Alzheimer’s disease genetic risk and changes in brain atrophy and white matter hyperintensities in cognitively unimpaired adults

Abstract Reduced brain volumes and more prominent white matter hyperintensities on MRI scans are commonly observed among older adults without cognitive impairment. However, it remains unclear whether rates of change in these measures among cognitively normal adults differ as a function of genetic risk for late-onset Alzheimer’s disease, including APOE-ɛ4, APOE-ɛ2 and Alzheimer’s disease polygenic risk scores (AD-PRS), and whether these relationships are influenced by other variables. This longitudinal study examined the trajectories of regional brain volumes and white matter hyperintensities in relationship to APOE genotypes (N = 1541) and AD-PRS (N = 1093) in a harmonized dataset of middle-aged and older individuals with normal cognition at baseline (mean baseline age = 66 years, SD = 9.6) and an average of 5.3 years of MRI follow-up (max = 24 years). Atrophy on volumetric MRI scans was quantified in three ways: (i) a composite score of regions vulnerable to Alzheimer’s disease (SPARE-AD); (ii) hippocampal volume; and (iii) a composite score of regions indexing advanced non-Alzheimer’s disease-related brain aging (SPARE-BA). Global white matter hyperintensity volumes were derived from fluid attenuated inversion recovery (FLAIR) MRI. Using linear mixed effects models, there was an APOE-ɛ4 gene-dose effect on atrophy in the SPARE-AD composite and hippocampus, with greatest atrophy among ɛ4/ɛ4 carriers, followed by ɛ4 heterozygouts, and lowest among ɛ3 homozygouts and ɛ2/ɛ2 and ɛ2/ɛ3 carriers, who did not differ from one another. The negative associations of APOE-ɛ4 with atrophy were reduced among those with higher education (P < 0.04) and younger baseline ages (P < 0.03). Higher AD-PRS were also associated with greater atrophy in SPARE-AD (P = 0.035) and the hippocampus (P = 0.014), independent of APOE-ɛ4 status. APOE-ɛ2 status (ɛ2/ɛ2 and ɛ2/ɛ3 combined) was not related to baseline levels or atrophy in SPARE-AD, SPARE-BA or the hippocampus, but was related to greater increases in white matter hyperintensities (P = 0.014). Additionally, there was an APOE-ɛ4 × AD-PRS interaction in relation to white matter hyperintensities (P = 0.038), with greater increases in white matter hyperintensities among APOE-ɛ4 carriers with higher AD-PRS. APOE and AD-PRS associations with MRI measures did not differ by sex. These results suggest that APOE-ɛ4 and AD-PRS independently and additively influence longitudinal declines in brain volumes sensitive to Alzheimer’s disease and synergistically increase white matter hyperintensity accumulation among cognitively normal individuals. Conversely, APOE-ɛ2 primarily influences white matter hyperintensity accumulation, not brain atrophy. Results are consistent with the view that genetic factors for Alzheimer’s disease influence atrophy in a regionally specific manner, likely reflecting preclinical neurodegeneration, and that Alzheimer’s disease risk genes contribute to white matter hyperintensity formation.


Introduction
Alzheimer's disease (AD) pathology and neurodegeneration, as measured by atrophy on magnetic resonance imaging (MRI), are present many years prior to the emergence of clinical symptoms when individuals are cognitively normal. 1,2lder adults without cognitive impairment also frequently have evidence of small vessel cerebrovascular disease, which most commonly manifests as white matter hyperintensities (WMHs) on MRI scans. 3Recent evidence suggests that WMH may also play a role in Alzheimer's disease, 4,5 with both vascular and Alzheimer's disease-specific pathways contributing to WMHs. 6,7Both brain atrophy 8,9 and WMH burden 10,11 among individuals with normal cognition have been shown to predict subsequent cognitive decline and impairment.It remains unclear, however, whether rates of brain atrophy and WMH accumulation among cognitively normal adults differ as a function of genetic risk for late-onset Alzheimer's disease and whether this relationship is influenced by other variables, such as age, sex, vascular risk factors and education.This is an important topic for investigation because an examination of non-modifiable and modifiable factors that influence longitudinal changes in atrophy and WMH may help identify ways to reduce brain deterioration and eventual cognitive decline in older persons.
Research among middle-aged and older individuals with normal cognition, however, has primarily included crosssectional studies that cannot address whether observed differences in brain volumes as a function of Alzheimer's disease-genetic risk reflect lifelong differences in brain structure, as opposed to differential atrophy that occurs during the preclinical phase of Alzheimer's disease.Results from crosssectional studies have been mixed, with some finding higher WMH burden 48,49 and lower volumes or thinner cortex 23,[50][51][52][53] among individuals at greater Alzheimer's disease-genetic risk (i.e.][56][57][58][59] For APOE-ɛ2 genetic status, results from cross-sectional studies have also been mixed. 22,52,54,60,61Few prior longitudinal studies have been conducted among cognitively unimpaired older individuals.Of these, two reported greater volume loss in Alzheimer's disease-vulnerable regions among APOE-ɛ4 carriers compared to non-carriers, 62,63 whereas two others found no APOE-ɛ4-related differences. 64,65dditionally, a relatively small study reported less hippocampal atrophy among older cognitively normal APOE-ɛ2 carriers relative to ɛ3 homozygouts, 66 consistent with a study that included individuals across the Alzheimer's disease-spectrum. 33o our knowledge, the relationship of AD-PRS and longitudinal atrophy rates or WMH accumulation have not been examined among middle-aged and older cognitively unimpaired individuals.Likewise, although cross-sectional studies across the AD-spectrum have found higher WMH burden in APOE-ɛ2 carriers relative to ɛ3/ɛ3 homozygotes, 40 the impact of the APOE-ɛ2 allele on longitudinal changes in WMH burden in cognitively normal individuals remains unclear. To address these gaps, the current study examined rates of change in regional brain volumes and WMH in a large, harmonized dataset of middle-aged and older individuals with normal cognition at baseline (mean MRI follow-up = 5.3 years, max = 24 years), with both APOE genotypes (N = 1541) and AD-PRS scores (N = 1093) available.The current study expands on prior ones in several ways.First, prior studies have been characterized by either short follow-up periods (i.e.mean follow-up 2-3.5 years), [63][64][65][66] or by small sample sizes (i.e.N < 110), 62,65,66 limiting their ability to draw inferences regarding less frequent alleles, including ɛ2 carrier status and ɛ4 homozygosity.Second, the large sample size allowed us to examine potential interactions between Alzheimer's disease-genetic risk and other variables in relationship to brain atrophy or WMH accumulation, including age, sex, education, vascular risk and progressor status (i.e.remained cognitively normal versus progressed to MCI or dementia).Third, we examined atrophy rates in three different measures: (i) a composite score of Alzheimer's disease-vulnerable regions derived from machine learning (SPARE-AD); (ii) hippocampal volume; and (iii) a composite score of regions sensitive to non-Alzheimer's disease-related aging (SPARE-BA), also derived from machine learning.This allows for a comparison of the influence of Alzheimer's disease-genetic risk on atrophy in Alzheimer's disease-vulnerable and non-vulnerable regions.Lastly, we examined the impact of AD-PRS on longitudinal brain atrophy and WMH change, as well as interactions between AD-PRS and APOE genotypes on these measures.

Participants
This study used data from the Preclinical Alzheimer's disease Consortium (PAC), a multi-site collaboration established to investigate the earliest phases of Alzheimer's disease.The PAC study includes harmonized cognitive, clinical, genetic, MRI and amyloid imaging data from five on-going longitudinal cohort studies: the Adult Children Study (ACS), 67 the Australian Imaging, Biomarker, and Lifestyle study (AIBL study), 68 the Biomarkers of Cognitive Decline Among Normal Individuals (BIOCARD) study, 69 the Neuroimaging Substudy of the Baltimore Longitudinal Study of Aging (BLSA) 70 and the Wisconsin Registry for Alzheimer's Prevention (WRAP). 71To be included in the PAC data files, each participant had to be cognitively normal at baseline and have at least one molecular biomarker (derived from cerebrospinal fluid or positron emission tomography) collected while they were cognitively normal.By design, at least half of the participants in each cohort, except BLSA, had a family history of dementia.Individuals with epilepsy, recent strokes or remote strokes with residual effects were excluded at baseline.0][71][72] Molecular biomarkers were not considered in the present analyses.Participants in all cohorts provided written informed consent according to the Declaration of Helsinki.The study protocols were approved by each site's local institutional review board.

Clinical and cognitive assessments
Participants in all cohorts undergo longitudinal clinical and cognitive assessments, as well as medical, neurologic and psychiatric evaluations at regular intervals, depending on the protocol for each site (e.g.every 12, 18 or 24 months).The cognitive assessments at each site include a comprehensive neuropsychological battery covering all major cognitive domains (for details, see Gross et al. 73 and Pettigrew et al. 74 ).All sites conduct regular consensus diagnoses for all participants using published criteria, e.g. the National Institute on Aging/Alzheimer's Association criteria for MCI 75 and dementia. 76he diagnostic process for each case is handled in a comparable manner at each site: (i) clinical data are examined pertaining to the medical, neurologic and psychiatric status of the subject; (ii) reports of changes in cognition by the subject and by collateral sources are examined, based on the Clinical Dementia Rating scale; and (iii) change in cognitive performance is established.This information is then used to determine whether the subject has become cognitively impaired, and determine the likely aetiology of the impairment.Clinical diagnoses were made without knowledge of the biomarker measures.To be included in the current analyses, participants had to be cognitively normal at the time of their first MRI scan (which is considered the 'baseline' in these analyses) and have non-missing APOE genetic and vascular risk score data (see below).Participants with a diagnosis of 'impaired not MCI' were included with the cognitively normal participants, as they do not meet criteria for MCI, consistent with prior publications. 69ummary vascular risk scores were calculated using a previously validated method, based on the presence or absence of five vascular risk factors: hypertension, diabetes, obesity (defined as a body mass index > 30 kg/m 2 ), hypercholesterolaemia and smoking within the 30 days prior to data collection. 77This information was obtained from medical history reports or medical records collected at visits coinciding with the MRI visits (±12 months).[80]
Details regarding the generation of the AD-PRS for the PAC dataset have been described previously. 74Briefly, each site generated GWAS data using various genotyping arrays and the raw GWAS data were imputed by chip using a standard pipeline that included variant filtering for genotyping efficiency (95%), minor allele frequency (>1%) and Hardy-Weinberg equilibrium (P > 1 × 10 −6 ).Given the racial and ethnic makeup of the included studies, all GWAS analyses were restricted to those of European ancestry that was confirmed using population principal component analysis.For the purpose of the AD-PRS analysis, we restricted all GWAS datasets to overlapping variants leaving a total of 6 739 456 common variants available in all five datasets for analysis.AD-PRS were generated using imputed GWAS data, leveraging the summary statistics provided by Kunkle et al. 16 that were regenerated for us removing PAC participants who were included in the original GWAS analysis (n = 93 220).AD-PRS were computed with PLINK using a previously published method. 18The current analyses only used AD-PRS without the APOE region (i.e. 1 MB upstream and downstream of the APOE gene) to assess the independent associations of APOE and other Alzheimer's disease risk genes on the MRI measures.AD-PRS were transformed to Z-scores to simplify interpretation, using the mean and standard deviation (SD) across all five datasets (see Pettigrew et al. 74 for the distribution of harmonized AD-PRS in the PAC cohorts).

Image acquisition
All PAC sites have collected structural MRI scans longitudinally, with the majority of scans acquired on 3 T scanners, but a subset on a 1.5 T scanner, since some of the studies began in the mid-1990s.See Supplementary Table 1 for details regarding the types of scanners and acquisition protocols for each site.

Image processing and harmonization
Processing of T 1 -weighted images included correction of intensity inhomogeneities, 82 skull stripping 83 and segmentation of the brain into a set of anatomical regions of interest (ROIs) using the Multi-atlas Region Segmentation Utilizing Ensembles (MUSE) software platform. 84This method was specifically designed for longitudinal studies to handle differences in scanners and imaging protocols over time and across sites and employs harmonized acquisition-specific atlases.For a detailed description of these methods, see Erus et al. 85 and Habes et al. 86 Briefly, MUSE uses a consensus labelling framework that combines an ensemble of labelled atlases in target image space by using multiple atlases reflecting a broad representation of anatomy.Scanner-specific atlases share the same ROI labels, imposing consistency of segmentations, while each atlas set preserves the image intensity characteristics of the specific scanner.The MUSE pipeline has been extensively validated against benchmark methods and applied in various cross-sectional and longitudinal studies. 64,84,86,87In comparison to most commonly used segmentation tools, such as FreeSurfer, MUSE has demonstrated significant improvement in accuracy and more consistent segmentations across scanners, particularly in segmentation of deep brain structures. 88The MUSE software package is freely available: https://www.med.upenn.edu/cbica/sbia/muse.html.
Additional statistical harmonization was applied to the ROI volumes based on the multivariate ComBAT-GAM method 89 to remove cohort-related effects and protocolspecific variability.This method simultaneously models scanner effects (unwanted sources of variation) and covariate associations (e.g.age and sex).This harmonization approach integrates a generalized additive model, with a smoothed non-linear term for age, using thin plate regression splines, and linear terms for sex and intra-cranial volume (ICV), thereby preserving age and sex differences across sites. 89uantification of WMH volumes from fluid attenuated inversion recovery (FLAIR) images was completed using an automated deep learning based segmentation method 90 that is built upon the UNet architecture, 91 with the convolutional network layers replaced by an Inception ResNet architecture. 92The network model uses inhomogeneity corrected and co-registered FLAIR and T 1 -weighted images as input, and has been trained using a multi-site training dataset with human-validated WMH labels, as published previously. 86The algorithm was applied to MRI scans of PAC participants to calculate binary WMH masks, and to extract regional WMH volumes.The current analyses used global WMH volumes.

Volumetric regions of interest and spatial patterns of atrophy
Harmonized volumes of the left and right hippocampus were normalized for head size by regressing the average of the left and right hemispheres on ICV.The standardized residuals (mean = 0, SD = 1) were used in analyses presented below.Hippocampal volumes were examined to enable direct comparison to many prior studies that have specifically focused on this structure.
Atrophy in regions vulnerable to Alzheimer's disease was measured using SPARE-AD scores, which represent an imaging signature of Alzheimer's disease-like neurodegeneration derived from machine learning, as previously described and validated. 28,93For SPARE-AD calculation, a support vector machine classifier with a linear kernel was trained to maximally differentiate between cognitively unimpaired participants and participants with AD-dementia, using a curated dataset of over 10 000 individuals, known as the iSTAGING consortium, 86 which includes the PAC sites.More positive SPARE-AD scores imply a more Alzheimer's disease-like brain structure (i.e. more AD-related atrophy).
We also calculated a brain signature of age-related brain atrophy, using SPARE-BA scores, to estimate structural brain changes due to aging.As published previously, this MRI approach uses a multivariate pattern regression method based on support vector regression to calculate brain aging scores for each participant. 86,94The model was trained with the T 1 -MR scans using harmonized ROI volumes for structures.In the present analyses, SPARE-BA scores were regressed on age at scan and the standardized residuals (referred to in the tables as SPARE-BA-resid) were used in the analyses presented below, with more positive scores indicating greater age-related atrophy compared to normative trends.This is comparable to 'brain age gap' scores estimated using related techiques. 95,96The regions contributing to the SPARE-BA scores are weighted optimally for estimating age, whereas the regions contributing to SPARE-AD scores are weighted optimally to distinguish between cognitively normal individuals and individuals with Alzheimer's diseasedementia.See Supplementary Fig. 1 for illustrations of the SPARE-AD and SPARE-BA masks and Supplementary Text 1 for additional information on SPARE-AD and SPARE-BA.

Statistical analyses
We used linear mixed effects models with random intercepts and slopes with an unstructured covariance to evaluate whether the trajectories of the MRI measures of atrophy (i.e.SPARE-AD and SPARE-BA), hippocampal volume or WMH burden differed based on AD-genetic risk.Separate models were run for each MRI outcome measure.The primary models evaluating APOE effects included the following predictors: baseline age, sex, education, dichotomous indicators for APOE-ɛ2 and APOE-ɛ4 (with ɛ3/ɛ3 as the reference group), indicators for site (to control for potential site differences), time and the interaction (cross-product) of each predictor with time.The primary models evaluating AD-PRS were the same as the APOE models, but additionally included the AD-PRS.The years of education variable was standardized to Z-scores separately for cohorts within versus outside the USA, given differences in the number of years of compulsory schooling.Final models examining the trajectories of the atrophy measures and hippocampal volume included a time 2 term to account for their statistically significant non-linear change over time.For the WMH models, the time 2 was not significant and therefore not included.The primary APOE models were re-run with APOE-ɛ4 status coded categorically (as described above) to evaluate whether the MRI trajectories differed between APOE ɛ3/ ɛ4 and ɛ4/ɛ4 carriers.In a sensitivity analysis, the primary models were also re-run excluding APOE ɛ2/ɛ4 carriers.
A second set of linear mixed effects models evaluated whether the results remained the same when additionally covarying both vascular risk summary scores (using all available measures over time) and participants' follow-up diagnostic status, based on their last (i.e.most recent) consensus diagnosis (coded as 0 = remained normal, or 1 = progressed to MCI or dementia).These models were identical to the primary models, but additionally included the vascular risk summary scores over time and binary indicators for progressor status, as well as their interactions with time.
Lastly, to evaluate whether the relationships between the Alzheimer's disease-genetic factors and the MRI measures were modified by the demographic or clinical variables, a third set of linear mixed effects models were run.These models were the same as the primary models, but additionally included three-way interaction terms for the genetic factors × demographic/clinical variable × time (e.g.APOE-ɛ2 × baseline age × time and ɛ4 × baseline age × time; or AD-PRS × baseline age × time), as well as the corresponding lowerorder interaction terms.These models were not adjusted for multiple comparisons because they were exploratory in nature and correcting for multiple comparisons in exploratory analyses can increase the likelihood of Type II errors and potentially obscure meaningful findings.
Data analysis was performed using STATA 17.0 and P-values of <0.05 were considered significant.

Results
Table 1 shows baseline characteristics of participants included in the MRI volumetric and WMH analyses, separately for participants in the APOE and AD-PRS analyses.For baseline characteristics by cohort, see Supplementary Tables 2 and 3. On average, participants were in their mid-60s at baseline, primarily White and highly educated.About one-third of participants were APOE-ɛ4 carriers and approximately two-thirds had one or more vascular risk factor.The mean number of volumetric MRI measures over time was 3 (max = 18), with a mean 5.3 years between the first and last MRI scan (max = 24 years).Out of the 1541 participants with volumetric data, 1348 also had one or more WMH measure (mean number of measures over time = 2, max = 10; mean time between baseline and last WMH measure = 2.9 years, max = 19 years).

APOE genotypes and MRI trajectories
Results from the primary model examining the binary APOE-ɛ2 and ɛ4 indicators in relationship to trajectories of the MRI measures are shown in Table 2 (with ɛ3/ɛ3 as the reference).APOE-ɛ4 carrier status was not associated with any MRI measure at baseline (all P ≥ 0.14).However, relative to ɛ3/ɛ3 carriers, ɛ4 carriers demonstrated greater increases in SPARE-AD (P < 0.001) scores and greater decreases in hippocampal volume (P ≤ 0.001) over time; they also showed greater increases in SPARE-BA (P = 0.025), though the effect appeared smaller (Z = 2.25 versus Z = 3.45 for SPARE-AD; see Supplementary Text 2 for formal model comparison).APOE-ɛ2 carrier status was also not associated with any baseline MRI measure (all P ≥ 0.12), but ɛ2 carriers showed greater increases in WMH volumes over time (P = 0.014), compared to ɛ3/ɛ3 carriers.There was no association between ɛ2 carrier status and rate of change in the other MRI measures (all P > 0.3).
Additionally, across all models (Table 2), older age was associated with higher SPARE-AD scores, smaller hippocampal volumes and greater WMH volumes at baseline and greater increases in SPARE-AD, SPARE-BA scores and WMH volumes over time, as well as greater decreases in hippocampal volume over time (all P < 0.0001).Participant sex was not associated with any baseline MRI measure, but females had smaller increases in SPARE-AD (P = 0.009) and SPARE-BA (P < 0.0001) scores over time than males.Years of education was not associated with the baseline or rate of change in any MRI measure (all P ≥ 0.17).These results remained the same when excluding APOE-ɛ2/ɛ4 carriers from the analysis (data not shown), except that the association between APOE-ɛ4 carrier status and rate of change in SPARE-BA was no longer significant (estimate = 0.010, SE = 0.005, P = 0.069).
The pattern of results was also the same when additionally adjusting for vascular risk scores and follow-up diagnosis (see Supplementary Table 4).In these models, higher vascular risk scores were associated with higher baseline SPARE-AD scores (P = 0.026), but were not related to the rate of change in any MRI measure (all P ≥ 0.05).Additionally, participants who progressed to MCI/dementia over time had higher baseline WMH volumes (P = 0.012), greater increases in WMH volumes (P = 0.012) and SPARE-AD scores (P < 0.0001) over time, and greater declines in hippocampal volume (P < 0.0001) after also adjusting for vascular risk and follow-up diagnosis.
Results from models examining whether the demographic and clinical variables modified the relationships between APOE genotypes and the MRI trajectories are shown in Table 3.The associations of APOE-ɛ4 genetic status with rates of change in SPARE-AD scores and hippocampal volume were modified by baseline age and years of education (all P for interaction terms of ɛ4 × (age or education) × time ≤ 0.039).Specifically, ɛ4 related atrophy in SPARE-AD regions and the hippocampus was greater among older participants and weaker among those with more years of education.These three-way interactions remained significant when excluding APOE ɛ2/ɛ4 carriers (all P ≤ 0.012) and are illustrated in Fig. 2. Additionally, the relationship between ɛ4 genetic status and the rate of decline in hippocampal volume was greater among participants with higher compared to lower vascular risk scores (P = 0.038) and among those who progressed to MCI or dementia over time compared to those who remained cognitively unimpaired (P = 0.001), see Table 3.However, the three-way interactions with vascular risk scores or progressor status were not significant when excluding APOE ɛ2/ɛ4 carriers (both P > 0.21).Among APOE-ɛ2 carriers, higher education was unexpectedly associated with greater increases in SPARE-AD over time (P = 0.032); however, this interaction was also not significant after exclusion of APOE ɛ2/ɛ4 carriers (P > 0.15).

Alzheimer's disease-polygenic risk score and MRI trajectories
In the primary models, there was no association between AD-PRS scores and baseline levels of the MRI measures (all P ≥ 0.12).By contrast, higher AD-PRS scores were associated with greater increases over time in SPARE-AD scores, greater decreases in hippocampal volume and greater increases in WMH volumes (all P ≤ 0.035, see Table 4 and Fig. 3).Results were similar using a dichotomous AD-PRS (see Supplementary Table 5).The associations between the AD-PRS score and rate of change in the MRI measures were independent of APOE genetic status and were not modified by age, sex, years of education, vascular risk scores or progressor status (see Table 3).Lastly, we explored potential interactions between the AD-PRS scores and APOE-ɛ4 and ɛ2 genetic status in relationship to change in the MRI measures.To simplify interpretation, APOE-ɛ2/ɛ4 carriers were excluded from these analyses.There were no interactions between the AD-PRS score and APOE-ɛ4 or APOE-ɛ2 genetic status with respect to the rate of change in SPARE-AD, SPARE-BA and hippocampal volume (all P > 0.47).However, for WMH volumes, there was an AD-PRS × APOE-ɛ4 × time interaction (P = 0.038, see Supplementary Table 6), suggesting a stronger association between APOE-ɛ4 genetic status and rate of increase in WMH among participants with higher compared to lower AD-PRS scores (Fig. 3C).

Discussion
The large sample size and substantial follow-up period of the current study provide the basis for several new insights on the relationship between genetic risk factors for late-onset Alzheimer's disease among cognitively normal individuals and changes in MRI measures of brain atrophy and WMH.First, the atrophy rates in a composite volume measure of Alzheimer's disease-vulnerable regions (SPARE-AD) and the hippocampus demonstrated an APOE-ɛ4 gene-dose effect, with greatest atrophy among ɛ4 homozygous participants, followed by ɛ4 heterozygous participants, and least among ɛ4 non-carriers.Second, both APOE-ɛ4 status and AD-PRS scores independently influenced rates of change in AD-vulnerable regions and the hippocampus, suggesting additive effects.Third, the negative impact of APOE-ɛ4 on atrophy in AD-vulnerable regions and the hippocampus was reduced among individuals with higher education and younger baseline ages.Fourth, ɛ4 carrier status was associated with greater increases in global WMH volumes over time, particularly among ɛ4 homozygous participants and those with high Alzheimer's diesease-polygenic risk scores.Fifth, ɛ2 carrier status did not influence atrophy rates in regions sensitive to aging or Alzheimer's disease, but was associated with greater increases in WMH volumes over time.In contrast, neither APOE nor AD-PRS scores showed robust associations with atrophy in a composite measure of regions sensitive advanced non-Alzheimer's disease-related brain aging (SPARE-BA), supporting prior evidence that this measure largely reflects age but not disease-related atrophy.Taken together, these results underscore the impact of AD-genetic risk factors on rates of change in MRI measures of neurodegeneration among middle-aged and older adults with normal cognition and point to potential interactions and synergistic effects between APOE and other Alzheimer's disease risk genes on WMH burden.

APOE-e4 genetic status and brain atrophy
The current results are consistent with prior longitudinal studies among participants across the clinical spectrum of Alzheimer's disease 33 and non-demented cohorts [30][31][32] that have reported elevated longitudinal atrophy in Alzheimer's disease-vulnerable regions among ɛ4 carriers compared to non-carriers.Our findings are also in line with, and extend prior work, among individuals with normal cognition 62,63,97 by documenting an APOE-ɛ4 gene-dose effect on atrophy rates in the hippocampus and a composite of Alzheimer's disease-vulnerable regions.This gene-dose effect likely reflects the fact that Alzheimer's disease pathology (i.e.][100] It is hypothesized that this earlier age of amyloid accumulation likely initiates an earlier onset of AD-related atrophy in selected brain regions.This is consistent with the view that subtle Alzheimer's disease-related atrophy begins during the preclinical phase of the disease, when individuals are cognitively normal. 8,9,101The lack of volumetric differences between ɛ4 carriers and non-carriers at baseline supports the view that APOE-ɛ4 primarily influences brain volumes during the preclinical phase of Alzheimer's disease, but has limited impact on volumes prior to midlife, though such differences have been demonstrated. 26,102Our results also demonstrate that APOE-ɛ4 related differences in hippocampal atrophy appear to be particularly evident among individuals who progress to MCI or dementia over time, in line with a prior study. 97The finding that ɛ4 was only weakly associated with atrophy in non-Alzheimer's disease regions that are sensitive to aging (SPARE-BA) underscores the specificity of ɛ4 to Alzheimer's disease-related atrophy and might reflect the fact that in some individuals, Alzheimer's disease pathology begins in more atypical regions.The rate of change in AD-related atrophy (estimate = 0.033, SE = 0.015, P = 0.028) and hippocampal volumes (estimate = 0.026, SE = 0.011, P = 0.014) was greater for individuals with two ɛ4 alleles compared to those with one ɛ4 allele, who in turn showed more atrophy than ɛ3/ɛ3 carriers (estimate = 0.017, SE = 0.006, P = 0.008 for SPARE = AD and (estimate = 0.013, SE = 0.004, P = 0.003 for hippocampus).Atrophy rates did not differ between ɛ2 carrier and ɛ3/ɛ3 carriers (all P > 0.3).WMH volumes (D) increased more over time among both ɛ4/ɛ4 carriers (estimate = 0.029, SE = 0.014, P = 0.039) and ɛ2 carriers (estimate = 0.015, SE = 0.008, P = 0.049) relative to ɛ3/ɛ3 carriers (see text for details).
Our results also showed that the association between APOE-ɛ4 genetic status and rate of atrophy in the hippocampus and the Alzheimer's disease-vulnerable regions increases with advancing age (Fig. 2C and D), in line with the age-related increase in Alzheimer's disease pathology accumulation 98 and the age-related increase in Alzheimer's disease-related cognitive impairment.7][58] We found no evidence that the association between APOE-ɛ4 and atrophy in AD-vulnerable regions or regions sensitive to aging (SPARE-BA) was influenced by overall levels of vascular risk, though for the hippocampus, higher vascular risk scores were associated with a stronger relationship between APOE-ɛ4 genetic status and atrophy over time (but this was not significant when excluding ɛ2/ɛ4 carriers).Although vascular risk factors have been consistently linked with smaller regional brain volumes 28,104,105 and atrophy rates, 47,97 little is known about whether APOE variants moderate this relationship.Additionally, the relationship between vascular risk and brain atrophy may differ by other factors, such as amyloid burden 106 and by type of vascular risk factor (e.g.hypertension versus obesity). 64Thus, summary scores using different risk factors, or differentially weighted risk factors, may potentially show stronger associations with brain atrophy than was observed here.and D).The negative effect of APOE-ɛ4 genetic status on rate of atrophy in Alzheimer's disease-vulnerable regions and the hippocampus was greater among older than young participants (C and D), as indicated by significant three-way interactions of ɛ4 × age × time for SPARE-AD (estimate = 0.002, SE = 0.001, P = 0.001) and for the hippocampus (estimate = −0.001,SE = 0.0004, P = 0.027), but was attenuated among individuals with more years of education (A and B), as indicated by significant ɛ4 × education × time interactions for SPARE-AD (estimate = −0.015,SE = 0.006, P = 0.017) and for the hippocampus (estimate = 0.009, SE = 0.004, P = 0.039).For illustration purposes, the 25th and 75th percentiles of the baseline education were used to show trajectories of high versus low education.
In this study, we also observed that years of education modified the association between APOE-ɛ4 genetic status and atrophy in the Alzheimer's disease-vulnerable regions composite and the hippocampus, such that participants with more years of education had less ɛ4-related atrophy than those with less education.These findings are in line with a recent study also using data from the PAC cohort (N = 1819), which reported that higher scores on a composite measure of years education and literacy attenuated the negative effect of APOE-ɛ4 genotype on the rate of decline in episodic memory and a global cognitive score. 74he present results suggest that this reduction in APOE-ɛ4 related cognitive decline among participants with more education may be mediated by reduced atrophy in Alzheimer's disease-vulnerable regions.Previous crosssectional studies have produced mixed results regarding the association of years of education and regional brain volumes among middle-aged and older cognitively unimpaired individuals (e.g.Launer et al., 107 Arenaza-Urquijo et al., 108 Liu et al. 109 and Vemuri et al. 110 ).2][113] The present results suggest that some inconsistencies across prior studies may be attributable to the fact that the education-related reduction in atrophy is relatively small and primarily evident among APOE-ɛ4 carriers, making it difficult to detect in studies with smaller samples.

APOE-ɛ2 genetic status and brain atrophy
Another important finding is that APOE-ɛ2 carriers demonstrated similar rates of brain atrophy over time as ɛ3 homozygous individuals and did not differ in terms of brain volumes at baseline.Given the relatively low prevalence of the ɛ2 allele, prior work on this subject has been limited and largely comprised of cross-sectional studies with small numbers of ɛ2 carriers (ranging from ∼12 to 85 compared to 184 in this study).Our results are consistent with two prior cross-sectional studies among cognitively normal individuals that also found no difference in volumetric measures as a function of ɛ2 carrier status among middle-aged and older adults. 54,61By comparison, a small-scale longitudinal study 66 and a few other cross-sectional studies 52,60,114 reported reduced 2-year atrophy and greater cortical thickness and volumes among older cognitively normal ɛ2 carriers compared to ɛ3 homozygotes in regions sensitive to Alzheimer's disease.A likely explanation for these discrepancies across studies is that ɛ2 carriers are less likely to harbour preclinical Alzheimer's disease pathology (due to a later age of amyloid accumulation).Consequently, they are less likely to have atrophy in Alzheimer's disease-sensitive regions during middle-and old age compared to ɛ3/ɛ3 carriers, which can appear as reduced atrophy or greater volume.Future studies will be able to test this possibility by covarying amyloid and tau burden when evaluating associations between APOE-ɛ2 status and atrophy.

APOE-ɛ2 and ɛ4 genetic status and white matter hyperintensities
Prior studies among non-demented participants as well as samples spanning the Alzheimer's disease-spectrum have reported higher WMH burden [37][38][39][40] and greater longitudinal increases in WMH burden 42,43 among APOE-ɛ4 carriers relative to carriers, with stronger associations for homozygous than heterozygous participants 38 (but see Habes et al., 44 Lyall et al., 45 Lane et al. 46 and Debette et al. 47 for negative results).Few studies, however, have examined ɛ4-related differences in WMH volumes among individuals with normal cognition. 48,49,115The present study found greater longitudinal increases in global WMH volumes among ɛ4 homozygous compared to ɛ4 heterozygous participants and ɛ3/ɛ3 carriers, who did not differ from one another.These findings are consistent with a cross-sectional study among cognitively normal middle-aged participants. 49iven that cognitively unimpaired middle-aged and older ɛ4/ɛ4 carriers likely harbour the highest level of brain amyloid, 98 these findings support the view that Alzheimer's disease-specific pathways contribute to the formation of WMH among individuals with normal cognition. 4,5,116his contribution may be subtle during the preclinical phase of AD, when pathology levels are low, and increase as the disease progresses, as evidenced by more robust associations of WMH burden with ɛ4 genetic status 37,43 or Alzheimer's disease-biomarker levels [116][117][118] among symptomatic individuals.Our finding of an interaction between APOE-ɛ4 and AD-PRS scores in relation to WMH trajectories further suggests that associations between APOE-ɛ4 and WMH load may be more evident among those with additional Alzheimer's disease risk genes, beyond APOE.
The current study is the first, to our knowledge, to demonstrate greater longitudinal increases in WMH burden among cognitively unimpaired APOE-ɛ2 carriers relative to ɛ3/ɛ3 carriers.This finding is consistent with and expands prior cross-sectional studies among non-demented and cognitively impaired cohorts that have also reported ɛ2-related elevations in WMH burden. 40,119Two crosssectional studies among cognitively normal middle-aged participants found no ɛ2-related WMH differences, 49,120 consistent with the absence of a baseline difference in WMH volumes by ɛ2 genetic status in this study.Altogether, these results support the view that the APOE-ɛ2 allele promotes cerebrovascular disease, though the mechanisms remain poorly understood. 121

Alzheimer's disease-polygenic risk, brain atrophy and WMH burden
Another important finding of this study was that higher AD-PRS scores were associated with greater atrophy over time in Alzheimer's disease-vulnerable regions, including the hippocampus, independent of APOE-ɛ4 status.This extends prior findings from longitudinal studies with participants across the Alzheimer's disease-spectrum 18,34,35 and from cross-sectional studies among non-demented corhorts 21,23,24,28,29 and suggest that higher Alzheimer's disease-polygenic risk scores increase the risk of neurodegeneration not only during the symptomatic phase of the disease but also among participants with normal cognition.Furthermore, AD-PRS-related atrophy was independent of age, sex, education and vascular risk scores, and not evident for regions that show non-Alzheimer's disease-related atrophy with age, consistent with a cross-sectional study. 28his suggests that APOE-ɛ4 and other AD risk genes influence atrophy in common Alzheimer's disease-susceptible brain regions, including the hippocampus, as early as midlife.
Higher AD-PRS were also associated with greater increases in global WMHs over time, particularly among APOE-ɛ4 carriers.This suggests that other Alzheimer's disease risk genes may exert some of their effects via cerebrovascular mechanisms, in addition to Alzheimer's disease-specific pathways.This interpretation is consistent with evidence linking higher AD-PRS scores to reductions in cerebral blood flow 122 and neuropathological markers of cerebrovascular disease, 34 though results might differ for PRS scores computed using other methodologies.Future studies that also include AD-biomarker assessments of amyloid and tau are needed to clarify how AD-PRS influence neurodegeneration.

Conclusion
The generalizability of findings from the current study to the broader population is limited because participants were primarily White, well-educated and enriched for a family history of Alzheimer's disease-dementia.For example, atrophy is more likely due to Alzheimer's disease in the current study than in the general population, and we may have overestimated associations between Alzheimer's diseasegenetic risk with atrophy and underestimated other factors, like vascular risk.Additionally, these analyses do not include biomarkers of Alzheimer's disease pathology or other measures of brain structure and function, which precludes inferences regarding the precise mechanisms by which Alzheimer's disease-genetic risk influences brain atrophy among unimpaired individuals.Also, the power to detect significant three-way interactions involving Alzheimer's disease-genetic variables for the observed effects was only low to moderate.Nonetheless, the study provides compelling evidence that APOE-ɛ4 and AD-PRS independently and additively influence longitudinal trajectories of neurodegeneration in Alzheimer's disease-sensitive regions and synergistically increase WMH accumulation among cognitively normal individuals.Conversely, APOE-ɛ2 primarily influences WMH accumulation, but not atrophy.These AD-genetic associations did not differ by participant sex, but in some cases were influenced by participant age, years of education and vascular risk, providing potential avenues for reducing the negative impact of AD risk genes on neurodegeneration prior to the development of cognitive impairment.Future studies are needed to examine the degree to which these Alzheimer's disease-genetic-related brain changes mediate changes in cognition.Estimates from mixed effects regression model showing independent associations between AD-polygenic risk scores and APOE-ɛ4 genetic status with longitudinal atrophy in AD-vulnerable regions (i.e.SPARE-AD scores, A, estimate = 0.007, SE = 0.003, P = 0.035) and the hippocampus (B, estimate = −0.006,SE = 0.003, P = 0.014).For global WMH volumes (C), APOE-ɛ4 genetic status was more strongly associated with increases in WMH volumes among those with higher compared to lower AD-polygenic risk scores (AD-PRS × APOE-ɛ4 × time interaction, estimate = 0.014, SE = 0.007, P = 0.038).For illustration purposes, the 25th and 75th percentiles of the baseline AD-PRS scores were used to show trajectories of high versus low PRS.

Figure 2
Figure 2 Longitudinal volumetric atrophy based on APOE-ɛ4 genetic status and participant education and age.Estimates from mixed effects regression model showing how years of education (A and B) and baseline age (C and D) modify the association between APOE-ɛ4 genetic status and rate of change of Alzheimer's disease-vulnerable regions, measured by SPARE-AD scores (A and C), and the hippocampus (Band D).The negative effect of APOE-ɛ4 genetic status on rate of atrophy in Alzheimer's disease-vulnerable regions and the hippocampus was greater among older than young participants (C and D), as indicated by significant three-way interactions of ɛ4 × age × time for SPARE-AD (estimate = 0.002, SE = 0.001, P = 0.001) and for the hippocampus (estimate = −0.001,SE = 0.0004, P = 0.027), but was attenuated among individuals with more years of education (A and B), as indicated by significant ɛ4 × education × time interactions for SPARE-AD (estimate = −0.015,SE = 0.006, P = 0.017) and for the hippocampus (estimate = 0.009, SE = 0.004, P = 0.039).For illustration purposes, the 25th and 75th percentiles of the baseline education were used to show trajectories of high versus low education.

TheFigure
Figure Longitudinal volumetric atrophy and WMH volumes as a function of AD-polygenic risk and APOE-ɛ4.Estimates from mixed effects regression model showing independent associations between AD-polygenic risk scores and APOE-ɛ4 genetic status with longitudinal atrophy in AD-vulnerable regions (i.e.SPARE-AD scores, A, estimate = 0.007, SE = 0.003, P = 0.035) and the hippocampus (B, estimate = −0.006,SE = 0.003, P = 0.014).For global WMH volumes (C), APOE-ɛ4 genetic status was more strongly associated with increases in WMH volumes among those with higher compared to lower AD-polygenic risk scores (AD-PRS × APOE-ɛ4 × time interaction, estimate = 0.014, SE = 0.007, P = 0.038).For illustration purposes, the 25th and 75th percentiles of the baseline AD-PRS scores were used to show trajectories of high versus low PRS.

Table 1 Participant characteristics at baseline Participants in APOE analyses with volumetric data Participants in APOE analyses with WMH data Participants in AD-PRS analyses with volumetric data Participants in AD-PRS analyses with WMH data
a Includes ɛ2/ɛ2 and ɛ2/ɛ3 carriers.b Includes ɛ2/ɛ4, ɛ3/ɛ4 and ɛ4/ɛ4 carriers.

Table 3 Results from mixed effects models testing whether associations between genetic risk factors and rate of change in MRI measures differ by demographic and clinical variables SPARE-AD (AD-related atrophy) SPARE-BA-resid (age-related atrophy) Hippocampus volume WMH volume Estimate (SE) P-value Estimate (SE) P-value Estimate (SE) P-value Estimate (SE) P-value
Three-way interaction terms for each demographic variable (i.e.baseline age, sex and years of education) or clinical variable (i.e.VRS score and progressed status) with APOE-ɛ4 and APOE-ɛ2 (or AD-PRS scores) and time were tested in separate models, adjusting for baseline age, sex, years of education, cohort indicators and interactions of all predictors with time.All lower-order interaction terms for the three-way interactions were also included.Bold values represent significant effects at P < 0.05.

Table 4 Mixed effects model results of AD-polygenic risk score and APOE genetic status in relationship to MRI measures SPARE-AD (AD-related atrophy) SPARE-BA-resid (age-related atrophy)
All models were adjusted by baseline age, sex, years of education, indicators for each cohort and included interactions of each predictor with time (e.g.terms for all genetic predictors × time and covariates × time).