Beyond resistance and tolerance: rethinking evolutionary responses to antibiotics from the perspective of individual bacterial cells

Abstract

Antibiotic resistance and tolerance are thought to be the key phenotypic traits that underpin evolutionary adaptation to antibiotics. However, the concepts of resistance and tolerance can be traced back to population-level assays developed by Fleming in the 1920s. Recent single-cell studies have shown that population responses to antibiotics are driven by a dynamic interplay between cell birth and death, challenging the concepts of resistance and tolerance at the level of individual cells. In this Perspective, we show that ecological models of population growth provide a tool to connect the population-level responses to antibiotics measured by classic assays (ie growth rate) with individual cellular phenotypes (ie birth and death rates). In this model, resistance and tolerance represent emergent properties of bacterial populations that change as an indirect consequence of selection on cell birth and death rates. We argue that this model challenges established concepts and highlights the need to better understand evolutionary responses to antibiotics from the perspective of cellular phenotypes.

Antibiotic resistance and tolerance are measured by determining responses of bacterial populations to antibiotic exposure. In this Perspective, MacLean & Peña-Miller propose that resistance and tolerance represent emergent properties of bacterial populations that arise from selection on cell birth and death rates.

Introduction

Resistance to antibiotics in pathogenic bacteria is driving a global crisis that is undermining human health, food security, and prosperity. Understanding the evolutionary biology of antibiotic resistance increases our ability to predict and prevent the emergence of resistance^1–5. In addition to this applied importance, antibiotics have been used as a powerful model to understand fundamental evolutionary problems related to adaptation, cooperation, and horizontal gene transfer.

Evolution by natural selection occurs when heritable variation in reproductive ability exists in populations, leading to changes in the phenotypic and genetic composition of populations over time. Two key phenotypic traits that are usually discussed in the context of antibiotics are “resistance” and “tolerance”^2,4–12. However, both resistance and tolerance are determined by measuring the responses of bacterial cultures to antibiotics using methods that can be traced back to the pioneering work of Fleming in the 1920s (Box 1). It is challenging to make inferences about the phenotypes of individual cells from responses that are measured from bacterial cultures because even clonal bacterial populations are phenotypically heterogeneous¹³. For example, one clear risk is that a minority of cells drives population-level responses that are not reflective of average cellular phenotypes within the population^13,14.

Imaging techniques that capture the dynamic responses of individual cells provide a way to overcome this limitation. A key insight from this approach has been that a dynamic interplay of cell birth and death occurs in bacterial populations under antibiotic exposure^15–22, a perspective first highlighted, to the best of our knowledge, in a pioneering paper by Coates and colleagues¹⁷. For example, cell elongation and cell division occurs in “sensitive” bacterial cultures that are being grown in media containing antibiotic concentrations above their MIC^15,17,19,23 and cell death occurs in cultures of resistant mutants growing at antibiotic concentrations well below their MIC¹⁴. The number of papers that have described cellular responses to antibiotics remains limited, but these examples clearly highlight the disconnect that exists between the responses of bacterial cells and populations to antibiotic treatment.

A simple way to connect population responses that are measured by MIC and tolerance assays to cellular phenotypes is to decompose population growth into per-cell birth and death rates. In the simplest ecological model, the net population growth rate, r, is given by r = b − d, where b and d denote per-cell birth and death rates, respectively. From this perspective, antibiotics shift bacterial populations in birth–death space: bacteriostatic antibiotics act mainly by decreasing b, whereas bactericidal antibiotics act mainly by increasing d¹⁷ (Fig. 1A). In the context of this model, broth microdilution assays identify antibiotic concentrations where r ≤ 0 (i.e. the MIC), whereas tolerance assays directly measure r, subject to the constraint that r < 0. This representation of antibiotic effects in birth-death space immediately highlights a key limitation of population-based assays: different combinations of b and d can give rise to the same value of r (Fig. 1B). As a consequence, population-based estimates of r cannot be used to infer b and d.

Fig. 1. Representing bacterial responses to antibiotics in birth-death space.

A Antibiotic effects represented as vectors in the (b, d) plane, defined by the per-capita cell division rate b and death rate d. The diagonal line b = d marks the boundary of zero net population growth (r = 0). For any fixed antibiotic concentration A, the net growth rate is r(A) = b(A) − d(A). The MIC is the critical concentration A* at which r(A*) = 0. The grey point indicates a bacterial population in the absence of antibiotic effects. Red arrows represent the effects of different antibiotic classes on this genotype, expressed as changes in effective birth and death rates. Arrow direction and magnitude reflect whether a drug primarily reduces division (bacteriostatic effects), increases death (bactericidal effects), or produces a combination of both. B Schematic dose–response curve showing the relationship between antibiotic concentration (in units of MIC) and net growth rate r = b − d. The solid black line shows a theoretical dose–response that crosses r = 0 at the MIC. Points indicate populations with the same net growth rate but different combinations of birth and death rates: high birth and death (blue), and low birth and death (green). Insets illustrate the corresponding branching processes, where yellow circles represent live cells and grey circles represent dead cells. C Alternative evolutionary trajectories in birth-death space that modify net population growth. Arrows represent two possible evolutionary trajectories under antibiotic pressure: one that increases the division rate b at constant d, and one that decreases the death rate d at constant b. The dashed line indicates the isoline of constant net growth passing through the resulting evolved populations. Although both evolutionary trajectories lead to the same net growth rate, they correspond to distinct changes in birth and death rates, illustrating that similar population-level outcomes can arise through different underlying cellular processes, and therefore different evolutionary dynamics.

Evolutionary change under antibiotic exposure can be visualized in this model as directional shifts in birth and death rates (Fig. 1C). Adaptations that reduce the probability of cell death correspond to downward shifts (decreasing d), whereas adaptations that increase cell division correspond to rightward shifts (increasing b). Because the same population growth rate can arise from different combinations of birth and death rates, populations can cross the r = 0 boundary (the MIC) by following different evolutionary trajectories. Resistance and tolerance are usually considered to be alternative evolutionary responses to antibiotics. However, from this perspective, resistance and tolerance are emergent population-level properties that reflect how antibiotics reshape cellular birth and death processes, rather than intrinsic phenotypes of individual cells.

Box 1 Resistance and tolerance.

The concepts of antibiotic resistance and tolerance can be traced back to the seminal paper reporting the discovery of penicillin⁵¹. Fleming measured the ability of bacterial populations to grow in media containing a range of penicillin concentrations to quantify variation in penicillin susceptibility across bacteria. The basic method reported in this paper has been refined and standardized over time, giving rise to broth microdilution assays that are now used to determine antibiotic resistance⁵⁴. Broth microdilution assays challenge a fixed inoculum (approx. 5 × 10⁵ bacterial cells) with a standardized culture medium containing a two-fold dilution series of antibiotics and determine the minimum antibiotic concentration (or MIC) needed to prevent bacterial population growth, as determined by culture turbidity.

Fleming also studied bacterial responses to antibiotics by measuring changes in the density of viable bacteria over time following antibiotic exposure⁵¹. Pioneering studies showed that the rate of decline in viable cell density following antibiotic exposure differed between bacterial strains with the same MIC, giving rise to the concept of antibiotic tolerance as a trait that is orthogonal to resistance^52,55,56. Tolerance continues to be measured by determining the rate of decline in bacterial population density following exposure to >MIC concentrations of antibiotic³¹.

Phenotypic heterogeneity

Evolutionary models of antibiotic resistance assume that resistance evolves due to selection for rare resistant variants in populations of antibiotic-sensitive cells^6,24–27. In this framework, inspired by classical population genetic models, phenotypes map perfectly to genotype. However, variation in the concentration of cellular components (ie metabolites, proteins, transcripts) generates phenotypic heterogeneity in genetically uniform bacterial populations^13,28. For example, bacterial cells stochastically switch between an actively growing state and a dormant persister state that is highly refractory to killing by antibiotics^29–31. Although persister cells can be very important for infection outcomes, they reflect a small subpopulation of cells (typically <1%), and phenotypic heterogeneity clearly extends beyond persisters^13,28. For example, sub-populations of cells with high efflux pump expression are more likely to survive antibiotic treatment²³ and the random partitioning of multi-copy plasmids during bacterial cell division generates extensive heterogeneity between cells in the expression of plasmid-carried antibiotic-degrading enzymes²¹.

Phenotypic heterogeneity in antibiotic susceptibility within populations is expected to accelerate evolutionary adaptation to antibiotics via two mechanisms. First, the ability of some cells to survive and reproduce under antibiotic treatment prevents population extinction and provides the opportunity for bacterial populations to acquire genetic adaptations to antibiotics (ie mutations or mobile genetic elements). Phenotypic states that are associated with underlying heterogeneity in gene expression show some degree of heritability as cellular components involved in gene expression (i.e. RNA, protein, plasmids) are shared between daughter cells at cell division¹³. Selection on this heritable phenotypic variation provides a simple mechanism for bacterial populations to adapt to antibiotic pressure. For example, exposing a population of bacteria carrying multi-copy plasmids to high doses of antibiotics rapidly selects for cells with a high plasmid copy number²¹.

Given the existence of phenotypic heterogeneity, populations are best represented as a distribution of phenotypes in birth–death space (Fig. 2). One key insight from this perspective is that the responses of populations are predicted to depend not only on mean birth and death rates, but also on the extent of phenotypic variation along each axis. Increased variance in either birth or death rates can increase the fraction of cells with positive net growth, even when the mean population growth rate is negative, thereby enhancing the ability of bacterial populations to adapt to antibiotic treatment. Persister cells, for example, may reflect the tail of the distribution of cell death rate under antibiotic treatment.

Fig. 2. Phenotypic heterogeneity shapes population responses to antibiotics.

Ellipses show the hypothetical distribution of cell birth and death rates in clonal bacterial populations with negative covariance between birth and death. Shading shows the region of the distribution where birth rates exceed death rates (r > 0). Marginal distributions are shown for birth rate (top) and eath rate (right). In all panels, grey indicates cells with r < 0 (susceptible), and yellow indicates cells with r > 0 (resistant).

A second prediction of this view of birth and death rates as continuous traits is that the covariance between birth and death rates can play a key role in shaping responses to antibiotics. Positive phenotypic correlations between cell birth and death rates are expected to constrain adaptation to antibiotics by minimizing the fraction of cells with r > 0. Interestingly, cells with a high rate of growth have an elevated risk of death under ciprofloxacin treatment¹⁹, and two observations suggest that this is a widespread phenomenon. First, many bactericidal antibiotics are most active against actively growing bacteria, demonstrating a link between cell division and antibiotic toxicity^32–34. Second, mutations that compromise metabolism, for example by disrupting amino acid biosynthesis^35,36, are an important mechanism of adaptation to antibiotics, suggesting that reducing growth rate protects cells against antibiotics^7,37,38. Collectively, these observations suggest that cell birth and death under antibiotic treatment are inherently coupled, and this may represent an important constraint on the evolution of antibiotic resistance. As a counter-example, rapid cell growth prevents the intracellular accumulation of the bacteriostatic antibiotic roxithromycin¹⁶, which may help to accelerate adaptation to this drug.

Genetic adaptation to antibiotics

Although phenotypic heterogeneity is important for determining short-term responses of bacterial populations to antibiotics, continued exposure to even low doses of antibiotics generates selection for de novo mutations or acquired mobile genetic elements that increase bacterial fitness under antibiotic treatment^39,40. Recent work has highlighted the importance of genome instability as a source of antibiotic resistance. For example, tandem amplification of chromosomal regions carrying genes that protect against antibiotics or localized hypermutability gives rise to resistance at a high rate^41–43. However, the spread of these novel variants is not guaranteed, even if they are associated with high fitness, because stochasticity in cell birth and death can lead to the extinction of nascent cell lineages of beneficial variants^14,17,18. This process of stochastic extinction can impose a strong evolutionary constraint – for example, the probability that a single cell carrying a plasmid associated with antibiotic-degrading enzymes establishes a successful cell lineage is <10% at antibiotic concentrations as low as 1/8 of the MIC of the plasmid-carrying strain¹⁴. Crucially, the probability of extinction of a nascent lineage can be approximated by d/b in simple branching-process models, providing a simple link between cellular phenotypes and the emergence of genetic resistance¹⁷. All else being equal, stochastic extinction will tend to eliminate nascent lineages with high death rates relative to their birth rate, biasing the evolutionary response to antibiotics towards “conservative” phenotypes with a low death rate (Fig. 3A).

Fig. 3. Evolutionary responses to antibiotic treatment.

A An ancestral bacterial population (grey) adapts to antibiotic pressure by acquiring a series of variants that lead to incremental increases in b, while r remains below zero (blue phenotypes). Resistance is only achieved after the acquisition of a large-effect variant (red phenotypes) that shifts the population to r > 0. In this trajectory, adaptation is characterized by increased tolerance followed by increased resistance. B Schematic Muller plot illustrating evolutionary trajectories. Phenotypes with increasing tolerance (shown in shades of blue) accumulate and diversify over time, while resistant phenotypes (red) appear later and sweep through the population.

Mutations and mobile genetic elements that are beneficial under antibiotic treatment are usually associated with a low growth rate⁴⁴, and this fitness cost plays a key role in restricting resistance at both a within-patient level and at an epidemiological scale^3,4,45–48. Compromised growth rate has often been seen as an unavoidable pleiotropic side-effect of antibiotic resistance that decreases fitness in both the presence and absence of antibiotics^3,44. However, this cost of resistance may emerge due to the adaptive benefits of reducing growth rate under antibiotic treatment and the extinction bias that promotes the success of phenotypes associated with low birth and death rates.

The longer-term dynamics of adaptation to antibiotics are expected to depend on the distribution of fitness effects of variants that increase fitness under antibiotic treatment²⁷. If mutations that lead to large increases in b and/or decreases in d are common, populations will rapidly evolve towards resistance (ie r>0). However, recent work has shown that the suite of genes that can be mutated to increase fitness under antibiotic treatment is much larger than the set of genes associated with increased MIC^7,49. When small-effect mutations are common, adaptation to antibiotics is expected to occur through a sequence of small steps in birth-death space (Fig. 3B). In this adaptive walk scenario, selection to optimize birth and death rates will drive populations to evolve increased tolerance before eventually evolving increased resistance, providing an explanation for the observation that antibiotic tolerance is an evolutionary stepping stone towards increased resistance^8–10,50.

Understanding the genetic basis of adaptation to antibiotics has both fundamental evolutionary importance and practical implications for infectious disease epidemiology. Resistance and tolerance have been considered as alternative traits associated with genes in these analyses^5,6,8,10. One important insight from our approach is that the impact of a gene may well be dictated by the context in which it occurs. For example, a mutation that leads to a small change in b and d will be associated with increased tolerance when it occurs in a low-fitness background and increased resistance when it occurs in a high-fitness background.

Outlook

The concepts of resistance and tolerance are ultimately derived from foundational work in bacteriology that was carried out before tools were available to visualize bacterial phenotypic heterogeneity at a cellular level^51–53. Much remains to be learned about cellular responses to antibiotics, but pioneering studies in this area have clearly shown that antibiotics lead to changes in bacterial birth and death rates that are not captured by conventional whole culture assays. We have argued that simple ecological models provide a way to link these cellular responses with the responses of populations. Although resistance and tolerance are considered to represent alternative phenotypic adaptations to antibiotics, we argue that they represent emergent properties of bacterial populations that change as an indirect consequence of selection on cell birth and death rates.

Important progress has been made by studying the evolutionary responses to antibiotics through the lens of resistance and tolerance. Our goal is not to invalidate or discredit this literature. Instead, we have argued that studying cellular responses to antibiotics can give a more nuanced understanding of the processes that have dominated evolutionary models of antibiotics, including the emergence of resistance, dynamics of adaptation, and trade-offs associated with antibiotic adaptation. One important challenge moving forward will be to uncover how changes to cell birth and death rates mediate bacterial adaptation to antibiotics. The evolution of phenotypic traits that are strongly associated with fitness is often highly repeatable in microbial populations, suggesting that cell birth and death rates will change in repeatable ways in response to antibiotic treatment. Uncovering the biochemical processes that create these genotype-phenotype associations could lead to the identification of novel targets for antimicrobial development and will provide fundamental insights into mechanisms of antibiotic toxicity and cellular robustness.

Author contributions

R.C.M. conceived the study. R.C.M. and R.P.M. authored and edited the manuscript.

Peer review

Peer review information

Nature Communications thanks Minsu Kim and Matthew Shepherd for their contribution to the peer review of this work.

Funding

This work was supported by the UKRI Frontiers grant (EP/Y031067/1) to R.C.M. and a PASPA fellowship from UNAM-DGAPA to R.P.M.

Competing interests

The authors declare no competing interests.