Evaluating regional and global diffusion measures as biomarkers for vascular contributions to cognitive impairment and dementia

Highlights

• •Vascular risk factors were linked to consistent frontal white matter changes across diffusion metrics.
• •All diffusion metrics showed predictive value for cognitive decline.
• •Global free water and composite vascular white matter offered low sample size for clinical trials.

Abstract

Background

Diffusion MRI (dMRI) has been proposed for quantifying early tissue changes in cerebral small vessel disease (SVD). We evaluated the regional dependance of predictors of white matter (WM) damage and compared the utility of longitudinal WM changes in the commonly available diffusion MRI measures for vascular contribution to cognitive impairment and dementia (VCID) prevention trials.

Methods

We included 718 participants (mean age: 71.1(9.6) years, 56 % males) with at least two dMRI scans, amyloid-PET, and structural imaging. We computed single-shell dMRI measures (fractional anisotropy, mean diffusivity, free water, peak-width skeletonized mean diffusivity (PSMD) and assessed: i) regional dependance of predictors of baseline WM damage using voxel-level analyses; ii) longitudinal associations between dMRI measures and cognition; and iii) sample size estimates for a hypothetical clinical trial considering the regional and global dMRI measures as markers for VCID. We also included white matter hyperintensities (WMH) and our recently proposed composite vascular WM score (combination of WMH and fractional anisotropy of the genu) as a comparison.

Results

Vascular risk was consistently associated with dMRI changes in the genu of the corpus callosum. All SVD markers correlated with cognitive performance longitudinally. Global free water and the composite score provided the smallest sample size estimates, especially in participants with prevalent vascular disease (aged 70–89).

Conclusions

dMRI markers had significant frontal lobe changes due to vascular risk and were sensitive to cognitive decline. The composite vascular WM score, global free water, and WMH emerged as promising VCID biomarkers, but further validation is needed in multiple populations.

Introduction

Cerebral small vessel disease (SVD) arises from disorders in either the brain vasculature itself or in upstream vessels and is a major contributor to cognitive impairment and dementia in the aging population[1]. The most common MRI features identified as SVD include white matter hyperintensities (WMH), lacunes, microbleeds, and more recently, enlarged perivascular spaces and microinfarcts[2,3]. WMHs on fluid-attenuated inversion recovery (FLAIR) are considered the hallmark feature of SVD that reflects demyelination, gliosis, and axonal loss[2,4]. Accumulating evidence suggests that diffusion MRI (dMRI) measures may be more sensitive to early WM injury due to diffuse SVD well before the appearance of WMH[5,6].

Given the sensitivity of dMRI markers to SVD, there has been tremendous focus on developing quantitative markers that can be used in aging and dementia studies. Consortiums like MarkVCID (https://markvcid.partners.org/) have been established to identify and validate SVD biomarkers that capture vascular contributions to cognitive impairment and dementia (VCID). Two global measures based on single-shell Diffusion Tensor Imaging (DTI) that have been proposed by MarkVCID are the Free Water (FW) fraction from Free Water Elimination (FWE)[7] and Peak Width Skeletonized Mean Diffusivity (PSMD)[8]. FW indicates the extracellular water molecules that are not restricted by the local environment and PSMD reflects the dispersion of MD values across the WM skeleton; both have shown association with vascular risk factors, vascular burden, clinical deficits, and SVD-related cognitive imapirment[[9], [10], [11]].

Recently, there has been great interest in using DTI measures as the surrogate endpoints of SVD clinical trials [12,13]; with most studies have shown baseline and change in global DTI markers to predict cognitive impairment and dementia conversion[14,15]. Notably, a multicenter collaboration for optimizing multimodal imaging markers in trials of VCID demonstrated the lowest sample sizes for MD median[16] or PSMD[16] for large SVD datasets. However, regional specificity of dMRI markers, especially in frontal lobes, has been reported in association with aging and SVD-related WM damage[[17], [18], [19]]. Previous evidence from us and others focused on fractional anisotropy (FA) or mean diffusivity (MD) of anterior corpus callosum fibres[[20], [21]]; it is plausible that analyzing all single shell genu dMRI measures (FA, MD, FW, and PSMD) may improves the prediction of subtle SVD-related WM damage. While our work has focused on regions less likely to be impacted by AD pathologies, it is important to note that vascular amyloid can contribute to SVD-related WM damage particularly in regions vulnerable to cerebral amyloid angiopathy (CAA)[22]. Although WMH has been used in SVD prevention trials to stratify patients and evaluate the efficacy of novel drugs, its inability to capture early SVD changes underscores the utility of composite DTI and WMH measures to map the SVD heterogeneity beyond early microstructural changes. Hence, there is an unmet need to compare the global, regional, and composite SVD markers.

In this study, first, we evaluated the regional specificity of baseline dMRI markers to capture WM damage associated with systemic vascular injury and AD biomarkers using voxel-wise analysis. Using both global and regional dMRI measures, we then examined the longitudinal association of dMRI markers with cognition and computed sample size estimates for a hypothetical clinical trial. We included WMH and our previously proposed composite score of dMRI measure and WMH score[23,24] as a comparison in all analyses.

Methods

Selection of participants

Participants included in the study were enrolled in the Mayo Clinic Study of Aging (MCSA), an ongoing population-based study of the residents of Olmsted County, MN. The Rochester Epidemiology Project (REP) medical records-linkage system was used to enumerate the MCSA sample population[25,26]. This study included 718 individuals, who were at least 50 years of age with two usable dMRI measurements, and a baseline Pittsburgh compound B (PiB)-PET (amyloid) scan. The study design and diagnostic criteria for the MCSA population have been described previously[27].

Standard protocol approvals, registrations, and patient consents

All the study procedures were approved by the Mayo Clinic and Olmsted Medical Center institutional review boards. Written informed consent was obtained from all participants/surrogates.

Systemic vascular health indicator

We computed a composite score of cardiovascular and metabolic conditions (CMC) as the sum of presence and absence of seven conditions: hypertension, hyperlipidemia, cardiac arrhythmias, coronary artery disease, congestive heart failure, diabetes mellitus, and stroke as previously described[19] and utilized it as a marker of systemic vascular risk. These conditions were abstracted by a nurse from the electronic health records.

Assessment of amyloid marker from PiB-PET scans

The details of acquisition and computation of amyloid biomarker in the MCSA population were discussed elsewhere[28]. A global PiB-PET standardized uptake value ratio (SUVR) was computed by normalizing target regions (prefrontal, orbitofrontal, parietal, temporal, anterior cingulate, and posterior cingulate/precuneus) to the cerebellar crus grey matter and utilized as our amyloid marker.

Assessment of WMH from flair scans

The detailed description of acquisition, segmentation and WMH volume calculation were described in Graff-Radford et al [29]. In brief, WMH on T2-FLAIR images were segmented using a semiautomated algorithm[30] and the custom-made masks were further edited by trained image analyst. WMH volume scaled by total intracranial volume and log transformed due to skewness was used as the conventional SVD biomarker and referred to as WMH in this work.

Assessment of dMRI markers

All MR images were acquired on 3T GE Signa HDxt and Discovery MR750 scanners. The DTI acquisition and analysis protocol were described previously[19,31]. Briefly, the dMRI scans were acquired using a single-shot spin-echo-planar imaging sequence with an isotropic resolution of 2.7 mm. The protocol included five non-diffusion weighted directions (b = 0 s/mm²) followed by 41 diffusion weighted gradient directions (b = 1000 s/mm²), field of view = 350 × 159 mm, echo time = 68 ms, and repetition time = 9951 ms. After denoising[32], head motion and eddy current distortion were corrected using FSL’s eddy[33]. We corrected for Gibbs ringing[34], and skull stripped the images. The Rician noise bias was then removed using the noise image from denoising and the procedure outlined in Koay et al [35].. Diffusion tensors were estimated using nonlinear least squares fitting and used to calculate Fractional Anisotropy (FA) and Mean Diffusivity (MD) images in dipy[8].

FW: We computed Free Water (FW) maps using MarkVCID scripts, which are based on single shell dMRI, as published previously[36]. Here, we only utilized FW fraction because it is the only proposed biomarker from MarkVCID kit. Additionally, we did not consider corrected FA and MD, due to the inherent limitations in the regularized gradient descent fitting method used in single shell free water, specifically the fixed MD value of 0.6 μm²/ms, which suppresses the variations of microscopic tissue properties by attributing them to disease-related FW changes[37]. Briefly, the diffusion signal was initially modeled as two coexisting compartments: a tissue compartment characterizing intra and extracellular molecules and a FW compartment characterizing the fractional volume of water molecules in spaces with unrestricted diffusion. Then the FA maps were coregistered to the FSL FA DTI (FMRIB 1-mm FA) template using linear and nonlinear transformations and the resulting transformation parameters were then applied to FW map. Further, we derived a WM mask from FSL FA template by applying a threshold of 0.3 to reduce the partial volume contamination due to CSF. Finally, the individual FW maps were masked by WM mask, and the mean FW was computed by averaging over all WM voxels.

PSMD: PSMD was also computed using the latest version (v1.8.3) of the fully automated freely available shell script (http://www.psmd-marker.com/). We used the preprocessed FA and MD as the input data for the PSMD pipeline. In brief, computation of PSMD involved two steps; skeletonization of the FA-map and histogram analysis of the MD masked by the FA skeleton, as published previously[8]. FSL’s TBSS software was used to skeletonize the FA maps by registering all FA volumes to standard space using the FMRIB 1 mm FA template and then applying a threshold of 0.2 on FA maps. Then the MD volumes were projected onto the FA skeleton, and the resulting MD skeleton was further masked with a custom-made template skeleton supplied with the PSMD scripts. A threshold of 0.2 was applied to avoid partial volume contamination from CSF in regions near ventricles (e.g., the fornix). Finally, the PSMD was derived as the difference of the 95th and 5th percentiles of the included voxels’ MD.

Global brain analyses

To identify the contribution of certain regions to systemic vascular health, we performed a voxel-wise analyses of all available diffusion maps. Briefly, each participants FA, MD, and FW maps were nonlinearly registered to a custom-made study-specific template using ANTs-SyN. Each of the normalized diffusion images were then smoothed with an 8-mm FWHM isotropic Gaussian kernel and analyzed per-voxel within each tissue-class mask, using SPM12. Then, we fit multiple regression models with vascular risk (cardiovascular and metabolic condition, CMC) and amyloid (based on PiB-PET SUVR) as predictor and each of the baseline dMRI markers (FA, MD, and FW) as outcomes after accounting for age and sex[38]. The generated SPM-T maps were corrected for multiple comparisons using false discovery rate (FDR) with pFDR < 0.001 and those not survived for same threshold were reported with pFDR < 0.05.

Regional brain analyses

To obtain the regional diffusion measures, each participant's conventional FA and MD images as well as FW and PSMD skeleton images were nonlinearly registered to an in-house modified version of Johns Hopkins University “Eve” WM atlas (slightly modified by fusing left and right portions of structures spanning the left-right midplane particularly for genu and pons) using Advanced Normalization Tools software. For clarity, we estimated “Genu-FW” and “Genu-PSMD” by restricting the input voxels of FW and PSMD to the genu of the corpus callosum of the Eve atlas.

Composite of dMRI and flair

To evaluate the added value of WMH in addition to a quantitative regional dMRI marker, we considered a composite score called Vascular WM Score (a weighted sum of WMH and Genu-FA) derived from previously validated principal component analysis[23]. This vascular summary score better captures dynamic WM changes that are central to VCID[23,24], and proved to be more useful than relying on a single SVD imaging measure.

Cognitive performance

All the MCSA participants were underwent neuropsychological test battery encompassing four cognitive domains such as memory, attention/executive function, language, and visuospatial skills[39] and the derived global cognitive z-score was used as the cognitive outcome measure.

Statistical analysis

The characteristics of the participants were summarized with means and standard deviations for the continuous variables and counts and percentages for the categorical variables. Amyloid and WMH were log transformed to reduce positive skew. Each SVD imaging marker was standardized to be on similar scales. To assess the interrelationship between baseline imaging measures (DTI, WMH, and Vascular WM Score), we ran unadjusted Pearson correlation analyses using corrplot package.

We dichotomized all participants into two subsets using a population-specific median vascular risk cutoff of 2 (normal; CMC < 2, abnormal; CMC ≥ 2), with our sample comprising 434 participants classified as having abnormal CMC. In addition, receiver operating characteristic curve (ROC) analysis was performed to identify the sensitivity of neuroimaging biomarkers in distinguishing high versus low CMC groups after adjusting for demographic variables (age and sex). These analyses aimed to evaluate whether baseline imaging features are sensitive to systemic vascular burden, which is relevant for early risk stratification, participant selection, and mechanistic insight in SVD clinical trials. Next, individual slopes were calculated for each participant with each longitudinal SVD imaging marker and cognition z-score, and their relationship was assessed by Pearson correlation.

To evaluate the utility of the neuroimaging markers as surrogates for recruiting participants in clinical trials for VCID, we studied their longitudinal variability in the context of vascular risk factors. We estimated the required sample size in a hypothetical clinical trial to slow down the worsening of the imaging changes in the overall cohort as well as in participants with abnormal CMC. Sample size estimates were calculated for all participants and dichotomized CMC groups based on the slopes of neuroimaging SVD markers with arbitrary effect sizes to show 50 % or 25 % reduction in worsening of imaging measures with 80 % power. We also performed a sensitivity analysis restricted to participants with a narrow age band of 70–89 by repeating the sample size estimation due to increased vascular disease prevalence after 70 years. The power.t.test function from the stats package in R was used to calculate the sample size estimates, and the confidence intervals were estimated using bootstrapping.

Data availability

The data used in this study will be made available upon reasonable request from the authors.

Results

The demographics, APOE4 status, intellectual enrichment variables, cognitive measures, and vascular biomarker values are summarized in Table 1. The mean age of the participants was 71.1 years, 56 % males, 29 % APOE4 positive, and mean education of 14.8 years.

Table 1.

Characteristics table of all serial DTI subjects with the mean (SD) listed for the continuous variables and count ( %) for the categorical variables.

	All Subjects n = 718	CMC Normal n = 284	CMC Abnormal n = 434	P-value
Demographics
Age, yrs	71.1 (9.6)	67.5 (9.2)	73.5 (9.1)	<0.001
Males, no. ( %)	399 (56 %)	141 (50 %)	258 (59 %)	0.010
APOE4, no. ( %)	208 (29 %)	86 (30 %)	122 (28 %)	0.57
Follow-up, yrs	3.3 (1.5)	3.2 (1.4)	3.4 (1.6)	0.067
Number of visits	2.4 (0.6)	2.3 (0.6)	2.4 (0.7)	0.16
Intellectual Enrichment
Education/Occupation	12.6 (2.6)	12.9 (2.5)	12.5 (2.6)	0.049
Education, yrs	14.8 (2.7)	15.1 (2.6)	14.6 (2.7)	0.008
Cognitive Measures
MMSE	28.2 (1.5)	28.4 (1.5)	28.1 (1.6)	0.012
Global z-score	0.11 (1.09)	0.32 (1.03)	−0.04 (1.10)	<0.001
Memory z-score	0.12 (1.13)	0.25 (1.15)	0.04 (1.11)	0.014
Language z-score	−0.00 (1.04)	0.16 (1.04)	−0.11 (1.04)	0.001
Attention z-score	−0.00 (1.10)	0.29 (0.96)	−0.20 (1.15)	<0.001
Visualspatial z-score	0.19 (1.03)	0.31 (1.01)	0.11 (1.03)	0.011
Vascular Markers
Cardio Metabolic Condition	1.97 (1.26)	0.74 (0.44)	2.78 (0.93)
Abnormal CMC (2+), no. ( %)	434 (60 %)	0 (0 %)	434 (100 %)
MI, no. ( %)	68 (9 %)	0 (0 %)	68 (16 %)	<0.001
Statins, no. ( %)	365 (51 %)	70 (25 %)	295 (68 %)	<0.001
Dyslipidemia, no. ( %)	579 (81 %)	150 (53 %)	429 (99 %)	<0.001
Dyslipidemia Duration, years	16.03 (9.52)	11.52 (9.69)	17.91 (8.81)	<0.001
Systolic BP	139.99 (17.63)	135.74 (16.21)	142.77 (17.99)	<0.001
Diastolic BP	76.33 (10.19)	76.13 (9.35)	76.46 (10.71)	0.68
Hypertension, no. ( %)	447 (62 %)	55 (19 %)	392 (90 %)	<0.001
BP Meds, no. ( %)	393 (55 %)	51 (18 %)	342 (79 %)	<0.001
Hypertension Duration, years	14.54 (12.08)	7.17 (12.41)	16.22 (11.37)	<0.001
FA GCC	0.60 (0.05)	0.61 (0.04)	0.59 (0.05)	<0.001
MD GCC	0.84 (0.05)	0.82 (0.04)	0.85 (0.06)	<0.001
FW GCC	0.27 (0.06)	0.25 (0.05)	0.28 (0.07)	<0.001
FW	0.21 (0.04)	0.20 (0.04)	0.22 (0.04)	<0.001
PSMD GCC	0.20 (0.06)	0.18 (0.04)	0.21 (0.07)	<0.001
PSMD	0.27 (0.06)	0.25 (0.05)	0.29 (0.06)	<0.001
Vascular WM Score	0.83 (0.64)	1.02 (0.59)	0.70 (0.64)	<0.001
WMH/TIV %	0.886 (1.023)	0.642 (0.770)	1.045 (1.132)	<0.001
Infarction, no. ( %)	124 (19 %)	33 (12 %)	91 (24 %)	<0.001
Microbleed, no. ( %)	72 (15 %)	21 (10 %)	51 (19 %)	0.005
AD Biomarkers
Cortical Thickness, mm	2.87 (0.17)	2.92 (0.15)	2.84 (0.17)	<0.001
Adjusted Hippocampus Volume	−0.35 (0.65)	−0.28 (0.62)	−0.41 (0.66)	0.009
PIB SUVR	1.52 (0.35)	1.45 (0.28)	1.56 (0.38)	<0.001
GENU FA Slope	−0.335 (0.765)	−0.349 (0.706)	−0.326 (0.802)	0.70
GENU MD Slope	0.353 (1.069)	0.280 (0.921)	0.401 (1.154)	0.14
GENU FW Slope	0.415 (0.955)	0.324 (0.858)	0.475 (1.010)	0.037
FW Slope	0.369 (0.571)	0.296 (0.548)	0.417 (0.582)	0.005
GENU PSMD Slope	0.307 (1.596)	0.169 (1.167)	0.397 (1.819)	0.062
PSMD Slope	0.337 (1.250)	0.196 (1.108)	0.429 (1.328)	0.014
WMH Slope	0.065 (0.119)	0.046 (0.098)	0.078 (0.129)	<0.001
Vascular WM Score Slope	−3.909 (6.385)	−3.016 (6.235)	−4.492 (6.422)	0.002

P-values for differences between groups come from t-test for continuous variables and a chi-squared test for categorical variables. Slopes were multiplied by 100.

Baseline associations

Predictors of WM damage using voxel-wise analysis: Fig. 1 show that the largest commissural WM tract in the frontal lobe, particularly genu of corpus callosum showed the most significant association with CMC across all the dMRI measures (p < 0.001, FDR). Additionally, superior longitudinal fasciculus, anterior thalamic radiation, and inferior longitudinal fasciculus were related to CMC. On the other hand, the temporal part of parahippocampal cingulum (top associated tract) was related to amyloid on MD and FW maps (Right panel of Fig. 1).

Fig. 1.

Voxel-wise SPM maps showing the association of dMRI measures with the vascular risk (P < 0.001, FDR corrected, K = 100) and amyloid (P < 0.05, FDR corrected, K = 100). FA-Fractional anisotropy, MD-Mean diffusivity, FW-Free water, CMC-cardiometabolic condition.

Left panel shows that the largest commissural WM tract in the frontal lobe, particularly genu of corpus callosum showed the most significant association with CMC across all the dMRI measures (red-yellow represents reduced FA whereas blue-light blue represents increased MD or FW). On the other hand, the parahippocampal cingulum was related to amyloid (Right panel) (blue-light blue represents increased FW).

Interrelationship between SVD imaging markers and their distribution by age

Fig. 2 illustrates the SVD imaging biomarkers used in the further analyses. We evaluated the correlation between all these baseline imaging biomarkers including genu measures (Genu-FA, Genu-MD, Genu-PSMD, and Genu-FW), global FW, PSMD, WMH, and Vascular WM Score using unadjusted Pearson correlation analyses. There were strong associations between genu-measures as well as between genu-measures and global diffusion measures, except for Genu-PSMD which was moderate (Fig. 3). All diffusion measures significantly associated with WMH and Vascular WM Score suggest their utility to capture SVD damage.

Fig. 2.

Example of neuroimaging markers used to capture SVD related white matter damage: Illustration showing dMRI measures in the genu of corpus callosum outlined as red in fractional anisotropy (FA), mean diffusivity (MD), MD skeleton overlaid on MD image using Parula colormap, and free water (FW) maps, as well as white matter hyperintensities on T2-weighted FLAIR MRI. The minimum and maximum thresholds are ranging from 0 to 1 for FA and FW whereas MD ranges from 0 to 0.003 mm²/s. PSMD-peak width of skeletonized mean diffusivity refers to the difference between 95th and 5th percentiles of MD values obtained using histogram analysis.

Fig. 3.

Pearson correlation between neuroimaging measures used in the study.

FA-Fractional anisotropy, MD-Mean diffusivity, FW-Free water, PSMD-peak width of skeletonized mean diffusivity, WM-white matter, WMH-white matter hyperintensity

The distributions (mean and standard errors) of SVD imaging markers as a function of age are shown in Fig. 4. We found decreased Genu-FA and Vascular WM score, increased Genu-MD, Genu-PSMD, FW, PSMD, WMH, and Genu-FW with increase of age.

Fig. 4.

Bar plots of dMRI measures in the genu of corpus callosum, WMH, and Vascular WM Score (composite of Genu-FA and WMH) with 95 % of confidence intervals by 5-year increments.

FA-Fractional anisotropy, MD-Mean diffusivity, FW-Free water, PSMD-peak width of skeletonized mean diffusivity, WM-white matter, WMH-white matter hyperintensity.

Sensitivity analysis of neuroimaging markers using ROC analysis: When we evaluated the sensitivity of the baseline neuroimaging measures for the prediction of CMC using ROC analyses after dichotomizing participants with a cut-off of 2 as well as adjusting for age and sex, all diffusion markers performed equally well with Area under the curve (AUC) between 0.69 and 0.71 (supplemental Table 1).

Longitudinal associations

Change in imaging biomarkers and change in cognition

The longitudinal analyses between dMRI measures and cognition are shown in Table 2. Change in all dMRI markers were correlated with change in cognition. Regarding the genu measures, all metrics showed significant association with the global cognition and subdomain scores (attention and memory) (p < 0.05). Importantly, a high Genu-FW was most significantly associated with the low global cognitive scores (r = −0.23, p < 0.001). As expected, stronger associations were identified for global FW, PSMD, and WMH across all cognitive domains (p < 0.005).

Table 2.

Relationship between change in dMRI (slope) and change in cognition (slope).

	Global Cognition		Attention		Memory
	Estimate	P-value	Estimate	P-value	Estimate	P- value
Genu-FA	0.15	<0.001	0.11	0.004	0.09	0.02
Genu-MD	−0.19	<0.001	−0.17	<0.001	−0.14	<0.001
Genu-FW	−0.23	<0.001	−0.21	<0.001	−0.16	<0.001
FW	−0.35	<0.001	−0.25	<0.001	−0.19	<0.001
Genu-PSMD	−0.21	<0.001	−0.16	<0.001	−0.16	<0.001
PSMD	−0.29	<0.001	−0.19	<0.001	−0.15	<0.001
WMH	−0.29	<0.001	−0.23	<0.001	−0.17	<0.001
Vascular WM Score	0.15	<0.001	0.13	<0.001	0.10	0.012

Utility of dMRI measures for clinical trials and research

Sample sizes for a clinical trial with arbitrary effect sizes of 50 % and 25 % with 80 % power were computed for longitudinal change in all SVD imaging markers, as shown in Fig. 5 and Supplemental Table 2. Across all metrics, global FW and Vascular WM Score (composite of WMH and Genu-FA) showed the lowest required sample in all comparisons. When we repeated these analyses in participants with an age range of 70–89, Vascular WM Score outperformed all individual imaging biomarkers and had the best sample size estimate (Table 3).

Fig. 5.

Plot of sample size estimate using annual rate of change of dMRI measures in the genu of the corpus callosum (Genu-FA, Genu-MD, Genu-FW, Genu-PSMD), global PSMD and FW, WMH, and Vascular WM Score in all participants and participants with abnormal levels of CMC (≥ 2). The blue lines represent sample size estimate for all participants while the red lines denote estimate for abnormal CMC groups. The solid lines represent 50 % reduction while the dashed lines represent the 25 % reduction.

FA-Fractional anisotropy, MD-Mean diffusivity, FW-Free water, PSMD-peak width of skeletonized mean diffusivity, WM-white matter, WMH-white matter hyperintensity.

Table 3.

Sample Size Estimates from participants with age 70–89.

50 % reduction in the rate of worsening at 80 % power				25 % reduction in the rate of worsening at 80 % power
	All	CMC Normal	CMC Abnormal	All	CMC Normal	CMC Abnormal
	N (95 % CI)	N (95 % CI)	N (95 % CI)	N (95 % CI)	N (95 % CI)	N (95 % CI)
Genu-FA	216 (147, 346)	103 (67, 169)	297 (178, 587)	861 (590, 1392)	409 (267, 678)	1185 (714, 2352)
Genu-MD	298 (197, 530)	211 (113, 464)	342 (207, 772)	1189 (784, 2123)	840 (451, 1862)	1362 (823, 3083)
Genu-FW	165 (118, 260)	122 (75, 226)	185 (121, 347)	656 (467, 1040)	484 (299, 906)	736 (482, 1394)
FW	95 (73, 128)	121 (73, 229)	86 (65, 125)	375 (289, 509)	478 (287, 911)	339 (255, 502)
Genu-PSMD	941 (319, 16,724)	534 (200, 4730)	1067 (251, 20,251)	3759 (1271, 66,893)	2130 (796, 18,911)	4265 (1003, 81,067)
PSMD	296 (201, 479)	303 (157, 1005)	290 (184, 504)	1178 (800, 1912)	1206 (625, 4015)	1155 (733, 2011)
WMH	116 (92, 148)	141 (99, 212)	108 (83, 149)	461 (370, 594)	558 (391, 845)	427 (328, 593)
Vascular WM Score	77 (57, 133)	61 (43, 103)	84 (58, 177)	302 (225, 526)	240 (160, 416)	330 (227, 700)

Discussion

In this study, we evaluated the regional specificity of all available single-shell dMRI markers to SVD-related brain changes and then compared genu dMRI markers with global imaging measures based on their association with cognition and the required number of participants for a hypothetical clinical trial. The main findings of this study were: (1) vascular risk (CMC) was associated with all single shell dMRI measures in the genu of the corpus callosum; (2) Change in all SVD imaging measures correlated with change in global and subdomain scores; (3) Global FW and Vascular WM Score (composite scores of dMRI and WMH) had the had the smallest required sample sizes for a hypothetical clinical trial designed to slow down VCID across the population as well as participants with higher vascular risk. Interestingly, Vascular WM Score required the lowest number of participants in the age band of 70–89 where the vascular disease is more prevalent.

Genu as a more specific marker of systemic vascular health

Vascular health plays an important role in maintaining neurovascular health and is associated with WM abnormalities[40]. dMRI serves as a sensitive marker of ischemic microstructural alterations in WM associated with SVD, which often precede lesions visible on conventional MRI sequences, such as WMH on FLAIR that typically reflect more extensive white matter damage[41,42]. Loss of integrity of genu of the corpus callosum has been increasingly used to characterize aging and age-related vascular damage[43] and in a series of studies we also proposed this frontal WM component as the surrogate marker of VCID in the elderly[20,38,44]. Similar to previous findings that showed a correlation between Genu-FA and WMH[23], we found that all single-shell genu measures were correlated with WMH. These associations confirm that dMRI measures of genu capture SVD changes. The more deleterious vascular effects seen in the (late-myelinating) genu tract[45] may be due to increased disconnection seen in late-myelinating WM structures due to their myelin structure[46] and greater sclerotic changes seen in frontal lobe medullary arteries due to hypertension[47]. These regional changes were also confirmed by a decline in glucose metabolism with increased vascular risk factors[17].

Effect of amyloid burden on WM health

It is well known that Aβ deposition can cause WM damage in SVD, particularly in CAA and atherosclerosis[48,49]. There is also evidence suggesting posterior WM damage in association with Aβ deposition[38,50]. The voxel-wise analyses in Fig. 1 confirmed that elevated amyloid PET signal was associated with increased MD as well as FW in the temporal WM tracts. Specifically, the association was greater in the parahippocampal cingulum, the tract known to play an important role in episodic autobiographical memory [51].

White matter health and cognition

We also found significant longitudinal association between all SVD imaging markers and cognition. Although the correlations were significant and closer across measures, FW outperformed other DTI metrics and WMH to capture cognitive decline. These findings were consistent with previous reports suggesting increase of extracellular FW in dMRI as a marker of cerebrovascular dysfunction and associated cognitive impairment [9,19,20,52,53]. Although PSMD and FW were treated as global measures of WM integrity in previous SVD studies, we calculated those values from genu of corpus callosum along with global measures. The association that we found for genu-FW with cognitive deficits are in line with the disconnection hypothesis, where the disruptions of largest interhemispheric WM pathway in the frontal lobe was responsible for the underlying cognitive dysfunction in global cognition and attention/executive dysfunction[38]. Although the prognostic ability of vascular composite score is expected to be limited by the multiple mechanisms associated with the components, our study demonstrated a similar performance as other vascular brain indices, further suggesting its utility as an independent marker of SVD.

Diffusion MRI measures for VCID clinical trials

Neuroimaging measures, particularly DTI, have been proposed as a surrogate marker for clinical trials to evaluate the efficacy of new therapeutic approaches in phase II studies[8,14]. While many of these dMRI markers have the potential to predict the SVD related clinical deficits such as cognition[[8], [9], [10],44,54], a true surrogate for clinical trials would also require low sample sizes such that clinical trials with a reasonable number of participants can show the slowing down of VCID with treatment. Here, we evaluated all single-shell dMRI markers based on longitudinal change using data from the population-based sample and among all and high vascular risk groups. With a focus on dMRI markers, FW had the lowest sample size estimates across the groups suggesting that these may be early markers of WM damage due to systemic vascular injury and worsen at a higher rate in those at higher risk of VCID.

WMH, though having low sample size estimates, is a more downstream marker of VCID and appears later than dMRI changes, especially Genu-FA, as we have recently shown[6]. Interestingly, the longitudinal changes in Vascular WM score (composite of Genu-FA and WMH) provided the lowest sample size estimate, suggesting that regional dMRI measures together with traditional WMH as biomarkers can capture VCID better than dMRI alone. Future work should be focused on optimizing composite scores for VCID clinical trials. Indeed, global FW, WMH, and PSMD showed the expected pattern i.e. smaller sample sizes in the CMC-abnormal group, supporting their sensitivity to vascular pathology. In contrast, samples sizes for genu FA were smaller in the CMC-normal group suggesting that it may be more sensitive to early damage with a plateauing effect once there is significant vascular pathology.

Strengths, limitations, and future work

The major strength of this study was the large sample of participants with longitudinal dMRI data from a population-based cohort that allowed the realistic prediction of WM decline. Limitations include the focus on only dMRI markers without consideration of microbleeds, lacunes, infarctions, perfusion, and perivascular spaces. It is important to note that FW from single shell dMRI makes additional assumptions that are not valid[55] across gray and white matter because of their different microstructural properties. Because multi-shell dMRI offers the opportunity to accurately measure FW fraction with fewer assumptions, we plan to extend our study in the future. Future work using biophysical models including NODDI, multi-shell FW data, and more reliable vascular composite from all genu markers and WMH would be needed to unravel the multifactorial etiology of SVD. Focusing on Genu that has less FW and developing measurements similar to FA which are more continuous may yield better measures for detecting subtle changes within or between individuals.

Funding sources

This work was supported by NIH grants R01 AG056366, U01 AG006786, P50 AG016574, RF1AG069052, UF1NS125417, R37 AG011378, and R01 AG041851 and the GHR Foundation grant. This study used the resources of the Rochester Epidemiology Project (REP) medical records-linkage system, which is supported by the NIA (AG 058738), by the Mayo Clinic Research Committee, and by fees paid annually by REP users. The content of this article is solely the responsibility of the authors and does not represent the official views of the NIH or the Mayo Clinic.

Potential conflicts of interest

The authors do not have any pertinent disclosures relevant to this study.

CRediT authorship contribution statement

Sheelakumari Raghavan: Writing – review & editing, Writing – original draft, Formal analysis, Conceptualization. Scott A. Przybelski: Writing – review & editing, Formal analysis. Robert I. Reid: Writing – review & editing, Formal analysis. Michael G. Kamykowski: Writing – review & editing, Formal analysis. Jonathan Graff-Radford: Writing – review & editing. Val J. Lowe: Writing – review & editing. David S. Knopman: Writing – review & editing. Clifford R. Jack: Writing – review & editing, Funding acquisition. Ronald C. Petersen: Writing – review & editing, Funding acquisition. Prashanthi Vemuri: Writing – review & editing, Writing – original draft, Methodology, Funding acquisition, Formal analysis, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.