Fighting Resistance with Resistance: How Scientists Are Turning Bacterial Evolution Against Itself

The dominant strategy against antibiotic resistance has been the same for 80 years: find a new molecule that kills bacteria, deploy it, and wait for resistance to evolve. Then find another. This arms race has produced every antibiotic we have — and the crisis we face when resistance outpaces discovery.

But a different approach is emerging. Instead of searching for new weapons, a growing body of research asks a subversive question: what if resistance itself has vulnerabilities?

Every adaptation has a cost. When bacteria evolve to resist one antibiotic, they often lose something else — a surface receptor, an efflux pump, a metabolic pathway. These losses can make them vulnerable to other drugs, to viruses that prey on bacteria, or to genetic systems that strip away resistance genes entirely. The trick is not to overpower evolution but to redirect it.

Three strategies now embody this idea, each at a different stage of maturity: phage steering, which has clinical proof in humans; collateral sensitivity, which has population-level surveillance validation; and CRISPR gene drives, which remain in the laboratory but represent the most radical reimagining of how we might fight resistance.

Phage Steering: Forcing Bacteria to Choose

Bacteriophages — viruses that infect and kill bacteria — have been used therapeutically for over a century. The modern renaissance of phage therapy adds a strategic dimension that goes beyond simply killing pathogens. It exploits a fundamental constraint: bacteria cannot resist phages and antibiotics simultaneously without paying an evolutionary price.

The mechanism is elegant. Many phages bind to bacterial surface structures that also serve as resistance machinery. The best-studied example is OMKO1, a phage that targets the outer membrane protein OprM in Pseudomonas aeruginosa. OprM is part of the MexAB-OprM efflux pump — one of the primary mechanisms by which P. aeruginosa expels antibiotics from its interior. When the bacterium mutates OprM to escape OMKO1, it cripples its own efflux system. The phage-resistant mutants become antibiotic-sensitive again.

This is phage steering: selecting phages that force bacteria into an evolutionary trade-off where resistance to the phage means vulnerability to drugs.

OMKO1 was the proof of concept, but it is no longer alone. At least six phage-pathogen systems now demonstrate analogous trade-offs:

The first clinical proof came in 2025. Benjamin Chan’s group at Yale treated nine adults with cystic fibrosis harboring multidrug-resistant or pan-drug-resistant P. aeruginosa with personalized nebulized phage cocktails. The phages were specifically selected for predicted evolutionary trade-offs — not just killing capacity, but the expectation that surviving bacteria would be weaker.

This was compassionate use, not a randomized controlled trial. The sample size was small. But it demonstrated that phage steering works in human lungs, against the hardest infections, in patients who had exhausted every other option.

The field is advancing rapidly. Armata Pharmaceuticals’ AP-SA02, a phage cocktail targeting Staphylococcus aureus bacteremia, received QIDP designation from the FDA in February 2026. The FDA confirmed Phase 3 readiness based on the diSArm Phase 2a data. If the superiority study initiating in the second half of 2026 succeeds, AP-SA02 would become the most advanced IV phage therapy program for bloodstream infections.

Honesty demands acknowledging the limits. Not all phage resistance involves beneficial trade-offs. Some phage-resistant mutants show enhanced biofilm formation, increased virulence, or heightened antibiotic resistance — so-called trade-ups. Phage 14/1 against P. aeruginosa and phage U115 against A. baumannii both select for mutants with greater fitness, not less. Phage steering requires careful selection of phages whose resistance mechanisms are genuinely costly to the pathogen. The wrong phage can make things worse.

Collateral Sensitivity: The Hidden Weaknesses in Resistance

Phage steering exploits trade-offs between phage resistance and antibiotic susceptibility. Collateral sensitivity exploits trade-offs between resistance to one antibiotic and sensitivity to another.

The concept is intuitive: a bacterium that evolves to resist drug A becomes more vulnerable to drug B. If you can map these relationships, you can design drug sequences that trap bacteria in an evolutionary dead end — every adaptation to survive the current treatment makes them more susceptible to the next one.

The problem has always been scale. Laboratory evolution experiments reveal collateral sensitivity patterns, but do these patterns hold in the messy reality of clinical infections, with their diverse genetic backgrounds, polymicrobial communities, and variable antibiotic exposures?

In March 2026, Tandar and colleagues provided the strongest evidence yet that they do. Using ISIS-AR, the Dutch national antimicrobial resistance surveillance system, they analyzed collateral sensitivity patterns across hundreds of thousands of clinical isolates. The results, published in The Lancet Microbe, confirmed that collateral sensitivity relationships observed in laboratory evolution hold at the population level across multiple bacterial species. They also identified new collateral sensitivity pairs not previously found in laboratory studies — the clinical data revealed patterns that controlled experiments had missed.

The Tandar study built on systematic laboratory work. In January 2025, Sakenova and colleagues at EMBL published the most comprehensive chemical genetics mapping of drug interactions in E. coli to date. Using high-throughput fitness assays, they identified 404 cross-resistance and 267 collateral sensitivity interactions — a threefold expansion of known relationships. They validated 64 of 70 predictions in independent experiments and introduced OCDM, a metric for quantifying collateral drug interactions that accounts for resistance levels.

The temporal dynamics add another layer of complexity. Maltas and colleagues at the University of Michigan showed that collateral sensitivity increases with evolution time — the longer bacteria are exposed to drug A, the more sensitive they become to drug B. But predictions made at day 2 of evolution only predicted 41% of the collateral sensitivity patterns observed at day 4. The dosing window matters. Start the second drug too early, and the collateral sensitivity may not have fully developed. Wait too long, and the bacterium may have found an escape route.

In the largest lab evolution study to date, researchers validated the tigecycline-to-polymyxin B collateral sensitivity pathway across 779 clinical uropathogenic E. coli isolates. The mechanism — involving Lon protease dysfunction and exopolysaccharide changes — was consistent across diverse clinical backgrounds. This is the kind of scale that bridges the gap between laboratory curiosity and clinical reality.

Species matters. A 2026 analysis across six ESKAPE pathogens found that collateral sensitivity profiles are species-specific — patterns reliable in E. coli may not hold in Klebsiella pneumoniae or Staphylococcus aureus. P. aeruginosa emerged as the most promising target for collateral sensitivity-based strategies, partly because its resistance mechanisms (particularly efflux pump overexpression) create the most consistent vulnerabilities.

The bridge between phage steering and collateral sensitivity was built in 2025 by Mu and colleagues, who published the first systematic collateral sensitivity-informed phage cocktail design. Working against carbapenem-resistant hypervirulent K. pneumoniae, they designed a three-phage cocktail exploiting dual-layer collateral sensitivity: the first phage selected for capsule loss (exposing LPS targets), while the second and third phages exploited LPS modifications that switched the O-antigen serotype. The result: 100% mouse survival in a lethal infection model. Phage steering and collateral sensitivity are not separate strategies — they can be integrated into a single therapeutic design.

Yet the most critical fact about collateral sensitivity in 2026 is this: zero clinical trials have been registered testing collateral sensitivity-based antibiotic cycling in human patients. The laboratory evidence is extensive. The surveillance data is supportive. The mathematical modeling is encouraging — Kline and colleagues showed that equal allocation of drugs exhibiting reciprocal collateral sensitivity outperforms or matches sequential deployment for delaying resistance. But no one has yet tested whether deliberately sequencing antibiotics based on collateral sensitivity patterns improves patient outcomes in a controlled clinical setting.

CRISPR Gene Drives: Disarming Resistance at the Genetic Level

Phage steering and collateral sensitivity work with evolution — redirecting it, exploiting its trade-offs. CRISPR gene drives propose something more radical: overwriting bacterial evolution entirely.

In February 2026, Ethan Bier and Justin Meyer at UC San Diego published a system called pPro-MobV — a conjugal CRISPR-based gene drive that spreads through bacterial populations and strips away antibiotic resistance genes. The concept is borrowed from insect gene drives, which spread genetic modifications through mosquito populations faster than Mendelian inheritance allows. Applied to bacteria, the idea is to introduce a CRISPR cassette that targets resistance genes on plasmids — the mobile DNA elements that carry most clinically relevant resistance — cuts them, and inserts itself in their place. Then, through bacterial conjugation (a natural process of DNA transfer between cells), the cassette spreads to neighboring bacteria, neutralizing their resistance genes in turn.

pPro-MobV demonstrated this in E. coli, targeting plasmid-borne ampicillin and kanamycin resistance. Crucially, it worked inside biofilms — the dense microbial communities that protect bacteria in chronic infections, on medical devices, and in hospital plumbing. The team built in a safety mechanism: a homology-based deletion system that allows the CRISPR cassette to be removed if necessary.

The vision is compelling: instead of killing resistant bacteria (which selects for more resistance), neutralize their resistance and let existing antibiotics work again. A genetic disarmament campaign, self-propagating through bacterial communities.

The reality is sobering. pPro-MobV has no in vivo data. Of the approximately nine studies that have ever tested any CRISPR antimicrobial in animals, none have used a conjugal gene drive approach. The challenges are formidable: delivery to infection sites, stability of the CRISPR cassette across generations, off-target effects in the broader microbiome, regulatory frameworks that do not exist for self-spreading genetic modifications in clinical pathogens, and the fundamental question of what happens when a self-propagating genetic system encounters the vast diversity of bacterial species in a patient or an environment.

CRISPR gene drives for AMR are years — perhaps a decade — behind phage steering in clinical maturity. But they represent the logical endpoint of the paradigm shift this article describes: from killing bacteria to reprogramming them.

The Gradient of Readiness

These three strategies exist on a gradient. Phage steering has treated patients. Collateral sensitivity has population-level surveillance data. CRISPR gene drives have laboratory proof of concept. None of them is ready to replace standard antibiotic therapy. All of them challenge the assumption that the only way to fight resistance is to find another molecule that kills.

The common thread is strategic humility. Instead of brute-forcing bacterial death, each approach asks: what does resistance cost the bacterium? And can we make that cost lethal?

This is not a new idea in evolutionary biology. Fitness trade-offs are a central concept in adaptive evolution. What is new is the systematic application of this thinking to clinical infectious disease — and the convergence of tools (engineered phages, high-throughput chemical genetics, CRISPR, computational modeling, national surveillance systems) that make it actionable.

The UK is positioning itself at the forefront of translating these approaches. A November 2025 meeting at the University of Liverpool, reported in Nature Microbiology, brought together researchers, clinicians, regulators, and industry to define the path toward phage therapy implementation in the NHS. The UK’s permanent antibiotic subscription model — launching April 1, 2026, with contracts up to £20 million per year per drug — could theoretically extend to phage products or collateral sensitivity-guided treatment protocols, though it has not yet done so.

The arms race is not over. We still need new antibiotics. But the most sustainable long-term strategy may not be more weapons. It may be learning to turn the enemy’s adaptations into vulnerabilities — fighting resistance with resistance.