APOE rs429358 is one tagging SNP, not an Alzheimer’s forecast.

What APOE is doing in a person

APOE is a gene that makes a protein involved in moving cholesterol and other fats. You inherit two copies. The common versions are named ε2, ε3, and ε4. Those names are haplotypes, not personality types.

ε4 is the most cited genetic association with late-onset Alzheimer’s disease. Association is a population fact: in groups, the allele is more common among people who develop the disease. It is not a clock, a certainty, or a reason to treat a lookup as a diagnosis.

Most people who carry ε4 never get a simple story from it. Many people with Alzheimer’s do not carry ε4. Environment, age, other genes, and luck still do most of the work. A spit-kit row cannot replace a clinician.

• APOE: a lipid-transport gene on chromosome 19
• ε2, ε3, ε4: common haplotypes from two SNPs
• ε4: a risk allele in populations, not a personal forecast

• Genotype vs phenotype
• What is a gene

APOE rs429358 is one tagging SNP, not a haplotype

APOE rs429358 names a site in APOE. Common ε2, ε3, and ε4 haplotypes are combinations of rs429358 and rs7412. Querying only rs429358 is incomplete for haplotype calling.

The public record is a pair of dbSNP clusters plus whatever ClinVar and gnomAD say about those sites. Your file, if you have one, names the two bases at each site, or it does not list the site at all.

Association with late-onset Alzheimer’s disease risk is a population-level finding. It is not a determination that a person will or will not develop disease. genome.sh will not turn a C or a T into a care plan.

• rs429358: one of two SNPs that tag common APOE haplotypes
• rs7412: the other SNP; required for ε2 versus ε3 versus ε4
• Personal status: only from called genotypes, preferably local

• rs429358 in dbSNP
• How to look up an rsID

rs7412 and the e2, e3, e4 combinations

Text often writes epsilon 2, 3, and 4. Those labels are haplotypes, not single-SNP genotypes. Tables that decode ε4 from one letter skip strand, build, and rs7412.

Do not decode APOE from a blog table without checking the file’s genome build and alleles. Consumer exports mix GRCh37 positions with rsIDs. The rsID is the stable key. A raw chr:pos is not.

If rs7412 is missing from the chip, you cannot call the common haplotype honestly. Write not called for the missing site. Do not invent ε3 as a default.

If you must print a haplotype, print both rsIDs, both genotypes, the assembly, and the allele strand your file used. If any of those is missing, print not called instead of e4.

• rs7412 in dbSNP
• gnomAD allele frequency

Look up frequency and ClinVar without a blog decoder

Query both rsIDs. Read ClinVar clinical significance and review status as submitted labels. Risk-association language and Mendelian pathogenicity are different kinds of statements. Treat the live ClinVar record as the source.

gnomAD frequencies for these sites are not ultra-rare in many groups. Frequency is context, not a forecast. A common allele can still be a research risk factor. A rare allele is not automatically causal.

AlphaMissense, if present, is a missense prior. It does not replace haplotype calling and it does not replace ClinVar.

If ClinVar uses risk-factor language, print that language. Do not upgrade it to Pathogenic in a stylesheet. Do not drop gnomAD because the word risk appeared. The two databases are still answering different questions.

• ClinVar clinical significance
• gnomAD browser

Privacy: APOE status stays on disk

APOE status is sensitive. Prefer local lookup. Do not paste genotypes into a prompt, a random web form, or a group chat.

Identifier lookup of rs429358 can go over HTTPS because the rsID is public. The two letters you carry are not public. Keep the export on the machine that reads it.

The HTTP API rejects genome files on purpose. The in-browser importer parses in the tab. A coding agent should run genome query on disk, not in a cloud paste.

Logs and chat transcripts keep strings forever. An rsID in a URL is fine. A genotype string next to APOE in Slack is how a sensitive trait leaks. Prefer genome query on disk.

Query rs429358 and rs7412 with genome.sh

Install genome-sh, then a standard or full database if you want gnomAD next to ClinVar. Query both sites. JSON is for scripts.

GET /v1/query/rs429358 and GET /v1/gnomad/rs429358 are the hosted forms. No API key. No file.

If you annotate a VCF, genome annotate file.vcf.gz --format json still only sees called sites. Missing APOE probes stay not called.

cargo install genome-sh
genome db install standard
genome query rs429358
genome query rs7412 --format json
curl -s https://api.genome.sh/v1/query/rs429358 | jq .
curl -s https://api.genome.sh/v1/gnomad/rs429358 | jq .

Association is not a personal forecast

Population association is compatible with many people who carry ε4 never developing Alzheimer’s disease, and with people who do not carry it who do. Lifestyle, age, and other genes sit outside this pair of SNPs.

genome.sh will not estimate lifetime risk. It will not advise testing. That is a question for a clinician.

If you print a PDF, print the public records, the called genotypes, and the not-called sites. Leave the forecast out.

• Do print: rsIDs, ClinVar review text, gnomAD AF, called genotypes
• Do not print: a homemade ε4 decoder from one SNP
• Do not print: a prediction that a person will develop disease

Missing chip sites stay not called

rs429358 is often on consumer chips. Often is not always. rs7412 is the one people forget to check.

A missing marker is not wild type. A no-call is not ε3. Write not called and stop.

This lookup is informational. It is not a clinical Alzheimer’s test and not medical advice.

Older exports and newer chips disagree about which APOE probes exist. Always grep both rsIDs in the file you actually have. A blog that says 23andMe always calls rs429358 is describing one manifest, not yours.

genome query rs429358 --format json | jq .
genome query rs7412 --format json | jq .
grep -w rs429358 23andme.txt || echo not called
grep -w rs7412 23andme.txt || echo not called

What consumer reports get wrong about this SNP

Many consumer PDFs label e4 from rs429358 alone. That is incomplete. rs7412 can flip the haplotype. Strand flips and GRCh37 versus GRCh38 mix-ups flip it again.

Another failure mode is treating a population odds ratio as a personal probability. genome.sh will not compute that probability. It will print AF and ClinVar text.

If you build your own report, include a sentence that a consumer chip is not a clinical Alzheimer's test. Include not-called rows. Keep the file off the network.

genome query rs429358 --format json | jq '{rsid, clinvar, gnomad}'
genome query rs7412 --format json | jq '{rsid, clinvar, gnomad}'
curl -s https://api.genome.sh/v1/query/rs7412 | jq .

• Keep the file local
• ClinVar clinical significance

Questions