Regional differences in the link between water exchange rate across the blood–brain barrier and cognitive performance in normal aging

The blood–brain barrier (BBB) undergoes functional changes with aging which may contribute to cognitive decline. A novel, diffusion prepared arterial spin labeling-based MRI technique can measure the rate of water exchange across the BBB (kw) and may thus be sensitive to age-related alterations in water exchange at the BBB. However, studies investigating relationships between kw and cognition have reported different directions of association. Here, we begin to investigate the direction of associations between kw and cognition in different brain regions, and their possible underpinnings, by evaluating links between kw, cognitive performance, and MRI markers of cerebrovascular dysfunction and/or damage. Forty-seven healthy older adults (age range 61–84) underwent neuroimaging to obtain whole-brain measures of kw, cerebrovascular reactivity (CVR), and white matter hyperintensity (WMH) volumes. Additionally, participants completed uniform data set (Version 3) neuropsychological tests of executive function (EF) and episodic memory (MEM). Voxel-wise linear regressions were conducted to test associations between kw and cognitive performance, CVR, and WMH volumes. We found that kw in the frontoparietal brain regions was positively associated with cognitive performance but not with CVR or WMH volumes. Conversely, kw in the basal ganglia was negatively associated with cognitive performance and CVR and positively associated with regional, periventricular WMH volume. These regionally dependent associations may relate to different physiological underpinnings in the relationships between kw and cognition in neocortical versus subcortical brain regions in older adults.


Introduction
Normal aging is associated with alterations in cerebrovascular function that negatively affect cognitive performance [1][2][3][4].An increasing number of studies indicate that disruption of the blood-brain barrier (BBB) may represent an early contributor to age-related vascular cognitive declines [5][6][7][8][9][10].The BBB is a highly regulated vascular interface between the blood and central nervous system.It is made up of the endothelial cells that limit permeability and is sheathed by perivascular mural cells, pericytes, and astrocytic endfeet [5,11,12].The endothelial cells of the BBB have tight junctions which serve to regulate the flow of ions and molecules between the blood and brain and prevent toxins and pathogens from entering the brain, supporting proper neuronal functioning [13,14].
BBB disruption can be assessed in vivo using imaging techniques such as dynamic contrastenhanced magnetic resonance imaging (DCE MRI [5]).DCE MRI can track paracellular leakage of gadolinium (Gd)-based contrast agents (GBCAs) as these pass between the blood and brain (k trans ), revealing important information about both advanced and more subtle tight junction disruption [5,15,16], as well as increases in transcellular bulk flow transcytosis across the BBB [17].Increased k trans is associated with cognitive dysfunction in mild cognitive impairment, Alzheimer's disease, and vascular cognitive impairment [7,18,19].However, GBCAs have relatively large molecular weights (Gd-DTPA 550 Da) which may necessitate significant structural damage to the BBB before extravasation occurs [5,20,21].
Recently, a non-invasive MRI method has been proposed to measure subtler BBB alterations in the transport of water across the BBB using diffusion prepared arterial spin labeling (DP-ASL) MRI [21][22][23].Unlike DCE-MRI, the DP-ASL technique does not measure alterations in the permeability of the BBB.Instead, it can estimate water exchange rate across the BBB (expressed as k w ) which indicates the rate of water molecule exchange between capillaries and brain tissue (i.e., reciprocal of water exchange time) [21].The majority of water molecules traverse the BBB via water-specific transport proteins known as aquaporins, with aquaporin-4 (AQP4) being the most prevalent aquaporin channel in the brain [24].Characterized by a diameter comparable to that of a single water molecule, AQP4 only allows for diffusion of one molecule at a time [24,25].This regulated water transfer across the BBB has physiological significance in safeguarding the brain against edema and swelling [26,27].DP-ASL has been validated in relation to the exchange of ASL tagged water with tissue water in the rat brain, using a hypercapnic challenge [28,29].For instance, in the rat brain BBB k w was shown to increase as a function of mannitol-induced osmotic effects on water movement across the BBB [30].
This means that the MRI-based metric of DP-ASL k w may be sensitive to decreases in cognitive performance associated with age-related alterations in water exchange across the BBB.To date, however, only a handful of studies have employed DP-ASL to investigate potential links between BBB k w and cognition [21,[31][32][33].In addition, the relationships reported between BBB k w and cognitive performance have not been entirely consistent.In particular, while most studies have reported a positive relationship between BBB k w and cognitive performance [31,32], Shao et al. [21] reported negative associations between BBB k w in whole brain, gray matter (GM), and white matter (WM) and cognitive performance in a cohort of older adults at risk of cerebral small vessel disease (cSVD).
Given the potential utility of BBB k w as a noninvasive MRI metric of alterations in water exchange across the BBB, it is important to gain insight into the direction of associations between BBB k w and cognitive function, as well as the underlying mechanisms involved.One possibility is that different physiological underpinnings may contribute to the direction of the k w -cognition associations.Previous research by our group and by other researchers hypothesized that the positive associations between BBB k w and cognition may reflect optimal BBB-related protein clearance functions [31][32][33].However, negative associations between BBB k w and cognition appear incompatible with the optimal protein clearance function hypothesis.These negative associations were reported in a cohort of older adults at risk of cSVD [21].Therefore, one possible explanation for these Vol.: (0123456789) negative BBB k w -cognition associations could be age-related regional differences in cerebrovascular dysfunction and/or vascular related tissue damage.
In the current study, we investigate this possibility in a cohort of healthy older adults.Specifically, we first used a voxel-wise approach to identify associations between BBB k w in all possible brain regions and composite measures of executive function (EF) and episodic memory (MEM).EF and MEM were selected because these cognitive domains are among the first to be affected in normal aging [34][35][36].The voxel-wise analysis identified both positive and negative associations between BBB k w and both EF and MEM, in different brain regions.
Second, to assess links between BBB k w and cerebrovascular dysfunction and/or vascular-related tissue damage and how these links relate to cognitive performance, additional regression analyses were conducted between BBB k w in all regions emerging from the voxel-wise analyses and 1) cerebrovascular reactivity (CVR), a validated marker of cerebrovascular compliance [37,38] and 2) white matter hyperintensity (WMH) volumes, an MRI marker of WM damage associated with cSVD [39,40].

Participants
Forty-seven healthy older adults participated in this study.All participants provided informed consent under a protocol approved by the Institutional Review Board of the University of Kentucky.Participants were recruited from an existing longitudinal cohort at the Sanders-Brown Center on Aging [41] and the broader Lexington, KY, community.All participants were cognitively intact based on clinical consensus diagnosis and scores from the Uniform Data Set (UDS-3) used by US ADCs (procedure outlined in [42,43]) or a score of 26 or higher on the Montreal Cognitive Assessment (MoCa [44]) for those participants recruited from the community.
Study exclusionary criteria were self-reported significant head injury (defined as loss of consciousness for more than five minutes), heart disease, neurological or psychiatric disorders, claustrophobia, pacemakers, the presence of metal fragments or any metal implants that are incompatible with MRI, diseases affecting the blood (anemia, kidney/heart disease, etc.), or significant brain abnormalities detected during imaging.Twenty-one of the participants selfreported hypertension, but indicated it was currently under control by prescription medication which they identified by name.Four participants not only indicated recent/active diabetes but also indicated it was currently under control by prescription medication which they identified by name.Detailed characteristics of the participant cohort are shown in Table 1.Thirty of the subjects in the current study also participated in [33].
Lastly, the DP-pCASL sequence was acquired with the following parameters: TR = 4 s, TE = 36.5 ms, FOV = 224 mm, matrix size = 64 × 64, 12 slices (10% oversampling), resolution = 3.5 × 3.5 × 8 mm 3 , label/control duration = 1500 ms, and centric ordering, and optimized timing of background suppression for grey matter (GM) and WM [47].A two-stage approach was used to measure arterial transit time (ATT) and k w : fifteen repetitions were acquired during the flow encoding arterial spin tagging (FEAST) scan at post-labeling delay (PLD) = 900 ms and diffusion weighting (b-value) of 0 and 14 s/mm 2 with a total acquisition time of 4 min for estimating ATT.The k w metric was calculated from scans acquired at PLD = 1800 ms, when the labeled blood reaches the microvascular compartment, with b = 0 and 50 s/mm 2 , respectively.Twenty repetitions were acquired for each b-value of the k w scan, and the total acquisition time was 6 min.The 12 slices of the DP-pCASL scan did not cover the entire brain.As such, before each DP-pCASL scan, the area of acquisition was manually positioned to include as much of the hippocampus as possible (98% average hippocampal coverage) without excluding any cortical brain regions.Inferior temporal lobe regions below the hippocampus were partially covered by the DP-pCASL scan (53% average inferior temporal lobe coverage).

BBB k w mapping
BBB k w maps were created using the procedure outlined in [21].More specifically, DP-pCASL control/label images were first corrected for rigid head motion using SPM12 [48].Residual motion artifacts resulting from temporal fluctuations in the control/ label image acquisition were further reduced using principal component analysis [49], which can help restore the details of gyral structures.Following motion correction, k w maps were calculated by a total-generalized-variation (TGV; [50]) regularized SPA model [23] using the tissue (or capillary) fraction of the ASL signal at the PLD of 1800 ms, incorporating ATT, T1 of arterial blood, and brain tissue as inputs for the algorithm [21].Arterial blood T1 was assumed to be 1.66 s, which is commonly used for CBF quantification [51].To highlight regional differences in BBB k w across the participant cohort, the k w maps were converted to Z-scores, representing the number of standard deviations that individual voxels deviated from the mean BBB k w for each participant.
Following the creation of the BBB k w maps, the four echoes of the T1-weighted structural images from the ME-MPRAGE scan were averaged into a root mean square (RMS) image and registered to the M0 image of the DP-pCASL scan.A semi-automatic, edge-based approach was then used for registration, implemented using the align_epi_anat function of AFNI [52].In more detail, AFNIs 3dLocalstat function was first used to create normalized 2D gradient versions of the k w and T1-weighted images, corresponding to the local spatial variance within each image.The local spatial variance was computed as the standard deviation of intensities within a local neighborhood of voxels, divided by their mean intensity.The optimal alignment was subsequently computed between the 2D gradient version of the T1-weighted/ M0 images and the resulting transformation matrix was used to align the original T1/M0 images.
This edge-based alignment approach was selected because it yielded the best T1/M0 image registration among all the available methods provided by AFNI, for the majority of participants (inspected manually).For those participants in which registration was sub-optimal, the T1-weighted images were manually registered to the M0 images and corresponding transformation matrices created using the nudge tool in FSLeyes.Our semi-automated registration approach allowed for accurate registration of high-resolution T1-weighted images to the low-resolution images of the DP-pCASL scan (3.5 × 3.5 × 8 mm 3 ).
Following registration, the BBB k w registered T1-weighted images were non-linearly warped/normalized to MNI152 space (MNI ICBM152 1-mm 6th-generation atlas; [45]) using the AFNI function auto_warp.py.The transformation matrices obtained in the previous step were used to warp the BBB k w maps to MNI152 space, using the AFNI function 3dNwarpApply and a wsinc5 cost function.The MNI152 warped BBB k w images were used in subsequent voxel-wise analyses.Lastly, FreeSurfer 6.0 was used with the recon-all option to segment each participant's k w aligned T1-weighted image, in native space, into 89 cortical/subcortical ROIs (ROIs iterated in [53]).These 89 ROIs were eroded by one voxel to minimize partial volume effects and then resampled to the BBB k w voxel resolution (i.e., from 1 mm 3 isotropic to 3.5 × 3.5 × 8 mm 3 voxels).
Average k w values from the native space, Z-scored BBB k w maps were then extracted from each of these resampled ROIs for each participant, under the constraint that at least 90% of the voxels in a given ROI overlapped with the BBB k w map.This constraint is important because the BBB k w maps did not cover the entire brain, whereas the T1-weighted images did.All 47 participants included in this study had greater than 90% coverage in all ROIs reported in subsequent sections.Extracted values were used in ROI-based analyses to more closely evaluate the relationship between BBB k w and cerebrovascular dysfunction.Lastly, FreeSurfer estimated intracranial volumes (eICV, in mm 3 ) were recorded for each participant.In all ROIbased regression analyses, eICV was included as a covariate to account for head size-dependent differences in ROI volumes across participants.

Neuropsychological assessment
Forty-two out of the 47 participants of the cohort underwent neuropsychological testing using the National Alzheimer's Coordinating Center's (NACC) UDS-3 [42,43].The UDS-3 includes a comprehensive battery of neuropsychological tests assessing global cognition, memory encoding, memory retrieval, semantic memory, working memory, attention, executive function, processing speed, and verbal retrieval.Additional neuropsychological measures specific to the UK-ADRC were also administered to the participants, including the California verbal learning test, 2 nd edition (CVLT-II) used to construct the episodic memory composite scores described in the section below.

Executive function (EF) and episodic memory (MEM) composite scores
Composite scores were created for the 42 participants who underwent neuropsychological testing, to assess EF and MEM.As such, all subsequent analyses that use these composite scores have a sample size of 42, whereas analyses that do not include these composite scores use the full participant cohort (n = 47).Better performance is reflected by higher composite scores for both EF and MEM.The EF composite scores were calculated using a validated software toolkit developed for the UDS3 [54] and scores from the following individual cognitive tests: category fluency (animals and vegetables), phonemic fluency (letter F and letter L), digit span backward, and trail making test A and B. Specifically, correct responses are used for category fluency, phonemic fluency, and digit span backward, whereas number of correct lines per minute are used for the trail making test.
The UDS-3 does not currently have a software toolkit for calculating MEM composite scores.As such, MEM composite scores were calculated using a factor score-based approach, comparable to the one used for the EF composite [54].In more detail, individual test scores were first selected corresponding to those UDS-3/UK-ADC cognitive tests that best represent long-term memory performance.That is, those cognitive tests with the longest time delay between memory encoding and memory retrieval.These included the total story units recalled, using verbatim scoring from the Craft test, the delayed copy scores from the Benson complex figure test, and the long delay raw scores from the CVLT-II.An exploratory factor analysis indicated that the selected cognitive tests all represent a single latent factor (presumably long-term memory).Factor scores were then created for the single factor obtained in the previous step, using the Bartlett approach.This approach was selected because it provides 1) high validity (the Vol:.(1234567890) factor scores are highly correlated to the estimated factor) and 2) unbiased estimates of factor score parameters [55].
CVR imaging procedure CVR was assessed using a previously described procedure [56][57][58].Briefly, participants were fitted with a mouthpiece and nose-clip and mild hypercapnic air (5% carbon dioxide, 74% nitrogen, and 21% oxygen) was administered using a Douglas bag, with a two-way non-rebreathing valve, enabling precise switching between room-air and hypercapnic air [58].Participants underwent blocked inhalation of hypercapnic air and room air while BOLD fMRI was acquired.The third author (C.B.) was present inside the scanner room throughout the experiment to manually switch the valve to control the breathing air type (either room air or hypercapnic air).The CVR paradigm used interleaved blocks comprised of 70 s room-air breathing and 50 s of hypercapnic air breathing, for a total scan time of 7.20 min.CO 2 concentration in the exhaled air was sampled at 100 Hz, and the resulting CO 2 trace was recorded using capnography (Philips Respironics NM3 Monitor, Model 7900, CT).CVR data processing CVR analysis was performed using a cloud-based processing tool, CVR-MRICloud [38,59] as follows: the BOLD data were first motion corrected and smoothed by an 8-mm Gaussian kernel using SPM12.The end-tidal CO 2 (Et-CO 2 ) was extracted from the CO 2 trace using an algorithm to identify the peak CO 2 of each exhaled breath and then the resulting Et-CO 2 curve was temporally aligned (global-shifted) to the whole-brain BOLD signal time course.Unscaled, whole-brain CVR values were subsequently obtained using a general linear model (GLM) between wholebrain averaged BOLD signal and global-shifted EtCO2 (Eq. 1 in [38]).Lastly, the raw CVR values obtained in the previous step were scaled to units of %BOLD signal change (in the CO 2 -enhanced vs. normal air condition) per mm of mercury (Hg) of Et-CO 2 change (%BOLD/mmHg CO 2 ) using Eq. 2 in [38].
The resulting, individual participant CVR maps were then co-registered to their corresponding RMS T1-weighted images, which allowed warping/normalization of the CVR data to MNI space.In addition, the RMS T1-weighted images were segmented by the MultiAtlas Segmentation toolbox on MRICloud into 287 ROIs using the procedure described in [60,61] and the "Adult50-90_287Labels_30atlases_M2_252_ V10A" atlas.The segmented RMS T1 images were used to calculate per-ROI CVR values by constraining the whole-brain CVR procedure described previously within each ROI separately.That is, the Et-CO 2 curve was temporally aligned to the BOLD signal time course within each ROI separately, followed by the GLM and scaling steps.Lastly, to provide an index of the overall quality of the CVR data, per ROI partial correlation coefficients were computed between the BOLD time course and temporally aligned EtCO 2 , after factoring out the linear drift.These correlation coefficients passed quality control (automatically evaluated by the CVR-MRICloud toolbox) in all ROIs reported in subsequent analyses.

White matter hyperintensity volume quantification
WMHs were identified using a validated 4-tissue segmentation method [62] as follows.Participants' FLAIR images were first registered to their corresponding RMS T1-weighted image using the FSL FLIRT function (FSL version 6.0.1;[63]), corrected for inhomogeneities using a previously published local histogram normalization [64], and then non-linearly warped to a standard atlas [62].WMH probabilities were then estimated on the standard-atlas warped FLAIR images using a Bayesian probability structure, based on histogram fitting and prior probability maps [65].The prior-probability maps were created from more than 700 individuals with semi-automatic detection of WMHs followed by manual editing [62].The estimated WMH probabilities from the previous step were then thresholded at 3.5 SDs above the mean WM signal intensity of the FLAIR image to create per-participant, binary WMH masks.These binary WMH masks were then back-transformed to each participants' native space FLAIR image.Lastly, the native space WMH masks were visually inspected and manually edited were necessary to assure quality.
Following visual inspection and manual editing of the WMH masks, ROI-based masking was used to split the native space WMH masks into periventricular (PV) and deep brain regions.PV WMHs were defined as being located within ~ 10 mm of the lateral ventricles and deep WMHs were defined as those located outside this radius [66].The PV and deep ROIs used for masking were created using a validated group mask of the lateral ventricles in MNI152 space (MNI ICBM152 1-mm 6th-generation atlas; [45]), namely, the Automatic Lateral Ventricle delIneatioN (ALVIN) mask, created using data from 275 healthy adults (age range 18-94 [67];).The ALVIN mask extends beyond the lateral ventricles into brain parenchyma (~ 7-11 mm in most participants) to ensure inclusion of the lateral ventricles across participants with varying brain size.As such, the ALVIN mask, inversely warped from MNI152 space to each participant's native space FLAIR image, can be used to delineate PV WMHs.Similarly, a whole-brain, binary mask in the same space that excludes the ALVIN mask can be used to delineate deep WMHs.
More specifically, the following procedure was used.First, each participant's RMS T1 image was skull-stripped using FreeSurfer 6.0 and aligned to their corresponding FLAIR image in native space, using AFNIs align_epi_anat.pyfunction and a local Pearson correlation cost function.The aligned and skull-stripped T1 images were then non-linearly warped to MNI152 space (MNI ICBM152 1-mm 6th-generation atlas; [45]) using the AFNI function auto_warp.py.The inverse of the transformation matrix obtained in the previous step was then used to inversely warp the ALVIN mask from MNI152 space to each participant's FLAIR image in native space, using the AFNI function 3dNwarpApply and a nearest neighbor interpolation method.The inversely warped ALVIN masks were subsequently resampled to the FLAIR image grid using the 3dresample function in AFNI.Following the previous step, a binary brain mask was also created from the skull-stripped and FLAIR aligned RMS T1 images and resampled to the FLAIR image grid.
Each participant's whole-brain WMH mask was then multiplied by their native-space ALVIN mask to create a PV WMH mask.Similarly, a deep WMH mask was created by multiplying each participant's whole-brain WMH mask with a mask created by subtracting the native-space ALVIN mask from the FLAIR aligned and resampled, T1-based brain mask.Lastly, WMH volumes in mm 3 were calculated for each participant by computing the number of voxels in the PV and deep WMH masks.

Statistical analyses
Separate, group-level analyses were conducted with EF and MEM composite scores as independent variables (as mean centered vectors), and voxel-wise, Z-scored BBB k w as the dependent variable, using linear mixed-effects models (3dLME; [68]).Covariates were age and years of education (as meancentered vectors), and gender (as a categorical, between-subjects variable).The resulting statistical maps were thresholded at qFDR < 0.05 using the false discovery rate approach for multiple comparison correction [69].
To assess associations between BBB k w and cerebrovascular dysfunction in brain regions where the voxel-wise analyses were significant, subsequent ROI-based analyses were conducted.These analyses focused on BBB k w and CVR values extracted from those FreeSurfer and MRICloud/MultiAtlas ROIs that overlapped with significantly active voxels, as well as PV/deep WMH volumes.ROI analyses were then conducted in SPSS 27 (IBM, Chicago, IL, USA) using linear regression models with CVR values as the dependent variable and BBB k w as the independent variable.Age, years of education, gender, and eICV values acted as covariates in all linear regression models.
For WMH volumes, two separate linear regression analyses were conducted, one for PV and another for deep WMH volumes.PV/deep WMH volumes acted as the dependent variable whereas average BBB k w from all FreeSurfer ROIs that overlapped with significantly active voxels acted as independent variables.As in all previous linear regression models, age, years of education, gender, and eICV values were used as covariates.The WMH distributions were skewed as is typical and were log transformed.CVR values were also skewed and therefore also log-transformed in order to pass the Shapiro-Wilk test of normality [70].
Vol:. (1234567890) All remaining variables were normally distributed.Variance inflation factors (VIF) are provided in all linear regression analyses in order to evaluate the degree of collinearity between independent variables.Multiple comparisons are reported using the Sidak correction unless stated otherwise.

Voxel-wise BBB k w vs. EF composite scores
Results from this analysis revealed significant positive correlations between EF composite scores and BBB k w in the left superior frontal gyrus and bilateral precuneus.In addition, negative correlations were observed between EF scores and BBB k w in bilateral portions of the putamen (Fig. 1).The MNI152 coordinates of peak correlation voxels and their corresponding anatomical labels are provided in Table 2.

Voxel-wise BBB k w vs. MEM composite scores
Results from this analysis revealed significant positive correlations between MEM composite scores and BBB k w in the left superior and right middle frontal gyrus, as well as negative correlations between these scores and BBB k w in the left caudate and putamen (Fig. 2).The MNI152 coordinates of peak correlation voxels and their corresponding anatomical label are provided in Table 3.

ROI-based analyses between BBB k w and CVR
To evaluate potential links between BBB k w within the brain regions identified in the voxel-wise analyses and MRI metrics of cerebrovascular function, linear regression analyses were first conducted between average BBB k w and CVR extracted from the following GM ROIs: bilateral superior frontal gyrus, middle frontal gyrus, and precuneus.BBB k w was also extracted from composite ROIs comprised of the bilateral caudate and putamen ROIs.All linear regression analyses controlled for age, years of education, gender, and eICV.Results are summarized in Table 4.No significant associations were identified between BBB k w and CVR in the superior frontal gyrus, middle frontal gyrus, and precuneus ROIs.However, in the basal ganglia ROIs, a significant negative association between these measures was identified, indicating that higher BBB k w was associated with lower CVR values.

ROI-based analyses between BBB k w and WMH volumes
To further evaluate links between BBB k w and vascularrelated tissue damage, linear regression analyses were also conducted between average BBB k w within the same ROIs used in the previous section and PV/deep WMH volumes, while controlling for age, years of education, gender, and eICV.Results are summarized in Table 5.A single significant, positive association was observed between PV WMH volumes and BBB k w within the basal ganglia ROIs, such that higher BBB k w was associated with larger PV WMH volumes.

ROI-based analyses between CVR, PV, and WMH volumes and cognitive performance in the basal ganglia
A significant negative association was found between BBB k w and CVR in the basal ganglia, while a positive association was observed between BBB k w in the same region and PV WMH volumes.These findings raise the question of whether CVR values in the basal ganglia and neighboring PV WMH volumes also correlate significantly with EF and MEM composite scores, suggesting possible mediation effects of cerebrovascular dysfunction or vascular-related tissue damage on the association between BBB k w and cognitive performance in the basal ganglia.To test this possibility, two separate linear regression analyses were conducted, one with EF and another with MEM as the dependent variable and average basal ganglia CVR values and PV WMH volumes as independent variables in the same model.Age, years of education, gender, and eICV were added as covariates in both regression models.Results are summarized in Table 6.A significant positive correlation was identified between CVR in the basal ganglia and EF; however, PV WMH did not correlate significantly with EF.Neither CVR nor PV WMH volumes correlated significantly with MEM.
The effect of CVR on the association between BBB k w and cognitive performance in the basal ganglia A significant negative association was found between BBB k w and CVR in the basal ganglia, while a positive association was observed between CVR in this region and EF.This raises the question of whether the BBB k w -EF associations we observed in the voxelwise analyses are driven by alterations in the transport of water across the BBB per se, or more general vascular dysfunction associated with CVR.To test between these alternatives, a mediation analysis was conducted, testing the impact of CVR as a mediator on the direct effect of BBB k w on EF.The mediation analysis was conducted in SPSS using the Andrew F. Hayes PROCESS (version 3.5; model 4) computational tool [71].The mediation results were evaluated at 95% confidence intervals using the default setting of 5000 bootstrapped samples.As in all previous analyses, age, years of education, gender, and eICV were added as covariates.The results are presented in Table 7.
As anticipated from the voxel-wise results presented previously, a negative direct effect was observed between BBB k w in the basal ganglia and EF composite scores (Table 7).However, CVR did not mediate the direct effect of BBB k w on EF.

Discussion
Our results revealed significant correlations between BBB k w and both executive function and episodic memory performance.However, the direction of these associations varied across brain regions.Furthermore, BBB k w in the basal ganglia, but not in frontoparietal brain regions, was negatively associated with CVR and positively associated with periventricular WMH volume.Our findings collectively indicate that BBB k w is a sensitive metric of cognitive function in older adults, but appears to show different directions of association in neocortical and basal ganglia regions.These regionallydependent associations may relate to different physiological underpinnings in the relationships between BBB k w and cognition in neocortical versus subcortical brain regions in older adults.

BBB k w and cognitive performance
Our voxel-wise results present novel evidence linking BBB k w to cognitive performance in task relevant brain regions in older adults, after controlling for age, gender, years of education, and eICV.Specifically, BBB k w in the frontal and parietal lobe brain regions and in the putamen was associated with executive function, while BBB k w in the frontal lobe brain regions, caudate, and putamen was associated with episodic memory.Frontoparietal cortical brain regions are consistently implicated in executive functions, such as cognitive control [72][73][74], and frontal lobe regions are known to play a role in episodic memory retrieval [75][76][77][78].
The basal ganglia is strongly associated with initiation of movement [79] and motor learning [80], but the dorsal striatum (caudate and putamen) of the basal ganglia also play key roles in higher cognitive functions such as executive function and episodic memory.For example, the caudate and putamen have been associated with planning and set-shifting, the ability to alter a response in the face of changing circumstances [81], and other forms of cognitive control [82].Additionally, the putamen has been linked to verbal episodic memory, particularly with performance on the CVLT-II [83], which is one of the cognitive tests used to create the episodic memory composite scores in the current study.

BBB k w and cognitive performance: direction of associations
The direction of the k w -cognition associations we observed varied across brain regions, with positive associations observed in the frontal and parietal brain regions and negative associations observed in basal ganglia regions.The positive associations observed between BBB k w and cognitive performance in frontal and parietal brain regions are in keeping with the BBB k w -cognitive performance trends previously reported by our group in the same cortical brain regions, in a cohort of cognitively normal older adults [33].Moreover, similar associations in the neocortical brain regions have been reported by other researchers exploring BBB k w in patient populations [31,32,84].Results from the present study provide further evidence that high BBB k w in the neocortical regions is associated with high cognitive performance.
In contrast to the positive associations observed between BBB k w and cognition in the frontoparietal regions, the present study observed negative associations between BBB k w and cognition in the basal ganglia.The basal ganglia is a region that is particularly susceptible to age-related vascular dysfunction and degradation [85][86][87][88].Specifically, the sole blood supply to the basal ganglia and neighboring WM comes from the lenticulostriate arteries (LSAs) [89], which are small and tortuous with no collaterals, rendering them more susceptible to flow disruption caused by arteriosclerosis and/or ventricular expansion associated with aging [90].Therefore, we considered the possibility that the negative association between BBB k w and cognition in the basal ganglia may be related to vascular dysfunction and/or vascular-related tissue damage.
Our results indicated a selective negative association between BBB k w and CVR in the basal ganglia, with no comparable association in frontoparietal Vol.: (0123456789) regions.Similarly, a selective positive association was observed between BBB k w in the basal ganglia and PV WMH volumes that was not present in frontoparietal regions.Notably, this positive association in the basal ganglia was only significant for regional, PV WMH volumes and not more distal deep WMH volumes.PV WMHs, the most prevalent type of WMHs [90], are believed to be in part a consequence of vascular damage to the LSAs.Thus, there is an established connection between the basal ganglia and neighboring PV WMHs in terms of the LSA blood supply, contributing to their common susceptibility to age-related vascular dysfunction and degradation.Importantly, this arterial supply is not shared by deeper WM regions, which are not supplied by the LSAs.Therefore, the selective positive association we observed between BBB k w in the basal ganglia and PV WMH volumes only (not deep WMHs) is 1) consistent with the basal ganglia being a region where vascular dysfunction and/or damage tends to be prominent and 2) suggests a link between k w in the basal ganglia and neighboring vascular-related structural tissue damage.

Potential physiological underpinnings of the BBB k wcognition associations in neocortex and in the basal ganglia
In our previous study, we found that high BBB k w in neocortical regions, but not hippocampus, was associated with high concentrations of amyloid beta (Aβ)42 in cerebrospinal fluid (CSF; indicative of low Aβ deposition in neuritic plaques).Thus, our previous results suggested that high BBB k w values in neocortical regions may be indicative of increased BBB Aβ clearance rates in older adults [33].Other studies exploring BBB k w in patient populations have adopted similar conclusions, linking high k w to optimal BBB clearance functions [31,32].Optimal Aβ clearance rates are likely to be of most relevance to cognition, at least in healthy older adults, when considering neocortical regions, as Aβ deposition begins in neocortex [91,92].Thus, the positive associations we observed between k w and cognition in frontoparietal regions in this study are not inconsistent with an optimal clearance function explanation.
In contrast, the basal ganglia is not an early site of common proteinopathies such as Aβ deposition [91,92] or Lewy body deposition (comprising aggregates of ubiquitin and alpha-synuclein) [93].However, the basal ganglia is a major site of age-related cSVD, which could contribute to the negative relationship between BBB k w and cognition we observed in this region.Indeed, the significant links between BBB k w in the basal ganglia, CVR, and PV WMH volumes we observed here suggest that the association between BBB k w and cognition in this region may be related to vascular dysfunction and/or damage.Notably, however, the negative association between BBB k w and cognitive performance was not mediated by regional CVR or neighboring PV WMH volumes.This latter finding suggests that BBB k w in the basal ganglia may be uniquely sensitive to abnormal BBB water exchange that is associated with but extends beyond general cerebrovascular dysfunction/damage.
One possibility is that age-related arterial stiffening and associated vascular degradation in the basal ganglia may trigger the overexpression of perivascular water channels, leading to higher BBB water permeability and increased BBB k w .The water channel protein aquaporin allows water molecules to pass through the BBB one at a time [94].This limits the rate of water exchange across the BBB [22,[95][96][97], which is essential for protecting the brain from swelling and edema [21,25].Vascular degradation can trigger overexpression of these channels [98][99][100], increasing the BBB's water permeability and leading to reduced cognitive performance.We also cannot eliminate the possibility that the negative association between BBB k w and cognition in the basal ganglia could in part reflect loss of BBB integrity or BBB leakage, for example, age-related endothelial degeneration leading to loss of tight junctions [101].To evaluate this possibility, biofluid markers of BBB integrity are necessary, such as measures of CSF/ plasma albumin ratio [5,10,102].This possibility will be explored in a future study.
Finally, it should be noted that some of our findings differ from those observed by [32].Specifically, [32] found negative associations between BBB k w and markers of cSVD in the whole brain, temporal lobe, normal appearing WM, and putamen, in patients with cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL).In contrast, in our study, we found positive associations between BBB k w and cSVD measures in the caudate and putamen and no relationship between BBB k w and measures of Vol:.(1234567890) cSVD in neocortical regions.This divergence may be attributed to reduced expression of the water channel protein aquaporin-4 (AQP4) in the BBB of CADASIL patients, leading to low BBB k w values.Previous pathological studies have shown that reduced AQP4 expression in the BBB is a characteristic of CADASIL, associated with both chronic hypoperfusion of brain tissue and NOTCH3 gene mutations which can lead to apoptosis of astrocytes and increased clasmatodendritic astrocytes with displaced AQP4 [103][104][105].It should also be noted that CADASIL and other cerebrovascular disorders may prolong arterial/arteriolar transit times which may contribute to lower BBB k w values.

Limitations
This was a cross-sectional study that is limited to describing correlations.Future longitudinal studies are necessary to evaluate causal relationships between BBB k w , cognitive ability, and cerebrovascular dysfunction in normal aging.Further, while this study provides evidence for different physiological underpinnings underlying the relationship between BBB k w and cognition in cortical versus subcortical brain regions, the specific mechanisms driving these associations require further investigation.Animal model physiological and/or histological validation studies of BBB k w are necessary to directly test specific biological mechanisms linking BBB function and/or degradation to the k w metric.The use of other independent water permeability techniques may also be helpful in further elucidating the role of BBB alteration in cognitive aging [106][107][108].Lastly, the current study employed a DP-ASL sequence with relatively lowresolution to assess BBB k w .The use of a relatively low-resolution DP-ASL sequence limits our ability to investigate potential lateralization effects in the relationship between k w and cognition, particularly affecting voxel clusters near the midline.Future research using higher-resolution DP-ASL sequences [109] will be better positioned to investigate potential lateralization effects.

Conclusions
The present study provides novel evidence suggesting that the non-invasive, MRI measure of BBB k w is associated with cognitive functioning in healthy older adults.Further, the current study provides preliminary evidence suggesting that BBB k w may be associated with different physiological underpinnings in neocortex compared to subcortical structures.In particular, k w in the basal ganglia maybe sensitive to alterations in the transport of water across the BBB that are associated with, but extend beyond, general cerebrovascular dysfunction.Future studies should more thoroughly explore the relationship between BBB k w and cerebrovascular dysfunction in the subcortical brain regions using biofluid markers of BBB integrity.
Vol.: (0123456789) Commons licence, and indicate if changes were made.The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material.If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.To view a copy of this licence, visit http:// creat iveco mmons.org/ licen ses/ by/4.0/.

Fig. 1
Fig. 1 Voxel-wise results of the BBB k w vs. EF composite score analysis.Positive correlations are depicted with warm (orange to yellow) colors and negative correlations are depicted with cool (blue to cyan) colors.Results are overlaid

Fig. 2
Fig. 2 Voxel-wise results of the BBB k w vs. MEM composite scores analysis.Positive correlations are depicted with warm (orange to yellow) colors and negative correlations are depicted with cool (blue to cyan) colors.Results are overlaid

Table 1
Group demographics and mean cognitive measures

Table 2
MNI152 coordinates, cluster volume, and anatomical labels of voxels showing peak correlation between BBB k w and EF composite scores Vol.: (0123456789)

Table 3 MNI152
coordinates, cluster volume, and anatomical labels of voxels showing peak correlation between BBB k w and MEM composite scores Vol:.(1234567890)

Table 4
Linear regression analyses results: BBB k w vs. CVR in the superior frontal gyrus, middle frontal gyrus, precuneus, and basal ganglia ROIs.Results are adjusted for multiple com-parisons using the Sidak correction such that p < 0.013 uncorrected corresponds to p < 0.05 corrected

Table 5
Linear regression analyses results: BBB k w in the superior frontal gyrus, middle frontal gyrus, precuneus, and basal ganglia ROIs vs. PV/deep WMH volume (mm 3 ).Results are adjusted for multiple comparisons using the Sidak correction such that p < 0.025 uncorrected corresponds to p < 0.05 corrected Bold values indicate significant results 1 Standardized coefficients SFG superior frontal gyrus, MFG middle frontal gyrus, Precun precuneus, BG basal ganglia, PV periventricular Vol.: (0123456789)

Table 6
Linear regression analyses results: basal ganglia CVR and PV WMH volumes vs. EF and MEM composite scores.Results are adjusted for multiple comparisons using the Sidak correction such that p < 0.025 uncorrected corresponds to p < 0.05 corrected

Table 7
Mediation analysis results (process 3.5): basal ganglia CVR as a mediator of the direct effect of BBB k w on EF composite scores Vol:. (1234567890)