Vascular health and diffusion properties of normal appearing white matter in midlife

In this study, we perform a region of interest diffusion tensor imaging and advanced diffusion complexity analysis of normal appearing white matter to determine the impact of vascular health on these diffusivity metrics in midlife adults. 77 participants (26 black, 35 female) at year 30 visit in the Coronary Artery Risk Development in Young Adults longitudinal study were scanned with an advanced diffusion-weighted imaging and ﬂuid-attenuated inversion recovery protocol. Fractional anisotropy and non-linear diffusion complexity measures were estimated. Cumulative measures across 30 years (9 study visits) of systolic blood pressure, body mass index, glucose, smoking and cholesterol were calculated as the area under the curve from baseline up to year 30 examination. Partial correlation analyses assessed the association between cumulative vascular health measures and normal appearing white matter diffusion metrics in these participants. Midlife normal appearing white matter diffusion properties were signiﬁcantly associated ( P < 0.05) with cumulative exposure to vascular risk factors from young adulthood over the 30-year time period. Higher cumulative systolic blood pressure exposure was associated with increased complexity and decreased fractional anisotropy. Higher cumulative body mass index exposure was associated with decreased fractional anisotropy. Additionally, in the normal appearing white matter of black participants ( P < 0.05), who exhibited a higher cumulative vascular risk exposure, fractional anisotropy was lower and complexity was higher in comparison to normal appearing white matter in white participants. Higher burden of vascular risk factor exposure from young adulthood to midlife is associated with changes in the diffusion properties of normal appearing white matter in midlife. These changes which may reﬂect axonal disruption, increased inﬂammation and/or increased glial proliferation, were primarily observed in both anterior and posterior normal appearing white matter regions of the corpus callosum. These results suggest that microstructural changes in normal appearing white matter are sensitive to vascular health during young adulthood and are possibly therapeutic targets in interventions focused on preserving white matter health across life.


Introduction
White matter hyperintensity volumes (WMH), as measured by fluid-attenuated inversion recovery (FLAIR) MRI, [1][2][3][4][5][6][7] have been associated with age related cognitive decline and functional impairment following ischaemic stroke.However, there has been growing interest to investigate normal appearing white matter (NAWM) microstructural integrity 8,9 that is not classified as lesioned tissue WMH via FLAIR. 1,4[12] However, DTI is just one modelling technique 13,14 to interpret diffusion properties of the white matter microstructure and only valid for a limited MRI data acquisition scheme. 15,16To more completely describe the diffusion-weighted signal, there have been attempts to implement a diffusion model that estimates kurtosis 17,18 as a measure of non-linear dynamics, but even this method has limitations on data acquisition requirements. 19More recently we have modelled these non-linear diffusion dynamics as anomalous subdiffusion 16,19 as a way to identify tissue microstructural complexity that has been shown to be sensitive to both axonal and glial cell morphology. 20n this study, we performed a region of interest DTI and complexity analysis of NAWM to determine the relationship between advanced diffusivity metrics and Graphical Abstract vascular risk factor exposure for a cohort of participants in midlife that have been enrolled in an ongoing 30-year study to identify Coronary Artery Risk Development in Young Adults (CARDIA). 21Vascular risk factors severely impact brain structure, 22 increase the presence of WMH burden 23 and are associated with cognitive impairment 24 and increased risk of dementia 25 in late life.The cumulative impact of increased vascular risk on brain structure is supported by midlife measures of vascular risk and late-life impairments 23 ; however, little is known about the impact on NAWM at midlife and its relationship to early life vascular risk.Therefore, a life-course approach may be necessary to identify the harmful consequences of vascular risk factors on neuroimaging markers of brain microstructure.We collected advanced diffusion MRI measures in 77 participants from the CARDIA cohort at the age of 55-60 years with longitudinal measures of vascular risk over the prior 30 years.In consideration of the low categorical instances of diabetes mellitus in this cohort and hypertension 26 in the white participants at year 30, we specifically focused on continuous measures of cumulative systolic blood pressure exposure (cSBP), 27 cumulative body mass index (cBMI), cumulative smoking exposure (cSmoke), cumulative glucose levels (cGlu) and cholesterol levels (cChol) from young adulthood to midlife to characterize potentially emerging vascular burden.

Participants
CARDIA began in 1985 across 4 field centres in the USA (Birmingham, AL; Chicago, IL; Minneapolis, MN and Oakland, CA). 21Black and white adults (N ¼ 5115), aged 18-30 years, were recruited and consequently followed with serial follow-up examinations through 30 years post their initial visit (Y 0 ).In the Year 30 visit (Y 30 ), the Chicago field centre invited and enrolled 202 of its participants in the Cerebral Small Vessels Disease in Motor and Cognitive Decline ancillary study approved by the institutional review board of Northwestern University.Separate written consent was obtained for each participant, where upon 77 subjects underwent a detailed assessment that included brain MRI with advanced diffusion-weighted imaging sequences.

Clinical measurements
At each of the nine visits during the 30 years of followup, systolic blood pressure (SBP) and diastolic blood pressure (DBP) was measured as per standardized CARDIA protocols previously described. 28,29Blood pressure measurements were the average of the last two of the three measurements taken at the brachial artery of subjects.For each participant, cumulative exposures to SBP (cSBP) and DBP (cDBP) were also calculated to examine the effect of blood pressure exposure from their participation in the stud at early adulthood to presently in midlife.These measures were calculated as the total sum of the product of the average millimetres of mercury recorded at two consecutive CARDIA visits and the number of years between those visits. 30The same calculations were performed for body mass index, smoking, glucose and cholesterol to produce the area under the curve cumulative metrics of cBMI, cSmoke (life-time pack years of cigarette smoking), cGlu and cChol exposure over the 30-year time-frame.

Diffusion-weighted image preprocessing, tensor fitting and complexity fitting
The diffusion-weighted images were first brain-extracted using the brain extraction toolbox in FMRIB software library (FSL). 31The data were then denoised using an estimate of the noise variance in CSF signal intensity of the right ventricle.The data were corrected for motion and eddy currents 32 by co-registering diffusion-weighted images to the image acquired with b ¼ 0 s/mm 2 using FSL.The motion correction transformation matrix was applied to the diffusion gradient directions to rotate them according to the registration algorithm.Using only the b ¼ 0 s/mm 2 and b ¼ 1000 s/mm 2 images, the preprocessed diffusion-weighted data were fitted to a tensor on a voxel-wise basis using DTIFIT in the FSL diffusion toolbox to produce estimates of FA, mean diffusivity (MD), radial diffusivity (RD) and axial diffusivity (AD).
Using the b ¼ 0 s/mm 2 , b ¼ 1000 s/mm 2 , b ¼ 2000 s/ mm 2 , b ¼ 3000 s/mm 2 images, the preprocessed diffusionweighted data were fitted to the Mittag-Leffler function (MLF) 33,34 using previously published procedures 19,20 and custom MATLAB codes. 35Prior to minimum leastsquared convergence to the MLF, a starting value for the classical diffusion coefficient, D, was estimated using a simple monoexponential function for the b ¼ 0 s/mm 2 and b ¼ 1000 s/mm 2 data.Using the b ¼ 0 s/mm 2 , b ¼ 1000 s/ mm 2 , b ¼ 2000 s/mm 2 , b ¼ 3000 s/mm 2 images the starting value for D and a starting value of a ¼ 1 were used to estimate a final value of D and the power law subdiffusion index, 0 < a 1 in Equation ( 1), where E a is the single parameter MLF, which is the characteristic functional form derived from the fractional order partial differential equation describing subdiffusion and power law dynamics. 19,36Following fitting of the MLF parameters, the powder average was computed to produce a mean value of a for each voxel.
The MLF is a special function that corresponds to specific functions for particular values of a. 33,34 For example, when a ¼ 1, the MLF is the simple monoexponential function, Another case is when a ¼ 0.5 and then MLF becomes the scaled complementary error function, Overall, a serves as a heterogeneity index to determine the deviation from simple homogeneous Gaussian diffusion (a $ 1), and the smaller the value of a, the more heterogeneous the diffusion, indicative of power law subdiffusive behaviour and an increasingly complex diffusion environment.For brevity of presentation, complexity, C, shall be defined as, such that the nonlinear diffusion signal captured by a value of a < 1 is interpreted as increased complexity.

WMH segmentation
FLAIR images were skull stripped, 37 denoised using a non-local means filter 38 and corrected for intensity nonuniformity. 39Cleaned images were input to the Lesion Prediction Algorithm of the Lesion Segmentation Toolbox (LST 2.0.15) 40for SPM12.Resultant voxelwise lesion probability maps were thresholded at greater than zero percent probability to form initial masks.These were manually corrected using freeview (Freesurfer v6.0) and approved by a neurologist.The brain-extracted FLAIR images were affinely registered to the b ¼ 0 s/mm 2 diffusion scan (3dAllineate AFNI), and the resultant transform was applied to the corrected mask with nearest-neighbor interpolation to obtain diffusion-space WMH masks.The FLAIR images were brain extracted and affine registered to the b ¼ 0 s/mm 2 diffusion scan using FLIRT in FSL.The same transformations were applied to the WMH mask images.

Statistical analyses
Tract-based spatial statistics (TBSS) were performed in FSL. 41FA maps were first linearly and then non-linearly registered to the FMRIB58_FA in Montreal Neurological Institute's (MNI) standard space.A mean FA image was then created from all individual FA images and used to generate a common group skeleton.A threshold was applied at 0.2 to minimize potential white matter/grey matter partial volume effects.Finally, each FA image was projected onto the common group skeleton for subsequent statistical analysis.The same transformations that were applied to the FA maps were also applied to the a, MD, RD, AD and WMH maps.In order to correct for voxel-wise raw statistic P-values, whole brain permutation-based testing (n ¼ 5000) was performed using the randomise and threshold-free cluster enhancement (TFCE) functions in FSL 42 in order to report family-wise error (FWE) corrected results.The WMH (cool) maps are shown in Figs 1-3 and are presented as a percentage of those individuals who had instances of WMH voxels with respect to the total group.4][45] NAWM is defined as the common group FA skeleton that does not overlap with the WMH probability map mask for this cohort.
To test for possible significant associations between the diffusion metrics and the individual cumulative vascular risk factors of cSBP, cBMI, cSmoke, cGlu and cChol in Table 1, partial correlation analyses (P < 0.05) were performed for each cumulative vascular risk factor while adjusting for age and all remaining cumulative vascular risk factor variables.Additionally, a two-group unpaired t-test (P < 0.05) was performed to compare the diffusion measures between the white participants with the black participants, who exhibited significantly different vascular risk factor history.This two-group unpaired t-test was adjusted for age and all remaining demographic variables in Table 1 in order to isolate, as best as possible, the neural microstructural morphology present between the two subsets of our cohort that encapsulate a multitude of individual vascular risk factor differences.Y 30 SBP, Y 30 DBP, hypertension risk, Y 30 BMI, Y 30 smoker status, diabetes mellitus, Y 30 glucose and Y 30 cholesterol are reported in Table 1 for completeness even though only cumulative measures were tested.Based on our previous work, 27,30 cSBP was chosen as the representative cumulative blood pressure exposure metric.The r and P values are reported as the average value of the significant voxels (P < 0.05) within the particular NAWM region of interest.

Data availability
Data can be made available upon reasonable request and in accordance with CARDIA policy and procedures.As to cumulative systolic blood pressure (cSBP) and FA with respect to cSBP.Green represents the white matter skeleton for the group comparison.Red represents voxels where there is a significant (P < 0.05) positive association between Complexity, C, and cSBP.Blue represents voxels where there is a significant (P < 0.05) negative association between FA and cSBP.Cool map represents the white matter hyperintensity (WMH) probability map from FLAIR images overlayed on the group FA mean brain (grey scale).Images are shown in rightÀleft radiological convention.Statistical analyses were adjusted for age and all other risk factor variables (cBMI, cSmoke, cGlu, cChol and WMH) shown in Table 1 as covariates and family-wise error corrected using permutation-based testing and threshold free cluster enhancement with respect to the whole brain.Green represents the white matter skeleton for the group comparison.Blue represents where there is a significant (P < 0.05) negative association between FA and cBMI.Cool map represents the white matter hyperintensity (WMH) probability map from FLAIR images overlayed on the group FA mean brain (grey scale).Images are shown in rightÀleft radiological convention.Statistical analyses were adjusted for age and all other risk factor variables (cSBP, cSmoke, cGlu, cChol and WMH) shown in Table 1 as covariates and family-wise error corrected using permutation-based testing and threshold free cluster enhancement with respect to the whole brain.   1 as covariates and family-wise error corrected using permutation-based testing and threshold free cluster enhancement with respect to the whole brain.
mentioned above, all analysis software has previously been made freely available for download.

Cohort characteristics
As shown in Table 1, the final analysis included 77 participants with a mean age of 57.0 years (SD 3.4), 35  were female and 26 were black.7][48] Specifically, the black participants were exposed to significantly higher burden blood pressure and BMI compared with the white participants, while the white participants were exposed to higher glucose levels compared with the black participants.Age was not significantly different between the black and white participants, but black participants had fewer years of education compared with the white participants.More specifically, WMH volume, cholesterol and cumulative smoking were not significantly different across black and white participants.

Analyses of non-significant parameters and sex demographics
Neither FA nor complexity, C, were significantly different when tested against cGlu, cChol, years of education and WMH adjusting for demographic and vascular risk factors in Table 1.A complete accounting of the ROI breakdowns for the significant voxels with respect to the JHU ICBM-DTI-81 white matter regions are available in Supplementary Tables 1 À 5.There have been numerous previous DTI studies [49][50][51][52][53][54][55][56] that have demonstrated a sexual dimorphism, particularly in the corpus callosum.The results in the present study confirm the previous work and demonstrate that the body, genu and splenium of the corpus callosum, in particular showed differences between females and males even when adjusting for all demographic covariates.

Discussion
We examined the relationship between cumulative vascular risk factor exposure across young adulthood and NAWM microstructure, as assessed by advanced MRI diffusion measures in midlife adults.Our results show that among vascular risk factors, cSBP and cBMI exposure were uniquely related to differences in these diffusion measures with some variance across race.
In line with previous structural and diffusion MRI studies, 45,57,64 our results also suggest that increased blood pressure and body mass index exposure are risk factors that may be deleterious to white matter structural integrity and health as evidenced by lower FA.Although the tissue determinants of diffusion measures are still not completely understood, as a global index of white matter organization, FA, combines information from all components of the diffusion tensor, and is largely dependent on fibre packing density and tortuosity, axonal membrane thickness and permeability, intercellular space size as well as myelin content and integrity. 65Lower white matter FA, which may represents loss of microstructural integrity, 13,14 could result from a degenerative process, such as axonal structural irregularities and degeneration or demyelination. 66The anterior and posterior regions of the corpus callosum and the peduncles demonstrated the majority of significantly changed FA voxels, indicating that these interhemispheric and brainstem white matter regions may be particularly sensitive to the longitudinal burden of these vascular risk factors.
The significant positive association between cSBP and the higher order measure of complexity, C, is a novel finding in our study.This observation would suggest that in those exposed to higher SBPs, the white matter microstructural environment is more heterogeneous, hence, the diffusion pattern appears more complex. 16While it could seem counterintuitive to observe an increased white matter diffusion complexity for those exposed to higher blood pressures, this relationship between advanced diffusion parameters and degenerative neuropathology is not without precedent.For example, a recent rodent study 67 estimated FA and complexity measures in both wild type mice and R6/2 mice, a well-established animal model of Huntington's disease.They showed that in R6/2 mice, FA was decreased and that complexity was increased in the corpus callosum when compared with wild-type mice.Corresponding to these diffusion metrics, pathologically there were dysmyelinated axons, but also an increased density of glial cells. 67While decreased FA likely represents the pathological loss of tissue organization, 66 increased diffusion complexity is likely corresponding to the proliferation of glial cells as an inflammatory response to neural injury and repair, hence, contributing to a heterogeneous white matter environment in the R6/2 mice.Similar observations have been reported in studies that estimated diffusion kurtosis differences in Alzheimer's disease 68 and demylenation/remylenation studies in mouse models. 18,69The complexity parameter, a estimated in this study and kurtosis, K, have a direct mathematical conversion. 19However, estimation of a is not limited by MRI acquisition choice 19 as is the case with classical techniques for estimating kurtosis. 17n the context of previous work, 18,[67][68][69] it possible that cBMI and cSBP associated with lower FA and cSBP associated with higher complexity, represent a combination of axonal disruption, inflammation and increased glial proliferation.As shown in Supplementary Tables 1  and 3, the majority of significantly changed voxels were observed in the corpus callosum, suggesting that the interhemispheric NAWM is particularly sensitive to longitudinal burden of vascular risk factors.While white matter characterization has focused on axonal morphology, 66 the contribution of glial cells to the MRI signal should not be ignored, as they comprise $40% of the volume in purely segmented white matter voxels. 70Therefore, in the context of the existing MRI measures linked to axonal and glial morphology, 67 and the observations of FA and complexity measures for this longitudinal cohort, advanced diffusion measures of NAWM may provide early measures of white matter health and resilience.In the clinical context, similar observations following acute stroke have shown that the NAWM microstructure as measured by diffusion MRI is linked to neural differences not only between those individuals with and without acute ischaemic stroke but also correlated with the severity of acute motor impairment. 12s demonstrated in Table 1, in our longitudinal cohort, the black participants had a higher burden of vascular risk exposure in comparison to the white participants.Therefore, we also sought to examine these advanced diffusion measures in this potentially at-risk demographic.We show that while lower FA was observed in 23% of the NAWM, there was higher complexity that was present in 43% of the NAWM in the black participants compared with the white participants.Moreover, there seemed to be an apparent spatial differentiation for complexity and FA with only a minor amount of spatial overlap of 27% between the significant FA and complexity voxels, and an apparent anterior presence for higher complexity and an apparent posterior presence for lower FA.Given that these observations parallel the diffusion patterns observed with cSBP and cBMI (Figs 1 and 2, respectively), they further support the deleterious impact of vascular risk burden on white matter diffusion properties reported in our study.The spatial differentiation between FA and complexity is particularly intriguing in the context of the 'retrogenesis' hypothesis, which postulates that the last areas to be fully developed, or myelinated, are the first area to be damaged in Alzheimer's disease as an example. 71,72In the context of this sub-analysis, when comparing the participants with the greatest cumulative exposure to vascular risk factors (black) to the ones with lower exposure burden (white), there is a heterogenous spatial distribution of non-overlapping significant FA and Complexity voxels that span both early (e.g.posterior limb of the internal capsule) and late (e.g.4][75] These observations, while novel, are preliminary, and must be interpreted with caution.However, one hypothesis may be that in NAWM inflammation and tissue reorganization are temporally and spatially separated.In other words, following the heterogeneous neuropathological temporal pattern of FA and Complexity changes reported in the C57/BL6J mice 18,69 and R6/2 mice, 67 inflammatory changes possibly representing injury and repair may occur across both early and late myelinating fibres, but tissue organizational degeneration may be preferential to certain late developmental white matter populations in this stage of the vascular risk exposure.In the context of cognitive changes associated with aging as it pertains to Alzheimer's disease, a potential target for further investigation, these results give merit to further investigations of the corpus callosum, which was a predominant ROI with significant changes in our analyses.][79] Moreover, multiple studies have also shown that FA in the corpus callosum is also significantly lower in patients with Alzheimer's as compared with mild cognitive impairment, 80,81 and mild cognitive impairment when compared with those with subjective cognitive complaints. 82As a new development in the current study, the corpus callosum 83 demonstrates not only passive structural changes as evidenced by lower anisotropy at midlife, but also active compensatory mechanisms via higher Complexity, both of which contribute to the microstructural progression of functional aging.The observation of reduced FA and increased complexity in the body of the corpus callosum with higher vascular risk exposure, may be related to ante-riorÀposterior differences in fibre myelination in this region of the corpus callosum, where larger myelinated fibres are located posteriorly.Further longitudinal data is needed to determine whether or not changes in complexity or FA have a temporal significance as potential targets for novel interventions directed at specific pathologies.

Limitations
While our finding represents an important initial step in characterizing changes in diffusion properties of NAWM exposed to vascular risk in a midlife cohort, our small sample size and cross-sectional imaging measures, warrant a cautious interpretation.This is particularly important for our demographic analysis across race where the impact of genetic, medication or other environmental influences cannot be fully investigated.Future work could investigate the testÀretest reliability of these analyses in which each participant undergoes two MRI scans at the cross-sectional timepoint of this longitudinal cohort to evaluate the stability of these parameters.Furthermore, it should be noted that the diffusivity metrics were tested independently and no additional adjustments were applied (e.g.Bonferroni correction) to the significance level, except for the FWE corrections to the reported Pvalues using permutation-based testing and TFCE.Additionally, the JHU white matter label atlas [43][44][45] covers only a subset of the white matter skeleton used for statistical analysis in this study.Therefore, there are significant voxels which are not represented in the ROI.

Conclusion
Midlife NAWM diffusion properties appear to be sensitive to higher cumulative exposure of vascular risk factors, specifically BMI and SBP.While increased cBMI over the 30-year time period was associated with lower FA, increased cSBP exposure over the same time period, was related to higher complexity, C, and lower FA.In the context of existing neuropathological data from experimental models, our results suggest that exposure to a higher burden of vascular risk factors across young adulthood is potentially associated with a combination microstructural morphological processes including axonal disruption, increased inflammation and increased glial proliferation in NAWM, particularly in the anterior and posterior regions of the corpus callosum.Diffusion properties of NAWM may have a differential temporal and spatial response to the burden of vascular risk exposure.However, this preliminary observation will need to be confirmed with longitudinal data.

Figure 1
Figure 1 Partial correlation analyses: tract-based Spatial Statistics Partial Correlation Analysis for Complexity, C, with respect

Figure 2
Figure 2 Partial correlation analysis: tract-based Spatial Statistics Partial Correlation Analysis for FA and cumulative body mass index (cBMI).Green represents the white matter skeleton for the group comparison.Blue represents where there is a significant (P < 0.05) negative association between FA and cBMI.Cool map represents the white matter hyperintensity (WMH) probability map from FLAIR images overlayed on the group FA mean brain (grey scale).Images are shown in rightÀleft radiological convention.Statistical analyses were adjusted for age and all other risk factor variables (cSBP, cSmoke, cGlu, cChol and WMH) shown in Table1as covariates and family-wise error corrected using permutation-based testing and threshold free cluster enhancement with respect to the whole brain.

Figure 3
Figure 3 Group comparison analyses: tract-based spatial statistics group comparison for complexity C (black) versus complexity C (white) and FA (black) versus FA (white).Green represents the white matter skeleton for the group comparison.Red represents significant voxels (P < 0.05) where complexity C (black) > complexity C (white).Blue represents significant voxels (P < 0.05) where FA (black) < FA (white).Cool map represents the white matter hyperintensity (WMH) probability map from FLAIR images overlayed on the group FA mean brain (greyscale).Images are shown in rightÀleft radiological convention.Statistical analyses were adjusted for all other demographic (age, sex, education and WMH) variables shown in Table1as covariates and family-wise error corrected using permutation-based testing and threshold free cluster enhancement with respect to the whole brain.
Vascular health and white matter diffusion BRAIN COMMUNICATIONS 2021: Page 9 of 12 | 9