Genes That Influence Longevity, and How to Test for Them

How much of lifespan is actually inherited

Start with the number, because it reframes everything below. Classical twin studies put the heritability of human lifespan at roughly 20 to 30 percent. That is lower than most people assume, and it is probably an overestimate rather than an underestimate.

A large analysis of genealogy records found that much of the apparent heritability disappears once you account for assortative mating, meaning the tendency of people to marry partners similar to themselves in health, wealth and habits. Spouses, who share no DNA, resemble each other in lifespan more than the genetic model predicts. That points to shared environment doing work that was previously credited to shared genes.

Ruby and colleagues put a number on it. Their structural equation modelling, accounting for assortative mating, gave an upper-bound estimate of 7%, and the paper concluded true heritability sits well below 10%.

That figure is now itself disputed. A 2025 analysis argues that correcting for extrinsic mortality and inconsistent minimum-age cutoffs more than doubles previous estimates, putting intrinsic lifespan heritability above 50%, in line with the roughly 50% typical of other human physiological traits. Its criticism of the pedigree work is that it pools self-reported data across roughly 300 years and wildly different environments. That analysis is a preprint and has not been peer-reviewed, so it belongs here as a live challenge rather than as a settled counter-figure.

So the honest range under active debate runs from 7% to above 50%, which is wider than any single number suggests.

The practical reading is not that genes do not matter. It is that the ceiling they set is high enough that almost nobody reaches it, so the binding constraint for most people is what they do rather than what they inherited. That reading holds at either end of the disputed range.

The genes with replicated evidence

The list below is deliberately short. A great many genes have been associated with longevity in a single study and then failed to replicate, which is the normal fate of candidate-gene findings in small cohorts. These are the ones that have held up across independent populations, with an honest note on how strong each is.

The shortness of that list is itself a finding. Deelen and colleagues ran a genome-wide association meta-analysis at large scale and confirmed only a very small number of loci. When a study of that size finds few hits, the honest conclusion is that few hits exist, not that the search was inadequate.

Two patterns are worth noticing. The genes with the strongest human evidence, APOE and LPA, are both lipid-related, which fits the broader picture that cardiovascular disease is the largest single cause of death in developed countries. And the genes with the most exciting mechanistic stories, SIRT6 among them, have the thinnest human population evidence.

APOE, and why it deserves separate treatment

APOE comes in three common versions: ε2, ε3 and ε4. Most people carry two copies of ε3. Carrying one copy of ε4 raises Alzheimer’s risk substantially, and carrying two raises it considerably further. The ε2 allele appears protective and is over-represented among centenarians, though ε2 homozygotes make up less than 1% of the general population, which is why the protective effect is much harder to study than the risk effect.

The numbers are worth seeing, with the caveat that they are population data rather than a personal forecast.

Relative risk against non-carriers runs at roughly three to four fold in heterozygotes and roughly nine to fifteen fold in homozygotes. Prospective cohort data from Whitehall II, following participants for twenty years, put the sub-distribution hazard ratios for dementia at 2.19 (95% CI 1.73–2.77) for heterozygotes and 5.97 (95% CI 3.85–9.28) for homozygotes. In that cohort 25% of people were heterozygous and 2% homozygous.

Lifetime risk is the figure most people actually want. Population-based cohort estimates put lifetime risk of mild cognitive impairment or dementia at roughly 30 to 35% for ε4 homozygotes, 20 to 25% for heterozygotes and 10 to 15% for non-carriers. Qian and colleagues, analysing Framingham and Rotterdam data, put lifetime incidence to age 80 to 85 for ε4/ε4 individuals in the 31 to 40% range.

Read those two paragraphs together and the point the rest of this section makes becomes concrete: around half of ε4 homozygotes will not develop Alzheimer’s dementia in their lifetimes.

It also explains why searching this topic returns wildly different numbers. Retrospective case-control studies have produced much higher figures, over 50% by age 85 for homozygotes, than prospective cohort studies do. The confidence intervals are wide, particularly for homozygotes, and estimates vary by ancestry: the association has been reported as weaker in African-American and Hispanic populations and stronger in Japanese populations relative to those of European ancestry.

This is the single most consequential result a longevity-motivated genetic test is likely to return, and it deserves thinking about before you test rather than after. It shifts probability without determining outcome. Many ε4 carriers never develop dementia, and many people who do develop it carry no ε4 at all.

There is no treatment that changes the risk conferred by the variant itself, though the therapeutic landscape around it is moving and APOE status has begun to carry clinical consequences of its own, particularly around eligibility and monitoring for newer Alzheimer’s therapies. What that leaves is a result with genuine psychological weight and no corresponding action, beyond the modifiable dementia risk factors that are worth addressing regardless of genotype. Some people find knowing useful for planning. Others find it a burden. Both are reasonable, and a genetic counsellor is the right person to discuss it with beforehand.

Is ε4 homozygosity a risk factor or a disease?

Since 2024 there has been a serious argument that carrying two copies of ε4 is better understood as a distinct genetic form of Alzheimer’s disease rather than a risk factor for it. The case rests on near-universal biomarker abnormality in homozygotes and a narrower, earlier onset window, with a mean age of 65.1 on the reanalysed data.

The counter-argument is the one made above. Prospective cohorts show a substantial fraction of homozygotes never progressing to mild cognitive impairment or dementia, which is difficult to square with a deterministic genetic disease model.

This is unresolved, and it matters more than most academic disagreements, because roughly 2% of people are homozygous. Anyone in that group deciding whether to test should know the framing itself is contested rather than settled.

One practical note. If you do carry ε4, the modifiable factors become more valuable rather than less, because you are starting closer to a threshold. That is the same logic that applies to elevated Lp(a).

How testing actually works

The word test covers several different things here, and the differences determine what you learn. Most disappointment with genetic testing comes from expecting one category to do the job of another.

The last row is the most useful one on this page. Lp(a) is roughly largely genetic, with heritability estimates of roughly 70 to 90 percent, and you do not need a gene panel to find out, because a standard blood test measures the actual concentration. It costs little, most people have never had it done, and it is worth knowing once in your life.

Two cautions on consumer arrays specifically. They genotype pre-selected positions rather than sequencing, so a rare variant simply will not appear. And third-party services that reinterpret raw array data have a well-documented false-positive rate, which means anything consequential should be confirmed with a clinical-grade test before you act on it.

What a result should actually change

For most people, honestly, very little. The variants above shift probabilities by amounts that are small relative to the factors you control, and knowing your FOXO3 status does not imply a different plan.

Two results are exceptions. An elevated Lp(a) is a reason to manage ApoB and blood pressure more aggressively than the population default, because your baseline risk is higher and unchangeable. APOE4 status is a reason to take the modifiable dementia risk factors seriously earlier, and for some people a reason to plan differently.

The general principle is worth stating plainly, because it inverts the intuition. Inherited risk you cannot change increases the value of the levers you can, since a fixed disadvantage means the remaining variables carry more weight. A genetic result is a reason to do more of what already works, not a verdict that makes the effort pointless.

That is not just logic. It has been measured, in exactly the marker this page uses as its example.

In EPIC-Norfolk, among people with Lp(a) above 50 mg/dL, those with few traditional risk factors had a one-third to two-thirds lower risk of an atherosclerotic cardiovascular event over 11.5 years of follow-up than those with an unhealthy lifestyle. Same unchangeable genetic burden, substantially different outcomes.

The same pattern runs the other way. Someone with high Lp(a) and none of the traditional risk factors may never develop atherosclerotic cardiovascular disease at all, while someone carrying several of them sees overall risk rise even within the 30 to 50 mg/dL grey zone. The genotype sets the starting line. The rest determines where you finish.

What to test, and what to do about the result

Consumer genetics marketing tends toward one of two failure modes: implying your DNA holds a personalised blueprint that ordinary advice cannot match, or reducing everything to a risk score with no accompanying action.

Get an Lp(a) blood test once. It is inexpensive, widely available, needs no genetics service, and roughly one in five people carry a level high enough to matter. Most people have never had it measured. Updated ESC/EAS dyslipidaemia guidance recommends measuring it at least once in every adult’s lifetime, so this is a guideline position rather than ours.

The 2022 European Atherosclerosis Society consensus statement sets the thresholds. Below 30 mg/dL (75 nmol/L) rules Lp(a)-attributable risk out. Above 50 mg/dL (125 nmol/L) rules it in. Between the two is an explicit grey zone, which exists both because assays vary near decision thresholds and because risk in that band depends on what other risk factors you carry.

Those cut-offs are a clinical convention rather than a biological cliff. The relationship between Lp(a) concentration and cardiovascular risk is continuous, with no true threshold effect. UK Biobank data in the same statement show it: against the median of 7 mg/dL, hazard ratios were 1.22 at 30 mg/dL and 1.40 at 50 mg/dL, rising further at higher concentrations.

Roughly 20% of people are above 50 mg/dL, and about 0.3% above 175 mg/dL (438 nmol/L). Testing rates run at around 1 to 2%. That gap between how common elevated Lp(a) is and how rarely anyone looks for it is the whole argument for the recommendation above.

Two practical notes. Units are not interchangeable: nmol/L is the preferred clinical standard, and mg/dL does not convert to it by a fixed factor, so check which unit a result is reported in before comparing across labs. And concentrations differ by ancestry, with Black and Hispanic populations generally showing somewhat higher levels than white and Asian populations.

An elevated result is currently unchangeable, which is why the advice is to manage everything around it. Several Lp(a)-lowering therapies are in late-stage clinical trials, so that may not remain true.

Think about APOE before testing rather than after. Decide in advance how you would use the result, and consider speaking to a genetic counsellor first.

The psychological weight is the part people anticipate. Three practical consequences get less attention and belong in the decision.

Insurance. How predictive genetic results may be used differs by country, and the protections that exist for health insurance and employment do not always extend to life, disability or long-term care cover. Worth checking the position where you live before testing rather than after. This is general information rather than legal or financial advice.

Family. A genetic result is information about relatives who did not consent to being tested. APOE status in particular carries implications for children and siblings, who may or may not want to know.

Irreversibility. A blood marker changes. A genotype does not, and you cannot un-know it.

Treat consumer array results as a starting point rather than a conclusion, and confirm anything consequential clinically. Be particularly wary of third-party tools that reinterpret raw data.

Whatever the result, the plan barely changes. What genetics does is shift where your personal starting line sits, and the modifiable factors carry more weight than any variant on this page. Inherited risk raises their value rather than diminishing it.

Sources

Key references for the claims on this page. Where a figure is attributed to a specific study or body, it is named here.

1. Herskind AM, et al. The heritability of human longevity: a population-based study of 2872 Danish twin pairs. Human Genetics, 1996. The classical twin heritability estimate.
2. Ruby JG, Wright KM, Rand KA, et al. Estimates of the heritability of human longevity are substantially inflated due to assortative mating. Genetics, 2018;210(3):1109–1124. Source of the 7% upper-bound estimate. DOI · PMID 30401766
3. Willcox BJ, et al. FOXO3A genotype is strongly associated with human longevity. PNAS, 2008. The original FOXO3 finding, since replicated across populations.
4. Deelen J, et al. A meta-analysis of genome-wide association studies identifies multiple longevity genes. Nature Communications, 2019. The large-scale GWAS confirming APOE and only a small number of other loci.
5. Corder EH, et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science, 1993. The APOE4 dose-response relationship.
6. Kronenberg F, Mora S, Stroes ESG, et al. Lipoprotein(a) in atherosclerotic cardiovascular disease and aortic stenosis: a European Atherosclerosis Society consensus statement. European Heart Journal, 2022;43(39):3925. Source of the thresholds, the grey zone, the dose-response figures and the EPIC-Norfolk finding.
7. Qian J, Wolters FJ, Beiser A, et al. APOE-related risk of mild cognitive impairment and dementia for prevention trials: an analysis of four cohorts. PLoS Medicine, 2017. Source of the lifetime risk figures.

Frequently asked

Related entries