Recent developments in the field of antibiotic pharmacodynamics have led to a better knowledge of the optimal dosage strategies to obtain maximal eradication of the infecting pathogen 
. Further, optimizing the duration and dosing of treatment to reduce the likelihood of resistance for a given level of antibiotic use, however, remains a challenge. Resistance to dozens of antibiotics has evolved in hundreds of bacterial species, but the relationships between dosage regimens, pharmacokinetics and therapeutic efficacy in the context of bacterial resistance are only now beginning to be understood 
. The public health response to various aspects of the resistance problem has largely focused on a single solution—the reduction of antibiotic overuse. A more holistic approach may be to build up a functional taxonomy, a classification of bacterial pathogens that are functionally similar for resistance evolution. A functional taxonomy would consider pathogen ecology, the dynamics of immunity along with other issues that are important determinants of transmission and of population structure. Those other important concerns may include the typical population sizes of colonizing bacteria, their intrinsic resistance to antibiotics, their tendency to accumulate resistance genes, and species-specific cross-resistance to dominant antibiotics.
Our research particularly emphasizes the public health importance of understanding the dynamics of bacterial pathogens in their dominant ecological reservoir, and pharmacodynamic and pharmacokinetic parameters operating in those tissues, as an important complement to concerns about optimizing treatment for bacterial infections.
In this paper, we show that 1) all antibiotic use—even under a “perfect” dosing strategy—exerts selection for resistance, and 2) for most infections the optimal dosing strategy for clinical treatment may not be optimal for preventing the spread of resistance, likely a consequence of those treatment regimens may not being focused on the dynamics of bacterial populations. The general rule suggested by our studies is that shorter duration of treatments is usually, but not always, optimal. More comprehensive guidelines that consider the important differences among bacteria in their ecological dynamics, as well as in the diversity of ecological dynamics for the same bacteria in different habitats in the body could contribute to dramatically reduce the volume of antibiotics consumed and selection for resistance.
An example is the treatment of AOM—the second most common infection among children after the common cold, and the single most important reason for antibiotic prescriptions in the United States 
. Here, no treatment or a short three-day course with antibiotic treatment has been shown to be no less effective than ten days with antibiotics 
. The motivation for the trend and steady progress in designing shorter drug regimens have been based on clinicians' experience rather than on systematic evaluation of the trade-off between treatment success and resistance, and yet, many physicians continue to recommend that patients complete the full 10 day course of antibiotic treatment 
, thereby potentially accelerating the rate with which resistance evolves and spreads from other bacterial populations. A guideline change toward shorter treatment durations with antibiotics would not only prevent resistance among individuals receiving antibiotics for self-limiting infections, but also among the large number of patients receiving antibiotics for nonbacterial infections 
For infections that cannot be handled by the immune system alone, i.e. for which high concentrations of antibiotics for long durations are required, it is important to consider the functional taxonomy of the drugs used. The mode of administering the antibiotic is important because it affects the concentrations of the drug in other parts of the body; antibiotics typically reach much higher concentrations in the gut if given orally rather than intravenously 
. For instance, there is evidence suggesting that the rise of vancomycin-resistant enterococci in the United States during the 1980s could have been driven by the use of oral vancomycin for C. difficile 
. In Europe, where vancomycin was mostly used intravenously, vancomycin concentrations and selection for resistance in gut commensals would be expected to be much lower, and vancomycin-resistant enterococci are much less frequent. Tuberculosis treatment does not impose an increased selection for resistance, despite extensive drug regimes, because Mycobacterium tuberculosis
does not form a part of the commensal flora, and because the two key agents (izoniazid and ethambutol) in the triple-drug combination are active only against mycobacteria, which are not a part of the commensal flora. In contrast, for the treatment of acne with broad-spectrum antibiotics (minocycline), sometimes for a year or more 
, our results suggest a significant selection pressure on the commensal flora. By leaving the gut flora, upper respiratory flora, and much of the skin flora unexposed, topical usage of antibiotics could be a solution. It has, however, long been discouraged and amounts to only 1% of systemic use 
Since populations of commensal flora can be extensive, resistant bacteria generated by otherwise rare mutations or genetic exchange events are likely to exist and comprise some part of the existing flora at the time of antibiotic treatment 
. Resistant commensal bacteria are therefore more likely to persist and cause resistant infections when they spread to other hosts. Even if they do not cause infection, commensal bacteria can transfer genetic material coding for resistance to other, more pathogenic bacteria.
It follows that the value of drug optimization is related to the full spectrum of drug pharmacokinetics and pharmacodynamics throughout the body, both in terms of the delivery mode and the organisms it targets. Optimization of treatment in the primary ecological reservoir for transmission is both the best measure of collateral damage in transmission models and the most important consideration for public health. Our results emphasize the need to reexamine topical, intravenous, or focal delivery of antibiotics for other drug and organism combinations.
Caveats and Limitations
Like all mathematical models of biological systems, the models analyzed here involve some assumptions and simplifications. Mutations in the model are generated through a one-step process. Genetic transfers typically result in higher levels of resistance, but are unlikely to occur in self-limiting infections unless there is a coinfection; the first documented case of infection caused by vancomycin-resistant S. aureus
is one important example 
. Allowing for mutations to high-level resistance at the colonization site increases the time span in which there is extensive selection for resistance, but does not alter our conclusions.
We assumed drug concentration were maintained at a constant level throughout treatment—an assumption which is unnatural for most situations other than for intravenous treatment 
. For oral and intramuscular administration, the concentrations of antibiotics will be in a continuous state of flux, and the dynamics could be further confounded by factors such as non-compliance. As result, the effect of periodic waning of antibiotic concentrations below the effective MICs are not considered in our analysis, but instead constant concentrations either below or above the effective MICs are considered. Accounting for periodic waning of antibiotic concentrations to below the effective MICs will affect the intensity of the selection for resistance and the time scale for the dynamics and should therefore be considered in future analyses for antibiotic and pathogen specific guidelines.
The MIC was used as a single pharmacodynamic parameter in the model. The pharmacodynamic function captures the effect of an antibiotic over a wide range of antibiotic concentrations and as a consequence of differences in the shape of these pharmacodynamic functions, antibiotics with the same MICs and pharmacokinetics may differ profoundly in their microbiological efficacy 
. This assumption may have to be relaxed for future predictions in the development of antibiotic treatment protocols for specific drug and organism combinations.
An important aspect of our study was that we explicitly considered the different dynamics of drugs and bacteria that could occur at the infection site and in other habitats, taking into account the effects of immune response on the emergence of and selection for resistance. Bacteria populations in different habitats in a body are not expected to have strong direct interactions, so the population responses would be different and largely independent. Because of the many interdependent mechanisms, such as different cell types and cytokines involved in the immune responses to bacterial infections, experiments to measure the contribution of various components have been difficult, and precise knowledge about these processes is still lacking 
. The focus of this study, however, was not prediction of resistance emergence for a specific infection or species of the microflora, but rather a conceptual framework for addressing questions about the impact of considering both the infection site and the commensal site in the optimization of antibiotic drug dosing regimens to prevent resistance.
In the model, the immune response depends on the total bacterial population. There is, however, evidence that some pathogens may escape the effects of an antibiotic by moving from the extracellular to the intracellular space. In contrast to our assumption, resistance to an antibiotic would then be correlated with increased resistance to the immune response 
. The evidence for this phenomena is still emerging and further investigation is needed. Were this observation to become established, however, the degree of resistance prior to treatment could affect the overall immune response.
There may be other important clinical outcomes to consider in the evaluation of dosing strategies. Some studies for instance have reported an increase in the number of complications from countries with lower rates of antibiotic prescribing for AOM 
. The incidence is low, however, and the risk of more serious sequelae has to be weighed against the risk and consequences of a strategy that generates more-resistant organisms.
Lastly, this study only captured one of the two overlapping problems related to resistance: the emergence and selection of resistant strains within the host. For many infections, primary resistance caused by the spread of resistant strains within a population is the most significant problem 
and it remains important for future studies to evaluate whether the gain of treating a patient outweighs the risk of resistance in the population.
Summary and Conclusions
Existing antibiotic treatment guidelines do not consider 1) important differences in the ecological dynamics among different bacterial species or 2) the diversity of ecological dynamics within the same bacterial species in different habitats in the body. Our results challenge the conventional clinical wisdom that long durations of antibiotic therapy are appropriate for common infections regardless of whether or not they are self-limiting, and regardless of the potential for selection for resistance in the commensal flora. In conclusion, one-size-fits-all dosing regimens do not fit all combinations of organisms and antibiotics, demanding a need to broaden the principles employed in selection of dosing regimens for both existing and future antibiotics to include not only successful treatment of the underlying infection but doing so in a manner that minimizes the pressure to select for increasingly more antibiotic resistant pathogens.