Network dysfunction in cognitively normal APOE ε4 carriers is related to subclinical tau

Abstract

INTRODUCTION:

Apolipoprotein E (APOE) ε4 allele status is associated with amyloid and tau-related pathological changes related to Alzheimer’s disease (AD). However, it is unknown whether brain network changes are related to Aβ and/or tau-related pathology in cognitively normal APOE ε4 carriers with sub-threshold amyloid (Aβ) accumulation.

METHODS:

Resting state functional connectivity (rs-fc) measures of network integrity were evaluated in cognitively normal individuals (n=121, mean age 76.6 ± 7.8 years, 15% APOE ε4 carriers, 65% female) with minimal Aβ per cerebrospinal fluid (CSF) or amyloid positron emission tomography (PET).

RESULTS:

APOE ε4 carriers had increased lateralized connections relative to callosal connections within the default-mode, memory, and salience networks (p = 0.02), with significant weighting on linear regression towards CSF total tau (p = 0.03) and CSF phosphorylated tau at codon 181 (p = 0.03), but not CSF Aβ₄₂.

DISCUSSION:

Cognitively normal APOE ε4 carriers with sub-threshold amyloid accumulation may have network reorganization associated with tau.

1. Introduction

Apolipoprotein E (APOE) ε4 allele status remains the most common genetic risk factor that alters an individual’s risk for Alzheimer disease (AD). The presence of a single APOE ε4 allele increases the risk for AD by ~3.7 fold, while two alleles increases the risk ~12 fold (1–3). The APOE ε4 allele has been linked to early and accelerated Aβ deposition (4–7), as well as amyloid-independent mechanisms of tau tangle formation, neuroinflammation, and ultimately neuronal death (8,9).

Yet it remains unclear which one or a combination of these effects is the cardinal event leading to early microcellular disruptions that may manifest as macro-scale reorganization of the functional connections within and/or between critical brain regions. Regions in the brain display coherent activity within different functional and cognitive domains, forming fundamental networks which are preserved across individuals (10–12). In AD, network reorganization may occur at varying spatial scales, affecting both local and long-range connections (13–15). Resting state functional connectivity (rs-fc) is a powerful method to explore these macro-scale network changes within a cohort of cognitively normal individuals at increased risk for developing AD, such as APOE ε4 carriers.

The evaluation of APOE-mediated cortical reorganization with sub-threshold amyloid burden is predicated on the ability to appropriately select at-risk individuals with normal cognition. Biomarker profiles, as determined by cerebrospinal fluid (CSF) and molecular imaging using positron emission tomography (PET), can provide an integrated framework for selection and stratifying individuals into AD stages (16,17). Biomarker profiles can quantify an individual’s amyloid and tau burden, thereby permitting direct comparison with rs-fc metrics of macro-scale reorganization in critical brain networks.

Rs-fc was used to compare cortical network integrity between cognitively normal APOE ε4 allele carriers and APOE ε4 non-carriers who were below established CSF and molecular imaging cutoffs (biomarker negative). Changes in the nature of underlying rs-fc were evaluated with respect to local and long-range (callosal) projections. Finally, observed changes in rs-fc were analyzed in relation to biomarker profiles of Aβ and tau.

2. Materials and Methods:

Participants

Two-hundred forty-eight participants enrolled at the Knight Alzheimer’s Disease Research Center (ADRC) at Washington University in St. Louis (WUSTL) were evaluated (Table 1). Only baseline data was utilized, due to limited availability of longitudinal data. All participants underwent imaging and CSF evaluation less than one year apart (average interval 91 days). Methods for recruitment have previously been described (18). All participants were cognitively normal (CDR® 0). This group was further subdivided by biomarker status using previously established cutoffs for PET PiB (19) and CSF Aβ₄₂ (20) into biomarker negative (i.e. subthreshold for both CSF amyloid and PET-PiB), CSF-amyloid+ (suprathreshold for CSF amyloid, but subthreshold for PET-PiB), and definitive-amyloid+ (i.e. suprathreshold for both CSF amyloid and PET-PiB). Of note, five individuals (mean age 83.4 ± 5.4, 0% APOE ε4 carriers, 80% female) who were amyloid PET positive but CSF amyloid negative were not included. This study was approved by the WUSTL Institutional Review Board and each participant provided signed informed consent.

Table 1:

Demographics of all cognitively normal participants, grouped by biomarker profile

	Biomarker Negativea	CSF amyloid+b	Definitive amyloid+c
No.	121	75	52
Age (years ± SD)	76.6 ± 7.8	69.9 ± 6.9	79.1 ± 7.4
Education (years ± SD)	15.7 ± 2.6	16.2 ± 2.3	16.1 ± 3.2
Sex (% Female)	65	61	44
APOE ε4 carries (%)	15	27	62
Race (%)
Asian	1 (0.8)	1 (1.3)	0 (0.0)
African-American	5 (4.1)	15(20)	2 (3.9)
Non-Hispanic White	115 (95.1)	59 (78.7)	50 (96.1)

^aSubthreshold per both CSF(Aβ₄₂ > 1098 and CSF p-tau/Aβ₄₂ < 0.0198) (15,16) and PET criteria (PET PiB SUVR < 1.42) (20,21)

^bSuprathreshold per CSF criteria (Aβ₄₂ > 1098 or CSF p-tau/Aβ₄₂ < 0.0198) (15,16) but subthreshold for PET criteria (PET PiB SUVR <1.42) (20,21)

^cSuprathreshold per both CSF (Aβ₄₂ > 1098 or CSF p-tau/Aβ₄₂ < 0.0198) (15,16) and PET criteria (PET PiB SUVR > 1.42) (20,21)

CSF = cerebrospinal fluid; APOE = Apolipoprotein E; SD = standard deviation; PET = positron emission tomography

Genetic analyses

DNA samples were collected at enrollment and genotyped using either an Illumina 610 or Omniexpress chip as previously published (21). Participants carrying one or more APOE ε4 allele were defined as “carriers”, while those lacking an ε4 were classified as “non-carriers”.

CSF acquisition and processing

CSF was collected using methodology as previously described (22). Briefly, participants underwent a lumbar puncture at 8 AM following overnight fasting. Approximately 25 ml of CSF was collected in a 50 ml polypropylene tube using a Sprotte 22-gauge spinal needle and gravity drip. Gentle mixing followed by low speed centrifugation pelleted any residual debris. CSF samples were then aliquoted into individual polypropylene tubes for storage at −80 °C until the time of assay. CSF Aβ42, total tau, and p-tau were measured with corresponding Elecsys immunoassays on the Roche cobas e601 analyzer (20).

PET image acquisition and processing

Amyloid PET images with [¹¹C] Pittsburgh Compound B (PiB) were acquired using methodology previously described (23–25). Dynamic scans were obtained after injecting participants with 12 – 15 mCI of PiB, with the window for quantification time 30 – 60 minutes post-injection. Raw PET data was then processed using a PET Unified Pipeline (github.com/ysu001/PUP). FreeSurfer 5.3 was employed for region of interest (ROI) segmentation. For each region, a tissue mask was generated based on segmentation, and partial volume correction performed (23). Regional target-to-reference intensity ratio, also known as the standard uptake ratio (SUVR), was evaluated for each region using the cerebral cortex as the reference region. The partial volume corrected SUVR derived from cortical regions was used as a summary value.

Magnetic resonance imaging (MRI) acquisition and processing

Magnetic resonance imaging (MRI) images were collected on a 3T Siemens Trio scanner (Erlangen, Germany). T1-weighted scans were segmented using FreeSurfer 5.3 (Martinos Center for Biomedical Imaging, Charlestown, Massachusetts, USA). Right and left hippocampal volume was combined and scaled by mean intracranial volume. Rs-fc data was acquired as previously been described (10–12). Post-processed rs-fc data was aggregated into 298 seed-based functional regions of interest (ROI). These 298 ROIS were sorted into 13 previously defined network and clustered into the following groups: default mode (DM)/Memory (MEM)/Salience (SAL) group, the attentional/alertness group (cingulo-opercular, frontopartieal, dorsal and ventral attention networks), sensory group (auditory, visual, somatosensory networks), and finally the subcortical group (subcortical and cerebellar networks). A cross-correlation matrix comparing pairwise Pearson’s correlations between each seed was generated for each participant, and then underwent Fisher’s r-to-z transformation to better normalize the correlations.

For a given participant, intra-network synchrony was defined as the mean correlation between all nodes within a given network (e.g. DMN-DMN; See Supplemental 1: Resting state Analysis Schema). In contrast, inter-network synchrony was defined as the mean correlation between any nodes representing different networks of a given group (e.g. DMN-SAL, DMN-MEM, MEM-SAL, but not DMM-visual). A given node may fall in the right hemisphere, left hemisphere, or rarely, midline. Masks were generated for the full 298×298 cross-correlation matrix reflecting pairs of nodes in the same hemisphere (within hemisphere or “lateralized” mask) or pairs of nodes in different hemisphere (between hemisphere or “callosal” mask). Midline nodes were not included in either mask.

Statistical Analyses

Group differences of rs-fc between APOE ε4 carriers and non-carriers were compared using unpaired t-tests following linear regression of known confounds of age, sex, and education (26). A given individual’s lateralized and callosal within-network synchrony was combined into a single scalar value termed the rs-fc interaction term (See Supplemental 1: Resting state Analysis Schema). The interaction term reflected the relative bias of lateralized to callosal synchrony of a given individual. Similarly, group differences in individual CSF or imaging biomarkers were also compared using unpaired t-tests following regression of known confounds of age, sex, and education. Linear modeling of the rs-fc interaction term for key cortical networks (DMN, SAL, MEM) was performed using the Matlab function LinearModel.fit. Models included continuous variables for CSF Aβ₄₂, PET PiB, CSF p-tau, CSF t-tau, normalized hippocampal volume, age, and years of education, as well as categorical variables for sex and APOE carrier status. Stepwise regression of the same rs-fc interaction term used the Matlab function stepwiselm. Models used the same parameters and included interactions, with insignificant terms being dropped in a stepwise, data-driven manner. All analyses and modeling was performed using Matlab 2019b software, MathWorks). Finally, partial correlations were used to directly correlate between terms while accounting for confounds of age, sex, and education.

3. Results

APOE ε4 carriers with subthreshold amyloid burden have disruptions in key networks

We first examined the effect of APOE ε4 carrier status on 13 key resting-state networks in 121 cognitively normal (clinical dementia rating® [CDR®] 0) adults. This cohort, termed the biomarker negative group, had biomarker levels below established cutoffs for preclinical AD (i.e. both CSF Aβ₄₂ > 1098, p-tau/Aβ₄₂ < 0.0198 (20,27) and PET Pittsburgh compound B (PiB) SUVR < 1.42 (23,24); Fig. 1A). These criteria were applied to rigorously exclude individuals with significant amyloid deposition (28), with CSF Aβ₄₂ included to account for non-AD dementias (29) (Table 1; with detailed demographic comparisons in Supplemental Table 1).

Fig 1. Resting-state group difference matrices comparing cognitively normal APOE ε4 non-carriers to carriers.

a-c) the mean correlation matrix of APOE ε4 carriers was subtracted from the mean correlation matrix of ε4 non-carriers to generate a difference matrix. Blue connections were more correlated in ε4 carriers and red connections were more anti-correlated in carriers. a) Difference matrix for the biomarker negative group. b) Difference matrix for the CSF amyloid+ group, who met CSF but not PET-PiB criteria for AD. c) Difference matrix for the definitive-amyloid+ group, who met both CSF and PET-PiB criteria for AD. Subtle qualitative changes were appreciated throughout the matrices. APOE = Apolipoprotein E; CSF = cerebrospinal fluid; PET = positron emission tomography; PiB = Pittsburgh compound B; AD = Alzheimer’s disease

Key networks, namely DMN, MEM, and SAL networks, that were previously shown to be disrupted in AD changed prior to cortical atrophy (10,30–36), were evaluated. These core networks were grouped into a cluster to quantify changes in rs-fc both within and between networks as a function of APOE ε4 carrier status after accounting for confounds of age, sex, and education (26) (See Methods)

Biomarker negative APOE ε4 carriers had insignificant increases in rs-fc within these three networks (i.e. intra-network; p = 0.118) compared to their biomarker negative non-APOE ε4 carrier peers (Fig. 2A). Biomarker negative APOE ε4 carriers had a significant loss of anti-correlations (p= 0.044) between the same networks (i.e. inter-network, between DMN-SAL, DMN-MEM, and SAL-MEM). This is concordant with a previous study of rs-fc between core elements of the DMN and MEM networks, namely the hippocampus and posterior cingulate cortex, in asymptomatic adults (37).

Fig 2. Mean synchrony of the default mode, memory, and salience networks.

Mean intra-network and inter-network synchrony between the DMN, MEM, and SAL networks were calculated for cognitively normal APOE ε4 carriers and non-carriers. Note inter-network synchrony was defined as only between DMN, MEM, or SAL networks, not any other networks. Biomarker negative APOE ε4 carriers had insignificant increase in mean intra-network synchrony, but a significant change in inter-network synchrony. This relationship was not observed in either the CSF amyloid+ group (who met CSF but not PET-PiB criteria for AD) or the definitive-amyloid+ group (who met both CSF and PET-PiB criteria for AD). DM = default-mode; MEM = memory; SAL = salience; APOE = Apolipoprotein E; CSF = cerebrospinal fluid; PET = positron emission tomography; PiB = Pittsburgh compound B; AD = Alzheimer’s disease

Changes in CSF amyloid levels were traditionally considered one of the earliest biomarkers to change in AD pathogenesis, and may be an important contributing factor (38). The same key networks were then examined in a separate group of 75 cognitively normal participants (mean age 69.9 ± 6.93 years, 27% APOE ε4 carriers, 61% female) with isolated biomarker positive CSF Amyloid (termed the CSF-amyloid+ group, Fig. 1B) and in a third group of 52 cognitively normal participants (mean age 79.1 ± 7.40 years, 62% APOE ε4 carriers, 44% female) with both CSF Amyloid and PET PiB positivity (termed the definitive-amyloid+ group, Fig. 1C). CSF biomarker positivity was defined as either CSF Aβ₄₂ < 1098 or p-tau/Aβ₄₂ > 0.0198 (20,27). All three groups represented a unique cohort of individuals with no overlap amongst groups (Table 1). Compared to the biomarker negative group, the CSF-amyloid+ and definitive-amyloid+ groups represented later preclinical AD stages, after biomarker conversion. No group differences were observed in either the CSF-amyloid+ or definitive-amyloid+ groups (Fig. 2B, C). This favors a transient phenomenon, because we would anticipate similar findings in both biomarker positive groups if the network changes represent a sustained, rather than transient, APOE ε4-mediated difference.

APOE-mediated changes were also specific to networks known to be disrupted in AD, with no similar group differences observed in either the attentional/alertness (cingulo-opercular, frontoparietal, dorsal and ventral attention; Supplemental Fig. 2A), sensory (auditory, visual, somatosensory; Supplemental Fig. 2B), or subcortical (subcortical and cerebellar; Supplemental Fig. 2C) networks for the biomarker negative, CSF-amyloid+, or definitive-amyloid+ groups.

Our findings suggest isolated cortical reorganization amongst the DMN, MEM, and SAL networks of cognitively normal APOE ε4 carriers who have biomarker levels below established cutoffs. We next sought to determine the type of the connections underlying these disruptions.

Lateralized connections are increased at the expense of callosal connections in APOE carriers

Previous work in cognitively normal APOE ε4 carriers demonstrated both marked laterality and disruption of synchronization between mirrored patches of cortex in the medial temporal lobe (14). We next focused on clarifying if lateralized or callosal connections underpinned the observed network disruptions in the DMN, SAL, and MEM networks in cognitively normal biomarker negative APOE ε4 carriers.

In biomarker negative APOE ε4 carriers (Fig. 2A), early increases in rs-fc within the DMN, SAL, and MEM networks (i.e. DMN-DMN, MEM-MEM, and SAL-SAL connections) were driven by a significant increase in lateralized connections (Fig. 3A, p = 0.05), relative to callosal connections (p = 0.34). An interaction term was created that combines both lateralized/callosal terms into a single scalar value and reflects an individual’s relative bias toward local at the expense of long-range connections (Fig. 3B, p = 0.024). This hyper-connectivity phase was only observed in the cognitively normal biomarker negative group. No differences in the rs-fc interaction term was observed in the CSF-amyloid+ and definitive-amyloid+ groups (Fig. 3A,B).

Fig 3. Lateralized and callosal connections underpin the observed changes of the mean synchrony of the default mode, memory, and salience networks.

a) Intra-network synchrony within the DMN, MEM, and SAL networks was masked to only examine lateralized or callosal connections. Increased lateralized synchrony in APOE ε4 carriers was observed only within biomarker negative, APOE ε4 carriers. b) The rs-fc interaction term, calculated by subtracting an individual’s lateralized and callosal intra-network synchrony, was also increased only in biomarker negative, APOE ε4 carriers. c) Significant anti-correlations were observed mainly in callosal, but not lateralized connections of biomarker negative APOE ε4 carriers only. DM = default-mode; MEM = memory; SAL = salience; APOE = Apolipoprotein E

Earlier, we demonstrated significant loss of anti-correlations amongst the DMN, SAL, and MEM networks (i.e. DMN-SAL, DMN-MEM, and SAL-MEM connections) when comparing cognitively normal biomarker negative APOE ε4 carriers to non-carriers (Fig. 2A). Observed changes were driven mainly by loss of callosal connections in cognitively normal biomarker negative APOE ε4 carriers (p = 0.017; Fig. 3C).

Together, these findings suggest an increase in local, lateralized connections over more longrange, callosal connections in biomarker negative APOE ε4 carriers, which is absent in biomarker positive individuals.

Preclinical APOE carriers have no significant differences in A/T/N biomarkers compared to non-carriers

We next examined if APOE ε4 carriers had differences in biomarkers representing amyloid, tau, and neurodegeneration (i.e. the A/T/N criteria) (17) compared to non-carriers in our three groups (Fig. 4). Cognitively normal biomarker negative APOE ε4 carriers had a trend toward decreased CSF Aβ₄₂ compared to cognitively normal biomarker negative APOE ε4 non-carriers ([1552 vs 1815], p = 0.08; Supplemental Table 2). No group difference was observed between the CSF-amyloid+ APOE ε4 carriers and non-carriers ([719.2 vs 809.3], p = 0.33), or between the definitive-amyloid+ APOE ε4 carriers and non-carriers ([715.3 vs 817.9], p = 0.16). No difference in PET-PiB was observed between APOE ε4 carriers and APOE non-carriers for any group. Similarly, no differences between APOE ε4 carriers and APOE non-carriers was observed for CSF total tau, CSF p-tau, or normalized hippocampal volume amongst the groups. In summary, no significant group differences were observed between APOE ε4 carriers and APOE ε4 non-carriers, regardless of biomarker status within this cohort.

Fig 4. Group biomarker profiles according to the amyloid (A), tau (T) and neurodegeneration (N) criteria.

Mean CSF and PET-PiB biomarker profiles of our biomarker-, CSF amyloid+, and definitive amyloid+ groups. Biomarkers were clustered by amyloid (“A” i.e. CSF Aβ₄₂ and PET-PiB), tau (“T” i.e. CSF p-tau), or neurodegeneration (“N” i.e. CSF t- tau and normalized hippocampal volume) (17). Established thresholds for AD were delineated in red (18–21). No statistically significant group differences were observed in any of the measures between APOE ε4 carriers and non-carriers. CSF = cerebrospinal fluid; PET = positron emission tomography; PiB = Pittsburgh compound B; p-tau = tau phosphorylated at 181; AD = Alzheimer’s disease; APOE = Apolipoprotein E

Tau is linked to changes in intra-network connectivity in the default-mode, salience, and memory networks in APOE carriers

We next used multivariate and regression modeling to clarify the relationship between network reorganization, an individual’s biomarker levels, and APOE carrier status within the cognitively normal biomarker negative group. The rs-fc interaction term (Fig. 3B) served as a surrogate for network reorganization in the DMN, SAL, and MEM networks, with known confounds such as age, sex, and education included (26).

The resulting linear model (Fstat 2.68, p = 0. 00741) revealed significant weighting on CSF p-tau (p = 0.03), CSF t-tau (p = 0.03), age (p = 0.045), sex (male-bias, p = 0.02), and a trend for APOE ε4 allele carrier status (p = 0.059), but not for CSF Aβ₄₂, PET-PiB, or normalized hippocampal volume (Fig. 5A). In other words, older female APOE ε4 carriers with the highest CSF p-tau relative to CSF t-tau levels tended to have the greatest lateralized synchrony relative to callosal synchrony in key AD networks (i.e. the rs-fc interaction term).

Fig 5. Modeling of the rs-fc interaction term as a function of CSF and PET-PiB biomarkers.

a) A simple linear model of the rs-fc interaction term for the biomarker negative group as function of biomarkers, APOE ε4 carrier status, and known confounds of age, sex, education (30). Significant weighting was observed for p-tau, t-tau, age, sex (female), and trend toward significance for carrier status. b) A stepwise regression model which included interaction terms of the same rs-fc interaction term. Notably, weighting was observed on t-tau, carrier status, the interaction of p-tau and age, along with age and sex (female). rs-fc = resting state functional connectivity; CSF = cerebrospinal fluid; PET = positron emission tomography PiB = Pittsburgh compound B; APOE = Apolipoprotein E; p-tau = tau phosphorylated at 181

Similar results were found using a model that did not include insignificant elements but added interaction terms using a stepwise regression (Fstat 5, p = 0.000139; Fig. 5B), with significant weighting on CSF t-tau (p = 0.039), APOE ε4 carrier status (p = 0.018), and sex (male-bias, p = 0.003), and a trend level for CSF p-tau (p = 0.06). Essentially, female APOE ε4 allele carriers with the highest CSF p-tau relative to CSF t-tau levels again demonstrated the greatest lateralized compared to callosal synchrony within core AD networks. In both models, no clear association was observed between the rs-fc interaction term and either CSF Aβ₄₂ or PET amyloid. The observation of a protective effect of the male sex is in line with recently published work (40). A similar relationship was observed when examining only the eighteen APOE ε4 carriers (Fstat 11.5, p = 0.0003). Despite a significantly truncated group size (error in degrees of freedom of 13), CSF t-tau (p = 0.00002) and age (p = 0.0002) remained significant. This observation was only within the cognitively normal biomarker negative group, as the same interaction terms were not significant in the model fit for either the CSF amyloid+ or definitive amyloid+ groups.

Finally, comparable results were obtained with a combined CSF p-tau /CSF t-tau variable, the inclusion of CSF tau cutoffs (i.e. to examine whether the relationship was driven by a subgroup of individuals with elevated CSF t-tau or CSF p-tau), through the use of stricter amyloid cutoffs, and aggregating the two amyloid+ groups. (See Supplemental 2: Confounds and Limitations Analyses).

4. Discussion

The link between APOE ε4, Aβ, and tau remains enigmatic and represents a complex interplay that varies over the natural history of AD. This study explores potentially early changes in the DMN, MEM, and SAL networks of cognitively normal APOE ε4 carriers with minimal biomarker evidence of amyloid accumulation. We demonstrated a relative strengthening of local compared to long-range connections underlying network disruptions within core networks. No changes were observed in cognitively normal individuals with biomarker evidence of amyloid deposition. Finally, our studies reveal that key brain network changes are more strongly associated with tau, and not amyloid deposition or neuronal degeneration. Together, these findings not only support a novel approach to biomarkers that includes APOE ε4 allele status, but also suggest a role of tau in pathogenic processes in preclinical AD with only subthreshold accumulation of amyloid.

The APOE ε4 allele is known to avidly bind to and accelerate early Aβ deposition by prompting early nucleation, seeding, and Aβ fibril formation (7,41,42). This combined with decreased clearance from the interstitial space across the blood-brain barrier and reduced glial degradation efficiency could result in earlier and faster Aβ regional accumulation (7,43). The rate of Aβ accumulation is more equivocal once an individual is biomarker positive for the different APOE genotypes. This may explain the relatively lower frequency of APOE ε4 carriers in the biomarker- group. Traditionally, increased aggregate burden is mainly noted in presence of concomitant amyloid pathology, suggesting tau (micro/macro) structural pathology manifests after an initial seeding by Aβ.

However, different APOE isoforms are known to have amyloid-independent interactions with tau, including unique binding affinities (44). Recent mouse models expressing human tau with different APOE allele knock-ins also reveal an association between the APOE ε4 allele and increased hippocampal and cortical atrophy, early elevations of soluble tau, and greater distribution of p-tau staining (8). P-tau staining patterns further correlated with levels of microglial and astrocyte activation, with APOE ε4 allele expressing mice demonstrating an exaggerated response relative to ε3 and ε2 alleles. The absence of APOE was protective. Complementary in vitro culture work revealed potentiated neuronal death when human-tau expressing mouse neurons were co-cultured with glial cells from APOE ε4 knock-in mice. Increased apoptosis was also observed in a separate model of cultured human astrocytes grown with CSF collected from AD patients with elevated CSF tau (45). More recently, microglial depletion was found to abrogate the APOE ε4 -mediated atrophy in the same human tau expressing mice (9). This suggests an additional tau-mediated, amyloid-independent neuroinflammatory process may be important. The pathophysiology of potentially associated neuroinflammation is complex, however, and likely varies through the stages of AD. Motta and colleagues demonstrated that APOE ε4 allele carriers without evidence of tau accumulation have significant elevations in key inflammatory cytokines. Intriguingly, this was protective, not detrimental for loss of cognition in their amyloid positive cohort (46). Finally, recent work also suggests that mesiotemporal tau accumulation is independent of Aβ (47).

Traditional Aβ-potentiating-tau paradigms may underappreciate these recently characterized amyloid-independent mechanisms. Aβ-based models also fail to explain occult cortical functional reorganization as documented here and in other studies (10,30–33,35–37), with complementary PET studies (48–51). Functional reorganization may form the basis of intrinsic compensation to maintain cognitive function in response to early seeding by one or as an interplay of these two core proteinpathies. In this alternative framing, early cognitive decline reflects overwhelming functional compensation in response to local stressors, rather than the direct accumulation of low-levels of a toxic proteinopathy. In this context, APOE ε4 may stratify individuals to be at greater risk for loss of compensation to maintain cognitive function.

Our work supports a key role of tau in cortical reorganization in older individuals with subthreshold amyloid burden. Even prior to significant tau accumulation, changes in t-tau and p-tau levels in neurons reflected by changes in CSF t-tau and p-tau may demonstrate altered neuronal signaling. This may potentially be mediated by exaggerated neuroinflammation in APOE ε4 carriers compared to their ε3/2 peers. This inflammatory state may result in an overall transient strengthening of local connections at the expense of long-range connections to maintain cortical synchronization. Neuronal stress in response to this pro-inflammatory state further amplifies p-tau relative to t-tau accumulation in a malignant feed-back loop. Rather than tau driving connectivity changes, this possible alternative model suggests reorganization of local and long-range connections may itself drive tau accumulation, leading to downstream inflammation and damage in an amyloid-independent manner. Supplemental modeling with a combined p-tau/tau terms suggests a disproportionate elevation in p-tau relative to t-tau is related to the observed connectivity changes. However, the lack of longitudinal data and high collinearity of CSF p-tau and t-tau limit further interpretation in this cohort. Future studies integrating plasma p-tau, however, hold promise in potentially further parsing these effects (52).

There are several additional limitations to this study. First, the explicit timing of the relative elevations in local versus long-range connectivity remains unclear given both the cross-sectional nature of this study and the predominately older population enrolled at the Knight ADRC. The older biomarker- cohort may have intrinsic resistance to amyloid which may be partially mediating the observations. The Knight ADRC database is unique given the greater representation of asymptomatic, preclinical individuals compared to other databases. Additional limitations include examining network-level effects in an a priori manner. Anatomical parcellations within a given network and a blinded, data-guided approach are future considerations. Further evaluation of the longitudinal evolution of network changes as an individual progresses through the stages of preclinical AD is also warranted. Integration of additional CSF biomarkers of neuroinflammation would aid in further disentangling the pathophysiologic underpinnings, as would longitudinal characterization as an individual progresses through AD. Finally, the addition of PET tau status and distribution, and future knowledge of time to biomarker conversion will aid in clarifying the relationship between key biomarkers.

These results support network disruption as a potentially early change, prior to CSF or PET biomarker positivity. This thereby suggests a non-invasive categorization of subtle AD pathobiology occurring with subthreshold changes in amyloid. It further supports the integration of APOE ε4 status into the current diagnostic framework for AD. Taken together, our results would be essential for the potential categorization of eligible participants for clinical trials that look to nullify tau-mediated changes.

Research in Context.

1: Background:

APOE ε4 is associated with Alzheimer’s disease (AD)-related pathological changes in amyloid-β (Aβ) deposition and tau-mediated neuroinflammation and neurodegeneration. It is unknown whether changes in cognitively normal APOE ε4 carriers without significant amyloid accumulation are related to Aβ- or tau-related pathology.

2. Interpretation:

Resting state functional connectivity (rs-fc) disruptions in APOE ε4 carriers occurred within the default-mode, memory, and salience networks and reflected increases in lateralized connections relative to callosal connections (p = 0.02). A linear regression revealed that rs-fc disruptions were significantly weighted for CSF tau and tau phosphorylated at 181 (ptau). The implications are APOE ε4 carriers may benefit from therapies that impact tau-related changes, key for the selection and timing of therapy in primary prevention trials.

3. Future Directions:

Future steps include examining the anatomical parcellations underlying the network reorganization and the longitudinal evolution of network change throughout AD.

Abbreviation:

APOE

Apolipoprotein E

AD

Alzheimer’s disease

Aβ

amyloid-β

rs-fc

Resting state functional connectivity

CSF

cerebrospinal fluid

DMN

default-mode

MEM

memory

SAL

salience

p-tau

tau phosphorylated at 181

PET

positron emission tomography

CDR®

clinical dementia rating

PiB

Pittsburgh compound B

tau ratio

CSF p-tau to CSF total tau ratio

ADRC

Knight Alzheimer’s Disease Research Center

WUSTL

Washington University in St Louis

ROI

region of interest

SUVR

standard uptake ratio

MRI

magnetic resonance imaging