One of the clearest demonstrations of a rule that governs all of this: the more people they looked at, the smaller the effect got. It started at 1.31 and finished at 1.17 — and the second number is the honest one.
Bipolar disorder is a psychiatric illness of episodes — sustained periods of elevated mood and energy alternating with depression. It is among the most heritable common conditions in medicine: twin studies put it far above diabetes or heart disease.
Which makes what happened to this variant instructive.
A 2011 study followed a single position through three progressively larger samples.
The association is real; it survived a replication ten times the size of the discovery. But the effect fell by nearly half between the first look and the second.
It has a name — the winner's curse — and it is a property of how discovery works rather than a fault in this study.
A variant is flagged because it crossed a threshold in one sample. Crossing a threshold takes a real effect plus, usually, a bit of luck. The real effect stays when you look again; the luck does not. So first estimates are biased upward, and it is the replication that tells you the size.
Anyone reading genetic findings should expect this. A striking odds ratio in a first report is a starting number, not a measurement.
NCAN encodes neurocan, a glycoprotein in the extracellular matrix — the scaffolding between brain cells rather than a neurotransmitter or a receptor. It is involved in cell adhesion and migration, and in mice its expression is concentrated in cortical and hippocampal areas.
Those are regions independently implicated in bipolar disorder by imaging and neuropsychology, which is the sort of convergence that makes a statistical finding more believable.
These two facts sit together and confuse people, so it is worth stating plainly.
Bipolar disorder runs in families strongly. It is also polygenic to an extreme degree: the inheritance is spread across a very large number of positions, each contributing an odds ratio near 1.1. An odds ratio of 1.17 against a population risk of around 1 percent moves an individual almost not at all.
Heritable does not mean predictable from any one position, and this is the condition that shows it most clearly.
Bipolar disorder is diagnosed by a clinician from a history of episodes over time — which is difficult, often delayed, and not something a genotype accelerates. No variant here is used in diagnosis, in prognosis, or in choosing between lithium, anticonvulsants and antipsychotics.
If any of this is a live question for you or someone close to you, the useful move is a psychiatric assessment. It is the only thing on this page that changes an outcome.
The source. Cichon et al. (Am J Hum Genet 2011) investigated 499,494 autosomal and 12,484 X-chromosomal SNPs in 682 patients with bipolar disorder and 1,300 controls, tested the 48 most significant SNPs in 1,729 patients and 2,313 controls, and meta-analysed. rs1064395 in NCAN reached genome-wide significance in 2,411 patients and 3,613 controls (p = 3.02 × 10⁻⁸, OR 1.31). A second follow-up in independent samples totalling 6,030 patients and 31,749 controls replicated the finding (p = 2.74 × 10⁻⁴, OR 1.12); combined analysis gave p = 2.14 × 10⁻⁹, OR 1.17.
Effect size and the winner's curse. The decline from 1.31 to 1.12 between discovery and independent replication is the expected consequence of selecting a variant on its having exceeded a significance threshold in a finite sample: the estimate at discovery captures both the true effect and the sampling deviation that helped it cross. The combined OR of 1.17 remains the better estimate, and it is the one to quote.
NCAN. Neurocan is a chondroitin sulphate proteoglycan of the extracellular matrix implicated in cell adhesion and migration. The authors reported murine expression localised to cortical and hippocampal areas, regions previously implicated in bipolar disorder through neuropsychological, neuroimaging and post-mortem work.
Architecture. Bipolar disorder has among the highest twin-study heritability estimates of any common disorder while individual common variants carry odds ratios near 1.1, which is the signature of an extremely polygenic architecture. Polygenic scores for bipolar disorder discriminate at the group level and are not clinically actionable for an individual.
Clinical boundary. Diagnosis is clinical, resting on the longitudinal course of mood episodes, and is frequently delayed by years. No genotype contributes to diagnosis, to prediction of course, or to the selection of mood stabilisers or antipsychotics.
What a 23andMe/AncestryDNA export or raw VCF can and can't tell you about Bipolar Disorder comes down to these specific, well-studied positions — not a diagnosis.
Because the first number included luck and the second did not. A variant gets flagged for crossing a threshold, and crossing takes a real effect plus, usually, a favourable sample. The effect stays when you look again; the luck does not. It is called the winner's curse and it happens to nearly every first report. The combined 1.17 is the number to quote.
Because the inheritance is spread very thin. A great many positions each add a little, and no single one carries much. Heritable and predictable-from-one-position are different claims, and bipolar disorder is the clearest case of the gap between them.
No. It is diagnosed from the course of mood episodes over time, by a clinician. Diagnosis is often delayed by years, which is a real problem, and no genotype shortens it. An assessment does.
It is less strange than it looks. Neurocan sits in the matrix between brain cells and is involved in how they adhere and migrate, and in mice its expression concentrates in cortical and hippocampal regions — the same regions imaging and neuropsychology had already implicated.
Free to reuse. This page's text is original writing from freely-available research, licensed CC BY 4.0 — reuse it, including commercially, with attribution to MyGeneLog. It's general research-derived information, not medical advice or a diagnosis — see Terms of Use.