A haptoglobin (HP) structural variant alters the effect of APOE alleles on Alzheimer’s disease

Abstract

Background:

Haptoglobin (HP) is an antioxidant of apolipoprotein E (APOE), and previous reports have shownHP binds with APOE and amyloid beta (Aβ) to aid its clearance. A common structural variant of the HP gene distinguishes it into two alleles: HP1 and HP2.

Methods:

HP genotypes were imputed in 29 cohorts from the Alzheimer’s Disease Genetics Consortium (N = 20,512). Associations between the HP polymorphism and Alzheimer’s disease (AD) risk and age of onset through APOE interactions were investigated using regression models.

Results:

The HP polymorphism significantly impacts AD risk in European-descent individuals (and in meta-analysis with African-descent individuals) by modifying both theprotective effect of APOE ε2 and the detrimental effect of APOE ε4. The effect is particularly significant among APOEε4 carriers.

Discussion:

The effect modification of APOE by HP suggests adjustment and/or stratification by HP genotype is warranted when APOE risk is considered. Our findings also provided directions for further investigations on potential mechanisms behind this association.

1 |. INTRODUCTION

Alleles of the apolipoprotein E (APOE) gene are the strongest genetic factor for sporadic Alzheimer’s disease (AD) and are classified as APOE ε2, APOE ε3, and APOE ε4. The protein products of these three alleles differ from each other by single amino-acid substitutions at positions 112 and 158.¹ This leads to conformational changes in two alpha helices and causes the APOE ε4 protein to be more compact, less protected, and less stable compared to APOE ε2 and APOE ε3.¹ Furthermore, these changes lead to functional differences in lipid binding.¹ The APOE ε4 allele increases the risk of AD relative to APOE ε3, while the APOE ε2 allele decreases the risk.^2,3 APOE ε4 is also associated with earlier age of AD onset by approximate 2.5 years.² The amyloid cascade hypothesis and the tau hypothesis are the two most commonly accepted hypotheses of AD pathology.^4,5 APOE is potentially involved in both hypotheses. In addition to altering lipid binding, APOE potentially plays an important role in amyloid beta (Aβ) protein deposition.^4,6 Human studies show that the APOE ε4 allele dosage is associated with increased Aβ plaques in AD patients,⁷ and APOE ε4 carriers who are middle-aged or elderly are more likely to have brain amyloid while APOE ε2 carriers rarely develop fibrillar Aβ.^8–10 In vitro experiments also show that the APOE protein binds to Aβ with high avidity,¹¹ and mouse model experiments suggest that APOE regulates Aβ metabolism, aggregation, and deposition.^12,13 APOE also clears soluble Aβ in mice, with APOE ε4 less efficient than APOE ε2 or APOE ε3.¹⁴ In addition, APOE affects tau neuropathological changes in AD brains.¹⁵ Abnormal tau phosphorylation was found in Apoe ε4 mice brains.¹⁶

The lipid clearance function of APOE is also influenced by the oxidative state of the protein, with oxidized APOE showing lower lipid-binding affinity.¹⁷ Haptoglobin (HP) is a hemoglobin scavenger that keeps free hemoglobin from causing oxidative damage to tissues, and HP is also a potential antioxidant of APOE. In vitro experiments have shown that the HP protein physically binds to APOE and this binding potentially protects the APOE protein against oxidation and preserves its lipid transport activity.^18,19

A common structural variant (SV) of the HP gene spans two tandem exons and distinguishes two HP alleles: HP1 (one copy of exons 3 and 4) and HP2 (two copies of exons 3 and 4).²⁰ This variant is not captured via genotyping arrays that are typically used in genome-wide association studies (GWAS) due to the complexity of the surrounding linkage disequilibrium and haplotype structures, so the effects of this variant have not been adequately explored in prior large-scale GWAS studies. This SV alters the α-subunit of HP, changing the quaternary structure of the final HP protein complex.²¹ HP1 proteins form dimers, HP2 forms multimers with cyclic conformations, and the HP1 and HP2 together form multimers with linear and cyclic conformations.^22,23 Though the hemoglobin binding affinity of the HP protein complex is not strongly influenced, due to their larger sizes, multimers demonstrate reduced binding capability, thereby leading to lower functional activity compared to dimers.²² This SV has been previously associated with plasma lipid levels, and (in our prior work) with neurocognitive deficits in people living with human immunodeficiency virus (HIV).^20,24,25 A small study also reported that the interaction of APOE ε4 and HP genotypes associated with longevity in a population in central Italy.²⁶

Given the antioxidant role of HP, previously reported associations to cognitive phenotypes, and its physical interactions with APOE, here we investigate the genetic interaction between HP and APOE polymorphisms and the effect on AD risk and age of onset in the largest collection of AD samples in the United States.

2 |. METHODS

2.1 |. Study cohorts and phenotype

The study data includes European-descent participants from 29 Alzheimer’s Disease Genetic Consortium (ADGC) cohorts with available genotyping data (Text S1 in supporting information). The sample size and other descriptive statistics are provided in Table 1. The detailed description of each cohort along with their diagnosis of AD can be found in Kunkle et al.²⁷ A detailed description of the ascertainment of age at AD onset along with the descriptive statistics of each cohort can be found in Naj et al.²⁸ All subjects were recruited under protocols approved by the appropriate institutional review boards. Written informed consent was obtained from study participants or, for those with substantial cognitive impairment, from a caregiver, legal guardian, or other proxy.

TABLE 1.

Study population

Overall
N			20512
Categorical variables [N (%)]
HP genotype
HP1/HP1			2969 (14.5)
HP1/HP2			9506 (46.3)
HP2/HP2			8037 (39.2)
APOE genotype
ε3/ε3			9995 (48.7)
ε2/ε2			84 (0.4)
ε2/ε3			1735 (8.5)
ε2/ε4			511 (2.5)
ε3/ε4			6713 (32.7)
ε4/ε4			1474 (7.2)
Sex
Female			12044 (58.7)
Male			8468 (41.3)
AD
Case			9824 (47.9)
Control			10688 (52.1)
Numerical variables [mean (SD)]
Age			75.79 (8.02)
Cross tabulation of HP and APOE genotypes
	HP1/HP1 (%)	HP1/HP2 (%)	HP2/HP2 (%)
ε3/ε3	1410 (0.069)	4630 (0.226)	3955 (0.193)
ε2/ε2	21 (0.001)	29 (0.001)	34 (0.002)
ε2/ε3	249 (0.012)	777 (0.038)	709 (0.035)
ε2/ε4	973 (0.047)	3143 (0.153)	2597 (0.127)
ε4/ε4	240 (0.012)	667 (0.033)	567 (0.028)
ε4/ε4	240 (0.012)	667 (0.033)	567 (0.028)
Cohort breakdown
Cohort	N (%)		HP hard-call rate (%)
ACT2	28 (0.1)		96.87
ADC1	1951 (9.5)		96.72
ADC2	667 (3.3)		96.98
ADC3	1056 (5.1)		97.28
ADC4	621 (3.0)		97.06
ADC5	735 (3.6)		97.3
ADC6	512 (2.5)		97.15
ADC7	1256 (6.1)		97.68
ADNI	421 (2.1)		96.68
BIOCARD	114 (0.6)		98.02
CHAP2	164 (0.8)		97.44
EAS	146 (0.7)		97.54
GSK	1117 (5.4)		82.19
NIA-LOAD	1798 (8.8)		96.85
MAYO	1605 (7.8)		97.85
MIRAGE	663 (3.2)		98.07
NBB	78 (0.4)		97
OHSU	264 (1.3)		96.71
RMAYO	212 (1.0)		97.45
ROSMAP1	890 (4.3)		96.97
ROSMAP2	116 (0.6)		95.87
TARCC	442 (2.2)		91.68
TGEN2	709 (3.5)		72.93
UKS	577 (2.8)		97.31
UPITT	1288 (6.3)		66.74
WASHU	480 (2.3)		97.61
WASHU2	102 (0.5)		97.45
WHICAP	594 (2.9)		96.91

Notes: The “N”s are the number of individuals that are included in the analyses after omitting for missing in all variables used. The “Age” variable is equal to age at onset for AD cases and age at last visit for controls. The “HP hard-call rate” is calculated as (number of HP hard-calls)/(number of total genotyped individuals) for each cohort.

Abbreviations: AD, Alzheimer’s disease; APOE, apolipoprotein E; HP, haptoglobin; SD, standard deviation.

2.2 |. Genotyping and quality control

Genotyping was performed on either Illumina or Affymetrix high-density single nucleotide polymorphism (SNP) microarrays. For Illumina chip data, a minimal call rate of 0.95 and a minor allele frequency (MAF) of 0.02 were used for filtering and for Affymetrix 0.98 and 0.01 were used, respectively.²⁹

Ancestry-based principal components (PCs) were computed from a combined dataset using the set of SNPs genotyped in all study cohorts.³⁰ After filtering SNPs with pairwise linkage disequilibrium (r2) < 0.20, 31,310 SNPs were evaluated using EIGENSTRAT.³⁰ The top three PCs from EIGENSTRAT were used as covariates in the joint analysis.³⁰

The APOE genotype was determined for different cohorts in multiple ways including using SNPs rs7412 and rs429358, Roche Diagnostics LightCycler 480 (Roche Diagnostics) instrument LightMix Kit ApoE C112R R158 (TIBMOLBIOL), pyrosequencing or restriction fragment length polymorphism analysis, and high-throughput sequencing of codons 112 and 158 in APOE by Agencourt Bioscience Corporation.³⁰

2.3 |. Imputation and relatedness filtering

After quality control, SNP data for chromosome 16 was extracted from available genotype data, submitted to the TOPMed imputation server,^31–33 and imputed to the TOPMed (version TOPMed-r2) genotype marker set for each cohort. This approach was validated using gene-tissue expression (GTEx) RNA expression data and genotyping platforms were found to not affect HP hard-calls (Text S2, Tables S1 and S2, Figure S1 in supporting information). To impute the HP SV, we used a customized version of a published imputation reference that was developed using droplet polymerase chain reaction and validated using RNA-sequencing,^20,25 [Supplement File – Imputation Reference]. After TOPMed imputation, we first extracted SNPs that are included in the HP imputation reference. We then filtered the SNPs for TOPMed imputation R² value and kept only markers that had an R² ≥ 0.8. We performed HP imputation with the customized reference panel using IMPUTEv2 software for each cohort.³⁴ Finally, we conducted hard-calling for the imputed HP allele dosages with a threshold of 0.9. Only imputation dosages ≥0.9 were kept for further analyses and dosages < 0.9 were removed. This approach has been previously validated.²⁵

Sample relatedness was checked by calculation of the genetic relationship matrix (GRM) using GCTA software.³⁵ After filtering for related individuals, 20,512 unrelated individuals remained for statistical analysis.

2.4 |. Association analysis

Association analyses were conducted jointly using data from all cohorts together. We sought to model the effect of APOE alleles comprehensively, so we combined both the detrimental effect of APOE ε4 and the protective effect of APOE ε2. To accomplish this, we assumed equal change in the AD risk between APOE ε2 to APOE ε3, and APOE ε3 to APOE ε4 alleles in our statistical analyses and encoded the ε2, ε3, and ε4 alleles into one variable in ascending order (denoted as “APOE ε2-3-4”). The value of this variable should increase with AD risk additively from a baseline risk of ε2. As each individual carries two APOE alleles, we modeled each APOE allele from the pair of alleles separately (referred below as “APOE1 ε2-3-4” and “APOE2 ε2-3-4”). Logistic regression models with AD case/control status as an outcome were fit for all individuals with an interaction term of HP2 allele count and APOE allele effects.

All regression analyses were adjusted for sex, age, and the top three ancestry-based PCs. Age is defined as the age at onset for AD cases and age at last visit for controls. All analyses were conducted using R (Metafor,³⁶ Survival³⁷). The statistical models used in analyses are as below:

We fit a logistic regression model using the full data with an APOE ε2-3-4 effect for each of the two APOE alleles:

$$\begin{array}{l}Outcome~HP2+APOE1\epsilon \mathrm{2-3-4}+APOE2\epsilon \mathrm{2-3-4}+Age+Sex\\ +PC1+PC2+PC3+HP2\times APOE1\epsilon 2-3-4+HP2\\ \times APOE2\epsilon \mathrm{2-3-4}+APOE1\epsilon \mathrm{2-3-4}\times APOE2\epsilon \mathrm{2-3-4}+HP2\\ \times APOE1\epsilon \mathrm{2-3-4}\times APOE2\epsilon \mathrm{2-3-4}\end{array}$$ ((1))

where the “HP2” represents the number of HP2 alleles, and “APOE1 ε2-3-4” and “APOE2 ε2-3-4” represent the individual “APOE ε2-3-4” effect of each APOE allele.

We also fit individual stratified logistic regression models for the three APOE strata (i.e., individuals who carry at least one ε2, ε3, or ε4 allele, respectively) as shown below:

$$\begin{array}{l}Outcome~HP2+APOE\epsilon \mathrm{2-3-4}+Age+Sex+PC1+PC2+PC3\\ +HP2\times APOE\epsilon \mathrm{2-3-4}\end{array}$$ ((2))

where the “APOE ε2-3-4” represents the “APOE ε2-3-4” effect of the remaining APOE allele within each stratus.

In addition, we conducted survival analysis using the Cox proportional hazards model to investigate the effects of HP on the age of AD onset. The model is as below:

$$\begin{array}{l}\log \left\{\frac{H_1(t)}{H_0(t)}\right\}~HP2+APOE1\epsilon \mathrm{2-3-4}+APOE2\epsilon \mathrm{2-3-4}+Sex+PC1\\ +PC2+PC3+HP2\times APOE1\epsilon 2-3-4+HP2\\ \text{×}APOE2\epsilon \mathrm{2-3-4}+APOE1\epsilon \mathrm{2-3-4}\times APOE2\epsilon \mathrm{2-3-4}\\ +HP2\times APOE1\epsilon \mathrm{2-3-4}\times APOE2\epsilon \mathrm{2-3-4}\end{array}$$ ((3))

where the “$H_i(t)$” represents the hazard. The sample size (N) for analyses may differ due to variable missingness within subsets.

3 |. RESULTS

To examine the HP SV for association to AD risk, we imputed the HP SV for European-descent ADGC participants across 29 cohorts originally genotyped with multiple different arrays (Table 1). We obtained a > 95% hard-call rate in 25/29 of the cohorts, with an overall hard-call rate of 93.52% (33,725/36,062). This call rate suggests high imputation quality and confidence despite heterogeneity in genotyping platform. We noted that 4/29 cohorts showed statistically significant deviations from Hardy–Weinberg equilibrium (HWE; Table S3 in supporting information); however, sensitivity analyses removing these cohorts showed no qualitatively different results.

We first studied the association between AD status and HP2 allele count adjusting for sex, age, and the top three ancestry-based PCs. No statistically significant effect of HP genotypes was found (P = 0.386). We also examined dominant and recessive models of the HP2 alleles and found no significant associations (P = 0.449 and P = 0.497, respectively).

Given prior evidence of their molecular interaction, we further investigated HP alleles in the context of APOE. In these models, for simplification we assumed an equal (linear) increase in AD risk from APOE ε2 to APOE ε3, and APOE ε3 to APOE ε4 (denoted as APOE ε23–4) (see Text S3 in supporting information for more details on how we arrived at our modeling strategy). We first fit a logistic regression model using AD status as the outcome with age, sex, the first three PCs, HP2 allele count, APOE ε2-3-4 for each of the two APOE alleles, and pairwise interactions and a three-way interaction of these genetic effects of the two APOE alleles and the HP genotype (Equation 1). We found significant main effects from each of the two APOE ε2-3-4 variables and HP2 allele count, along with their two-way and three-way interaction terms (Table 2). These interaction effects can be observed in Figure 1A, whereby AD risk decreases dramatically with each HP2 allele in APOE ε2/ε4 (purple) individuals, yet increases in APOE ε4/ε4 (dark green), APOE ε2/ε3 (light green), and APOE ε2/ε2 (yellow) individuals. The risk remained fairly equal in APOE ε3/ε3 (magenta) and APOE ε3/ε4 (orange) individuals regardless of HP2 allele count (Figure 1A).

TABLE 2.

Effect of HP and both APOE alleles on AD risk (N = 20,512)

Variable	OR	95% CI	P-value
(Intercept)	0.49	(0.28, 0.87)	0.016
HP2	1.54	(1.10, 2.18)	0.013
APOE1 ε2-3-4	2.69	(1.66, 4.36)	5.48e-05
APOE2 ε2-3-4	4.4	(3.21, 6.05)	5.20e-20
Sex: Female	0.98	(0.92, 1.04)	0.447
Age	0.97	(0.97, 0.98)	7.61e-43
PC1	0.71	(0.28, 1.79)	0.463
PC2	0.55	(0.21, 1.42)	0.219
PC3	1.68	(0.64, 4.37)	0.291
HP2 x APOE1 ε2-3-4	0.63	(0.45, 0.88)	0.007
HP2 x APOE2 ε2-3-4	0.72	(0.57, 0.90)	0.004
APOE1 ε2-3-4 x APOE2 ε2-3-4	0.78	(0.58, 1.04)	0.092
HP2 x APOE1 ε2-3-4 x APOE2 ε2-3-4	1.41	(1.14, 1.74)	0.001

Notes: Effects are from a logistic regression model described in Equation 1. HP2 was included as the count of HP2 alleles. “APOE1 ε2-3-4” and “APOE2 ε2-3-4” represent the individual APOE ε2-3-4 effects from the two copies of APOE alleles, respectively. “Age” represents the age at onset for AD cases and age at last visit for controls. The bold lines are effects of interests that are statistically significant.

Abbreviations: AD, Alzheimer’s disease; APOE, apolipoprotein E; CI, confidence interval; HP, haptoglobin; OR, odds ratio.

FIGURE 1.

Effect of apolipoprotein E (APOE) and haptoglobin (HP) genotype on Alzheimer’s disease (AD) risk and age. A, Trend lines show the fitted effect estimates and standard errors from the logistic regression model in Table 2. B, Jitter plot and trend lines show the linear relationship between APOE ε2-3-4 and predicted AD probability stratified by HP genotypes for all APOE strata. Note the increase in slope of the trendline with each additional HP2 allele for ε4 carriers (left to right of the top row) and the decrease in slope of the trendline with each additional HP2 allele for ε2 carriers (left to right of the bottom row). C, Trend lines show the fitted effects estimates and standard errors from the Cox proportional hazards regression model in Table 4

To more easily describe the interaction effects detected in this model, we also fit individual logistic regression models stratified by participants’ APOE genotype, as illustrated in Equation 2. This approach isolates the effect of one APOE allele on the background of ε2, ε3, or ε4. Among APOE ε4 carriers (i.e., people with APOE ε2/ε4, ε3/ε4, and ε4/ε4, N = 8698), the APOE ε2-3-4 effect of the remaining allele leads to increased AD risk as expected (P = 1.891e-5; Table 3). Each HP2 allele further increases this APOE ε2-3-4 effect on AD risk significantly (P = 0.008; Table 3, Figure 1B Top). This can be seen as an increasing slope in the trend of the average risk (seen as increasing steepness of the gray trend lines in Figure 1B Top from left to right) with each additional HP2 allele. Among APOE ε2 carriers (N = 2330), given one APOE ε2 allele, the APOE ε2-3-4 effect of the remaining allele also leads to an increased AD risk (P = 3.415e-17; Table 3). However, unlike the APOE ε4 carriers, each HP2 allele decreases this APOE ε2-3-4 effect on AD risk (P = 0.012; Table 3, Figure 1B Bottom). This can be seen as decreasing slope in the trend of the average risk (seen as decreasing steepness of the gray trend lines in Figure 1B Bottom from left to right) with each additional HP2 allele. Analyses of APOE ε3 carriers did not show any significant effects of the HP alleles or any significant HP–APOE interactions (Table 3, Figure 1B Middle). Thus, the significant interactions in our model are due to APOE ε4 and APOE ε2 carriers, where the HP alleles show opposite modifying effects of APOE on AD risk.

TABLE 3.

Effect of HP and APOE ε2-3-4 alleles on AD risk stratified by APOE genetic background

	APOE ε2 (N = 2330)			APOE ε3 (N = 18443)			APOE ε4 (N = 8698)
Variable	OR	95% CI	P-value	OR	95% CI	P-value	OR	95% CI	P-value
(Intercept)	0.13	(0.05, 0.38)	1.874e-04	1.47	(1.04, 2.08)	0.03	34.28	(19.44, 60.44)	2.61e-34
HP2	1.49	(1.02, 2.19)	0.040	1.00	(0.89, 1.12)	0.969	0.79	(0.66, 0.95)	0.012
APOE ε2-3-4	5.78	(3.84, 8.69)	3.415e-17	2.89	(2.58, 3.22)	4.508e-79	1.60	(1.29, 1.98)	1.891e-05
Sex: Female	1.00	(0.83, 1.20)	0.969	0.97	(0.91, 1.03)	0.374	0.93	(0.85, 1.03)	0.164
Age	0.99	(0.98, 1.00)	0.046	0.97	(0.97, 0.98)	5.692e-37	0.96	(0.95, 0.96)	3.696e-38
PC1	3.94	(0.25, 63.01)	0.333	0.66	(0.25, 1.70)	0.386	0.97	(0.21, 4.41)	0.964
PC2	0.04	(0.00, 0.58)	0.018	0.72	(0.27, 1.91)	0.511	0.33	(0.07, 1.59)	0.166
PC3	5.58	(0.30, 102.72)	0.248	1.77	(0.66, 4.73)	0.255	3.07	(0.63, 15.02)	0.166
HP2 x APOE ε2-3-4	0.69	(0.52, 0.92)	0.012	1.00	(0.92, 1.08)	0.962	1.24	(1.06, 1.45)	0.008

Notes: Effects are from a logistic regression model described in Equation 2. HP2 was included as the count of HP2 alleles. The APOE ε2-3-4 variable represents the effect of the remaining APOE allele given the APOE genetic background and is encoded in ascending order for ε2, ε3, and ε4 alleles. “Age” represents the age at onset for AD cases and age at last visit for controls. N is the number of individuals in this model after omitting missing values in all model variables. The bold lines are effects of interests that are statistically significant.

Abbreviations: AD, Alzheimer’s disease; APOE, apolipoprotein E; CI, confidence interval; HP, haptoglobin; OR, odds ratio.

We have previously shown that genetic interaction models can show false positive associations due to deviation from the model’s genotype effect assumption,³⁸ which in our case is congruent to the dose-response assumption for APOE ε2-3-4. Therefore, we further explored the impact of our linear APOE ε2-3-4 assumption on the model fitting. While this simplifying assumption lets us assess the change in APOE effect due to the HP alleles, prior literature suggests that the protective effect of ε2 is smaller in magnitude than the risk effect of ε4. We conducted a sensitivity analysis exploring the individual effect of ε3 and the results demonstrated that the HP effect of APOE on AD risk is not due to misspecification of APOE main effects (Text S3, Tables S4–S6 in supporting information).

Given the known association of APOE to age of AD onset, we hypothesized that the HP–APOE interactions might also influence/modify the APOE effect on the age of AD onset. Thus, we performed survival analysis using a Cox proportional hazards regression model to investigate the effects of APOE ε2-3-4 from both APOE alleles individually and HP2 allele count on the age at AD onset (Equation 3) with controls included as censored. Similar significant effects were found for our age of onset analysis as were found for AD risk, though some coefficients were of marginal significance (Table 4, Figure 1C). The directions of these effects (positive/negative) were congruent to our findings in logistic regression models.

TABLE 4.

Effect of HP and both APOE alleles on age at AD onset (N = 20512)

Variable	HR	95% CI	P-value
HP2	1.24	(0.94, 1.63)	0.136
APOE1 ε2-3-4	1.91	(1.28, 2.85)	0.002
APOE2 ε2-3-4	2.77	(2.19, 3.50)	1.35e-17
Sex: Female	0.91	(0.87, 0.95)	2.72e-06
PC1	1.17	(0.64, 2.11)	0.611
PC2	0.57	(0.31, 1.07)	0.081
PC3	2.07	(1.11, 3.86)	0.021
HP2 x APOE1 ε2-3-4	0.78	(0.59, 1.02)	0.073
HP2 x APOE2 ε2-3-4	0.87	(0.74, 1.03)	0.102
APOE1 ε2-3-4 x APOE2 ε2-3-4	1	(0.80, 1.24)	0.986
HP2 x APOE1 ε2-3-4 x APOE2 ε2-3-4	1.17	(1.01, 1.37)	0.039

Notes: Effects are from a Cox proportional hazards regression model described in Equation 3. “HR” represents the hazard ratio. HP2 was included as the count of HP2 alleles. “APOE1 ε2-3-4” and “APOE2 ε2-3-4” represent the individual APOE ε2-3-4 effects from the two copies of APOE alleles, respectively. N is the number of individuals in this model after omitting missing values in all model variables. The bold lines are effects of interests that are statistically significant.

Abbreviations: AD, Alzheimer’s disease; APOE, apolipoprotein E; CI, confidence interval; HP, haptoglobin.

An eQTL (expression quantitative trait locus), rs2000999, is associated with the RNA level of HP that is independent of the HP SV.^{19, 20} We evaluated the impact of the genotypes of this eQTL and found no impact on our results (Text S4, Table S7 in supporting information).

As prior work suggests divergent evolutionary histories of this HP variant and potentially different effects of HP alleles on neurocognition by ancestry,²⁰ we also evaluated the HP–APOE interaction in African American ADGC cohorts. We imputed the HP genotypes for 12 ADGC AA cohorts (N = 4429, Text S5 in supporting information). Although we are underpowered to replicate the HP–APOE interaction effect in the AA cohorts (Table S8 in supporting information), a random effects meta-analysis for both European and AA cohorts’ data showed that the detected associations had consistent directions of effect and retained statistical significance (Figure 2, Text S6 in supporting information). Also, both the main effect of HP2 and the interaction effect of HP2 and APOE ε2-3-4 became more significant in APOE ε2 individuals (Table S9, S10 in supporting information) and less significant in APOE ε4 individuals, which is consistent with a previous report that APOE ε4 confers a lower risk to AD in AA individuals.³⁹

FIGURE 2.

Random effects meta-analysis of European and African ancestry (AA) data. European population effects are from the logistic regression model shown in Table 2. AA population effects are from the logistic regression model shown in Table S8 in supporting information. Meta-analysis effects are from a random effects meta-analysis of both European and AA data shown in Table S9 in supporting information

4 |. STRENGTHS AND LIMITATIONS

The ADGC dataset provided a large sample size to statistically evaluate this important HP–APOE molecular hypothesis, and we have achieved a high accuracy in our validated SV imputation process. The strength of our reported effect is also bolstered by a meta-analysis across different ancestry groups, and is strongly supported by prior mechanistic evidence of HP protein interactions with known components of AD risk. This study design and analysis has limitations. All the study subjects have an age of > 60, thus we can make no inference on the role of HP in early-onset forms of AD. As with prior studies of ADGC cohorts, we noticed large heterogeneity in age and case/control distributions across cohorts due to differences in geography and ascertainment strategy. This heterogeneity did not impact HP imputation; however, this could impact individual analyses within each cohort. The HP genotypes from 4 out of 29 cohorts are significantly deviated from HWE. However, similar effects were replicated in the 25 cohorts with no qualitative change in results suggesting that the effects were not driven by this deviation. We and others have also reported associations between the HP SV and serum and cerebrospinal fluid (CSF) HP levels in which the genetic effect of the HP SV remains independent of the CSF HP levels. Due to the limited availability of this data in our AD cohorts, we cannot eliminate HP levels as a potential confounder of our association.

5 |. DISCUSSION

The epsilon alleles of the APOE gene have been repeatedly shown to impact AD risk in a dose-dependent fashion. As a result, many studies adjust for the APOE effect by enumerating the number of ε4 alleles and often ignoring the more infrequent protective ε2 alleles. Here, we show that a functional SV in the HP gene alters the effect of APOE alleles on AD risk in ways that defy this convention. For example, the APOE ε2/ε4 genotype is generally considered to have higher risk for AD given the detrimental effect of ε4. However, individuals with APOE ε2/ε4 and HP2/HP2 are closer in risk to APOE ε3/ε3 individuals than APOE ε3/ε4 or APOE ε4/ε4 individuals. The modification effect on APOE by HP is larger among APOE ε2 carriers but more significant among APOE ε4 carriers. These results suggest that for some scenarios in which APOE stratification or adjustment is needed, the inclusion of HP genotypes and interactions is likely to improve predictions of AD risk. Given allele frequency differences of HP by ancestry, this association may explain some ancestry-specific differences in APOE risk.

There are multiple potential biological hypotheses that could drive the statistical interaction we detected. Both APOE and HP are related to inflammation. In vitro experiments using human APOE knock-in mice showed that Apoe ε4 mice are more susceptible to inflammation compared to Apoe ε2 and Apoe ε3.⁴⁰ The APOE peptide inhibits inflammation processes in isolated microglia;⁴⁰ furthermore, microglia with an APOE ε4 background demonstrated a greater release of proinflammatory cytokines.⁴¹ HP, as a free hemoglobin scavenger, reduces oxidative stress by preventing the release of free heme iron, thereby reducing inflammation, generation of reactive oxygen species, and oxidative damage to surrounding tissues. It is hypothesized that HP alleles exhibit differing antioxidant activities; therefore, the production of and protection from inflammatory products could be balanced by the differential activities of the APOE and HP alleles in a way that produces the interaction effect we observed. Studies have also shown associations between HP alleles and vascular complications of diabetes,⁴² which may point to a role for this interaction in the vascular contributions to dementia.

More directly, HP physically binds to and is an antioxidant of the APOE protein, likely protecting APOE function and activity. HP2 offers lower oxidation protection compared to HP1. However, due to the conformational differences in APOE structure from allelic variants, the HP antioxidant activity may also depend on the APOE allele. If true, this hypothesis would explain why we observe a statistical interaction effect of HP and APOE on AD risk rather than an independent effect of HP.

HP may also play a role in the binding of APOE with Aβ. Spagnuolo et al.⁴³ found that HP promotes formation of stable Aβ complexes with APOE proteins. Immunoassays showed that HP binds to Aβ in brain tissue from AD patients;⁴³ Shi et al. found that APOE and HP, along with four more proteins, are consistently associated with high amyloid burden.⁴⁴ Esiri et al. showed that HP facilitates the binding of APOE and Aβ.⁴⁵ It was also reported that in the human glioblastoma–astrocytoma cell line U-87 MG, HP impairs Aβ uptake and limits the toxicity of this peptide on these cells.⁴⁶ Therefore, another mechanistic hypothesis from our findings could be that these HP alleles may alter the promotion of binding of Aβ to APOE in an APOE-allele specific manner.

This variant affecting HP function may alter strategies for using APOE as a therapeutic target for AD,⁴⁷ or efforts to modify APOE–Aβ binding.⁴⁸ Haptoglobin has been used as a therapeutic agent in some settings for more than three decades and has also been explored as a way to mitigate oxidative damage from hemoglobin-driven pathology in the brain.⁴⁹ Given its prior clinical use and the ability to synthesize both HP functional alleles,⁵⁰ a form of haptoglobin therapy could potentially be tailored to individuals based on APOE genotype.

Supplementary Material

Additional supporting information can be found online in the Supporting Information section at the end of this article.

RESEARCHINCONTEXT.

1. Systematic Review: The authors reviewed the published literature (e.g., PubMed) on statistical and molecular interactions between apolipoprotein E (APOE) and haptoglobin (HP), and their influence on Alzheimer’s disease (AD) and related neurocognitive phenotypes, which suggest that interactions between APOE and HP exist.

2. Interpretation: Our result shows for the first time that accounting for a functional HP exon deletion variant alters the influence of APOE alleles on the risk of AD. This effect is particularly significant among APOE ε4 carriers. This suggests inclusion of the HP variant when APOE is considered an evaluation of AD risk.

3. Future Directions: Molecular studies are necessary to further elucidate the mechanism by which HP alters the APOE effects, and additional epidemiological studies are needed to examine any clinical or pathological differences in AD presentation based on HP alleles.