Binary

In September 2026, Jacob Choby and colleagues at Emory published a finding that should unsettle anyone who trusts an antibiotic susceptibility report. They took an Enterobacter cloacae complex isolate — the kind of organism that causes hospital-acquired infections in exactly the patients who can least afford treatment failure — and showed that its resistance to fosfomycin depended on something no laboratory routinely measures: how much glucose was in the environment.

The mechanism was precise. Glucose represses GlpT, the transporter that imports both glycerol and fosfomycin into the bacterial cell. In low-glucose conditions, GlpT expression is high and fosfomycin enters freely — the isolate tests susceptible. In high-glucose conditions, GlpT expression drops. But GlpT expression isn't uniform across the population: some cells express it, some don't. The cells that don't — the ones with low GlpT — survive fosfomycin. They're not genetically resistant. They're metabolically invisible.

Combined with the fosfomycin resistance gene fosA, this heterogeneous expression generates a subpopulation of resistant cells within an otherwise susceptible culture. In diabetic mice, the frequency of this resistant subpopulation increased. The host's metabolic condition was creating resistance that the standard test — run in Mueller-Hinton broth at pH 7.2, without glucose — would never detect.

The Test and the Patient

This is the gap that heteroresistance exploits. Heteroresistance — the coexistence of susceptible and resistant subpopulations within a single bacterial isolate — is not a rare laboratory curiosity. It is staggeringly common. In a systematic assessment by the Andersson group, 27.4% of 766 bacteria-antibiotic combinations tested were heteroresistant. Not 2.7%. Twenty-seven percent.

And the numbers get worse the closer you look. Among Pseudomonas aeruginosa isolates, imipenem and meropenem heteroresistance has been detected in 54% and 73% of isolates, respectively. In Staphylococcus aureus, gentamicin heteroresistance reaches 69.2%. In carbapenem-resistant Acinetobacter baumannii, 59% of isolates are cefiderocol-heteroresistant. In A. baumannii bloodstream infections, one study found 32.2% colistin heteroresistance — meaning nearly one in three BSI isolates harbored resistant subpopulations that standard susceptibility testing would miss.

The Stepping Stone

Heteroresistance isn't just a testing artifact. It's an evolutionary mechanism. In January 2026, researchers showed that A. baumannii colistin heteroresistance serves as a stepping stone to stable, full resistance — ISAba-mediated eptA hyperexpression converts transient heteroresistant subpopulations into permanently resistant ones. The bacteria pay a fitness cost (38 vs 22 minute doubling time), but once the selection pressure arrives, they're already there.

The same logic applies to gene amplification — unstable tandem duplications that expand under antibiotic pressure and revert without it. Jonsson and colleagues used nanopore sequencing to resolve these amplification dynamics in real time, revealing how bla_SHV copy number expansion drives piperacillin-tazobactam heteroresistance in E. coli. Standard AST, which reports a single MIC value averaged across the entire population, cannot see this.

Nor can it see what Choby found: that the patient's own biology — specifically, the glucose concentration in the infection microenvironment — determines whether the resistant subpopulation is 1 in 10,000 or 1 in 100.

The Paradigm Call

Resistance is not a fixed property of a bacterium. It is context-dependent — modulated by temperature, oxygen, pH, nutrient availability, and the conditions of the infection environment. Standard AST, performed at fixed pH 7.2 under aerobic conditions, captures one point in a multidimensional landscape.

— Frank Aarestrup et al., Trends in Microbiology, September 22, 2026

When Frank Aarestrup writes something, the field notices. The architect of DANMAP — the surveillance system that proved the link between agricultural antibiotic use and human resistance — is arguably the most influential AMR scientist alive. His September 2026 perspective in Trends in Microbiology calls for a fundamental reconceptualization: resistance is not binary. It is a spectrum that shifts with the environment. And our testing infrastructure was built on the binary assumption.

The implications cascade. Breakpoints — the MIC thresholds that divide "susceptible" from "resistant" — are set using standardized conditions. Treatment guidelines are built on breakpoints. Surveillance data aggregates S/I/R categories. Every layer assumes the test result reflects what the drug will do inside the patient. When 27% of isolates harbor resistant subpopulations that the test misses, every layer is making decisions on incomplete information.

The Tools That Could See It

The diagnostic technology exists. The DnD assay, published in Nature Communications, uses microfluidic confinement and time-lapse microscopy to detect resistant subpopulations at frequencies as low as 1 in 100 million — in 36 minutes. Resistell's nanomotion technology delivers AST results in 1–2 hours with 97.6% sensitivity. QuantaMatrix's dRAST is commercialized across 26 countries and pursuing FDA 510(k) clearance in 2026. An EU HERA consortium is spending €8.85 million to build a point-of-care AST device that delivers results in under an hour.

None of these tools specifically report heteroresistance profiles as standard output. They accelerate the binary question — susceptible or resistant, faster — without fundamentally changing the question being asked. The DnD assay is the exception, designed specifically to resolve subpopulation heterogeneity, but it remains a research tool, not a clinical standard.

The Patient We're Missing

Roughly 35% of hospitalized patients with serious infections have diabetes. In the United States, 38 million people live with diabetes and another 98 million have prediabetes. These patients have higher infection rates, worse outcomes, and longer hospitalizations — consequences we have attributed primarily to immune dysfunction.

Choby's finding adds a second mechanism. The hyperglycemic environment doesn't just impair the immune response. It actively shifts the bacterial population toward resistance phenotypes that our standard tests — designed for glucose-free Mueller-Hinton broth — cannot detect. The clinician orders fosfomycin for a susceptible isolate. The lab report says S. But inside the diabetic patient, the glucose-repressed GlpT population is already enriched. The drug enters fewer cells. The resistant subpopulation expands. Treatment fails, and the failure is classified as unexplained.

This is one drug, one transporter, one pathogen, one metabolic condition. The same principle — environment-dependent resistance — likely operates across dozens of drug-pathogen-patient combinations we haven't examined. Aarestrup's perspective identifies temperature, oxygen tension, pH, and carbon source availability as variables that modulate resistance phenotypes. Each variable adds a dimension that standard AST, locked at its single standardized point, does not explore.

What This Changes

The S/I/R system is not wrong. It is incomplete. It captures the average behavior of a population under standardized conditions, and for most infections in most patients, that average is close enough. But "close enough" fails systematically in the patients who are hardest to treat: the diabetic, the critically ill, the immunocompromised, the patient with a deep-seated abscess at pH 5.5 where aminoglycosides lose activity, the ventilated patient whose lung microenvironment differs from anything Mueller-Hinton broth approximates.

Heteroresistance is not a new concept — the term dates to the 1940s. What is new is the convergence: systematic prevalence data (27%), mechanistic understanding (gene amplification, metabolic heterogeneity, ISAba stepping stones), environmental modulation (glucose, pH, oxygen), clinical association (treatment failure, relapse), and the authoritative call to change (Aarestrup). The field is ready for a paradigm shift. Whether the clinical infrastructure — breakpoint committees, diagnostic manufacturers, treatment guidelines — can absorb it is a different question.

For now, the gap between what we test and what the patient experiences remains unmeasured, unmonitored, and in most clinical settings, unacknowledged. One in four bacteria-antibiotic combinations harbors a population our tests call susceptible that is already preparing to survive.