The centerpiece of conventional susceptibility testing is the minimum inhibitory concentration, or MIC, the lowest drug concentration that prevents visible bacterial growth under standardized conditions. The MIC is enormously useful as a comparative benchmark, but it compresses a rich dynamical story into a single number. It says little about how quickly a bacterial population declines under a given dose, whether that decline will reverse if the population shrinks below a critical size, or how the drug itself degrades over the course of an experiment. Textbook pharmacodynamic models, the researchers note, tend to neglect these transient dynamics and density-dependent effects even though both are ubiquitous in natural and clinical settings. In an actual infection, where the bacterial population is the entity being treated, ignoring such effects raises the risk of undertreatment.

One especially stubborn phenomenon is the inoculum effect: the observation that the apparent potency of an antibiotic decreases as the starting density of bacteria increases. A drug concentration that wipes out a small inoculum may fail against a dense one, a pattern with obvious consequences for treating established, high-burden infections. Related complications include rate effects, which concern how fast a population is killed rather than whether it is killed, and yield effects, which describe the total biomass a culture can sustain under drug pressure. Classic microbiology has documented these phenomena for decades, but quantifying them rigorously requires data that standard endpoint assays simply do not provide.

To close that gap, the team generated high-resolution optical density time series for Pseudomonas aeruginosa, a Gram-negative bacterium notorious for its role in hospital-acquired infections, cystic fibrosis airway disease, and its formidable antibiotic resistance. The experimental design was deliberately comprehensive: three antibiotics, twelve doses of each, seven different inoculum sizes, and fourfold replication, producing a dense grid of growth curves that captures how population trajectories respond jointly to drug concentration and starting density. Optical density measurements, taken continuously, record the turbidity of each culture as a proxy for bacterial abundance, yielding thousands of data points per condition rather than the single endpoint that a conventional assay would produce.

That density of measurement is what makes the new analysis possible. From the time series, the researchers estimated time derivatives, the instantaneous rates of change of population size, using gradient estimation techniques. Derivatives are the currency of dynamical systems modeling: ordinary differential equation models make explicit claims about how growth and death rates depend on population size, drug concentration, and time, and those claims can only be tested against rate data rather than raw abundance curves. By evaluating candidate population-scale ordinary differential equation models directly against the estimated derivative data, the pipeline could distinguish between mechanistic structures that produce similar-looking growth curves but imply very different biology.

The second pillar of the pipeline is a classification strategy for transient dynamics. Rather than reducing each growth curve to summary statistics, the researchers clustered derivative trajectories across the dose-inoculum space using unsupervised clustering, letting the data itself reveal which combinations of drug concentration and starting density produce qualitatively similar dynamical behavior. This approach maps out distinct regimes of response, for example regions where populations crash irreversibly, regions where they recover after an initial decline, and boundary zones where small changes in dose or inoculum flip the outcome. The clustering thus converts a continuum of curves into an interpretable landscape of dynamical regimes.

Applied to their Pseudomonas data, the pipeline identified an ordinary differential equation model with two key structural features: a saturating antibiotic-loss term and a threshold-dependent weak Allee term. The saturating antibiotic-loss term captures the fact that antibiotics degrade or are consumed over time, and that the rate of this loss saturates rather than scaling indefinitely, meaning the drug environment itself is dynamic during the experiment. The weak Allee term encodes density-dependent cooperation: at low population densities, per-capita growth becomes negative below a threshold, so a battered population can spiral toward extinction even after drug levels fall, whereas a denser population can recover from the same insult. Together, these components recapitulated and quantified the classic rate, yield, and inoculum effects that microbiologists have long observed but struggled to formalize within a single predictive framework.

The payoff is not merely descriptive. The fitted model and the clustering analysis together suggest a set of novel metrics that define thresholds separating distinct dynamical regimes in dose-inoculum space. Where the MIC asks a binary question, whether growth is inhibited at a given concentration, these metrics ask richer ones: how far a given dose-inoculum combination sits from the boundary between recovery and collapse, how the trajectory of a population will unfold over time, and how sensitive that outcome is to the initial burden of bacteria. Such thresholds could eventually help clinicians reason more quantitatively about dosing in situations where inoculum size is known to matter, and help researchers compare antibiotic candidates not just by potency but by the geometry of the dynamical regimes they induce.

The methodological contribution extends well beyond antibiotics. The authors emphasize that their approach, a derivative-based fitting algorithm paired with clustering of derivative trajectories, is applicable to any biological time series collected under controlled perturbations with variable initial conditions. That covers a vast swath of experimental biology, from microbial growth under nutrient stress to population responses to environmental change. Wherever researchers can measure a system repeatedly as it responds to graded perturbations from different starting states, the pipeline offers a way to move from curves to mechanisms, testing differential equation models against rate estimates and organizing the resulting behaviors into regimes without imposing preconceived categories.

The study arrives at a moment when the limits of conventional susceptibility metrics are under increasing scrutiny, as treatment failures in high-density infections continue to challenge the assumption that a single laboratory number can predict clinical response. By treating bacterial populations as dynamical systems rather than static targets, Sundius and colleagues have shown that the inoculum effect and its kin are not experimental annoyances but structured, quantifiable features of drug-pathogen interaction, with boundaries that can be mapped and modeled. The work, published in PLOS Computational Biology on May 10, 2026, under DOI 10.1371/journal.pcbi.1014827, suggests that the future of susceptibility testing may lie not in richer endpoint measurements but in the shape of the transient, the often-overlooked interval between exposure and outcome where the fate of an infection is actually decided.

Subject of Research: Quantifying antibiotic susceptibility and inoculum effects in Pseudomonas aeruginosa using transient dynamics modeling

Article Title: Quantifying antibiotic susceptibility and inoculum effects using transient dynamics of Pseudomonas aeruginosa

Article References: Sundius, S., Farrell, J., Eick, K. L., Kuske, R., & Brown, S. P. (2026). Quantifying antibiotic susceptibility and inoculum effects using transient dynamics of Pseudomonas aeruginosa. PLOS Computational Biology, 22(10), e1014827. https://doi.org/10.1371/journal.pcbi.1014827

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014827

Keywords: antibiotic susceptibility, Pseudomonas aeruginosa, inoculum effect, minimum inhibitory concentration, ordinary differential equations, transient dynamics, pharmacodynamics, unsupervised clustering, time-series analysis, Allee effect, computational biology, dose-response