The research shows a mutant protein named tau is poisoning brain cells, and that blocking its production may allow some of those sick neurons to recover. It worked in demented mice who, to the scientists' surprise, regained memory fairly rapidly.

The work is years away from being useful in people. There are no drugs yet to block tau, and most of the recent search for Alzheimer's treatments has focused instead on another protein, called beta-amyloid.

But Thursday's report, published in the journal Science, is sure to refocus attention on finding ways to attack this second culprit, too.

"There basically are two prongs and we need to deal with both," said lead researcher Karen Ashe, a University of Minnesota neurologist. "What we're showing is that there are neurons which are affected (by Alzheimer's) but not dead."

It's important research because it bolsters the notion of targeting those sick neurons in hopes of one day reversing at least some of dementia's damage, said William Thies, scientific director of the Alzheimer's Association. Today's Alzheimer's drugs only treat symptoms.

"If you can actually rescue some of these sick cells, that really brings the possibility of return of some function, which would be of tremendous value," he said.

No one knows exactly what causes Alzheimer's, a creeping brain degeneration that afflicts about 4.5 million Americans and is on the rise as the population ages.

The leading theory is that something spurs abnormal production of beta-amyloid, which forms sticky clumps that coat brain cells and kill them -- plaque that is the disease's hallmark. But tau clearly plays some role: A mutant form of this protein forms fibrous tangles in brain cells of Alzheimer's patients, and tau seems to be primarily responsible for another form of dementia.

To see if the tangles themselves are a cause or symptom of dementia, Ashe and colleagues specially engineered a mouse to mimic the kind of tau-and-tangle formation seen in Alzheimer's patients' brains.

Sure enough, as the rodents aged, more tangles built up and more brain cells died -- and the mice showed dramatic memory loss.

How could they tell? Mice don't like water and thus quickly learned how to swim out of a water maze. But as they became demented, it took longer to get out of the water until eventually the mice just swam aimlessly.

The mice were bred so that eating a certain antibiotic would switch off a gene responsible for producing the bad tau.

Here's the first surprise: As tau production plummeted, the rodents' memory loss didn't just stop -- they regained some memory. It wasn't a full recovery -- dead brain cells can't be brought back -- but after repeated retesting to confirm the results, Ashe concluded that memory function improved to about half the predemented state. Also, neuron death stopped.

The second surprise: Those fibrous tangles continued to form even as the mice got better.

That suggests the tangles aren't killing brain cells, but the mutant tau itself is. Perhaps the tangles form as the brain tries to fight off poisonous tau by sequestering it, Ashe said.

Studies using mice who overproduce that other Alzheimer's culprit, beta-amyloid, also have suggested that blocking that protein might reverse memory loss, with or without getting rid of the accompanying amyloid plaques, Thies noted. That's why numerous drug companies are hunting medicines to target amyloid production.

Similar efforts to block tau have lagged because until now there hasn't been a good animal model of tau-caused dementia to test, he said.

That now is likely to change, and Ashe foresees one day attacking both proteins simultaneously.

She adds a caution: Don't think antibiotics are the key -- Ashe just happened to use one to switch off a gene specially bred into these mice.