Abstract
Background:
It is critical to identify individuals at risk for Alzheimer’s disease (AD) earlier in the disease time course, such as middle age and preferably well prior to the onset of clinical symptoms, when intervention efforts may be more successful. Genome-wide association and candidate gene studies have identified single nucleotide polymorphisms (SNPs) in APOE, CLU, CR1, PICALM, and SORL1 that confer increased risk of AD.
Objective:
In the current study, we investigated the associations between SNPs in these genes and resting-state functional connectivity within the default mode network (DMN), frontoparietal network (FPN), and executive control network (ECN) in healthy, non-demented middle-aged adults (age 40 –60; N = 123; 74 females).
Methods:
Resting state networks of interest were identified through independent components analysis using a template-matching procedure and individual spatial maps and time courses were extracted using dual regression.
Results:
Within the posterior DMN, functional connectivity was associated with CR1 rs1408077 and CLU rs9331888 polymorphisms (p’s < 0.05). FPN connectivity was associated with CR1 rs1408077, CLU rs1136000, SORL1 rs641120, and SORL1 rs689021 (p’s < 0.05). Functional connectivity within the ECN was associated with the CLU rs11136000 (p < 0.05). There were no APOE- or PICALM-related differences in any of the networks investigated (p’s > 0.05).
Conclusion:
This is the first demonstration of the relationship between intrinsic network connectivity and AD risk alleles in CLU, CR1, and SORL1 in healthy, middle-aged adults. These SNPs should be considered in future investigations aimed at identifying potential preclinical biomarkers for AD.
INTRODUCTION
Alzheimer’s disease (AD) is the most common cause of dementia and the fifth-leading cause of death in adults aged 65 years and older [1]. With the increasing population age, prevalence of AD is projected to more than double by 2050 [2], posing a significant public health crisis. Unfortunately, currently available treatments for AD are largely ineffective, most likely because treatments are delivered following diagnosis, when significant neurodegeneration has already occurred and cannot be reversed [3]. Identifying biomarkers to detect individuals earlier in the trajectory of the disease, such as middle age, prior to significant pathological brain changes, may improve intervention and treatment efforts. Studying asymptomatic individuals who are at genetic risk for AD is one method that may be useful in elucidating potential AD biomarkers. Further, this strategy may reveal intermediate phenotypes between genetics and behavioral symptoms, pinpointing neural systems that are directly influenced by genetic risk variants [4].
The greatest genetic risk factor for AD is conferred by apolipoprotein E (APOE) [5]. Located on chromosome 19q13.2, APOE encodes apolipoprotein E which binds to and assists with amyloid-β (Aβ) clearance [6], one of the two neuropathological hallmarks of AD. The three common alleles of APOE include ɛ2, ɛ3, and ɛ4, where the ɛ4 allele increases risk and decreases age of onset for AD [3], while the ɛ2 allele is considered to be protective [7]. APOE ɛ4 genotype has been linked to multiple AD biomarkers including cerebrospinal fluid Aβ and tau levels [8–10], reduced glucose metabolism [11, 12], and medial temporal lobe atrophy [13, 14] in patients with mild cognitive impairment (MCI) and AD.
Additional AD susceptibility variants, including single nucleotide polymorphisms (SNPs) in SORL1, PICALM, CLU, and CR1, have been identified using genome-wide association studies (GWAS) [15, 16] or candidate gene [17, 18] approaches. Polymorphisms within these genes are proposed to increase AD risk through amyloid-dependent mechanisms [19, 20]. Similar to APOE ɛ4, risk alleles in SORL1, PICALM, CLU, and CR1 have been linked to alterations in medial temporal lobe integrity, including hippocampal volume [21–23], and entorhinal thickness [21, 24], regions affected early in AD pathogenesis [25].
Recent evidence suggests that resting state functional magnetic resonance imaging (rs-fMRI) could serve as an AD biomarker because connectivity patterns of functional networks change throughout the AD spectrum [26–32], connectivity alterations from baseline can be predictive of later conversion to AD [33, 34], and system-level alterations are detectable prior to significant neurodegeneration [35]. Therefore, rs-fMRI may serve as a key endophenotype associated with AD, prior to clinically significant symptom expression. Here, we focus on three networks associated with cognitive function (i.e., the default mode network (DMN), executive control network (ECN), and frontoparietal network (FPN)), containing regions that correspond to some of the earliest sites where AD neuropathology accumulates [36–38] and ultimately affects connectivity during preclinical stages [38, 39] and later in AD [26, 40–42]. Moreover, these networks are associated with memory and executive functioning, cognitive domains that are among the first affected in the disease [43].
The DMN, which is active in rest conditions and inactive during task performance, can be separated into functionally-unique anterior (aDMN) and posterior (pDMN) subsystems [41, 45]. The aDMN, which includes medial PFC, contributes to self-referential thought, while the posterior DMN (pDMN), which includes the posterior cingulate cortex (PCC) and precuneus, is associated with memory performance [44]. Disease status divergently affects the DMN subsystems; a cross-sectional investigation of DMN connectivity demonstrated that compared to non-demented adults, AD patients show lower pDMN connectivity but higher aDMN connectivity [35]. The ECN consists of medial-frontal areas, and the FPN includes the lateral frontal and temporoparietal regions [46]. The ECN and FPN are implicated in multiple cognitive functions including working memory and executive function [46]. Functional connectivity within the ECN is higher in AD [26], whereas FPN connectivity in AD is varied, as frontal regions show higher connectivity and temporoparietal regions show lower connectivity [26, 42].
Connectivity within the DMN, ECN, and FPN networks are also modulated by genotype. Zhu and colleagues [32] demonstrated an interaction between APOE genotype and disease status on DMN intrinsic connectivity. Compared to non-ɛ4 carriers, ɛ4 carriers exhibited approximately opposite patterns of connectivity in the DMN across the disease spectrum, an effect that was particularly pronounced in patients with MCI. For instance, MCI APOE ɛ4 patients showed lower connectivity in right frontal regions of the DMN but higher connectivity in the left hippocampus, middle temporal gyrus, and right precuneus compared to non-ɛ4 carriers [32]. Alterations in DMN connectivity are also evident prior to significant amyloid deposition in non-demented APOE ɛ4 carriers, particularly in the posterior regions of the DMN [35, 47–49].
While the association between APOE genotype and connectivity has been established [32, 47], relatively little work has been done to characterize the relationships between SNPs in PICALM, CLU, CR1, and SORL1 and intrinsic network connectivity [50–52]. Further, most previous work characterizing the relationships between polymorphisms implicated in AD risk and connectivity have largely excluded middle age. Given that AD intervention may be more successful during middle age, when AD pathophysiology has likely begun but clinical symptoms are not yet expressed, it is critical to explore this time period.
In the current study, we characterized the relationships between SNPs in five candidate genes (see Table 1) associated with AD (i.e., APOE, SORL1, PICALM, CR1, and CLU) and functional connectivity within three resting state networks (i.e., DMN, ECN, and FPN) in healthy, non-demented, middle-aged adults. We predicted that risk alleles in each of the candidate genes would be associated with lower DMN functional connectivity, particularly in posterior portions of the network, higher ECN connectivity, and higher connectivity within frontal regions but lower connectivity within temporoparietal regions of the FPN network compared to non-carriers.
Distribution of SNPs within sample.
*ɛ4 row displays ɛ4 allele-carriers (N = 34) and non-carriers (N = 88). This table displays the alleles for each SNP, minor allele in the current sample, and sample size for each genotype.
MATERIALS AND METHODS
Participants
Sample included 123 community-dwelling adults (age 40–60; M = 50.03, SD = 6.03; 74 females). Participants did not have a history of neurological (e.g., dementia, Parkinson’s disease, epilepsy) or psychiatric disorders (e.g., schizophrenia, bipolar disorder), severe cardiac disease (e.g., myocardial infarction, coronary bypass surgery, angioplasty), metastatic cancer, or substance use disorder. Participants were screened for global cognitive impairment and depression using the Mini-Mental State Exam (MMSE; MMSE≥25) [53], the Mattis Dementia Rating Scale Second Edition (DRS-2; DRS-2≥136) [54], and the Geriatric Depression Scale (GDS; GDS≤10) [55]. All participants provided written informed consent and were given financial compensation for their participation. The study was carried out in accordance with the guidelines of the institutional review boards at the University of Wisconsin-Milwaukee and the Medical College of Wisconsin.
Genotyping
The five genes examined in the current study (APOE, CLU, CR1, PICALM, SORL1) were selected based on their promising association with the neuropathology of dementia [19]. A small sample of blood (10 mL) was drawn from each participant to obtain DNA. SNPs in APOE (rs7412, rs429358, rs157580, rs157582, rs405509, rs405697, rs429358, rs439401, rs8106922), CLU (rs11136000, rs1532278, rs9331888, rs9331908), CR1 (rs1408077, rs3818361), PICALM (rs3851179, rs541458, rs592297), and SORL1 (rs1010159, rs1131497, rs1614735, rs1699102, rs2282649, rs3824968, rs641120, rs668387, rs689021) were sequenced at the University of Wisconsin Biotechnology Core Facility.
MR imaging
MR was conducted on a GE Signa 3T scanner (Waukesha, WI) with a quad split quadrature transmit/receive head coil. The 8-min rs-fMRI was acquired during a multimodal imaging session that lasted 1 h and 15 min. All participants were screened for any contraindications to MRI.
A T2*-weighted functional scan was acquired with an echo-planar pulse imaging (EPI) sequence (28 axial slices, 20×20 cm2 FOV, 64×64 matrix, 3.125 mm×3.125 mm×4 mm voxels, TE = 40 ms, TR = 2000 ms). The fMRI scan was acquired under a task-free condition during which individuals were instructed to relax, keep their eyes closed and to “not think about anything in particular”.
Data processing and analysis
The resting state data was processed and analyzed using Analysis of Functional NeuroImages (AFNI) [56] and FMRIB Software Library (FSL) [57] similar to Korthauer et al. [58] in accordance with methods used in the Human Connectome Project [59]. Preprocessing included removing the first four slices of the EPI data, slice timing corrections, despiking, registration of each volume to the first volume, and removal of non-brain tissue from EPI volumes. A high pass filter (0.01 Hz) was applied to remove low-frequency drift, and the data were spatially smoothed using a 6-mm FWHM Gaussian filter.
Independent component analysis (ICA) was conducted on each subject’s 4D preprocessed dataset to identify artefactual components using FSL’s MELODIC tool [60]. The output of ICA includes statistically independent spatial maps with associated time courses for each subject, which was inspected by two independent raters. Inter-rater agreement for identification of artefacts was high (Cohen’s κ= 0.85). Individual datasets were denoised by regressing out these artefactual components.
Denoised data were then entered into a group ICA manually set to yield 20 independent components. The DMN, ECN, and FPN were identified through visual inspection of the resulting components spatial pattern and validated using a template-matching procedure [46]. Briefly, average z-score of voxels within the template and voxels outside of the template were extracted. The best-fit component was the component with the highest score resulting from subtracting the average outside-of-template voxels z-score from the average of within-template voxels z-score, as previously described in the literature [40, 58]. Following identification of the networks, dual regression [61, 62] was performed to calculate participant-specific spatial maps and time courses which were used to examine intrinsic connectivity of each network.
Statistical analysis
The relationship between genotype and functional connectivity was tested voxel-wise using 5000 permutations of nonparametric permutation testing through FSL’s randomize tool [63]. Each analysis was restricted to voxels contained within the resting state network of interest output from the group ICA. Similar to past work [58], for each SNP, separate ANOVAs were constructed. A dominant genetic model (i.e., two groups; minor allele carriers [homozygote or heterozygote] versus major allele homozygotes) was used for SNPs with fewer than 10 minor allele homozygote carriers. For SNPs with 10 or more minor allele homozygotes, an additive genetic model was utilized (i.e., three groups; minor allele homozygotes, heterozygotes, and major allele homozygotes). Resulting statistical maps were cluster-corrected using Threshold-Free Cluster Enhancement (TFCE) [64]. We report results at family-wise error corrected p < 0.05 (voxel level) given a priori hypotheses (more stringent Bonferroni correction yields a p-value of 0.05/26 = 0.002 based on the number of SNPs tested), and a cluster size of ≥10 voxels [28].
G*Power [65] was used to conduct a power analysis for ANOVA. Assuming an additive genetic model and alpha = 0.05, the study had a power of 0.84 to detect a moderate-sized effect (d = 0.3). Thus, the sample size was adequate to power the analyses and detect the association between genotype and intrinsic connectivity.
RESULTS
ICA results
Of the 20 components output from the group ICA, four corresponded to the DMN, ECN, and FPN (see Fig. 1). The DMN was separated into anterior and posterior subnetworks. The pDMN consisted of the precuneus, PCC, hippocampi, middle temporal gyri, angular gyri, lateral occipital cortices, middle frontal gyri, and frontal pole. The aDMN consisted of the superior frontal, middle frontal, inferior frontal, PCC, and temporal pole. The FPN included the middle frontal gyri, superior frontal gyri, superior parietal lobes, angular gyrus, lateral occipital gyri, PCC, and ACC. The ECN contained the medial frontal cortex, the orbitofrontal cortex, and frontal pole.
Resulting spatial maps from group ICA of resting-state networks of interest: A) posterior default mode network (DMN); B) anterior DMN; C) frontoparietal network; and D) executive control network. The figure displays three orthogonal slices of each network on the Montreal Neurological Institute standard brain, in radiological convention.
Associations between SNPs and functional connectivity
See Fig. 2 for a summary of significant results by SNP. For statistically significant comparisons, cluster-wise results are reported in Table 2 and depicted in Fig. 3. Given the a priori hypotheses, we report the results at TFCE family-wise error corrected p < 0.05 along with a Bonferroni-corrected threshold (p = 0.002). The pattern of results remains largely the same when we control for age, sex, and education, with one exception-CLU rs9331888 was no longer associated with pDMN connectivity (p > 0.05) and a significant relationship CR1 rs1408077 and aDMN connectivity emerged (p < 0.05). None of the findings survive a stringent Bonferroni correction for multiple comparisons (p’s > 0.002).
Associations between SNPs in APOE, CLU, CR1, PICALM, and SORL1 and resting-state connectivity in (a) posterior default mode network (DMN), (b) anterior DMN, (c) frontoparietal network, and (d) executive control network.
Summary of functional connectivity findings by resting state network
Summary of threshold-free cluster enhancement family-wise error corrected results at p < 0.05 (voxel level). RSN, resting state network; k, cluster size; pDMN, posterior default mode network; FPN, frontoparietal network; ECN, executive control network; min, minor allele homozygote; het, heterozygotes; maj, major allele homozygotes.
Associations between SNPs in CLU, CR1, and SORL1 and intrinsic connectivity in the pDMN, ECN, and FPN. Depicted in red are the significant family-wise corrected clusters from T-map comparisons (ps < 0.05) superimposed atop binarized clusters colored in blue (ps < 0.20) to aid visualization. The genotype contrast is indicated below each SNP. pDMN, posterior default mode network; FPN, frontoparietal network; ECN, executive control network; min, minor allele homozygotes; het, heterozygote; maj, major allele homozygotes.
Anterior default mode network
There were no significant associations between SNPs in APOE, CLU, CR1, PICALM, or SORL1 and connectivity within the aDMN (p’ > 0.05).
Posterior default mode network
Minor allele homozygotes of CLU rs9331888 had higher connectivity than major allele homozygotes in left superior temporal gyrus of the pDMN (p < 0.05). Major allele homozygotes of CR1 rs1408077 had higher connectivity of the pDMN in the right precuneus than minor allele carriers (p < 0.05). There were no associations between SNPs in APOE, PICALM, or SORL1 with pDMN connectivity.
Frontoparietal network
Minor allele homozygotes of CLU rs11136000 had higher connectivity than major allele homozygotes in the left lateral occipital lobe of the FPN (p < 0.05). Major allele homozygotes of CR1 rs1408077 had higher connectivity of the FPN in the left frontal pole than minor allele carriers (p < 0.05). Major allele homozygotes of SORL1 rs641120 had higher connectivity than minor allele homozygotes in the FPN within the right precentral gyrus (p < 0.05). Minor allele homozygotes of SORL1 rs689021 had higher FPN connectivity than heterozygotes in the left frontal orbital cortex (p < 0.05). There were no associations between SNPs in APOE or PICALM with FPN connectivity.
Executive control network
Minor allele homozygotes of CLU rs11136000 exhibited higher connectivity than heterozygotes in the left frontal lobe of the ECN (p’s < 0.05). There were no associations between SNPs in APOE, CR1, PICALM, or SORL1 with ECN connectivity.
DISCUSSION
The purpose of the current study was to examine the relationships between functional connectivity and SNPs in genes associated with AD risk in healthy, non-demented middle-aged adults. Our findings revealed associations between polymorphisms in CLU, SORL1, and CR1 and altered connectivity within the pDMN, FPN, and ECN. There were no relationships, however, between functional connectivity and APOE [66, 67] or PICALM [50] in any of the resting state networks of interest in the current study compared to existing reports.
CLU rs1136000 and rs9331888 and CR1 rs1408077 are associated with higher risk for AD [15, 69]. Although neuroimaging studies examining associations of brain integrity with these risk alleles are limited, literature supports structural differences for AD risk allele carriers, including lower white matter microstructural integrity of the fornix and fronto-temporo-parietal tracts [70] as well as lower entorhinal cortex integrity [21, 71]. Our results extend these findings to demonstrate disrupted functional connectivity in resting state networks associated with memory and executive functioning. The current study also contributes to the mounting evidence in support of the relationships between SORL1 and AD-associated biomarkers [18, 72–74], although the literature on this specific topic is still somewhat equivocal [75, 76].
Disruptions to posterior regions of the DMN and connectivity between the DMN and FPN are among the earliest connectivity alterations identified in relation to AD progression [35, 38]. The DMN in particular features a few major cortical hubs, including the PCC, precuneus, and orbitofrontal cortex, which are highly connected to other regions of the brain and serve to integrate diverse information supporting cognitive function [36]. It is hypothesized that Aβ accumulation may begin in regions of the brain with hub-like qualities due to their high level of activity and metabolic demand [77]. This hypothesis is supported by evidence in the animal literature showing that regional Aβ deposition is governed by region-specific neuronal activity [78] and that regions of high metabolism in young adulthood exhibit more PET-detected Aβ pathology in older adulthood [79]. Indeed, there is high correspondence between regions of the DMN and the topography of Aβ deposition [36, 37] but intriguingly alterations in posterior DMN connectivity are evident prior to significant Aβ burden in the brain [38] suggesting that dysfunction in major hub regions of the network may precede and propagate Aβ aggregation. Although the exact mechanisms of risk conferred by single nucleotide polymorphisms in SORL1, CLU, and CR1 are unknown, it is hypothesized to be related to alterations in processing of amyloid-β protein precursor or clearance of Aβ [19], potentially exacerbating Aβ-burden. Our data extend the previous literature by demonstrating that even in middle age, years prior to the onset of clinical symptoms, relationships between risk alleles in CLU, CR1, and SORL1 and functional connectivity in cortical hubs of resting state networks vulnerable to AD neuropathology are evident. Polymorphisms in CLU, CR1, and SORL1 should be considered in future investigations examining preclinical biomarkers of AD to better understand the link between early amyloid pathology, network disruption, and decline to AD.
Disruptions in DMN and ECN connectivity in middle-aged APOE ɛ4 carriers have been reported; however, findings are equivocal, with some studies reporting higher connectivity among ɛ4 carriers [67], while others reporting lower ɛ4-associated connectivity [66, 80] or no significant differences between ɛ4 carriers and non-carriers [58, 81]. Our findings align with the literature suggesting that APOE-related connectivity differences are not evident through resting state network analyses in middle age. While inconsistencies in the literature could be a simple byproduct of methodological differences between studies, such as statistical power, analytical approaches, or the age groups sampled, they could also be explained by the proposed antagonistic pleiotropic effects of APOE ɛ4 status [82]. This model suggests that ɛ4 status differentially impacts evolutionary fitness throughout the lifespan and proposes that ɛ4 exerts beneficial effects in young adulthood but detrimental effects in older age. Studies of rs-fMRI connectivity support this proposal, as young adult ɛ4 carriers show stronger connectivity in the DMN than ɛ3 carriers [83], but healthy, non-demented older adult ɛ4 carriers exhibit lower connectivity in the DMN [35, 48]. Middle age could represent a transition between these two stages of ɛ4-related effects on fitness, and thus alterations in connectivity due to genotype during middle age could be subtle, if present at all. Given that we are capturing a cross section during middle age in the current study, individuals may be at different transition points of the antagonistic pleotropic trajectory, and thus, may offset any findings in this age range. More sensitive techniques may be necessary to detect such differences. As demonstrated by Korthauer and colleagues [58], ɛ4-related differences in functional connectivity in middle age were only evident in graph theoretical metrics on a combined functional-structural network but not conventional resting-state network analyses. Longitudinal studies are needed to better characterize the effects of APOE genotype on functional connectivity in middle age, to understand at what point in the aging process APOE ɛ4 begins to exert detrimental effects.
We report no relationships between polymorphisms in PICALM and connectivity within FPN, DMN, or ECN. Variants in PICALM have been associated with several neuroimaging measures of structural and functional brain integrity, including entorhinal cortex thickness [24] and altered DMN connectivity [84]. However, these studies were conducted in older adult samples. It could be that effects of PICALM polymorphisms do not become evident until late life; however, more work is necessary in order to understand the associations of endophenotypes with PICALM risk.
Our study is not without limitations. Candidate gene approaches have received much criticism for generating false positives due to the risk of inflated type I error [85]. Although we had sufficient power to detect modest SNP-related differences in functional connectivity, studies with larger samples should be conducted to replicate and expand on our preliminary findings, albeit replication is the current gold standard for genetic associations regardless of sample size or the approach (GWAS versus candidate gene) [86]. Our study was also limited by the cross-sectional design. Longitudinal analyses are necessary to map the clinical implications of resting state network reorganization and to determine which risk allele carriers will develop AD. Finally, in the current study we examined the independent influence of each gene on intrinsic connectivity but did not have the statistical power to pursue gene-gene interactions, which we believe is important in studying multi-factorial disorders with multiple risk genes, such as AD. Future investigations should employ a polygenic approach (e.g., gene interaction, polygenic risk score) to examine associations with resting-state network connectivity. This approach may better model disease variance and reveal important associations with endophenotypes that are not evident when examining each polymorphism in isolation.
Nonetheless, what are candidate gene studies good for? Information gathered through candidate gene studies reveal associations between biologically relevant genetic variants and disease traits which may contribute to our understanding of pathways that impact the observed heterogeneity in AD. To the best of our knowledge, this study is the first to demonstrate relationships between polymorphisms in CLU, CR1, and SORL1 associated with AD risk and resting-state network connectivity in healthy, non-demented middle-aged adults. Our results suggest that differences in major hubs of the pDMN, ECN, and FPN connectivity may be evident years prior to the onset of clinical symptoms. Neuroimaging genetic approaches show promise in identifying brain alterations early, and seemingly prior to onset of any potential clinical symptoms, and as such may have utility as biomarkers aiding in early AD detection and intervention.
