APOE ɛ4 exacerbates age-dependent deficits in cortical microstructure

Abstract The apolipoprotein E ɛ4 allele is the primary genetic risk factor for the sporadic type of Alzheimer’s disease. However, the mechanisms by which apolipoprotein E ɛ4 are associated with neurodegeneration are still poorly understood. We applied the Neurite Orientation Dispersion Model to characterize the effects of apolipoprotein ɛ4 and its interactions with age and education on cortical microstructure in cognitively normal individuals. Data from 1954 participants were included from the PREVENT-Dementia and ALFA (ALzheimer and FAmilies) studies (mean age = 57, 1197 non-carriers and 757 apolipoprotein E ɛ4 carriers). Structural MRI datasets were processed with FreeSurfer v7.2. The Microstructure Diffusion Toolbox was used to derive Orientation Dispersion Index maps from diffusion MRI datasets. Primary analyses were focused on (i) the main effects of apolipoprotein E ɛ4, and (ii) the interactions of apolipoprotein E ɛ4 with age and education on lobar and vertex-wise Orientation Dispersion Index and implemented using Permutation Analysis of Linear Models. There were apolipoprotein E ɛ4 × age interactions in the temporo-parietal and frontal lobes, indicating steeper age-dependent Orientation Dispersion Index changes in apolipoprotein E ɛ4 carriers. Steeper age-related Orientation Dispersion Index declines were observed among apolipoprotein E ɛ4 carriers with lower years of education. We demonstrated that apolipoprotein E ɛ4 worsened age-related Orientation Dispersion Index decreases in brain regions typically associated with atrophy patterns of Alzheimer’s disease. This finding also suggests that apolipoprotein E ɛ4 may hasten the onset age of dementia by accelerating age-dependent reductions in cortical Orientation Dispersion Index.


Introduction
The APOE ɛ4 allele is the strongest genetic risk factor for both sporadic early and late-onset Alzheimer's disease. 1,2 growing body of neuroimaging studies has delineated the imaging phenotypes of APOE ɛ4 in cognitively normal individuals as a means to detect preclinical changes that may occur before cognitive impairment (for systematic reviews, see 3,4 ) Some notable findings in middle-aged APOE ɛ4 carriers include (i) grey matter atrophy in typical Alzheimer's disease regions, [5][6][7] (ii) diffusion tensor imaging (DTI) changes in white matter tracts, 8 (iii) functional deficits in cerebral perfusion and glucose metabolism [9][10][11] as well as (iv) disruptions in both task-based functional MRI 12,13 and intrinsic connectivity within resting-state networks.14 Nonetheless, definitive evidence for a APOE ɛ4 'signature' remains elusive due to null results and inconsistent observations.In addition to methodological discrepancies, small APOE ɛ4 effect sizes and sample variability across cohorts, contradictory results may also be attributed to the complex midlife effects of the APOE ɛ4 gene as well as its interactions with other pathological markers of Alzheimer's disease (i.e.amyloid and tau).The associations between APOE ɛ4 and brain imaging changes and cognition in midlife adults may to some extent be modulated by sex and cognitive reserve.15,16 Similarly, a growing body of evidence suggests that the presence of APOE ɛ4 may exacerbate the trajectory of age-related biomarker changes, with older APOE ɛ4 homozygotes exhibiting an accelerated decline in white matter microstructural integrity, 8 myelination (i.e.MRI T 1 -weighted/T 2 -weighted ratio), 17 grey matter volumes and hippocampus shape.18,19 It is also conceivable that the deleterious effects of APOE ɛ4 are more pronounced at the cellular level of neurite morphology (i.e.dendritic arborization) and so would not be easily detected by macroscale analysis of standard MRI biomarkers. Thi notion is reinforced by histological evidence linking APOE ɛ4 to several synaptic and neuritic abnormalities.20 Thus, to the extent that neuritic abnormalities precede major cell death or brain atrophy, imaging markers of grey matter microstructure may reveal some of the earliest effects related to APOE ɛ4 genotype.To this end, the mean diffusivity (MD) parameter from DTI has been shown to be sensitive to dendritic alterations in the hippocampus, 21 and its applications have expanded to include the analysis of hippocampal subfields and cortical grey matter in patients with autosomal dominant Alzheimer disease, 22 mild cognitive impairment (MCI) and Alzheimer's disease.[23][24][25] Collectively, these findings underscore the potential for diffusion MRI to characterize subtle cortical microstructural deficits.However, given the inherent differences in image resolution between typical DTI (2 mm) and T 1 -MRI sequences (1 mm), the cortical MD signal could be sensitive to overestimation due to the spill-in of high MD values from unrestricted 'free water' diffusion in the CSF.26 Recent advances in diffusion-weighted imaging have led to the development of biophysical models, expanding the application of diffusion imaging beyond white matter to grey matter microstructure. 27 ne particularly promising technique for the quantification of neurite morphology at the level of axons and dendrites is neurite orientation dispersion and density imaging (NODDI), which features a threecompartment model to distinguish signal from intra-neurite, extra-neurite and CSF.The Orientation Dispersion Index (ODI) from NODDI reflects the complexity or dispersion in brain tissue.Histological correlations between ODI and post-mortem measures of orientation complexity suggest that these NODDI and ODI may be used to characterize in vivo dendritic branching.,31 A recent study by our group demonstrated a close relationship between ODI and synaptic density in individuals with primary tauopathies, lending credence to the interpretation of ODI as a proxy of synaptic function.32 Given that Alzheimer's disease pathology accumulates for years before irreversible and observable macrostructural atrophy, 33 establishing NODDI as a measure of microstructural abnormalities during the asymptomatic period will be essential for early identification, improving prognosis, accurate disease-staging, and treatment monitoring.Though NODDI has shown promise in studies of normal ageing, 8,30 early-onset Alzheimer's disease, 34 and other tauopathies, 32 there is currently insufficient data to draw conclusions about the influence of APOE ɛ4 on grey matter NODDI measurements in cognitively normal midlife adults. In tis multi-site study, the NODDI model is applied to characterize the effects of APOE ɛ4 and its interactions with age and education on cortical microstructural changes in cognitively normal individuals recruited from the PREVENT-Dementia 35 and ALFAprojects.36 Both PREVENT-Dementia and ALFA are prospective observational cohort studies with the overarching goal of clarifying the pathophysiology and pathogenic factors occurring during the preclinical phase of Alzheimer's disease.The goals of this research were to (i) compare APOE ɛ4 carriers and non-carriers on measures of ODI measured within the lobes and vertex-wise, (ii) evaluate the extent to which APOE ɛ4 interacts with age and education to determine ODI, and (iii) investigate whether NODDI provides novel insights beyond that provided by conventional biomarkers such as cortical thickness.

Study participants
The recruitment for ALFA subjects consisted of two steps.First, 2743 cognitively healthy volunteers aged between 45 and 76 years were enrolled in the ALFA.Exclusion criteria included performance exceeding the established cut-off for a number of cognitive tests and the presence of a psychiatric diagnosis.Second, after APOE ɛ4 genotyping, all participants homozygous for the ɛ4 allele as well as carriers of the ɛ2 allele were invited to undergo MRI scanning along with APOE ɛ4 heterozygotes and non-carriers, matched for age and sex.The ALFA+ study was approved by the Independent Ethics Committee 'Parc de Salut Mar', Barcelona.All participants signed the informed consent form that had also been approved by the Independent Ethics Committee 'Parc de Salut Mar', Barcelona; 701 participants were recruited in the PREVENT-Dementia study from 5 study sites: West London (n = 210), Edinburgh (n = 223), Cambridge (n = 100), Oxford (n = 68) and Dublin (n = 100).The main entry criteria were age 40-59 and the absence of dementia or other neurological disorders.Approval for the study has been given by the NHS Research Ethics Committee London Camberwell St-Giles.

Preprocessing
Image preprocessing and vertex-wise analyses were performed on a dedicated cluster composed of CPU nodes [Cambridge Service for Data-Driven Discovery (CSD3): 2×Intel Xeon Skylake 6142 processors, 2.6 GHz 16-core, 192GiB RAM per node], and GPU servers for the GPU-optimized parts of the processing pipeline (NVidia Tesla K40/K80 GPUs).

Structural MRI
T 1 structural MRI datasets from ALFA and PREVENT were preprocessed using identical procedures.MRI scans were first corrected for intensity inhomogeneities with the Advanced Normalization Toolbox N4 algorithm. 37utomated surface-based reconstruction of the T 1 -MRI datasets was performed using FreeSurfer v7.2 (http://surfer.nmr.mgh.harvard.edu/). 38Visual inspections of the scans were performed by two authors (E.M. and E.T.Z) to ensure that there were no severe errors in the surface constructions (i.e.global misclassification of grey and white matters).Minor edits were made conservatively on datasets with white matter or pial surface errors when necessary (see Supplementary Materials for an example of topological correction).Estimates of cortical thickness were determined using Freesurfer by computing the average distance between grey matter/white matter and grey matter/CSF surfaces at each vertex across the cortical mantle.These thickness measurements were then projected to the inflated surface of the rebuilt brain of each subject.

Diffusion MRI Preprocessing
Diffusion datasets were visually inspected for optimal coverage and to ensure minimal eddy-current distortions.Subjects were excluded if there were excessive Echo Planar Imaging (EPI) distortions, poor image quality or insufficient field-of-view coverage.First, the dwidenoise tool from MRTRIX (https://www.mrtrix.org) was used to perform denoising on the raw 4D DWI volumes to improve signal-to-noise ratio. 39Second, the mrdegibbs tool was used to apply Gibbs ringing correction to get rid of ringing artefacts.Subsequently, the DWI datasets were stripped of nonbrain tissue using the Brain Extraction Tool.To confirm the accuracy of the masks, a visual inspection was undertaken, and manual modifications were performed as needed.Eddy currents and head movements were corrected with ed-dy_openmp in FSL (Version 6.0.5).Quantitative identification of slices with signal loss was performed using the -repol option, and outlier volumes were replaced by nonparametric predictions using the Gaussian process. 40The b-matrix was subsequently reoriented by applying the rotational part of the affine transformation used during eddy correction. 41

Neurite orientation dispersion and density imaging
Following eddy-current correction, we applied the NODDI model 27 using the Microstructure Diffusion Toolbox 42 to generate ODI whole-brain maps.The diffusion signal within each brain voxel is described by NODDI as a combination of intracellular, extracellular and CSF components.The intracellular compartment is modelled as a set of sticks with restricted diffusion perpendicular to the orientation of the axonal bundles and unimpeded diffusion along them in order to capture neurite membranes and myelin sheaths.The extracellular compartment is assumed to encompass the region surrounding the neurites, which is made up of glial cells and somas.Isotropic diffusion is used to represent the CSF compartment.The ODI is a tortuosity measure computed as the neurite dispersion coefficient.Values closer to 0 indicate less dispersion or well-aligned neurites, but values closer to 1 indicate greater dispersion, which is often associated with increased dendritic branching or complexity in grey matter. 29The original paper established the feasibility of modelling the ODI from single-shell diffusion measurements. 27Furthermore, single-shell ODI studies have been demonstrated to reproduce group-level differences from multi-shell DWI data, particularly analyses were limited to the grey matter. 43The neurite density (NDI) maps were excluded from all analyses since prior research found significant discrepancies between the NDI maps from multi-and single-shell DWIs. 27,43

Coregistration and estimation of lobar ODI
The b0 volumes were coregistered with the T 1 -MRI using Freesurfer's Boundary-Based Registration. 44In order to warp the cortical lobar labels to native DWI space using the inverse transformation that was utilized to coregister the b0 to the T 1 -MRI, we used the mri annotation2label command with the '-lobesStrict' flag, followed by mri label2vol and mri vol2vol.The average ODI values for the parietal, occipital, temporal, cingulate and insula lobes were generated using 'mri segstats'.Given the inherent resolution differences between DWI and T 1 -MRI, the lobar methodology depends less on the perfect alignment of fine sulcal areas and provides a robust method for detecting APOE ɛ4-related alterations in the presence of potential image-coregistration difficulties Furthermore, it has been shown that lobe averaged values of imaging derived cortical thickness levels were comparable to the range of values as reported by von Economo's histological measurements. 45,46

Vertex-wise ODI surfaces
In addition, we performed a vertex-wise analysis to provide a more nuanced understanding of the correlations between APOE ɛ4 and topological ODI changes.In order to accomplish this, the ODI intensities were sampled from the midpoint of the cortical ribbon (i.e.50% of the cortical thickness along the surface normal to the grey matter/white matter surface) and projected to an inflated surface using the mri vol2surf function in Freesurfer, before smoothing at 15-mm Guassian kernel.

ComBat harmonization
To attenuate site-related effects of ODI, we used ComBat to lobar ODI and vertex-wise ODI datasets using R and custom MATLAB scripts, respectively. 47Age, sex and APOE ɛ4 status were specified as biological factors to retain the biological variability of the data, as demonstrated by the expected correlations with age (Supplementary Materials).The outcomes of the multi-site harmonization and the effects of ComBat on ODI are depicted in subject-specific box plots of lobar ODI, ordered by the site (Supplementary Materials).After using ComBat, site effects on ODI were no longer statistically significant.

Statistical analyses
All statistical analyses were performed in R version 4.2.0 (22  April 2022).Demographic variables and characteristics of the cohort were reported as mean (standard deviation, SD) or number (percentage), and differences between the APOE ɛ4 carriers and non-carriers were compared using ANOVA, Kruskal-Wallis, and χ 2 tests, where appropriate.Individuals with the ɛ2/ɛ4, ɛ3/ɛ4 and ɛ4/ɛ4 genotypes were classified as APOE ɛ4 carriers and compared against the non-carriers (ɛ2/ɛ2, ɛ2/ɛ3 and ɛ3/ɛ3 genotypes).Our primary analyses focused on the (i) main effects of APOE ɛ4, and the (ii) interactions of APOE ɛ4 with age on lobar and vertexwise ODI.Non-parametric permutation models were computed using Permutation Analysis of Linear Models (PALM; https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/PALM).In addition, the quadratic term of age was examined in order to model nonlinear changes.Thus, a significant main effect of APOE ɛ4 would imply group-level differences in ODI, whereas a significant interaction term would be consistent with a 'accelerating' effect of APOE ɛ4 on age-dependent ODI changes.The R package 'emmeans' was used to estimate marginal effects and to estimate the earliest age at which differences between APOE ɛ4 carriers and noncarriers could emerge.(iii) Non-parametric permutation models were used to assess three-way interactions between formal education years, age, and APOE ɛ4 on ODI to determine the potential role of cognitive reserve in moderating the effects of APOE ɛ4 on ODI changes.Statistical results were reported after correcting for multiple comparisons using family wise error (FWE).The TFCE method was used to adjust for multiple comparisons in all vertex-wise analyses.

Sample characteristics
Participant characteristics are summarized in Table 1.

APOE ɛ4 × age
There were significant APOE ɛ4 × age interactions in the frontal, parietal, temporal, cingulate and insula lobes, Plots of marginal effects showed steeper age-related decreases in ODI for APOE ɛ4 carriers compared with noncarriers Fig. 2. To estimate the hypothetical inflection points of divergent ODI age trends, we calculated between-group differences of the predicted ODI from age 40 to 80 and determined the earliest age at which the differences emerged (Fig. 3).In general, lobar ODIs were similar between APOE ɛ4 carriers and non-carriers from age 40 to 60, after which significant reductions in ODI became more apparent in APOE ɛ4 compared with the non-carriers (temporal: age = 60-74, cingulate: age = 64-76, frontal: age = 63-80, insula: age = 63-77 and parietal: age = 61-80).
Next, we used an unbiased vertex-wise approach to map the topography of the APOE ɛ4 age interactions across the cortex.Significant clusters were found primarily in the temporo-parietal cortices and the frontal lobe (Fig. 4). Figure 5 tabulates the summary statistics and MNI coordinates of the peak effects in each cluster.There were no significant interactions between APOE ɛ4 and age on lobar cortical thickness (P FWE > 0.05).

APOE ɛ4 × age × education interactions
Three-way interactions of APOE ɛ4 × age × education were found in the frontal, parietal lobes (P FWE < 0.05), indicating steeper age-related reductions in ODI among APOE ɛ4 carriers with lower years of education (mean-1 SD).Post hoc stratified analyses in APOE ɛ4 carriers and non-carriers confirmed that the age × education interaction was significant only among APOE ɛ4 carriers [frontal: T = 2.68, parietal T = 2.37 and P FWE < 0.05 (Fig. 6)].
The fronto-parietal lobar patterns of APOE ɛ4 × age × education interactions were corroborated by vertex-wise permutation analyses, which revealed peak effects in the rostral left middle frontal and precentral gyri (Fig. 7; TFCE P FWE < 0.05).The summary statistics and MNI coordinates of the peak effects in each cluster are shown in Fig. 8.There were no significant three-way interactions between APOE ɛ4 × age × education on lobar cortical thickness.

Figure 3 Differences in standardized lobar ODI between APOE ɛ4 carriers and non-carriers (y-axis) as a function of age (x-axis).
Pairwise comparisons were calculated based on marginal effects to predict the earliest hypothetical age at which APOE ɛ4 carriers would show decreased lobar ODI relative to non-carriers, highlighted by the vertical dotted lines on the x-axis.The shaded cyan region represents the 95% confidence interval.For each lobe, the earliest age of significant ODI reductions in APOE ɛ4 carriers is illustrated in the choropleth brain map.Abbreviations: APOE, Apolipoprotein E gene; ODI, Orientation Dispersion Index.

Sensitivity analyses
A series of sensitivity analyses were performed to determine the robustness of our key findings against potential confounds.(i) All analyses were repeated after excluding the APOE ɛ2ɛ4 carriers due to the genetic ambiguity of contrasting risk profiles.(ii) To investigate whether the APOE ɛ4 effects on ODI are independent of atrophy, we additionally adjusted for mean cortical thickness.All the findings were unchanged.Detailed results for these analyses are provided in the Supplementary Materials.

Discussion
Our findings demonstrate more pronounced age-related declines in cortical ODI among APOE ɛ4 carriers compared with non-carriers.The interactions of APOE ɛ4 and age were most prominent in the temporo-parietal lobes, which are known to be more vulnerable to early neurodegeneration and accumulation of Alzheimer's disease abnormalities.Second, our analyses of the three-way interactions between APOE ɛ4, age, and education showed that the detrimental effects of APOE ɛ4 with increased age were most strongly expressed in ɛ4 carriers with lower levels of education, suggesting that high levels of cognitive reserve could mitigate against the accelerated ODI decline in ɛ4 carriers.Finally, these findings were independent of cortical atrophy.While conclusions are limited due to the cross-sectional design, our findings imply that APOE ɛ4-related changes in ODI may precede macroscopically visible changes in cortical atrophy and may be a more sensitive marker of incipient Alzheimer's disease.
A growing body of evidence suggests that APOE ɛ4 is a potent moderator of age-related trends for imaging biomarkers. 17,19,48,49Indeed, we observed widespread interactions between age and APOE ɛ4 on ODI, characterized by steeper age-dependent ODI reductions in APOE ɛ4 carriers compared with non-carriers.Insofar as ODI is a valid proxy  for dendritic complexity, 28,29 our findings could also be interpreted as an in vivo extension of a previous histological study in which Alzheimer's disease APOE ɛ4 carriers exhibited a marked loss of dendritic outgrowth and synaptic count relative to non-carriers. 50Loss of dendritic arborization is a common phenomenon associated with the ageing brain, 51 and previous NODDI studies have demonstrated age-related decline with ODI. 30,52The main novel aspect of our finding, however, is that the presence APOE e4 allele may further exacerbate the age-dependent decline in ODI.The interactions between APOE e4 and age are also broadly consistent with data from other imaging modalities, including accelerated grey matter atrophy, 53 alterations in myelination, 17 hippocampal shape, 18,19 glucose metabolism 54 and cerebral perfusion. 10,55More broadly, these studies suggest that the previously outlined inconsistencies in the literature could be clarified by taking into account the non-uniform effects of APOE4 across age in the future.
Marginal effect analyses revealed that cortical ODI trajectories diverged between APOE 4 allele carriers and non-carriers around the age of 60 (i.e.temporal lobe).While additional longitudinal studies are required to validate these results, our data may hint at the existence of an inflection point (age = 60) prior to the appearance of more severe grey matter deficits in APOE ɛ4 carriers.Multiple lines of research seem to support this notion.For instance, it has been shown that APOE ɛ4 is more closely associated with Alzheimer's disease neuropathology after the age of 60. 56 Within the context of current biomarker models in Alzheimer's disease, 57 our results positioned the onset of APOE ɛ4-related ODI alterations (age = 60) in close temporal proximity with the exponential build-up of amyloid (age = 60), 58 roughly a decade before the onset of cognitive impairment APOE ɛ4 (age = 69). 59,60These studies, including our own findings, lend support to the hypothesis that the detrimental consequences of APOE ɛ4 are more pronounced after the sixth decade.
Although the molecular pathways by which age and APOE ɛ4 interact to influence cortical microstructure remain poorly understood, several lines of research support the biological plausibility of our findings.(i) First, APOE ɛ4 may influence age-related ODI declines via tau-related pathways. 58,61Recent evidence suggests that cognitively normal APOE 4 carriers accumulate tau and amyloid at a quicker rate beginning at age 60, which coincides with our findings in the temporal lobe ODI (earliest divergence between APOE 4 and non-carriers at age 60). 58Indeed, our vertex-wise analyses demonstrated that the most prominent effects were localized within the temporal cortices (e.g.entorhinal cortex, inferior temporal cortex)-early predilection sites for neurofibrillary tangles. 62Previous data from transgenic mice 63 and neuropathological studies have also shown that APOE ɛ4 is related to increased tau deposition in the temporal cortex. 64In addition, lower cortical ODI has been observed in rTg4510 transgenic mice model. 65ur group has recently demonstrated in vivo reductions of ODI in patients with primary taupathies, 32 and lower ODI has been correlated with increased uptake of the PET [18F] THK5351 tracer in an Alzheimer's disease sample. 66ii) Second, amyloid deposition may also play a role, since 10-15% of APOE ɛ4 participants in our cohorts are anticipated to have elevated amyloid burden. 67,68It has been shown that dendritic branching in close proximity to amyloid deposits exhibits bending abnormalities, more abrupt branch terminations and spine loss. 69Future research integrating multi-modal imaging with CSF or plasma biomarkers 70 would be well-suited to distinguish the unique contributions of Alzheimer's disease pathologies to changes in cortical ODI.In order to establish whether our vertex-wise findings reflect the 'penumbra' of subsequent atrophy, longitudinal studies comprising a wider spectrum of cognitively normal APOE ɛ4 carriers, individuals with MCI and Alzheimer's disease are warranted.
A secondary aim of this study was to determine the degree to which education may attenuate the adverse effects of APOE ɛ4 on age-related ODI changes.To that end, our lobar and vertex-wise analyses indicated a significant threeway interaction between age, APOE ɛ4 and education years in the fronto-parietal cortices; i.e. steeper age-dependent ODI reductions were observed in APOE ɛ4 carriers with less education.In contrast, the impact of APOE ɛ4 on age-related ODI changes in individuals with more years of education was negligible.As such, our findings lend support to previous findings indicating that more years of education was a significant predictor of less cognitive change over time among ApoE e4 carriers 71 and are aligned with post-mortem evidence, suggesting that higher education may mitigate the deleterious effects of senile plaques 72 The involvement of fronto-parietal cortices may also be interpreted in light of their roles as key neural substrates of working memory. 73ndeed, a recent study has shown that education beyond high school may mitigate the effects of APOE ɛ4 on working  memory. 15Collectively, these findings support a scenario in which a diminished 'cognitive reserve' may lead to greater age-related susceptibility to the APOE ɛ4 allele. 16otably, our ODI findings persisted after controlling for cortical thickness.Moreover, we did not find any significant effects of APOE ɛ4 carriership on cortical thickness (i.e.effects may only appear in APOE ɛ4 homozygotes at this age range).The null results of APOE ɛ4 on cortical thickness could be anticipated in light of the inconsistent findings and modest effect sizes in the literature.Grey matter atrophy is also thought to reflect downstream neurodegenerative changes occurring later in the course of Alzheimer's disease. 74Under the assumption that a cross-sectional estimate of a biomarker represents its accumulated pathological burden, the combination of ODI changes and the absence of cortical thickness effects is compatible with early loss of dendritic arborization before cell death and atrophy. 75,76imilarly, greater magnitudes of changes seen in NODDI relative to that of atrophy have been demonstrated in studies of people with primary tauopathies, 32 young-onset Alzheimer's disease, 34 MCI and Alzheimer's disease. 77aken together, our findings suggest that NODDI may serve as a promising biomarker to reveal unique insights into subtle grey matter deficits in preclinical dementia.
There are several caveats to consider when interpreting our results.(i) Our cross-sectional design precludes definitive conclusions regarding the role of APOE ɛ4 in predicting age-related trends of ODI reduction.Therefore, conclusions of group differences at different ages should be considered preliminary.Prospective longitudinal studies are necessary to determine if the intra-individual rates of ODI reductions are (ii) significantly accelerated in APOE ɛ4 carriers and (iii) whether they are associated with cognitive decline.While our findings revealed robust associations between APOE ɛ4 and accelerated reductions in ODI, the pathological processes underpinning this relationship remain unclear.Additional investigations using fluid biomarkers or PET imaging would be required to characterize the pathogenic processes that may underlie the APOE ɛ4-related ODI alterations.Leveraging multi-modal data will facilitate analyses to investigate potential interactions of APOE ɛ4 with other pathological markers on ODI.The lack of amyloid status on our participants is another limitation, and future studies are needed to determine whether findings are consistent in amyloid-negative APOE ɛ4 carriers.(iv) The ODI parameters were determined from diffusion-weighted single-shell datasets as opposed to multi-shell NODDI sequences.Nonetheless, the feasibility of estimating ODI from singleshell DWI had been demonstrated in the seminal NODDI study, 27 and subsequent groups have confirmed that ODI maps calculated using multi-shell and single-shell (b = 1000 s/mm 2 ) are highly comparable and exhibit a high degree of overlap with respect to group differences. 43v) While we have interpreted ODI reductions as suggestive of dendritic arborization loss, histological validation of NODDI measurements remains confined to post-mortem spinal cord specimens from patients with multiple sceleroides. 29More research triangulating data from ex vivo imaging, post-mortem samples and PET imaging would be required to draw firmer conclusions on the specific cellular abnormalities revealed by ODI.(vi) Lower education years increased the detrimental effect of APOE ɛ4 on age-related ODI declines; however, interpretations should also account for the extensive relationships of education with general health, environmental enrichment and socioeconomic status, all of which could significantly moderate the associations between education and APOE ɛ4-related ODI decreases.(vii) Another limitation of the current study is our exclusive focus on older adults with normal cognitive function.By restricting our sample to non-impaired individuals, we aimed to delineate structural brain changes associated with APOE ɛ4 in the preclinical stage, prior to the onset of overt symptoms.Therefore, our findings may not generalize to later stages of dementia, where relationships between APOE ɛ4 status and brain structure may be more complex.
In summary, we have conducted a comprehensive analysis of APOE ɛ4 carriership and its impact on age-related patterns of ODI alterations in one of the largest datasets of cognitively normal persons in midlife.We found novel evidence that APOE ɛ4 worsened age-related ODI decreases in brain regions typically associated with atrophy patterns of Alzheimer's disease.This finding also suggests that APOE ɛ4 may hasten the onset age of dementia by accelerating age-dependent reductions in cortical ODI, although additional studies are needed to verify this hypothesis.Finally, our research lends methodological support to the use of NODDI as a sensitive surrogate for changes in grey matter microstructure.

Figure 2
Figure 2 Marginal effect graphs depicting the two-way interactions between APOE ɛ4 and age (x-axis) on lobar ODI values (y-axis).Individuals with the APOE ɛ4 allele showed significantly steeper age-related reductions in ODI compared with non-carriers in the cingulate, frontal, insula, parietal and temporal lobes.Statistical significance was determined in PALM using non-parametric permutation models (5000 permutations, P FWE < 0.05, adjusted for gender, years of formal education and sites).APOE, Apolipoprotein E gene; FWE, family wise error; ODI, Orientation Dispersion Index; PALM, Permutation Analysis of Linear Models.

Figure 4
Figure 4 Vertex-wise interactions between APOE ɛ4 and age on cortical ODI.Results were obtained using non-parametric permutation models, implemented using PALM on ComBat harmonized ODI surface maps, and significance was established with TFCE P FWE < 0.05, adjusted for sex, years of formal education and site.Abbreviations: APOE, Apolipoprotein E gene; TFCE, Threshold Free Cluster Enhancement; FWE, Family Wise Error; ODI, Orientation Dispersion Index; PALM, Permutation Analysis of Linear Models.

Figure 5
Figure 5 Summary statistics and MNI152 coordinates significant cluster associated with the APOE ɛ4×age interaction on vertex-wise ODI (TFCE P FWE < 0.05, adjusted for sex, years of formal education, and site).Abbreviations: APOE, Apolipoprotein E gene; TFCE, Threshold Free Cluster Enhancement; FWE, family wise error; ODI, Orientation Dispersion Index; MNI152, Montreal Neurological Institute 152.

Figure 6
Figure 6 Three-way interactions between APOE ɛ4 × age × education.Marginal effects are depicted for the frontal and parietal lobes, stratified by APOE ɛ4 status.Lower years of education (mean-1 SD) are associated with steeper age-dependent reductions of ODI among the ɛ4 carriers only (frontal: T = 2.68, parietal: T = 2.37, P FWE < 0.05, adjusted for site and sex).Abbreviations: APOE, Apolipoprotein E gene; FWE, family wise error; SD, standard deviation; ODI, Orientation Dispersion Index.

Figure 7
Figure 7 Vertex-wise three-way interactions between APOE ɛ4 × age × education.Marginal effects are depicted for the frontal and parietal lobes, stratified by APOE ɛ4 status.Lower years of education (mean-1 SD) are associated with steeper age-dependent reductions of ODI among the ɛ4 carriers only (P FWE < 0.05, adjusted for site and sex).Abbreviations: APOE, Apolipoprotein E gene; FWE, family wise error; SD, standard deviation; ODI, Orientation Dispersion Index.

Figure 8
Figure 8 Summary statistics and MNI152 coordinates for each significant cluster associated with the three-way interactions of APOE ɛ4×age × education interaction on vertex-wise ODI.Abbreviations: APOE, Apolipoprotein E gene; ODI, Orientation Dispersion Index; MNI152, Montreal Neurological Institute 152.