Research
Why cutting a normal brain protein can prevent prion disease symptoms in animals
EVERY MAMMAL makes a small protein that sits on the outer surface of its brain cells, and makes it continuously throughout life. Healthy animals and healthy people carry it, and animals bred without it grow up and behave much like their littermates. In a group of rare and fatal brain diseases, some copies of that protein fold into an abnormal shape, and a copy in the abnormal shape causes normal copies that touch it to take the same shape. The abnormal copies accumulate, brain cells die, and the illness advances over months or years. The protein at the center of this is the prion protein, and the diseases are named after it.
That origin story has an unusual consequence. The damaging material is not an invading organism with its own genes. It is made out of a protein the body supplies. A sheep or goat version of the illness, called scrapie, a cattle version, and the human versions all depend on the same supply. If the supply stops, there is nothing left to convert. If the supply is merely reduced, there is less raw material, and the conversion should proceed more slowly.
Those two predictions can be separated, and animals have now tested both. The clean test of the first came from a herd of Norwegian dairy goats that carry a natural change in the gene for the protein. The change inserts a stop signal early in the gene's instructions, so cells reading those instructions halt before the protein is built. Goats with the change on both copies of the gene make none of the protein at all. Goats with it on one copy make less. Since these animals occur naturally, no laboratory engineering is involved in producing them.
Øyvind Salvesen, Cecilie Ersdal and colleagues at the Norwegian University of Life Sciences, with coworkers at the Norwegian Veterinary Institute and at a veterinary school in Toulouse, published the resulting experiment in Veterinary Research in 2020. They took 12 female Norwegian dairy goats, four of each of the three genotypes, and injected 400 microliters of a brain preparation from a goat that had died of natural scrapie directly into each animal's midbrain. That route skips the gut and lymph tissue that an orally acquired infection passes through first, so it is a severe challenge rather than a realistic one.
The four goats with two working copies of the gene fell ill first. Clinical signs appeared at a mean of 601 days after inoculation, with a standard deviation of 14 days, and the group was euthanized at 615 days. The four goats with one working copy fell ill later, at a mean of 773 days with a standard deviation of 46 days, which the authors report as a 29% longer incubation. Their illness also unfolded more gradually once it began, over 136 days against 14 days in the animals with two working copies. Measured in hippocampal tissue, the gene's activity in the one-copy goats ran at 54% of the level in the two-copy goats. Half the normal supply, in other words, bought those four animals roughly a third more time and a slower decline, and none of them escaped.
The four goats that made none of the protein were a different matter. They stayed healthy, with no clinical signs of prion disease, until they were euthanized at 1,260 days, more than twice the incubation time of the animals with two working copies. Examined afterward by tissue staining, by an enzyme immunoassay, and by a laboratory test that measures whether a sample can seed new misfolding, their brains and peripheral tissues showed no abnormal protein. Four animals per group is a small experiment, and the authors describe it as a proof-of-principle study. The direction of the result, though, matches what engineered mice lacking the protein had already shown, in a species that catches scrapie on its own.
This sets up the question that matters for treatment. Having none of the protein prevents the disease, and having half of it only delays the disease. A drug cannot remove a protein completely from a living brain. So how far down does it have to push the supply, and does the cut still work once an animal is already infected?
A preprint posted on bioRxiv on November 12th, 2026, is the most direct attempt yet to answer that. Juliana E. Gentile is the first author, and the corresponding investigators are Sonia M. Vallabh and Eric Vallabh Minikel, whom the paper lists at the Broad Institute of MIT and Harvard and at Massachusetts General Hospital. The work has not yet been reviewed by other scientists, which is worth keeping in mind for every figure that follows.
Their tool is a short piece of double-stranded RNA designed to match the message that carries the protein's instructions from gene to factory. Cells have machinery that recognizes such a match and destroys the message, so less of the protein gets built. Short RNAs used this way are siRNA molecules. This one carries a fatty chemical tag that helps it get into brain cells, and it was injected into the fluid-filled spaces inside the brains of mice. The researchers describe it as a tool for experiments rather than a candidate medicine, and say its tolerability is imperfect.
Measuring the result required a better assay than the team's earlier one, which lost accuracy below about a fifth of normal protein levels. The new version responds reliably down to 5% of normal. With two loading doses 30 days apart, uninfected mice were down to 11.7% of normal protein in whole brain hemispheres by day 60, and a further dose at day 120 kept them at 14.1% by day 150. The share of the protein still present after treatment is its residual level, and the authors round theirs to about 12%. They also caution that the assay may not count every pool of the protein in the brain.
Then they infected mice, by injection into the brain, and began treatment before any symptoms appeared. In the first experiment the animals received RML prions, a long-standing laboratory preparation, and dosing began 60 days after infection. Eight mice entered that arm. Two were censored after surgical complications, two were found dead of unknown causes with no obvious preceding symptoms, and one was euthanized because of dermatitis. The remaining three reached the study's pre-specified end at 500 days after infection with no prion disease symptoms, active and well groomed. For comparison, mice given artificial cerebrospinal fluid reached a median survival of 165 days, and mice given a non-targeting control RNA 156 days. Three animals is a thin result, and the losses, though unrelated to prion symptoms, took most of the arm with them.
A second experiment used Fukuoka-1 prions. Different preparations of misfolded protein behave as distinct and reproducible agents, each with its own speed and pattern of damage, and each is a prion strain. Fukuoka-1 had been the harder of the two to influence with modest lowering. Untreated mice given it survived to a median of 181 days, and untreated mice with one disrupted copy of the gene to 273 days, 51% longer. Of 11 mice treated from 56 days after infection, 2 were censored for surgical complications and 1 was euthanized for dermatitis, and the remaining 8 reached 500 days without prion disease symptoms, gaining weight and building nests normally. On the authors' reading, the two strains replicate at indistinguishable rates and differ instead in how much damage they do, and a deep enough cut handles both.
Timing turned out to matter as much as depth. In mice whose weight gain and nest-building had already measurably declined, treatment starting at 126 days after infection extended survival significantly, whether given as a single dose or repeatedly. It did not undo what had happened. Some groups regained a little weight, nest-building did not recover, and the authors state that reversal of existing symptoms was not seen in any group. Starting later still, at 140 days, when the paper notes the animals had about 21 days left to live, repeated dosing did not significantly extend survival beyond the 161-day median of controls. The explanation the authors offer is mechanical. The protein already present turns over with a half-life of about five days, so roughly a month passes before an RNA drug reaches its full effect on protein levels, and an animal three weeks from death does not have that month.
The infectious material itself was reduced, not removed. Using the seeding test, terminally ill animals averaged 10.6 log10 seeding units per gram of brain. Early-treated RML mice that reached 500 days averaged 8.5, about 100-fold lower, and early-treated Fukuoka-1 survivors averaged 9.1, about 10-fold lower than their terminal controls. Mice treated later, from 126 days, that survived to 500 days averaged 9.1. Those are large reductions and they are not zero. The authors draw the cautious conclusion. If some level of the protein exists below which a brain clears prions it has already made, it lies under about 12%, and they raise the possibility that no such level above zero exists at all, with the steady-state amount of infectious material simply tracking how much protein remains.
Several other boundaries belong with these numbers. The monitoring was designed to catch obvious prion disease symptoms rather than subtle ones, so the healthy-looking survivors cannot be called symptom-free in a fine-grained sense. The strains were mouse-adapted rather than human, a deliberate choice the authors explain by the deaths of researchers who acquired prion disease at work. The seeding test is not the older and more demanding method of injecting material into fresh animals to measure infectivity.
The study also includes primates, and they were not infected. Cynomolgus macaques received a single spinal injection of a version of the RNA aimed at the human gene. Protein in their spinal fluid fell to 43% of baseline at the lower dose and 22% at the higher one by 84 days, and at autopsy the higher dose had cut protein to 6% in the prefrontal cortex, 4% in the hippocampus, 21% in the thalamus and 18% in the hypothalamus. That is evidence the approach reaches a larger brain, and it is not evidence of benefit, because these animals had no disease to be protected from. The authors note that the lowering achieved in the monkey brain was not as deep as in mice, and that not all brain regions were examined.
The commercial and legal context is stated in the paper. Fourteen of its authors are employees and shareholders of Alnylam Pharmaceuticals, and four are employees and shareholders of Regeneron Pharmaceuticals. Gentile, Minikel and Vallabh are listed as co-inventors on a patent application filed by UMass Chan Medical School covering siRNA molecules for prion disease, and other authors are co-inventors on two further applications. The work was funded by two National Institutes of Health grants, the Prion Alliance and a Broad Institute fund.
What the goats and the mice together support is a statement about quantity and timing rather than about a cure. No supply of the protein means no disease. Half the supply means a slower disease that still arrives. About an eighth of the supply, started before symptoms, kept some infected mice outwardly well for 500 days while leaving measurable infectious material in their brains, and the same treatment given to animals already declining lengthened their lives without returning them to health. Lowering this protein is already being tested in people by two other methods, an antisense oligonucleotide and a different siRNA design, as the authors note. Whether a human brain can be pushed as deep as a mouse brain, and at what dose and how often, is what they say remains to be determined in those clinical studies.