The Centers for Disease Control and Prevention (CDC) keeps a Most Wanted list for flu viruses. The agency evaluates every potentially dangerous strain, and gives them two scores out of 10—one reflecting how likely they are to trigger a pandemic, and another that measures how bad that pandemic would be. At the top of the list, with scores of 6.5 for emergence and 7.5 for impact, is H7N9.

Influenza viruses come in many flavors—H5N1, H1N1, H3N2, and so on. The H and N refer to two proteins on their surface, and the numbers refer to the versions of those proteins that a particular virus carries. H1N1 was responsible for both the catastrophic pandemic of 1918 that killed millions of people, and the most recent (and much milder) one from 2009. H5N1 is the bird-flu subtype that has been worrying scientists for almost two decades. But H7N9? Until recently, it had flown under the radar.

H7 viruses infect birds, and only very rarely jump into humans. H7N9 in particular had never been known to infect humans at all before 2013, when it caused an unexpected epidemic in China. It was billed as low-pathogenic (or “low-path”) because it only caused mild disease in chickens. But in humans, the story was different: Of the 135 people infected, around a quarter died.

Every year since, there’s been a new epidemic, and the current one is the worst. H7N9 has evolved, acquiring mutations that allow other flu strains to reproduce more effectively in both birds and mammals. It has started killing birds. In one year, H7N9’s highly pathogenic (“high-path”) strains have caused as many human infections as the previous four epidemics put together. As of September 20, there have been 1,589 laboratory-confirmed cases, and 39 percent of those people have died. “It was a matter of time,” says the flu expert Yoshihiro Kawaoka, from the University of Wisconsin-Madison. “It wasn’t surprising to see this change.”

Kawaoka and his colleagues have now studied the new high-path strains collected from one of the people who died this year. They’ve shown that these strains reproduce efficiently in mice, ferrets, and monkeys, and cause more severe disease than their low-path ancestors. They can spread through the air between captive ferrets, and in some cases, kill the animals they land in. Perhaps most worrying, some strains have already evolved the ability to resist Tamiflu, a frontline drug that’s used to treat flu infections.

These are, of course, just animal studies, and they’re an imperfect reflection of how the high-path viruses behave in humans. “The little data available to date does not reveal an obvious increase in virulence for humans,” says Malik Peiris, from the University of Hong Kong, “but this is very difficult to assess because we only see the more severe infections who present to hospitals. This is an issue that needs to be closely monitored in the upcoming winter season.”

“When you compare H5 and H7 viruses, I think H7 are more worrisome,” says Kawaoka. That’s because the H5 viruses need several further mutations to spread between mammals, as Kawaoka showed in controversial lab experiments where he engineered strains with those mutations. But H7 strains apparently don’t need such tweaks. The strains that are out there right now are already capable of spreading between ferrets.

And yet, there’s no strong evidence that they’re hopping from person to person. Some of the cases this year have occurred in family groups, but it’s hard to say if they passed H7N9 between them or simply acquired it from the same birds. For now, the CDC still notes that “the risk to the general public is very low,” since most people who were infected had been in direct contact with birds, whether in poultry markets, vehicles, or their own homes.

“Clearly this is a virus that we don’t want to become any more transmissible between humans,” says Wendy Barclay, from Imperial College London. “But it’s not already transmissible enough to cause a pandemic—otherwise, we would have seen one.” She also notes that, in Kawaoka’s study, the high-path strains didn’t spread any more easily between ferrets than their low-path cousins. Even though this year’s epidemic is unprecedentedly big, the viruses don’t seem to be any more transmissible than when they first emerged in 2013.

There’s also a silver lining to the Tamiflu-resistant strains that Kawaoka identified. The mutation behind this resistance works by changing the shape of a protein on the virus’s surface—a protein that Tamiflu normally attacks. But the same protein is also part of the infection process; by changing its shape, the strains weaken themselves. They cause milder disease in both mice and ferrets (although they still spread with the same ease as the drug-sensitive strains).

That’s good news, but it’s no reason to rest on our laurels. In 1999, scientists discovered a mutation called H274Y that made H1N1 strains resistant to Tamiflu, but that also reduced their ability to infect mouse and ferrets. The scientists thought that this mutation was “unlikely to be of clinical consequence.” They were wrong. H1N1 picked up other mutations that compensated for H274Y, creating flu strains that were infective and resistant. By 2008, almost all the seasonal strains of H1N1 had become resistant to Tamiflu. With H7N9, history could well repeat itself.

But Tamiflu isn’t our only weapon against influenza. There’s an experimental new drug called Avigan (or favipiravir) that, rather than going after a surface protein, attacks an enzyme that the virus uses to copy its genetic material. Even Tamiflu-resistant strains of H7N9 fall to this drug, as do other kinds of flu that Kawaoka has looked at—at least in animals. “Whether that’s also the case in humans, we don’t know,” he says.

The viruses could eventually evolve to resist this new drug, too. But, Kawaoka says, “many people, including us, have looked for viruses that are resistant to favipiravir, and I don’t think anyone has found one yet.” And Barclay suggests that scientists should start running clinical trials that test both drugs together. “It still astonishes me that we continue to treat flu patients with a single drug when we know that the virus is highly mutable,” she says. “It’s almost inevitable that drug-resistant viruses can evolve.”

In the meantime, vaccines are being developed to match the viruses seen in the fifth and current epidemic. Other control measures have waxed and waned. When the first of the epidemics struck, Chinese health ministries closed markets and slaughtered birds. But as Helen Branswell reports in STAT, some of those containment efforts became more lax in 2015 and 2016.

Again, there is some good news: H7N9 infects chickens very well, but unlike H5N1, it seems to avoid ducks. That matters because Chinese ducks are often housed outside, and domestic birds can mingle with wild ones. Aboard ducks, bird flu can easily spread from one infected farm to other parts of the world. “That may be a major difference that may make it easier to control H7N9 compared to H5N1.”

It might also be a blessing in disguise that the high-path strains have emerged. The low-path strains were very hard to detect because they didn’t cause symptoms. But the high-path viruses kill infected birds, which means “they might be easier to eradicate from chickens since they can be more easily detected,” says Adolfo García-Sastre, from the Icahn School of Medicine at Mount Sinai in New York. “However, one would need a very well-organized eradication campaign to eliminate them from poultry before they spread to other areas beyond China. I’m afraid that this will not happen, since it did not happen with the H5N1 viruses, which were first detected in 1997, and finally disseminated to most of the rest of the world starting in 2003.”