Skip to main content
Premium Trial:

Request an Annual Quote

Statistical Tool Helps Select Antibiotic Regimens to Reverse Bacterial Drug Resistance


NEW YORK (GenomeWeb) – Researchers from the University of California, Merced and American University have developed a statistical tool called Time Machine that is able to suggest candidate antibiotic treatment regimens that can reverse the evolution of resistance-causing mutations in bacteria.

The researchers believe that their work could have major implications for doctors that are attempting to keep patient infections at bay by using 'antibiotic cycling,' an approach in which a handful of different antibiotics are used on a rotating basis to treat infections to reduce the risk of the bacteria becoming resistant to treatment.

In their paper, which was published this week in PlOS One, the investigators described the statistical techniques that they used to explore combinations of 15 common antibiotics and identify optimal regimens that could coax drug resistance mutations in gram-negative bacteria to revert to the wild-type, making them more responsive to existing treatments. 

Drug resistance is inevitable with long-term drug use. This is especially true for antibiotics and is particularly pertinent in hospital settings where lots of antibiotics are routinely used in patient care. Managing resistance in any drug environment is extremely difficult because, among other reasons, bacteria evolve and adapt quickly to changes in their environments, picking up new mutations in the process that can render them highly resistant to existing treatments. Moreover, bacteria are able to share these resistant genes across species, further fueling the spread of antibiotic resistance.

To compensate for bacterial evolution, a doctor fighting infections in an intensive care unit, for example, may reduce, rotate, or discontinue different antibiotics to get them to be effective in the short term. However, in current practice, "doctors don't take an ordered approach when they rotate antibiotics," Miriam Barlow, an associate professor at UC Merced and a co-author on the paper, said in a statement. "Our goal was to find a precise, ordered schedule of antibiotics that doctors could rely on and know that in the end, resistance will be reversed, and an antibiotic will work," she said.

Elaborating on this, Barlow told GenomeWeb this week that they also hoped to help some of the more "human friendly" antibiotics, which have been drained of their efficacy by repeated heavy use over time,  become more reliable treatments again.

Barlow first became interested in exploring methods of reversing antibiotic resistance about 12 years ago, when she was studying mechanisms that actually cause such resistance. She noticed that as new resistance developed in the bacteria, there were tradeoffs in some of the old resistance phenotypes suggesting that it might be possible to develop antibiotic cycling regimens that could capitalize on those tradeoffs and help address the drug resistance problem in bacteria, she told GenomeWeb. She then spent the next several years further investigating the feasibility of the task as well as designing appropriate experimental assays.

For the recently published study, the researchers looked at 16 genotypes of TEM beta-lactamases, a class of enzymes commonly found in gram-negative bacteria and responsible for resistance to various antibiotics. These genotypes varied from the wild type genotype — TEM 1 — by different combinations of four specific amino acid substitutions which are known to confer resistance advantages to antibiotics such as penicillins and cephalosporins. The goal, according to the paper, was to find treatments that would favor the reversal of the substitutions within the evolved TEM genotype, thus increasing the number of potential treatment options. More specifically, the researchers aimed to use selective pressures to return TEM-genotypes to the TEM-1 wild-type state, as it was first observed in 1963.

The researchers measured the growth rates of each test genotype when treated with each of 15 β-lactam antibiotics and then used these to calculate the probability of each amino acid substitution for each treatment using correlated probability and equal probability statistical models. For each of the 15 treatments tested in the study, they searched for substitution paths that led from each of the16 genotypes back to the wild type TEM-1, identifying optimized treatment paths that had the highest probabilities of "electing for reversions of amino acid substitutions and returning TEM to the wild type state.

Essentially this is a simple optimization problem with a simple strategy, Kristina Crona, a co-author on the paper and an assistant professor in American University's department of mathematics and statistics, explained to GenomeWeb. Basically, she added, the goal is to test all possible combinations of drugs and mark which ones have the highest probability of returning the mutated gene to wild type.

The starting point is to measure the degree of resistance of each type of bacteria to each antibiotic in the proposed test cohort. Those fitness measurements serve as the input to the algorithm which then computes the probability that a given combination of two or more treatments — in this case up to six drugs before the combinatorial space gets too large — is likely to reverse the mutation. The idea is to compute probabilities for all possible combinations of drugs in the cohort and see which work the best. "The final output is a sequence of drugs that is optimal in the sense that the probability to return to the wild type is maximal," she said.

For their next steps, Barlow and her colleagues are developing a microfluidic chip that mimics a possible hospital environment, providing an opportunity to test the antibiotic cycles and predictions made using their method. As part of those efforts, she said, they'll also try to assess how much drug resistance happens in hospitals versus what may be coming into hospitals from surrounding communities to get a better sense of the scale on which this approach should be used, she said.

They are also working out the details of how the predicted treatment regimen cycles might actually work in real-world settings, she said. The way this is envisioned to work is that a doctor might prescribe a sequence of four antibiotics for two or three months and then change to a different set of treatments for another few months, and then — depending on the infection in question — potentially change to a new set of treatments or re-prescribe the first set of treatments. However the duration of each treatment regimen in the cycle would still need to be determined, she said.

"A further issue is that if new genotypes arise, the treatment plan may fail," the researchers wrote. As such "the inclusion of more resistance genes in this type of approach may aid in the creation of robust treatment plans that are effective even when unexpected genotypes arise."

There's also room for development on the algorithmic side. Finding optimal treatment plans becomes more challenging as the space of possible drug combinations grows, Crona noted. It's an opportunity for mathematicians interested in tackling biological problems to jump in and develop better algorithms that can provide optimal plans for larger numbers of antibiotics and different kinds of bacteria, she said.

While this paper focused specifically on antibiotic resistance in gram-negative bacteria, the approach can be used to come up with treatment cycles for gram-positive bacteria as well as to design HIV treatment plans, the researchers wrote in their paper.