How Dan I. Andersson and his colleagues are tracing the evolutionary escape routes that allow apparently susceptible bacteria to survive antibiotics
The meaning of Heteroresistance: “An infection can contain millions or billions of bacteria. Even in a population that looks fully susceptible, random mutations mean a very small fraction of cells (say, one in a million) may carry a change that lets them survive the antibiotic. This is called heteroresistance and it is a problematic killer!”
A bacterium does not have to be fully resistant to make an antibiotic fail. Sometimes, it merely needs a few cells that are different. Hidden inside an apparently susceptible bacterial population may be a minute resistant minority, perhaps one cell among a million. A routine laboratory test can classify the infection as treatable. The clinician prescribes accordingly. Yet the antibiotic eliminates the susceptible majority while giving the rare resistant cells room to multiply.
This phenomenon, known as heteroresistance, has become the organising problem in a substantial strand of research led by Dan I. Andersson, professor of medical bacteriology at Uppsala University. His programme connects bacterial evolution, clinical microbiology and diagnostic engineering. Its significance lies not in a single discovery announced in September 2026, but in the gradual assembly of evidence that a clinically important form of resistance may often be present before conventional testing can see it.
The distinction is important. A university profile published on 24 September marked Andersson’s receipt of the 2026 Olof Rudbeck Prize and the start of a five-year research project. It was not, in itself, the announcement of a newly released treatment, diagnostic test or peer-reviewed breakthrough.
What it described was more consequential, if less dramatic: a mature research programme moving from evolutionary theory towards clinical evidence and potentially deployable diagnostics.
A career built around bacterial adaptation
Andersson has studied bacterial evolution and antibiotic resistance for more than three decades. From 1994 to 2004, he worked at the former Swedish Institute for Infectious Disease Control, where his research became more closely connected to applied infection biology. After returning to Uppsala University as a professor, he continued to combine experimental evolution with questions about resistance, treatment and diagnosis.
The programme’s breadth is important. It examines how genetic variation arises, how resistance changes bacterial fitness and virulence, how bacteria compensate for the biological costs of resistance, and how antibiotics interact when used in combination. The group works with pathogens including Escherichia coli, Salmonella Typhimurium, Staphylococcus aureus, Mycobacterium tuberculosis, Klebsiella pneumoniae and Pseudomonas aeruginosa, using bacterial genetics, experimental evolution, molecular biology, sequencing and mathematical modelling.
That approach has repeatedly challenged simplified assumptions about antibiotic resistance.
Resistance is often imagined as a binary property: a bacterium is either susceptible or resistant. Andersson’s work instead depicts bacterial populations as genetically variable systems in which resistance can emerge, recede, persist at low frequency or evolve through routes that later become difficult to reconstruct.
His earlier research helped show that resistance can be selected at antibiotic concentrations far below those required to stop bacterial growth. Other work demonstrated that bacterial adaptation to environments without antibiotics can incidentally reduce susceptibility to several drug classes. In one experiment, E. coli and Salmonella entericapopulations were propagated for 500 to 1,000 generations without antibiotic exposure; mutations selected for improved growth in those environments also produced reduced susceptibility to antibiotics.
The implication was unsettling. Antibiotic exposure is a principal driver of resistance, but it is not the only evolutionary process capable of producing resistance-related traits.
The population behind the laboratory result
Heteroresistance pushes this evolutionary argument into the diagnostic laboratory.
A standard susceptibility report generally assigns one result to an isolate. Heteroresistance means that the result may describe the majority while missing the minority that determines what happens during treatment. The resistant subgroup may occur at frequencies between roughly one in ten million and one in ten thousand cells, according to a 2026 diagnostic study.
A major mechanism is tandem gene amplification. A small number of bacteria acquire extra copies of a resistance gene, raising their resistance sufficiently to survive antibiotic pressure. Because amplifications can be unstable, the extra copies may disappear when the drug is removed. That instability makes the phenotype difficult both to detect and to reproduce.
This is not the same as persistence or tolerance. In heteroresistance, the subpopulation can grow at antibiotic concentrations that inhibit the majority. The problem is not simply that some cells temporarily stop growing. It is that a minority has measurably lower susceptibility and can be selected during treatment.
By 2019, Andersson, Hervé Nicoloff and Karin Hjort had synthesised evidence from laboratory experiments, animal models, mathematical modelling and clinical studies suggesting that these resistant minorities could be enriched during antibiotic exposure and contribute to treatment failure. They also stressed that the evidence base remained incomplete and that inconsistent definitions and methods complicated comparisons between studies.
That combination of biological plausibility and diagnostic uncertainty has defined the field ever since.
From a hidden phenotype to patient outcomes
The strongest clinical development in the Uppsala programme came not from the September 2026 profile but from a retrospective cohort study published in The Lancet Microbe in 2025.
Researchers examined preserved E. coli isolates from 255 people treated for bloodstream infections in the Uppsala healthcare region in 2014 and 2015. They reassessed the bacteria for heteroresistance to cefotaxime, gentamicin and piperacillin-tazobactam, then compared the laboratory findings with information from patient records.
The prevalence varied sharply by antibiotic. Breakpoint-crossing heteroresistance was found in fewer than 1 per cent of isolates for cefotaxime, 43 per cent for gentamicin and 9 per cent for piperacillin-tazobactam. Of 125 isolates displaying this form of heteroresistance, 120, or 96 per cent, had been classified as susceptible by ordinary disk-diffusion testing.
Among patients who received the antibiotic to which their infecting strain was heteroresistant, gentamicin heteroresistance was associated with higher odds of intensive-care admission and death. Piperacillin-tazobactam heteroresistance was associated with higher odds of admission to intermediate care. These findings move the programme beyond laboratory mechanism, but they do not settle the clinical question.
The study was retrospective and examined a geographically and temporally defined cohort. Its primary outcome, length of hospital stay, was not associated with heteroresistance. Admission to higher levels of care and mortality can be influenced by illness severity, underlying disease, treatment timing and many other factors. Although statistical adjustment can reduce confounding, it cannot remove every source of bias.
In a published response, Gabriele Maria Leanza, Rita Murri, Maurizio Sanguinetti and Carlo Torti described the work as important but questioned whether length of stay and escalation of care could reliably isolate the effect of heteroresistance. They also asked whether recurrent infections developed full resistance after exposure to an antibiotic against which the original isolate had been heteroresistant.
The responsible conclusion is therefore narrower than the strongest publicity claims. The study provides clinically concerning evidence of association, particularly when patients received the implicated drug. It does not prove that routine heteroresistance screening would improve survival, nor does it establish how clinicians should alter therapy when the phenotype is detected.
Prospective, multicentre studies and intervention trials are still needed.
Building a test that can find one cell in a million
Clinical significance means little if the phenotype cannot be detected quickly enough to influence treatment. The present reference method, population-analysis profiling, is labour-intensive and poorly suited to routine, time-sensitive clinical microbiology. The Uppsala programme has consequently expanded into microfluidics, imaging and machine learning.
In a 2025 Science Advances study, Sagar N. Agnihotri, Nikos Fatsis-Kavalopoulos, Jonas Windhager, Maria Tenje and Dan I. Andersson placed bacteria into microscopic droplets containing growth medium and antibiotics. Growth of resistant cells was detected through changes in the droplets. The method identified resistant subpopulations occurring at frequencies as low as one in a million, using approximately 200 to 300 droplets, and was evaluated with three clinically important antibiotics in E. coli isolates from bloodstream infections.
A subsequent 2026 paper combined droplet microfluidics with image-texture quantification to detect rare resistant populations from bloodstream infections. The work represents a genuine diagnostic advance at the research-validation stage, but it should not be confused with a test already established in routine care.
The commercial translation effort is being pursued through Singular Diagnostics Sweden AB. Its RAREST® platform uses droplet-based microfluidics and is intended to identify heteroresistance within a 12-to-24-hour clinical decision window. The project was included on the Royal Swedish Academy of Engineering Sciences’ 2026 list of research projects judged to have commercial and societal potential.
That is promising, but the wording is crucial. RAREST® could support earlier treatment adjustment. Publicly accessible material reviewed for this article does not establish that it has completed the full pathway of prospective clinical validation, regulatory clearance, health-economic assessment and routine implementation across hospital laboratories.
The technology also has declared intellectual-property interests. The researchers reported a pending patent covering the microfluidic approach. That does not invalidate the science, but it increases the importance of independent replication and transparent validation.
Where machine learning may help, and where it may not
Another branch of the programme has explored whether whole-genome sequencing can predict heteroresistance.
In work summarised in Sofia Jonsson’s 2025 doctoral thesis, researchers trained machine-learning models on 467 clinical E. coli isolates assessed for piperacillin-tazobactam heteroresistance. The best-performing model reportedly achieved 100 per cent sensitivity and 84.6 per cent specificity in that dataset. Increased resistance-gene copy number, caused by tandem amplification or changes in plasmid copy number, emerged as a principal mechanism.
The result is technically striking, but it exposes a central difficulty. Gene amplifications can be unstable. A sequencing result may depend on how the sample was cultured, when it was collected and whether the relevant minority remained present at a detectable frequency. A model trained on one organism, antibiotic and laboratory workflow may not retain its performance in other hospitals or patient populations.
Machine learning may therefore complement phenotypic testing rather than replace it.
External validation against prospectively collected isolates will be decisive.
The evolution of an invisible intermediate
The September 2026 university profile stated that Andersson was beginning a five-year European Research Council project on the evolution of heteroresistance. The Advanced Grant provides €2.5 million, approximately SEK 27.2 million at the university’s reported exchange rate, for a project examining how resistant minorities arise, persist and disappear.
The profile was accurate as a programme update. Subsequent research, however, strengthened its scientific context.
On 9 October 2026, Brandon A. Berryhill and Dan I. Andersson posted a preprint modelling unstable heteroresistance as a stepping stone towards stable resistance. Their simulations suggested that, under beta-lactam exposure, bacteria could initially survive either by amplifying a resistance gene or by acquiring compensatory mutations. Over weeks or months, both routes could converge on a stable, low-copy resistant state. At that point, genome sequencing might reveal no trace of the earlier amplified intermediate.
If confirmed experimentally and clinically, the idea could solve an evidential puzzle. Heteroresistance may facilitate stable resistance while erasing the genetic signature of its own contribution. But this is not yet an established clinical finding. The October work is a preprint based on mathematical and computer simulations. It has not yet passed peer review, and it does not demonstrate the proposed evolutionary route in patients.
It nevertheless answers one part of the editorial question raised by the September item: later material did appear that made the programme more concrete. It still did not transform the university profile into a discrete treatment or diagnostic breakthrough.
Beyond heteroresistance
Andersson’s clinically oriented contributions extend beyond detecting rare resistant cells.
The CombiANT method, introduced in 2020, uses a three-dimensional printed insert in an agar plate to generate diffusion patterns for three antibiotics. Automated analysis measures whether each antibiotic pair is synergistic, additive or antagonistic. Initial technical validation in E. coli, Pseudomonas aeruginosa and Staphylococcus aureusproduced results comparable with conventional checkerboard testing while reducing complexity.
Later testing of 232 isolates from five Gram-negative species produced 696 measured antibiotic interactions. Most were additive, 11.6 per cent were antagonistic and only 0.3 per cent were synergistic. Some interactions varied between isolates of the same species, reinforcing the argument that combination therapy cannot always be predicted from species-level assumptions.
Here, too, the boundary between useful finding and clinical implementation must be respected. CombiANT offers a practical research and potentially diagnostic method. The available evidence does not mean that personalised combination testing has become a universal component of routine infection care.
A programme at the translational threshold
The most accurate description of the work is neither “basic research only” nor “clinical breakthrough”.
Several levels of evidence now coexist:
- Mechanism: Extensive laboratory research supports gene amplification, copy-number variation and compensatory evolution as routes into and out of heteroresistance.
- Patient association: A retrospective Uppsala cohort links some forms of heteroresistance with adverse outcomes, while leaving substantial questions about causation and generalisability.
- Diagnostic validation: Microfluidic methods can identify extremely rare resistant cells in clinical isolates, but routine implementation has not yet been demonstrated.
- Commercial development: RAREST® aims to produce clinically actionable results within 12 to 24 hours, but its clinical, regulatory and economic performance remains to be established publicly.
- Evolutionary projection: The October 2026 preprint offers a plausible model in which unstable heteroresistance becomes stable resistance and then disappears from the historical genetic record.
The accomplishment is therefore cumulative. Andersson and a widening network of microbiologists, clinicians, engineers and computational researchers have made an obscure bacterial phenotype measurable, mechanistically intelligible and clinically testable.
They have not yet shown precisely when a heteroresistance result should overrule an established treatment guideline. Nor have they demonstrated, in a prospective interventional study, that detecting it changes therapy and improves outcomes.
That is the programme’s next threshold.
Antibiotic resistance is usually recognised after bacteria have crossed a defined susceptibility breakpoint. The Uppsala work asks whether medicine has been looking too late, after a heterogeneous population has already been simplified into a single laboratory category.
The most important resistance may not always be the resistance a test reports. It may be the resistance the test leaves behind.
Source assessment
- Uppsala University, researcher profile, 24 September 2026. Primary institutional source for the prize, career narrative and new ERC-funded programme. Reliable for biographical and programme information, but promotional in form and not evidence of a new scientific breakthrough. [uu.se], [uu.se]
- Uppsala University, ERC grants announcement, 23 June 2026; European Research Council award list.Primary institutional and funder sources for the €2.5 million Advanced Grant. [uu.se], [erc.europa.eu]
- Heyman and colleagues, The Lancet Microbe, 2025. Peer-reviewed retrospective cohort study based on 255 bloodstream infections. The most important direct clinical evidence, but observational, geographically limited and vulnerable to residual confounding. [thelancet.com]
- Jonsson, Uppsala doctoral thesis, 2025. Useful synthesis of the cohort, machine-learning work and amplification mechanisms. A university-reviewed dissertation, but not a substitute for independent external validation. [uu.diva-portal.org]
- Leanza and colleagues, The Lancet Microbe, 2025. Published critical correspondence identifying uncertainties in outcome selection and causal interpretation. Important for balancing the clinical claims. [thelancet.com]
- Agnihotri and colleagues, Science Advances, 2025; npj Digital Medicine, 2026. Peer-reviewed diagnostic-development research demonstrating highly sensitive droplet-microfluidic detection. Promising validation evidence, not proof of routine clinical effectiveness. [science.org], [pmc.ncbi.nlm.nih.gov], [nature.com]
- Singular Diagnostics Sweden AB and the Royal Swedish Academy of Engineering Sciences. Primary commercial and innovation-assessment sources for RAREST®. Useful for technology status and intended performance, but commercial claims require independent clinical and regulatory confirmation. [iva.se], [singulardi…ostics.com]
- Berryhill and Andersson, bioRxiv, 9 October 2026. Relevant subsequent development proposing that heteroresistance can become stable resistance without leaving a later genomic trace. It is a non-peer-reviewed modelling preprint and should not guide clinical care. [biorxiv.org], [biorxiv.org]
- Andersson, Nicoloff and Hjort, Nature Reviews Microbiology, 2019. Authoritative peer-reviewed review of heteroresistance mechanisms and clinical relevance. Strong for conceptual context, although later clinical studies refine its conclusions. [nature.com], [pubmed.ncb…lm.nih.gov]
- CombiANT studies, 2020–2024. Peer-reviewed methodological and experimental evidence concerning isolate-specific antibiotic interactions. Relevant to the programme’s wider translational development, but not evidence that personalised combination testing is standard clinical practice. [journals.plos.org], [journals.asm.org], [pubmed.ncb…lm.nih.gov]
