Colistin is an antibiotic of last resort, one of the final options left when all other drugs fail. It is an older antibiotic and sometimes toxic to the kidneys. Yet precisely because colistin is not a particularly safe drug and thus rarely used, bacteria didn’t develop resistance to it.

Until they did, of course. At first, the occasional resistance mutation popped up, here and there. Then in 2015, scientists surveilling Chinese pig farms reported the discovery of Escherichia coli bacteria with colistin resistance in a form that can spread with frightening ease. The resistance gene, which they called mcr-1, lived on a free-floating loop of DNA called a plasmid. Bacteria—even bacteria of different species—can swap plasmids back and forth. Just seven months later, another group in Belgium found a second, similar gene, mcr-2. And this week, the original group of scientists reported a third, this time more distinct, colistin-resistance gene, mcr-3, also on a plasmid and also found in E. coli from a Chinese pig farm. That plasmid where the researchers found mcr-3 also contained 18 other resistance genes against other antibiotics.

Researchers think they have another reason to worry about mcr-3. This resistance gene is very similar to a naturally occurring gene in Aeromonas, a type of bacteria ubiquitous in fresh and brackish waters. It could be that mcr-3 developed in Aeronomas in the first place, and though it’s not yet confirmed, these bacteria may be a reservoir of colistin resistance. “mcr-3 might exist everywhere,” says Yang Wang, a biologist at China Agricultural University and an author on the new report. Colistin resistance could be even more widespread than we thought.

The case of mcr-3 illuminates the complex interplay between antibiotics and the natural environment, which scientists are only just beginning to understand. It makes sense to find resistance genes in hospitals or on farms, where antibiotics are used to treat humans or animals. But why would antibiotic resistance genes turn up in bacteria from the natural environment—even in an isolated cave or 30,000-year-old permafrost? Does the natural environment harbor a reservoir of antibiotic resistance genes, waiting to spring into action?

First, some history about colistin. Decades ago, as colistin fell out of favor for human medicine, farmers started using it. Small doses of antibiotics can fatten pigs and chickens, so colistin became a growth promoter added to feed. China has been a major user of colistin in agriculture, and it’s Chinese scientists who first detected the mcr-1 resistance. But the drug has also been used worldwide in various ways, from promoting growth to preventing and treating diarrheal diseases in animals. (This year, China banned the use of colistin as a growth promoter, and Europe is cutting down on its use in prevention. Treating sick animals with it is still allowed, though.)

Infections resistant to multiple drugs—the kind that might require resorting to colistin—are thankfully still rare, and they’re mostly a concern for the already sick and immunocompromised. But if colistin resistance becomes more common, these patients would lose one of their only remaining options.

So the bombshell discovery of mcr-1 set off a search through bacteria collections around the world. And soon enough, researchers found mcr-1 in dozens of countries, even in decades-old samples of Enterobacteriaceae, a group of bacteria that includes E. coli and Salmonella. The gene had spread around the world before scientists even knew to look for it.

Unfortunately, the same thing has probably already happened with mcr-3. In fact, when Wang and his coauthors went to compare the DNA sequence of mcr-3 to previously sequenced bacteria, they found three 100 percent matches—in Enterobacteriaceae from a Malaysian pig in 2013, human pus in Thailand in 2015, and human stool in the U.S. in 2008.

What makes mcr-3 different from mcr-1 is the existence of mcr-3-like genes in a whole different group of more distantly related water bacteria, the Aeromonas. For example, one bacteria sample from Malaysian lake water had 94.1 percent similarity to the enzyme encoded by mcr-3. And some Aeromonas species seem to have intrinsic resistance to colistin’s class of antibiotics. Wang is now working to isolate the mcr-3-like gene in Aeromonas to figure out if it is indeed what gives the bacteria resistance to colistin.

What could Aeromonas be doing with a colistin-resistance gene out in the environment? This kind of scenario is actually quite common. “Environmental bacteria are just chock full of resistance genes,” says Gerry Wright, a biochemist at McMaster University, who has looked in places like permafrost and isolated caves for resistance genes. One answer could be that bacteria are protecting themselves. Many antibiotics actually come from microbes, which may be creating toxins to fight off their microbe competitors. In fact, colistin was first isolated from a flask of bacteria in Japan.

But chemical warfare between bacteria isn’t the only way to think about the existence of natural antibiotics. The antibiotics we’ve isolated from microbes often don’t exist at high enough natural concentrations to actually kill other bacteria. Perhaps, what we think of as antibiotics are really signaling molecules that bacteria use to communicate with each other. “These are very complex ecosystems that have hundreds and thousands of species in some way communicating with each other,” says Justin Donato, a biochemist at University of St. Thomas. And what we think of resistance genes might just be there to modify signaling molecules.

Of course, in high enough concentrations, the compounds that may originally have been for signaling and that humans use as antibiotics, become lethal. Then natural selection kicks in. The genes used in bacterial communication could then be co-opted for antibiotic resistance. That may be what happened with mcr-3 and colistin. Perhaps a pig drank water with Aeromonas carrying mcr-3 or an mcr-3-like gene, which then encountered E. coli in the pig’s gut and eventually passed along the gene on a plasmid, giving the E. coli resistance to colistin.

In one view, the case of mcr-3 illustrates how little we know about complex bacterial communities, and how sources of antibiotic resistance may lurk in unexpected places. But seen another way, it shows just how predictable the larger pattern is: Bacteria always find a way to become resistant.