APOE AND ALZHEIMER DISEASE: A MAJOR GENE WITH SEMI-DOMINANT INHERITANCE

Abstract

Apolipoprotein E (APOE) dependent lifetime risks (LTRs) for Alzheimer Disease (AD) are currently not accurately known and odds ratios (ORs) alone are insufficient to assess these risks. We calculated AD lifetime risk in 7,351 cases and 10,132 controls from Caucasian ancestry using Rochester (USA) incidence data. At the age of 85 the LTR of AD without reference to APOE genotype was 11% in males and 14% in females. At the same age, this risk ranged from 51% for APOE44 male carriers to 60% for APOE44 female carriers, and from 23% for APOE34 male carriers to 30% for APOE34 female carriers, consistent with semi-dominant inheritance of a moderately penetrant gene. Using PAQUID (France) incidence data, estimates were globally similar except that at age 85 the LTRs reached 68% and 35 % for APOE 44 and APOE 34 female carriers, respectively. These risks are more similar to those of major genes in Mendelian diseases, such as BRCA1 in breast cancer, than those of low-risk common alleles identified by recent GWAS in complex diseases. In addition, stratification of our data by age- groups clearly demonstrates that APOE4 is a risk factor not only for late- onset but for early- onset AD as well. Together, these results urge a reappraisal of the impact of APOE in Alzheimer disease.

Introduction

Since the initial report of an enrichment of the APOE 4 allele of the apolipoprotein E (APOE) gene among Alzheimer disease (AD) patients ¹, the strength of the association between different APOE genotypes and the disease is reported as odds ratios (ORs). Taking as a basis the most frequent genotype (APOE 33), the odds ratios are estimated to be 3.2 for APOE 34 and 14.9 for APOE 44 whereas the APOE 2 allele has a protective effect in Caucasian subjects ². Note that these values obtained in clinical samples are probably slightly underestimated since larger ORs have recently been reported in a neuropathologically confirmed sample ³. However, ORs, that are basically epidemiological measures, are of limited interest in medical practice. What the physician (and the carrier of an “at risk“ genotype) want to know is not the magnitude of the increased risk conferred by this particular genotype with respect to the most frequent genotype in the population but the actual probability to develop the disease according to age and sex. To address this issue, we ⁴ and others ^5–7 previously attempted to calculate genotype-dependent AD lifetime risks (LTRs), i.e. the risk to develop the disease between birth and a given age ⁸. The LTR of a given APOE genotype could also be seen as the age-dependent penetrance of this APOE genotype or the probability that a randomly selected individual with this APOE genotype will develop AD by that age assuming that he does not die of another cause before that age ⁹. However, due to the limited sample sizes available at that time, 95% confidence intervals (95% CI) were huge and precluded any accurate estimate of LTRs especially in APOE 4 homozygotes. Taking advantage of the large case/control sample used in a recent European AD genome wide association study (GWAS)¹⁰ and adding two novel case/control cohorts, we undertook a novel attempt to estimate these values.

Material and methods

Patients

All subjects included in the GWAS as well as 2,971 new subjects (1,398 controls and 1,573 AD patients) originating from the west of France and the USA and ascertained according to the same criteria used for the GWAS subjects were included in this study. Therefore, a total of 7,531 cases and 10,132 controls from seven different sub-studies were available. All these subjects were Caucasians. Demographic characteristics of the samples are summarized in Table 1.

Table 1.

Summary of the characteristics of the different samples included in this study.

	Cases			Controls
Study	Number	Mean Age (y) (SD)	Male:Female	Number	Mean Age (y) (SD)	Male:Female
SPAIN	755	75.31 (9.33)	0.54	849	76.05 (12.37)	0.53
FINLAND	586	71.38 (7.50)	0.47	663	69.14 (6.09)	0.67
FRANCE 1	2 025	73.66 (8.91)	0.53	5 328	73.83 (5.42)	0.63
ITALY	1 513	76.62 (8.72)	0.48	1 338	70.76 (11.83)	0.80
USA	934	72.86 (6.26)	0.48	866	73.85 (6.36)	0.68
FRANCE 2	639	63.99 (9.88)	0.60	532	66.13 (12.01)	0.89
BELGIUM	1 079	78.56 (8.10)	0.51	556	64.20 (15.29)	0.75
TOTAL	7 531	74.03 (9.27)	0.51	10 132	72.37 (9.06)	0.67

Lifetime risks computation

Age-dependent penetrance for the different APOE genotypes was computed using a method similar to those described in Bickeboller et al. ⁴ and Satagopan et al.⁹ The only difference is that in our computation we accounted for mortality rates over the different age and genotype categories. We first calculated age-specific incidence rates in carriers of a given APOE genotype. Let I_i,g,s denote this probability over the relevant age category i (i=1 to 4 for the 4 age categories considered here; i.e., 59 and less, 60–69,70–79 and 80 and more) for individuals carrying the APOE genotype g of sex s and let S_i,g,s denote the probability for an individual of sex s and APOE genotype g of surviving over the age category i. The penetrance of APOE genotype g at the end of the a^th age-interval in individuals of sex s is then given by

$$P_{a,g,s}=1-\prod _{i=1}^a(1-I_{i,g,s}\times S_{i,g,s})$$ (1)

To estimate the values of S_i,g,s that account for differences in mortality rates for the different APOE genotypes, we used a model similar to the one proposed by Gerdes et al. ¹¹ based on genotype frequency differences between age groups. Briefly, if we denote by S_i,s the probability for an individual of sex s of surviving over the age category i that spans from age t₁ to age t₂, then we can derive S_i,g,s from S_i,s and the APOE genotype distribution observed in the sample using the following equation:

$$S_{i,g,s}=S_{i,s}\times \frac{P(G=g\mid \text{a}ge\ge t_2)}{P(G=g\mid \text{a}ge<t_1)}$$ (2)

The probabilities P(G = g|age ≥ t₂) and P(G = g|age < t₁) are the APOE g genotype frequencies in the overall population with age ≥ t₂ and age < t₁ respectively. These probabilities can be derived from the observed genotype frequencies in controls assuming that controls are representatives of the overall population.

The survival probabilities S_i,s over the different age and sex categories were derived from the 2005 Actuarial life table from the US Social Security Department (http://www.ssa.gov) and are reported in Table S1.

The values of I_i,g,s were calculated based on the age- and sex-specific incidences I_i,s over the corresponding age interval i (see Supplementary Table S1) and the APOE genotype frequencies observed in the sample in the different age-groups using Bayes’ formula:

$$I_{i,g,s}=P_{i,s}(D\mid G=g)=\frac{P_i(G=g\mid D)\times P_{i,s}(D)}{P_i(G=g)}$$ (3)

Where P_i,s (D)=I_i,s is the incidence rate for individuals of sex s in age group i,, P_i (G = g|D) is the prevalence of genotype g among those who develop the disease in age category i (note that it is not dependent on the sex as we did not find any genotype frequency differences between males and females) and P_i (G = g) is the prevalence of genotype g among all individuals in age category i (that can be approximated by the prevalence of g in controls).

To obtain confidence intervals for the penetrances, we used the bootstrap method as in Satagopan et al. ⁹ Penetrances were calculated repeatedly using regenerated samples of cases and controls where the subjects were sampled with replacement from their original populations (we kept the original number of individuals from the different studies). One thousand boostrap samples were generated using a script written in Rv2.10.1 (R_Development_Core_Team 2009) (http://www.R-project.org).

Results

Individuals were stratified into 4 age categories (less than 60, between 60 and 69, between 70 and 79 and more than 80 years old) depending on either the age of onset of the disease for patients or the current age for controls. APOE genotype frequency estimates and their 95% CI are shown in Supplementary Figure S1 for cases and controls in the different samples stratified by age (males and females were considered together as no frequency difference was found between gender). After stratification on age, only a limited heterogeneity was observed between samples and we thus decided to pool them all. Table 2 gives the ORs (taking APOE 33 as reference) obtained for the different APOE genotypes on the pooled data in the different age categories and over all age categories. The OR of APOE 44 in the whole sample (OR=14.49; 95%CI = [11.91; 17.64]) is consistent with previous estimates ² but varies significantly with age, ranging from 5.6 (95%CI = [3.17; 9.89]) when the onset is before the age of 60 years old to 35.07 (95%CI = [23.8; 51.68]) when the onset is between 60 and 69. The same is true for APOE 34 with ORs in the range between 2.09 (95%CI= [1.61; 2.71]) for patients with an onset before 60 years and 4.18 (95%CI= [3.59; 4.88]) for the patients with an onset between 70 and 79 years. For the other APOE genotypes, ORs are more homogeneous across age-groups and are estimated to be 2.64 [2.13; 3.27] for APOE 24 and 0.56 [0.49; 0.64] for APOE 22 and 23 considered together.

Table 2.

Odds-Ratios [95% Confidence Interval] of the different APOE genotypes (using APOE 33 genotype as reference).

Age-group	Sample Size (Cases/Controls)	APOE 44	APOE 34	APOE 24	APOE 22+23
Less than 60	593/610	5.60 [3.17;9.89]	2.09 [1.61;2.71]	1.29 [0.61;2.71]	0.43 [0.24;0.77]
60 to 70	1511/2614	35.07 [23.8;51.68]	4.18 [3.59;4.88]	2.2 [1.32;3.68]	0.59 [0.43;0.82]
70 to 80	3244/5175	15.75 [11.93;20.8]	4.54 [4.09;5.05]	2.79 [2.05;3.8]	0.51 [0.41;0.63]
More than 80	2183/1733	7.87 [4.28;14.46]	3.16 [2.68;3.72]	3.77 [2.28;6.23]	0.59 [0.47;0.74]
All	7531/10132	14.49 [11.91; 17.64]	3.63 [3.37; 3.90]	2.64 [2.13; 3.27]	0.56 [0.49; 0.64]
p-valuea		1.05×10−7	3.40×10−8	0.11	0.64

^aP-value of the Woolf test of homogeneity of OR between the different age categories.

The AD LTR estimates according to age, sex and APOE genotype and computed using the Rochester (USA) incidence data ¹² are presented in Table 3A. At age 85, AD-LTR reached 51 % (95%CI= [41;70]) and 60% (95%CI= [47;84]) for APOE 44 male and female carriers respectively and 23% (95%CI= [22;25]) and 30% (95%CI= [28;32]) for APOE 34 male and female carriers respectively, consistent with semi dominant inheritance of a moderately penetrant gene. To study the impact of incidence estimates on LTR, we recomputed LTR using the French PAQUID incidence rates ¹³ (Table 3B). For the PAQUID data, incidence rates were only available for individuals older than 65 years. In order to use our method however we need to compute incidence rates among individuals of less than 60 years old and individuals between 60 years old and 64 years old. Given that in the age category 65–69 years, incidence rates obtained from the Rochester Epidemiological study and from PAQUID were not significantly different, we used Rochester data for individuals younger than 65. Estimated values were similar to those reported above except that the LTRs reached 68% and 35 % at age 85 for APOE 44 and APOE 34 female carriers, respectively. We thus can be confident that our results are globally valid for Caucasian populations. They can be compared with the lifetime risk of AD as a function of age, without reference to APOE genotype, that is 10–11% for males and 14–18 % for females by age 85 depending on whether Rochester or PAQUID incidence rates are used.

Table 3.

Age specific LTR estimates [95% Confidence Interval] (%) of the different APOE genotypes in males and females after accounting for APOE genotype differential effect on mortality. A: Rochester incidence rates. B: PAQUID incidence rates. The first column provides LTR estimates irrespective of APOE genotype.

A: Results using Rochester incidence rates
Age	Gender		APOE 44	APOE 34	APOE 24	APOE 33	APOE 22+23
65	Male	0.54 [0.29;0.79]	4.14 [1.47;10.10]	1.03 [0.83;1.26]	0.28 [0.08;0.79]	0.28 [0.25;0.32]	0.21 [0.13;0.34]
	Female	0.30 [0.08;0.53]	1.95 [0.85;4.57]	0.57 [0.48;0.70]	0.17 [0.08;0.45]	0.17 [0.15;0.19]	0.11 [0.07;0.17]
75	Male	3.19 [2.11;4.26]	28.35 [21.41;42.91]	6.96 [6.38;7.58]	4.39 [3.10;6.74]	1.78 [1.70;1.87]	1.12 [0.91;1.37]
	Female	3.01 [2.19;3.85]	27.81 [20.11;44.30]	7.03 [6.41;7.73]	4.52 [3.11;7.06]	1.69 [1.62;1.77]	1.01 [0.84;1.23]
85	Male	11.19 [8.34;13.98]	50.93 [40.60;70.51]	23.22 [21.72;25.02]	19.99 [14.48;29.67]	7.76 [7.49;8.04]	4.87 [4.15;5.70]
	Female	14.42 [12.27;16.57]	60.37 [47.42;84.43]	30.09 [28.02;32.62]	26.82 [19.07;40.79]	10.23 [9.84;10.64]	6.40 [5.38;7.51]

B: Results using PAQUID incidence rates
Age	Gender		APOE 44	APOE 34	APOE 24	APOE 33	APOE 22+23
65	Male	0.38 [0.12;0.65]	2.92 [1.04;7.39]	0.73 [0.59;0.89]	0.20 [0.06;0.55]	0.20 [0.18;0.22]	0.15 [0.09;0.23]
	Female	0.19 [0.00;0.43]	1.95 [0.86;4.67]	0.57 [0.48;0.70]	0.17 [0.08;0.45]	0.17 [0.15;0.19]	0.11 [0.07;0.16]
75	Male	3.62 [1.87;5.35]	33.19 [24.17;51.75]	8.30 [7.59;9.09]	5.35 [3.70;8.31]	2.02 [1.93;2.12]	1.23 [1.02;1.50]
	Female	3.24 [1.77;4.75]	31.85 [22.78;50.82]	8.07 [7.36;8.87]	5.21 [3.59;8.17]	1.92 [1.84;2.02]	1.15 [0.95;1.40]
85	Male	10.22 [5.68;14.64]	51.62 [40.71;74.42]	21.69 [20.43;23.15]	17.53 [13.17;24.81]	6.85 [6.64;7.08]	4.23 [3.67;4.88]
	Female	16.66 [12.04;21.19]	67.81 [53.63;91.44]	34.87 [32.43;37.81]	31.29 [22.33;47.66]	11.98 [11.54;12.46]	7.50 [6.33;8.87]

Discussion

Clearly these values indicate that the effect of APOE on Alzheimer disease is more similar to the one of major genes in Mendelian diseases such as BRCA1 in breast cancer than the one of low-risk common alleles identified by recent GWASs in complex diseases. For a comparison, the lifetime risk of breast cancer in BRCA1 mutation carriers by age 70 is estimated around 57% (95% CI, 47% to 66%), which is similar to our estimated APOE 44 penetrances by age 85¹⁴.

The frequency of the APOE 4 allele in human populations ranges from 0.09 to 0.30 ^2,15. These values are well above those found for other deleterious alleles responsible for Mendelian diseases, even those involved in most recessive disorders. This apparent paradox might be explained by the fact that the APOE 4 allele, that is probably the ancestral APOE allele in humans ¹⁶, exerts its deleterious effect mainly in elderly individuals. Its impact on the reproductive fitness of the carriers is probably very limited. However, its reduced frequency compared to the APOE 3 allele, if not due to genetic drift, suggests that it might slightly reduce the fitness.

As discussed by Yang et al.¹⁷, lifetime risk estimates suffer from uncertainties in the population incidence rates, genotype frequencies and effect sizes. By using a bootstrap method, we were able to account for the latter two uncertainties but since we had to rely on published records for incidence rates, we were not able to model uncertainties in incidence rates. To address this issue, we used incidence data from two different studies including Caucasian populations and show that LTR estimates are indeed consistent. We also had to account for the fact that mortality rates vary depending on APOE genotypes. In particular, the APOE 4 allele increases cardiovascular mortality ¹⁸ and a recent GWAS confirmed that the APOE locus plays a role in longevity as this locus was among these loci that showed the most extreme allele frequency differences in centenarians compared to younger controls ¹⁹. To account for mortality differences depending on APOE genotypes in LTRs computations, we built a new model that assumes that, for a given APOE genotype, survival probabilities are proportional to the genotype frequency differences observed over the different age categories in the sample. Finally, in addition to statistical and population-specific uncertainty, one has to acknowledge that incidence rates are likely to change with time. Incidence rates for current 70-year old subjects may not apply when a current 30-year old will be 70 years old in 40 years.

In conclusion, we think that our results urge for a shift of category of the APOE gene from “risk factor” to “major gene”. This shift is not a pure semantic exercise as it has profound implications for patients. In Caucasian populations roughly 2% of the population bears the APOE 44 genotype. Considering the major risk conferred by this genotype (roughly 30% by age 75 and >50% by age 85), it would be appropriate to target in priority these individuals, as well as PSENs or APP mutation carriers, in clinical trials aimed at developing novel preventive therapeutics. In addition, as advocated by the REVEAL study ²⁰ whose aim is to examine the effects of APOE genotype disclosure, genetic counseling based on accurate LTR estimates should also probably be considered in the future for these individuals.