Evolution of antibiotic resistance under subtherapeutic dosing — the concentration zone that kills susceptible bacteria while actively selecting for resistant mutants
Every antibiotic-bacteria pair is characterized by more than a single potency number. The minimum inhibitory concentration (MIC) tells us how much drug is needed to stop the visible, dominant, drug-susceptible population from growing. But real bacterial populations are never genetically uniform — spontaneous mutation constantly generates a small subpopulation with reduced susceptibility. Suppressing that subpopulation requires a second, higher threshold: the mutant prevention concentration (MPC). Understanding both thresholds — not just the MIC — is the foundation of the mutant selection window hypothesis.
The MIC is measured by broth microdilution or agar dilution: a standardized bacterial inoculum (~10⁵–10⁶ CFU/mL) is exposed to serial two-fold dilutions of antibiotic, and the MIC is the lowest concentration that prevents visible turbidity (growth) after ~18–24 hours.
Crucially, MIC testing uses an inoculum far too small to reliably contain a resistant mutant. Spontaneous resistance mutations occur at rates of roughly 10⁻⁶ to 10⁻⁹ per cell division depending on the gene and drug. At 10⁵–10⁶ CFU, a standard MIC test may contain zero resistant cells — so the MIC reflects only the susceptibility of the wild-type, drug-sensitive majority.
Clinically, this means a concentration at or modestly above the MIC reliably stops the dominant population from growing and looks "effective" by conventional susceptibility testing — while doing nothing to prevent a much smaller, less-susceptible subpopulation already present in a real infection (10⁸–10¹⁰ CFU in an established infection) from surviving and expanding.
The mutant prevention concentration is measured differently: a very large bacterial inoculum (≥10¹⁰ CFU, enough to statistically guarantee the presence of pre-existing resistant mutants) is plated onto agar containing various antibiotic concentrations. The MPC is the lowest concentration that prevents growth of ANY colony — including the least-susceptible first-step mutant in that large population.
Because MPC testing deliberately captures the rare resistant subpopulation that MIC testing misses, MPC is always ≥ MIC, and the gap between them can span several-fold in concentration. This gap is not a testing artifact — it reflects a real biological truth: killing the majority population and killing the resistant minority require genuinely different drug exposures.
The MPC is sometimes difficult to determine precisely for slow-growing or fastidious organisms, so it is often estimated or bracketed between the highest concentration permitting growth and the lowest preventing it.
Traditional dosing philosophy asks: "is the concentration above the MIC?" If yes, treatment is considered adequate. The two-threshold framework asks a sharper question: "is the concentration positioned to suppress the resistant subpopulation too, not just the susceptible majority?"
This reframing matters most for infections with high bacterial burden (pneumonia, endocarditis, osteomyelitis) where the sheer number of bacteria present all but guarantees pre-existing resistant mutants are already there before treatment even starts. In these settings, exceeding the MIC alone is necessary but not sufficient — the goal must be to spend as little time as possible in the space between MIC and MPC.
A drug concentration can be many-fold above the MIC — and by conventional interpretation look like a strong, effective dose — while still sitting well below the MPC. "Above MIC" and "resistance-safe" are not the same statement.
Plot bacterial growth response against antibiotic concentration and three zones appear: below MIC, growth continues largely unimpeded; above MPC, essentially the entire population — susceptible and resistant alike — is suppressed; and in between lies the mutant selection window (MSW), the range from MIC to MPC. Within this window, susceptible bacteria are killed while resistant mutants are not — the very conditions that convert a rare mutant into a dominant subpopulation.
At a concentration inside the MSW, the antibiotic exceeds what the susceptible majority can tolerate — those cells stop dividing, or die outright, and their numbers fall. But the concentration remains below what is needed to stop the resistant minority — those cells continue to grow, sometimes essentially unimpeded, sometimes at a somewhat reduced rate depending on how far above MIC (but still below MPC) the concentration sits.
The result is a population inversion: a bacterial community that started out overwhelmingly susceptible (resistant mutants typically <0.01–0.001% of the total at baseline) can, over the course of days of window-range exposure, become predominantly resistant — not because resistance was "created" by the drug, but because the drug removed all the competition that had been holding the resistant subpopulation in check.
It is tempting to picture the window as a rigid box between two exact numbers. In reality MIC and MPC are themselves statistical estimates with measurement variability, and different resistance mechanisms within the same species can have different MPC values. The practical implication is the same regardless: concentrations that hover in the general vicinity between typical susceptible-population and resistant-population thresholds carry elevated selective risk, and the risk rises the closer concentration sits to the MIC end and falls the closer it approaches the MPC end.
The window also is not a fixed number in absolute drug concentration — it is fundamentally about the RATIO between drug exposure and both threshold values, which is why the same absolute plasma concentration can be safely above the window against one pathogen and squarely inside it against a less-susceptible one.
Under the traditional MIC-only lens, a regimen that clears symptoms but leaves a resistant subpopulation behind might look like a treatment success followed by an unrelated "new" resistant infection later. Under the MSW lens, this is recognized as the same episode: the drug did its job against the susceptible majority (symptoms improve) while simultaneously enriching for the resistant minority that eventually breaks through, sometimes during the same course, sometimes at relapse weeks later.
This reframing is why regulatory and stewardship guidance increasingly evaluates candidate antibiotic regimens not only for whether they clear infection, but for how much cumulative time predicted drug-concentration profiles spend inside each pathogen's selection window.
The mutant selection window hypothesis, formalized by Karl Drlica and colleagues, reframes resistance emergence: it is not merely that "underdosing lets bacteria survive" — it is that a specific, definable concentration range actively converts survival into a competitive advantage for the resistant genotype.
Every sizeable bacterial population already contains rare pre-existing resistant mutants, generated continuously by spontaneous mutation before any antibiotic is ever given — resistance is not induced by the drug, it is selected by it. Inside the mutant selection window, this pre-existing minority is transformed into the dominant surviving population purely through differential survival and reduced competition, a textbook case of natural selection operating on a bacterial timescale of hours.
A foundational (and often misunderstood) point: antibiotics do not "cause" a susceptible bacterium to mutate into a resistant one in direct response to drug exposure. Resistance-conferring mutations — altered drug targets, upregulated efflux pumps, drug-inactivating enzymes — arise stochastically during ordinary DNA replication, independent of whether antibiotic is present. This was demonstrated definitively by the Luria-Delbrück fluctuation test in 1943, a landmark experiment in evolutionary biology.
What the antibiotic does is act as a selective filter after the fact: in the absence of drug, resistant mutants have no growth advantage and remain a vanishingly rare minority, often out-competed by the fitter wild-type. Inside the mutant selection window, the filter flips — susceptible cells are suppressed and resistant cells are not — and the rare minority becomes free to expand into the ecological space vacated by its dying competitors.
Bacterial populations compete for finite local resources — nutrients, attachment sites, oxygen tension, physical space within a biofilm or tissue niche. Under normal (no-drug) conditions, resistant mutants — which frequently carry a fitness cost from their resistance mechanism (an efflux pump consumes energy; an altered ribosomal protein may translate slightly less efficiently) — are often out-competed and held at low frequency by the much larger susceptible population.
Once drug concentration inside the window kills off the susceptible majority, this competitive suppression vanishes — a phenomenon ecologists call "competitive release." Freed from competition, resistant cells access more nutrients per cell and can proliferate toward carrying capacity, even if their intrinsic growth rate is somewhat slower than the wild type's was pre-treatment.
Because this is a selection process acting over successive bacterial generations, the outcome scales with time spent in the window, not merely with whether concentration ever entered it. A brief transient dip into window-range concentration during otherwise adequate dosing carries materially less selective risk than concentration that lingers in the window for many hours or days, allowing many generations of differential survival and regrowth to accumulate.
This is why pharmacokinetic/pharmacodynamic (PK/PD) indices used in antibiotic dosing — AUC/MIC, Cmax/MIC, %T>MIC — have MSW-aware counterparts that track exposure relative to MPC as well: %T>MPC (percentage of the dosing interval above MPC) and the "window duration" or “time in window” it complements. Regimens engineered to maximize %T>MPC (or minimize time-in-window) are specifically designed to blunt this selective-amplification process.
Selective amplification inside the window is a race against generation time: the longer bacteria remain in the window, the more doublings the resistant subpopulation gets while its competitor is suppressed — turning a starting frequency of 1-in-a-million into a majority population within days.
In principle, avoiding the mutant selection window sounds simple: dose high enough to stay above the MPC. In practice, drug concentrations in the body are never flat lines — they rise after each dose and decay steadily until the next one, following each drug's elimination pharmacokinetics. Under-dosing, missed or delayed doses, poor adherence, drug interactions that accelerate clearance, or simply a normal trough near the end of a long dosing interval can all let concentration fall into window range for meaningful stretches of time — and every hour spent there is an hour of active selection.
A drug concentration profile drifting into the mutant selection window can arise from many distinct clinical scenarios, often in combination:
• Under-prescribing: a dose or interval chosen conservatively (e.g. to limit toxicity) that never reaches a peak comfortably above the MPC for the target pathogen • Non-adherence: missed doses, early discontinuation once symptoms improve, or irregular dose timing that extends the low-concentration trough period • Pharmacokinetic variability: faster-than-average renal or hepatic clearance, reduced absorption (food interactions, GI dysfunction), or altered volume of distribution (obesity, edema, critical illness) that lowers achieved concentrations relative to a standard dose • Site-of-infection penetration: even adequate serum concentrations may fail to reach window-safe levels in poorly perfused tissue, abscesses, biofilms, or across the blood-brain barrier • The natural decay phase itself: even a well-designed regimen spends part of every dosing interval declining from peak toward trough — if that trough sits inside window range before the next dose is due, some window exposure is essentially built into the regimen
It is intuitive to think of an inadequate dose as simply a weaker version of an adequate one — somewhat less effective, but still moving in the right direction. The mutant selection window hypothesis shows this intuition is incomplete and, in the specific window range, actually backwards.
A concentration profile that lingers in window range does not just fail to fully suppress the population — it actively reshapes that population's genetic composition toward resistance, an outcome strictly worse than doing nothing at all in terms of the pathogen's future susceptibility. A patient who takes no antibiotic leaves the resistant-to-susceptible ratio roughly where it started; a patient dosed squarely inside the window can leave with a population enriched for resistance even though the treatment nominally "helped" in the short term.
The probability that a given infection already harbors a resistant mutant before treatment even begins scales directly with bacterial burden. At 10⁶ CFU and a mutation rate of 10⁻⁸, roughly one hundredth of a resistant mutant is expected to be present on average — often zero. At 10¹⁰ CFU (typical of an established pneumonia or abscess), the same mutation rate predicts roughly 100 pre-existing resistant cells even before the first dose is given.
This means subtherapeutic, window-range dosing is far more dangerous in high-burden infections (pneumonia, endocarditis, large abscesses, chronic biofilm-associated infections) than in low-burden ones, because the raw material for selection — pre-existing resistant mutants — is essentially guaranteed to already be present in the former and may genuinely be absent in the latter.
A regimen that "sort of works" — clearing most bacteria, easing symptoms, but leaving trough concentrations inside the mutant selection window for hours each day — can be a worse long-term outcome for antimicrobial resistance than either full-dose therapy or no therapy at all.
If time spent between MIC and MPC is what drives resistance selection, then rational dosing strategy follows directly: design the dose, interval, and course length so that drug concentration at the site of infection spends as little time as possible in that range — ideally staying above the MPC for as much of the dosing interval as achievable, and never being stopped early while a resistant subpopulation might still be present but suppressed.
The most direct lever is simply dosing high enough that peak (and ideally trough) concentrations clear the MPC, not just the MIC, for the specific pathogen being treated. For concentration-dependent antibiotics (aminoglycosides, fluoroquinolones), this means favoring higher, less frequent doses that achieve a high Cmax/MPC ratio. For time-dependent antibiotics (beta-lactams), it means dosing and interval combinations engineered to keep concentration above the relevant threshold for a large fraction of the dosing interval, sometimes via extended or continuous infusion strategies specifically to avoid trough concentrations drifting into window range.
Dose selection informed by MPC (rather than MIC alone) is increasingly incorporated into breakpoint-setting and stewardship guidance for pathogens and drug classes where resistance emergence during therapy is a recognized clinical problem, such as fluoroquinolones against Mycobacterium tuberculosis and Pseudomonas aeruginosa.
Even a regimen with an adequate peak can still spend substantial time in window range during the decay phase between doses if the dosing interval is too long relative to the drug's elimination half-life. Shortening the interval (more frequent dosing of the same total daily amount), or using extended-infusion / continuous-infusion delivery for time-dependent drugs, compresses the trough period and reduces cumulative %T-in-window without necessarily requiring a higher total dose.
Therapeutic drug monitoring (TDM) — measuring actual patient drug levels rather than assuming population-average pharmacokinetics — allows individualized interval adjustment for patients whose clearance is unusually fast (young, augmented renal clearance in critical illness) and who would otherwise spend more time in window range than a standard regimen predicts.
Stopping therapy as soon as symptoms improve — a common but risky pattern of non-adherence — can leave a partially-reduced but still-present bacterial population, including any resistant subpopulation that was suppressed but not eliminated, free to regrow once drug concentration falls away entirely. Completing the prescribed course (calibrated to actually clear the infection, not just relieve symptoms) closes this reopening of selective opportunity.
Combination therapy — two antibiotics with independent mechanisms and independent resistance mutations dosed together — offers an orthogonal defense: a bacterium would need to acquire resistance mutations to both drugs simultaneously to escape, a probability approximated by multiplying each drug's individual mutation rate (e.g. 10⁻⁷ × 10⁻⁷ = 10⁻¹⁴), making it far less likely that any single-window exposure allows full escape. This is the same logic underlying combination regimens for tuberculosis and HIV, where mutation escape from any single drug is common but simultaneous escape from several is not.
The mutant selection window hypothesis converts a qualitative stewardship slogan — "take your full course, don't skip doses" — into a quantitative pharmacological target: keep concentration above the mutant prevention concentration for as much of every dosing interval as the regimen can achieve, and never let the course end while a suppressed resistant subpopulation might still be viable.