A new high-throughput imaging method that quantifies antibiotic killing at single-cell resolution predicts mycobacterial treatment outcomes more accurately than minimum inhibitory concentrations, according to research published in Nature Microbiology.

The technique, called Antimicrobial Single-Cell Testing (ASCT), tracked more than 140 million mycobacteria across approximately 20,000 time-kill curves and demonstrated that drug tolerance—the delayed killing of bacterial populations independent of resistance—is a heritable, genetically encoded trait that correlates with clinical failure in individual patients.

"Even in the absence of antibiotic resistance, treatment outcomes for bacterial infections, including urinary tract, respiratory and bloodstream infections, are often poor," wrote corresponding author Lucas Boeck, PhD, of the University of Basel in Switzerland, and colleagues. "This gap between in vitro growth inhibition and in vivo efficacy motivated us to develop strategies beyond standard susceptibility testing to better predict treatment outcomes."

Method and Validation

ASCT immobilizes bacteria in agar pads containing propidium iodide within 1,536-well plates, then captures brightfield and fluorescence images of more than 10,000 fields every 2 to 4 hours for up to 7 days. The workflow generates up to 1 million images per experiment, which are processed using sparse and low-rank decomposition for background correction, supervised random forest classifiers for bacterial segmentation and viability classification, and custom tracking algorithms based on object position and homology.

The researchers validated propidium iodide as a viability marker by treating Mycobacterium abscessus with antibiotics for 24 hours, then tracking more than 30,000 single bacteria for an additional 24 hours following washout. Only one propidium iodide-positive bacterium resumed growth—a misclassified doublet—while 1% to 11% of propidium iodide-negative bacteria regrew.

Using an inducible green fluorescent protein reporter strain, the researchers demonstrated that most propidium iodide-negative bacteria retained protein synthesis capacity following antibiotic exposure, far exceeding the fraction capable of regrowth (4% for cefoxitin, 3% for moxifloxacin, 3% for amikacin). "Importantly, the fraction of bacteria capable of GFP synthesis far exceeded the fraction of regrowing bacteria...demonstrating that regrowth assays underestimate survival and that most metabolically competent cells fail to regrow in standard media conditions," the researchers wrote.

Tuberculosis Regimen Prediction

For Mycobacterium tuberculosis, the researchers tested 65 drug regimens in 2 avirulent strains under nutrient-rich and starvation conditions, dosing each drug at maximum blood concentrations achievable during therapeutic dosing. While regimens containing isoniazid, rifampicin, or ethambutol killed growing M tuberculosis most effectively, only killing under starvation conditions predicted outcomes in mouse models and human clinical trials.

Time-kill kinetics of starved bacteria discriminated between drug combinations classified as similar to standard of care vs better than standard of care across multiple endpoints: relapsing mouse models, bactericidal mouse models of common strains, the granulomatous C3HeB/FeJ mouse strain, and phase 2 clinical trial bactericidal activity. Areas under the receiver operating characteristic curve ranged from 76% to 94%.

In contrast, colony-forming unit-based killing assessments, median MICs, and drug interaction measures showed no statistically significant association with in vivo outcomes. "Notably, all combinations contained at least one drug administered at concentrations above the MIC, highlighting the value of MICs for identifying active agents, even though regimens with very low MICs did not translate into improved outcomes," the researchers noted.

M abscessus Strain Variation

Extending ASCT to 405 clinical M abscessus isolates from patients across Europe and Australia, the researchers generated 18,244 time-kill curves against eight antibiotics at 2 concentrations each. The findings revealed highly variable but reproducible killing kinetics under identical drug exposure.

Heritability analysis using 1.3 million genomic variants demonstrated that tolerance phenotypes are substantially genetically determined, with heritability estimates of 32% to 97% across all drugs—far exceeding the 1% expected by random chance. "While drug resistance is well established as a genetic trait, drug tolerance has traditionally been viewed as a largely phenotypic characteristic," the researchers wrote.

Among eight tested drugs, only macrolide MICs associated with clinical outcomes. However, tolerance to amikacin, cefoxitin, and imipenem at both concentrations correlated with M abscessus clearance. Incorporating a single tolerance measure alongside macrolide resistance improved clinical outcome prediction, increasing the area under the receiver operating characteristic curve from 69% to 78%.

"Notably, the effect sizes of drug tolerance in predicting infection outcomes were comparable to those of macrolide MICs, currently the main in vitro measure guiding M abscessus treatment decisions," the researchers wrote.

Mechanistic Insights

Principal component analysis of tolerance phenotypes revealed clustering by antibiotic target, with distinct patterns for drugs targeting protein synthesis, DNA, and the cell wall. Genome-wide association identified multiple genes linked to tolerance, including MAB_0233, a putative phage tail tape measure protein.

Knockout of MAB_0233 increased tolerance to translation-targeting antibiotics (amikacin, tigecycline, linezolid) without affecting cell wall-active drugs or altering MICs. Complementation restored the low-tolerance phenotype for amikacin and tigecycline.

The researchers also identified a tigecycline low-tolerance clade within the dominant circulating clone of M abscessus subspecies massiliense—strains that often carry high-level aminoglycoside and macrolide resistance. "Our finding of low tigecycline tolerance highlights a potential vulnerability in otherwise highly drug-resistant M abscessus, where treatment success rates can fall below 20%," they wrote.

Limitations

The researchers acknowledged that propidium iodide directly reflects killing by cell wall damage but only indirectly captures other killing mechanisms, with an approximately 3-hour delay observed in M abscessus for non-lytic antibiotics. Clinical outcomes are also influenced by host immunity, drug penetration, toxicity, and adherence—factors ASCT cannot capture.

"By studying millions of single-cell fates across hundreds of conditions, our approach provides a scalable framework to translate in vitro killing into in vivo efficacy, opening new avenues for drug development and personalized therapy," the researchers concluded.

The researchers reported no competing interests.

Reference: Nature Microbiology