APOE ε4 gene dose drives a canonical and predictable topography of amyloid deposition.

APOE ε4 gene dose drives a canonical and predictable topography of amyloid deposition MENDES et al.

Abstract

INTRODUCTION

Apolipoprotein E ε4 (APOE ε4) carriership is associated with higher amyloid burden in Alzheimer's disease (AD), but whether it constrains the spatial predictability of amyloid deposition at the individual level remains unknown.

METHODS

We analyzed voxel‐wise amyloid positron emission tomography data from 781 amyloid‐positive individuals across two independent cohorts (Alzheimer's Disease Neuroimaging Initiative and Geneva Memory Center), spanning the full clinical continuum. Individual topographic similarity to canonical AD amyloid maps and within‐group spatial homogeneity were assessed via voxel‐wise correlations and linear regression adjusted for age, sex, and global amyloid burden.

RESULTS

APOE ε4 carriers exhibited amyloid distributions more aligned with canonical AD topographies than non‐carriers (p < 0.001), with a clear gene‐dose effect on within‐group spatial homogeneity: Homozygotes showed the highest inter‐subject consistency, followed by heterozygotes and non‐carriers. Effects were robust across tracers, clinical stages, and cohorts.

DISCUSSION

APOE ε4 gene dose constrains the spatial predictability of amyloid accumulation, informing genotype‐stratified approaches in anti‐amyloid trials.

Highlights

APOE ε4 gene dose increases the spatial predictability of amyloid deposition.

APOE ε4 carriers show greater alignment with canonical AD amyloid topographies.

Homozygotes exhibit the highest inter‐individual spatial homogeneity of amyloid.

Effects were replicated in both cohorts and independent of global amyloid burden.

Findings support APOE ε4‐stratified strategies in anti‐amyloid trials.

1INTRODUCTION

Why amyloid deposition follows predictable spatial trajectories in some individuals but not others is a fundamental, unresolved question in Alzheimer's disease (AD) research, with direct consequences for the emerging era of anti‐amyloid therapies. The classical amyloid cascade hypothesis posits a largely deterministic sequence of events whereby amyloid deposition triggers tau aggregation, followed by neurodegeneration and, ultimately, progressive cognitive decline.1 However, growing evidence points to substantial heterogeneity in the onset, regional distribution, and rate of progression of AD pathology, as well as in its clinical manifestations, suggesting that this process may vary substantially across individuals.2, 3

To address this complexity, several variants of the deterministic version of the amyloid cascade hypothesis have been proposed. One example is the probabilistic model of AD, which reinterprets the cascade by integrating both genetic and stochastic factors into the sequence of events leading to cognitive impairment and progression.4 Within this framework, apolipoprotein E ε4 allele (APOE ε4) carriership, the strongest common genetic risk factor for sporadic AD,5 is a key biological modifier of the cascade's penetrance. The model proposes that the degree of determinism in the amyloid‐to‐tau‐to‐neurodegeneration sequence varies according to the strength of the genetic factors. Accordingly, three major forms of AD can be distinguished: autosomal dominant AD (ADAD), APOE ε4‐related sporadic AD, and APOE ε4‐unrelated sporadic AD. These categories represent a continuum of decreasing pathophysiological determinism and increasing influence of stochastic factors, such as lifestyle behaviors and exposures.4 In ADAD, due to the presence of a deterministic mutation, amyloid accumulation, and subsequent pathological events in the AD cascade occur inevitably and follow a relatively predictable trajectory. In contrast, in APOE ε4‐unrelated sporadic AD, amyloid deposition and the downstream pathological processes are increasingly heterogeneous and less predictable, reflecting a greater influence of non‐genetic and modifiable factors.4 Furthermore, a growing body of evidence has emphasized the gene‐dose effect of APOE ε4, as homozygosity confers a markedly higher penetrance and earlier onset of pathology.6, 7 This has raised the question of whether APOE ε4 homozygosity constitutes a genetic form of AD.8, 9, 10, 11

Despite previous evidence that APOE ε4 modulates global amyloid burden12 or regional differences between APOE ε4 genotypes,13 it has not been demonstrated whether APOE ε4 constrains the topographic predictability of amyloid deposition at the individual level. Prior studies have characterized regional differences in amyloid uptake between APOE ε4 genotype groups, but group‐level comparisons do not capture the degree to which each individual's amyloid positron emission tomography (PET) pattern resembles a canonical AD spatial template, nor do they assess within‐genotype spatial consistency across subjects. These are distinct and complementary dimensions of topographic organization that remain uncharacterized. Addressing this gap could clarify how APOE ε4 gene dose constrains the spatial substrate on which amyloid‐targeting therapies act and inform the design of genotype‐stratified clinical trials.

In this study, we conducted topographical analyses of amyloid‐PET data from two independent cohorts spanning cognitively unimpaired (CU), mild cognitive impairment (MCI), and dementia stages to quantify individual similarity to canonical AD amyloid maps and within‐group spatial homogeneity across APOE ε4 genotypes. We hypothesized that APOE ε4 carriers, particularly homozygotes, would exhibit a more biologically constrained pattern of amyloid accumulation, reflected by greater alignment with canonical AD topographies and higher within‐group spatial consistency compared with non‐carriers.

RESEARCH IN CONTEXT

Systematic review of the literature: We searched the literature on APOE ε4 and amyloid‐PET, identifying studies examining genotype‐related differences in amyloid burden and regional distribution. Prior work established that APOE ε4 increases global burden and shifts deposition toward temporoparietal hubs, but no study has quantified individual‐level topographic similarity to canonical AD maps or within‐genotype spatial consistency.

Interpretation: APOE ε4 gene dose constrains the spatial predictability of amyloid accumulation. Carriers, in particular homozygotes, show greater alignment with canonical AD topographies and higher inter‐individual homogeneity than non‐carriers, independent of global burden, revealing topographic stereotypy as a novel dimension of APOE ε4‐related amyloid pathology.

Future directions: Longitudinal amyloid and tau PET studies should determine whether topographic stereotypy predicts amyloid accumulation rate and downstream tau propagation. Clinical trials of anti‐amyloid therapies should prospectively test whether APOE ε4 homozygosity predicts more uniform spatial treatment response, informing genotype‐stratified eligibility criteria.

2METHODS

2.1Participants

Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). ADNI was launched in 2003 as a public‐private partnership, led by Principal Investigator Michael W. Weiner, MD, with the primary goal of developing biomarkers for the early detection and tracking of AD. Further details are available at www.adni‐info.org. We included amyloid‐positive individuals from ADNI spanning the CU, MCI, and dementia stages. Amyloid positivity was determined using amyloid‐PET scans with florbetapir ([1
8F]‐AV‐45), applying a standardized uptake value ratio (SUVR) cutoff of 1.11 using the whole cerebellum as the reference region, as recommended by the authors of the original analysis.14 Demographic, genetic, and clinical data were obtained from the ADNI database. APOE genotyping was performed as part of the ADNI baseline protocol, and participants were categorized according to the number of ε4 alleles as non‐carriers (zero ε4 alleles), heterozygotes (one ε4 allele), or homozygotes (two ε4 alleles).

To validate our findings, we analyzed an independent sample of amyloid‐positive individuals with MCI or dementia from the Geneva Memory Center (GMC). Amyloid positivity in the GMC cohort was assessed visually by an expert in nuclear medicine (VG).

Ethical approval was obtained from the Institutional Review Boards of all participating ADNI institutions and the Cantonal Ethics Committee of Geneva for the GMC cohort (PB_2016‐01346 and 2020_00403).

2.2Imaging acquisition and processing

2.2.1ADNI acquisition

PET imaging scans in the ADNI cohort used [1
8F]florbetapir (AV‐45) tracer. Participants received an intravenous bolus injection of 370 MBq (10 mCi ± 10%) of AV‐45, followed by a 30‐min uptake period in a resting state. Image acquisition began 50 min after injection and consisted of a dynamic 3D scan comprising four consecutive 5‐min frames. Scans were performed using PET or PET/computed tomography (CT) systems with appropriate attenuation correction, via either transmission scans (PET) or low‐dose CT scans (PET/CT). Subjects were closely monitored and repositioned between frames to minimize motion artifacts. All images were reconstructed using scanner‐specific parameters, quality‐checked for artifacts, and uploaded to the Laboratory of Neuroimaging for centralized storage and analysis.

For this study, we downloaded the fully preprocessed Digital Imaging and Communications in Medicine (DICOM) set 4# (“Co‐reg, Avg, Std Img, and Vox Siz, Uniform Resolution”).15 These standardized images were harmonized to a uniform spatial resolution using scanner‐specific 3D Gaussian smoothing kernels developed by the University of Michigan team, based on Hoffman phantom scans performed at each acquisition site.16 The target resolution was determined by the lowest‐performing scanners within ADNI and set to 6 mm full width at half maximum to ensure consistency across systems.17 Resulting preprocessed PET DICOMs have a standardized voxel size and spatial resolution.

2.2.2GMC acquisition

The amyloid‐PET images collected in GMC were acquired using [1
8F]flutemetamol tracers using a protocol previously described in detail.18 [1
8F]‐flutemetamol images were acquired 90 min after injection of 150 MBq for 20 min and were obtained on Siemens Biograph and Biograph Vision scanners (Siemens, Washington, DC), reconstructed using a 3D ordered subset expectation maximization iterative reconstruction, corrected for randoms, dead time, normalization, scatter, attenuation, and sensitivity. An in‐house workflow based on Statistical Parametric Mapping (SPM12; Wellcome Department of Cognitive Neurology, London, UK) was used for PET image processing.18

In both cohorts, voxel‐wise SUVR images were vectorized for each subject after resampling to 2 mm isotropic resolution and normalized to the default Montreal Neurological Institute (MNI) template using SPM12. All spatial co‐registrations were visually inspected by an experienced rater (MP) to ensure accurate alignment and consistent spatial resolution across all PET images. Two canonical amyloid PET reference patterns were used: one derived from AD patients included in the Centiloid Project19 and a second derived from an independent cohort of amyloid‐positive cognitively impaired individuals from the GMC who did not have available APOE ε4 genotyping. Information about the cohorts used to generate the canonical maps is provided in Table S1 and in the original study.19 Mean SUVR images were generated using the Image Calculator function in SPM12 to represent canonical AD amyloid distributions and were processed following the same preprocessing pipeline described earlier.18

2.3Statistical analysis

Regional differences in amyloid‐PET uptake among the three APOE ε4 groups were assessed using voxel‐wise analyses. Gray matter‐restricted SUVR images, obtained by applying a probabilistic gray matter mask from SPM and normalized to MNI space, were entered into analyses of variance (ANOVAs) followed by pairwise two‐sample t tests in SPM12 to compare the three APOE ε4 groups and each group pair, respectively. Age and sex were included as nuisance covariates, and the statistical threshold was set at α < 0.05 with a cluster extent threshold of >100 voxels.

For each participant, similarity to the canonical AD distribution was quantified by computing Pearson correlation coefficients between individual amyloid‐PET maps and each reference (i.e., Centiloid and GMC maps) using vectorized data restricted to the gray matter. Group differences in correlation coefficients across APOE ε4 groups were tested using linear regression models entering the correlation coefficient as dependent variable and APOE ε4 genotype as the primary predictor (i.e., three levels: homozygotes, heterozygotes, and non‐carriers), adjusting for age, sex, and global amyloid burden. Estimated marginal means (EMMs) and pairwise contrasts between APOE ε4 groups were performed when results reached significance (α < 0.05), using the emmeans package in R with Tukey's correction for multiple comparisons. Then we used spatial correlation analysis to evaluate the local similarity of amyloid distribution across groups. To this end, we extracted local amyloid patterns within clusters of 5 × 5 × 5 voxels and compared them to the corresponding pattern in the canonical map using correlation analysis (Pearson correlation). This approach is similar to that used in imaging studies to measure structural similarity20 or regional homogeneity in functional magnetic resonance imaging (fMRI) data.21 Maps of local amyloid similarity were then averaged within each APOE ε4 group to derive group‐level maps. From these correlation maps between APOE ε4 groups and canonical maps, we applied a winner‐take‐all approach to assign each voxel to the APOE ε4 group showing the highest similarity value. In parallel, we computed voxel‐wise difference maps (e.g., homozygotes vs non‐carriers and heterozygotes vs non‐carriers) derived from the same local‐correlation maps, allowing us to identify brain regions where one group exhibited stronger or weaker correspondence to the canonical pattern compared with another.

To explore within‐group similarity, we computed inter‐subject correlation matrices of amyloid‐PET patterns for each APOE ε4 group. Specifically, we applied the same correlation approach as above (Pearson correlation) on the amyloid‐PET maps of each pair of subjects, using vectorized data restricted to the gray matter. From these matrices, all unique subject pairs – n(n−1)/2 – were extracted and entered in a linear regression model to test for APOE ε4‐related differences in similarity. Correlation values were entered as the dependent variable, APOE ε4 group as predictor, and age, sex, and global amyloid burden as covariates. Since each matrix element corresponds to a pair of subjects, it is not possible to assign individual covariates; therefore, we computed combined covariates based on the mean age, sex, and Centiloid values of each matrix pair. This analysis follows the inter‐subject similarity approach used in fMRI studies, where dyads are the unit of analysis and pair‐level covariates are used to explain inter‐subject similarity.22 In case of significance, we applied the same methodology described earlier, using EMM with Tukey correction for post hoc comparisons. The degree of spatial homogeneity of amyloid‐PET patterns within each APOE ε4 and clinical stage groups was also evaluated voxel‐wise by the computation of standard deviation (SD) maps. For each group, we extracted gray‐matter‐masked PET values for all subjects and, at each voxel, calculated the SD across subjects. These metrics quantify the intersubject variability of the amyloid signal at the voxel level: lower values indicate greater homogeneity (i.e., subjects within the group show more similar regional uptake), whereas higher values indicate greater heterogeneity. The same winner‐take‐all approach was applied to SD maps to identify, for each voxel, the APOE ε4 group with the highest homogeneity.

Analyses were conducted in the overall cohort and stratified by clinical stage within both the ADNI and GMC cohorts. However, due to smaller subgroup sizes in the GMC validation cohort, statistical comparisons were limited to APOE ε4 carriers (with one or two alleles) versus non‐carriers. All statistical analyses and visualization were performed in MATLAB R2025a, R (version 4.2.3), and MRIcroGL (Python API).

3RESULTS

3.1Participants

A total of 676 amyloid‐positive participants from the ADNI cohort were included, spanning CU (n = 160), MCI (n = 331), and dementia (n = 185) stages. The mean age of ADNI participants was 74 years (SD = 7), and 49% were female. Regarding APOE ε4, carriers represented 63% of the sample, including 49% heterozygotes (n = 329) and 14% homozygotes (n = 98). When comparing APOE ε4 groups, significant differences emerged in age (p < 0.001), MMSE scores (p < 0.001), AV45 SUVR (p < 0.001), and hippocampal volume (p = 0.010). Homozygotes were younger, were more cognitively impaired, showed the highest amyloid burden, and had the smallest hippocampal volumes. Clinical stage distribution also differed across groups (p < 0.001), with homozygotes presenting fewer CU individuals and more dementia cases than non‐carriers and heterozygotes. Demographic and clinical characteristics are summarized in Table 1 by APOE ε4 and in Table S2 stratified by clinical stage.

An independent validation cohort comprised 105 amyloid‐positive individuals with MCI (n = 86) and dementia (n = 19) from the GMC. In this cohort, 59% were APOE ε4 carriers (n = 62; heterozygotes n = 50, homozygotes n = 12), and 41% were non‐carriers (n = 43). A summary of demographic characteristics and amyloid burden across APOE ε4 groups in the GMC cohort is provided in Table S3.

3.2Topography of amyloid across APOE ε4 groups

ANOVA analyses revealed widespread regional differences in amyloid uptake across APOE ε4 genotype groups (Figure 1), with prominent effects in lateral and medial frontal cortices, lateral temporal regions, precuneus, posterior cingulate cortex, and striatum. Pairwise voxel‐wise t tests showed that both APOE ε4 homozygotes and heterozygotes exhibited significantly higher amyloid deposition than non‐carriers across these regions, with the most extensive and robust effects observed in heterozygotes. Analyses conducted within each clinical stage showed similar effects in MCI, while in CU the effect was present only in heterozygotes. No detectable differences were observed in the dementia group (Supplementary Results and Figure S1). In the GMC validation cohort, voxel‐wise t tests comparing APOE ε4 carriers and non‐carriers showed higher amyloid burden in carriers, predominantly in the orbitofrontal and medial frontal cortices, as well as scattered neocortical regions (Figure S2).

3.3Similarity to canonical Alzheimer's disease amyloid maps

APOE ε4 genotype was a strong predictor of amyloid‐PET pattern similarity for both canonical amyloid maps. In ADNI, for the canonical Centiloid map, APOE ε4 explained a significant proportion of variance in similarity scores (F[2,669] = 52.8, p < 0.001), with post hoc contrasts indicating significantly lower similarity in non‐carriers compared with both APOE ε4 heterozygotes (β = −0.03, SE = 0.01, t[669] = −4.92, p < 0.001) and homozygotes (β = −0.05, SE = 0.01, t[669] = −5.46, p < 0.001). The difference between heterozygotes and homozygotes showed a trend toward statistical significance (p = 0.07) (Figure 2). A similar pattern was observed for the GMC canonical map, where APOE ε4 genotype was again a significant predictor (F[2,669] = 27.78, p < 0.001) (Figure S3). Pairwise comparisons revealed that non‐carriers displayed lower similarity than heterozygotes (β = −0.024, SE = 0.006, t[669] = −4.09, p = 0.001) and homozygotes (β = −0.035, SE = 0.009, t[669] = −4.09, p = 0.001), while similarity did not differ significantly between heterozygotes and homozygotes (p = 0.37). For both canonical maps, age did not contribute significantly to similarity scores (p > 0.5), whereas sex and global amyloid load remained significant predictors (all p < 0.001), indicating that the APOE ε4 effect persists beyond overall amyloid burden. When analyses were stratified by clinical stage, APOE ε4 genotype showed a consistent association with amyloid‐PET pattern similarity, with the strongest effects observed in earlier disease stages, namely, CU and MCI (full description in Supplementary Results and Figure S4). When we relied on a winner‐take‐all approach to assign each voxel to the APOE ε4 group showing the highest similarity to the canonical maps, we confirmed the foregoing results and observed clearer group‐level effects (Figure 2). Full statistical outputs, including model summaries and all pairwise post hoc comparisons, as well as amyloid‐PET spatial maps, are provided in Tables S4 to S25 and Figures S5 to S7.

Findings were further validated in the independent GMC cohort. In this cohort, APOE ε4 carriership was associated with higher similarity to the canonical Centiloid map (F[1,100] = 5.49, p = 0.021) and for the canonical GMC map (F[1,100] = 4.76, p = 0.031) (Figure 2). Across all models, global amyloid burden was a significant covariate, while age and sex showed smaller or inconsistent effects. In participants with dementia, the same models did not reveal a significant effect of APOE ε4 carriership (p > 0.05), whereas analyses conducted in the MCI group yielded results comparable to those observed in the whole sample (Figure S8).

3.4Within‐group spatial similarity of amyloid deposition

In the ADNI whole cohort, APOE ε4 genotype was strongly associated with within‐group spatial similarity of amyloid‐PET patterns (F[2, 33,796] = 902.5, p < 0.001). Pairwise Tukey‐adjusted comparisons showed that heterozygotes exhibited higher spatial homogeneity than non‐carriers (β = −0.023, SE = 0.001, t[33,796] = −21.15, p < 0.001), and homozygotes showed even higher homogeneity than non‐carriers (β = −0.056, SE = 0.002, t[33,796] = −25.1, p < 0.001). Homozygotes also demonstrated higher spatial similarity than heterozygotes (β = −0.033, SE = 0.002, t[33,796] = −15.7, p < 0.001), indicating a clear dose‐dependent effect of APOE ε4 allele on inter‐subject amyloid pattern convergence (Figure 3). Across clinical stages, the APOE ε4 genotype continued to influence within‐group spatial homogeneity of amyloid‐PET patterns, although the magnitude and direction of effects varied across disease progression. In CU individuals, this was the only stage without a clear dose‐dependent increase in spatial homogeneity across APOE ε4 genotypes, a pattern likely influenced by the very small number of APOE ε4 homozygotes (n = 7) in this group. In contrast, both MCI and dementia showed a dose‐dependent effect, with spatial similarity progressively increasing from non‐carriers to heterozygotes and homozygotes (full description in Supplementary Results and Figure S9). Winner‐take‐all maps revealed that APOE ε4 homozygotes exhibited the greatest spatial homogeneity across widespread cortical and subcortical regions, whereas non‐carriers showed the highest homogeneity primarily in frontal and striatal areas (Figure 3). Model summaries with pairwise post hoc comparisons, and within‐group amyloid‐PET spatial maps are provided in Tables S26 to S36 and Figures S10 to S11.

Results were replicated in the independent GMC validation cohort, where APOE ε4 carrier status was strongly associated with spatial homogeneity (F[1,2189] = 140.3, p < 0.001), with carriers showing higher within‐group similarity compared with non‐carriers (Figure 3). A similar effect was observed among participants with MCI and dementia separately, where APOE ε4 carriership was significantly associated with greater homogeneity of amyloid deposition, with carriers again exhibiting higher within‐group correlations (Figure S12). Across all models, greater global amyloid burden and older age were consistently associated with increased within‐group similarity, while sex effects were minimal.

4DISCUSSION

This study demonstrates that APOE ε4 shapes not only the magnitude but also the spatial predictability of amyloid deposition, in a gene–dose‐dependent manner. Beyond greater accumulation in AD‐vulnerable hubs, APOE ε4 carriers, and homozygotes in particular, exhibit amyloid patterns that are more tightly aligned with canonical AD topographies and more homogeneous across individuals, independent of overall amyloid burden. These effects were dose‐dependent, robust across PET tracers, and replicated across two independent cohorts.

The regions showing greater similarity and homogeneity among APOE ε4 carriers were localized in classical hubs of amyloid deposition, including the posterior cingulate cortex, precuneus, inferior parietal, and lateral temporal cortices, which are also the core regions used for Centiloid quantification.19 This spatial distribution aligns with prior studies showing that APOE ε4 carriers exhibit higher amyloid burden in these areas across disease stages. In CU individuals, APOE ε4 has been linked to early amyloid deposition in medial temporal and inferior parietal regions,13 while in mild to moderate AD, higher amyloid levels have been observed in temporoparietal regions,23 and in dementia, relatively lower frontal amyloid load has been reported among APOE ε4 carriers compared with non‐carriers.24, 25 Consistent with this literature, our findings revealed pronounced topographical differences mainly in the CU and MCI stages, suggesting a possible ceiling effect at the dementia stage where widespread amyloid deposition may mask APOE ε4‐related spatial distinctions.11, 26 While cortical saturation in dementia reduces the specificity of correlations with the canonical amyloid pattern, the persistence of high within‐group homogeneity in APOE ε4 carriers suggests that this saturation occurs in a highly synchronized and stereotyped manner. This is particularly relevant for APOE ε4 carriers, especially homozygotes, who not only exhibit higher overall amyloid burden but also tend to reach positivity thresholds much earlier, both in PET and cerebrospinal fluid measures,11, 12 and tend to plateau in amyloid accumulation sooner than non‐carriers.27 Indeed, amyloid positivity is two to three times more prevalent in APOE ε4 carriers compared with non‐carriers among CU individuals,12 and prior studies suggested that this group continued to accumulate amyloid even after surpassing the positivity threshold.28 Together, these observations support the idea that APOE ε4 shapes amyloid deposition both spatially and temporally, exerting its strongest effects in the early phases of the disease.

These findings provide topographic evidence in support of the probabilistic model of AD, demonstrating that APOE ε4 gene dose constrains the spatial patterning of amyloid accumulation, such that carriers, and homozygotes in particular, follow a less variable and more stereotyped trajectory. Whether individuals with ADAD would exhibit even greater topographic stereotypy than APOE ε4 homozygotes, as the probabilistic model predicts, remains an important hypothesis for future work.4 Nonetheless, conceptually these results align with the aforementioned model postulating a gradient of amyloid predictability driven by genetic determinants, in which amyloid precursor protein (APP), presenilin 1 (PSEN1), and PSEN2 mutations define a nearly fixed biological trajectory of pathology,29, 30 followed by APOE ε4 homozygotes and heterozygotes, and finally non‐carriers,13 whose amyloid accumulation may be more influenced by stochastic or environmental factors. This interpretation is biologically plausible because the genes responsible for autosomal dominant AD and APOE ε4 share amyloidogenic mechanisms. Mutations in APP, PSEN1, and PSEN2 promote early and widespread fibrillar amyloid deposition in highly stereotyped cortical networks,29 whereas APOE ε4 primarily promotes impaired clearance, enhanced aggregation, and increased seeding of amyloid.31 Despite acting through different molecular mechanisms (i.e., overproduction vs impaired clearance), both pathways converge on a common downstream phenotype of excessive fibrillar amyloid deposition that follows a remarkably conserved topographical pattern across genetic and sporadic forms of AD. In line with this, our observation of a gene–dose effect, with homozygotes showing greater amyloid spatial predictability than heterozygotes, further supports this gradient of biological determinism. This pattern is consistent with prior evidence showing that the frequency of amyloid positivity6, 12, 32 and the earlier onset of amyloid accumulation7, 27 both increase with the number of APOE ε4 alleles, reinforcing the idea that APOE ε4 dosage modulates not only the risk but also the spatial expression of amyloid pathology.

Our approach fits within a growing body of work that moves beyond regional amyloid burden to characterize the topographic organization of pathology at the individual level. Prior analyses applied similarity‐based methods in structural MRI and resting‐state fMRI to capture inter‐individual convergence of brain organization33, 34 and the spatial heterogeneity of tau and cortical atrophy using cluster‐based subtyping approaches.35 Building on conceptually similar approaches in neuroimaging, our work introduces spatial similarity and within‐group homogeneity of amyloid‐PET distributions as quantitative measures of pathological stereotypy, an aspect of AD biology that may serve as a sensitive readout for stratifying patients by the predictability of their disease trajectory.

Although this study advances our understanding of how APOE ε4 influences amyloid topography, several limitations should be considered when interpreting the findings. First, the cross‐sectional design prevents inference on the longitudinal pattern of amyloid pathology, a common constraint in AD studies using PET imaging.9 This limitation also extends to the disease timing, as the APOE ε4 gene dose reduces the age of onset of amyloidosis; therefore, chronological age, included as a covariate in our models, does not consistently serve as a proxy for time since amyloid onset. Although we stratified analyses by clinical stage to approximate disease progression, longitudinal investigations will be essential to establish whether the similarity and homogeneity measures identified here are detectable at the individual level. Second, the limited number of APOE ε4 homozygotes, particularly among CU individuals (n = 7), may have reduced statistical power to detect dose‐dependent effects in the earliest disease stages. Third, differences in amyloid tracers across datasets should also be acknowledged. ADNI participants were scanned with [1
8F]florbetapir, whereas the validation cohort from GMC used [1
8F]flutemetamol. Despite this methodological variability, the APOE ε4‐related effects on spatial similarity and homogeneity were replicated across cohorts, and both canonical reference maps, each derived from datasets using different tracers, produced consistent results. This cross‐tracer reproducibility strengthens the robustness and generalizability of our findings. Finally, future studies should incorporate longitudinal and multimodal imaging to determine whether the increased spatial homogeneity observed here predicts faster tau propagation or cognitive decline and whether it does so independently of the earlier onset of amyloidosis associated with APOE ε4 carriage. Such work will also help clarify whether the phenotypic convergence of amyloid patterns in APOE ε4 carriers extends to tau pathology and neurodegeneration. Recent longitudinal PET evidence indicates that APOE ε4 amplifies amyloid‐related tau accumulation, suggesting that its effects on tau spread are mostly mediated by amyloid pathology,28, 36, 37 and that this relationship is moderated by age and clinical stage.11 Conversely, other evidence points to a partially amyloid‐independent pathway, in which APOE ε4 may directly promote early tau deposition in initial Braak regions,28, 38, 39 while at later stages its effect on tau spread becomes increasingly dependent on amyloid pathology.40 These observations underscore the need to extend similarity and homogeneity analyses to longitudinal and tau‐PET data, to test whether the spatial convergence we identified in amyloid patterns is also present in tau and whether it reflects a modulatory role of APOE ε4 in the trajectory of the amyloid cascade.

Overall, the findings of this study contribute to the broader debate on the deterministic versus stochastic nature of AD pathology.4 Although APOE ε4 is the strongest common genetic risk factor for sporadic AD, it is not fully deterministic as many carriers remain CU even at advanced ages.9, 10 Nevertheless, the greater spatial homogeneity and stereotyped amyloid topographies observed in APOE ε4 carriers suggest that this allele also confers a more biologically constrained and predictable course of amyloid deposition compared with non‐carriers. This pattern supports the notion of a more deterministic progression of AD pathology in the early disease stages among APOE ε4 carriers, particularly homozygotes. These findings are directly related to the amyloid substrate that anti‐amyloid treatments target: A more homogeneous and canonically distributed amyloid accumulation reduces variability among individuals in the therapeutic target and reinforces the rationale for stratifying anti‐amyloid trials by APOE ε4 genotype. Whether this translates into differential treatment benefit remains unclear: Pooled analyses across anti‐amyloid trials have suggested a comparable or slightly more favorable effect in APOE ε4 carriers,41 whereas the pivotal trials showed little evidence of benefit in homozygotes, even without adjustment for baseline amyloid burden or clinical severity.42, 43 What our findings do show is that amyloid accumulates along a more predictable spatial course in APOE ε4 carriers, and particularly in homozygotes, a pattern that could be useful for understanding how amyloid spreads in AD.

CONFLICTS OF INTEREST STATEMENT

GBF has received consulting fees through his institution from Biogen, Diadem, Roche, Eisai, Eli Lilly, Ac Immune, Novo Nordisk, Schwabe, Bromatech, AtonRâ, World Clinical Trials, and J&J Innovative Medicine. GBF has received payment or honoraria for lectures, presentations, speaker bureaus, manuscript writing, or educational events through his institution from Biogen, Roche, Novo Nordisk, GE HealthCare, and Vifor Pharma. VG received research support and speaker fees through her institution from GE Healthcare, Siemens Healthineers, Novo Nordisk, Janssen, and Novartis. All other authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

In both cohorts, written informed consent was obtained from all participants or their legally authorized representatives at the time of enrollment, and all research was conducted in accordance with the Declaration of Helsinki.

Supporting information

REFERENCES

1Hardy JA , Higgins GA . Alzheimer's disease: the amyloid cascade hypothesis . Science (1979) . 1992 ;256 (5054 ):184 ‐185 . doi:10.1126/SCIENCE.1566067
1566067

2Mendes AJ , Ribaldi F , Pievani M , Boccalini C , Garibotto V , Frisoni GB . Validating the Amyloid cascade through the revised criteria of Alzheimer's association workgroup 2024 for Alzheimer disease . Neurology . 2025 ;104 (11 ):e213675. doi:10.1212/WNL.0000000000213675 PMC12079574 40359457

3Jagust WJ , Teunissen CE , DeCarli C . The complex pathway between amyloid β and cognition: implications for therapy . Lancet Neurol . 2023 ;22 (9 ):847 ‐857 . doi:10.1016/S1474-4422(23)00128-X
37454670

4Frisoni GB , Altomare D , Thal DR , et al. The probabilistic model of Alzheimer disease: the amyloid hypothesis revised . Nat Rev Neurosci . 2022 ;23 (1 ):53 ‐66 . doi:10.1038/S41583-021-00533-W
PMC8840505 34815562

5Corder EH , Saunders AM , Strittmatter WJ , et al. Gene dose of apolipoprotein E type 4 Allele and the risk of Alzheimer's disease in late onset families . Science (1979) . 1993 ;261 (5123 ):921 ‐923 . doi:10.1126/SCIENCE.8346443
8346443

6Therriault J , Pascoal TA , Benedet AL , et al. Frequency of biologically defined Alzheimer disease in relation to age, sex, APOE ϵ4, and cognitive impairment . Neurology . 2021 ;96 (7 ):E975 ‐E985 . doi:10.1212/WNL.0000000000011416
PMC8055338 33443136

7Betthauser TJ , Bilgel M , Koscik RL , et al. Multi‐method investigation of factors influencing amyloid onset and impairment in three cohorts . Brain . 2022 ;145 (11 ):4065 ‐4079 . doi:10.1093/BRAIN/AWAC213
PMC9679170 35856240

8Genin E , Hannequin D , Wallon D , et al. APOE and Alzheimer disease: a major gene with semi‐dominant inheritance . Mol Psychiatry . 2011 ;16 (9 ):903 ‐907 . doi:10.1038/mp.2011.52
PMC3162068 21556001

9Reiman EM , Ghisays V , Langbaum JB . The risk of Alzheimer disease in APOE4 homozygotes . JAMA Neurol . 2025 ;82 (9 ):879 ‐880 . doi:10.1001/JAMANEUROL.2025.0639
PMC12439451 40227662

10Pappalettera C , Carrarini C , Cappa S , et al. Challenges to identifying risk versus protective factors in Alzheimer's disease . Nat Med . 2024 ;30 (11 ):3094 ‐3095 . doi:10.1038/s41591-024-03158-5
39025966

11Fortea J , Pegueroles J , Alcolea D , et al. APOE4 homozygosity represents a distinct genetic form of Alzheimer's disease . Nat Med . 2024 ;30 (5 ):1284 ‐1291 . doi:10.1038/s41591-024-02931-w
PMC13310155 38710950

12Jansen WJ , Janssen O , Tijms BM , et al. Prevalence estimates of amyloid abnormality across the Alzheimer disease clinical spectrum . JAMA Neurol . 2022 ;79 (3 ):228 ‐243 . doi:10.1001/JAMANEUROL.2021.5216
PMC12138908 35099509

13Cacciaglia R , Salvadó G , Molinuevo JL , et al. Age, sex and APOE‐ε4 modify the balance between soluble and fibrillar β‐amyloid in non‐demented individuals: topographical patterns across two independent cohorts . Mol Psychiatry . 2022 ;27 (4 ):2010 ‐2018 . doi:10.1038/s41380-022-01436-7
PMC9126807 35236958

14Joshi AD , Pontecorvo MJ , Clark CM , et al. Performance characteristics of amyloid PET with Florbetapir F 18 in patients with Alzheimer's disease and cognitively normal subjects . J Nucl Med . 2012 ;53 (3 ):378 ‐384 . doi:10.2967/JNUMED.111.090340
22331215

15Landau SM , Harrison TM , Baker SL , et al. Positron emission tomography harmonization in the Alzheimer's disease neuroimaging initiative: a scalable and rigorous approach to multisite amyloid and tau quantification . Alzheimers Dement . 2025 ;21 (1 ):e14378. doi:10.1002/ALZ.14378
PMC11772732 39559932

16Joshi A , Koeppe RA , Fessler JA . Reducing between scanner differences in multi‐center PET studies . Neuroimage . 2009 ;46 (1 ):154 ‐159 . doi:10.1016/J.NEUROIMAGE.2009.01.057
PMC4308413 19457369

17Weiner MW , Veitch DP , Aisen PS , et al. The Alzheimer's disease neuroimaging initiative 3: continued innovation for clinical trial improvement . Alzheimers Dement . 2017 ;13 (5 ):561 ‐571 . doi:10.1016/J.JALZ.2016.10.006
PMC5536850 27931796

18Dodich A , Mendes A , Assal F , et al. The A/T/N model applied through imaging biomarkers in a memory clinic . Eur J Nucl Med Mol Imaging . 2020 ;47 (2 ):247 ‐255 . doi:10.1007/S00259-019-04536-9
31792573

19Klunk WE , Koeppe RA , Price JC , et al. The centiloid project: standardizing quantitative amyloid plaque estimation by PET . Alzheimers Dement . 2015 ;11 (1 ):1 ‐15.e4 . doi:10.1016/J.JALZ.2014.07.003
PMC4300247 25443857

20Tijms BM , Seris P , Willshaw DJ , Lawrie SM . Similarity‐based extraction of individual networks from gray matter MRI scans . Cereb Cortex . 2012 ;22 (7 ):1530 ‐1541 . doi:10.1093/CERCOR/BHR221
21878484

21Zang Y , Jiang T , Lu Y , He Y , Tian L . Regional homogeneity approach to fMRI data analysis . Neuroimage . 2004 ;22 (1 ):394 ‐400 . doi:10.1016/J.NEUROIMAGE.2003.12.030
15110032

22Parkinson C , Kleinbaum AM , Wheatley T . Similar neural responses predict friendship . Nat Commun . 2018 ;9 (1 ):332 . doi:10.1038/s41467-017-02722-7 PMC5790806 29382820

23Drzezga A , Grimmer T , Henriksen G , et al. Effect of APOE genotype on amyloid plaque load and gray matter volume in Alzheimer disease . Neurology . 2009 ;72 (17 ):1487 ‐1494 . doi:10.1212/WNL.0B013E3183A2E8D0
19339712

24Lehmann M , Ghosh PM , Madison C , et al. Greater medial temporal hypometabolism and lower cortical amyloid burden in ApoE4‐positive AD patients . J Neurol Neurosurg Psychiatry . 2014 ;85 (3 ):266 ‐273 . doi:10.1136/JNNP-2013-305858
PMC3946299 23965289

25Ossenkoppele R , Van Der Flier WM , Zwan MD , et al. Differential effect of APOE genotype on amyloid load and glucose metabolism in AD dementia . Neurology . 2013 ;80 (4 ):359 ‐365 . doi:10.1212/WNL.0B013E31827F0889
23255822

26Wang Y , Li F , Qin Q , et al. Influence of APOE ε4 on performance of CSF biomarkers in differentiating clinical Alzheimer's disease . J Prev Alzheimers Dis . 2025 ;12 (4 ):100065. doi:10.1016/J.TJPAD.2025.100065
PMC12184029 39827005

27Ghisays V , Goradia DD , Protas H , et al. Brain imaging measurements of fibrillar amyloid‐β burden, paired helical filament tau burden, and atrophy in cognitively unimpaired persons with two, one, and no copies of the APOE ε4 allele . Alzheimers Dement . 2020 ;16 (4 ):598 ‐609 . doi:10.1061/J.ALZ.2019.08.195
PMC7187298 31831374

28Young CB , Johns E , Kennedy G , et al. APOE effects on regional tau in preclinical Alzheimer's disease . Mol Neurodegener . 2023 ;18 (1 ):1 ‐14 . doi:10.1186/S13024-022-00590-4
PMC9811772 36597122

29Gordon BA , Blazey TM , Su Y , et al. Spatial patterns of neuroimaging biomarker change in individuals from families with autosomal dominant Alzheimer's disease: a longitudinal study . Lancet Neurol . 2018 ;17 (3 ):241 ‐250 . doi:10.1016/S1474-4422(18)30028-0
PMC5816717 29397305

30Levitis E , Vogel JW , Funck T , et al. Differentiating amyloid beta spread in autosomal dominant and sporadic Alzheimer's disease . Brain Commun . 2022 ;4 (3 ):fcac085. doi:10.1093/BRAINCOMMS/FCAC085
PMC9116976 35602652

31Raulin AC , Doss SV , Trottier ZA , Ikezu TC , Bu G , Liu CC . ApoE in Alzheimer's disease: pathophysiology and therapeutic strategies . Mol Neurodegener . 2022 ;17 (1 ):1 ‐26 . doi:10.1186/S13024-022-00574-4
PMC9644639 36348357

32Saddiki H , Fayosse A , Cognat E , et al. Age and the association between apolipoprotein E genotype and Alzheimer disease: a cerebrospinal fluid biomarker–based case–control study . PLoS Med . 2020 ;17 (8 ):e1003289. doi:10.1371/JOURNAL.PMED.1003289
PMC7446786 32817639

33Chen TY , Zhu JD , Tsai SJ , Yang AC . Exploring morphological similarity and randomness in Alzheimer's disease using adjacent grey matter voxel‐based structural analysis . Alzheimers Res Ther . 2024 ;16 (1 ):1 ‐14 . doi:10.1186/S13195-024-01448-1
PMC11036786 38654366

34Bellana B , Liu ZX , Diamond NB , Grady CL , Moscovitch M . Similarities and differences in the default mode network across rest, retrieval, and future imagining . Hum Brain Mapp . 2017 ;38 (3 ):1155 ‐1171 . doi:10.1002/HBM.23445
PMC6866996 27774695

35Jeon S , Kang JM , Seo S , et al. Topographical heterogeneity of Alzheimer's disease based on MR imaging, tau PET, and amyloid PET . Front Aging Neurosci . 2019 ;10 (JUL ):458419. doi:10.3389/FNAGI.2019.00211/BIBTEX
PMC6710378 31481888

36Cai Y , Du J , Li A , et al. Initial levels of β‐amyloid and tau deposition have distinct effects on longitudinal tau accumulation in Alzheimer's disease . Alzheimers Res Ther . 2023 ;15 (1 ):1 ‐14 . doi:10.1186/S13195-023-01178-W
PMC9903587 36750884

37Ferrari‐Souza JP , Bellaver B , Ferreira PCL , et al. APOEε4 potentiates amyloid β effects on longitudinal tau pathology . Nat Aging . 2023 ;3 (10 ):1210 ‐1218 . doi:10.1038/s43587-023-00490-2
PMC10592050 37749258

38Ferrari‐Souza JP , Lussier FZ , Leffa DT , et al. APOEε4 associates with microglial activation independently of Aβ plaques and tau tangles . Sci Adv . 2023 ;9 (14 ):eade1474. doi:10.1126/SCIADV.ADE1474
PMC10075966 37018391

39Dincer A , Chen CD , McKay NS , et al. APOE ε4 genotype, amyloid‐β, and sex interact to predict tau in regions of high APOE mRNA expression . Sci Transl Med . 2022 ;14 (671 ):eabl7646. doi:10.1126/SCITRANSLMED.ABL7646
PMC9912474 36383681

40Salvadó G , Grothe MJ , Groot C , et al. Differential associations of APOE‐ε2 and APOE‐ε4 alleles with PET‐measured amyloid‐β and tau deposition in older individuals without dementia . Eur J Nucl Med Mol Imaging . 2021 ;48 (7 ):2212 ‐2224 . doi:10.1007/S00259-021-05192-8
PMC8175302 33521872

41Evans CD , Sparks JD , Andersen SW , et al. APOE ε4's impact on response to amyloid therapies in early symptomatic Alzheimer's disease: analyses from multiple clinical trials . Alzheimers Dement . 2023 ;19 (12 ):5407 ‐5417 . doi:10.1002/ALZ.13128
37204338

42Sims JR , Zimmer JA , Evans CD , et al. Donanemab in early symptomatic Alzheimer disease: the TRAILBLAZER‐ALZ 2 randomized clinical trial . JAMA, J Am Med Assoc . 2023 ;330 (6 ):512 ‐527 . doi:10.1001/JAMA.2023.13239
PMC10352931 37459141

43van Dyck CH , Swanson CJ , Aisen P , et al. Lecanemab in early Alzheimer's disease . N Engl J Med . 2023 ;388 (1 ):142 ‐143 . doi:10.1056/NEJMOA2212948
36449413