Introduction

The mechanisms by which APOEε4 imposes a genetic risk factor for sporadic Alzheimer’s disease are not fully understood. While early work linked APOEε4 with both amyloid-β and tau pathologies, much of the focus of the role of APOEε4 has been in relation to amyloid-β [1]. The APOEε4 allele is associated with increased production [2, 3] as well as diminished clearance of cerebral amyloid-β [4, 5]. Individuals with the APOEε4 genotype also demonstrate increased amyloid-β PET uptake [6], with amyloid positivity beginning earlier in life in APOEε4 carriers than noncarriers [7].

Together, these findings are interpreted to suggest that the mechanism through which the APOEε4 allele confers risk for Alzheimer’s disease is by leading to increased cerebral amyloid-β burden, considered to be the central pathological event in Alzheimer’s disease [8]. However, recent work has demonstrated that the APOEε4 allele is also related to inflammation and neurodegeneration in mouse models, as well as faster disease progression in humans [9]. Furthermore, previous observational studies have reported that the APOEε4 allele modifies the relationship between amyloid-β and cognitive decline [10,11,12,13], though the precise mechanisms underlying this relationship remain unclear.

Given its close association with cognitive deficits [14,15,16], aggregation of tau pathology presents a potential mechanism for APOEε4 modifying the relationship between amyloid-β and cognitive decline [17]. While recent tau-PET studies have reported inconsistent effects of APOEε4 on tau-PET uptake [14, 15, 18], no studies have assessed whether APOEε4 potentiates the relationship between amyloid-β and tau pathologies. Thus, we aimed to determine if tau pathology depends on the synergistic interaction between APOEε4 and amyloid-β, rather than the sum of their independent effects. We hypothesize that APOEε4 synergistically interacts with amyloid-β to drive tau aggregation.

Materials and methods

Participants

TRIAD

The Translational Biomarkers in Aging and Dementia (TRIAD) cohort aims at modeling biomarker trajectories and interactions as drivers of dementia. TRIAD was launched in 2017 as part of the McGill Centre for Studies in Aging. We assessed cognitively unimpaired (n = 138), mild cognitive impairment (n = 26), and Alzheimer’s disease dementia (n = 47) subjects who underwent amyloid-β PET with [18F]AZD4694, tau-PET with [18F]MK6240, structural MRI and genotyping for APOEε4. All subjects had detailed clinical assessments including Mini-Mental State Examination (MMSE), Clinical Dementia Rating (CDR), and cerebrovascular disease risk with the Hachinski Ischemic scale [19]. Cognitively unimpaired controls had a CDR of 0, mild cognitive impairment subjects had a CDR of 0.5, and Alzheimer’s disease participants had a CDR of 1 or 2, in addition to meeting standard diagnostic criteria [20]. Similar to other large-scale cohort studies of aging and Alzheimer’s disease [13], the TRIAD Cohort is enriched for APOEε4 carriers. This study’s protocol was approved by the Institutional Review Board at McGill University and informed consent was obtained from every subject or their caregiver.

ADNI

Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 as a public–private partnership led by principal investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial MRI, PET, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment and early Alzheimer’s disease. In this study, we assessed cognitively normal (n = 157), amnestic mild cognitive impairment (n = 83), and Alzheimer’s disease dementia (n = 24) individuals from ADNI cohort who underwent amyloid-β PET with [18F]Florbetapir, tau-PET with [18F]Flortaucipir (also known as [18F]T807 and [18F]AV1451), structural MRI and genotyping for APOEε4. We also examined a third independent sample of cognitively normal (n = 104), amnestic mild cognitive impairment (n = 283), and Alzheimer’s disease (n = 100) individuals from ADNI cohort who underwent amyloid-β PET with [18F]Florbetapir, lumbar puncture, structural MRI, and genotyping for APOEε4. Cognitively normal controls had a CDR of 0, MCI subjects had a CDR of 0.5, and Alzheimer’s disease participants had a CDR of 1 or greater in addition to meeting standard diagnostic criteria [20]. Full information regarding the ADNI inclusion and exclusion criteria can be accessed at http://adni.loni.usc.edu/ (accessed April 2019). The ADNI study was approved by the Institutional Review boards of all of the participating institutions. Informed written consent was obtained from all participants at each site.

Genetic and CSF analyses

TRIAD

Determination of APOE genotypes for subjects enrolled in the TRIAD cohort was performed using the polymerase chain reaction amplification technique, followed by restriction enzyme digestion, standard gel resolution, and visualization processes. Full details of this procedure can be found elsewhere [21].

ADNI

Determination of APOE genotypes for ADNI subjects took place at the University of Pennsylvania Alzheimer’s Disease Biomarker Laboratory. CSF measurements of tau phosphorylated at threonine 181 were assessed using the multiplex xMAP Luminex platform (Luminex, Austin, TX, USA) with INNOBIA AlzBio3 (Innogenetics, Ghent, Belgium) immunoassay kit-based reagents. The CSF biomarker data sets used in this study were obtained from the ADNI files ‘UPENNBIOMK5-8.csv’. We considered a subject positive for tau hyperphosphorylation if the CSF p-tau value was above the ADNI published threshold (0.23 pg/mL) [22, 23]. Complete details of CSF methods employed in ADNI can be accessed at http://adni.loni.usc.edu/data-samples/clinical-data/.

PET image acquisition and processing

TRIAD

All subjects had a T1-weighted MRI which was used for coregistration. PET scans were acquired with a Siemens High Resolution Research Tomograph. [18F]MK6240 images were acquired 90–110 min post injection and scans were reconstructed with the OSEM algorithm on a 4D volume with 4 frames (4 × 300 s) [24]. [18F]AZD4694 images were acquired 40–70 min post injection and scans were reconstructed with the OSEM algorithm on a 4D volume with three frames (3 × 600 s) [25]. Immediately following each PET acquisition, a 6-min transmission scan was conducted with a rotating 137Cs point source for attenuation correction. Furthermore, the images underwent correction for dead time, decay, and random and scattered coincidences. T1-weighted images were nonuniformity and field-distortion corrected and processed using an in-house pipeline. Then, PET images were automatically registered to the T1-weighted image space, and the T1-weighted images were linearly and nonlinearly registered to the ADNI template space. Next, a PET nonlinear registration was performed using the linear and nonlinear transformations from the T1-weighted image to the ADNI template space and the PET to T1-weighted image registration. The PET images were spatially smoothed to achieve a final resolution of 8 mm full-width at half maximum. PET image partial volume correction was carried out using the PETPVC toolbox [26]. The region-based voxel-wise correction technique was used to perform partial volume correction using ten tissue priors with a gaussian kernel with the FWHM of 2.4 mm [27]. [18F]MK6240 standardized uptake value ratio (SUVR) maps were generated using the inferior cerebellar gray matter as a reference region and [18F]AZD4694 SUVR maps were generated using the cerebellar gray matter as a reference region. A global [18F]AZD4694 SUVR value was estimated for each participant by averaging the SUVR from the precuneus, prefrontal, orbitofrontal, parietal, temporal, anterior, and posterior cingulate cortices.

ADNI

Full information regarding acquisition of PET data in ADNI is provided at http://adni.loni.usc.edu/data-samples/pet/. Preprocessed PET images downloaded from ADNI underwent spatial normalization to the ADNI standardized space using the transformations of PET native to MRI native space and MRI native to the ADNI space. Partial volume correction was carried out using the PETPVC toolbox [26] described above in an effort to diminish off-target binding to the choroid plexus. [18F]Flortaucipir SUVR maps were generated using the inferior cerebellar gray matter as a reference region [28] and [18F]Florbetapir SUVR maps were generated using the cerebellar gray matter as a reference region. A global [18F]Florbetapir SUVR value was estimated for each participant by averaging the SUVR from the precuneus, prefrontal, orbitofrontal, parietal, temporal, anterior, and posterior cingulate cortices.

Statistical analyses

To measure tau pathology in vivo, we used CSF and PET measurements. Previous studies have demonstrated associations between CSF p-tau and tau-PET uptake in Alzheimer’s disease-related brain regions [29,30,31]. The primary outcome measure of the study was tau pathology as measured by voxel-wise [18F]MK6240 SUVR (TRIAD cohort), [18F]Flortaucipir SUVR (ADNI tau-PET cohort), and CSF phosphorylated tau (ADNI CSF cohort). Three independent samples were investigated cross-sectionally: (1) the McGill cohort assessed with [18F]MK6240 and [18F]AZD4694 (2) an ADNI cohort assessed with [18F]Flortaucipir and [18F]Florbetapir and (3) an ADNI cohort assessed with [18F]Florbetapir and lumber puncture for CSF phosphorylated tau. In each cohort, we tested the hypothesis that the synergistic interaction between amyloid-β and APOEε4 is related to tau pathology, i.e., the interaction between amyloid-β and APOEε4 is greater than the sum of the additive effects of amyloid-β and APOEε4 [32, 33]. In each cohort, we measured the effect of one APOEε4 allele or two APOEε4 alleles, meaning that the comparison is to individuals who do not carry an APOEε4 allele. The results displayed in this manuscript are multiple comparisons corrected t-values (Random Field Theory with a cluster threshold of p < 0.005). From RFT-corrected significant clusters, we subsequently extracted the beta estimates and standard errors.

Baseline demographic and clinical data were assessed using t tests and χ2 tests. Neuroimaging analyses were carried out using the VoxelStats toolbox (https://github.com/sulantha2006/VoxelStats), a MATLAB-based analytical framework that allows for the execution of multimodal voxel-wise neuroimaging analyses [34]. Other statistical analyses were performed using the R Statistical Software Package version 3.0.2 (http://www.r-project.org/). In models with interaction terms, predictor variables were centered on the mean for improved interpretability of coefficients and to improve numerical stability for estimation associated with multicollinearity [35]. Given the large number of covariates in the statistical models, model diagnostics were carried out using the car package in R to determine the presence of multicollinearity. We computed the Variance Inflation Factor (VIF), a measurement of how much variance in regression coefficients are inflated due to multicollinearity in the statistical models [36]. All neuroimaging analyses described below were repeated using partial volume corrected data. To provide further confirmation that the interaction term is statistically significant, we computed the absolute values of the beta estimates from RFT-corrected statistically significant clusters, e.g., absolute value of (beta(X1) + beta(X2) + beta(X1 × X2)) > absolute value of (beta(X1) + beta(X2)) [32, 33].

In the TRIAD cohort, the voxel-based interaction model outlined below was built to test whether main and interactive effects between APOEε4 carriership and [18F]AZD4694 SUVR at every voxel are associated with [18F]MK6240 uptake. To ensure that the results are not driven by an effect of clinical status, we adjusted the model for clinical diagnosis. The model was also adjusted for age. Because APOEε4 is related to amyloid-PET uptake, amyloid-β was included as a covariate in every analysis. Statistical parametric maps were corrected for multiple comparisons using a random field theory [37] threshold with a cluster threshold of P < 0.005. The analysis was repeated using partial volume corrected data. The analysis was also repeated excluding a subject with the ε2/ε4 genotype.

$${\mathrm{Tau}}\;{\mathrm{PET}}\;{\mathrm{SUVR}} = \, \beta _0 + \beta _1\,\left( {{\mathrm{Amyloid}}\;{\mathrm{PET}}\;{\mathrm{SUVR}}} \right)\\ \!+ \,\beta _2\,\left( {APOE_{\varepsilon 4}} \right) +\, \beta _3\,\left( {{\mathrm{Amyloid}}\;{\mathrm{PET}}\;{\mathrm{SUVR}}\;x\;APOE_{\varepsilon 4}} \right) \\ \! +\, \beta _4\,\left( {\mathrm{Clinical}}\;{\mathrm{Status}} \right) + \beta _5\,({\mathrm{Age}})\; + \; \in.$$

Next, we aimed to investigate a possible gene-dose relationship in the ADNI database, a larger cohort containing more homozygous APOEε4 carriers. In these gene-dose analyses, APOE status was treated as a categorical variable with three levels (Noncarriers < Heterozygotes < Homozygotes). No individuals in the ADNI tau-PET cohort were also included in the ADNI CSF analyses. This model was also adjusted for cerebral amyloid-β, age, and clinical status. Statistical parametric maps were corrected for multiple comparisons using a random field theory [37] threshold with a cluster threshold of P < 0.005. The analysis was repeated using partial volume corrected data.

$${\mathrm{Tau}}\;{\mathrm{PET}}\;{\mathrm{SUVR}} = \;\beta _0 + \beta _1\,\left( {{\mathrm{Amyloid}}\;{\mathrm{PET}}\;{\mathrm{SUVR}}} \right) \\ \!+ \beta _2\,\left( {APOE_{\varepsilon 3\varepsilon 4}} \right)\\ + \beta _3\,\left( {APOE_{\varepsilon 4\varepsilon 4}} \right) \\ \!+ \beta _4\,\left( {{\mathrm{Amyloid}}\;{\mathrm{PET}}\;{\mathrm{SUVR}}\;x\;APOE_{\varepsilon 3\varepsilon 4}} \right)\\ \!+ \beta _5\,\left( {{\mathrm{Amyloid}}\;{\mathrm{PET}}\;{\mathrm{SUVR}}\;x\;APOE_{\varepsilon 4\varepsilon 4}} \right) \\ \!+ \beta _6\,\left( {Clinical\;Status} \right) + \beta _7\,(Age) \,+ \in.$$

In order to gain a better understanding of the similarities between cohorts, we repeated the analyses in ADNI using the same APOEε4 carrier/noncarrier framework as conducted in the TRIAD cohort.

We followed up the tau-PET analyses by testing the hypothesis in an independent sample of 487 individuals who underwent amyloid-β PET with [18F]Florbetapir and lumbar puncture, with CSF phosphorylated tau as an outcome measure. No individuals in the ADNI CSF cohort were also included in the ADNI tau-PET analyses. APOE status was treated in a dose-dependent manner and the model was adjusted for age, clinical diagnosis, and the main effect of amyloid-β.

$${\mathrm{CSF}}\;{\mathrm{phosphorylated}}\;{\mathrm{Tau}} = \,\, \beta _0 + \beta _1\,\left( {{\mathrm{Amyloid}}\;{\mathrm{PET}}\;{\mathrm{SUVR}}} \right) \\ + \beta _2\, \left( {APOE_{\varepsilon 3\varepsilon 4}} \right)\\ + \beta _3\,\left( {APOE_{\varepsilon 4\varepsilon 4}} \right) \\ + \beta _4\,\left( {{\mathrm{Amyloid}}\;{\mathrm{PET}}\;{\mathrm{SUVR}}\;x\;APOE_{\varepsilon 3\varepsilon 4}} \right)\\ \\ + \beta _5\,\left( {{\mathrm{Amyloid}}\;{\mathrm{PET}}\;{\mathrm{SUVR}}\;x\;APOE_{\varepsilon 4\varepsilon 4}} \right) \\ + \beta _6\,\left( {\mathrm{Clinical}}\;{\mathrm{Status}} \right) + \beta _7\,(Age) \,+ \in.$$

Results

Demographic and clinical information for the three samples examined in this study is summarized in Table 1. VIFs for all variables in all cohorts are presented in Supplementary table 1. VIFs for all variables were below 4, indicating that problematic levels of multicollinearity are not present in our analyses [36]. Table 2 presents the estimates of main and interactive effects of amyloid-PET and APOEε4 on tau pathology in the three independent samples. Standardized estimates are presented in Supplementary Table 2. The brain regions displayed in Table 2 correspond to regions that were statistically significant after correction for multiple comparisons.

Table 1 Demographic and key characteristics of the samples.
Full size table
Table 2 Main and interactive effects of amyloid-PET and APOEε4 on tau-PET uptake and CSF p-tau.
Full size table

We tested the hypothesis that amyloid-β and APOEε4 are related to tau pathology, where the interaction between amyloid-β and APOEε4 is greater than the sum of the independent effects. Voxel-wise analyses revealed a synergistic interaction between APOEε4 and local [18F]AZD4694 SUVR on [18F]MK6240 uptake across the cerebral cortex (Fig. 1) independent of age and clinical diagnosis. The interaction between local [18F]AZD4694 and APOEε4 was related to increased [18F]MK6240 in the posterior cingulate, precuneus, occipital, and inferior parietal cortices. The results remained similar when employing partial volume corrected data. Scatterplots representing the associations between [18F]AZD4694 SUVR and [18F]MK6240 SUVR stratified by APOEε4 genotype are displayed in Supplementary Fig. 1.

Fig. 1: Cerebral tau aggregation depends on the synergistic interaction between amyloid-β and APOEε4.
figure 1

The synergistic interaction between [18F]AZD4694 and APOEε4 carriership was related to increased [18F]MK6240 uptake in the posterior cingulate, precuneus, occipital, and inferior parietal cortices. T-statistical parametric maps were corrected for multiple comparisons using a Random Field Theory cluster threshold of P < 0.005, overlaid on the ADNI reference template. Age, clinical diagnosis, and amyloid-β SUVR were employed as covariates the model.

Full size image

When investigating a gene-dose relationship, different effects were found for APOEε4 heterozygotes and homozygotes. The interaction between local [18F]Florbetapir and one APOEε4 allele was associated with higher levels of [18F]Flortaucipir uptake in posterior cingulate, precuneus, posterior parietal, lateral temporal, temporooccipital, and orbitofrontal cortices. The interaction between local [18F]Florbetapir and two APOEε4 alleles was related to increased [18F]Flortaucipir uptake in the posterior cingulate, anterior cingulate, precuneus, posterior parietal, medial prefrontal, and orbitofrontal cortices (Fig. 2). Tau-PET uptake in the temporooccipital cortex was observed only for the interaction between [18F]Florbetapir SUVR and one APOEε4 allele. Effects of homozygosity were observed in the tau-PET uptake in the precuneus, anterior cingulate, and medial prefrontal cortices were observed only for the interaction between [18F]Florbetapir SUVR and two APOEε4 alleles. These effects were independent of age and clinical diagnosis. Again, results remained similar when employing partial volume corrected data. Scatterplots representing the associations between [18F]Florbetapir SUVR and [18F]Flortaucipir SUVR stratified by APOEε4 genotype are displayed in Supplementary Fig. 2. When investigating the carrier/noncarrier framework in ADNI (as was conducted in the TRIAD cohort), we observed that the interaction between local [18F]Florbetapir and APOEε4 carriership was associated with higher levels of [18F]Flortaucipir uptake in posterior cingulate, precuneus, inferior parietal, lateral temporal temporooccipital, and orbitofrontal cortices (Supplementary Fig. 3).

Fig. 2: APOEε4 exerts a gene-dose effect on tau aggregation when interacting with amyloid-β.
figure 2

a The interaction between [18F]Florbetapir and a single APOEε4 gene was related to increased [18F]Flortaucipir uptake in the posterior cingulate posterior parietal, lateral temporal temporooccipital, and orbitofrontal cortices. b Homozygous ε4 carriers demonstrated a more widespread relationship between [18F]Florbetapir and [18F]Flortaucipir uptake, with [18F]Flortaucipir uptake in the posterior cingulate, precuneus, posterior parietal, medial prefrontal, and orbitofrontal cortices. Tau-PET uptake in the temporooccipital cortex was observed only for the interaction between [18F]Florbetapir SUVR and one APOEε4 allele. Effects of homozygosity were observed in the tau-PET uptake in the precuneus, anterior cingulate, and medial prefrontal cortices were observed only for the interaction between [18F]Florbetapir SUVR and two APOEε4 alleles. T-statistical parametric maps were corrected for multiple comparisons using a Random Field Theory cluster threshold of P < 0.005, overlaid on the ADNI reference template. Age, clinical diagnosis, and amyloid-β SUVR were employed as covariates in each model. Results remained comparable when using partial volume corrected PET data.

Full size image

In a third sample of subjects with CSF measurements of phosphorylated tau, the synergistic effect between APOEε4 and neocortical [18F]Florbetapir SUVR, rather than the sum of their independent effects, was related to increased CSF phosphorylated tau. While the heterozygotes (β4 = 20.31, se = 6.59, p < 0.0001) had a milder slope than the homozygotes (β5 = 33.01, se = 14.24, p = 0.01), this difference in slopes was not statistically significant (p = 0.07) (Fig. 3). These results were independent of age, clinical diagnosis, and the main effect of amyloid-β.

Fig. 3: Interaction between amyloid-β and APOEε4 is related to increased CSF phosphorylated tau pathology.
figure 3

The synergistic effect between APOEε4 and neocortical [18F]Florbetapir SUVR was related to increased CSF p-tau. While the heterozygotes (β4 = 20.31, se = 6.59, p < 0.0001, represented in green) had a milder slope than the homozygotes (β5 = 33.01, se = 14.24, p = 0.01, represented in blue), this difference in slopes was not statistically significant (p = 0.07). Age, clinical diagnosis, and amyloid-β SUVR were employed as covariates.

Full size image

Discussion

This study presents in vivo evidence that APOEε4 potentiates the relationship between amyloid-β and tau pathologies. This potentiation, revealed by the synergistic interaction between APOEε4 and amyloid-β, was associated with higher levels of tau pathology in the precuneus, posterior cingulate, anterior cingulate, inferior parietal, and basolateral temporal cortices, regions known to exhibit tau accumulation and neurodegeneration across the Alzheimer’s disease spectrum [14, 16, 38, 39]. Homozygous ε4 carriers had a more widespread pattern of cerebral tau pathology compared with heterozygous ε4 carriers. The interaction between amyloid-β and one ε4 allele was related to tau aggregation in the inferior parietal, lateral temporal, orbitofrontal, and posterior cingulate cortices, while the interaction between amyloid-β and two ε4 alleles was also related to tau aggregation in additional regions including the precuneus, medial prefrontal, and anterior cingulate cortices. In the independent sample of subjects who underwent lumbar puncture, the synergistic interaction between APOEε4 and amyloid-β on CSF phosphorylated tau was also observed. To the best of our knowledge, this is the first in vivo study demonstrating a synergistic interaction between APOEε4 and amyloid-β on tau pathology.

The effects of APOEε4 on tau pathology with the presence of amyloid-β may help explain faster disease progression [9, 40] as well as the stronger relationship between amyloid-β and cognitive decline in APOEε4 carriers [10,11,12]. Tau-PET uptake in temporal and parietal regions reported in our study are associated with impaired cognitive function [15], and longitudinal studies demonstrated that elevated neocortical tau-PET predicts cognitive decline [41]. Post-mortem studies have also reported that Alzheimer’s disease patients who are APOEε4 carriers have increased tau pathology compared with noncarriers [42]. Previous studies have also reported a lack of association between primary age-related tauopathy and the APOEε4 genotype [43], suggesting that the effect of APOEε4 on tau pathology may be related to its interaction with amyloid-β (Fig. 4). Similarly, post-mortem studies reported that APOEε4 was associated with increased paired helical filament (PHF) tau in individuals with concomitant amyloid-β pathology, while no association between APOEε4 and tau was observed in individuals without amyloid-β pathology [44]. Furthermore, recent reports have suggested that amyloid-β synergistically interacts with tau to determine disease progression [45, 46], supporting a framework in which Alzheimer’s disease is characterized by multiple pathological interactions rather than the sequential aggregation of different pathologies.

Fig. 4: APOEε4 exerts a double hit in Alzheimer’s disease.
figure 4

Schematographic representation of the pathological process presented in this manuscript. APOEε4 exerts a double hit on Alzheimer’s disease risk through its relationship to amyloid-β aggregation, and by potentiating the relationship between amyloid-β and tau pathologies. It is important to note that this figure is intended to illustrate the process described in the present manuscript and is not intended to be a complete description of the roles of APOEε4 or amyloid-β in Alzheimer’s disease.

Full size image

The brain regions in which tau pathology was related to an amyloid-β × APOEε4 interaction were concentrated to brain regions known to accumulate tau in Alzheimer’s disease [47]. While both TRIAD and ADNI cohorts demonstrated relationships between amyloid-β × APOEε4 interactions and tau pathology in the posterior cingulate, precuneus and inferior parietal cortices, small differences between cohorts existed as well. In particular, the medial occipital uptake observed in the TRIAD cohort could be attributable to the AD individuals who also meet criteria for Posterior Cortical Atrophy (PCA), a condition characterized by occipital and posterior parietal tau-PET uptake [15]. Differences in the properties of tau imaging agents could cause the minor differences between cohorts in our manuscript. [18F]MK6240 has a 5-fold higher Bmax/Kd (concentration of available binding sites/equilibrium dissociation constant) ratio than [18F]Flortaucipir in AD brains post-mortem [48]. Correspondingly, it is conceivable that [18F]MK6240 captured tau pathology that was below the detection threshold of [18F]Flortaucipir. However, head-to-head studies of tau-PET radioligands are needed to clarify this issue as these minor regional discrepancies could also be due to population differences: the TRIAD cohort includes more early onset AD subjects, who have greater cortical tau pathology compared with late onset AD subjects [49].

The present results provide support for an emerging framework in which APOEε4 exerts pathophysiological effects beyond its involvement in increased cerebral amyloid-β burden [9]. In fact, apoE-immunoreactivity has been demonstrated to aggregate in neurons bearing neurofibrillary tangles [50] and increased expression of apoE in neurons is related to increased tau phosphorylation [51,52,53]. ApoE3, but not apoE4, has been demonstrated to bind to the microtubule-binding repeat region of tau implicated in the self-assembly of tau into PHFs [54], suggesting that there may be isoform-dependent relationships between apoE and tau pathology [55]. Furthermore, truncated apoE4 fragments stemming from stress- or injury-related proteolytic cleavage of apoE are also related to increased tau hyperphosphorylation and neuronal cytoskeletal disruption [56, 57]. APOEε4 has also been associated with cerebral hypometabolism independently of cerebral amyloid-β burden [58]. Taken together, these studies suggest the need for a reassessment of the role of APOEε4 throughout the stages of Alzheimer’s disease pathogenesis.

Our study has important methodological limitations. The first is that this study is phenomenological and was not designed to discover a biological mechanism underlying the relationship between APOEε4, amyloid-β, and tau. Secondly, despite correcting our analyses for clinical status, longitudinal studies are needed to disentangle whether APOEε4 carriers had more advanced disease pathophysiology. A methodological strength of the study is the replication of results obtained with a first-generation tau-PET tracer with a second-generation tau-PET tracer [59]. Future work is needed to determine whether the effects of APOEε4 and amyloid-β on tau result in increased phosphorylation, conformational changes or increased cortical spreading. Because of the different responses of APOEε4 carriers to disease-modifying pharmaceutical agents [60], a more complete understanding of the involvement of APOEε4 in Alzheimer’s disease will help guide the development and design of future disease-modifying therapeutic trials.