Longitudinal Relationships of White Matter Hyperintensities and Alzheimer Disease Biomarkers Across the Adult Life Span

Abstract

Background and Objectives

White matter hyperintensities (WMH) correlate with Alzheimer disease (AD) biomarkers cross-sectionally and modulate AD pathogenesis. Longitudinal changes have been reported for AD biomarkers, including concentrations of CSF β-amyloid (Aβ) 42, Aβ40, total tau and phosphorylated tau181, standardized uptake value ratio from the molecular imaging of cerebral fibrillar Aβ with PET using [¹¹C] Pittsburgh Compound-B, MRI-based hippocampal volume, and cortical thickness. Correlations between established AD biomarkers and the longitudinal change for WMH have not been fully evaluated, especially among cognitively normal individuals across the adult life span.

Methods

We jointly analyzed the longitudinal data of WMH volume and each of the established AD biomarkers and cognition from 371 cognitively normal individuals whose baseline age spanned from 19.6 to 88.20 years from 4 longitudinal studies of aging and AD. A 2-stage algorithm was applied to identify the inflection point of baseline age whereby older participants had an accelerated longitudinal change in WMH volume, in comparison with the younger participants. The longitudinal correlations between WMH volume and AD biomarkers were estimated from the bivariate linear mixed-effects models.

Results

A longitudinal increase in WMH volume was associated with a longitudinal increase in PET amyloid uptake and a decrease in MRI hippocampal volume, cortical thickness, and cognition. The inflection point of baseline age in WMH volume was identified at 60.46 (95% CI 56.43–64.49) years, with the annual increase for the older participants (83.12 [SE = 10.19] mm³ per year) more than 13 times faster (p < 0.0001) than that for the younger participants (6.35 [SE = 5.63] mm³ per year). Accelerated rates of change among the older participants were similarly observed in almost all the AD biomarkers. Longitudinal correlations of WMH volume with MRI, PET amyloid biomarkers, and cognition seemed to be numerically stronger for the younger participants, but not significantly different from those for the older participants. Carrying APOE ε4 alleles did not alter the longitudinal correlations between WMH and AD biomarkers.

Discussion

Longitudinal increases in WMH volume started to accelerate around a baseline age of 60.46 years and correlated with the longitudinal change in PET amyloid uptake, MRI structural outcomes, and cognition.

Cerebrovascular diseases (CVDs) often co-occur with Alzheimer disease (AD), sharing many risk factors such as age, APOE genotype, and hypertension. Contributions from CVD risk factors to cognitive impairment and dementia have been well recognized.¹ White matter hyperintensities (WMH), also known as leukoaraiosis in clinical manifestation and visualized as elevated intensities on T2-weighted MRI scans, is a radiographic biomarker for small vessel CVD. While gray matter loss–related brain atrophy has been long noted on the AD pathogenesis continuum, the links between white matter microstructural changes and AD neuropathologic changes remain unclear. A recent study on autosomal dominantly inherited AD suggests that WMH are a core feature of AD.² Accumulating evidence from mostly cross-sectional studies also suggests that, in late-onset AD, WMH predict dementia^3,4 and cognitive impairment,^1,4-8 accompany gray matter loss^8,9 and brain atrophy,^7,10,11 correlate with established AD biomarkers,¹² and are associated with the risk of AD progression.^12,13 Longitudinal changes in well-established AD biomarkers, including those from the CSF, MRI, and PET imaging using [¹¹C] Pittsburgh Compound-B (PiB), have also been well reported.^14-16 Their longitudinal relationships with longitudinal changes in WMH, however, remain unknown, especially among cognitively normal individuals across the adult life span.

This study aimed, first, to characterize the longitudinal trajectory of WMH volume and then, to examine the longitudinal relationships between WMH and major established AD biomarkers, as functions of baseline age and APOE genotype, among cognitively normal individuals across almost the entire adult life span. The biomarkers to be longitudinally correlated with WMH include the CSF concentrations of amyloid (β-amyloid 1–42 [Aβ42] and the Aβ42/β-amyloid 1–40 [Aβ40] ratio), total tau (t-Tau), and phosphorylated tau at 181 (pTau181); regional and composite measures of PET amyloid load using PiB¹⁴; structural/volumetric measures from the MRI, and cognition.

Methods

Participants

The study included participants from 4 longitudinal studies of aging and AD at Washington University School of Medicine (WUSM): the Knight Alzheimer Disease Research Center (Knight ADRC), the Healthy Aging and Senile Dementia (HASD) program project, the Adult-Children Study (ACS), and the Dominantly Inherited Alzheimer Network (DIAN).^17,18 All were longitudinally assessed for AD biomarkers (CSF, MRI, and PET imaging), cognition, and everyday function.¹⁹ Individuals with comorbid conditions that could interfere with cognitive assessments were excluded. The Knight ADRC and HASD enrolled participants aged 65 years or older at baseline, and the ACS enrolled participants in the age group of 45–75 years. The DIAN enrolled individuals older than 18 years from families with a parent carrying an AD-causing variant in the amyloid precursor protein, presenilin 1 or 2, but this study included only variant noncarriers. Other inclusion criteria for this study were cognitive normality at baseline and availability of longitudinal data on WMH volume and additionally, on at least 1 of 4 other modalities, including CSF biomarkers, PiB PET regional and composite measures of amyloid, MRI structural outcomes, and cognition. Details of the study design are presented in Figure 1.

Figure 1. Study Flowchart.

A total of 371 cognitively normal participants who had at least 2 longitudinal assessments of WMH volume were included for analyses, listing the number of participants with at least 2, 3, or 4 visits for each modality. Aβ = β-amyloid; ACS = Adult-Children Study; ADRC = Alzheimer Disease Research Center; CDR = Clinical Dementia Rating; DIAN = Dominantly Inherited Alzheimer Network; HASD = Healthy Aging and Senile Dementia; PiB = Pittsburgh compound B; pTau181 = tau phosphorylated at 181; t-Tau = total tau; WMH = white matter hyperintensities.

Standard Protocol Approvals, Registrations, and Patient Consents

All the participants provided written informed consent at study enrollment and agreed to research sharing of specimens and data. This study was approved by the WUSM Institutional Review Board.

Clinical and Cognitive Assessment

The clinical assessment protocols of the 4 studies were similar to and largely consistent with that of the National Alzheimer Coordinating Center Uniform Data Set.²⁰ Cognitive normality at baseline was defined by a global Clinical Dementia Rating²¹ score of 0. Comprehensive cognitive batteries were similar across Knight ADRC, HASD, and ACS cohorts but different from that of the DIAN. We selected 5 cognitive tests that represented the major cognitive domains and were shared across all studies: animal fluency (60 seconds), Boston naming test,²² Mini-Mental State Examination,²³ Wechsler Adult Intelligence Scale-R digit symbol, and Wechsler Memory Scale-R and Wechsler Memory Scale-III logical memory delayed.¹⁹ Each test was standardized to a Z score by subtracting the baseline mean and dividing the difference by the baseline SD. A cognitive composite was constructed by averaging the individual Z scores only when at least 3 of the 5 tests were available.

MRI and Analysis

Whole-brain structural MRI scans were acquired using prequalified MRI scanners with a field strength of 3.0T across the 4 studies. Scan acquisition and quality examinations primarily and consistently followed Alzheimer Disease Neuroimaging Initiative protocols. All the MRI scans were centrally processed by the Washington University Neuroimaging Core. T1-weighted images were processed using FreeSurfer (version 5.3 HCP-patch; Athinoula A. Martinos Center for Biomedical Imaging, Charlestown, MA) to obtain regional volumes. An AD cortical thickness signature was defined as previously described.²⁴ WMH volume was quantified from a T2-weighted fluid-attenuated inversion recovery sequence using the automated lesion segmentation toolbox implemented in SPM8. Total WMH volume was calculated as the sum of the labeled voxels multiplied by voxel dimensions.

PET Imaging and Analysis

[¹¹C] PiB PET was used to measure cerebral amyloid deposition in brain regions. As previously detailed,²⁵ the standardized uptake value ratio (SUVR) was calculated based on the regions of interest (ROI) defined by FreeSurfer. Data underwent partial volume correction using a geometric transfer matrix approach and used a cerebellar gray as the reference region. PiB cortical mean SUVR was calculated as the mean SUVR across cortical ROI including prefrontal cortex, precuneus, and temporal cortex, as previously described.^24,25 We analyzed PiB cortical mean SUVR as the metric for the global Aβ burden and the regional PiB SUVR at the precuneus (PiB precuneus SUVR) in this study.

CSF Collection and Analysis

CSF specimens were obtained by lumbar puncture from participants at approximately 8 am after overnight fasting. The clinically validated and automated immunoassay (LUMIPULSE G1200; Fujirebio, Malverne, PA) was used to measure CSF concentrations of Aβ42, Aβ40, t-Tau and pTau181, following standardized operating procedures.²⁶ All the longitudinal specimens of the same participant were assayed in the same run to minimize potential interrun artifacts. The CSF Aβ42/Aβ40 ratio was derived as the ratio of CSF Aβ42 to Aβ40.

APOE Genotyping

DNA from peripheral blood samples or buccal swabs was extracted following standard procedures and used for APOE genotyping. APOE genotype was defined as “APOE ε4 positive” if carrying 1 or more ε4 alleles vs “APOE ε4 negative” if none.

Statistical Analysis

We included only CSF and imaging biomarker data that passed standard quality control procedures. Participant characteristics were summarized by mean (SD) and median (range) for quantitative variables or count (percentage) for categorical variables. Longitudinal analyses of WMH were motivated by its spaghetti plot (Figure 2), which clearly indicated that at some point of baseline age, those who were older at baseline had an accelerated longitudinal change than the younger individuals. A 2-stage approach was implemented to pinpoint the inflection point. A random intercept and random slope linear mixed-effects model (LMM) was fitted to the longitudinal data of WMH. Then, the participant-level random slopes were modeled as a piecewise linear function of baseline age to estimate the inflection point of baseline age, using the algorithm by Muggeo and Davies.^27,28 Subsequently, the participants fell into the younger and older subcohort as having a baseline age before and after the estimated inflection point, respectively.

Figure 2. Longitudinal Trajectory and Inflection Point of WMH Volume.

(A) Longitudinal spaghetti plot of WMH volume (in raw scale) as a function of age in the longitudinal follow-up. (B) Estimated participant-level annual rate of change (slope in raw scale) as a piecewise linear function of baseline age, indicating an inflection point of baseline age at 60.46 years. WMH = white matter hyperintensities.

To assess the longitudinal relationship between WMH and each of the AD biomarkers, we implemented bivariate LMMs (BLMMs) to jointly model the longitudinal data of WMH and each biomarker, all converted into Z scores by using the baseline mean and SD. The models incorporated the fixed effects of intercept and slope for WMH and the AD biomarker and the participant-level random intercepts and random slopes with an unstructured covariance matrix. The longitudinal correlation between the annual rates of change of WMH and the AD biomarker was derived from the estimated covariance matrix and tested against zero using an asymptotic standard normal test, similar to what we previously implemented.¹⁴ These analyses were repeated after accounting for the main effects of major covariates: baseline age, race, sex, APOE ε4 status, education, intracranial volume, family history of dementia, body mass index (BMI), hypertension, hypercholesterolemia, and diabetes. Because of the inflection point in the longitudinal trajectory of WMH, we further examined whether the longitudinal rates of change in WMH and AD biomarkers, and their correlation, are different between the younger and older subcohorts, by including the subcohorts as an additional fixed effect on the slope and allowing heteroscedasticity (e.g., cohort-specific unstructured covariance matrix on the random effects). The goodness-of-fit of models with and without accounting for the young and old subcohort was evaluated by the Akaike information criterion (AIC). Similar analyses were conducted by stratification of APOE ε4–positive and APOE ε4–negative participants. Covariate-adjusted analyses were performed for sensitivity check by: (1) constraining to the participants who remained cognitively normal throughout the follow-up; (2) censoring the follow-up measurements post the inflection point for the young subcohort; and (3) constructing cognitive composite requiring ≥4 tests available. We repeated the analyses by incorporating the study (ADRC/DIAN/ACS/HASD) into the model for possible study-to-study variation. Residual plots after LMM fitting revealed no major deviations from the normal distributions.

All the statistical tests were 2-sided. The 2-stage algorithm was performed in R (version 4.1.1), and the LMMs were performed using SAS (version 9.4; SAS Institute, Cary, NC) PROC MIXED.

Data Availability

Anonymized data not published within this article will be available by request from any qualified investigators.

Results

Participant Demographics

A total of 371 participants were cognitively normal at baseline, had WMH data from at least 2 visits, and were included in the analyses (Figure 1). Participants were longitudinally followed up for a median of 3.52 (interquartile range 2.99–5.79) years, with 145 (39.1%) having data for ≥3 visits. Baseline WMH assessment had a median time difference of 2.0 years from baseline Aβ42/Aβ40, 2.20 years from baseline CSF t-Tau and pTau181, 0.2 years from baseline PiB SUVR, and 0.49 years from baseline cognition. Study-specific participant characteristics at baseline are summarized in Table 1. Overall, the baseline age spanned from 19.6 to 88.2 years, covering almost the entire adult life span with a median of 63 years. White individuals constituted nearly 90% of the cohort. Nearly 40% of participants were male, approximately one-third were APOE ε4 positive, and approximately 78% had a family history of dementia. The mean BMI was 27.3 kg/m². Approximately 66% had a positive vascular risk score due to obesity (i.e., BMI ≥30 kg/m²), hypercholesterolemia, hypertension, and diabetes. All participants had data from ≥3 markers. More than half (N = 196) had data from all the markers. The number of participants with at least 2, 3, and 4 longitudinal assessments is provided in Figure 1. A total of 55 participants dropped out during longitudinal follow-up, but only 3 because of AD progression, and hence, the chance of informative dropout due to AD presented no major concern.

Table 1.

Summary of Baseline Characteristics, Overall and by Study

Cross-sectional Correlations of WMH With AD Biomarkers at Baseline

At baseline, the unadjusted analyses indicated that WMH correlated with most of the AD biomarkers and cognition (eTable 1, links.lww.com/WNL/C815). Specifically, WMH was negatively correlated with CSF Aβ42/Aβ40 (r = −0.191, p = 0.0019), MRI hippocampal volume (r = −0.391, p = 7.49E-15), cortical thickness (r = −0.428, p = 8.64E-18), and cognition (r = −0.250, p = 2.20E-06). WMH was positively correlated with CSF t-Tau (r = 0.239, p = 0.0001), pTau181 (r = 0.313, p = 2.92E-07), and PiB SUVR at precuneus (r = 0.174, p = 0.0088). After adjusting for the effects of baseline age, sex, APOE ε4 status, education, and family history of dementia, however, the magnitudes of these correlations were dramatically reduced. There were no significant correlations between WMH and amyloid and tau biomarkers (eTable 1). Only correlations of WMH with cortical thickness and cognition remained statistically significant.

Longitudinal Trajectory of WMH Volume

Figure 2 (left panel) displays the spaghetti plot of WMH against age, which clearly indicates that the longitudinal rate of change accelerated after the baseline age of 60.46 years (Figure 2, right panel). Participants older than this inflection point at baseline exhibited a dramatically faster (Davies test²⁸ p < 0.0001) longitudinal rate of increase (slope/SE = 83.12/10.19 mm³ per year), in comparison with the younger subset whose longitudinal increase was not statistically significantly different from zero (slope/SE = 6.35/5.63 mm³ per year, p = 0.2599). To evaluate the longitudinal relationships between WMH and AD biomarkers, subsequent analyses were conducted including all participants and older and younger subcohorts defined by the inflection point of 60.46 years at baseline (N = 159/212 in the younger/older subcohort).

Longitudinal Trajectories of AD Biomarkers

The estimated annual rates of longitudinal change for the CSF and imaging biomarkers are summarized in Table 2. Although all participants were cognitively normal at baseline, there were already significant longitudinal changes for all CSF and imaging biomarkers among younger participants except for CSF Aβ42/Aβ40 (p = 0.06) and CSF Aβ42 (p = 0.41). The longitudinal changes were accelerated among the older participants for all AD biomarkers with the exception of CSF Aβ42, leading to significant differences in the annual rates of change between older and younger participants (p values ranging from 0.004 to <0.0001, detailed slope estimates in Table 2). The smallest magnitude of acceleration on the annual rate of change occurred for MRI cortical thickness whose annual rate among the older participants nearly doubled (p < 0.0001), in comparison with that among the younger participants. The largest magnitude of acceleration occurred for PiB cortical mean SUVR whose annual rate of change from the older participants was approximately 6 times that from the younger participants (0.133 per year vs 0.021 per year, p < 0.0001). For the cognitive composite, there was a practice effect for both the younger (annual rate = 0.004 per year, SE = 0.003) and older (annual rate = 0.007 per year, SE = 0.003) participants, which did not differ between the age groups (p = 0.462). After adjusting for all major covariates (Table 3), significant accelerations in the longitudinal changes remained from younger to older participants (at baseline) for all except for CSF Aβ42 and the cognitive composite.

Table 2.

Estimated Annual Rates (Block 3) of Change (in Z Score) for All AD Biomarkers and Their Correlations (r) With the Annual Rate of Change for WMH in the Entire Cohort (Block 1) and in the Young and Old Subcohorts (Block 2), as Defined by the Baseline Age of 60.46 Years

Table 3.

Multiple Covariate-Adjusted Annual Rate (Slope) of Change (in Z Score) for All AD Markers and Their Correlation (r) With Longitudinal Change for WMH, Overall and in the Young and Old Subcohorts

Longitudinal Correlations Between WMH Volume and AD Biomarkers

The estimated longitudinal correlations (r) in the annual rates of change between WMH and the CSF biomarkers, imaging biomarkers, and cognition are summarized in Table 2 for all participants and separately for the older and younger groups. To compare the BLMM with and without age groups in the fixed and random effects, log likelihood and AIC are listed in eTable 2 (links.lww.com/WNL/C815). The adjusted estimates of the longitudinal correlations between the annual rate of change in WMH and the AD biomarkers are summarized in Table 3.

There were no significant differences in the longitudinal correlations of WMH with the AD biomarkers between the younger and older participants. Of importance, among the younger participants, the annual increase in WMH was already correlated with the annual decrease in hippocampal volume (r/SE = −0.622/0.115, p < 0.0001), cortical thickness (r/SE = −0.378/0.149, p = 0.011), and cognitive composite (r/SE = −0.357/0.122, p = 0.003).

Among all the participants, a significant positive correlation in the annual rates of change was found between WMH and CSF t-Tau (r/SE = 0.254/0.122, p = 0.038). The annual increase in WMH was associated with the annual increase in PiB cortical mean SUVR (r/SE = 0.287/0.081, p = 0.0004) and PiB precuneus SUVR (r/SE = 0.297/0.081, p = 0.0002), the annual decrease in MRI hippocampal volume (r/SE = −0.282/0.112, p = 0.012), and marginally associated with the decrease in MRI cortical thickness (r/SE = −0.270/0.156, p = 0.083).

After adjusting for all major covariates (Table 3), there remained significant positive correlations (both r = 0.22, p = 0.004) in the annual rates of change between WMH and each of the 2 PiB biomarkers among all participants. In addition, the adjusted analyses confirmed that the longitudinal change in WMH was negatively correlated with the longitudinal changes in MRI hippocampal volume (r/SE = −0.6953/0.1294, p < 0.0001), cortical thickness (r/SE = −0.5157/0.2103, p = 0.0142), and cognition (r/SE = −0.5413/0.1342, p < 0.0001), among the younger participants. Adjusted sensitivity analyses constrained to the participants who remained cognitively normal throughout yielded similar results (eTable 3, links.lww.com/WNL/C815). The significant findings remained after censoring by the inflection point with negligible changes in the old individuals and small slope and correlation attenuation for some markers in the young individuals (eTable 4); When requiring ≥4 component tests available, slope difference for cognition was significant between the old and young subcohort, while the longitudinal correlation enhanced in both and remained significant in the young individuals (eTable 5).

Effect of APOE ε4 on Longitudinal Correlations Between WMH Volume and AD Biomarkers

Cognitively normal participants carrying at least 1 APOE ε4 allele had a faster annual increase (p = 0.0383) in WMH (slope = 0.0964 per year in Z score, SE = 0.0143), in comparison with APOE ε4–negative participants (slope = 0.0607 per year, SE = 0.0093). Similarly, carrying at least 1 APOE ε4 allele was associated with faster rates of change (in magnitude) in all the AD biomarkers except for CSF t-Tau, MRI cortical thickness, and cognition.

The annual rate of change in WMH was negatively correlated with the rates of change in MRI biomarkers (hippocampal volume, cortical thickness) and positively correlated with that of PiB precuneus SUVR and cortical mean SUVR for both APOE ε4–negative and APOE ε4–positive participants (with a trend for PiB cortical mean SUVR among the negative), as summarized in Table 4. No differences were observed by APOE ε4 status in the longitudinal correlations of WMH with the AD biomarkers.

Table 4.

Estimated Annual Rates (Slope) of Change (in Z Score) for All AD Markers and Their Correlations (r) With Longitudinal Change for WMH as a Function of APOE ε4 Status (Positive vs Negative)

Discussion

Accumulating evidence from mostly cross-sectional studies suggests that, in late-onset AD, WMH predict dementia^3,4 and cognitive impairment,^1,4-8 accompany gray matter loss^7,8 and brain atrophy,^7,10,11 correlate with established AD biomarkers,¹² and are associated with the risk of AD progression.^12,13 There have been few longitudinal studies of WMH; these have all focused on the association with brain structural changes and cognitive outcomes only. They found that WMH volume was associated with subsequent cortical atrophy among non-Hispanic Black older adults²⁹ and linked the increase in WMH to more rapid cortical thinning and cognitive decline.^30,31 Neuropathologic studies have found that, at end of life, AD neuropathologic changes and vascular pathology were the main pathologic correlates of dementia, and most patients had mixed pathologies.^4,32 A recent clinicopathologic study³³ found a direct association between WMH and increased odds of having AD neuropathology.

In this longitudinal study of 371 cognitively normal individuals whose baseline age spanned from 19.6 to 88.2 years, we first characterized the longitudinal trajectory of WMH volume by identifying an inflection point of 60.46 years at baseline. We found a dramatically accelerated longitudinal change in not only WMH but also in almost all AD biomarkers among the older participants, in comparison with the younger participants. Of importance, although the longitudinal increase in WMH was not statistically significant in the younger subcohort, there were already significant longitudinal changes for almost all CSF and imaging biomarkers. This suggests that AD pathology already started to accumulate in the brain among cognitively normal individuals before WMH started to increase, consistent with that observed in a previous report.²

We found no significant differences in the longitudinal correlations of WMH with all the AD biomarkers between the younger and older participants and significant longitudinal correlations between WMH and PiB SUVR, MRI structural outcomes, and cognition (but not CSF biomarkers) from the entire cohort. Some correlations (between WMH and MRI markers and cognition) were numerically stronger among younger participants, likely due to the fact that variability of WMH trajectories increased dramatically among the older individuals, as demonstrated by Figure 2. Given a significant correlation between CSF Aβ42/Aβ40 and PiB cortical SUVR (r = −0.48, p < 0.0001), the discrepancy between CSF and PiB SUVR in our findings is likely due to the large variation of the CSF biomarkers across participants and hence reduced statistical power, in comparison with PET amyloid measures. The annual rates of change in WMH and the cognitive composite were already negatively correlated in the younger subcohort, suggesting that, although the average practice effect in cognition and increase in WMH were subtle over time, considerable heterogeneity remained, and participants with larger increase in WMH were associated with attenuated practice effect or even cognitive decline. This is in contrast to our findings that, although many of the CSF and PiB PET imaging biomarkers already exhibited significant longitudinal changes in the younger cohort, their longitudinal correlations with WMH were not significant. One possibility for this inconsistency is lack of statistical power in the younger cohort, especially given that we did observe significant longitudinal correlations between WMH and PiB SUVR when evaluating the entire cohort. If the negative findings on the longitudinal relationships between WMH and amyloid and tau biomarkers are true, however, they would suggest independent effect(s) of WMH on early development of AD pathologies among the younger participants, consistent with multiple recent reports.^2,4,34 Hence, the observed longitudinal effect of WMH on cognition in the younger age window may be through different pathways that are not directly related to AD pathologies, likely linked to brain atrophy and structural changes, given the significant longitudinal correlations between WMH and MRI structural outcomes identified in this study. Larger studies on cognitively normal individuals younger than 60.46 years are needed to further address these important questions.

It is important to distinguish cross-sectional and longitudinal correlations between WMH and AD biomarkers. At baseline, almost all AD biomarkers were cross-sectionally correlated with WMH, consistent with multiple recent reports.^11,32 However, after we adjusted for the effects of age and other covariates, many were no longer statistically significant, indicating that the cross-sectional correlations may be largely attributed to the wide age span and age effect in our cohort. On the contrary, the significant longitudinal correlations we observed remained largely consistent and significant even after adjusting for the well-known covariates including baseline age. Longitudinal correlations can allow within-subject change of WMH volume to be linked to the within-subject change of AD biomarkers, whereas cross-sectional correlations cannot represent the association in within-subject changes. Our results highlight the importance of adjusting for effects of important covariates in these analyses. Of importance, correlations reflect only co-occurrence but not causal relationships. Future longitudinal data with a sufficiently long follow-up may warrant more complex modeling for better understanding of the relationships.

Our findings of longitudinal correlations of WMH with multiple AD biomarkers and cognition are consistent with several recent reports suggesting that WMH are subject to many different brain spatial patterns and lesion sizes, and some may be even directly linked to AD (through AD-related and aging-related WMH) beyond vasculature (cerebral disease–related and cerebral amyloid angiopathy–related WMH).^35-37 Our findings hence support the concept that WMH, which have multiple risk factors that are modifiable,³⁶ offer another target for preventing cognitive decline and AD. The associations between WMH and vascular risk factors have been well reported, and our own data also showed a highly significant correlation between a cardiovascular risk score³⁴ and WMH (r = 0.42, p < 0.0001). Multiple epidemiologic studies suggested a link between accumulating vascular risk factors early in life to the severity of WMH in later life.³⁸ For instance, smoking was associated with elevated WMH³⁹ from a large study of the UK Biobank samples, and intensive blood pressure control slowed down white matter progression.⁴⁰ Smoking, blood pressure, and other vascular risk factors may exert their effects through endothelial damage, oxidative stress, and inflammatory mechanisms and may have potential for dementia prevention due to their modifiable nature.^1,41 In fact, WMH have already been targeted in multiple randomized trials for reducing the risk of dementia. However, both the EVA trial⁴¹ and the PreDIVA trial^42,43 reported reduced WMH progression in the arm with intensive vascular care, but no difference in global cortical atrophy or dementia risk,⁴⁴ compared with that in the control arm. The reasons may be multifold. For example, it may be too late to target WMH in patients with a definite AD diagnosis, especially given that our study demonstrated the profound AD biomarker and WMH changes already occurring among cognitively normal participants, and the region/size of WMH lesions may also need to be considered in designing and analyzing these trials.^45,46

Many open questions are to be answered before WMH can be effectively targeted for reducing the risk of AD dementia⁴: what interventions are appropriate? when is the appropriate age window for intervention? what population should be targeted? Our findings shed light on these important questions and have significant implications on the design and analyses of prevention trials for AD. First, because AD, similar to many biological processes, reflects a complex multistep combination of physiologic events, AD treatments targeting amyloid and tau pathology may need to be combined with drugs for WMH and vascular diseases for a better chance of trial success. Our findings of longitudinal correlations between WMH and multiple AD biomarkers and cognition support the use of treatments that improve vascular health. Second, the findings that most amyloid and tau biomarkers from CSF and PET imaging already showed a significant longitudinal change in the younger cohort suggest that primary prevention trials of AD may enroll individuals younger than 60.46 years. Our estimated annual rates of change in the AD biomarkers may serve as pilot data for designing such trials. The observed longitudinal correlation of WMH with PiB amyloid SUVR suggests that WMH may serve as a key secondary efficacy endpoint in these trials. Third, the much accelerated longitudinal increase in WMH among participants older than 60.46 years and the significant longitudinal correlations with brain atrophy change and MRI structural measures suggest that secondary prevention trials, which often aim at cognitive efficacy, may have the best chance of success with participants older than 60.46 years. Given that brain atrophy and structural changes have been reported to be “closer” to, and longitudinally correlated with the cognitive change,^14,47 secondary prevention trials may consider WMH as a major stratum in treatment randomization and adjust for its effect in efficacy analyses, consistent with previous recommendations.⁴⁸ Fourth, we observed a significant practice effect on the cognitive composite in addition to a significant longitudinal correlation between WMH and cognition. For secondary prevention trials, the longitudinal practice effect of cognitive tests continues to present a challenge to the design and analyses.^49,50 Our findings suggest that the inclusion and exclusion criteria of future secondary prevention trials of AD may consider WMH and lesions, in addition to amyloid and tau, to ensure cognitive decline in the placebo arm.

Major strengths of the study include a longitudinal design, a large cohort whose baseline age spanned almost the entire adult life span, and availability of longitudinal data on both WMH and all major AD biomarkers. Our focus on cognitively normal individuals at risk of AD ensures that results may inform future prevention trials, and the use of advanced BLMM allowed potentially bidirectional relationships between WMH and AD biomarkers. Our study also has several limitations. The longitudinal follow-up of the biomarkers and WMH was relatively short. The participants were predominantly White, and thus, our findings may not generalize to the broader and diverse populations. The retrospective nature of our study dictated that not all relevant data (e.g., periventricular or deep WMH, history of alcohol/drug abuse, major depression, or other comorbid neurodegenerative diseases, CVD, and treatments) were collected and available for analyses. In particular, important cerebral small vessel disease radiologic markers (e.g., cerebral microbleeds, lacunes, dilated perivascular spaces) are not readily available, preventing us from assessing their impact. Informative dropout due to reasons other than AD remains a possibility, which we did not investigate. Another limitation is the use of Z scores in the cognitive composite that ignored the likely different distributions across tests.

In summary, we identified an accelerated longitudinal increase of WMH around the baseline age of 60.46 years and estimated the longitudinal rates of change of WMH and major AD biomarkers and cognition for both younger and older participants. We comprehensively characterized the longitudinal correlations in the annual rates of change of WMH and all major established AD biomarkers. These results provide support for WMH as a potential target in future AD prevention trials and inform the design and analyses of these trials involving WMH.

Glossary

Aβ

β-amyloid

ACS

Adult-Children Study

AD

Alzheimer disease

ADRC

AD Research Center

AIC

Akaike information criterion

BLMM

bivariate LMM

BMI

body mass index

CVD

cerebrovascular disease

DIAN

Dominantly Inherited Alzheimer Network

HASD

Healthy Aging and Senile Dementia

LMM

linear mixed-effects model

PiB

Pittsburgh compound B

pTau181

tau phosphorylated at 181

ROI

region of interest

SUVR

standardized uptake value ratio

t-Tau

total tau

WMH

white matter hyperintensities

WUSM

Washington University School of Medicine

Appendix 1. Authors

Appendix 2. Coinvestigators

Study Funding

This study was supported by the National Institute on Aging (NIA) grant R01 AG053550 (C. Xiong) and NIA grants P50 AG005681, P01AG026276, and P01 AG0399131 (J.C. Morris) and UF1AG032438 (R.J. Bateman). Image processing was supported partly by the Neuroimaging Informatics and Analysis Center (1P30NS098577) and R01 EB009352 (T.L.S. Benzinger). G.S. Day's research is supported by NIH (K23AG064029, U01AG057195, U19AG032438), the Alzheimer's Association, and Chan Zuckerberg Initiative. Data collection and sharing for this project was supported by The Dominantly Inherited Alzheimer Network (DIAN, U19AG032438) funded by the National Institute on Aging (NIA), the Alzheimer Association (SG-20-690363-DIAN), the German Center for Neurodegenerative Diseases (DZNE), Raul Carrea Institute for Neurological Research (FLENI), Partial support by the Research and Development Grants for Dementia from Japan Agency for Medical Research and Development, AMED, and the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), Spanish Institute of Health Carlos III (ISCIII), Canadian Institutes of Health Research (CIHR), Canadian Consortium of Neurodegeneration and Aging, Brain Canada Foundation, and Fonds de Recherche du Québec–Santé. This article has been reviewed by DIAN Study investigators for scientific content and consistency of data interpretation with previous DIAN Study publications.

Disclosure

The authors report no disclosures relevant to the manuscript. The following disclosures should be noted. J.C. Morris consults for Barcelona Brain Research Center (BBRC) Int'l Advisory Board, TS Srinivasan-NIMHANS Knowledge Conclave (Chennai, India) International Advisory Board; Center for Brain Research (Bangalore, India) Native Alzheimer Disease-Related Resource Center in Minority Aging Research, Ext Adv Board, and Cure Alzheimer Fund, Research Strategy Council Diverse VCID Observational Study Monitoring Board LEADS Advisory Board, Indiana University, and received honoraria from Montefiore Grand Rounds, NY Tetra-Inst ADRC seminar series, Grand Rds, NY. A.M. Fagan has consulted for Biogen, Centene, Fujirebio, and Roche Diagnostics, is a member of the scientific advisory boards for Roche Diagnostics, Genentech, and AbbVie, and consults for Araclon/Grifols, Diadem, and DiamiR. T.L.S. Benzinger received research support from Avid Radiopharmaceuticals and participates in clinical trials sponsored by Avid Radiopharmaceuticals, Eli Lilly, and Hoffman-LaRoche and has received honoraria from Biogen, Eisai, and Genetech. R.J. Bateman has equity ownership interest in C2N Diagnostics and may receive income based on technology licensed or optioned by Washington University to C2N Diagnostics, receives income from C2N Diagnostics for serving on the scientific advisory board, received honoraria as a speaker/consultant/advisory board member from Amgen, Eisai, and Roche. B. Gordon participates in clinical trials sponsored by Avid Radio-pharmaceuticals (a wholly owned subsidiary of Eli Lilly), Eli Lilly, and Hoffman-LaRoche. G.S. Day is supported by the NIH/National Institute on Aging (K23AG064029), serves as a topic editor on dementia for DynaMed Plus (EBSCO Industries, Inc.), is a consultant for Parabon NanoLabs, is the clinical director for the Anti-NMDA Receptor Encephalitis Foundation (uncompensated), has provided record review and expert medical testimony on legal cases pertaining to the management of Wernicke encephalopathy, and holds stocks (>$10,000) in ANI Pharmaceuticals (a generic pharmaceutical company). C. Xiong consults for Diadem. There are no conflicts with this work. Go to Neurology.org/N for full disclosures.