Abstract
The apolipoprotein (APOE) ε4 allele is a strong genetic risk factor for Alzheimer's disease (AD). Intrinsic fluctuations of brain activity measured by fMRI during rest may be sensitive to AD-related neuropathology. In particular, functional connectivity of the default-mode network (DMN) has gained recent attention as a possible biomarker of disease processes and associated memory decline in AD. Here, we tested the hypothesis of APOE-related alterations in DMN functional connectivity in 95 healthy individuals between 50 and 80 years of age, including 33 carriers of the ε4 allele. Based on previous studies, we hypothesized increased hippocampal DMN synchronization in APOE ε4 carriers. This was supported using independent component analysis in combination with a dual-regression approach for analysis of resting state data. Whole-brain analysis suggested effects also in other areas, including the posterior cingulate cortex, parietal cortex, and parahippocampal regions. DMN synchronization showed a negative correlation with performance on a test of memory functioning, suggesting a neurocognitive significance of the brain activity patterns during rest. Our findings indicate that increased genetic vulnerability for AD is reflected in increased hippocampal DMN synchronization during rest several years before clinical manifestation. We propose that the results reflect ε4-related failure in hippocampal decoupling, which might elevate the total hippocampal metabolic burden and increase the risk of cognitive decline and AD. The results provide an important confirmation of specific genotype effects on intrinsic fluctuations and support the use of functional connectivity indices as imaging-derived endophenotypes in the emerging field of imaging genetics.
Introduction
Resting state functional connectivity (RSFC) studies have revealed specific functional neuronal networks (Biswal et al., 1995), including low-level sensorimotor and higher-order attention and cognitive control networks (Zhang and Raichle, 2010). Resting state networks (RSNs) are identifiable in the absence of contextual demands, but show high correspondence with task-evoked networks (Smith et al., 2009). The coherence and spatial distribution of the RSNs persist across wakefulness and sleep and also under anesthesia (Zhang and Raichle, 2010). Demonstrations of intrinsic functional connectivity in the monkey (Vincent et al., 2007; Margulies et al., 2009) and infant brain (Fransson et al., 2007, 2011) support that the patterns of RSFC are evolutionarily conserved traits of brain functional architecture.
A heavily investigated RSN is the default-mode network (DMN) (Raichle et al., 2001). Altered DMN RSFC has been observed in clinical conditions including mild cognitive impairment and Alzheimer's disease (AD) (Greicius et al., 2004; Sorg et al., 2007). The close neuroanatomical relations between the DMN and the loci of AD neuropathology are compelling. Based on the notion that early AD pathology forms preferentially throughout areas that make up the DMN (Braak and Braak, 1991; Buckner et al., 2005), it has been suggested that DMN metabolism may be related to disease processes and associated memory decline (Buckner et al., 2008). In line with this hypothesis, several studies have reported disease- and age-related variability in the intrinsic fluctuations of the DMN (Greicius et al., 2004; Damoiseaux et al., 2008).
Carriers of the APOE ε4 allele have a twofold to threefold increased risk of developing late-onset AD if they are heterozygous and ∼12-fold if homozygous for the ε4 allele (Roses, 1996; Bertram et al., 2007). Evidence linking APOE ε4 to RSFC in healthy individuals is limited, and previous studies either did not include a pure resting condition (Persson et al., 2008; Fleisher et al., 2009a; Pihlajamäki and Sperling, 2009) or included relatively few participants (Filippini et al., 2009; Fleisher et al., 2009a), limiting the statistical power. Filippini et al. (2009) demonstrated altered RSFC in retrosplenial, medial–prefrontal, and medial temporal lobe (MTL) regions including the hippocampi in young (20–35 years old) subjects carrying the ε4 allele. Interestingly, ε4 carriers also showed increased hippocampal activation during memory encoding, but neither result was explained by differences in memory performance, brain morphology, or resting cerebral blood flow (Filippini et al., 2009). Effects of APOE ε4 on RSFC have not been demonstrated in middle-aged and elderly individuals using independent component analysis and dual regression.
Therefore, the aim of the present study was to test the hypothesis of specific ε4-related alterations in DMN synchronization in a sample of 95 healthy individuals between 50 and 80 years of age, including 34 carriers of the ε4 allele. Based on one study using a similar analytical approach (Filippini et al., 2009), we hypothesized that MTL regions would show increased DMN synchronization in carriers compared with noncarriers beyond what can be explained by hippocampal volumes and cortical gray matter (GM) density. We further hypothesized that MTL–DMN synchronization during rest would be inversely associated with performance on a standardized test of memory function.
Materials and Methods
Subjects.
Healthy individuals were invited through an advertisement to take part in the first wave of a longitudinal study of cognitive aging. Subjects with a history of substance abuse, present neurologic or psychiatric disorder, or other significant medical conditions were excluded from the study. In this first wave, all participants were examined according to an extensive neuropsychological test protocol, including assessment of memory function, and delivered blood for genotyping (see below). The present study included participants who were invited to a follow-up study after 3 or 4 years, of which 112 participants took part in an MRI examination including 3D MRI and resting state fMRI. All subjects' scans were evaluated by an experienced neuroradiologist. Brain tumors, cysts, recent infarctions, or gross regional or global signal abnormalities were exclusion criteria. No participants were excluded based on the neuroradiological evaluation. Twelve subjects were excluded because of lack of genotyping data. In line with previous studies (Persson et al., 2008; Filippini et al., 2009), two individuals with genotype ε(2,4) were excluded as the ε2 variant is reported to protect against cognitive decline (Wilson et al., 2002) and AD (Benjamin et al., 1994). Further, two individuals with MMSE < 27 were excluded (Folstein et al., 1975). One final participant was excluded because of low estimated general intellectual abilities (IQ = 64). The final sample thus comprised 95 healthy participants aged 50–80 years [mean age = 63.8 years (SD = 7.2), 61 women], including 33 carriers of the APOE ε4 allele. Group characteristics are shown in Table 1. There were no significant differences between the genotype groups with respect to sex (Fischer's exact two-sided test, p = 0.66) or age (independent two-sample t test, p = 0.347).
Group characteristics
All subjects provided informed consent according to the Declaration of Helsinki. The project was approved by the Regional Committee for Research Ethics of Western Norway, and a biobank for storage of personalized data was approved by the Department of Health.
Measures of memory function.
All subjects were tested with the Norwegian translation of the California Verbal Learning Test, second version (CVLT-II) (Delis et al., 1987). A list of 16 words (List A) was presented five times. Immediately after the fifth trial, the participants were read a new list (List B) and asked to recall it. Then, the participants were asked to recall the words from List A, immediately after the recall of List B and ∼20 min later (the long-delayed condition). Subsequently, a recognition trial was presented in which the participants were asked to identify the 16 items from List A from a larger list that contained various distractor items. In the present study, we included the following three CVLT measures: learning, defined as the number of hits across the five learning trials; recall, defined as the number of words recalled in the free long-delayed condition; and recognition discriminability, reflecting the difference in SD units between the examinee's hit rate and false-positive rate (d′) (Delis et al., 1987).
Genotyping.
The whole-blood samples were collected at the Department of Biological and Medical Psychology at the University of Bergen by a bioengineer right after the neuropsychological examination; samples were immediately frozen and sent to Department of Medical Biochemistry, Oslo University Hospital, Rikshospitalet, Norway, for ApoE analysis. Genotyping was performed by real-time PCR with allele-specific fluorescence energy transfer probes and melting curve analysis on the LightCycler system (Roche Diagnostics). DNA was extracted from 300 μl of whole blood using MagNA Pure LC DNA Isolation Kit–Large Volume on the MagNA Pure LC (Roche), eluted and diluted to 1 ml, of which 5 μl was applied in each assay. Typing of the APOE ε2, ε3, and ε4 genotypes was performed using the LightCycler APOE Mutation Detection Kit (Roche). The assay was performed as specified by the supplier, except for scaling down the total assay volume from 20 to 10 μl. The laboratory participates in an external quality assurance program (Equalis) that includes APOE genotyping. Allele frequency combinations are shown in Table 2. The genotype frequencies are in Hardy–Weinberg equilibrium, and the observed carrier frequency of 34.7% homozygotes or heterozygotes for ε4 is in accordance with the known high frequency of this allele in the northern European population (Gerdes, 2003).
APOE genotype distribution
MRI acquisition.
Imaging was performed on a 1.5 T GE Signa Echospeed Scanner, using a standard eight-channel head coil. For volumetry, we obtained two T1-weighted 3D inversion recovery-prepared fast spoiled gradient-recalled acquisitions in a steady state in succession (to improve signal-to-noise ratio and segmentation accuracy), with the following pulse sequence parameters: repetition time (TR)/echo time (TE)/inversion time (TI)/flip angle (FA) = 9.11 ms/1.77 ms/450 ms/7°; voxel size, 0.94 × 0.94 × 1.40 mm; 124 (sagittal) slices; scan time ∼6 min. Resting state BOLD fMRI data were collected for each subject with a T2*-weighted single-shot gradient echo EPI sequence with the following parameters: TR/TE/FA = 2000 ms/50 ms/90°; voxel size, 3.75 × 3.75 × 5.0 mm; 256 volumes (25 axial slices); scan time ∼8 min. Participants were instructed to lie still in the scanner with their eyes closed, to think of nothing in particular, and not to fall asleep. Cushions and headphones were used to reduce subject motion and scanner noise.
MRI processing and analysis.
The T1-weighted 3D MR images were processed using FreeSurfer (http://surfer.nmr.mgh.harvard.edu) enabling automated volumetric segmentation of various neuroanatomical structures including the hippocampi. The segmentation procedure has been described previously in detail (Dale et al., 1999; Fischl and Dale, 2000; Fischl et al., 2001, 2002, 2004). Briefly, the processing scheme includes motion correction and averaging of the two acquisitions, intensity normalization (Sled et al., 1998), removal of nonbrain tissue using a hybrid watershed/surface deformation procedure (Ségonne et al., 2004), automated Talairach transformation, segmentation of the subcortical white matter and deep gray matter volumetric structures (including hippocampus, amygdala, caudate, putamen, and ventricles) (Fischl et al., 2002, 2004), tessellation of the gray/white matter boundary, automated topology correction (Fischl et al., 2001; Ségonne et al., 2007), and surface deformation following intensity gradients to optimally place the gray/white and gray/CSF borders at the location where the greatest shift in intensity defines the transition to the other tissue class (Dale and Sereno, 1993; Dale et al., 1999; Fischl and Dale, 2000).
Individual GM density maps were computed from the T1-weighted 3D MR images using the script feat_gm_prepare, which is distributed with FSL, and FMRIB's Automated Segmentation Tool (FAST) (Zhang et al., 2001). Briefly, individual maps were registered into standard space, smoothed to suit the functional acquisition (σ = 2.63), demeaned, and added as voxelwise explanatory variables in the statistical models (Filippini et al., 2009).
Resting-state fMRI analysis of the 256 volumes in time were performed using Multivariate Exploratory Linear Optimized Decomposition into Independent Components (MELODIC) (Beckmann et al., 2005) implemented in FSL (Smith et al., 2004; Woolrich et al., 2009) (http://www.fmrib.ox.ac.uk/fsl). Individual processing included discarding the first four volumes to let the scanner reach equilibrium due to progressive saturation, motion correction, spatial smoothing using a Gaussian kernel of FWHM of 6 mm, and high-pass temporal filtering equivalent to 150 s (0.007 Hz). fMRI volumes were registered to the subject's skull-stripped T1-weighted scan processed in FreeSurfer using FMRIB's Linear Image Registration Tool (FLIRT) (Jenkinson and Smith, 2001; Jenkinson et al., 2002). The T1-weighted volume was registered and warped to Montreal Neurological Institute 152 standard space (MNI-152) using FMRIB's Nonlinear Image Registration Tool (FNIRT) (Andersson et al., 2007a, 2007b), and the resulting nonlinear transform was applied to the fMRI data. Next, the processed functional data were temporally concatenated across subjects to create a single 4D dataset.
The between-subjects analysis was performed using dual regression (Filippini et al., 2009), allowing for voxelwise comparisons of resting functional connectivity. This specific method has been proven more consistent and reliable than template-matching approaches in its ability to estimate individual-level RSNs from group-level independent component analysis (gICA) spatial maps (Zuo et al., 2010). The procedure comprises three steps: First, gICA is applied to the concatenated resting-state data. RSFC is a relatively new area of research, and the optimal number of components reflecting the true biological dimensionality in the brain is currently unknown. Thus, the choice of dimensionality in resting fMRI studies is arbitrary and will to a certain degree influence the spatial distribution of the DMN. Here, underfitting and overfitting of the components were minimized by using the Laplace approximation to the Bayesian evidence for a probabilistic principal component (Minka, 2000; Beckmann and Smith, 2004).
Second, the dual-regression algorithm (Filippini et al., 2009) is applied to identify subject-specific time courses and spatial maps. The procedure employs a set of gICA spatial maps in a linear model fit against the individual fMRI dataset. This results in matrices describing the temporal dynamics of the corresponding RSN for each of the 95 subjects. These time course matrices are normalized by their variance and used in a linear model fit against the individual fMRI dataset. This temporal regression results in subject-specific spatial maps. These maps reflect degree of synchronization, which is not simply a coherence measure, as they reflect both amplitude and coherence across space (Roosendaal et al., 2010). Here, 19 gICA maps reflecting various RSNs were included in the dual regression.
Third, the different synchronization maps are collected across subjects into 4D files (one file per original ICA map, with the fourth dimension being subject identification) and submitted to voxel-based statistical testing (see below). To explore potential bias inflicted by the unequal group sizes, we correlated the component in question with the analogous IC estimated using equal-sized groups (14 men/20 women in each, no group differences with respect to age and sex). The two components showed very high correlation (Pearson's r = 0.90). Thus, we performed the dual regressions on the gICA maps estimated from the full sample (95 subjects). Effects of APOE status were tested within the DMN, and we performed similar analysis on a control network (primary visual network), where we did not expect to see any effects of genotype, to test for specificity.
Statistical analysis.
Effects of APOE genotype on functional synchronization were tested voxelwise by means of nonparametric permutation testing using 5000 permutations (Nichols and Holmes, 2002) as implemented in randomise, part of FSL (Smith et al., 2004). Threshold-free cluster enhancement was used to avoid the arbitrariness in defining smoothing level and initial cluster-forming thresholds (Smith and Nichols, 2009). Effects were regarded significant at p < 0.05, corrected for multiple comparisons across space. We statistically accounted for effects of age, sex, and hippocampal volumes by including these variables as subjectwise covariates in the statistical models. Here, we used standardized residuals of mean volume of left and right hippocampus after regressing out estimated intracranial volume (Buckner et al., 2004). To account for effects of other structural differences between the two groups, we used GM density maps as voxelwise covariates. The three independent components spanning the DMN and the component spanning visual cortex were dealt with separately for the third part of dual regression.
Because of our a priori hypothesis of specific MTL effects of APOE status and to decrease the number of tests, we restricted our initial analyses to MTL regions (hippocampus + amygdala) as defined by the digitalized probabilistic Harvard–Oxford subcortical atlas provided with FSL. The masks were thresholded at 5% probability level, combined, and used as spatial masks in the statistical analysis. Next, to explore APOE-related variability outside the restricted masks, we performed full-brain explorative follow-up analysis. We computed Cohen's d (Cohen, 1992) in each of the significant clusters to characterize the regional variability in effect sizes.
Statistical analyses of the three verbal memory measures were performed using the package PASW Statistics 18.0. To explore the neurocognitive significance of the RSFC, DMN synchronization within the clusters showing significant effect of group (p < 0.05, corrected) was correlated (Pearson's r) with memory performance while partialing out age and sex. For sociodemographic variables, neuropsychological test scores, and brain structure volumes, we used independent-samples t tests. Pearson's χ2 tests were used for categorical variables (sex and APOE status).
Results
Verbal memory function
Mean CVLT-II scores on the learning, recall, and recognition discriminability measures within each genotype group are presented in Table 1. Independent-samples t tests revealed no statistically significant differences between the two APOE groups on learning and recall. However, we found decreased recognition discriminability scores in carriers compared with noncarriers (t = 2.27, p < 0.03), indicating superior performance in noncarriers. We also found a significant negative effect of age on discriminability scores across the APOE groups (r = −0.29, p < 0.01). Follow-up analysis showed similar effects of age within noncarriers (r = −0.38, p < 0.005) and carriers (r = −0.24, p > 0.05) of the ε4 allele. Fisher's z test revealed no significant difference in age effects on performance between the APOE groups (z = −0.71, p > 0.05).
Hippocampal volumes
Table 1 summarizes mean volume of the left and right hippocampus within and across groups and genders. Mean volume of the left and right hippocampus for noncarriers was 3300.4 (SD = 400.8) mm3 and 3275.4 (SD = 398.3) mm3, respectively. Mean volume for the left and right hippocampus for carriers was 3370.4 (SD = 229.2) mm3 and 3401.5 (SD = 331.7) mm3, respectively. Independent-samples t tests revealed no significant effect of APOE ε4 on hippocampus volumes (p values > 0.05).
Independent component analysis of fMRI data
Spatial gICA (MELODIC) produced 41 independent components estimated using the Laplace approximation to the Bayesian evidence for a probabilistic principal component (Beckmann and Smith, 2004). By visual inspection, 22 of the components were identified as artifactual, caused by variation in subjects' head sizes, head movement, or other physiological fluctuations. By visual inspection, we identified the DMN as the spatial map comprising PFC, ACC, and posterior cingulate cortex (PCC), lateral parietal cortex (LPC), inferior and medial temporal gyri, and thalamic nuclei extending to MTL regions (Raichle et al., 2001; Greicius et al., 2003; Fox et al., 2005; Damoiseaux et al., 2006; Boly et al., 2008). Figure 1A shows the three independent components identified as DMN constituents, and Figure 1B shows the primary visual component.
A, In red-yellow to the left is the DMN comprising three components from the gICA. In red-yellow to the right is the posterior DMN component in which we found increased functional synchronicity in APOE ε4 carriers. This component spans ACC, PCC, precuneal, and thalamic regions. The two DMN components that showed no significant effect of APOE are shown in blue and pink. B, The component spanning the primary visual cortex used as control. C, Significant (p < 0.05, corrected) effects of APOE on the posterior DMN from the analysis restricted to hippocampal and amygdala areas. D, Significant (p < 0.05, corrected) effects of APOE on the posterior DMN from the full-brain analysis. Numbers correspond to the z value of each slice in 1 mm MNI-152 space. The effects extend into thalamic regions, posterior parts of hippocampus, lateral parts of the frontal lobe, the PCC, parietal cortex, and parahippocampal regions. All images are shown in radiological orientation, where the x value increases in left direction, y value increases in anterior direction, and z value increases in superior direction. All images were transformed from 4 to 1 mm space for visualization purposes.
DMN connectivity
Figure 1C shows the spatial distribution of the p statistics from the main MTL analysis in the posterior DMN comprising the precuneus, ACC, PCC, and the thalamus. Significantly increased synchronization in carriers compared with noncarriers was found in large parts of the right MTL (p < 0.05, corrected). Clusterwise mean values for carriers and noncarriers and effect sizes (Cohen's d) are shown in Table 3. Briefly, we found moderate regional variability in effect sizes. The largest effect size was found in the right hippocampus/amygdala (d = 1.23) and the smallest in the left amygdala (d = 0.81). No significant effects of APOE ε4 status were shown in the control RSN (primary visual network).
Increased synchronization with the posterior DMN
Figure 1D shows the results from the explorative full-brain analysis. Increased synchronization in carriers was found in the same DMN component in several areas, mostly in the right hemisphere, including amygdala, hippocampus, PCC, and precuneus. See Table 3 for details regarding the spatial distribution and clusterwise effect sizes. The largest effect size was found in the insular and temporal pole (d = 1.23) and the smallest in occipital fusiform and lingual gyrus (d = 0.61). None of the other DMN components showed effects of the APOE ε4 genotype. Including GM density maps as voxelwise covariates in the statistical models only moderately altered the results. Full-brain analysis using age, sex, hippocampal volumes, and GM maps as covariates reduced the total number of significant voxels from 622 to 399, and the number of clusters increased from 9 to 10. The spatial distribution was very similar to what is displayed in Figure 1D, indicating that the effects of APOE status on RSFC in the posterior DMN cannot be explained by group differences in GM density.
Correlation between DMN synchronization and memory performance
We found a significant negative correlation (r = −0.32, p < 0.002) between recognition discriminability and RSFC in the hippocampal cluster from the main analysis (Fig. 1C), while partialing out age and sex. Figure 2 displays a scatter plot of discriminability performance as a function of hippocampal RSFC in the total sample, indicating decreasing DMN synchronization with increasing recognition discriminability.
Total discriminability raw score plots as a function of hippocampal DMN synchronization. Red, Carriers; blue, noncarriers. The fit line illustrates the significant negative linear association (r = −0.32, p < 0.01, partialing out age and sex). When we removed the outlier (synchronization > 20), the relationship was weaker but still significant (r = −0.27, p < 0.01).
To examine other regions outside hippocampus, we correlated synchronization within all clusters from the full-brain analysis with the recognition discriminability performance. Here, we also found a significant negative relationship (r = −0.27, p < 0.01). To further explore the spatial variability, we correlated all significant clusters from the full-brain analysis with the recognition discriminability independently. All clusters showed a weak negative relationship between synchronization and the performance on memory, but only the correlations in right precentral gyrus (r = −0.21, p < 0.05) and right frontal orbital cortex (r = −0.20, p < 0.05) were significant.
Effect of age on RSFC within the APOE ε4 carrier group
We tested whether the RSFC in hippocampus and amygdala would fall off with age or remain high within the APOE ε4 carrier group. Briefly, we found no effects of increasing age in this group. Voxelwise statistics were performed using only the independent component spanning the posterior DMN and were regarded as significant at p < 0.05, corrected for multiple comparisons. Results revealed no significant effects of age across APOE carrier groups. No significant effect was found when contrasting with APOE3 × age—negative and positive slopes—nor APOE4 × age—negative and positive slopes. We found one significant voxel when contrasting APOE3 × age > APOE4 × age, but correcting for outliers removed these effects. General linear model analysis on each reported cluster (Table 3) revealed no significant interaction of age and APOE ε4 carrier status after outlier corrections.
Discussion
We tested the effects of APOE ε4 on DMN synchronization in 95 healthy individuals between 50 and 80 years of age using gICA and dual regression. In line with our hypothesis, we demonstrated increased hippocampal synchronization in carriers relative to noncarriers of the ε4 allele in an RSN spanning the posterior DMN. Whole-brain analysis revealed extended effects into the PCC, parietal, and parahippocampal regions. The lack of effects in other RSNs indicates specific effects of APOE on the functional coherence of the posterior DMN, including MTL and retrosplenial cortices, known to be early targets in AD. Furthermore, we report a negative relationship between memory performance and resting hippocampal DMN synchronization, which indicates that intrinsic functional connectivity patterns have neurocognitive correlates. Implications of the results are discussed in detail below.
APOE ε4 is linked to accelerated brain aging and disease (Small et al., 2000; Deary et al., 2002; Wilson et al., 2002; Bertram et al., 2007; Espeseth et al., 2008). As a risk factor for AD, there is evidence suggesting that the role of the apolipoprotein in amyloid β (Aβ) aggregation and clearance is more important than its fat-transport properties (Mahley and Rall, 2000; Kim et al., 2009; Herrup, 2010). It has been suggested that the main mechanism of APOE in AD is via the effects on Aβ metabolism (Kim et al., 2009). This is supported by findings of low concentrations of Aβ42 in the CSF of cognitively normal middle-aged ε4 carriers (Sunderland et al., 2004), indicative of earlier cerebral amyloid deposition.
Previous studies have documented a connection between Aβ deposition and DMN connectivity. Pittsburgh Compound-B (PiB) (Klunk et al., 2004) PET studies have revealed retention of Aβ in posterior parietal regions near PCC and retrosplenial cortex in AD (Nordberg, 2004; Archer et al., 2006; Nordberg et al., 2010). Amyloid burden in posterior regions making up the memory network suggests a mechanism by which amyloid toxicity might disrupt memory function (Buckner et al., 2005). Similar Aβ deposition have been observed in subjects with increased genetic risk for AD (Reiman et al., 1996). A PET study by Buckner et al. (2005) demonstrated links between deposition of Aβ, cortical atrophy, and hypometabolism in PCC. This suggests that PCC is part of an intrinsic network specifically disrupted in early stages of AD. This hypothesis is supported by three recent studies combining PiB-PET and fMRI (Hedden et al., 2009; Sperling et al., 2009; Sheline et al., 2010) reporting aberrant DMN activation in individuals with high amyloid burden. Consequently, increased DMN synchronization in ε4 carriers may reflect increased Aβ deposition. However, without CSF biomarker data, such interpretations should be made with caution.
The functional significance of the observed increased hippocampal synchronicity with the DMN in carriers is not clear, but it seems reasonable to hypothesize that it could stem from a failure in hippocampal decoupling. A possible scenario is that aberrant hippocampal decoupling in ε4 carriers results in increased hippocampal DMN synchronization during rest. This notion is partly supported by findings suggesting that age-related memory impairment may be primarily related to loss of deactivation in medial parietal regions (Miller et al., 2008). Also, a recent longitudinal study reported no difference in fMRI activity over 2 years in healthy individuals, whereas preclinical subjects demonstrated a decrease in hippocampal activity during the same period (O'Brien et al., 2010). Individuals showing more rapid cognitive decline demonstrated both increased hippocampal activation at baseline and accelerated longitudinal loss of hippocampal activation (O'Brien et al., 2010), suggesting that hippocampal hyperactivation is predictive of future cognitive decline. Decreased hippocampal decoupling during rest might increase the total hippocampal metabolic burden in carriers and modulate the present increased DMN synchronization. This is in line with our findings of a negative correlation between hippocampal DMN synchronization and performance on a test of memory function. We propose that aberrant hippocampal decoupling in ε4 carriers might be linked to an increased Aβ burden. However, it remains unclear whether Aβ mainly exerts its effects via intracellular/intercellular signaling (Sanchez-Mejia et al., 2008), through decreased white matter myelination (Bartzokis et al., 2007), or by perturbing the receptor/ligand efficacy (Yun et al., 2005).
Whereas APOE ε4 may modulate task-related BOLD signal independently of AD risk and pathology (Ringman et al., 2011), increased resting metabolic burden in the memory system of carriers could be related to increased susceptibility of cognitive impairment later in life. This was partly supported by the negative correlation between the hippocampal DMN synchronization and memory performance in the current study and is consonant with a previous task–fMRI study (Celone et al., 2006). Because the DMN correlate negatively with task-related networks (Raichle and Snyder, 2007), increased hippocampal activity during rest might decrease hippocampal activation relative to baseline during memory tasks, explaining an apparent deficiency in memory-related hippocampal activation (Fleisher et al., 2009b). However, further studies delineating APOE effects on the functional interactions between task and rest are needed (Trachtenberg et al., 2010). Also, longitudinal studies are warranted to determine whether increased DMN synchronization is associated with higher risk of pathological brain and cognitive changes at a later stage.
The present study has limitations. Like most studies using convenience sampling, our sample represents a highly functioning subgroup of the population. The stability of the specific DMN component showing effect of APOE in the present analysis has been demonstrated in several studies (Damoiseaux et al., 2006; Buckner et al., 2008; Smith et al., 2009) and is almost identical to a component with high intrasession and intersession test–retest reliability (Zuo et al., 2010). At model order 42, this component showed the highest reproducibility along with a medial and lateral visual network (Zuo et al., 2010). This indicates a reliable intrinsic RSN.
However, choice of analytical approach could have influenced the results. Decreased DMN connectivity within MTL structures in elderly individuals with high amyloid burden has been demonstrated (Hedden et al., 2009; Sheline et al., 2010). These findings partly contrast with our and previous findings of increased DMN connectivity in ε4 carriers (Filippini et al., 2009; Fleisher et al., 2009a). Importantly, the two former studies used seed-based correlation analysis, which differs from data-driven approaches in several aspects. ICA and dual regression represent a data-driven multivariate approach that effectively regresses out the common variance from all other included components, including RSNs showing various degree of spatial overlap. Furthermore, seeds based purely on anatomical definitions might yield results highly sensitive to the exact region of interest (ROI) placement and, therefore, vulnerable to operator-dependent variability and decreased interstudy reliability. This is in line with a recent comprehensive functional network analysis that demonstrated that functionally inaccurate ROIs may damage the network estimation, suggesting that results derived from inappropriate ROI definitions should be regarded with caution (Smith et al., 2011). Employing a data-driven approach like ICA in conjunction with dual regression might minimize the impact of inappropriate seeds.
In line with this reasoning, the present results are very similar to the findings of Filippini et al. (2009) using almost identical methods. This indicates reliable effects of genotype within the same analytical framework. However, we emphasize that the development of sensitive and reliable measures of functional connectivity is an active area of research, and future studies are needed to uncover the stability of APOE effects across analytical approaches and imaging modalities, including measures of microstructural integrity (Heise et al., 2010) and the relation to resting BOLD coherence. Several non-Aβ-mediated mechanisms are significant mediators of AD pathology (Herrup, 2010), and future studies might also be able to test the hypothesis that APOE effects are not specifically related to AD pathology but to some unidentified function of the gene, for example, on the cerebral vascular reactivity (Ringman et al., 2011).
Conclusively, we have documented increased resting state functional hippocampal coupling to the posterior DMN in healthy middle-aged and elderly carriers of the APOE ε4 allele that could not be explained by hippocampal volumes or GM density. Posterior DMN synchronization was negatively correlated with memory performance, suggesting a neurocognitive significance of the findings. RSFC measures might thus be sensitive to genotypic differences related to increased risk of AD several years before progression of AD-related neurocognitive decline. Our findings are in agreement with the notion that APOE plays an important role in brain function, notably within the DMN and memory systems, which are tightly coupled to the loci of AD-related pathology. The hippocampal loci possibly reflect decreased decoupling during rest, which could in turn lead to elevated metabolic burden and increased vulnerability of the memory system in ε4 carriers. The significant negative correlation between hippocampal DMN synchronization and memory performance supports this interpretation. Our findings provide an important replication and extension of specific genotype effects on intrinsic functional coherence, supporting the use of RSFC as an endophenotype in imaging genetics studies (Filippini et al., 2009; Glahn et al., 2010).
Footnotes
-
This study was supported by Grant 911397 from the Western Norway Regional Health Authority to A.J.L., by Grant 911593 from Helse Vest (MedViz/Quantitative Brain MR Imaging in Aging and Neurodegenerative Disorders) to A.L., and by the Research Council of Norway to L.T.W. We thank Dr. Jonn-Terje Geitung and the Department of Radiology at the Haraldsplass Deaconess Hospital for providing access to the MRI facility and the participants who made this study possible.
-
The authors declare no competing financial interests.
- Correspondence should be addressed to Dr. Erling Tjelta Westlye, Neuroinformatics and Image Analysis Laboratory, Department of Biomedicine, University of Bergen, Jonas Lies vei 91, N-5009 Bergen, Norway. Erling.Westlye{at}student.uib.no
