCI group: A retrospective chart review was conducted to obtain clinical neuropsychological data on MCI participants. These data include medical history and data obtained from neuropsychological evaluations (e.g., demographic variables, MCI diagnosis, and neuropsychological test data). MCI participants completed comprehensive neuropsychological examinations as part of routine clinical workup. From these evaluations, common testing data elements were identified and entered into the study database including the Trail Making Test (Parts A and B) [27], as well as letter and category fluency tasks (composite score was created from one of three different versions [27 , 30]). Results from word list memory measures were also entered and included data from one of the following assessments: Hopkins Verbal Learning Test-Revised [31], RAVLT [25 , 32], or California Verbal Learning Test-2^nd or 3^rd Edition (CVLT-II or 3) [33, 34]. For the current study, we combined standardized delayed recall scores from these measures to have a single delayed recall common data element for between-group comparisons (“combined delayed recall” standardized score).

Data analysis

Imaging data were analyzed using a combination of AFNI [35], FSL [36], and Advanced Normalization Tools (ANTS) [37]. Data preprocessing followed the procedures outlined in Cohen et al. [10] using the Human Connectome Project (HCP) minimal preprocessing pipeline [38], modified to account for the multiple echo data [9].

Anatomical imaging processing was completed using the modified PreFreeSurferPipeline.sh scripts from the HCP pipeline. The MPRAGE image was first anterior commissure– posterior commissure (ACPC) aligned using aff2rigid in FSL. Next, a brain mask was generated by linearly registering the MPRAGE image to Montreal Neurological Institute (MNI) space using flirt in FSL [39, 40], and then nonlinearly refining the registration using fnirt in FSL [41]. Using these transformations, a reference brain-only image in MNI space was inverse warped to native space and used to mask the MPRAGE image. This brain-only MPRAGE image was then linearly registered to MNI space using flirt with 12 degrees of freedom [39] and then nonlinearly refined using fnirt. To control for potential variability in regional atrophy, individual grey matter (GM) density maps were also estimated using ANTS and the script antsCorticalThickness.sh and applied as the voxelwise covariate in further data analysis [42, 43].

The first eight volumes of the BH fMRI data were discarded to allow the signal to reach equilibrium. Then, the remaining volumes of fMRI were registered to the first echo. The subsequent echoes were registered using the transformation matrices from the first echo. Finally, the three echoes were combined using the $T_{2}^{*}$ -weighted approach [44].

Multi-echo independent component analysis (ME-ICA) was used to denoise the data using the open source Python script tedana.py version 0.0.11 (https://tedana.readthedocs.io/en/0.0.11) [45 –47]. As described in detail in previous work, ME-ICA identifies BOLD-independent components based on whether their amplitudes are linearly dependent on TE [45 , 48]. Following our previous work [10], an additional preprocessing step was added to avoid the BH task being erroneously classified as noise and regressed from the data [10]. In this step, the task frequency (f = 1/56 s) was bandpass filtered out of the data. ME-ICA was then run on the filtered data, and the noise components, including physiological and thermal noise, were regressed from the original, unfiltered dataset.

The ME-ICA-denoised BH fMRI dataset was then registered to the ACPC-aligned MPRAGE image using epi_reg in FSL, and next registered to MNI space using the previously computed anatomical transformations. Finally, the data were smoothed using a 6 mm FWHM Gaussian kernel. No additional nuisance regressors were removed from the MBME data.

The BH response was evaluated using a general linear model (GLM) with 3dDeconvolve in AFNI. After 3dDeconvolve, a restricted maximum likelihood model (3dREMLfit) was used to model temporal autocorrelations in the data. BH regressors were generated by convolving a square wave, with ones during BH periods and zeros otherwise, with the respiration response function (Equation 1) [49]. To account for the spatially varying and delayed BH hemodynamic response, which can fluctuate by as much as±8 s [49 –52], the BH regressor was shifted from –8 s to 16 s in steps of 2 s. The regressor that resulted in the highest positive t-score was chosen for each voxel. Finally, CVR was calculated as the percent signal change (PSC) of the BH task response by dividing the beta coefficient of the BH response by the mean signal. $RRF (t) = 0.6 \cdot t^{2.1} e^{- t / 1.6} - 0.0023 \cdot t^{3.54} e^{- t / 4.25}$ (1)

Voxelwise CVR comparisons were made between controls and MCI, controls and aMCI, controls and naMCI, and aMCI and naMCI subjects using 3dttest++ in AFNI with age, biological sex, and GM density as covariates. Multiple comparisons were controlled for using a cluster correction technique and 3dClustSim in AFNI [53]. The residuals of the 3dttest++ analysis were used to estimate the spatial autocorrelation function. Group data were thresholded at p < 0.05 and p < 0.01, and p < 0.005 and cluster-wise corrected at α<0.05 meaning there is less than a 5% chance of each cluster being a false positive cluster. Minimum cluster sizes varied with comparison and ranged from 1064 to 1201 for p < 0.05, 259 to 283 for p < 0.01, and 163 to 179 forp < 0.005.

A significant age difference was observed between the control and MCI groups (p = 0.025) in our sample. To investigate the potential effects of age on the CVR metric, an ANCOVA analysis was run using 3dMVM in AFNI [54]. The overall effects of age and the group by age interaction effect were estimated and results were cluster-corrected using 3dClustSim as above.

Regions-of-interest (ROIs) analysis

Further quantitative comparisons were examined by extracting individual mean CVR values from each of the 7 ROIs from the Yeo functional network template [55] overlapping with the voxelwise clusters computed above showing significant differences between control versus MCI group analysis at cluster-wise corrected p < 0.05, α<0.05. To investigate subregions of the DMN, frontoparietal and salience networks, the AAL3 atlas [56] was used to select the following regions: middle frontal gyrus (MFG); superior frontal gyrus medial (SFGmedial); ventromedial prefrontal cortex (PFCventmed); posterior cingulate gyrus (PCC); superior parietal gyrus (SPG); inferior parietal gyrus (IFG); anterior cingulate gyrus (ACC); insula; precuneus. As for the Yeo ROIs, individual mean CVR values were extracted from these ROIs overlapping with voxel clusters showing significant differences between control versus MCI group analysis at cluster-wise corrected p < 0.05, α<0.05. Groups were compared on ROI-averaged CVR and cognitive assessments using ANCOVA analyses. Partial correlations (controlled for age and biological sex) between ROI CVR values and cognitive measures were conducted. In order to control for false positives associated with multiple comparisons, post-hoc Bonferroni calculations were made for ANOVA/ANCOVA analyses, and a false-discovery rate (FDR) correction, the Benjamini– Hochberg procedure, was performed for correlational analyses.

RESULTS

Mean age was statistically significantly greater in the MCI group (M = 73.71 years, SD = 5.80) compared to controls (M = 69.00, SD = 7.15), t(39)=–2.324, p = 0.025. There was no difference in education between the groups (controls: M = 17.00 years, SD = 1.92; MCI: M = 16.43, SD = 2.23), t(39)=0.878, p > 0.05. Means, SDs, and post-hoc comparisons for cognitive measures across groups are displayed in Table 2. With the exception of letter fluency, statistically significant differences were seen between the MCI and control groups for cognitive measures (Trail Making A and B, category fluency, and delayed recall), with higher mean values for the control group. The same was true for the control subjects versus MCI phenotypes. The mean difference score was significant between aMCI and naMCI on the memory measure but no other cognitivemeasures.

Table 2

Means, SDs, and Post-hoc Comparisons for Cognitive Measures for Control versus MCI Phenotypes

Measure	Controls	aMCI	naMCI	MCI	Controls	Controls	aMCI versus
					versus aMCI	versus naMCI	naMCI
M (SD)	M (SD)	M (SD)	M (SD)	M Diff.	M Diff.	M Diff.
TMT A	116.15 (10.45)	99.91 (11.13)	96.3 (16.46)	98.19 (13.68)	16.24*	19.85**	3.61
TMT B	113.10 (13.58)	86.09 (18.35)	82.80 (16.01)	84.52 (16.93)	27.01**	30.30**	3.29
Letter Fluency	101.05 (18.52)	92.82 (15.48)	92.44 (16.74)	92.65 (15.63)	8.23	8.61	0.374
Category Fluency	107.00 (14.47)	77.09 (14.00)	86.22 (17.09)	81.20 (15.74)	29.91**	20.78*	9.13
Delayed Recall Combined	113.25 (13.40)	62.82 (9.23)	92.80 (18.29)	77.10 (20.70)	50.43**	20.45*	29.98**

Bonferroni correction; *≤0.01, **≤0.001; TMT, Trail Making Test.

Respiratory traces averaged across subjects for the control and MCI groups separately are shown in Supplementary Figure 2. Of the 15 participants excluded from all analyses, five control subjects and three MCI subjects were excluded due to lack of compliance with the BH task.

Depicted in Fig. 1 are mean CVR maps generated from a one-sample t-test for the control and MCI groups, and also the MCI group subdivided into aMCI and naMCI groups, with darker regions demonstrating lower CVR. Qualitatively, higher CVR was seen in the control group compared to the combined MCI group and aMCI and naMCI groups. CVR was also qualitatively higher for the naMCI group compared to the aMCI group. The results of the two-sample t-tests for group comparisons with age, biological sex, and GM density as covariates are also displayed in Fig. 1. The total MCI group had significantly lower CVR compared to controls at both of the cluster-wise corrected p < 0.05 and cluster-wise corrected p < 0.01 thresholds, mainly in the frontal and parietal regions and along the default mode network. For the cluster-wise corrected p < 0.005, threshold significant differences were seen in the insula, inferior parietal and inferior frontal regions, and supplementary motor area (SMA; significant but not visible in Fig. 1). The aMCI group also had significantly lower CVR compared to controls at both the cluster-wise corrected p < 0.05 and cluster-wise corrected p < 0.01 thresholds in similar regions including the frontal and parietal lobes, and specifically in the DMN and somatomotor cortex for cluster-wise corrected p < 0.05. For the cluster-wise corrected p < 0.005 threshold significant differences were only seen in parietal regions including bilateral precuneus and PCC. No voxelwise significant differences were observed between the controls and naMCI groups or the aMCI and naMCI groups. Results of the voxelwise ANCOVA analysis revealed no significant effects of age or group by age interactions. Uncorrected voxelwise age and group by age interaction images are shown in Supplementary Figure 3.

Fig. 1

One sample-test generated group mean CVR maps are displayed for control subjects and the whole MCI group (left, top) and for the MCI group subdivided into aMCI and naMCI groups (left, bottom). Globally, CVR was higher for the control group compared to MCI group and for the naMCI group compared to the aMCI group. Figure at right: t-test results of group comparisons with age, biological sex and voxelwise grey matter density as covariates. Voxelwise group t-test results showing control versus the whole MCI group (top) and control versus the aMCI group (bottom) for cluster-wise corrected p < 0.005, p < 0.01, and p < 0.05 (from left to right). Control versus naMCI and aMCI versus naMCI t-tests showed no significant regional difference at cluster-wise corrected p < 0.05.

ANCOVA analysis revealed group differences between MCI and controls in mean CVR ROI values for all Yeo functional networks ps < 0.001, with lower values for the MCI group. Post-hoc pairwise comparisons revealed that group differences were most notable between controls and aMCI mean values, which is consistent with the voxelwise t-tests results. However, controls did have statistically significantly higher mean CVR values compared to naMCI across all 7 networks. Mean CVR values for ROIs for each group are displayed in Table 3, along with corrected p-values for the post hoc comparisons between aMCI and controls and naMCI and controls. Illustrated in Fig. 2, naMCI showed an intermediate pattern of low CVR across all functional networks between controls and aMCI. Results in Supplementary Table 1 show a similar pattern of differences between controls and MCI phenotypes in key hubs within the frontoparietal, salience, and default mode networks.

Table 3

Yeo 1-7 Network Means and SDs for Controls and MCI Subgroups, and p values for Post-hoc Comparisons between Controls and MCI phenotypes

Regions of	Controls	aMCI	naMCI	Controls versus	Controls versus
Interest	M (SD)	M (SD)	M (SD)	aMCI p	naMCI p
Yeo 1, Visual Network	1.63 (0.350)	1.03 (0.332)	1.25 (0.462)	<0.001	0.027
Yeo 2, Somatomotor Network	1.19 (0.331	0.749 (0.290)	0.890 (0.313)	0.002	0.021
Yeo 3, Dorsal Attention Network	1.19 (0.276)	0.756 (0.251)	0.901 (0.372)	0.001	0.025
Yeo 4, Ventral Attention Network	1.38 (0.311)	0.897 (0.303)	1.07 (0.356)	<0.001	0.024
Yeo 5, Limbic Network	1.52 (0.655)	0.935 (0.269)	1.04 (0.495)	0.020	0.047
Yeo 6, Frontoparietal Network	1.45 (0.320)	0.890 (0.257)	1.07 (0.415)	<0.001	0.018
Yeo 7, Default Model Network	1.37 (0.361)	0.838 (0.274)	1.05 (0.388)	0.001	0.041

p-values for Bonferroni correction with age and biological sex as covariates. There were no significant differences between naMCI and aMCI groups.

Fig. 2

Yeo ROI averaged CVR compared between controls, non-amnestic MCI (naMCI), and amnestic MCI (aMCI), where naMCI shows an intermediate pattern of low CVR across all functional networks between controls and aMCI. Yeo1, Visual Network; Yeo2, Somatomotor Network; Yeo3, Dorsal Attention Network; Yeo4, Ventral Attention Network (including the Salience Network); Yeo5, Limbic Network; Yeo6, Frontoparietal Network; Yeo7, Default Mode Network.

As evidenced in Table 4, there are significant positive partial correlations between regional CVR along functional networks and cognitive measures— as CVR increases, cognitive scores increase.

Table 4

Partial correlation data for clinical assessment measures and Yeo networks for all participants

	1, Visual		2, Somatomotor		3, Dorsal Attn		4, Ventral Attn		5, Limbic		6, Frontoparietal		7, DMN
	r	p	r	p	r	p	r	p	r	p	r	p	r	p
TMT A	0.391*	0.015	0.406*	0.011	0.421*	0.008	0.379*	0.019	0.498*	0.001	0.416*	0.009	0.399*	0.013
TMT B	0.298	0.069	0.238	0.151	0.261	0.113	0.256	0.120	0.249	0.132	0.303	0.064	0.271	0.100
Letter Fl.	0.035	0.835	-0.107	0.523	-0.006	0.974	0.017	0.921	-0.062	0.712	0.001	0.995	-0.101	0.546
Sem. Fl.	0.510*	0.001	0.472*	0.003	0.516*	<0.001	0.519*	<0.001	0.336	0.039	0.533*	<0.001	0.454*	0.004
Memory	0.397*	0.014	0.440*	0.006	0.432*	0.007	0.467*	0.003	0.273*	0.098	0.465*	0.003	0.451*	0.004

p values listed are not corrected values. An asterisk was included for those correlations that remained significant after the B-H correction was applied, *<0.05; B-H values were never lower than 0.03. TMT, Trail Making Test; Letter Fl., Letter Fluency; Sem. Fl., Semantic Fluency; Memory, Delayed Recall; Yeo 1, Visual Network; Yeo 2, Somatomotor Network; Yeo 3, Dorsal Attention Network; Yeo 4, Ventral Attention Network (including the Salience Network); Yeo 5, Limbic Network; Yeo 6, Frontoparietal Network; Yeo 7, Default Mode Network.

DISCUSSION

To the best of our knowledge, this is the first neuroimaging study to examine CVR differences in aMCI and naMCI phenotypes. Consistent with previously reported MCI and dementia studies [15 , 22] we found that CVR is lower within MCI compared to cognitively normal controls providing further evidence that CVR is a physiological marker that differs based on clinical diagnosis.

Whole-brain analyses of CVR revealed relatively widespread regions of lower CVR within aMCI compared to controls, with clusters in the frontoparietal lobe and default mode network reaching significance across multiple thresholds. Similarly, regional CVR explored using ROI-based analyses (Yeo 1–7) was lower in aMCI and naMCI for all Yeo ROI networks, though the largest mean differences were between aMCI and controls.

While group differences between naMCI and aMCI did not reach statistical significance, when regional CVR mean values were visually inspected, aMCI had lower mean values across all Yeo ROIs compared to naMCI. Unfortunately, our study was underpowered and future research is needed to determine the potential importance of regional CVR for examining differences between MCI phenotypes and controls. Although further research is needed to elucidate CVR differences in MCI phenotypes, it is interesting to visually appreciate that aMCI may have the lowest CVR. Increasingly, there is recognition that vascular factors play an important role in AD. For instance, previous research that shows the highest prevalence of cerebrovascular disease in AD, compared to other neurodegenerative cohorts [57 –60]. One study found in middle-aged adults, a genetic risk factor for AD (APOE ɛ4 genotype) modified the relationship between CVR and cognition, suggesting early changes in CVR may be present and contributory to AD development.

The regional differences in naMCI revealed statistically significantly lower CVR in frontoparietal and dorsal and ventral attention areas, for example, which might not be entirely unexpected if naMCI is partly driven by a vascularly mediated mechanism. Overall, though, the etiology of naMCI is more heterogenous than aMCI. In vivo MRI shows that naMCI is associated with greater white matter abnormalities in bilateral posterior cingulum, right superior fasciculus, and left inferior fasciculus than aMCI [61]; however, structural and functional MRI studies of naMCI are sparse. Non-amnestic MCI is an understudied group and establishing neuroimaging biomarkers for naMCI phenotypes is crucial for determining outcome, probable etiological diagnosis/es, and neuroprotective treatments for these patients.

Lastly, we demonstrated a relationship between cognitive measures and regional CVR in controls and MCI. While prior studies used screeners, present results showed neuropsychological measures of processing speed, semantic language, and verbal memory functions were positively correlated with regional CVR in multiple ROIs, suggesting sensitivity of CVR to predict cognitive deficits across different cognitive constructs. Further interrogation of the relationship between specific cognitive domains and CVR ROIs was not possible in the current study due to limited statistical power.

The current study has several limitations. First, post-hoc analyses using MCI subgroups (amnestic and non-amnestic) were underpowered given the small sample sizes in individual groups. These issues were somewhat mitigated by the use of a MBME imaging sequence which has shown improved BOLD sensitivity and reliability compared to single band approaches [10]. Given power restrictions, we could not examine possible/probable MCI etiologies. Second, while assessment of cognitive functioning was conducted with gold-standard neuropsychological measures (reflecting an improvement compared to prior related research that used cognitive screening measures), valuable amyloid biomarker data frequently included in AD research were not available. Third, in the present study CVR was measured using the breath holding task. This approach is safer and less expensive; however, it has been noted that the quantity of blood CO2 detected will be impacted by an individual’s metabolic rate and lung functioning. Despite this weakness, overall, this approach has been shown to reliably correlate with inhalation of CO2, and we attempted to address confounds in a couple ways: first, respiration rates were examined and individuals that did not successfully complete the task were excluded from analyses. Further, individual differences were examined for potential outliers. In addition, potential confounding factors for CVR including smoking and alcohol drinks were not evaluated in this study. Also, we did not study the association of amyloid and other neurodegenerative biomarkers with CVR which merits further investigation. Finally, we did not collect heart rate (HR) data. Heart rate changes are known to affect the fMRI signal; however, we did not anticipate systematic changes in HR across groups. Furthermore, the ME-ICA denoising approach, coupled with the high temporal resolution used here, allow for physiological noise to be effectively removed from the data [10, 48].

While prior research supports a positive correlation between CVR and cognition, future research is needed to further appreciate the biological underpinnings of the relationship between CVR dysfunction and cognition, as well as further interrogation of the relationship between specific ROIs and cognitive domains in MCI clinical phenotypes. Longitudinal studies are also critical and should incorporate APOE genotypic data, as well as beta amyloid and tau AD biomarker data.

Conclusions

The findings of the current study provide further support that CVR is lower in MCI than in cognitively normal controls. We provide initial evidence that CVR is possibly lower in aMCI compared to naMCI and our ROI-based analyses between MCI phenotypes warrant further examination. Correlations between ROI-based CVR values and specific cognitive domains may provide novel mechanistic links between cerebrovascular disease and cognitive functions and also merit further investigation.

ACKNOWLEDGMENTS

The authors would also like to thank Montina Kostenko and Rachel Beers (Department of Radiology at the Medical College of Wisconsin) for study coordination and management.

FUNDING

This work was partially funded by a GE Healthcare research grant to Y.W. and an internal seed grant to L.G.U. L.G.U. and Y.W. are partially supported by NIA R21 AG075501-01.

CONFLICT OF INTEREST

This work was partially funded by a GE Healthcare research grant to Y.W. and an internal seed grant to L.G.U.

DATA AVAILABILITY

The data supporting the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Footnotes

The supplementary material is available in the electronic version of this article: .

References

Takeda

, Rakugi

, Morishita

(2020) Roles of vascular risk factors in the pathogenesis of dementia. Hypertens Res 43, 162–167.

Vemuri

, Lesnick

, Przybelski

, Knopman

, Preboske

, Kantarci

, Raman

, Machulda

, Mielke

, Lowe

, Senjem

, Gunter

, Rocca

, Roberts

, Petersen

, JackCR, Jr. (2015) Vascular and amyloid pathologies are independent predictors of cognitive decline in normal elderly. Brain 138, 761–771.

Viticchi

, Falsetti

, Vernieri

, Altamura

, Bartolini

, Luzzi

, Provinciali

, Silvestrini

(2012) Vascular predictors of cognitive decline in patients with mild cognitive impairment. Neurobiol Aging 33, 1127e1121–1129.

, Wang

, Xiao

, Tu

, Zhu

, He

, Liu

(2017) Molecular mechanisms of vascular dementia: What can be learned from animal models of chronic cerebral hypoperfusion? Mol Neurobiol 54, 3670–3682.

Salvadori

, Poggesi

, Pracucci

, Chiti

, Ciolli

, Cosottini

, Del Bene

, De Stefano

, Diciotti

, Di Donato

, Ginestroni

, Marini

, Mascalchi

, Nannucci

, Orlandi

, Pasi

, Pescini

, Valenti

, Federico

, Dotti

, Bonuccelli

, Inzitari

, Pantoni

, VMCI-Tuscany Study Group. (2018) Application of the DSM-5 criteria for major neurocognitive disorder to vascular MCI patients. Dement Geriatr Cogn Dis Extra 8, 104–116.

Korte

, Nortley

, Attwell

(2020) Cerebral blood flow decrease as an early pathological mechanism in Alzheimer’s disease. Acta Neuropathol 140, 793–810.

Tsvetanov

, Henson

RNA

, Rowe

(2021) Separating vascular and neuronal effects of age on fMRI BOLD signals. Philos Trans R Soc Lond B Biol Sci 376, 20190631.

Abdelkarim

, Zhao

, Turner

, Sivakolundu

, Lu

, Rypma

(2019) A neural-vascular complex of age-related changes in the human brain: Anatomy,hysiology, and implications for neurocognitive aging. Neurosci Biobehav Rev 107, 927–944 .

Cohen

, Wang

(2019) Improving the assessment of breath-holding induced cerebral vascular reactivity using a multiband multi-echo ASL/BOLD sequence. Sci Rep 9, 5079.

10.

Cohen

, Jagra

, Visser

, Yang

, Fernandez

, Banerjee

, Wang

(2021) Improving the breath-holding CVR measurement using the multiband multi-echo EPI sequence. Front Physiol 12, 619714.

11.

Pillai

, Mikulis

(2015) Cerebrovascular reactivity mapping: An evolving standard for clinical functional imaging. AJNR Am J Neuroradiol 36, 7–13.

12.

Catchlove

, Pipingas

, Hughes

, Macpherson

(2018) Magnetic resonance imaging for assessment of cerebrovascular reactivity and its relationship to cognition: A systematic review. BMC Neurosci 19, 21.

13.

Catchlove

, Parrish

, Chen

, Macpherson

, Hughes

, Pipingas

(2018) Regional cerebrovascular reactivity and cognitive performance in healthy aging. J Exp Neurosci 12, 1179069518785151.

14.

Sorond

, Hurwitz

, Salat

, Greve

, Fisher

(2013) Neurovascular coupling, cerebral white matter integrity, and response to cocoa in older people. Neurology 81, 904–909.

15.

Tarantini

, Tran

CHT

, Gordon

, Ungvari

, Csiszar

(2017) Impaired neurovascular coupling in aging and Alzheimer’s disease: Contribution of astrocyte dysfunction and endothelial impairment to cognitive decline. Exp Gerontol 94, 52–58.

16.

Palmer

, Kaufman

, Vidoni

, Honea

, Burns

, Billinger

(2022) Cerebrovascular response to exercise interacts with individual genotype and amyloid-beta deposition to influence response inhibition with aging. Neurobiol Aging 114, 15–26.

17.

Nelson

(2022) Peripheral pathways to neurovascular unit dysfunction, cognitive impairment, and Alzheimer’s disease. Front Aging Neurosci 14, 858429.

18.

Richiardi

, Monsch

, Haas

, Barkhof

, Van de Ville

, Radu

, Kressig

, Haller

(2015) Altered cerebrovascular reactivity velocity in mild cognitive impairment and Alzheimer’s disease. Neurobiol Aging 36, 33–41.

19.

McKetton

, Sobczyk

, Duffin

, Poublanc

, Sam

, Crawley

, Venkatraghavan

, Fisher

, Mikulis

(2018) The aging brain and cerebrovascular reactivity. Neuroimage 181, 132–141.

20.

Kim

, Lim

, Jeong

, Kang

, Park

(2019) The effect of clinical characteristics and subtypes on amyloid positivity in patients with amnestic mild cognitive impairment. Dement Neurocogn Disord 18, 130–137.

21.

Lee

, Jung

, Lee

(2007) Decreased vasomotor reactivity in Alzheimer’s disease. J Clin Neurol 3, 18–23.

22.

Silvestrini

, Pasqualetti

, Baruffaldi

, Bartolini

, Handouk

, Matteis

, Moffa

, Provinciali

, Vernieri

(2006) Cerebrovascular reactivity and cognitive decline in patients with Alzheimer disease. Stroke 37, 1010–1015.

23.

Peng

, Chen

, Li

, Rodrigue

, Park

, Lu

(2018) Age-related changes in cerebrovascular reactivity and their relationship to cognition: A four-year longitudinal study. Neuroimage 174, 257–262.

24.

Haight

, Bryan

, Erus

, Davatzikos

, Jacobs

, D’Esposito

, Lewis

, Launer

(2015) Vascular risk factors, cerebrovascular reactivity, and the default-mode brain network. Neuroimage 115, 7–16.

25.

Geffen

, Moar

, O’hanlon

, Clark

, Geffen

(1990) Performance measures of 16– to 86-year-old males and females on the auditory verbal learning test. Clin Neuropsychol 4, 45–63.

26.

Ivnik

, Malec

, Smith

, Tangalos

, Petersen

, Kokmen

, Kurland

(1992) Mayo’s Older Americans Normative Studies: WAIS-R norms for ages 56 to 97. Clin Neuropsychol 6, 1–30.

27.

Reitan RM, Wolfson D (1985) The Halstead-Reitan neuropsychological test battery: Theory and clinical interpretation, Reitan Neuropsychology.

28.

Galvin

(2015) The Quick Dementia Rating System (QDRS): A rapid dementia staging tool. Alzheimers Dement (Amst) 1, 249–259.

29.

Sue Baron

(2004) Delis-Kaplan executive function system. Child Neuropsychol 10, 147–152.

30.

Ivnik

, Malec

, Smith

, Tangalos

, Petersen

(1996) Neuropsychological tests’ norms above age 55: COWAT, BNT, MAE token, WRAT-R reading, AMNART, STROOP, TMT, and JLO. Clin Neuropsychol 10, 262–278.

31.

Brandt

, Benedict

(2001) Hopkins verbal learning test–revised: Professional manual, Psychological Assessment Resources.

32.

Taylor

(2013) Psychological Appraisal of Children with Cerebral Defects, Harvard University Press.

33.

Delis

, Kramer

, Kaplan

, Ober

(2017) California Verbal Learning Test Third Edition (CVLT 3) Manual, Pearson.

34.

Delis

, Kramer

, Kaplan

, Ober

(2000) Manual for the California Verbal Learning Test, (CVLT-II), The Psychological Corporation, San Antonio, TX, .

35.

Cox

(1996) AFNI: Software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 29, 162–173.

36.

Jenkinson

, Beckmann

, Behrens

, Woolrich

, Smith

(2012) Fsl. Neuroimage 62, 782–790.

37.

Avants

, Tustison

, Song

(2009) Advanced normalization tools (ANTS). Insight J 2, 1–35.

38.

Glasser

, Sotiropoulos

, Wilson

, Coalson

, Fischl

, Andersson

, Xu

, Jbabdi

, Webster

, Polimeni

, Van Essen

, Jenkinson

(2013) The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage 80, 105–124.

39.

Jenkinson

, Bannister

, Brady

, Smith

(2002) Improved optimization for the robust and accurate linear registration and motion correction of brain images. Neuroimage 17, 825–841.

40.

Jenkinson

, Smith

(2001) A global optimisation method for robust affine registration of brain images. Med Image Anal 5, 143–156.

41.

Andersson

, Jenkinson

, Smith

(2007), FMRIB Centre, Oxford, United Kingdom.

42.

Tustison

, Avants

, Cook

, Song

, Das

, van Strien

, Stone

, Gee

(2013) inMedical Imaging 2013: Biomedical Applications in Molecular, Structural, and Functional Imaging SPIE, pp. 126–129.

43.

Wang

, Risacher

, West

, McDonald

, Magee

, Farlow

, Gao

, O’Neill

, Saykin

(2013) Altered default mode network connectivity in older adults with cognitive complaints and amnestic mild cognitive impairment. J Alzheimers Dis 35, 751–760.

44.

Posse

, Wiese

, Gembris

, Mathiak

, Kessler

, Grosse-Ruyken

, Elghahwagi

, Richards

, Dager

, Kiselev

(1999) Enhancement of BOLD-contrast sensitivity by single-shot multi-echo functional MR imaging. Magn Reson Med 42, 87–97.

45.

Kundu

, Brenowitz

, Voon

, Worbe

, Vertes

, Inati

, Saad

, Bandettini

, Bullmore

(2013) Integrated strategy for improving functional connectivity mapping using multiecho fMRI. Proc Natl Acad Sci U S A 110, 16187–16192.

46.

Kundu

, Inati

, Evans

, Luh

, Bandettini

(2012) Differentiating BOLD and non-BOLD signals in fMRI time series using multi-echo EPI. Neuroimage 60, 1759–1770.

47.

DuPre E, Salo T, Markello R, Kundu P, Whitaker K, Handwerker D, ME-ICA/tedana: 0.0.6: https://doi.org/10.5281/zenodo.2558498.

48.

Olafsson

, Kundu

, Wong

, Bandettini

, Liu

(2015) Enhanced identification of BOLD-like components with multi-echo simultaneous multi-slice (MESMS) fMRI and multi-echo ICA. Neuroimage 112, 43–51.

49.

Birn

, Smith

, Jones

, Bandettini

(2008) The respiration response function: The temporal dynamics of fMRI signal fluctuations related to changes in respiration. Neuroimage 40, 644–654.

50.

Bright

, Murphy

(2013) Reliable quantification of BOLD fMRI cerebrovascular reactivity despite poor breath-hold performance. Neuroimage 83, 559–568.

51.

Bright

, Bulte

, Jezzard

, Duyn

(2009) Characterization of regional heterogeneity in cerebrovascular reactivity dynamics using novel hypocapnia task and BOLD fMRI. Neuroimage 48, 166–175.

52.

van Niftrik

, Piccirelli

, Bozinov

, Pangalu

, Valavanis

, Regli

, Fierstra

(2016) Fine tuning breath-hold-based cerebrovascular reactivity analysis models. Brain Behav 6, e00426.

53.

Cox

, Chen

, Glen

, Reynolds

, Taylor

(2017) fMRI clustering and false-positive rates. Proc Natl Acad Sci U S A 114, E3370–E3371.

54.

Chen

, Adleman

, Saad

, Leibenluft

, Cox

(2014) Applications of multivariate modeling to neuroimaging group analysis: A comprehensive alternative to univariate general linear model. Neuroimage 99, 571–588.

55.

Yeo

, Krienen

, Sepulcre

, Sabuncu

, Lashkari

, Hollinshead

, Roffman

, Smoller

, Zollei

, Polimeni

, Fischl

, Liu

, Buckner

(2011) The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol 106, 1125–1165.

56.

Rolls

, Huang

C-C

, Lin

C-P

, Feng

, Joliot

(2020) Automated anatomical labelling atlas 3. Neuroimage 206, 116189.

57.

Brickman

, Guzman

, Gonzalez-Castellon

, Razlighi

, Gu

, Narkhede

, Janicki

, Ichise

, Stern

, Manly

, Schupf

, Marshall

(2015) Cerebral autoregulation, beta amyloid, and white matter hyperintensities are interrelated. Neurosci Lett 592, 54–58.

58.

Sarro

, Tosakulwong

, Schwarz

, Graff-Radford

, Przybelski

, Lesnick

, Zuk

, Reid

, Raman

, Boeve

, Ferman

, Knopman

, Comi

, Filippi

, Murray

, Parisi

, Dickson

, Petersen

, Jack

Jr. , Kantarci

(2017) An investigation of cerebrovascular lesions in dementia with Lewy bodies compared to Alzheimer’s disease. Alzheimers Dement 13, 257–266.

59.

Sudre

, Cardoso

, Frost

, Barnes

, Barkhof

, Fox

, Ourselin

, Alzheimer’s Disease Neuroimaging I (2017) APOE epsilon4 status is associated with white matter hyperintensities volume accumulation rate independent of AD diagnosis. Neurobiol Aging 53, 67–75.

60.

Toledo

, Arnold

, Raible

, Brettschneider

, Xie

, Grossman

, Monsell

, Kukull

, Trojanowski

(2013) Contribution of cerebrovascular disease in autopsy confirmed neurodegenerative disease cases in the National Alzheimer’s Coordinating Centre. Brain 136, 2697–2706.

61.

Zhang

, Schuff

, Camacho

, Chao

, Fletcher

, Yaffe

, Woolley

, Madison

, Rosen

, Miller

, Weiner

(2013) MRI markers for mild cognitive impairment: Comparisons between white matter integrity and gray matter volume measurements. PLoS One 8, e66367.