Phagoburn trial a phage therapy bust; or, why kinetics matter

A hundred years of phage therapy has produced many claims for its utility. Almost none of these claims are backed up by compelling evidence. By “compelling” I mean randomized controlled trials, rather than anecdotes or case studies.

The problem with case studies in infectious diseases is that most patients get well on their own, even for refractory infections. We don’t do many placebo controlled trials these days for infectious diseases – withholding a known effective treatment (ie., antibiotics) is unethical. But scouring the historical literature, Brad Spellberg has  estimated cure rates for infections in the early antibiotic era. Among skin infections, cure rates for placebos were 66% for cellulitis, 36% for wounds and ulcers, 76% for major abscesses. Antibiotics improved these rates to 83-98%. The point is, spraying some phage on an infection and seeing it improve is no evidence of efficacy. There’s a good chance the patient would have gotten better anyway.

All this is in preface to the publication of disappointing results in the Phagoburn trial. This was a Phase I/II trial of safety and efficacy for burn wound patients suffering Pseudomonas aeruginosa infections. Pseudomonas is a common wound pathogen that often proves recalcitrant to antibiotic therapy due to its propensity for forming biofilms.

From Antimicrobial resistance in the respiratory microbiota of people with cystic fibrosis

Bacteriophage commonly express enzymes that break down biofilms, allowing them to attack and kill the bacteria within. These properties make them attractive candidates for treating persistent infections, either alone or as a complement to antibiotics.

The trial was designed to treat patients with burn wounds that were infected by Pseudomonas. One arm received a standard of care (sulfadiazene silver cream), the other received a cocktail of 12 anti-Pseudomonas bacteriophages soaked in a dressing. The primary endpoint (the pre-determined metric by which the treatments are evaluated) was a sustained reduction in wound bacterial loads.

The top line result is that phage treatment (PP1131) reduced bacterial loads, but at a much slower rate than the standard of care:

From Efficacy and tolerability of a cocktail of bacteriophages to treat burn wounds infected by Pseudomonas aeruginosa (PhagoBurn): a randomised, controlled, double-blind phase 1/2 trial.

It took about 45 hours for half the patients treated with sulfadiazene silver to see a reduction in bacterial load versus 145 hours for phage therapy. That’s a big difference, and not in the right direction. There was no placebo arm to compare whether phage are faster than no treatment at all.

Why the failure? The authors cite bacterial resistance to phage. They contend that the subset of phage responders differed from non-responders in that the bacteria infecting their wounds were susceptible to being killed by the phage in the cocktail. That is, that treatment failure occurred primarily because none of the phage in the cocktail effectively attacked the specific strains of bacteria in the wounds. I think this is at best only partially true, for reasons I’ll go into further below.

Simply put, I think the investigators blew it by not using enough phage. They didn’t because like most biologists (and nearly all clinicians) they don’t understand reaction kinetics in general or the kinetics of phage infection in particular. Biologists tend to be people who like science but hate math. We think in pictures, not numbers. As a result, we tend to be beguiled by pretty pictures – especially structures – and dismiss abstract quantitative evidence, no matter how rigorously argued.

One of my grad school mentors, Robert Thompson, would harangue me at length on this point. “Structures are all very nice” he would say in his round cheerful Lincolnshire accent, “but you never know if they mean anything. A helix, a cleft, an apparent bridging interaction – they could just be artifacts of the visualization process, or they could be real but irrelevant. But kinetics – how molecules interact in time – are always relevant. Living systems must move forward in time; when they stop, they die.”

Bacteriophage are much larger than other therapeutic agents and thus find their targets much more slowly. A typical antibiotic weighs the equivalent of 600 protons (a unit termed the Dalton). Antibodies are much larger, with a mass of about 180,000 Daltons, and they are very slow at finding their targets. Bacteriophage are a thousand times bigger than antibodies, a 100 million or more Daltons, and are correspondingly slower.

From Protein Diffusion in Living Skeletal Muscle Fibers: Dependence on Protein Size, Fiber Type, and Contraction

In the graph above, you can see that a typical drug molecule (10²-10³ Daltons) diffuses at least a hundred times faster than something the size of a bacteriophage (10⁸ Daltons). This effect is even more pronounced in tissues (filled symbols) than in solution (open symbols).

Bacteria are the targets of bacteriophage. When phage are in excess of bacteria, as they should be in any therapeutic scenario, they find their targets at a rate of 1 billionth milliliter per minute. That is, if you have a solution of 1 billion phage per mL (a high but typical amount for purified preparations), it will take about a minute for the bacteria in the solution to be attacked by the phage (10^-9 mL/min x 10⁹ phage/mL = 1/min). If you are at 1 million phage per mL, it takes about a thousand minutes (16.7 hours) to attack all the bacteria. That’s in ideal conditions, in a test tube. In a clinical environment, such as a wound, where plenty of things destroy or sequester or block entities like phage, the actual rate will be much lower. That’s why I am skeptical of claims that phage therapy cured an untreatable case of an Acinetobacter infection – there just wasn’t enough phage to do the job.

How many phage were applied in the Phagoburn study? The protocol was to soak the dressings with a phage solution of 1 million phage per mL. Per the rough calculations above, you can see that this is less than ideal, and possibly not enough phage to be effective. But in practice it was worse. Because of slow accrual rates, and because phage are subject to degradation over time, the amount of active phage applied was 100- to 1000-times lower than nominal for most patients . Starting with a marginally effective dose and cutting it by more than a hundred is a recipe for failure, and failure is just what they got.

Kinetics will not be mocked. It is always relevant, and it will not be ignored.

And ignore kinetics is just what the authors have done. They investigated resistance of the wound bacteria to the bacteriophage preparation via a semi-quantitative spot assay dilution series. Bacteria are spread out on a plate, and form a hazy “lawn” as they grow. Drops of phage are spotted on the lawn, and where they kill the bacteria, clear spots develop. Phage-resistant bacteria are not killed, and there is no spot on the lawn. Here is an example from the paper:

See the problem? An intermediately susceptible bug looks like a susceptible bug at low concentrations of phage (ie, top spots in the intermediate column look like lower spots in the susceptible column). One can convert an intermediate bug to a resistant bug, or a susceptible bug to an intermediate bug, just by lowering the phage concentration.

And that is what the researcher did, albeit unintentionally. They started low and went lower. No doubt there are intrinsic differences between Pseudomonas strains in their susceptibility to phage, and it is likely that these differences played a role in clinical outcomes, as claimed by the authors. But it is not the whole story.

A key tenet of troubleshooting, which is something you do all the time in commercial operations, is that once you find one problem, you never assume it is the only problem, and that if you fix it your job is done. Unfortunately the authors did just that. They saw a difference in intrinsic susceptibility to phage that correlated to clinical outcome and were satisfied to stop there. But their own data (per the above) show that strains can be converted from one category to another by raising or lowering the phage concentrations.

You know what is directly dependent on concentrations? Kinetics. Low concentrations of phage result in slow or nonexistent rates of bacterial killing. By ignoring kinetics, they not only failed to help patients, they set back a promising therapeutic mode. They ignored the kinetics, and wasted an entire clinical study.