In 2012, Rebecca Vega Thurber looked at the results of the large underwater experiment she had been running for three years—and was disappointed.

Since 2009, her team had been traveling to the coral reefs of the Florida Keys. In some spots, they exposed the corals to nitrogen and phosphorus, to simulate the agricultural runoffs that often pollute these reefs. In other areas, they used wire mesh to keep fish away, mimicking the effects of overfishing. They wanted to know if these sources of stress disrupt the relationship between the corals and the trillions of microbes that live with them—and whether these disruptions lead to the corals’ demise.

Scientists have conducted hundreds of similar studies in humans. They compare healthy and sick people and look for differences in their microbiomes—the vast community of bacteria and other microbes that share our bodies. They aren’t looking for a specific disease-causing bug, like the ones behind classic infections like plague, leprosy, or tuberculosis. Instead, they’re looking for imbalances, where certain species rise to the fore, others slink into obscurity, and the entire community changes for the worse.

That’s what Vega Thurber expected to find in the corals. But that’s not what her postdoc Jesse Zaneveld found when he analyzed the results. The extra nutrients and the missing fish both changed the coral microbiomes—but not in any consistent ways. “It was a pretty dark day after three years of work,” says Vega Thurber. “A lot of students would have thrown up their hands and cried a bit. But Jesse said: You know what, I think I see something strange. It’s a pattern but one we didn’t predict.”

The microbiomes of the stressed corals had become more varied. They didn’t shift in any particular direction—they changed in every direction. And shortly after Zaneveld realized this, he spotted the same pattern—but this time in chimpanzees. Researchers at Yale University had studied the gut microbiomes of chimps that were infected with an HIV-like virus, and found that their microbiomes had also become more variable.

Zaneveld and Vega Thurber now think that this trend applies to all kinds of microbiomes, whether in corals, chimps, or humans. All of these hosts use a range of tactics to control which species and strains get to share their bodies. When the hosts are stressed or diseased, their control breaks down, and their microbiomes start to change. But they change randomly, rather than predictably. They don’t shift to any one specific unhealthy state. Rather, they veer off in unpredictable directions and enter a wide range of new states.

Since each disrupted microbiome is its own unique little snowflake, “we were originally going to call this the snowflake hypothesis,” says Vega Thurber, “but there’s a lot of negative connotation around that word now.” So, instead, Zaneveld has called the idea the Anna Karenina Hypothesis, after the opening lines of Tolstoy’s novel: “All happy families are alike; each unhappy family is unhappy in its own way.” Similarly, every unhealthy microbiome is unhealthy in its own way.

The hypothesis is something of a Grand Unified Theory of Unhealthy Microbiomes. “We started in corals but we started seeing the same pattern in human studies,” says Vega Thurber. “Now it’s something I can’t unsee. Whenever I see a microbiome paper that compares healthy and sick individuals, this pattern jumps out.”

By scouring the data from past studies, she and Zaneveld have found Anna Karenina effects in the bodies of human alcoholics and cigarette smokers, mice that are fighting off gut infections, and sponges living in acidifying water. The immune system is critical for controlling the microbiome, so unsurprisingly, Anna Karenina effects can also be triggered by diseases where the immune system is suppressed (like HIV) or hyperactive (like type 1 diabetes, Crohn’s disease, and other autoimmune disorders).

“Have I missed this trend in my own data? Yes and no,” says Valerie McKenzie from the University of Colorado Boulder, who studies how the skin microbiomes of frogs react to a killer fungus. “We have definitely seen greater dispersion of the community following infection. I recall presenting that finding in a talk a while ago, but I didn't know how to explain it. Having read this paper, I will revisit that question.”

Not every past study conforms to this pattern. In several cases, scientists have found that stress or disease can change the microbiome in specific ways. But Vega Thurber says that such results often come from reductionist experiments, involving laboratory animals that are studied under tightly controlled conditions. And it’s the messiness of the real world that contributes to Anna Karenina patterns. If a host can’t control its microbiome, its body could be invaded by microbes that are in the environment, or that come in through food. Those potential invaders would vary significantly from one setting to another, which is why no disturbed microbiome changes in quite the same way as the others.

This might help to explain some puzzling discrepancies in microbiome research. For example, early studies showed that obese people differ from lean ones in the ratio of two major groups of gut bacteria—they had more Firmicutes and fewer Bacteroidetes. But later studies found that this “F/B ratio” is not consistently linked to obesity—and nor is anything else. The gut microbiomes of obese and lean individuals look different in any one study, but none of these differences apply across all studies. That’s always seemed a little weird, but if the Anna Karenina hypothesis is right, it’s exactly what you’d expect.

“It is a good time to make this point,” says Katherine Pollard from the Gladstone Institutes, who led a study that reanalyzed the obesity data. She notes that microbiome studies are starting to track volunteers over time, and they’re finally getting big enough that they can reliably detect Anna Karenina effects. She now wants to know if the concept applies beyond the species of microbes in an individual’s body. “Are the absolute levels of microbes more variable?” she wonders. “What about levels of genes and the chemicals/biomolecules the microbiota produce?” And are changes in stressed microbiomes truly random, adds McKenzie, or are these traits that would help us predict how they’re likely to shift?

All of this has important implications for microbiome research. Many scientists are looking for ways of treating diseases—a huge list of them, including allergies, diabetes, inflammatory bowel disease, cancer, asthma, and more—by changing the microbiome, whether by designing better probiotics, or by transplanting entire communities from healthy donors into sick recipients. But if the microbiome is ruled by randomness, then it might be hard to determine whether a particular community is unhealthy, and to develop standardized, effective ways of steering it back on course.

Zaneveld and Vega Thurber’s idea “suggests that microbiomes can’t be classified as simply ‘good’ versus ‘bad,’” says Maureen Coleman from the University of Chicago. “And it suggests that a one-size-fits-all approach is unlikely to work when we try to manipulate our microbiomes.”

“It’s also important to remember that there is a lot of variability between healthy microbiomes,” adds Pollard. People’s microbiomes vary far more than their genomes do, and we don’t really know why. Two big studies, which looked at hundreds of factors like diet, medical history, age, and weight, could only explain between 8 and 19 percent of the variation in the human microbiome. Which means, as I reported last year, we’re still largely in the dark about what makes my microbiome different from yours, let alone whether one is healthier than the other.

As Pollard says, “We shouldn’t assume that there is just one way for our microbes to be a happy family.”

Related Video