Abstract
Background:
Disease-modifying treatments for Alzheimer’s disease (AD) may be more successful if interventions occur early, prior to significant neurodegeneration and subsequent to the onset of clinical symptoms, potentially during middle age. Polymorphisms within BDNF, COMT, and KIBRA have been implicated in AD and relate to episodic memory and executive functioning, two domains that decline early in AD.
Objective:
The purpose of the current study was to use an endophenotype approach to examine in healthy, non-demented middle-aged adults the association between polymorphisms in BDNF, COMT, and KIBRA and functional connectivity within networks related to episodic memory and executive function (i.e., default mode network (DMN), executive control network (ECN), and frontoparietal network (FPN)).
Methods:
Resting state networks were identified using independent component analysis and spatial maps with associated time courses were extracted using a dual regression approach.
Results:
Functional connectivity within the DMN was associated with polymorphisms in BDNF (rs11030096, rs1491850) and KIBRA (rs1030182, rs6555791, rs6555802) (ps < 0.05), ECN connectivity was associated with polymorphisms in KIBRA (rs10475878, rs6555791) (ps < 0.05), and FPN connectivity was associated with KIBRA rs6555791 (p < 0.05). There were no COMT-related differences in functional connectivity of any of the three networks investigated (ps > 0.05).
Conclusion:
Our study demonstrates that in middle age, polymorphisms in BDNF and KIBRA are associated with altered functional connectivity in networks that are affected early in AD. Future preclinical work should consider these polymorphisms to further elucidate their role in pathological aging and to aid in the identification of biomarkers.
Keywords
INTRODUCTION
Dementia is characterized by impairment in multiple cognitive domains including learning and memory, language, and executive functioning, which significantly impacts independence [1]. Alzheimer’s disease (AD), the most common form of dementia, is associated with two neuropathological hallmarks, amyloid-β (Aβ) plaques and neurofibrillary tangles, along with progressive cognitive decline, particularly in learning and memory [1]. The treatments that are currently available are largely ineffective because by the time of diagnosis, when treatments are typically delivered, pathology and neurodegeneration are significant and may be irreversible [2]. As the prevalence of AD is projected to more than double by 2050 [3], identifying preclinical biomarkers that can pinpoint individuals most susceptible to AD will be essential to improve early intervention and treatment outcomes.
Episodic memory and executive function, two complex polygenic cognitive domains [4, 5], are affected early in the course of AD, with accelerated decline in each cognitive domain evident three to seven years prior to diagnosis [6, 7]. Single nucleotide polymorphisms (SNP) in genes that contribute to such cognitive functions, including BDNF, COMT, and KIBRA, have also been implicated in AD [8–13] by altering the functioning or availability of gene products, ultimately affecting cognitive function in domains vulnerable to AD. KIBRA and BDNF are both highly expressed in the hippocampus [14, 15], an area that supports learning and memory. Polymorphisms in KIBRA and BDNF are related to the variability observed in episodic memory performance [15, 16]. COMT, which encodes an enzyme that breaks down dopamine in the prefrontal cortex, is related to executive function [13]. While the measurable impact of polymorphisms within BDNF, COMT, and KIBRA on cognition is expected to be minimal, investigating intermediate phenotypes, such as functional brain integrity, may reveal more direct associations between genes and behavior and uncover unique neural pathways that underlie cognitive dysfunction.
Mounting evidence suggests the promise of resting state functional magnetic resonance imaging (fMRI) as a biomarker of AD during preclinical and clinical stages [17–19]. Examining functional connectivity of distinct brain networks which underlie memory and executive function, such as the default mode network (DMN), executive control network (ECN), and frontoparietal network (FPN), could be a particularly useful intermediate phenotype. The DMN contains a few major cortical hubs, including the precuneus, posterior cingulate cortex (PCC), and prefrontal cortex, and is the most-well characterized network in aging and AD [20]. The DMN can be parcellated into two subnetworks consisting of anterior (aDMN) and posterior (pDMN) divisions [21–23]. Each subnetwork is differentially related to cognition; the aDMN supports self-referential thought while the pDMN is related to episodic memory [21]. Compared to healthy controls, AD patients exhibit higher aDMN but lower pDMN connectivity [22, 24–26]. The PCC and precuneus, two hub regions in the pDMN, are especially vulnerable to AD pathology [27], and connectivity of these regions is among the first to be disrupted in AD [26], though the direction of findings in preclinical stages (i.e., hyperconnectivity versus hypoconnectivity) is inconsistent [23, 28]. In addition to the DMN, both the FPN and ECN are related to executive function and working memory [29], and functional connectivity within these networks is disrupted in disease [24, 31]. Compared to healthy controls and mild cognitive impairment patients, individuals with AD show higher connectivity within the ECN [24, 31], while connectivity of the FPN is altered in both directions, as frontal regions exhibit hypoconnectivity while parietal regions exhibit hyperconnectivity [24].
Most existing studies employing a genetic imaging approach examined the relationship between APOE ɛ4 (APOE4), the greatest genetic risk factor for AD, and functional brain integrity [32–36], though findings in middle age are equivocal. Our group recently demonstrated that in a healthy middle-aged sample, APOE4-related differences were only evident when using more sensitive techniques such as examining graph properties on a combined functional-structural network [35] rather than conventional resting state network metrics [36]. Moving beyond APOE4, few studies have examined the relationship between polymorphisms in genes associated with cognitive function (BDNF, COMT, KIBRA) and network connectivity [37–40], especially in the context of aging and AD. Moreover, middle age, a time period when AD pathophysiology has likely begun but clinical symptoms are not yet expressed, has largely been omitted from investigations to date. In the current study, the purpose was to expand our previous work and examine genes outside of APOE associated with cognitive function in domains vulnerable to AD, including BDNF, COMT, and KIBRA, and their relationship with functional connectivity within the DMN, ECN, and FPN in healthy, non-demented middle-aged adults (see Table 1). We predicted that healthy, non-demented middle-aged carriers of alleles in BDNF, COMT, and KIBRA that have been linked to worse cognitive performance and implicated in AD would show altered connectivity within the DMN, ECN, and FPN, with risk carriers exhibiting more AD-like connectivity patterns compared to non-risk carriers. Identifying genotype-related alterations in connectivity in middle age is critical for improving the characterization of biomarkers with potential for preclinical detection of AD.
Sample SNP Information
Dominant models consisted of two groups including minor allele carriers (i.e., minor allele homozygotes and heterozygotes) and major allele homozygotes and were employed when the sample size of minor allele homozygotes was less than 10. Additive models consisted of three groups (i.e., minor allele homozygotes, heterozygotes, major allele homozygotes) and were conducted when minor allele homozygote group size was greater than 10. APOE4 carrier status is defined as individuals carrying 1 or 2 ɛ4 alleles. Complete observations indicate the number of complete datapoints (i.e., participants with full profile of genotype and covariate [age, sex, education, APOE4] data) entered to the regression for each SNP.
MATERIALS AND METHOD
Participants
Participants were 123 adults from the Milwaukee community (age 40–60; M = 50.03, SD = 6.03, 49 males). Participants were in general good health and did not have any history of neurological conditions (e.g., stroke, dementia, Parkinson’s dis-ease, epilepsy), psychiatric disorders (e.g., schizoph-renia, bipolar disorder), severe cardiac disease (e.g., angioplasty, coronary bypass surgery), metastatic cancer, or substance use disorder. All participants met the following inclusion criteria: cognitively normal based on Mattis Dementia Rating Scale Second Edition [41] ≥136; Mini-Mental State Exam [42] ≥25; and non-depressed (Geriatric Depression Scale [43] ≤10. All adults gave informed consent and were compensated financially for their participation. The study was approved by and conducted in accordance with the guidelines of the University of Wisconsin-Milwaukee and the Medical College of Wisconsin institutional review boards.
Genotyping
The three genes were selected for analysis based on their promising associations with cognitive domains vulnerable to pathological aging and risk for AD [8–13, 16]. Haplotype tagging SNPs (tSNPs) were selected from the International Haplotype Map project (https://www.hapmap.org) using a bulk download of tag SNP data. To ensure robust haplotype tagging within the cohort, SNPs were selected based upon a pairwise r2 of greater than 0.8 and a minor allele frequency of greater than 0.2, all based on CEPH Caucasian data typed as a part of the HapMap project. The regions of the genome containing all exons of BDNF, KIBRA, and COMT plus at least 20kb of 5’ and 3’ flanking sequence were selected for tSNP analysis. This produced a list of 28 SNPs required to tag the genes of interest. In addition to these SNPs, we also assayed polymorphisms that make up the common ɛ2, ɛ3, and ɛ4 APOE genotypes (rs7412 and rs429358) and other well-established non-synonymous alterations in COMT (rs4680) and BDNF (rs6265) to produce a total of 32 SNPs to be assayed (Table 1).
All assays were performed at the University of Wisconsin Biotechnology Core Facility. DNA was obtained from a blood sample (10 mL) from each participant. DNA concentration was verified using the Quant-iT™ PicoGreen® dsDNA kit (Life Technologies, Grand Island, NY). DNA samples were first normalized/standardized to 0.5 ng/uL using epMotion 5075 and pH 7.5 10 mM Ultrapure TM Tris-HCl (Life Technologies, Grand Island, NY). KASPar assays were amplified using the Eppendorf Mastercycler pro384 and analyzed using a Synergy 2 (BioTek®) plate reader to measure the fluorescence of the samples using Gen5 TM software. All SNPs were in Hardy-Weinberg equilibrium, with the exceptions of KIBRA rs764221, which was removed from subsequent analyses. See Table 1 for a list of SNPs, minor allele frequencies, and sample size information.
MR image acquisition
MRI acquisition was conducted using a quad split quadrature transmit/receive head coil on a GE Signa 3T scanner (Waukesha, WI). The resting state fMRI scan (8 min) was acquired during a multimodal imaging session that lasted 1 h and 15 min. All participants were screened for MRI contraindications.
A T2*-weighted functional scan was acquired using 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 resting state fMRI scan was task-free, and individuals were told to close their eyes and “not think about anything in particular”.
Resting state fMRI processing and analysis
Preprocessing was carried out in a similar fashion to what was previously published [35, 36] using FSL [44] and AFNI [45], and in line with the methods used by the Human Connectome Project [46]. The first four volumes of the EPI data were removed followed by slice timing correction and despiking. Each volume was registered to the first volume and non-brain tissue was removed from the EPI data. To remove low-frequency drift from the data a high pass filter (0.01 Hz) was applied, and the data were smoothed spatially using a 6-mm FWHM Gaussian filter.
To identify artefactual components in the data, independent components analysis (ICA) using FSL’s MELODIC tool [47] was used on each participant’s preprocessed data and the output for each participant included statistically independent spatial maps with associated time courses. This output was inspected for artefacts by two independent raters, and inter-rater agreement was high (Cohen’s κ= 0.85). To denoise the data, these artefactual components were regressed out of each individual’s data. The denoised data was then entered into a group ICA and 20 components were extracted. Of the 20 components, four components were visually identified as the aDMN, pDMN, FPN, and ECN (Supplementary Figure 1) and verified using a template-matching procedure [29, 36]. Dual regression [48, 49] was next performed to obtain individual time courses and spatial maps for statistical comparison.
Statistical analysis
The association between genotype and functional connectivity was examined voxelwise using FSL’s randomize tool [50] with 5000 permutations of nonparametric permutation testing. Each analysis was constrained to the voxels contained within the output ICA spatial map of the resting state network of interest. Consistent with past work [35, 36] separate ANOVAs were created for each SNP (Table 1). For SNPs with fewer than 10 minor allele homozygote carriers, dominant genetic models were used (i.e., two groups; minor allele carriers, including both homozygotes and heterozygotes, versus major allele homozygotes). An additive genetic model was used (i.e., three groups: minor allele homozygotes, heterozygotes, and major allele homozygotes) for SNPs with 10 or more minor allele homozygotes. For all models, age, sex, education, and APOE4 genotype were entered as covariates. SNP sequencing was not successful for all participants, leading to different sample sizes for each SNP of interest (Table 1). To account for missing data for each SNP test, one column per missing data point was added to the regression matrix and was considered as a variable of no interest (covariate) in the analysis. For any given SNP, this effectively removes the resting-state data and associated degrees of freedom from each participant with missing genotyping. Statistical maps output from FSL’s randomise tool were cluster-corrected using threshold-free cluster enhancement [51]. Results are reported at family-wise error corrected p < 0.05 (voxel level) given a priori hypotheses, although strict Bonferroni correction would bring the p-value of 0.05/30 = 0.002 to account for the total number of SNPs tested, and a cluster size of ≥10 voxels [52].
RESULTS
Genotype differences in connectivity
A summary of the genotype-associated differences in connectivity can be found in Fig. 1. Cluster-wise results for each statistically significant comparison are reported in Table 2 and voxel-wise comparison maps are presented in Supplementary Figure 2. All results are reported at p < 0.05 (cluster and family-wise error corrected) after controlling for age, sex, education, and APOE4 genotype given the a priori hypotheses; with Bonferroni correction for multiple comparisons, p = 0.002. Only the associations between KIBRA rs6555791 and pDMN connectivity survived stringent Bonferroni corrections (p < 0.002).
Plot displays statistical associations between SNPs and functional connectivity within each resting state network of interest. Each dot represents a polymorphism in BDNF, COMT, or KIBRA (see Table 1 for listing). Dots above the p-value line of 0.05 represents SNPs that showed statistically significant associations with connectivity in the (A) aDMN; (B) pDMN; (C) ECN; and (D) FPN. Dots above the p-value line of 0.05/30 represent statistical associations that survived Bonferroni correction. All models controlled for age, sex, education, and APOE genotype. Connectivity within the aDMN, pDMN, ECN, and FPN are related to polymorphisms within BDNF (rs11030096, rs1491850) and KIBRA (rs1030182, rs10475878, rs6555791, rs6555802). Findings related to BDNF rs11030096 and aDMN connectivity were not reported in the results section as the cluster size was less than 10 voxels. aDMN, anterior default mode network; pDMN, posterior default mode network; ECN, executive control network; FPN, frontoparietal network.
Additional models were run excluding APOE4 genotype as a covariate (Supplementary Table 1). The results remain largely the same with a few exceptions. The associations between KIBRA rs6555802 and rs6555791 and connectivity were no longer observed (ps > 0.05). A relationship between KIBRA rs1030182 and aDMN connectivity was observed (p < 0.05), such that major allele homozygotes showed higher connectivity than minor allele homo-zygotes within the right superior frontal gyrus and right crus.
Anterior default mode network
Major allele homozygotes of BDNF rs1491850 exhibited higher functional connectivity of the aDMN within the left frontal pole than heterozygotes (p < 0.05; additive model). Major allele homozygotes showed higher connectivity of the left putamen, posterior middle temporal gyrus, caudate, middle frontal gyrus, and anterior middle temporal gyrus than minor allele homozygotes within the aDMN (ps < 0.05; additive model). Heterozygotes of KIBRA rs6555791 exhibited higher functional connectivity of the aDMN compared to major allele homozygotes, with significant clusters dispersed across the network in frontal, parietal, and temporal lobes (see Table 2 for specific regions; additive model). There were no significant associations between aDMN connectivity and SNPs in COMT.
Summary of resting state fMRI functional connectivity findings by network and SNP
Threshold-free cluster enhancement and family-wise error corrected clusters significant at p < 0.05. k, cluster size; aDMN, anterior default mode network; pDMN, posterior default mode network; ECN, executive control network; min, minor allele homozygote; het, heterozygotes; maj, major allele homozygotes; BA, Brodmann Area.
Posterior default mode network
Major allele homozygotes of BDNF rs11030096 exhibited higher pDMN connectivity compared to heterozygotes in the left precuneus and cerebellum (ps < 0.05; additive model). KIBRA rs1030182 heterozygotes and major allele homozygotes exhibited higher connectivity than minor allele homozygotes in the right thalamus (ps < 0.05; additive model). Heterozygotes of KIBRA rs6555791 had higher connectivity than major allele homozygotes in the right posterior parahippocampal gyrus and inferior lateral occipital cortex and left superior lateral occipital cortex and posterior temporal fusiform cortex (ps < 0.05; additive model). Minor allele homozygotes of KIBRA rs6555791 had higher connectivity within the precuneus, left superior lateral occipital cortex and intracalcarine cortex than major allele homozygotes (ps < 0.05; additive model). Heterozygotes of KIBRA rs6555802 exhibited lower connectivity than major allele homozygotes within the right lingual gyrus of the pDMN (p < 0.05; additive model). There were no significant associations between connectivity within the pDMN and SNPs in COMT.
Executive control network
Minor allele carriers of KIBRA rs10475878 had lower functional connectivity than major allele homozygotes in the left frontal pole, caudate, and insular cortex of the ECN (ps < 0.05; dominant model). Heterozygotes of KIBRA rs6555791 exhibited higher connectivity in the left frontal cortex, frontal pole, and caudate (ps < 0.05; additive model) compared to major allele carriers. There were no significant associations between SNPs in BDNF or COMT and ECN connectivity.
Frontoparietal network
Heterozygotes of KIBRA rs6555791 exhibited higher connectivity in the right middle frontal gyrus and left inferior frontal gyrus and bilaterally in the inferior temporal and superior lateral occipital gyrus (ps < 0.05; additive model) compared to major allele homozygotes. There were no significant associations between SNPs in BDNF or COMT and connectivity within the FPN (ps > 0.05).
DISCUSSION
The purpose of the current study was to characterize the relationships between SNPs in BDNF, COMT, and KIBRA and functional connectivity of the DMN, ECN, and FPN in healthy, non-demented, middle-aged adults. We did not find an association between polymorphisms in COMT and functional connectivity in any of the networks investigated, but our results suggest relationships between connectivity within the FPN, ECN, aDMN, and pDMN and BDNF and KIBRA genes after controlling for age, sex, education, and APOE4 genotype.
Our findings demonstrate a relationship between risk variants in KIBRA and BDNF and connectivity within the pDMN, aDMN, and ECN. The literature examining KIBRA has largely focused on the rs17070145 polymorphism, a common T to C substitution within the ninth intron of the gene. This SNP is associated with episodic memory performance [15] and moderately with AD [12]. BDNF has received attention as a candidate risk gene for AD due to its role in synaptic plasticity and regulation of long-term potentiation [8, 54]. This proposition is supported by the evidence showing that alterations in BDNF levels can affect brain integrity and ultimately cognition. For example, lower expression of BDNF proteins evident in mild cognitive impairment and AD patients is linked to worse cognitive function [55]. A common polymorphism within BDNF (Val66Met), which causes a significant decrease in activity-dependent release of BDNF, is associated with altered hippocampal integrity [16, 57], DMN functional connectivity [58], and worse episodic memory [16, 59]. Driscoll and colleagues [9] recently extended these findings by demonstrating a relationship between other KIBRA (rs6555802 and rs10475878) and BDNF (rs1491850) polymorphisms and probable dementia.
In the present study, risk allele carriers of KIBRA polymorphisms (rs6555802 [A allele]; rs10475878 [A allele]) previously associated with dementia risk [9] show hypoconnectivity within the posterior DMN (i.e., cluster encompassing precuneus) and ECN. Specifically, we observed that heterozygotes (AC) of KIBRA rs655802 had lower connectivity within the precuneus region of the pDMN than major allele homozygotes (CC; tested using additive model). Lower connectivity of the precuneus is largely in line with the existing literature reporting functional brain alterations in AD [17, 60]. The precuneus region serves as a cortical hub, working to integrate and rapidly disseminate functionally specific information [20]. This greatly connected hub region has high metabolic demands [61] and is vulnerable to early accumulation of Aβ pathology [27, 62]. As a result, the precuneus is among one of the earliest regions to exhibit disrupted connectivity in AD [26]. We also report lower frontal connectivity of the ECN in minor risk allele carriers (AA/AG) of KIBRA rs10475878 compared to major allele homozygotes (GG; tested using dominant model). Though we expected to observe higher frontal connectivity in the ECN networks, based on the literature reporting altered patterns of connectivity in AD [24, 31], our results are in accordance with reports of lower frontal connectivity observed in static resting state fMRI in the aging literature [63, 64]. In older adulthood, lower static connectivity of the anterior regions is shown to be attributed to changes in the dynamics of anterior connections. More specifically, dynamic approaches, which take advantage of the time-varying aspect of functional connectivity rather than averaging across the full session, find that frontal regions display increased variability (i.e., flexibility) in their connectivity profiles across the scan period, which when analyzed using an average static approach, appeared as lower global static FC [64]. This shift in anterior brain dynamics (i.e., widespread frontal recruitment) is proposed to be a compensatory mechanism meant to accommodate for alterations in posterior connectivity (e.g., precuneus/PCC) and is in line with evidence from the task-based fMRI literature showing that frontal regions are more widely recruited in older compared to young adults to support successful cognitive performance [65, 66]. Taken together, our results suggest that variants in KIBRA may increase risk for AD through advancing neural age, such that frontal systems encompass a more compensatory signature of activity earlier in the aging trajectory (i.e., middle age) to overcome lower connectivity in critical posterior hubs (i.e., precuneus). Though more work is needed to discern the mechanisms underlying risk of pathological age-related cognitive decline conferred by SNPs in KIBRA, our results provide evidence that even after controlling for APOE4, the strongest genetic risk factor for AD, risk variants in KIBRA genes are associated with disrupted connectivity in functional networks related to episodic memory and executive function that can be observed in middle age.
In addition to disruptions in connectivity related to KIBRA and BDNF variants, we also report altered DMN, ECN, and FPN connectivity in less commonly studied polymorphisms of BDNF (rs11030096) and KIBRA (rs1030182, rs6555791). BDNF rs11030096 was associated with differences in DMN connectivity within the precuneus while KIBRA SNPs, in particular rs6555791, were associated with widespread alterations across all networks assessed. Although more work is necessary to characterize the role of BDNF rs11030096 and KIBRA rs1030182 and rs6555791 in aging and AD, our results suggest that functional connectivity measures are sensitive to genotype-related differences in major resting state networks underlying cognition and implicated in disease, in a healthy, non-demented middle-aged sample.
In the current study, we found no COMT-rela-ted differences in network connectivity in healthy middle-aged adults. COMT has been posited as a candidate gene for dementia considering its role in modulating prefrontal cognition, such as executive function, through degradation of dopamine [67]. The relationship between dopamine and cognition is proposed to exhibit an inverted U-shape, where dopamine activity above and below an optimal threshold negatively impacts prefrontal cognitive performance [68, 69]. Polymorphisms within COMT, including the Val to Met substitution at codon 158 on chromosome 22q11 (Val158Met), affect the level of dopamine activity. The Met allele has diminished catabolism of dopamine compared to Val resulting in prolonged dopamine exposure, benefitting prefrontal performance in Met carriers [70]. Young adulthood and middle age may mark a period in the lifespan where COMT polymorphisms exert minimal influence [70, 71]. Our findings add to this growing literature and suggest that the influence of COMT polymorphisms on intermediate phenotypes of cognition are not evident in middle age (age 40–60), and that more sensitive techniques may be necessary to detect any subtle effects. Compared to young adulthood and middle age, the literature suggests that the magnitude of the Val158Met effect on cognition is more pronounced in older adulthood [69] due to age-related changes to the dopamine system. More work is needed to determine the relationship between COMT, brain integrity, and cognition across the lifespan but particularly in older adulthood, when the dopamine system experiences age-related alterations and individuals are at greater risk for AD.
This study has several important limitations. Our study is limited by sample size. Overall, we had 123 participants with a comprehensive panel of genetic data (Table 1), though the number of complete observations for any given SNP was as low as 90. However, genetic-phenotype studies are typically conducted in larger samples (e.g., hundreds to thousands) to provide sufficient power [9]. It will be important to confirm the endophenotypes observed in the current investigation in larger, independent samples. The cross-sectional design is an additional limitation of the study. It will be critical for this line of research to follow individuals longitudinally to identify AD-converters and correspondingly, which genotype-related alterations in connectivity preclinically may represent pathological aging. Nevertheless, the limitations should not undermine the strengths of the current investigation including the use of a hypothesis-driven neuroimaging genetics approach in middle age to examine intermediate phenotypes of genetic polymorphisms that have been associated with AD. While some have criticized the candidate gene approach for potentially generating false positives because of the risk of an inflated type I error [72], hypothesis-driven candidate gene studies are important to validate GWAS findings and explore the intermediate biological phenotypes between behavior and disease.
In conclusion, we report significant associations between polymorphisms in BDNF and KIBRA with aDMN, pDMN, ECN, and FPN connectivity. This novel investigation documents the relationships between genetic polymorphisms in BDNF and KIBRA with connectivity in networks related to episodic memory and executive function in middle age. Further work using an endophenotype approach is necessary to better understand how genetic variants potentially contribute to AD risk and to identify preclinical biomarkers.
