APOE4: the 'Alzheimer's gene', and what that phrase gets wrong

Two letters make three common alleles

The three versions of the protein differ at positions 112 and 158. ApoE3, the common one, has a cysteine at 112 and an arginine at 158. ApoE4 has an arginine at both; apoE2 has a cysteine at both^[1]. Each swap comes from a single changed letter in the gene. The letter behind position 112 is the variant catalogued as rs429358, where a T becomes a C; the letter behind position 158 is rs7412, where a C becomes a T^{[2, 3]}. The two sit 138 bases apart on chromosome 19.

Everyone has two copies of chromosome 19, so a genotype is two alleles: e3/e3, e3/e4, e4/e4 and so on. A raw DNA file does not list alleles. It lists the two letters it found at each position, without saying which copy each came from. For five of the six common genotypes that does not matter, because only one pair of alleles fits the letters. The exception is one letter of each kind at both positions, which fits e2/e4 and also the rare e1/e3; clinical laboratories and our report call it e2/e4 and say so.

# rsid      chromosome  position  genotype
rs429358    19          45411941  TC
rs7412      19          45412079  CC

What the case-control studies measured

Most of what is known about APOE and Alzheimer's comes from comparing the genotypes of people with the disease against people of similar age without it. The result is an odds ratio: how much more common a genotype is among cases, set against the e3/e3 reference. The study that fixed the standard numbers was a 1997 meta-analysis by Lindsay Farrer and colleagues, pooling 5,930 patients and 8,607 controls from 40 research teams^[5]. In white participants from clinic and autopsy series, it found odds ratios of 3.2 for e3/e4 and 14.9 for e4/e4. Among Japanese participants the e4 effect was stronger; among African Americans it was weaker and differed between studies^[5].

Two things in the table matter more than any single row. The first is that an odds ratio is relative. Fourteen times the odds sounds like a sentence until it is set against how common Alzheimer's is at a given age, which is what the absolute figures in the next section do. The second is that the size of the effect moves with who was studied. Most of the data are from people of European ancestry, the effect is larger in Japanese studies and smaller in African American ones^[5], and the Copenhagen cohort's own authors note that it included only people of white European background^[8].

Sex moves it too, in a narrower way than is often said. Neu and colleagues pooled 27 data sets and found almost no difference between men and women with e3/e4 across ages 55 to 85, odds ratios of 3.09 and 3.31; women had the higher risk between 65 and 75^[7]. Belloy's pooled analysis of 68,756 people placed the excess for women at ages 60 to 70^[9].

At the other end, e2 is the least common of the three alleles^[3] and the one linked to lower risk. In Belloy's data, carrying e2 with no e4 went with about half the odds of e3/e3 in white participants, a weaker effect in Black participants and no clear effect in Hispanic ones^[9]. In about 5,000 people whose brains were examined after death, e2/e2 had odds of 0.13 against e3/e3; in clinically diagnosed groups the figure was 0.52^[10].

Two copies of e4: from 6 to 90 percent

About 2 percent of people of European ancestry have two copies of e4 (in gnomAD, 676 of 33,958 non-Finnish Europeans^[2], our arithmetic). It is the genotype that gives rise to the most alarming headlines and the widest spread of published figures. Absolute risk is the share of people with the genotype who are diagnosed within some span of time, and each study below measured a different span, in different people.

The Copenhagen figures are the most directly comparable to a general population: 104,537 people from two Danish cohorts, followed forward, with the ten-year risk broken down by age and sex^[8]. A ten-year window is short, so the numbers look small. A lifetime to 85 is long, so Genin's are large: 51 percent of men and 60 percent of women with e4/e4^[11]. The 90 percent is from Corder's 1993 study of 42 families chosen because late-onset Alzheimer's ran in them, which is why it is the highest; in the same families the average age at onset fell from 84 with no e4 to 68 with two^[4].

In 2024 Juan Fortea and colleagues went further. Using autopsy records from the US National Alzheimer's Coordinating Center and biomarker data from five cohorts, they found that people with e4/e4 had more abnormal Alzheimer's biomarkers than people with e3/e3 from about age 55 and developed symptoms at an average age of 65.1. Describing near-full penetrance of the disease's biology, they proposed that two copies of e4 be treated as a distinct genetic form of the disease^[12]. Most of the people studied were of European ancestry^[12]. The proposal is about the biology; it does not change the Copenhagen or Rochester counts of who received a diagnosis within a given time.

Finland carries more e4 than the rest of Europe

The e4 letter is common everywhere and unevenly spread. In the gnomAD v4.1 genomes, the C at rs429358 is on 19.5 percent of Finnish chromosomes, against 13.8 percent of non-Finnish European ones; among the large groups only people of African or African American ancestry carry it more often, at 21.6 percent^[2]. The e2 letter, the T at rs7412, goes the other way: 4.2 percent of Finnish chromosomes against 7.9 percent of non-Finnish European ones^[3].

A higher frequency in Finland is a statement about the population, and it does not raise the risk that goes with any one genotype. The odds ratios in the table were measured within populations, against e3/e3 people from the same population.

APOE also moves blood cholesterol

APOE was a lipid gene long before it was an Alzheimer's gene. The protein is a ligand for the LDL receptor, the route by which the liver clears cholesterol-carrying particles from the blood, and the three versions bind and behave differently^[1]. A 2007 meta-analysis by Bennet and colleagues found LDL cholesterol rising roughly in step from e2 to e4: people with e2/e2, at one end, had LDL cholesterol about 31 percent lower than people with e4/e4 at the other. Against e3/e3, coronary disease was 20 percent less likely in e2 carriers (odds ratio 0.80) and slightly more likely in e4 carriers (1.06, an interval that crossed 1)^[13].

Two copies of e2 have a rarer consequence. ApoE2 binds the LDL receptor poorly, and e2/e2 is the genotype behind most type III hyperlipoproteinaemia, also called dysbetalipoproteinaemia, in which remnant particles of cholesterol and triglyceride build up^{[1, 14]}. Most people with e2/e2 never develop it: it takes a second factor, genetic or acquired, and only a minority of homozygotes have the raised lipids^[14].

Why APOE is not a diagnostic test

A diagnostic test answers whether someone has a disease. An APOE genotype answers neither that nor whether they will have it. Many e4 carriers never develop Alzheimer's, as the Copenhagen and Rochester figures show^{[8, 11]}, and the disease is common in people with no e4: the Choosing Wisely rationale published in American Family Physician describes e4 as neither necessary nor sufficient for it^[15].

There is one setting in which the genotype now has a defined clinical job. The anti-amyloid drug lecanemab (Leqembi) can cause swelling and small bleeds in the brain, called ARIA, and its US label says these are more frequent in e4 homozygotes, including the symptomatic and serious forms; it states that e4 status should be tested before treatment starts, to inform that risk^[17]. In the European Union the drug was authorised in 2025 only for people with one copy of e4 or none^[18]. In both cases the genotype is used to weigh a drug's safety in someone already diagnosed, which is a different question from predicting the disease.

How our report handles APOE

Every report built from a VCF, a gVCF or a chip export runs the APOE module under the one processing consent the service asks for; there is no separate tick box for it. What changes is how the result reaches the page. The APOE section opens with a warning headed 'Read this before you reveal it', which says what APOE is, that it is not a diagnosis, and that most e4 carriers never develop Alzheimer's. Below it is a closed panel with a button reading 'Reveal my APOE result'. The panel is closed every time the page loads, and nothing about the genotype is shown until it is opened.

Three other parts of the product keep the same line. The question-answering feature that works from a report leaves APOE out of what it can draw on, so a chat cannot reveal it. When a later re-analysis changes the APOE result, the Genome Watch timeline says only that a sensitive result you keep hidden was updated, without naming the gene. The downloadable PDF is the exception: it cannot know whether the panel was opened on screen, so it prints the result open. The card in the sample report shows the layout, and the free file check says whether a file is a chip export or a sequencing file before anything is paid for.