A Flowing Human ‘Micro-Bladder’ Reveals Why UTIs Persist — and Where Phage Therapy Could Help

Urinary tract infections are often presented as straightforward bacterial infections: identify the pathogen, prescribe an antibiotic to which it appears susceptible, and expect the infection to disappear. Yet for millions of patients, particularly women affected by recurrent UTIs, the reality is far more complicated. Bacteria can survive apparently appropriate antibiotic treatment, symptoms can return weeks or months later, and standard laboratory susceptibility tests often provide an incomplete picture of what is happening inside the bladder.

A new study published in Nature Communications on 4 September 2026 offers a striking explanation for part of this discrepancy. Researchers from University College London, the University of Oxford, the University of Leicester and collaborating institutions have built a physiologically sophisticated human bladder microtissue system in which uropathogenic Escherichia coli can be studied under human urine and realistic fluid-flow conditions. The platform revealed that the physical microenvironment of the bladder can profoundly alter bacterial behaviour and reduce the effectiveness of both antibiotics and bacteriophages. At the same time, phages showed a particularly interesting ability to reduce intracellular bacterial communities that survived conventional antibiotic treatment.

Microscopic 3D Microtissue Bladder Model

The work, led by Ramon Garcia Maset and colleagues and jointly supervised by Dario Carugo and Jennifer L. Rohn, brought together expertise in bladder biology, biomedical engineering, microbiology, imaging and phage research. Contributors included researchers from UCL's Centre for Kidney and Bladder Health, Oxford's Institute of Biomedical Engineering and Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, and the University of Leicester's Becky Mayer Centre for Phage Research, alongside collaborators from Sheffield and Airlangga University.

The scale of the problem explains why such models matter. The study estimates that urinary tract infections account for around 400 million cases annually worldwide, with approximately 260,000 deaths associated with antimicrobial-resistant UTIs. Uropathogenic E. coli, or UPEC, is responsible for roughly 80% of community-acquired infections. Recurrence or relapse can affect up to 30% of cases.

One potential explanation for recurrent infection lies in what UPEC does after reaching the bladder. These bacteria do not necessarily remain freely suspended in urine. They can attach tightly to the urothelium, invade bladder epithelial cells and establish intracellular bacterial communities, or IBCs. These protected populations are particularly important because they can occupy niches that are far less accessible to antibiotics and extracellular immune defences. A patient may therefore experience an apparent reduction in bacterial load while a fraction of the infecting population survives inside the bladder wall, potentially providing a reservoir from which infection can re-emerge.

That biology is poorly reproduced by conventional antimicrobial susceptibility testing. In a clinical microbiology laboratory, bacterial susceptibility is commonly measured in a nutrient-rich, static culture medium. The result is summarized by values such as the minimum inhibitory concentration, or MIC. But a bacterium inside a real urinary tract is exposed to a radically different environment: urine has a different nutrient content, osmolarity and pH; bladder cells create a physical surface to which bacteria can adhere and invade; and urine is continuously produced, accumulates during filling and is periodically expelled during urination.

The central achievement of the new study was therefore not simply testing another phage cocktail. It was progressively rebuilding these biological conditions and asking how each added layer changed treatment performance.

The researchers first examined UPEC in conventional laboratory medium and in pooled human urine. Nine E. coli strains were tested, including well-characterized UTI89 and CFT073 strains alongside clinical UPEC isolates and asymptomatic urinary strains. Nitrofurantoin, one of the most widely used first-line antibiotics for uncomplicated UTI, remained highly active in standard susceptibility assays. For most strains, MIC values ranged between 4 and 16 μg/mL, and changing from conventional medium to 70% pooled human urine did not dramatically alter those MIC measurements.

But more detailed killing experiments already showed that MIC alone was not telling the entire story. In both conventional medium and 70% human urine, concentrations above the MIC were necessary for complete eradication. In the urine-containing condition, a higher nitrofurantoin concentration was required to achieve the same bactericidal effect, indicating that even one relatively simple environmental change could alter treatment performance without changing the conventional susceptibility classification.

The researchers then moved to their three-dimensional urine-tolerant human urothelial model, known as 3D-UHU. Unlike a monolayer of cells growing on plastic, this system produces a differentiated and stratified human urothelium that can tolerate an apical environment consisting of human urine. UTI89 was allowed to infect the microtissue for 12 hours before antimicrobial treatment was introduced.

Here, the difference between conventional susceptibility testing and a host-like infection environment became much clearer.

Nitrofurantoin was administered at 256 μg/mL, corresponding to 16 times the measured MIC and a concentration chosen to reflect levels achievable in human urine. The antibiotic produced an approximately three- to four-log reduction in planktonic bacteria after treatment. But the population associated with the urothelium — including attached and intracellular bacteria — fell by only around two logs. The infection had already compromised epithelial barrier integrity, measured by TEER, and antibiotic treatment did not restore it during the experimental period.

An even more revealing observation emerged at a lower but still supratherapeutic concentration of seven times the MIC. Nitrofurantoin failed to significantly reduce the attached and intracellular bacterial burden even though the bacteria recovered afterwards still had the same MIC and minimum bactericidal concentration as before the experiment. No new stable antibiotic-resistance phenotype was required to explain the failure.

This distinction is central to the study. Bacteria can survive treatment without necessarily becoming genetically resistant. Their location, physiological state and relationship with host tissue can produce forms of tolerance or protection that conventional antimicrobial susceptibility testing does not capture.

The researchers next introduced bacteriophages.

The LCPR1 cocktail used in the experiments was designed against antibiotic-resistant urinary isolates and contains nine phages: three targeting E. coli and six targeting Klebsiella pneumoniae. For UTI89, the relevant E. coli phages included Leic_001, JK03 and UP17. The large cocktail reflects a strategy aimed at broad coverage across urinary pathogens rather than reliance on a single phage.

The first finding was itself a cautionary lesson for phage therapy. Phages that appeared highly active in nutrient-rich laboratory conditions performed less effectively in urine-containing media. At an MOI of 100, LCPR1 reduced UTI89 by approximately four logs after 24 hours in conventional medium but by around three logs in 70% urine-containing medium. Leic_001 and UP17 also showed evidence of reduced stability in the urine-containing environment, although the effect was phage-dependent and JK03 behaved differently.

This is important for therapeutic phage screening. A phage that performs exceptionally well in a conventional plaque assay or liquid culture may not retain the same activity at the actual site of infection. Urine composition, bacterial physiology and the surrounding tissue can all alter the phage-host interaction.

When LCPR1 was tested alone in the static 3D bladder microtissue, the result became even more nuanced. The cocktail did not significantly reduce the total bacterial burden associated with the tissue. It therefore would be inaccurate to describe the experiment simply as phages sterilizing the infected bladder model. But LCPR1 produced several other effects that may prove biologically important.

Most notably, phage treatment reduced intracellular bacterial communities.

The team developed a three-dimensional image-analysis pipeline specifically to distinguish extracellular bacteria from bacterial objects enclosed within urothelial cells. Confocal imaging, cellular segmentation and additional antibody and fluorescent controls were used to validate the identification of IBCs. Nitrofurantoin reduced overall bacterial CFU but did not significantly impair established IBC formation. LCPR1 showed the opposite pattern: it reduced IBCs despite producing no measurable reduction in total attached and intracellular CFU.

That may be one of the study's most intriguing observations.

The Oxford announcement describes the phages as reaching bacterial reservoirs hidden inside bladder tissue. The primary paper supports the broader conclusion that phage treatment reduces intracellular bacterial communities, but the precise mechanism remains unresolved. The experiments do not directly demonstrate individual phage particles penetrating intact urothelial cells and lysing bacteria inside established reservoirs. The reduction could involve direct access to intracellular populations, interference with the formation or maintenance of IBCs, killing bacteria during transitions between intracellular and extracellular states, or more complex interactions with the urothelium.

The distinction matters because it transforms an exciting observation into a scientific question rather than an already solved mechanism.

Phage treatment also affected the host tissue. LCPR1 alone reduced infection-associated cytotoxicity and modified inflammatory signalling. Exposure was associated with changes in TNFα, CXCL2 and G-CSF, while during UTI89 infection the presence of LCPR1 reduced IL-1β but increased several signals involved in neutrophil recruitment or maturation, including CXCL2, IL-8 and G-CSF.

The authors propose that phage-mediated lysis could release bacterial components that activate Toll-like receptors and NF-κB-associated inflammatory pathways. But they explicitly treat this as a hypothesis requiring more complex experimental models. Direct interactions between phages and urothelial cells cannot yet be excluded either.

The most encouraging antimicrobial results in the static model came when LCPR1 and nitrofurantoin were combined.

The combination completely eliminated detectable planktonic bacteria from the apical compartment and produced an approximately four-log reduction in the bacteria associated with the microtissue compared with untreated controls. However, total tissue-associated bacterial clearance was not substantially better than with nitrofurantoin alone.

Instead, the combination appeared to combine different strengths of the two treatments. Nitrofurantoin was particularly effective at reducing total bacterial burden, while the phage cocktail was better at reducing intracellular bacterial communities. Together they decreased both attached bacterial numbers and IBC abundance. The authors therefore suggest that the benefit of combination therapy may come from complementary targeting of distinct bacterial niches rather than simple classical synergy producing greater overall killing.

Then the researchers added perhaps the most important variable of the entire experiment: movement.

They developed a new mesofluidic system called P-FLO, or Plug-Flow Linked Organoid. Built from relatively inexpensive 3D-printed components and standard Transwell cultures, P-FLO allows controlled fluid flow across the surface of the 3D-UHU bladder microtissue.

During simulated bladder filling, urine-like fluid was delivered at 0.02 mL/min. Every two hours, the model underwent a one-minute simulated voiding event in which flow rose to 0.7 mL/min. This generated changing wall shear stresses intended to approximate some of the mechanical forces experienced by the bladder epithelium during filling and urination.

The consequences for UPEC were substantial.

Flow initially reduced the amount of freely suspended and attached bacteria following voiding events. But UTI89 adapted. By six hours, bacterial numbers had recovered to levels similar to those seen in static cultures. More strikingly, the bacteria changed their morphology.

Under flow, UTI89 cells became markedly elongated and filamentous. This phenotype was visible by confocal and electron microscopy and became increasingly pronounced over time. Bacteria recovered from the flowing environment also showed lower overall metabolic activity.

Filamentation is not simply a cosmetic change in bacterial shape. Elongated UPEC phenotypes have previously been associated with adaptation during infection and reduced susceptibility to some immune-clearance mechanisms. In this model, flow therefore acted as a biological signal capable of reshaping bacterial physiology.

The response was also strain-dependent. Another clinical UPEC isolate, EC3, showed pronounced filamentation under flow, whereas COM2, isolated from asymptomatic urine, tended instead to form aggregates. Fluid dynamics therefore did not create one universal UPEC phenotype but altered different strains in different ways.

When infections were allowed to develop for longer, the effects became even more relevant to recurrence. At 20 hours, UTI89 covered approximately 22% of the urothelial surface under flow compared with about 10% in static conditions, despite broadly comparable urothelium-associated CFU counts. Intracellular bacterial communities were also more abundant under mechanical stress.

In other words, urine flow did not simply wash the bacteria away. It altered the infection.

This finding has major implications for the way antimicrobial treatments are evaluated. A static culture may predict that a drug or phage works extremely well, while the same bacterial population exposed to tissue, urine and shear stress behaves differently enough to reduce treatment effectiveness.

That is exactly what happened when the researchers repeated the phage and antibiotic treatments under flow.

Nitrofurantoin still worked. It reduced both planktonic and urothelium-associated bacterial populations by approximately three logs relative to untreated infection. But despite increased total antibiotic exposure generated by the flowing system, its activity against planktonic bacteria was lower than in static conditions.

LCPR1 also lost efficacy under flow and did not reduce bacterial burden as effectively as expected from simpler conditions.

The combined treatment produced perhaps the clearest demonstration of how much the model mattered. Under static conditions, LCPR1 plus nitrofurantoin had completely eradicated detectable planktonic bacteria. Under flow, the same strategy produced only an approximately four-log reduction.

Four logs remains a substantial antibacterial effect, but the difference between a large reduction and complete eradication is particularly relevant when discussing recurrent infection.

Moreover, the combination did not outperform nitrofurantoin alone in terms of overall bacterial CFU under flow. Yet once again, its effect on intracellular reservoirs told a different story. Nitrofurantoin reduced intracellular bacterial communities, LCPR1 also limited their formation, and the combination reduced both total bacterial burden and IBC abundance. The combination produced a greater reduction in IBCs than nitrofurantoin alone despite comparable effects on total bacterial numbers.

This may be the most clinically interesting message of the paper.

The goal in recurrent UTI may not simply be to obtain the largest immediate reduction in bacteria floating in the urine. If a small protected population remains inside urothelial tissue, that population could potentially seed future episodes. Treatments that target extracellular bacteria and intracellular reservoirs through different mechanisms might therefore offer advantages that are invisible if efficacy is measured only through bulk CFU or a conventional MIC.

But this work is still preclinical, and the distinction is essential.

The study does not demonstrate that phage therapy prevents recurrent UTIs in patients. It does not establish an optimal phage dose or delivery route, and it does not show that LCPR1 plus nitrofurantoin will outperform antibiotics in a clinical trial. Even the micro-bladder is not a complete human bladder.

P-FLO reproduces important elements of urine exposure and wall shear stress but does not recreate all of the mechanical and volumetric events of human bladder filling and micturition. It lacks the full stretching of the urothelium, pressure fluctuations and changing urine volume that occur in vivo. The model also cannot reproduce the full complexity of a patient's systemic immune response.

The main treatment experiments were furthermore centred on UTI89, a widely studied UPEC strain. The additional clinical isolates demonstrated that flow responses can differ substantially between strains, reinforcing the need to validate therapeutic conclusions across more diverse bacterial collections.

These limitations do not diminish the study's central achievement. They illustrate why increasingly sophisticated models are needed before promising laboratory antimicrobial results are translated into patients.

One of the practical strengths of P-FLO is that the researchers deliberately designed it to be more accessible than many conventional organ-on-chip platforms. It uses commercially available Transwells and inexpensive 3D-printed components, and the team has made the system's designs available for other laboratories to adopt. This could make physiologically relevant fluid-flow experiments much more widely available in infection biology, not only for UTIs but potentially for other tissues in which mechanical flow influences microorganisms and treatment.

The study was supported primarily by the UK Engineering and Physical Sciences Research Council through the Beyond Antibiotics Programme Grant EP/V026623/1. Phage-related work at Leicester also received support through the Institute for Precision Health and Leicester Drug Discovery and Diagnostics via an MRC Impact Accelerator Account, alongside additional NIHR and international funding.

The collaboration is particularly notable because it connects two areas that are often studied separately. On one side is tissue engineering and the attempt to reproduce the human urinary microenvironment. On the other is bacteriophage biology and the search for antimicrobial strategies capable of complementing antibiotics. The results suggest that progress in phage therapy may depend as much on improving the models used to test phages as on finding new phages themselves.

A candidate therapeutic phage that performs brilliantly in rich laboratory medium but poorly in urine may never become an effective UTI treatment. A phage that produces only a modest change in total bacterial CFU but preferentially interferes with intracellular communities may, by contrast, possess a clinically valuable property that a conventional assay would overlook.

The study therefore changes the question from simply asking whether phages can kill UPEC.

The more important questions may be where they kill it, under which physical conditions, in which bacterial physiological state, and whether they can complement antibiotics by reaching or disrupting bacterial populations that conventional drugs leave behind.

For recurrent urinary tract infections, those hidden populations may be among the most important bacteria of all.

Sources :

The primary study is Ramon Garcia Maset et al., Effect of human urinary microenvironment and fluid flow on antibiotic and phage therapy efficacy against uropathogenic Escherichia coli, Nature Communications 17, 8974 (2026), published 4 September 2026.
Nature Communications — full research article

University of Oxford Engineering Science, Micro-bladder model offers clues to stopping recurrent UTIs, 4 September 2026.
University of Oxford — official news release

The study is open access under a Creative Commons Attribution 4.0 licence, meaning figures from the paper can generally be reused with appropriate attribution and indication of any modifications, unless a specific third-party credit states otherwise.