brugada.net · What’s broken
Updated 16 July 2026
That is genuinely the whole problem. The rest of this page is just detail about which clip, and how we know.
Your heart beats because cells pass electrical signals to each other. The signal starts when a tiny gate in the cell wall snaps open and lets sodium rush in. That gate is a protein called Nav1.5, and it is built from a blueprint gene called SCN5A.
A protein is a long chain that folds itself into a shape. The shape is the whole point — a misfolded protein is a machine assembled wrong, and the cell throws it away rather than installing it.
In one corner of Nav1.5 there is a small fold held together by what is essentially a clip. One part of the chain carries a positive charge; another carries a negative charge; they attract, and they snap together. The positive half is at position 104 — an amino acid called arginine. The negative half is at position 84.
The variant this site is about swaps that arginine for a glutamine, which is a perfectly nice amino acid that happens to be electrically neutral. The positive half of the clip is gone. The negative half is still there, buried in the middle of the protein, holding on to nothing.
The fold loosens. The protein doesn’t assemble properly. The cell doesn’t install it. Fewer gates reach the cell wall, so less sodium gets in, so the signal starts weakly — and a weak signal spreading through heart muscle is what Brugada syndrome looks like electrically. Written down, the variant is SCN5A p.Arg104Gln, or R104Q for short.
Worth being precise about one thing: the gate itself isn’t broken. The hole the sodium goes through is fine. What’s broken is a folding problem somewhere else in the protein, which means the finished gate never turns up. That distinction is the reason this project exists — you don’t fix a delivery problem with a better door.
Reasonably sure, and from a few directions that don’t depend on each other — which is the only kind of agreement worth anything.
Not infer it — see it. A real microscope image of the human protein (structure 8VYJ) shows those two charges sitting 3.79 Å apart, close enough to be gripping each other.
That arginine is the same in 7 out of 7 mammals checked. Positions that tolerate change, change. This one hasn’t, in millions of years.
A protein language model, asked what belongs at 104, answers arginine — then lysine. Both positive. It independently wants a positive charge there, without knowing why.
In cells, the variant produces roughly 29% of normal current. A 2013 lab result, not a prediction. This is the only line here from a real experiment.
The honest caveat. There are eleven computer predictors on file that all call this variant damaging, and it is tempting to describe that as eleven independent votes. It isn’t. Most of them are trained on overlapping data and partly on each other — they agree the way siblings agree. Counted honestly it is about three independent lines of evidence: conservation, a language model, and a stability calculation. Three is still three. It just isn’t eleven.
Structure. R104–D84 measures 2.65 Å in the minimised model and 3.79 Å in 8VYJ, an experimental cryo-EM structure. In the mutant model Q104–D84 opens to 3.21 Å — an H-bond at best, the ionic pair gone.
It’s a cluster, not a residue. propka3.5 puts D84 at pKa 4.39 and its neighbour D82 at 5.70 — both shifted, both buried, 5.1 Å apart. The lesion is a buried acidic cluster.
Stability. ThermoMPNN ΔΔG +1.19 kcal/mol. A rescuer has to offset roughly that.
The language model. ESM-2 log-likelihood ratio −3.57. Its top picks at position 104: R (−0.25), then K (−2.24) — both positive charges.
The correlated predictors. AlphaMissense 0.869 · REVEL 0.967 · CADD 28.1 · PolyPhen-2 0.998 · SIFT 0 · phyloP 7.86. Impressive-looking, but these share training data and features — treat them as roughly one line of evidence, not six.
Population. gnomAD v4: 5 / 1,461,406, allele frequency 3.4×10−6, zero homozygotes.
Function. ~0.29× wild-type current in oocytes; no measurable current in HEK293 cells (Gütter, Benndorf & Zimmer 2013, PMID 23805106). The one line here from a real experiment.
microenv.json · R104Q_dossier.md · r104q_for_pka.pka · thermompnn_ddg.json · esm_score.json · table_pred.csv · table_pop.csv
Feed that 29% into a mathematical model of a heart-muscle cell and the signal’s rising edge collapses by about 61% — while the cell’s recovery time barely changes, under 2%. In plain terms: the signal gets slower to spread, not longer to reset. That is a conduction problem, and it’s the fingerprint of Brugada syndrome, arrived at from the molecule rather than assumed at the start.
A model of a heart cell is a set of equations, not a heart. It tells you the shape of the consequence. It cannot tell you what any particular person’s heart does.
An O’Hara–Rudy ventricular myocyte model, driven at the measured ~29% sodium current.
Upstroke velocity (dV/dtmax, a conduction proxy) falls from ~315 to ~123 V/s — −61%. Action-potential duration moves <2%.
That asymmetry is the whole point: conduction collapses, repolarisation timing doesn’t. It’s a depolarisation/conduction disorder, consistent with loss of function, derived rather than assumed.
07_variant_metrics/digital_twin/
onwards.