Association of APOE alleles and polygenic profiles comprising APOE‐TOMM40‐APOC1 variants with Alzheimer's disease neuroimaging markers

Abstract

INTRODUCTION

TOMM40 and APOC1 variants can modulate the APOE‐ε4‐related Alzheimer's disease (AD) risk by up to fourfold. We aim to investigate whether the genetic modulation of ε4‐related AD risk is reflected in brain morphology.

METHODS

We tested whether 27 magnetic resonance imaging‐derived neuroimaging markers of neurodegeneration (volume and thickness in temporo‐limbic regions) are associated with APOE‐TOMM40‐APOC1 polygenic profiles using the National Alzheimer's Coordinating Center Uniform Data Set linked to the AD Genetic Consortium data.

RESULTS

All brain regions studied using structural phenotypes were smaller in individuals with AD. The ε4 allele was associated with smaller limbic (entorhinal, hippocampus, parahippocampus) brain volume and cortical thickness in AD cases than controls. There were significant differences in the associations for the higher‐risk and lower‐risk ε4‐bearing APOE‐TOMM40‐APOC1 profiles with temporo‐limbic region markers.

DISCUSSION

The APOE‐AD heterogeneity may be partly attributed to the modulating role of the TOMM40 and APOC1 genes in the APOE cluster.

Highlights

• The ε4 allele is associated with smaller values of neuroimaging markers in AD cases.

• Larger values of neuroimaging markers may protect against AD in the ε4 carriers.

• TOMM40 and APOC1 variants differentiate AD risk in the ε4 carriers.

• The same variants can differentiate the links between ε4 and neuroimaging markers.

1. BACKGROUND

The extent of the role of apolipoprotein E (APOE) on Alzheimer's disease (AD) risk has been unclear, most likely because of the complex interplay of nearby genes. Many studies, especially large‐scale genome‐wide association studies (GWAS), have emphasized the role of APOE ε2/ε3/ε4 polymorphism in AD risk. ¹ Meanwhile, it has been also evidenced that the APOE alleles can be modulated by other genetic variants. ² , ³ , ⁴ , ⁵ We showed a strong joint association of minor alleles of ε4‐encoding rs429358 of APOE, TOMM40 rs2075650, and APOC1 rs12721046 with odds of AD. ⁶ , ⁷ , ⁸ , ⁹

Neurodegeneration was proposed by the National Institute on Aging (NIA) and Alzheimer's Association Research Framework ¹⁰ , ¹¹ as an AD biomarker. Neuroimaging markers of cortical thinning have characterized disease‐specific morphometry change in AD, ¹² , ¹³ and are present even in early stages of cognitive decline. ¹⁴ It still unknown to what extent the APOE cluster is associated with neurodegeneration. Multivariate GWAS of Alzheimer's Disease Neuroimaging Initiative (ADNI) data have associated APOE, TOMM40, and APOC1 genes with eight AD‐relevant subcortical measures. ¹⁵ GWAS and phenome‐wide association studies (PheWAS) using hippocampal volume, regional, and whole brain volumes have associated TOMM40 rs2075650 with brain imaging markers of AD risk. ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ In several smaller datasets, we have identified clear patterns of atrophy and neuropathological biomarkers indicative of AD risk related to TOMM40 and APOE. ²⁰ , ²¹ , ²² We recently found that the TOMM40 rs2075650 G‐allele was associated with decreased sulcal depth, increased gyrification index, and decreased gray matter volume in cognitively unimpaired individuals. ²³ In addition, whole genome studies of brain‐wide imaging phenotypes (PheWAS) have shown the value of using brain morphometry to tease out complex genetic relationships with AD. ¹⁷ , ¹⁸

However, no studies have looked at the risk of the polygenic variant complex APOE‐TOMM40‐APOC1 with brain imaging risk markers, especially in a large sample like the National Alzheimer's Coordinating Center (NACC) database. There is also evidence from transcriptomic data that TOMM40 may functionally impact temporal cortex thickness, ²⁴ and thus a goal of this study was to more comprehensively assess effects of the APOE‐TOMM40‐APOC1 cluster variants on cortical volume and thickness in a large group of cognitively unimpaired and mild cognitive impairment (MCI)/AD individuals. Particularly, we aimed to determine if the polygenic risk profiles of the APOE‐TOMM40‐APOC1 gene cluster impact AD‐related morphological variation in the brain in regions associated with risk for AD.

RESEARCH IN CONTEXT

1. Systematic review: A literature review identified publications examining roles of the APOE ε4 allele in Alzheimer's disease (AD)‐related brain morphometry. No studies looked at the risk of the polygenic variant complex APOE‐TOMM40‐APOC1 with brain imaging risk markers. These relevant citations are appropriately cited.

2. Interpretation: The results of our analyses show that TOMM40 and APOC1 variants can dissect heterogeneity along the AD‐APOE axis by differentiating the impacts of the APOE ε4 allele on brain imaging phenotypes in temporo‐limbic regions.

3. Future directions: Examining AD‐related phenotypes from a polygenic perspective may provide insight into the role of the TOMM40 and APOC1 genes in early signs of risk for AD. An important direction of research would be to investigate relationships between TOMM40, APOE, and APOC1 impact markers of mitochondrial health and link those markers to aging and AD.

2. METHODS

2.1. Data source and AD phenotype

We conducted a cross‐sectional secondary analysis of the NACC Uniform Data Set (UDS) using data downloaded in October 2023 and spanning the period from 2005. Initiated in 2005, the NACC UDS is a longitudinal dataset of data collected from yearly assessments of study participants at the NIA‐funded Alzheimer's Disease Research Centers (ADRCs) across the country. ²⁵ The UDS and neuroimaging data examined for this study were submitted voluntarily to the NACC from 15 ADRCs. UDS data were collected by trained clinicians and personnel using standardized evaluation and uniform methods for each study subject. The UDS incorporates longitudinal demographics; family and health history; and clinical, neuropsychological, and diagnostic data including medications. ²⁶ For this study we used matched demographic and diagnostic data to a single neuroimaging scan dataset by date.

The source of genetic data and AD status on subjects of European ancestry was seven cohorts (ADC1 to ADC7) from the NIA Alzheimer's Disease Centers (ADCs) linked to NACC, which are a part of the Alzheimer's Disease Genetics Consortium (ADGC) initiative. ²⁷ The overlapping sample between UDS and ADGC/NACC included 676 subjects (Table 1). The ADC cohorts included autopsy‐ and clinically‐confirmed AD‐affected and cognitively normal subjects who were ascertained by the clinical and neuropathology cores of the NIA‐funded ADCs according to the National Institute of Neurological and Communicative Disorders and Stroke and the AD and Related Disorders Association. ²⁸

TABLE 1.

Descriptive statistics for genotyped subjects of European ancestry with information on Alzheimer's disease (AD) and brain imaging phenotypes.

Variable	Abbreviation	Total sample	AD cases	AD controls	p‐value
N		676	227 (33.6%)	449 (66.4%)
Age		74.9 (0.3)	77.2 (0.4)	73.7 (0.4)	4.05E‐08
Women		385 (57.0%)	105 (46.3%)	280 (62.4%)	7.24E‐05
APOE ε2ε2 genotype		0
APOE ε2ε3 genotype		66 (9.8%)	14 (6.2%)	52 (11.6%)	2.69E‐02
APOE ε2ε4 genotype		10 (1.5%)	4 (1.8%)	6 (1.3%)	6.14E‐01
APOE ε3ε3 genotype		335 (49.6%)	78 (34.4%)	257 (57.2%)	3.13E‐08
APOE ε3ε4 genotype		214 (31.7%)	91 (40.1%)	123 (27.4%)	8.49E‐04
APOE ε4ε4 genotype		51 (7.5%)	40 (17.6%)	11 (2.5%)	4.76E‐12
Lower‐risk compound genotype	LR‐CompG	46 (7.7%)	20 (9.6%)	26 (6.6%)	2.63E‐01
Higher‐risk compound genotype	HR‐CompG	219 (36.5%)	111 (53.1%)	108 (27.6%)	1.13E‐09
Total Gray Matter Volume, cm3	GrayVol	585.2 (2.4)	565 (4.1)	595.3 (2.8)	3.19E‐09
Total White Matter Volume, cm3	WhiteVol	441.3 (2.3)	433.2 (3.8)	445.3 (2.9)	1.17E‐02
Total Hippocampus Volume, cm3	HippoVol	6.0 (0.04)	5.3 (0.06)	6.4 (0.04)	2.05E‐41
Left Hippocampus Volume, cm3	LHippo	3.0 (0.02)	2.6 (0.03)	3.2 (0.02)	2.50E‐41
Right Hippocampus Volume, cm3	RHippo	3.1 (0.02)	2.7 (0.03)	3.2 (0.02)	1.13E‐35
Left Entorhinal Volume, cm3	Lent	4.0 (0.03)	3.6 (0.05)	4.2 (0.03)	6.49E‐23
Left Entorhinal Thickness, mm	LEnt_m	3.2 (0.03)	2.7 (0.05)	3.5 (0.03)	2.06E‐35
Left Middle Temporal Gray Matter Volume, cm3	LMidTemp	12.4 (0.08)	11.5 (0.1)	12.9 (0.09)	1.91E‐17
Left Middle Temporal Gray Matter Thickness, mm	LMidTemp_m	2.4 (0.01)	2.3 (0.03)	2.5 (0.02)	1.23E‐14
Left Parahippocampal Gray Matter Volume, cm3	LParHip	3.8 (0.02)	3.6 (0.03)	3.9 (0.02)	3.50E‐16
Left Parahippocampal Gray Matter Thickness, mm	LParHip_m	1.7 (0.01)	1.6 (0.02)	1.8 (0.01)	1.09E‐25
Left Precuneus Gray Matter Volume, cm3	LPrecun	9.5 (0.06)	9.0 (0.09)	9.7 (0.07)	5.57E‐10
Left Precuneus Gray Matter Thickness, mm	LPrecun_m	1.9 (0.01)	1.8 (0.02)	1.9 (0.01)	1.79E‐08
Left Superior Temporal Gray Matter Volume, cm3	LSupTem	15.1 (0.08)	14.3 (0.1)	15.5 (0.1)	5.27E‐12
Left Superior Temporal Gray Matter Thickness, mm	LSupTem_m	2.0 (0.01)	1.9 (0.02)	2.1 (0.01)	1.20E‐20
Right Entorhinal Volume, cm3	Rent	3.8 (0.03)	3.4 (0.05)	4.0 (0.03)	1.21E‐26
Right Entorhinal Thickness, mm	REnt_m	3.4 (0.03)	2.8 (0.05)	3.7 (0.03)	2.79E‐33
Right Middle Temporal Gray Matter Volume, cm3	RMidTemp	12.3 (0.07)	11.5 (0.1)	12.8 (0.08)	2.20E‐15
Right Middle Temporal Gray Matter Thickness, mm	RMidTemp_m	2.4 (0.02)	2.3 (0.03)	2.5 (0.02)	5.57E‐08
Right Parahippocampal Gray Matter Volume, cm3	RParHip	3.9 (0.02)	3.7 (0.03)	4.1 (0.02)	6.65E‐18
Right Parahippocampal Gray Matter Thickness, mm	RParHip_m	1.8 (0.01)	1.6 (0.02)	1.9 (0.01)	1.78E‐24
Right Precuneus Gray Matter Volume, cm3	RPrecun	9.1 (0.06)	8.6 (0.09)	9.4 (0.07)	1.04E‐10
Right Precuneus Gray Matter Thickness, mm	RPrecun_m	1.9 (0.01)	1.8 (0.02)	1.9 (0.01)	3.23E‐13
Right Superior Temporal Gray Matter Volume, cm3	RSupTem	13.5 (0.07)	12.8 (0.1)	13.8 (0.08)	2.19E‐10
Right Superior Temporal Gray Matter Thickness, mm	RSupTem_m	2.1 (0.01)	2.0 (0.03)	2.2 (0.01)	9.15E‐16
AD Signature Volume, cm3	ADSV	0.19 (0.0005)	0.18 (0.0009)	0.19 (0.0004)	1.86E‐21
Left Hippocampal Occupancy Score	LHOS	0.16 (0.003)	0.10 (0.002)	0.19 (0.004)	4.28E‐52
Right Hippocampal Occupancy Score	RHOS	0.18 (0.003)	0.12 (0.003)	0.2 (0.004)	4.80E‐52
Hippocampal Occupancy Score Averaged	HOSA	0.17 (0.003)	0.11 (0.003)	0.2 (0.004)	1.69E‐54

Note: The numbers in parentheses without percentage symbol show standard errors. Significance of the differences (p‐value) in mean values between AD cases and controls was evaluated using chi‐square test and unequal variance t‐test. A suffix “_m” represents thickness measured in millimeters.

2.2. NACC brain imaging data

Brain imaging data were from the NACC UDS dataset, linked to genetic data from ADGC. We used NACC's volumetric summary data for global and regional measures (Table 1). Calculations were performed for NACC by the IDeA Lab (Director: Charles DeCarli, MD; University of California, Davis; http://idealab.ucdavis.edu), following ADNI protocols (http://adni.loni.usc.edu). Of the standard regions provided by NACC, we focused on six regions of the brain (hippocampus, parahippocampus, entorhinal cortex, middle temporal gyrus, superior temporal gyrus, and the precuneus) for which we downloaded left and right volumes and thicknesses, as well as total gray matter and white matter volume. We chose to use specific imaging biomarkers that are sensitive to disease progression and AD genetic risk. ²⁹ , ³⁰ In addition to those regions downloaded, we also computed a hippocampal occupancy score (HOS) as another estimate of medial temporal lobe atrophy, by determining the ratio of hippocampal volume to the sum of the hippocampal and interior lateral ventricular volumes. We included left HOS (LHOS), right HOS (RHOS), and averaged HOS (HOSA) scores in our measurements. HOS has been shown to have a higher discriminative accuracy than the standard hippocampal volume measure. ³¹ We also calculated an AD signature volume (ADSV) composite, which includes the gray matter volumes of the medial temporal lobe regions (hippocampus, parahippocampus gyrus, entorhinal cortex), as well as the middle, inferior, and superior temporal cortices (ADSV composite is summarized in Figure 1; hippocampus not shown). ³² , ³³

FIGURE 1.

Brain imaging regions comprising the AD signature volume (ADSV) composite.

Among the 29 total brain imaging measurements and composites, the left and right hippocampus volumes (LHippo, RHippo) were highly correlated with the total hippocampus volume (HippoVol). However, LHippo and RHippo were less correlated with each other (Figure 2). Similarly, the LHOS and RHOS showed high correlation with the HOSA, but less so with each other (Figure 2). Due to greater variability between the left and right measurements/scores compared to their averages, we retained them for further analyses and excluded the bilateral measurements of HippoVol and HOSA.

FIGURE 2.

Correlation matrix between 29 brain imaging measurements. ADSV, Alzheimer's disease signature volume; GrayVol, total gray matter volume; HippoVol, total hippocampus volume; HOSA, hippocampal occupancy score averaged; LEnt, left entorhinal volume; LEnt_m, left entorhinal thickness; LHippo, left hippocampus volume; LHOS, left hippocampal occupancy score; LMidTemp, left middle temporal gray matter volume; LMidTemp_m, left middle temporal gray matter thickness; LParHip, left parahippocampal gray matter volume; LParHip_m, left parahippocampal gray matter thickness; LPrecun, left precuneus gray matter volume; LPrecun_m, left precuneus gray matter thickness; LSupTem, left superior temporal gray matter volume; LSupTem_m, left superior temporal gray matter thickness; REnt, right entorhinal volume; REnt_m, right entorhinal thickness; RHippo, right hippocampus volume; RHOS, right hippocampal occupancy score; RMidTemp, right middle temporal gray matter volume; RMidTemp_m, right middle temporal gray matter thickness; RParHip, right parahippocampal gray matter volume; RParHip_m, right parahippocampal gray matter thickness; RPrecun, right precuneus gray matter volume; RPrecun_m, right precuneus gray matter thickness; RSupTem, right superior temporal gray matter volume; RSupTem_m, right superior temporal gray matter thickness; WhiteVol, total white matter volume.

2.3. APOE variants

We focused on the APOE ε2 and ε4 alleles encoded by minor alleles of rs7412 (C/t; upper/lower case denotes here major/minor allele) and rs429358 (T/c) single nucleotide polymorphisms (SNPs), respectively. By excluding carriers of the minor allele of rs429538, the ε2 allele in our sample is equivalent to the APOE ε2/ε3 genotype, as there were no carriers of the ε2/ε2 genotype. In the sample without carriers of the minor allele of rs7412, the ε4 allele is defined by carrying the ε3/ε4 or ε4/ε4 genotype.

2.4. Compound genotypes comprising APOE, TOMM40, and APOC1 variants

To better understand the relationships between the APOE ε4 allele and the brain imaging measurements, we also considered combinations of genotypes—herein referred to as compound genotypes (CompGs)—comprising the APOE rs429358, TOMM40 rs2075650 (A/g), and APOC1 rs12721046 (G/a) SNPs, which are in moderate linkage disequilibrium with Pearson correlation coefficients of r²  = 0.56 for rs429358 and rs2075650, r²  = 0.59 for rs429358 and rs12721046, and r²  = 0.65 for rs2075650 and rs12721046. These SNPs were selected because we previously identified that carriers of the ε4 allele and minor alleles of the TOMM40 and APOC1 SNPs were under fourfold higher risk of AD compared to carriers of the ε4 allele and major allele homozygotes of those two SNPs. ⁹ , ³⁴ That is, depending on the carriage of minor alleles of the TOMM40 and APOC1 SNPs, carriers of the ε4 allele may be at substantially lower and higher risk of AD. Accordingly, given the available samples (Table 1), we constructed three aggregated CompGs, excluding carriers of the ε2 allele (Figure 3). One group included the APOE ε3/ε3 genotype, which represented an aggregated CompG comprising variants with no ε4 and ε2 alleles regardless of the presence or absence of the minor alleles of rs2075650 and rs12721046. The second was an aggregated CompG characterized by a lower AD risk. It included one or two copies of the ε4 allele and no minor alleles of rs2075650 and rs12721046 SNPs and was denoted as lower risk (LR)‐CompG. The third was the aggregated CompG characterized by a higher AD risk, denoted as higher risk (HR)‐CompG. It also included one or two copies of the ε4 allele, in addition to at least one minor allele of rs2075650 or rs12721046.

FIGURE 3.

Definition of compound genotypes (CompGs) used in the analyses. Group 1 (neutral) included carriers of the apolipoprotein E (APOE) ε3ε3 genotype and any allele of rs2075650 and rs12721046. Group 2 represents the lower‐risk CompG (LR‐CompG) comprising one or two copies of the ε4 allele and major allele homozygotes of rs2075650 and rs12721046 SNPs. Group 3, representing the higher‐risk CompG (HR‐CompG), also included one or two copies of the ε4 allele, in addition to at least one minor allele of rs2075650 or rs12721046. Upper and lower case (for AA, Ag, etc) denotes here major and minor allele, respectively.

2.5. Analysis

We examined associations of our genotype variable of interest (ie, the APOE ε4 allele and CompGs) with the brain imaging phenotypes using their natural scales (Table 1). The models were adjusted for the first five principal components, sex of the subject (male as reference), and age at time of imaging (standardized). The models run were ordinary least squares regression, using the lm() function in R. Results reported are the beta estimates and associated standard errors and p‐values from the regression model. Additional models were run with adjustments for whoever had AD status, in AD cases and controls separately, and to examine interactions between the genotype of interest and AD. We used p < 0.05 as the threshold for significance given the robust evidence linking APOE, TOMM40, and APOC1 variants to AD risk, and scientifically justified measures of the brain morphometry to look at AD‐related changes.

The analyses were conducted contrasting the effects for the genotype of interest to the reference genotype. We used a common reference represented by the APOE ε3/ε3 genotype. In addition, the models were run contrasting the effects between the LR‐CompG and HR‐CompG (Figure 3), with the LR‐CompG as a reference.

3. RESULTS

3.1. Descriptive statistics

Table 1 shows that one‐third of the total sample (n = 676) consisted of AD cases, who were older than controls. There were more AD‐affected men than women, whereas the opposite relationship held for controls. Additionally, Table 1 has p‐values of chi‐square and t‐tests to show the difference in proportions and means between the AD and non‐AD groups. As expected, carriers of the ε2/ε3 genotype have lower risk of AD, as evidenced by their larger proportion in controls than in AD cases (11.6% vs 6.2% respectively, p < 0.05), and the opposite relationship is observed for carriers of the ε3/ε4 and ε4/ε4 genotypes (p < 0.05 for both). In addition, as also expected, carriers of the LR‐CompG had lower absolute risk of AD (0.43 = 20(N_AD)/46(N_total)) than carriers of the HR‐CompG (0.56 = 111/219). ⁹ , ³⁴ All brain regions studied were significantly larger in controls than in AD cases.

3.2. Associations of the ε4 allele with brain measurements

The analysis with no adjustment for AD status identified that 20 of 27 brain measurements were significantly smaller (volume, thickness, or composite) in carriers of the APOE ε4 allele (Figure 4, APOE ε4; Table 2). No significant associations were observed for seven measurements, including global volumes for gray and white matter (GrayVol, WhiteVol), bilateral (left and right) measures of middle temporal gray matter thickness (LMidTemp_m, RMidTemp_m) and superior temporal gray matter volume (LSupTem, RSupTem), and right superior temporal gray matter thickness (RSupTem_m), although it attained marginal significance.

FIGURE 4.

Patterns of the associations of genetic variants with 27 brain imaging measurements. Genetic variants are represented by the apolipoprotein E (APOE) ε4 allele, the lower‐risk compound genotype (LR‐CompG), and the higher‐risk compound genotype (HR‐CompG). For APOE ε4, LR‐CompG, and HR‐CompG, the reference was the APOE ε3ε3 genotype. For HR versus LR (HR vs. LR), the reference was the LR‐CompG. The x‐axis shows abbreviations for the 27 brain imaging measurements. Thickness indicated by the suffix “_m” is measured in millimeters. Color indicates associations with β < 0 (red) and β > 0 (green). Color shades indicate the significance provided in the figure legend. Alzheimer's disease (AD) unadj, AD adj, Interactions, AD cases, and Controls denote the results from models for AD unadjusted, adjusted, interaction, and AD stratification analyses, respectively. Detailed numerical estimates are provided in Tables S1‐S6. ADSV, Alzheimer's disease signature volume; GrayVol, total gray matter volume; LEnt, left entorhinal volume; LEnt_m, left entorhinal thickness; LHippo, left hippocampus volume; LHOS, left hippocampal occupancy score; LMidTemp, left middle temporal gray matter volume; LMidTemp_m, left middle temporal gray matter thickness; LParHip, left parahippocampal gray matter volume; LParHip_m, left parahippocampal gray matter thickness; LPrecun, left precuneus gray matter volume; LPrecun_m, left precuneus gray matter thickness; LSupTem, left superior temporal gray matter volume; LSupTem_m, left superior temporal gray matter thickness; REnt, right entorhinal volume; REnt_m, right entorhinal thickness; RHippo, right hippocampus volume; RHOS, right hippocampal occupancy score; RMidTemp, right middle temporal gray matter volume; RMidTemp_m, right middle temporal gray matter thickness; RParHip, right parahippocampal gray matter volume; RParHip_m, right parahippocampal gray matter thickness; RPrecun, right precuneus gray matter volume; RPrecun_m, right precuneus gray matter thickness; RSupTem, right superior temporal gray matter volume; RSupTem_m, right superior temporal gray matter thickness; WhiteVol, total white matter volume.

TABLE 2.

Associations of the APOE ε4 allele with 27 brain imaging measurements in AD unadjusted and adjusted models.

Outcome	AD unadjusted			AD adjusted
	β	SE	p‐value	Β	SE	p‐value
GrayVol	−4.027	4.293	3.49E‐1	5.959	4.242	1.61E‐1
WhiteVol	−0.898	3.750	8.11E‐1	2.828	3.877	4.66E‐1
LHippo	−0.232	0.038	1.29E‐9	−0.092	0.034	6.32E‐3
RHippo	−0.198	0.038	2.08E‐7	−0.067	0.034	5.34E‐2
LEnt	−0.264	0.051	4.17E‐7	−0.108	0.049	2.68E‐2
LEnt_m	−0.324	0.058	2.81E‐8	−0.129	0.053	1.57E‐2
LMidTemp	−0.416	0.147	4.89E‐3	0.013	0.141	9.29E‐1
LMidTemp_m	−0.037	0.032	2.59E‐1	0.039	0.032	2.23E‐1
LParHip	−0.153	0.040	1.42E‐4	−0.062	0.039	1.18E‐1
LParHip_m	−0.111	0.024	6.62E‐6	−0.048	0.024	4.26E‐2
LPrecun	−0.249	0.114	2.96E‐2	−0.035	0.115	7.61E‐1
LPrecun_m	−0.068	0.023	2.53E‐3	−0.033	0.023	1.52E‐1
LSupTem	−0.257	0.163	1.16E‐1	0.139	0.161	3.88E‐1
LSupTem_m	−0.070	0.025	4.37E‐3	−0.005	0.024	8.45E‐1
REnt	−0.202	0.053	1.39E‐4	−0.027	0.049	5.75E‐1
REnt_m	−0.317	0.061	2.79E‐7	−0.114	0.057	4.34E‐2
RMidTemp	−0.306	0.141	3.03E‐2	0.068	0.137	6.22E‐1
RMidTemp_m	−0.025	0.035	4.62E‐1	0.034	0.035	3.32E‐1
RParHip	−0.151	0.040	1.49E‐4	−0.056	0.039	1.49E‐1
RParHip_m	−0.107	0.025	2.90E‐5	−0.043	0.025	8.10E‐2
RPrecun	−0.226	0.110	4.03E‐2	−0.003	0.110	9.76E‐1
RPrecun_m	−0.045	0.022	4.16E‐2	0.002	0.022	9.32E‐1
RSupTem	−0.186	0.141	1.89E‐1	0.151	0.139	2.78E‐1
RSupTem_m	−0.056	0.029	5.75E‐2	0.013	0.029	6.44E‐1
ADSV	−0.005	0.001	6.25E‐6	−0.002	0.001	6.92E‐2
LHOS	−0.017	0.006	4.51E‐3	0.001	0.006	8.26E‐1
RHOS	−0.016	0.006	1.04E‐2	0.003	0.006	6.37E‐1

Note: Total sample was N = 600, which included 209 AD cases; the number of subjects carrying at least one ε4 allele was N = 265. The results for the interaction and AD stratified analyses are provided in Table S1.

Abbreviations: AD, Alzheimer's disease; ADSV, AD signature volume; APOE, apolipoprotein E; GrayVol, total gray matter volume; LEnt, left entorhinal volume; LEnt_m, left entorhinal thickness; LHippo, left hippocampus volume; LHOS, left hippocampal occupancy score; LMidTemp, left middle temporal gray matter volume; LMidTemp_m, left middle temporal gray matter thickness; LParHip, left parahippocampal gray matter volume; LParHip_m, left parahippocampal gray matter thickness; LPrecun, left precuneus gray matter volume; LPrecun_m, left precuneus gray matter thickness; LSupTem, left superior temporal gray matter volume; LSupTem_m, left superior temporal gray matter thickness; REnt, right entorhinal volume; REnt_m, right entorhinal thickness; RHippo, right hippocampus volume; RHOS, right hippocampal occupancy score; RMidTemp, right middle temporal gray matter volume; RMidTemp_m, right middle temporal gray matter thickness; RParHip, right parahippocampal gray matter volume; RParHip_m, right parahippocampal gray matter thickness; RPrecun, right precuneus gray matter volume; RPrecun_m, right precuneus gray matter thickness; RSupTem, right superior temporal gray matter volume; RSupTem_m, right superior temporal gray matter thickness; WhiteVol, total white matter volume.

After adjustment for AD, only five of 20 significant measurements in the AD‐unadjusted model remained significant (left hippocampus volume [LHippo], left entorhinal volume [LEnt], left entorhinal thickness [LEnt_m], left parahippocampal gray matter thickness [LParHip_m], right entorhinal thickness [REnt_m]). None of the remaining seven measurements attained significance.

The interaction analysis showed that the ε4 allele is associated with smaller limbic brain volume and cortical thickness in AD cases than controls (characterized by six measurements; Figure 4, model for interactions, APOE ε4; Table S1). The same direction of the effects for all six measurements was observed in the AD‐unadjusted model, but only the LEnt_m was significantly associated with the ε4 allele in the AD‐adjusted model, β = −0.129, p = 1.57 × 10⁻².

The AD‐stratified analysis identified six measurements significantly associated with the ε4 allele in AD cases. These were the LEnt_m, REnt_m, LEnt, right entorhinal cortex volume (REnt), LHippo, and left precuneus gray matter thickness (LPrecun_m). Three of them, LEnt_m, LPrecun_m, and REnt underscored significant ε4 × AD interaction effects. The most significant association in AD cases was observed for LEnt_m, β = −0.387, p = 3.40 × 10⁻⁴. The only significant association in controls was with the RMidTemp_m, β = 0.082, p = 3.97 × 10⁻², which was not significant in the AD‐unadjusted models.

3.3. Associations of the ε2 allele with brain measurements

In contrast to the ε4 allele, the ε2 allele was not significantly associated with brain measurements either in the AD‐unadjusted or AD‐adjusted model (Table S2). However, we identified that AD significantly modulated the association of the ε2 allele with WhiteVol with the effect for interaction β_int = 39.912, p = 6.38 × 10⁻³. The lack of significance for the association with WhiteVol in AD‐unadjusted and AD‐adjusted models is due to antagonistic associations in AD cases (β = 30.079, p = 2.64 × 10⁻²) and controls (β = −13.192, p = 5.34 × 10⁻²). In addition, the ε2 allele was associated with ADSV in controls, β = −0.003, p = 4.52 × 10⁻².

3.4. Associations of LR‐CompG and HR‐CompG with brain measurements

3.4.1. AD‐unadjusted model

The analysis of the LR‐CompG and HR‐CompG individually showed that the HR‐CompG was associated with smaller values of brain measurements for 19 of 20 ε4‐associated measurements, whereas the LR‐CompG was associated with smaller values for 11 of them (Figure 4; Table S3). Contrasting the associations for the HR‐CompG to the associations for the LR‐CompG revealed no significant differences between them.

3.4.2. AD‐adjusted model

The HR‐CompG was significantly associated only with LHippo (Figure 4; Table S4), which was among the five measurements adversely associated with the ε4 allele in the AD‐adjusted model (LHippo, LEnt, LEnt_m, LParHip_m, REnt_m). In contrast, the LR‐CompG was associated with smaller values of five measurements (LHippo, RHippo, LEnt_m, REnt_m, right parahippocampal gray matter thickness [RParHip_m]). However, two of them (RHippo, RParHip_m) attained only suggestive‐effect significance (p < 0.1) in the associations with the ε4 allele.

Contrasting the associations of the HR‐CompG and LR‐CompG with brain measurements, we identified significant differences for left superior temporal gray matter thickness (LSupTem_m), REnt_m, and right middle temporal gray matter volume (RMidTemp) (Figure 5). For the LSupTem_m and RMidTemp, the differences were underscored by the opposite directions of the effects in the associations with HR‐CompG (positive direction) and LR‐CompG (negative direction). The REnt_m was associated with the LR‐CompG (β = −0.304, p = 6.25 × 10⁻³), but not the HR‐CompG (β = −0.068, p = 0.253). These associations indicate that the lower‐risk group (LR‐CompG) had smaller values of temporal and limbic measurements than the higher‐risk group (HR‐CompG).

FIGURE 5.

Figural representation of the focused brain regions. The regions shown have significantly smaller values of the brain imaging measurements in carriers of the lower‐risk compound genotype (LR‐CompG) than in carriers of the higher‐risk compound genotype (HR‐CompG) in the AD adjusted model. They include the left superior temporal cortex thickness (LSupTem_m; red), right middle temporal gray matter volume (RMidTemp; yellow) and right entorhinal thickness (REnt_m; yellow, upper medial view). Detailed numerical estimates from this analysis represented here are provided in Table S4.

3.4.3. The analysis of interactions of CompGs with AD

Significant interactions were observed only between the HR‐CompG and AD for the same six measurements as in the analysis of the ε4 allele (ie, LEnt_m, left middle temporal gray matter volume [LMidTemp], LPrecun_m, REnt, RPrecun_m, LHOS) (Figure 4; Table S5). These interactions were characterized by negative directions of the interactive effects.

3.4.4. Associations with brain measurements in AD cases

Further insights into the role of AD in the associations of CompGs with brain measurements can be gained from the AD‐stratified analysis. This analysis showed that all six measurements significantly associated with the ε4 allele in AD cases (LHippo, LEnt, LEnt_m, LPrecun_m, REnt, REnt_m) had significantly smaller values of brain measurements in HR‐CompG carriers (Figure 4; Table 3). For the LR‐CompG carriers, the only significant association underscored smaller LHippo. No significant differences in the associations of the HR‐CompG and LR‐CompG with brain measurements were identified (Figure 4; Table S6).

TABLE 3.

Significant associations of compound genotypes (CompGs) with eight brain imaging measurements in AD cases and controls.

Outcome	Beta	SE	p‐value
Associations for HR‐CompG in AD cases, N = 111
LHippo	−0.154	0.072	3.51E‐2
LEnt	−0.202	0.101	4.68E‐2
LEnt_m	−0.386	0.114	8.35E‐4
LPrecun_m	−0.117	0.049	1.73E‐2
REnt	−0.206	0.100	4.04E‐2
REnt_m	−0.252	0.119	3.50E‐2
Associations for LR‐CompG in AD cases, N = 20
LHippo	−0.273	0.118	2.32E‐2
Associations for HR‐CompG in controls, N = 108
RMidTemp_m	0.083	0.041	4.55E‐2
Associations for LR‐CompG in controls, N = 26
LEnt	−0.221	0.105	3.64E‐2
REnt_m	−0.277	0.125	2.76E‐2
ADSV	−4.01E‐3	1.94E‐3	4.05E‐2
HR‐CompG versus LR‐CompG in controls
LEnt_m	0.252	0.126	4.76E‐2
REnt_m	0.281	0.136	4.06E‐2

Note: LR‐CompG denotes a compound genotype with a lower risk of AD; HR‐CompG denotes a compound genotype with a higher risk of AD. Models for LR‐CompG and HR‐CompG were contrasted by the APOE ε3ε3 genotype. HR‐CompG versus LR‐CompG denotes a model contrasting the effect of the HR‐CompG by the LR‐CompG. The sample included 71 AD‐affected and 220 unaffected subjects. More details are provided in Table S6.

Abbreviations: AD, Alzheimer's disease; ADSV, AD signature volume; LEnt, left entorhinal volume; LEnt_m, left entorhinal thickness; LHippo, left hippocampus volume; LPrecun_m, left precuneus gray matter thickness; REnt, right entorhinal volume; REnt_m, right entorhinal thickness; RMidTemp_m, right middle temporal gray matter thickness.

3.4.5. Associations with brain measurements in controls

Counterintuitively, in controls, LR‐CompG was significantly associated with smaller values of brain measurements for three brain regions, LEnt, REnt_m, and the ADSV, while HR‐CompG was associated with larger RMidTemp_m (Figure 4; Table 3). Moreover, contrasting the associations for the HR‐CompG and LR‐CompG, we identified significantly smaller LEnt_m and REnt_m in carriers of the LR‐CompG than in carriers of the HR‐CompG (Table 3).

4. DISCUSSION

The APOE ε4 allele is the strongest single genetic factor linked to AD risk in populations of multiple ancestries, ³⁵ and has been extensively studied for three decades. ³⁶ , ³⁷ Intense interest in APOE is driven by its large population‐level impact. Indeed, the APOE ε4 allele is observed in about 50% of AD patients. ³⁸ Lifetime risks of AD estimated at 85 years in the ε4 unselected population of European ancestry are about 11% in men and 14% in women. These risks double for carriers of one ε4 allele (23% for men and 30% for women) and increase fivefold in carriers of its two copies (51% for men and 60% for women). ³⁹ However, despite these impacts, the APOE ε4 allele is still not a causal variant. This means that even people homozygous for the ε4 allele can survive to very old ages—and even become centenarians—and not develop AD during their lifetimes. ⁴⁰ The huge population‐level impact and the lack of apparent causal links between APOE and AD suggest a key role of heterogeneity in the APOE‐related mechanisms of AD pathogenesis.

We analyzed whether variations in the polygenic variant complex APOE‐TOMM40‐APOC1 were associated with brain imaging risk markers in the NACC dataset. The results of our analyses using AD‐related structural brain imaging phenotypes argue that the APOE‐AD heterogeneity can be partly attributed to the modulating role of the TOMM40 and APOC1 genes in the APOE gene cluster. Specifically, our analysis of 27 AD‐relevant brain measurements in AD‐affected and unaffected subjects facilitate two main conclusions.

First, as expected we found that individuals with an ε4 allele had smaller values of 20 brain measurements compared to those with the ε3/ε3 genotype, and this was in an AD‐unadjusted model (Figure 4). AD stratification analysis showed that these associations were more pronounced in AD‐affected subjects than in controls, attaining significance for six of them, including bilateral entorhinal cortex volume and thickness, LHippo, and left precuneus. Interestingly, none of the 20 neuroimaging measurements were significantly associated with the ε4 allele in AD‐unaffected subjects. This AD‐specific result fits our previous analysis of the TOMM40‐APOE haplogroup in a different sample in which we identified that TOMM40 rs10524523 (“523”), specifically the poly‐T short allele, was associated with brain volume, whereas APOE ε4 carriers had a significantly higher rate of AD‐related brain morphological markers in AD subjects but not cognitively healthy individuals. ²⁰ Moreover, several studies have shown that APOE ε4 may not impact limbic volume in cognitively unimpaired individuals. ⁴¹ , ⁴² This may suggest that having larger volume or thickness of AD‐relevant brain regions represents a compensatory mechanism prior to possible disease onset in carriers of the ε4 allele.

Second, stratifying carriers of the ε4 allele into the lower‐ and higher‐risk groups (LR‐CompG and HR‐CompG) dissects further heterogeneity underlying the association between the ε4 allele and AD‐relevant brain measurements. Specifically, the AD‐adjusted analysis highlighted a seemingly counterintuitive result that carriers of LR‐CompG tend to have thinner left superior temporal gyrus and right entorhinal cortex, and lower right middle temporal volume (Figures 4 and 5) than carriers of HR‐CompG, although carriers of LR‐CompG have lower AD risk than carriers of HR‐CompG. ⁹ Furthermore, AD‐stratified analysis showed that AD‐unaffected carriers of LR‐CompG have significantly thinner left and right entorhinal cortex compared to AD‐unaffected carriers of HR‐CompG. These results indicate that carriers of LR‐CompG are more tolerant to neurodegeneration of these brain regions compared to carriers of HR‐CompG, implying that carriers of LR‐CompG may not develop AD or develop it at older ages than carriers of HR‐CompG. Our results perhaps argue that having the major rs2075650 TOMM40 and rs12721046 APOC1 homozygotes allows people carrying the ε4 allele to stay in the control group despite having smaller brain volumes compared to those characteristics for carriers of HR‐CompG. This pattern is consistent with the hypothesis that TOMM40/APOC1 SNPs are contributing in a complex way to resilience rather than resistance in some individuals. Because LR‐CompG and HR‐CompG are differentiated by TOMM40 and APOC1 variants, our results indicate plausible genes modulating AD resilience during neurodegeneration in carriers of the ε4 allele.

While there have been no comparable studies investigating these higher and lower risk genetic compound genotypes using brain imaging, there have been quite a few studies investigating the relationship between individual genetic influences of TOMM40 rs2075650 and APOE ε4 on measures of brain health. Several large PheWAS studies done in the ADNI sample identified rs2075650 and APOE as key variants associated with hippocampal volume reductions as well as the amygdala. ¹⁷ , ¹⁸ However, they did not specify whether it was the high or low risk variant in those genes that was associated with volumetric reductions. Of note, in those gene‐phenotype analyses APOC1 was not identified, and in general APOC1 has been studied the least, perhaps because it has not shown a large signal for brain structure specifically. There has been one study from the ADNI cohort looking at the TOMM40‐APOC1 region which identified a collective impact of rs4420638, rs56131196, and rs157582 on hippocampal atrophy rate; however, this did not include the APOC1 SNP studied here. ⁴³ Interestingly, rs12721046 in APOC1 has been identified as a shared genetic factor between type 2 diabetes and AD, which may speak to its role in lipid metabolism while still leaving much to learn about the impact on brain structure. ⁴⁴

There is a potential relationship between TOMM40 and mitochondrial dysfunction associated with AD, due to its encoding of the TOM40 protein. TOM is key for mitochondrial energy metabolism, cellular processes, and homeostasis. Moreover, mitochondrial dysfunction elicited from genetic and environmental influences appears to be a key to understanding primary and secondary components to the cascades that lead to AD. ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ It will be important to investigate whether complex relationships between TOMM40, APOE, and APOC1 impact markers of mitochondrial health and link those markers to aging and AD.

There are several limitations to our analysis. First, as this was the first approach to looking into these higher and lower risk compound genotypes and brain imaging, we chose to use NACC volumetric data from key temporo‐limbic regions commonly associated with AD risk and progression. An inherent advantage to this is its generalizability to the small but meaningful literature on TOMM40, APOE, and brain health. However, future analyses should delve into fluorodeoxyglucose‐positron emission tomography (FDG‐PET), amyloid‐PET and other neuroimaging biomarkers of AD risk to gain a more comprehensive picture of the phenotypes of AD and their relationship to these compound genotypes. Moreover, characterizing atrophy and morphometric change with multiple time points and scans longitudinally will be key for understanding these risk and protective mechanisms. Beyond that, we acknowledge somewhat less than optimal power of the analyses of the compound genotype groups in the AD‐stratified samples (66% in AD cases for the lower‐risk group). Larger samples will be key to replicating data reported here.

CONFLICT OF INTEREST STATEMENT

The authors declare no competing interests. Author disclosures are available in the supporting information.

CONSENT STATEMENT

Consent statement was not necessary.

Supporting information

Supporting Information

Supporting Information

Kulminski AM, Jain‐Washburn E, Nazarian A, et al. Association of APOE alleles and polygenic profiles comprising APOE‐TOMM40‐APOC1 variants with Alzheimer's disease neuroimaging markers. Alzheimer's Dement. 2025;21:e14445. 10.1002/alz.14445