"I found out I carry two copies of the Alzheimer's gene."

0.02%.

That’s the fraction of my genome that 23andMe actually sequenced. Less than a fifth of one percent. Out of 3 billion base pairs in the human genome, that little test looks at roughly 700,000 pre-selected positions and calls it a day.

And yet — somewhere inside that 0.02%, I found something that changed how I think about the next decade of my life.

---

What I found

Two years ago, I took the 23andMe Health + Ancestry test. I was already deep into menopause research, already paying close attention to hormones and brain health, already suspicious that the Women’s Health Initiative had done a generation of women a profound disservice by scaring us off estrogen for 20 years. I added the health reports mostly out of curiosity.

My APOE result came back: ε4/ε4. Homozygous. Two copies of the APOE4 allele — one from each parent.

If you haven’t heard of APOE4, here’s the plain English version: it’s a gene variant associated with Alzheimer’s disease. Most people have two copies of APOE3, which is basically neutral. About 22–25% of the population carries one copy of APOE4, which raises lifetime Alzheimer’s risk to roughly 25–30%. About 2–3% of people — like me — carry two copies. That’s APOE4 homozygous, and the risk profile is in a different category entirely.

In May 2024, a landmark paper in Nature Medicine did something the field had been moving toward for a while: it reclassified APOE4 homozygosity as a distinct genetic form of Alzheimer’s disease — not just a risk factor, but an actual cause, analogous to the rare early-onset familial Alzheimer’s genes. The findings were stark: by age 65, more than 95% of APOE4/4 carriers show abnormal amyloid accumulation in their cerebrospinal fluid. Average age of symptom onset: about 65. That’s 7–10 years earlier than non-carriers across every milestone.

Here’s what I need you to understand about that 95% figure — because mainstream media got this badly wrong, and I spent two days in a dark place because of it.

What the paper actually found is that 95%+ of APOE4/4 carriers will have elevated amyloid levels in their cerebrospinal fluid by age 65. That is a biomarker finding. Amyloid accumulation is a measurable biological sign — a precursor process that happens in the brain, sometimes for 10 to 20 years, before any cognitive symptoms appear. And a meaningful percentage of people who accumulate amyloid never develop clinical dementia at all. “Biomarker positive” is not the same as “you will get Alzheimer’s.”

The paper reclassified APOE4 homozygosity as a distinct genetic form of Alzheimer’s because of the near-universal biomarker pattern — not because 95% of carriers will show symptoms. That distinction is everything. But most of the mainstream headlines collapsed it. They reported it as: APOE4/4 carriers are virtually certain to develop Alzheimer’s. That is not what the paper said.

I read those headlines. I panicked. I spent a day or two genuinely distressed — not the low-grade anxiety of elevated risk, but the kind of fear where you start doing math about how many good years you might have left. It was bad.

Then I went back to the actual paper. Not the summaries. The paper itself — because AI helped me find it, pull it up, and work through what the data actually showed rather than what someone else decided to tell me it showed. And when I read it carefully, I found the distinction that the headlines had dropped. I’m almost certainly accumulating amyloid. That’s serious. That’s real. But accumulating amyloid is not the same as being destined for dementia.

This is exactly why reading primary sources with AI help matters: the nuance gets lost in translation, and the stakes — when we’re talking about your own brain — are too high to outsource to a headline writer’s word count.

The actionable reframe: knowing about amyloid accumulation gives you a window to act. Estrogen timing matters — there’s a “critical window” hypothesis that estrogen has neuroprotective effects when started close to menopause, and that window closes. Lifestyle interventions with strong evidence (exercise, sleep, metabolic health) are not optional when you have this genotype. Clinical trials specifically for APOE4/4 carriers exist — Banner Alzheimer’s Institute runs several. And monitoring tools — blood tests for amyloid-beta ratios, PET scans — are becoming more accessible. The information isn’t a sentence. It’s a map.

I am 57.

My mother has Alzheimer’s. She lives in the guest house at my home in Ajijic, Mexico. She still knows I’m her daughter. She just can’t remember my name.

I’m not going to spend a lot of time in that paragraph, because this article is about action, not grief. But I want to be honest: getting that result wasn’t abstract. I know what this disease does. I’ve watched it for years, up close, in a person I love. And now I know I’m almost certainly on the same biological timeline if I don’t do everything I can to slow or interrupt it.

That’s what motivates everything that follows.

---

---

What 23andMe actually is — and what it’s not

Here’s what I wish I’d understood before I took that test: 23andMe is not really a sequencing test. It’s a genotyping array.

The difference matters. A genotyping array is like a checklist. Scientists identify certain positions in the genome that are known to vary between people — single nucleotide polymorphisms, or SNPs — and they build a chip that checks those specific spots. 23andMe’s chip looks at about 600,000–700,000 of these pre-selected positions. Out of 3 billion base pairs. That’s the 0.02%.

The Health Add-on doesn’t change this. It adds interpretation of those same positions — not more sequencing. You’re still looking at the same tiny window.

For ancestry and common risk variants, that’s often enough. Common variants are common precisely because they show up reliably in SNP arrays. The chip was designed to find them.

But there’s an enormous amount the chip can’t see:

• Rare variants that aren’t on the chip’s checklist at all
• Structural variants — large rearrangements and deletions that only sequencing can detect
• Copy number variants — gene duplications and deletions the chip misses entirely
• BRCA1/2 full mutation screening — 23andMe tests exactly 3 BRCA variants. The ones most common in people of Ashkenazi Jewish descent. There are over 1,000 known pathogenic BRCA variants. Someone of non-Ashkenazi descent with a different BRCA mutation gets a falsely reassuring “no variants detected.” This is not a small asterisk.
• Pharmacogenomics — how your body processes specific medications (relevant to anyone managing complex conditions, planning surgery, or potentially taking Alzheimer’s drugs in the future)

A whole genome sequence identifies over 900 times more known genetic variants than a 23andMe array. That’s not a marketing number — it’s the math of 700,000 positions versus 4–5 million variants per person at 30x coverage.

---

---

The chip problem — and why my APOE result might need a second look

Here’s something I learned after the fact that I think every 23andMe user with an APOE result should know.

Accurate APOE typing requires two SNPs in combination: rs429358 and rs7412. The combination of both tells you which alleles you actually have. But 23andMe’s older chip versions — V3 and V4 — only tested rs429358, not rs7412.

What this means practically: someone tested on a V3 or V4 chip could be reported as ε3/ε4 (heterozygous, one copy of APOE4) when they are actually ε4/ε4 (homozygous, two copies). The newer V5 chip reportedly includes both SNPs, but a lot of people tested before V5 rolled out.

This is not a small distinction. We’re talking about the difference between a 25–30% lifetime Alzheimer’s risk and what the 2024 Nature Medicine paper now calls a nearly certain biological cause.

I checked — I’m on V5. My APOE4/4 result is accurate. (I found it in my account settings under “Ancestry Features” — it shows your chip version and genotyping date. Mine: Version 5, April 5, 2018. V5 tests both rs429358 and rs7412, so the homozygous call is reliable.)

Reader action item: If you’ve tested with 23andMe and care about your APOE status, check which chip version you have (it’s in your account settings under “Ancestry Features” or “Raw Data”). If you’re on V3 or V4, your APOE report may not be accurate — specifically, you may have been reported as heterozygous when you’re actually homozygous. You can check your raw data directly for both rs429358 and rs7412 (there are guides online for how to do this), or verify via whole genome sequencing. The V5 chip includes both SNPs; the older chips do not.

---

AncestryDNA users: your APOE result doesn’t exist

If the 23andMe chip problem made you uncomfortable, this one is worse.

AncestryDNA uses the Illumina OmniExpress chip — about 730,000 SNPs. The chip technically includes a probe for rs429358, one of the two SNPs required for APOE typing. But here’s the thing: it has a systematic calling failure on that probe. It doesn’t produce an unreliable result. It doesn’t flag the position as ambiguous. It outputs T/T — homozygous reference — every single time, regardless of what your actual genotype is.

Read that again: regardless of what your actual genotype is.

That means if you have one copy of APOE4, AncestryDNA will tell you T/T. If you have two copies of APOE4, AncestryDNA will still tell you T/T. It will never — not once, not for any customer — return a result that suggests APOE4 is present.

This isn’t a mystery. The APOE4 community has documented this failure since 2018. It has not been fixed as of 2025. That’s more than 40 million AncestryDNA customers who cannot determine their APOE status from their raw data, full stop. It doesn’t matter which third-party tool you run it through — Promethease, SelfDecode, Genetic Genie, any of them. Garbage in, garbage out. The underlying data is wrong at the source.

To be absolutely clear about what you’re getting from each company:

• 23andMe pre-V5 problem: Could report ε3/ε4 (heterozygous — one copy of APOE4) when you’re actually ε4/ε4 (homozygous — two copies). It tells you something, but possibly the wrong degree of something. That’s a serious error, but at least the test is registering that APOE4 is present.
• AncestryDNA problem: Always reports ε3/ε3 (zero copies of APOE4) regardless of reality. You could be APOE4/4 — like me — and walk away thinking you have zero copies. That’s not a miscalibrated result. That’s complete silence, in the wrong direction.

23andMe V5 correctly tests both rs429358 AND rs7412 and includes APOE carrier status directly in their Health reports. Clinical APOE genotyping through LabCorp, Quest, or Invitae is also fully diagnostic and doesn’t depend on consumer chip arrays at all.

Reader action item: If you tested with AncestryDNA and care about your APOE status — or if you’ve run your AncestryDNA raw data through a third-party tool and received an “ε3/ε3” result — that result is not interpretable for this specific question. It is not evidence that you don’t carry APOE4. It is evidence that AncestryDNA’s chip cannot answer this question. If your APOE status matters to you (and at our age, it might), you need either 23andMe V5 or clinical genotyping. Don’t let a systematically broken probe give you false reassurance about something this important.

---

In Part 2, I explain what whole genome sequencing actually covers, why the pricing surprised me, how to think about genetic privacy after the 23andMe bankruptcy, and what I’m doing with all of this information.

[Read Part 2 → What Whole Genome Sequencing Actually Tells You (and What I’m Doing Next)]

— Annette

Annette Thompson is 57, the founder of adoption.com, and a menopause advocate writing about evidence-based women’s health.

---

P.S. — If you or someone you know is APOE4 positive and want community, the APOE4.info community at apoe4.info is a remarkable resource — researchers, carriers, and caregivers all in one place. Worth bookmarking.