The gospel according to clinicians has been to attack pathogens with as much killing power as possible. In the case of bugs resistant to antimicrobial drugs, they bring out the big guns, combining drugs that together are more powerful than the sum of their individual effects.
Counterintuitively, however, the better weapon in the long run may be combinations of drugs that actually decrease one another’s effectiveness. Antagonistic pairs, though they do not have the same immediate lethality as synergistic blends, are much better at both slowing and preventing the evolution of resistant strains, according to two studies from the lab of Roy Kishony, HMS associate professor of systems biology.
“It’s a revolutionary concept,” said Robert Moellering, the Shields Warren-Mallinckrodt professor of medical research at HMS and Beth Israel Deaconess Medical Center. On the basis of this work, “we need to re-look at the gospel in the community.”
The inspiration for Kishony’s studies came from his lab’s earlier work showing that certain antagonistic drug combinations selected against resistance (see Focus, April 20, 2007). The research, which hinted at a broader principle, also suggested a fundamental approximation called the scaling assumption. This simplifying hypothesis suggests that a mutation that confers resistance to a drug is equivalent to decreasing the concentration of that drug.
“Adopting this assumption connects the effect of resistant mutations to the interaction between the drugs,” said Kishony. For example, for synergistic drugs, which enhance one another in combination, a mutation that effectively decreases the concentration of one drug also decreases the synergistic effect.
This assumption enabled two teams in Kishony’s lab to mathematically characterize their experimental observations. As a result, the teams devised two distinct models of how drug combinations affect the evolution of resistance in bacteria.
One group, which included the first author, doctoral candidate Matt Hegreness, and postdoctoral fellow Noam Shoresh, monitored Escheri-chia coli populations over time to see how different drug combinations would affect the rate at which resistance evolves. He used a novel robotic system that allowed him to grow hundreds of populations in parallel for approximately 170 generations. Without this “technological tour de force,” said co-author Daniel Hartl, Harvard professor of biology, this study would have been impossible.
Hegreness grew wild-type E. coli cells in 96-well grids, varying the concentrations of two drugs across the plate. In one plate, he added doxycycline and ciprofloxacin, which are strongly antagonistic; in another, doxycycline and erythromycin, which are strongly synergistic. He also tested combinations of antibiotics with degrees of synergy that vary depending on each drug’s concentration.
Shoresh analyzed the resulting growth data. He calculated the rate of adaptation by periodically measuring the optical density of the fluorescently labeled bacteria to determine the growth rate and then tracking the change in that growth rate over time. Those populations that were not immediately wiped out by the drugs ended the 15 days of the experiment growing at a rate close to that of the wild-type strain thriving in drug-free conditions.
The researchers found that the more synergistic the drug pairing, the more quickly the bacteria evolved resistance. “There’s an assumption that synergy is better,” said Hegreness, that if synergistic drugs inhibit growth more, they must also inhibit resistance more. But, as the scaling assumption suggests, drug interactions are more complicated than that. For antagonistic drugs, “when you get resistance to one drug, you remove the antagonism of the other drug, which is worse for the bacteria,” he said.
Based on their observations, the team proposed a qualitative model of how those drug–drug interactions influence evolution. The model, appearing online Aug. 25 in Proceedings of the National Academy of Sciences, “is trying to capture the simplest defining features” of different drug combinations as a way “to guide intuition,” said Shoresh.
The other team, which included HMS postdoctoral fellow Pamela Yeh and doctoral candidate Jean-Baptiste Michel, worked up a more quantitative model. Their model predicts the probability that a resistant mutation will occur in a population depending on the concentrations of drugs in combination.
The model takes a measure called the mutant selection window (MSW), which has been studied extensively for single drugs, and expands it to drug pairs. The MSW is a range of doses that, at its lowest, kills all wild-type bacteria, and at its highest, essentially destroys all mutants. One goal in therapeutics is to shrink this window, thereby shrinking the likelihood that resistance will evolve.
Yeh and Michel based their model on the results of experiments with Staphylococcus aureus. They placed billions of bacteria cells derived from a sensitive strain onto agar plates containing two drugs with concentrations varying over 11 levels. Then they allowed the bacteria to grow for five days. Throughout, they used a bank of office scanners to capture images and an image-processing algorithm to count surviving colonies, each one growing because of a spontaneous mutation that confers resistance to the drug pair. From this data, they calculated the frequency of resistance for each drug pair.
Their resulting model predicts that synergistic drug pairs widen the MSW and, in turn, favor the evolution of resistance. While more research is required before translating this finding to the clinic, said co-author Moellering, “it will make people take another look at the combinations they use.”
Other scientists can use this model, which is described in the Aug. 18 PNAS online edition. They can predict the frequencies of resistance for any drug pair by measuring a small number of frequencies and fitting the data to the model, said Michel. Such an approach could be used to predict the evolution of resistance in Mycobacterium tuberculosis, a pathogen that is more and more often being treated with different drug combinations. Because tuberculosis tends to gain resistance through spontaneous mutations, said Yeh, this model is particularly relevant.
While both models have their limitations—the first does not account for cross-resistance, such as the evolution of efflux pumps; the second does not factor in horizontal gene transfer; and neither considers the interactions that influence bacterial growth and drug metabolism in a living body—the models and experimental data together suggest that it may be time for a new attitude: “Antagonism between drugs, which normally is avoided in clinical use, should actually get more attention,” said Kishony.
Conflict Disclosure: The authors declare no conflicts of interest.
Funding Sources: The National Institutes of Health
