Longitudinal brain imaging in preclinical Alzheimer disease: impact of APOE ε4 genotype

The longitudinal impact of apolipoprotein E ɛ4 on individual disease trajectories remains unclear. Mishra et al. show that group level effects on atrophy and β-amyloid are driven by ɛ4 carriers being more likely to develop Alzheimer’s disease, doing so earlier, and accumulating more β-amyloid even when matched for baseline pathology.

Abstract

While prior work reliably demonstrates that the APOE ɛ4 allele has deleterious group level effects on Alzheimer disease pathology, the homogeneity of its influence across the lifespan and spatially in the brain remains unknown. Further it is unclear what combinations of factors at an individual level lead to observed group level effects of APOE genotype. To evaluate the impact of the APOE genotype on disease trajectories, we examined longitudinal MRI and PET imaging in a cohort of 497 cognitively normal middle and older aged participants. A whole-brain regional approach was used to evaluate the spatial effects of genotype on longitudinal change of amyloid-β pathology and cortical atrophy. Carriers of the ɛ4 allele had increased longitudinal accumulation of amyloid-β pathology diffusely through the cortex, but the emergence of this effect across the lifespan differed greatly by region (e.g. age 49 in precuneus, but 65 in the visual cortex) with the detrimental influence already being evident in some regions in middle age. This increased group level effect on accumulation was due to a greater proportion of ɛ4 carriers developing amyloid-β pathology, on average doing so at an earlier age, and having faster amyloid-β accumulation even after accounting for baseline amyloid-β levels. APOE ɛ4 carriers displayed faster rates of structural loss in primarily constrained to the medial temporal lobe structures at around 50 years, although this increase was modest and proportional to the elevated disease severity in APOE ɛ4 carriers. This work indicates that influence of the APOE gene on pathology can be detected starting in middle age.

Introduction

Alzheimer’s disease is the most common cause of dementia in individuals above the age of 60. The most significant genetic modifier of late-onset Alzheimer disease is the apolipoprotein E (APOE) gene (Corder et al., 1993; Saunders et al., 1993; Strittmatter et al., 1993; Reiman et al., 1996). APOE is a protein that functions as a ligand in receptor-mediated endocytosis of lipoprotein particles in the CNS and the periphery (Holtzman and Herz, 2012). The ɛ4 allele of APOE increases the risk of developing Alzheimer’s disease and lowers the age of onset, while the ɛ2 allele is associated with a lower risk of Alzheimer’s disease (Corder et al., 1993; Reiman et al., 2009). The association between the presence of an APOE ɛ4 allele and the clinical risk of Alzheimer’s disease has been well established in the literature (Farrer et al., 1997). While the exact mechanism of the relationship is still an active area of research (Mahoney-Sanchez et al., 2016), one of the major effects of the APOE protein on Alzheimer’s disease risk is that it influences amyloid-β aggregation and clearance (Huynh et al., 2017).

The pathological hallmarks of Alzheimer’s disease are neurofibrillary tangles and amyloid-β plaques (Braak and Braak, 1991), and significant work has been done evaluating the effect of APOE ɛ4 allele on these biomarkers. Previous studies of pathology, CSF and PET imaging of the brain consistently find a relationship between APOE ɛ4 genotype and elevated amyloid-β burden (Schmechel et al., 1993; Morris et al., 2010; Rowe et al., 2010; Johnson et al., 2013; Mathis et al., 2013; Murphy et al., 2013; Villemagne and Rowe, 2013; Risacher et al., 2015; Gottesman et al., 2016). Whereas studies overwhelmingly find that the APOE ɛ4 genotype is associated with worse amyloid-β pathology, its effect on markers of neurodegeneration such as CSF tau, fludeoxyglucose PET, and structural MRI are less consistent. Studies have reported both an influence of the APOE ɛ4 allele (Reiman et al., 1996; Galasko et al., 1998; Liu et al., 2010; Jagust et al., 2012; Hostage et al., 2013; Lehmann et al., 2013; Roussotte et al., 2014) and no effect (Soininen et al., 1995; Jack et al., 1998; Reiman et al., 1998; Sunderland et al., 2004; Drzezga et al., 2009; Fan et al., 2010).

Perhaps most importantly, the fundamental interpretation of APOE ɛ4 effects on Alzheimer’s disease pathology has focused on general group effects while ignoring parallel ways such effects could arise. For example, it is unclear from the literature whether population level APOE ɛ4 effects on amyloid-β pathology are a result of a higher proportion of APOE ɛ4 carriers developing Alzheimer’s disease pathology, ɛ4 carriers developing Alzheimer’s disease pathology at an earlier age, whether ɛ4 carriers have an increased rate of amyloid-β accrual, or a combination of the three. These questions are critical to understanding the influence of APOE on Alzheimer’s disease as well as for clinical trials that are using ɛ4 genotype or PET amyloid-β levels to select participants (Reiman et al., 2011; Sperling et al., 2014). Disentangling these questions may be best addressed through the investigation of APOE ɛ4 effects using longitudinal data in a population that spans both middle and older age.

Reports on the relationship of APOE ɛ4 carrier status on longitudinal accumulation of amyloid-β pathology have been mixed, with studies showing that the APOE ɛ4 genotype increases the rate of change of amyloid-β pathology (Grimmer et al., 2010; Villemagne et al., 2011; Jack et al., 2013b), while others find no genotype effect (Vlassenko et al., 2011; Resnick et al., 2015). When examining APOE genotype effects across the lifespan, prior works have reported a significant ɛ4 and age interaction, (Fleisher et al., 2013; Scheinin et al., 2014; Jansen et al., 2015) although this is not always the case (Rodrigue et al., 2012).

The most common neurodegenerative biomarker studies with APOE have been structural MRI. Both cross-sectional (Hashimoto et al., 2001; Liu et al., 2010; Hostage et al., 2013; Manning et al., 2014) and longitudinal (Geroldi et al., 1999; Mori et al., 2002; Van De Pol et al., 2007; Morra et al., 2009; Risacher et al., 2010; Hostage et al., 2014; Manning et al., 2014) studies have reported significant effects of APOE genotype on hippocampal volume or cortical thickness in diseased populations, but other studies have shown no effect (Soininen et al., 1995; Jack et al., 1998; Reiman et al., 1998; Cohen et al., 2001; Lemaitre et al., 2005; Drzezga et al., 2009; Schuff et al., 2009; Fan et al., 2010; Leung et al., 2013). A similar pattern is seen within populations of only cognitively normal older adults, with studies finding differences between carriers and non-carriers on rates of change in medial temporal lobe structures (Moffat et al., 2000; Chen et al., 2007; Risacher et al., 2010; Lu et al., 2011; Cohen et al., 2013), while other work does not (Du et al., 2006; Taylor et al., 2014).

This heterogeneity in the literature may be attributed to inconsistencies in the populations studied, insufficient longitudinal follow-up, and relatively modest participant cohorts. The majority of the studies evaluating the effect of APOE ɛ4 on longitudinal changes in amyloid-β have either exclusively looked at impaired (Alzheimer’s disease or mild cognitive impairment) patients or combined healthy controls with cognitively impaired individuals (Jack et al., 2009, 2013b; Grimmer et al., 2010; Villemagne et al., 2011; Bilgel et al., 2016). Changes in Alzheimer’s disease pathology can begin decades before the onset of dementia (Bateman et al., 2012; Jack et al., 2013a). The focus of APOE studies on older and already impaired individuals has left the question of how the APOE ɛ4 genotype modulates early changes in Alzheimer’s disease pathology largely unanswered.

Pathology is not uniformly distributed, but has a spatial evolution in the brain as the disease progresses (Benzinger et al., 2013; Gordon et al., 2014). The majority of studies in the literature have looked at whole-brain summary measures of amyloid-β (Morris et al., 2010; Rowe et al., 2010; Johnson et al., 2013; Mathis et al., 2013; Risacher et al., 2015; Gottesman et al., 2016). Of those that have used a regional approach, some have not seen regional variability in ɛ4 effects (Murphy et al., 2013), while others have found effects in primarily frontal cortex regions (Reiman et al., 2009; Scheinin et al., 2014) or more posterior temporal-parietal regions (Fleisher et al., 2013). The APOE ɛ4 relationship on structural MRI has primarily focused on hippocampus or surrounding medial temporal lobe structures, with a minority of studies utilizing a whole-brain voxel-wise or regions of interest approach (Geroldi et al., 1999; Lu et al., 2011; Tosun et al., 2011; Hostage et al., 2014). While the literature demonstrates that the ɛ4 allele has an overall deleterious effect on amyloid-β pathology and neurodegenerative biomarkers, there are unanswered questions about where in the brain these effects spatially manifest.

The objective of this study was to utilize longitudinal data to characterize the effect of APOE ɛ4 on amyloid-β pathology as measured by PET and neurodegeneration as measured by structural MRI. Here, we attempt to understand the nature of the APOE ɛ4 effects by using a whole-brain regional approach to evaluate the spatial effects of APOE ɛ4 on longitudinal rates of change of pathology. Additionally, we explore whether ɛ4 carriers develop Alzheimer’s disease pathology at a higher proportion, at an earlier time point, or a faster rate than non-carriers. These analyses address shortcomings in the literature to better understand the role that the APOE ɛ4 genotype plays on regional pathology across the lifespan.

Materials and methods

Participants

Participants between the ages of 45 and 90 were included from ongoing studies on ageing and dementia from the Knight Alzheimer Disease Research Center at Washington University. Participants who had one or more ɛ4 allele were assigned a positive APOE ɛ4 status, while those with no ɛ4 allele were assigned a negative APOE ɛ4 status.

The longitudinal analyses of amyloid-β PET and structural MRI were analysed only in individuals who were cognitively normal at baseline [clinical dementia rating (CDR) of 0 (Morris, 1997)]. In this initial cohort 249 participants were identified that had two or more amyloid-β PET scans and 497 participants had two or more MRI sessions. Demographics for the entire cohort are shown in Table 1. Individuals with both an ɛ4 and an ɛ2 allele were excluded, yielding a final sample for statistical analyses of 242 with longitudinal PET and 482 with longitudinal MRI. Demographic comparisons between ɛ4 carrier and ɛ4 non-carrier groups are presented in the online Supplementary material. There were no significant demographic differences between carriers and non-carriers.

Table 1.

Longitudinal cohort demographics table for estimating rates of biomarker change

	PIB	MRI
Participants, n	249	497
Age (SD)	64.8 (9.4)	66.8 (10.0)
Gender, n male (%)	83 (33.3)	189 (38.0)
Education (SD)a	15.9 (2.5)	15.8 (2.5)
MMSE (SD)	29.1 (1.1)	29.1 (1.1)
CDR Sum of Boxes (SD)	0.01 (0.08)	0.02 (0.10)
Scans, n, mean (SD)	2.5 (0.6)	3.1 (1.3)
Years of follow-up (SD)	4.9 (2.1)	5.6 (3.1)
n APOE ɛ4a (ɛ34 or ɛ44) (%)	74 (29.7)	150 (30.2)
ɛ22/ɛ23/ɛ33/ɛ24b/ɛ34/ɛ44, n	3/30/135/7/63/11	4/59/269/15/127/23
[%]	[1.2/12.0/54.2/2.8/25.3/4.4]	[0.8/11.9/54.1/3.0/25.6/4.6]
n that converted to CDR >0	13 (5.2%)	89 (17.9%)

^aEducation values were not available for 23 individuals.

^bIndividuals with both an ɛ2 and ɛ4 allele were excluded from analysis.

MMSE = Mini-Mental State Examination.

It was of additional interest to estimate population level frequencies of an abnormal amyloid-β PET scan across middle and older age. To do this we examined the baseline visits from the 242 cognitively normal individuals included in the longitudinal PET data, as well as another 333 (total n = 575) individuals with only one PET session. Individuals carrying both an ɛ4 and an ɛ2 were excluded from this sample. From this total population 443 were CDR = 0, 70 CDR = 0.5, 40 CDR = 1, 19 CDR = 2, and 3 CDR = 3. Cognitively impaired individuals were included in this one analyses as excluding them would erroneously underestimate the true percentages of the population with abnormal amyloid-β PET scans. Demographics for this cross-sectional cohort are shown in Table 2.

Table 2.

Cross-sectional demographics table for estimating population rates of amyloid-β positivity

Participants, n	575
Age (SD)	67.7 (9.9)
Gender, n male (%)	239 (41.6)
CDR	443 CDR 0, 70 CDR 0.5, 40 CDR 1, 19 CDR 2, 3 CDR 3
ɛ22/ɛ23/ɛ33/ɛ34/ɛ44, n	4/69/297/176/29
[%]	[0.7/12.0/51.7/30.6/5.0]
n APOE ɛ4+ (ɛ34 or ɛ44) (%)	205 (35.6)

MRI acquisition and processing

Structural magnetization-prepared rapid gradient-echo (MPRAGE) images were acquired on either a 1.5 T (n = 336 sessions) or 3 T (n = 1199 sessions) Siemens. Scans had a resolution of either 1 × 1 × 1.25 mm or 1 × 1 × 1 mm. Structural scans were processed with FreeSurfer (Fischl, 2012) using the Desikan atlas. For each hemisphere, cortical thickness values were obtained for all FreeSurfer cortical regions of interest, and volumes were obtained for all FreeSurfer subcortical regions of interest. Cortical thickness was calculated as the shortest distance between the cortical grey/white boundary to the grey/CSF boundary (Fischl and Dale, 2000). All subcortical region volumes were adjusted for intracranial volume using a regression approach (Buckner et al., 2004). Left and right values for each region of interest were averaged together.

PIB PET acquisition and processing

Amyloid-β PET imaging was completed using ¹¹C-Pittsburgh compound B (PIB). PET data from the 30–60-min post-injection window were analysed using regions of interests derived from Freesurfer (Fischl, 2012; Su et al., 2013) (PET Unified Pipeline, https://github.com/ysu001/PUP). Regional estimates were transformed into standardized uptake value ratios (SUVRs) with cerebellar cortex as the reference region. Partial volume correction was performed using a regional spread function technique (Rousset et al., 1998; Su et al., 2015). As with the MRI data, regions were averaged across hemispheres before being entered into statistical analyses. For an examination of population level frequencies of abnormal amyloid-β deposition, participants were additionally classified as having abnormal (amyloid-β positive) or normal (amyloid-β negative) levels using a previously established PIB mean cortical SUVR cut-off of 1.42 (Sutphen et al., 2015; Brier et al., 2016; Vlassenko et al., 2016; Mishra et al., 2017), which is the linear projection of a mean cortical binding potential (MCBP) cut-off previously established in this cohort of 0.18 (Mintun et al., 2006; Vlassenko et al., 2011; Gordon et al., 2015).

Genotyping

Genomic DNA was isolated from peripheral blood samples using standard procedures. APOE genotyping was performed as previously described (Talbot et al., 1994). The distribution of alleles was in Hardy-Weinberg equilibrium.

Statistical analysis

We used multivariate linear mixed effects models (LME) to evaluate longitudinal changes in amyloid-β PET and structural MRI. LME models provide a statistical approach that is flexible and can accommodate unequal numbers of measurement points or sampling intervals. Considering multiple regions of interest simultaneously in a multivariate approach provides the ability to account for correlations between regional measurements. Within each modality, we modelled the cortical and subcortical regions separately. One model fit the 34 FreeSurfer cortical regions simultaneously and one model fit the seven FreeSurfer subcortical regions simultaneously. For MRI, cortical thickness was used for the cortical regions’ model and intracranial volume-corrected volumes were used for the subcortical regions’ model. Models were fit on mean-centred and standard deviation (SD) scaled data. Within each model (e.g. cortical PIB) for each individual region of interest, the model included fixed effects of baseline age, time, and APOE ɛ4 status. To allow for non-linear associations with baseline age, we modelled baseline age as a restricted cubic spline with three knots. Restricted cubic spline functions allow for the flexibility to vary non-linearly without forcing the relationship into a particular polynomial fit and have been used previously for modelling imaging biomarkers with respect to age or time in longitudinal analyses (Vemuri et al., 2010; Jack et al., 2015; Jansen et al., 2015). In our final model we included the two terms for baseline age from the spline, APOE ɛ4 status, time, and all two-way and three-way interactions as fixed effects and also subject-varying slope and intercept random effect terms. Both the subject-level intercepts and the subject-level slopes were allowed to correlate across all brain regions included in a model.

To fit each model, we used the software package Stan (mc-stan.org) (Gelman et al., 2015; Carpenter et al., 2017) implemented in R using rstan version 2.14.1, to perform Markov Chain Monte Carlo (MCMC) analyses. A Bayesian approach was selected as the resultant credible intervals provide a richer, more informative estimate than classical confidence intervals obtained from null-hypothesis significance testing approaches. We used a normal distribution (mean = 0, SD = 5) as a prior for the fixed effects beta-estimates and used a normal distribution (mean = 0, SD = 1) as a prior for the random effects gamma-estimates of the scaled data. We used uniform priors for the variances and the lkj correlation density function (mu = 1.0) as a prior for the correlation matrix (Lewandowski et al., 2009). We implemented MCMC with eight chains of 10 000 iterations each (5000 warm-up, thinning 1 in 10 iterations) (Kruschke, 2014; Sorensen and Vasishth, 2016). Credible intervals were defined as the range between the 0.5% and 99.5% estimates to represent 99% credible intervals. To test for convergence, the Gelman-Rubin convergence statistic (Ř), the ratio of between-chain variance to within-chain variance was used (Gelman and Rubin, 1992). A value close to 1 indicates convergence. Ř for each parameter estimate was within 0.01 of 1.0. The code for the models is included in the Supplementary material.

We report the rate of change of PIB SUVR or structural MRI data as a function of baseline age for APOE ɛ4-positive and APOE ɛ4-negative individuals. The ages at which the slopes between groups diverge were determined as the first age at which the 99% credible interval of the distribution of differences between groups did not overlap zero. To improve reliability and avoid spurious points, a time point was only considered significant if the distribution continued to not overlap zero for 2 years. To understand whether ɛ4 carriers accumulate pathology at a faster rate relative to non-carriers at a similar point in the disease, both the PIB longitudinal rate of change in the precuneus and the volume longitudinal rate of change in the hippocampus were additionally evaluated as a function of baseline disease burden, measured by the mean cortical SUVR (mcSUVR) from the baseline PIB scan. A linear mixed effects model was fit including fixed effects of baseline mcSUVR, time, and APOE ɛ4 status, along with all two-way and three-way interactions and subject-varying slope and intercept random effect terms. To again allow for non-linear associations, mcSUVR was modelled as a restricted cubic spline with three knots. MCMC analyses were completed as previously described to derive PIB slope and volume slope estimates as a function of baseline PIB mcSUVR, with 99% credible intervals. As exploratory analyses, we additionally ran models for precuneus PIB and hippocampal volume stratifying ɛ4 carriers into homozygotes (ɛ44) and heterozygotes (ɛ43 or ɛ34) and including all appropriate additional terms in the linear mixed effects models. These results were considered exploratory rather than as main analyses due to the small number of homozygotes with PIB (n = 11) and MRI (n = 23) data.

Finally, to characterize abnormal levels of amyloid-β PET across the lifespan, the cross-sectional cohort of 575 individuals with at least one PET scan was used. Each participant was characterized as amyloid-β positive or negative based upon previously defined cut-offs. The proportion of all ɛ4 carriers and all non-carriers who were amyloid-β positive was determined as a function of age across the span of 42 to 89. The resultant curves were smoothed using a local regression method, using the loess method implemented in R, with a span = 0.7.

Results

We found a statistically significant difference in the rate of amyloid-β accumulation in APOE ɛ4 carriers in comparison to non-carriers in 33/34 FreeSurfer cortical regions. Figure 1 shows an example of the results in the precuneus. Figure 1A shows a spaghetti plot of the model fits for the individual participant SUVR trajectories. Figure 1B shows the model fits for the SUVR rate of change as a function of baseline age and APOE genotype. Figure 1C shows the difference in the longitudinal rates of change of amyloid-β accumulation between the APOE ɛ4 carriers and non-carriers, along with the 99% credible intervals of this difference. The first point in this difference where credible intervals do not overlap 0 is the first baseline age where rates of accumulation significantly diverge between groups. In our data, APOE ɛ4 carriers demonstrate significantly accelerated accumulation of amyloid-β deposition in the precuneus, in comparison to non-carriers, at a baseline age of 49 years old. The difference in slopes between carriers and non-carriers is significantly greater than zero throughout the majority of the lifespan represented by our cohort. Figure 2 shows the age at which APOE ɛ4 carriers first have significantly greater longitudinal change in PIB binding than non-carriers for each of the cortical regions. These data are presented numerically in Table 3. In all cortical regions where there was a statistically significant difference between ɛ4 carriers and non-carriers, the earliest age of divergence in rates of amyloid-β accumulation between carriers and non-carriers varied between 49 and 68 years. Two out of seven of the FreeSurfer subcortical regions showed significant differences between groups in rates of amyloid-β plaque accumulation across the adult lifespan, the caudate (age 56) and the putamen (age 63). Graphs for all regions are depicted in the Supplementary material.

Figure 1.

Modelling longitudinal change in precuneus PIB (A–C) and hippocampal volume (D–F). A and D represent spaghetti plots of the model fits for the individual participant SUVR and volume trajectories, respectively. B and E show the longitudinal rates of change between the APOE ɛ4 carriers and non-carriers, along with the 99% credible intervals. For reference, individual random effect slope estimates are also plotted. C and F depict the difference in rate of biomarker change between APOE ɛ4 carriers and non-carriers across the course of the sampled lifespan. Shaded regions represent 99% credible intervals..

Figure 2.

Regional differences in emergence of PIB. The colour scale represents the first age where the rate of amyloid-β accrual in the cortical regions is significantly different between APOE ɛ4 carriers and non-carriers.

Table 3.

First age of accelerated PIB binding in ɛ4 carriers versus non-carriers

Region of interest	Age of first detectable APOE effect
Cortical
Inferior parietal cortex	49.28
Precuneus cortex	49.32
Middle temporal gyrus	50.21
Superior frontal gyrus	50.29
Superior temporal gyrus	50.6
Supramarginal gyrus	50.69
Caudal anterior cingulate cortex	50.71
Posterior cingulate cortex	50.93
Rostral anterior cingulate cortex	50.94
Banks of the superior temporal sulcus	50.96
Rostral middle frontal cortex	50.99
Medial orbitofrontal cortex	51.1
Pars triangularis	51.29
Caudal middle frontal cortex	51.44
Fusiform gyrus	51.79
Inferior temporal cortex	51.89
Pars opericularis	52.06
Superior parietal cortex	52.32
Frontal pole	52.44
Paracentral cortex	53
Lateral orbitofrontal cortex	53.02
Transverse temporal gyrus	53.51
Parahippocampal cortex	53.67
Insula cortex	54.28
Lateral occipital cortex	54.5
Postcentral gyrus	54.71
Pars orbitalis	56.69
Precentral gyrus	57.19
Pericalcarine cortex	65.28
Isthmus cingulate	66.74
Temporal pole	67.41
Lingual gyrus	67.62
Entorhinal cortex	68.45
Cuneus cortex	-
Subcortical
Nucleus accumbens	-
Amygdala	-
Caudate	56.76
Hippocampus	-
Pallidum	-
Putamen	64.04
Thalamus	-

In contrast to the effects on amyloid-β deposition, only 3/34 cortical regions showed a statistically significant difference in the rate of cortical atrophy, with carriers having faster rates of cortical thinning than non-carriers in the insula (age 47), parahippocampal cortex (age 51), and entorhinal cortex (age 71). All of these effects were very modest and bordered on significance (see regional graphs available in the Supplementary material). For the subcortical analyses 4/7 of the regions showed a significantly greater rate of subcortical atrophy in APOE ɛ4 carriers relative to non-carriers: the hippocampus, amygdala, putamen, and accumbens at ages of 57, 66, 70, and 71, respectively. Plots of the model fits of hippocampal volume are shown in Fig. 1D–F. The non-carriers’ hippocampal volumetric rate of change shows a linear acceleration throughout the lifespan, while APOE ɛ4 carriers demonstrate even greater accelerated cortical atrophy in midlife followed by a similar rate of acceleration to that of the non-carriers later in the lifespan (Fig. 1E). The slope difference between carriers and non-carriers significantly differs from zero between ∼55 to 80 years, after which there is not a significant difference in the rates (Fig. 1F). Exploratory analyses examining the effect of ɛ4 dosage are presented in Supplementary Fig. 1. These results indicate that ɛ4 homozygotes had earlier and greater rates of amyloid-β accumulation and hippocampal atrophy than APOE ɛ4 heterozygotes and non-carriers.

The temporal evolution of pathology as related to the APOE ɛ4 genotype differs regionally. Figure 3A and B shows the differences in the longitudinal rate of change of amyloid-β deposition and cortical atrophy, respectively, between carriers and non-carriers in two cortical regions prominently studied in Alzheimer’s disease. In comparison to the precuneus, which demonstrates early acceleration followed by a plateau in the rate of change of amyloid-β accumulation, the entorhinal cortex shows no difference in amyloid-β accumulation rates between carriers and non-carriers at younger ages but APOE ɛ4 carriers demonstrate increased rates of amyloid-β deposition in the entorhinal cortex at older ages. In contrast, the differences in cortical atrophy between carriers and non-carriers in the entorhinal cortex and precuneus are not significantly different across the lifespan sampled in this population.

Figure 3.

Regional differences in biomarker trajectories. Top: The difference in the longitudinal rates of change of PIB measurements (A) and cortical thickness (B) between APOE ɛ4 carriers and non-carriers in two cortical regions. Bottom: The difference in the longitudinal rates of change of PIB (C) and volume (D) between APOE ɛ4 carriers and non-carriers in two subcortical regions.

Figure 3C and D shows the differences in the longitudinal rate of change of amyloid-β accumulation and structural atrophy, respectively, between carriers and non-carriers in two subcortical regions. The differences in amyloid-β accumulation between carriers and non-carriers in the hippocampus and caudate did not differ significantly across the lifespan. When examining structural MRI, in the hippocampus ɛ4 carriers have accelerated volumetric loss early in the lifespan, but there is no significant difference in rate of change of volume loss between carriers and non-carriers in the caudate across the sampled lifespan.

The difference in amyloid-β accumulation in the precuneus and atrophy in the hippocampus between APOE ɛ4 carriers and non-carriers was evaluated as a function of disease burden, measured by baseline PIB mean cortical SUVR. Figure 4A shows no difference between hippocampal atrophy as a function of disease burden in ɛ4 carriers, in comparison to non-carriers. Figure 4B demonstrates greater longitudinal amyloid-β accumulation as a function of disease burden in ɛ4 carriers, in comparison to non-carriers. These findings show that APOE ɛ4 carriers experience accelerated amyloid-β accumulation, but not atrophy, relative to non-carriers after matching individuals for baseline levels of pathology.

Figure 4.

Longitudinal change in PIB as a function of disease status. The longitudinal rates of change of PIB in the hippocampus (A) and the precuneus (B) for both APOE ɛ4 carriers and non-carriers are shown as a function of mean cortical SUVR from the baseline PIB assessment. Shaded areas represent 99% credible intervals.

The population level frequencies of abnormal PIB levels stratified by genotype are shown across the ages of participants in the cross-sectional cohort (Fig. 5). At a population level, APOE ɛ4 carriers begin having elevated amyloid-β pathology at an earlier age, and at higher frequencies (steeper slope) after age 60. The proportion of APOE ɛ4 carriers that eventually develop pathological amyloid-β PET levels approaches 1 by age 90, while the proportion of non-carriers that develop pathological amyloid-β levels plateaus at less than 0.5, or half of the population.

Figure 5.

Characterization of abnormal levels of amyloid-β PET across the lifespan. The proportion of APOE ɛ4 carriers and non-carriers who are amyloid-β PET positive is depicted as function of age in (A) non-demented individuals only and (B) the combined sample of non-demented and demented individuals.

Discussion

The purpose of this study was to better characterize the role of the APOE ɛ4 allele on amyloid-β pathology and neurodegeneration across the lifespan of middle-aged to elderly cognitively normal adults. This study illustrates that the presence of the APOE ɛ4 allele results in increased longitudinal rates of change of amyloid-β plaque accumulation diffusely through the cortex. While APOE ɛ4 carriers show near ubiquitous accelerated amyloid-β accumulation throughout the brain relative to non-carriers, the temporal emergence of this pathological accumulation differs across brain regions (between ages 49 and 70). In addition, the present work is the first to show that APOE ɛ4 carriers have not only increased rates of amyloid-β accumulation earlier but also have faster rates of amyloid-β accumulation in comparison to non-carriers after adjusting for baseline levels of pathology. APOE ɛ4 effects on neurodegeneration were modest, with carriers displaying greater rates of structural loss in medial temporal lobe structures as early as 50 years.

The literature has shown that APOE ɛ4 carriers develop Alzheimer’s disease more often and earlier than non-carriers, and the APOE ɛ4 allele has deleterious effects on amyloid-β pathology. The majority of work on amyloid-β pathology has been cross-sectional (for a review see Fouquet et al., 2014), with a limited subset of work examining longitudinal change (Grimmer et al., 2010; Villemagne et al., 2011, 2013; Vlassenko et al., 2011; Jack et al., 2013b; Resnick et al., 2015; Bilgel et al., 2016). In general, longitudinal studies have found that APOE ɛ4 status did not significantly affect the rates of amyloid-β accumulation after adjusting for age (Vlassenko et al., 2011; Jack et al., 2013b; Villemagne et al., 2013; Resnick et al., 2015; Bilgel et al., 2016). Relatively modest longitudinal follow-ups, and study populations that contain older adults and demented individuals, but do not include middle-aged participants, may drive such null results. While prior work indicates there is an undeniable effect of the ɛ4 allele on group level measures of amyloid-β pathology, it is unclear how the ɛ4 allele produces this effect. It has remained to be shown whether these group differences can solely be attributed to a higher proportion of APOE ɛ4 carriers developing the disease, whether they develop it at an earlier age, or whether APOE ɛ4 carriers are developing pathology at a faster rate even relative to individuals with preclinical Alzheimer’s disease without the ɛ4 allele.

The current study found that at the group level ɛ4 carriers had greater longitudinal accumulation of amyloid-β than non-carriers and that this deposition accelerated with increasing baseline age. When considering population frequencies of PIB positivity (Fig. 5) we found that ɛ4 carriers had abnormal PIB scans earlier in time and at an elevated population frequency relative to non-carriers. Finally, in addition to looking at slope differences as a function of age, the present study shows that the accumulation of amyloid-β pathology as a function of baseline amyloid-β burden is also higher in APOE ɛ4 carriers than non-carriers (Fig. 4). This study examining cognitively normal middle-aged and older adults provides evidence that APOE ɛ4 positivity not only shifts the onset of amyloid-β deposition earlier, but also accelerates accrual of amyloid-β deposits once this process has begun. In contrast, previous work suggesting that APOE ɛ4 status primarily affects the age at which people begin to develop amyloid-β pathology, but not the rate of amyloid-β deposition (Jack et al., 2013b; Bilgel et al., 2016) has primarily been conducted using older adults and impaired individuals. Our findings suggest that genetic factors in Alzheimer’s disease may change the trajectory of disease progression, in contrast to the hypothesis that all individuals at a given disease burden will progress similarly. This may have important implications for clinical trial selection and therapeutic targets for APOE ɛ4 carriers.

The current study examined regional differences in the temporal evolution of amyloid-β accumulation in APOE ɛ4 carriers versus non-carriers. The vast majority of prior studies have used summary metrics in cross-sectional and longitudinal studies looking at APOE ɛ4 effects. In the present study, we consider the longitudinal rates of change of amyloid-β in all cortical and subcortical FreeSurfer anatomical regions of interest. The present study shows that the age at which APOE ɛ4 carriers display increased rates of accumulation in comparison to non-carriers is as early as 49 years old in regions that are affected early by amyloid-β pathology (i.e. precuneus). The presence of temporal differences in different regions introduces a novel aspect of genetic influences on late-onset Alzheimer’s disease.

The relationship between APOE ɛ4 and volumetric changes longitudinally has conflicting evidence in the current literature. Structural changes from neurodegeneration is a late biomarker change in the progression of Alzheimer’s disease (Bateman et al., 2012; Benzinger et al., 2013; Jack and Holtzman, 2013; Jack et al., 2013a). The present study shows that cognitively normal APOE ɛ4 carriers demonstrate more rapid volume loss in medial temporal lobe structures, in comparison to non-carriers, first detected around 60 years of age. This is consistent with prior work indicating significant effects of APOE genotype on structural MRI (Geroldi et al., 1999; Mori et al., 2002; Van De Pol et al., 2007; Morra et al., 2009; Risacher et al., 2010; Hostage et al., 2014; Manning et al., 2014) and in contrast to the studies that have shown no APOE ɛ4 effect on volumetric changes (Jack et al., 1998; Moffat et al., 2000; Du et al., 2006; Fan et al., 2010; Taylor et al., 2014).

Despite the large sample and high number of longitudinal visits the observed APOE ɛ4 effect on neurodegenerative biomarkers was quite subtle and the modest size of this effect may explain the heterogeneity in the field. The ɛ4 effect on neurodegenerative biomarkers may also be most prominent when including both middle-aged and older adults. Further, when looking at hippocampal atrophy as a function of baseline disease status, there was no difference between APOE ɛ4 carriers than non-carriers. These findings suggest that, in contrast with the APOE ɛ4 effect on amyloid-β, the neurodegeneration seen in both the APOE ɛ4 carriers and non-carriers was proportional to baseline levels of amyloid-β. This suggests that when considering populations matched for clinical severity or amyloid-β pathology, the ɛ4 genotype may have minimal effects on atrophy.

There are some limitations of our study. Our results reflect the detectable genotype differences given our sample. Larger samples with greater power would likely detect differences even earlier or detect effects in additional regions. Our results also focus on cognitively normal individuals. The effects of APOE genotype likely vary in demented cohorts. The earliest ages at which we can detect APOE ɛ4 carriers and non-carriers diverging are limited by the ages represented in our cohort, which begins at age 45. Including even younger ages could reveal that the influence of the ɛ4 allele begins even earlier in the lifespan. Prior work also indicates that the ɛ4 allele has a dose effect (Corder et al., 1993), with homozygotes having greater risk of developing Alzheimer’s disease. Even in our larger population with longitudinal MRI only 23 individuals were homozygous for the ɛ4 allele. This limits our ability to accurately model the interaction of ɛ4 dosage across the studied age ranges. Results from our exploratory analyses in the precuneus and hippocampus do suggest that ɛ4 homozygotes have greater and earlier amyloid-β accumulation and atrophy. However, our samples size is very modest. Future work integrating multiple cohorts would be able to better estimate how ɛ4 homozygosity or the ɛ2 alleles modify biomarker trajectories starting in middle age. Our cohort consisted of adults with longitudinal imaging biomarker data who were cognitively normal at baseline. The influence of the ɛ4 allele may be different as cognition declines. This study focused only on PIB PET and MRI. Evaluation of longitudinal changes of other imaging biomarkers of neurodegeneration, such as FDG and tau PET would be of interest as these may better predict cognition. Finally, our work represents the influence of the APOE ɛ4 allele in only one cohort of middle-aged and older adults. Replication of our results by other groups in an independent cohort would strengthen the interpretation of our findings.

In summary, the present study provides evidence that APOE ɛ4 is linked to increased rates of change of amyloid-β accumulation detectable in late middle-aged cognitively normal adults. In addition to APOE ɛ4 carriers having earlier amyloid-β accumulation, in comparison to non-carriers, they also have faster rates of accumulation as a function of baseline disease status. Through using a regional approach, the present study provides the first evidence that the temporal evolution of APOE ɛ4-related amyloid-β accumulation varies regionally. Finally, this study shows that cognitively normal APOE ɛ4 carriers show increased rates of volumetric change in medial temporal lobe structures as early as 60 years of age, in comparison to non-carriers.

Glossary

Abbreviations

CDR

Clinical Dementia Rating

PIB

¹¹C-Pittsburgh compound B

SUVR

standardized uptake value ratio