The Phages Did Not Eradicate Anything, and the Patients Improved

Phage therapy has a measurement problem. The intuitive endpoint is eradication: count the pathogen before treatment, count it after, and declare success if it is gone. That framing has governed how phage trials are designed, and it may be the reason the field keeps producing results that look like failures. A small case series from Swiss and German hospitals is interesting less for what it cured than for the argument it makes about what should have been counted.

Three women with recurrent urinary tract infections, histories running from 5 to 44 years and requiring between four and ten courses of antibiotics in the preceding year, received a two-phage cocktail against Escherichia coli, given orally for two days and then directly into the bladder for six more. Two also received a faecal microbiota transplant. The phages did not eradicate the bacteria. At 24 months, the patients were nonetheless substantially better.

Why it matters: Recurrent UTI accounts for more than 15 percent of all outpatient antibiotic prescriptions, which makes it one of the larger single contributors to community antibiotic pressure. An intervention that reduces recurrence without sterilizing the reservoir would still be worth having, but only if trials are designed to detect it.

What the phages did and did not do

The pharmacology was clean. Oral administration alone never produced detectable phage in the urine, which is a useful negative result for anyone designing a protocol. Instillation directly into the bladder did, and once established, titres persisted between doses. Bacterial counts fell sharply on the first bladder instillation, and two patients showed a seven-log reduction in urinary E. coli.

Then the bacteria came back. Follow-up samples were positive again, and durable eradication was not achieved in any patient. On the eradication endpoint, this is a negative study. On the clinical endpoint it is not: symptom scores fell from an average of 5 to 1, quality-of-life scores from 2 to 0, and across 24 months one patient had no further episodes at all, one had fewer with no further antibiotics, and the third, who declined the transplant, reported reduced severity but unchanged frequency.

The authors make the inference explicit, writing that sustained eradication with phage alone may not be achievable or even necessary for a clinical effect. As McCallin and colleagues argue in Nature Microbiology, if clinical and microbiological endpoints come apart, the clinical one should take precedence.

Why a transplant was part of it

The design pairs two interventions aimed at different reservoirs, and the logic is worth separating out. Phages attack the bacteria in the bladder. But the gut is where uropathogenic E. coli persists between episodes, and clearing the bladder does nothing about the source. Faecal transplant was included to shift the intestinal community toward one less hospitable to that persistence, something phages cannot do. Sequencing confirmed the recipients' gut communities moved toward their donors' composition, which establishes that the transplant engrafted without establishing that engraftment is what helped.

The pattern across the three patients is at least consistent with it mattering. The two who received transplants had the better outcomes, including the one patient with no recurrences at all; the patient who declined it reported milder episodes at unchanged frequency. With three people that is an observation, not a finding, and the authors treat it as one.

The strain that was not resistant

One observation deserves attention beyond this study. In the second patient, phage susceptibility dropped mid-treatment, which is exactly the signature of evolved phage resistance and would normally be reported as such. Whole-genome sequencing showed it was not. The bacterium had been replaced by a phylogenetically distinct strain differing at 2,076 of 2,513 typed alleles, one that lacked the original's fluoroquinolone resistance and carried F1C fimbriae, a less virulent attachment apparatus than the type 1 fimbriae of the strain it displaced.

Without sequencing, that event would have been recorded as treatment failure through resistance. It was closer to the opposite: a resistant, more virulent strain displaced by a susceptible, less virulent one. The lesson generalizes past phages, since any therapy targeting one organism in a colonized site can be scored wrong if the readout cannot distinguish an old strain surviving from a new strain arriving.

What the study can't say yet

Three patients, no control arm, no blinding, and outcomes that include self-reported symptom severity. The authors say so plainly and report against case-series guidelines. Recurrent UTI also fluctuates on its own, and patients who have pursued experimental treatment after decades of failed therapy are not a group from whom unblinded symptom improvement can be read confidently. Every patient additionally received co-trimoxazole, so no observation here isolates the phages from the antibiotic.

The prior evidence tempers it further. The one completed phase II randomized trial of phage therapy for UTI found phage non-inferior to antibiotics but not superior to placebo, which is the result this case series would need to overturn rather than assume. What this work supplies is not evidence that the treatment works but a specific, testable argument about how the next trial should be scored.

Quick questions

Is phage therapy ready for recurrent UTI? No. This is three patients without a control group, and the strongest existing randomized evidence did not beat placebo.

Why give phages into the bladder rather than by mouth? Because oral dosing produced no detectable phage in urine at all. Reaching the site of infection required direct instillation, which is a practical constraint on how this could ever be delivered.

What's the one-line takeaway? Phages failed to eradicate E. coli in all three patients yet recurrences and antibiotic use fell over two years, which the authors argue means eradication is the wrong endpoint to be scoring.