A phenotypic state variable for antimicrobial resistance: MIC-distribution entropy exposes population heterogeneity and phenotypically quiet genomic reservoirs
Abstract
Antimicrobial-resistance surveillance reports an isolate as susceptible, intermediate, or resistant, and a genome as a list of resistance genes; neither shows whether a population is uniformly wild-type or harbours a resistant minority. Both discard the shape of the minimum inhibitory concentration (MIC) distribution, which encodes within-population resistance heterogeneity. We treated the Miller-Madow-corrected Shannon entropy of that distribution as a phenotypic state variable and asked whether it carries information beyond the resistance genome. Entropy was computed on a fixed 19-bin MIC ladder for 1,612 species-drug distributions (European Committee on Antimicrobial Susceptibility Testing [EUCAST], 252 species) and 11,259 species-drug-year series (Antimicrobial Testing Leadership and Surveillance [ATLAS], 241 species, 2004–2024), paired with resistance-gene annotations (Comprehensive Antibiotic Resistance Database) for 64 species (6,382 genomes). Entropy rose with the non-wild-type fraction defined by epidemiological cut-offs (Spearman ρ = 0.72), indexing coexisting subpopulations. Across 50 paired species, resistance-gene content explained less than 2% of the variance in phenotypic distributional complexity (ordinary least squares R2 = 0.014, p = 0.406). This decoupling was the most reproducible feature, recovered under phylogenetic correction (Pagel-λ phylogenetic generalised least squares p = 0.891), a mechanism-matched comparison (pooled R2 = 0.002), and a 145-specification multiverse, although only |r| ≥ 0.39 could be excluded. In 33,862 Bacterial and Viral Bioinformatics Resource Center genomes with MICs, catalogued genes explained a minority of isolate-level entropy (median 20.7%, 95% confidence interval [CI] 18.5–25.2%). Population MIC complexity is therefore not reconstructable from the genome, and the residual is descriptive but stable: 14% of species expressed more complexity than their gene content predicts, 16% less, and 70% remained aligned, including all six ESKAPE species (Enterococcus, Staphylococcus, Klebsiella, Acinetobacter, Pseudomonas, Enterobacter). Entropy was reproducible across estimators (Spearman ρ ≥ 0.92) and discriminated later resistant-subpopulation emergence in walk-forward testing (area under the curve [AUC] = 0.695, 95% CI 0.629–0.756), similar to a momentum baseline (0.665), but did not forecast escalation in a hold-out split at the record’s midpoint (AUC = 0.535, 95% CI 0.472–0.594). MIC-distribution entropy therefore resolves a phenotypic dimension of resistance, heterogeneity consistent with heteroresistance, and silent genomic reservoirs, that is invisible to breakpoint categories and gene counts.
Related
- Antimicrobial resistance at the animal / environment interface: phenotypic detection and persistence of ESBL and carbapenemase producing Enterobacterales in Irish cattle systems
- Phenotypic and genomic characterization of antimicrobial resistance in Escherichia coli and Klebsiella spp. from livestock in Namibia
- Antimicrobial resistance in Mannheimia haemolytica from diseased cattle: phenotypic and genotypic insights over five years (2018-2022).
- Data heterogeneity and algorithmic bias in AI-based antimicrobial resistance prediction: a systematic review and mitigation framework
- A panel of genotypically and phenotypically characterised WHO priority Gram-negative bacteria to facilitate antimicrobial and diagnostic development
BibTeX
@article{ulaganathan2026phenotypic,
title = {{A phenotypic state variable for antimicrobial resistance: MIC-distribution entropy exposes population heterogeneity and phenotypically quiet genomic reservoirs}},
author = {Siva Bharathi Ulaganathan and C. Shreya and Karthikeyan Sivashanmugam},
journal = {Frontiers in Microbiology},
year = {2026},
month = nov,
doi = {10.3389/fmicb.2026.1976579},
url = {https://www.frontiersin.org/articles/10.3389/fmicb.2026.1976579},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A phenotypic state variable for antimicrobial resistance: MIC-distribution entropy exposes population heterogeneity and phenotypically quiet genomic reservoirs
AU - Siva Bharathi Ulaganathan
AU - C. Shreya
AU - Karthikeyan Sivashanmugam
PY - 2026
DA - 2026/11/04
JO - Frontiers in Microbiology
DO - 10.3389/fmicb.2026.1976579
UR - https://www.frontiersin.org/articles/10.3389/fmicb.2026.1976579
AB - Antimicrobial-resistance surveillance reports an isolate as susceptible, intermediate, or resistant, and a genome as a list of resistance genes; neither shows whether a population is uniformly wild-type or harbours a resistant minority. Both discard the shape of the minimum inhibitory concentration (MIC) distribution, which encodes within-population resistance heterogeneity. We treated the Miller-Madow-corrected Shannon entropy of that distribution as a phenotypic state variable and asked whether it carries information beyond the resistance genome. Entropy was computed on a fixed 19-bin MIC ladder for 1,612 species-drug distributions (European Committee on Antimicrobial Susceptibility Testing [EUCAST], 252 species) and 11,259 species-drug-year series (Antimicrobial Testing Leadership and Surveillance [ATLAS], 241 species, 2004–2024), paired with resistance-gene annotations (Comprehensive Antibiotic Resistance Database) for 64 species (6,382 genomes). Entropy rose with the non-wild-type fraction defined by epidemiological cut-offs (Spearman ρ = 0.72), indexing coexisting subpopulations. Across 50 paired species, resistance-gene content explained less than 2% of the variance in phenotypic distributional complexity (ordinary least squares R2 = 0.014, p = 0.406). This decoupling was the most reproducible feature, recovered under phylogenetic correction (Pagel-λ phylogenetic generalised least squares p = 0.891), a mechanism-matched comparison (pooled R2 = 0.002), and a 145-specification multiverse, although only |r| ≥ 0.39 could be excluded. In 33,862 Bacterial and Viral Bioinformatics Resource Center genomes with MICs, catalogued genes explained a minority of isolate-level entropy (median 20.7%, 95% confidence interval [CI] 18.5–25.2%). Population MIC complexity is therefore not reconstructable from the genome, and the residual is descriptive but stable: 14% of species expressed more complexity than their gene content predicts, 16% less, and 70% remained aligned, including all six ESKAPE species (Enterococcus, Staphylococcus, Klebsiella, Acinetobacter, Pseudomonas, Enterobacter). Entropy was reproducible across estimators (Spearman ρ ≥ 0.92) and discriminated later resistant-subpopulation emergence in walk-forward testing (area under the curve [AUC] = 0.695, 95% CI 0.629–0.756), similar to a momentum baseline (0.665), but did not forecast escalation in a hold-out split at the record’s midpoint (AUC = 0.535, 95% CI 0.472–0.594). MIC-distribution entropy therefore resolves a phenotypic dimension of resistance, heterogeneity consistent with heteroresistance, and silent genomic reservoirs, that is invisible to breakpoint categories and gene counts.
SN - 1664-302X
ER -