Part 1 · The disease
Genes and inheritance
Most Alzheimer's is not inherited in a simple way. One common gene, APOE, shifts the odds. A few rare genes all but guarantee it, and an extra chromosome 21 does too. Pick a genotype, then play the family below.
APOE: one gene, three common versions
Everyone carries two copies of APOE, one from each parent. The three common versions, ε2, ε3 and ε4, differ at only two positions in the protein (112 and 158). 531 Choose a pair to see how often it turns up in people with Alzheimer's compared with ε3/ε3.
Odds ratios compare the odds of Alzheimer's with ε3/ε3, on a log scale. They are group averages, not anyone's personal chance.
Odds ratio is not lifetime risk, and the estimates disagree
An odds ratio of 15 does not mean a 15-times or a near-certain chance. Researchers have turned the odds into lifetime risk two ways, and get different answers. Rows for your chosen genotype are highlighted.
Why they disagree. Case-control and memory-clinic samples carry referral bias; the population cohorts measured MCI or dementia rather than Alzheimer's alone; and most cohorts are of European ancestry, where the ε4 effect is larger than in African ancestry. 21,47,420
What both agree on. ε4 raises risk but does not decide it: many ε4/ε4 people never develop dementia, and Alzheimer's often occurs without ε4. Only 38% of APOE alleles in people with Alzheimer's are ε4. 194,420,531
Dominant inheritance: APP, PSEN1 and PSEN2
A mutation in APP, PSEN1 or PSEN2, or an extra copy of APP, is the only known cause that all but guarantees Alzheimer's. 40 Pick the gene, play the family, and press "Another family" to see that 50% is a chance for each child, not a quota.
Schematic family. DIAN years are cross-sectional estimates relative to the parent's age at onset.
Down syndrome: a third copy of APP
APP, the gene for the protein amyloid beta is cut from, sits on chromosome 21. People with Down syndrome usually have three copies of chromosome 21, so three copies of APP. Drag the age, then switch to the rare case with only two APP copies.
Ages are from separate cohorts; positions on the ladder are to scale.
Resilience, risk scores and testing
APOE3 Christchurch
A woman with the Colombian PSEN1 E280A mutation did not develop MCI until her seventies, about three decades late. She had two copies of APOE3 Christchurch, very high amyloid, and little tau. Among 1,077 carriers, the 27 with one Christchurch copy began to decline at a median 52 years against 47, about 5 years later. 26,421
RELN-COLBOS, 2023
A man with the same PSEN1 mutation stayed cognitively intact until 67 without Christchurch. He carried a gain-of-function variant in Reelin, which binds the same receptors as APOE; in mice it reduced tau phosphorylation. Both resilient people had very high amyloid but little entorhinal tau, suggesting protection acted at the tau stage. 26,323
A673T: protection in APP itself
An Icelandic APP variant next to the beta-secretase site cut amyloid production by about 40% in the lab, and protected against Alzheimer's and against decline in older people without it. 262 See Amyloid, animated.
The Colombian kindred
A 1997 survey of five extended families in Antioquia (nearly 3,000 people) found 128 affected, with onset at a mean 46.8 years. PSEN1 E280A traces to one founder about 15 generations ago. In 449 carriers, median ages were 44 for MCI, 49 for dementia and 59 for death. 8,296,322
GWAS: 75 places in the genome
The largest study, Bellenguez 2022 in Nature Genetics (111,326 cases, 677,663 controls), found 75 risk loci, 42 new, pointing to amyloid and tau processing, microglia and TNF-α signalling. Rare TREM2 R47H raised risk about 2.9-fold. 46,263
Polygenic scores
A score built with APOE, age and sex reached an AUC of 78.2%. Beyond age and APOE ε4, the Bellenguez score separated the top and bottom tenths by only 1.6- to 1.9-fold. Scores discriminate poorly in the general population, work less well outside European ancestry, and have no established clinical use. 46,144,309
Twins: heritability 58–79%
In 11,884 Swedish twin pairs, heritability of Alzheimer's was 58% in the full model and 79% in the best-fitting one; the rest came from non-shared environment. Identical twins also differed less in age at onset. The authors stress non-genetic factors still matter. 190
Is ε4/ε4 its own genetic form?
In 2024, Fortea and colleagues reported Alzheimer's biology in almost all ε4/ε4 people, with symptoms at a mean of 65, and proposed it as a distinct genetic form. Critics answer that near-full penetrance of biology is not near-full penetrance of dementia, citing lifetime risks of about 30–55%. 21,184
Testing: what the guidelines say
Predictive APOE testing
A 1995 ACMG/ASHG statement said APOE should not be used for predictive testing. The 2011 ACMG/NSGC guideline calls APOE susceptibility testing "not clinically recommended" and advises against direct-to-consumer testing; if a person still insists after counselling, it says testing should follow the Huntington disease protocol. 7,201,202
Families with a dominant gene
The guideline says testing for APP, PSEN1 and PSEN2 should be offered to people with early-onset Alzheimer's and a family history, to families with a dominant pattern, and to relatives of a known carrier, using a Huntington-style protocol. Children should not be tested. Only 8.4% of at-risk people in one series asked for predictive testing. 40,201
Genotyping before anti-amyloid drugs
US labels for lecanemab and donanemab say ApoE ε4 testing "should be performed prior to initiation of treatment", because ARIA (brain swelling or bleeding on MRI) is more common in ε4/ε4. This is a different purpose from predicting Alzheimer's. 103,154,155,423 See Anti-amyloid antibodies.
What disclosure did: REVEAL
162 adult children of people with Alzheimer's were randomised to learn their APOE result or not. At 1 year anxiety, depression and distress did not differ. ε4-positive people more often changed health behaviours (adjusted OR 2.73) and were 5.76 times more likely to change long-term care insurance. 72,208,539