From Petri dishes to patients to populations: scales and evolutionary mechanisms driving antibiotic resistance

Abstract

Tackling the challenge created by antibiotic resistance requires understanding the mechanisms behind its evolution. Like any evolutionary process, the evolution of antimicrobial resistance is driven by the underlying variation in a bacterial population and the selective pressures acting upon it. Importantly, both selection and variation will depend on the scale at which resistance evolution is considered (from evolution within a single patient to the host population level). While laboratory experiments have generated fundamental insights into the mechanisms underlying antibiotic resistance evolution, the technological advances in whole genome sequencing now allow us to probe antibiotic resistance evolution beyond the lab, and directly record it in individual patients and host populations. Here we review the evolutionary forces driving antibiotic resistance at each of these scales, highlight gaps in our current understanding of AMR evolution, and discuss the future steps towards evolution-guided interventions.

1. Introduction

The challenge created by antibiotic resistance

The ability to treat bacterial infections safely and efficiently is a cornerstone of modern medicine and it is currently threatened by the rise of antibiotic resistance. Recent work estimates the annual burden of AMR at nearly 5 million associated deaths (105) and resistance to all classes of antibiotics can now be found in pathogenic bacteria (32). Relying on developing new antibiotics has proven to be insufficient to counter resistance, as the pipeline to develop new antibiotics remain clogged (120), and antibiotics in development often exhibit cross-resistance with existing ones (150). Nevertheless, the rise of antibiotic resistance is not unescapable. Deaths from resistant infections in U.S. hospitals fell by 28% between 2013 and 2019, which can be linked to a combination of interventions including infection control and improved antibiotic stewardship (23). However, determining which specific mechanism led to this decline remains challenging. Even simple interventions, such as reducing antibiotic use, often fail to reduce resistance prevalence (47, 146). Similarly, promising strategies to contain resistance evolution in the lab, such as antibiotic cycling (70), struggle to translate into tangible results in clinical trials (43). Therefore, it is crucial to obtain a better understanding of the evolutionary forces shaping antibiotic resistance to determine which interventions will maximize the efficacy of the antibiotics we already have or will develop.

Evolutionary biology of antibiotic resistance

Antibiotic resistance, defined as the heritable ability of bacteria to survive antibiotic exposure, is an evolution problem. It is a classic example of evolution by natural selection: in the presence of antibiotics, resistant bacteria have a higher chance of survival and replication than susceptible cells. In this context, antibiotic resistant cells are said to have a higher evolutionary fitness, defined as the probability of reproductive success of a genotype in a specific environment. Antibiotic resistance evolution is dictated by the combination of variation in resistance traits existing in the population (the initial presence of resistant genes and mutations) with the selective forces determining the fitness of the bacteria bearing these traits (which cells are more likely to survive in a population).

Evolutionary mechanisms driving resistance emergence

Several evolutionary and ecological mechanisms can generate genetic variation in a bacterial population, and therefore give rise to the presence of antibiotic resistance traits. These mechanisms can be classified in four categories (Figure 1A). First, de-novo evolution of resistance, such as spontaneous mutations due to errors in DNA replication and repair. In this category, we also include structural rearrangements or intra-cellular movement of mobile genetic elements to consider all sources of variation that can occur in a clonal population. Second, bacteria can undergo horizontal gene transfer (HGT), where they acquire DNA from unrelated cells, driven by mobile genetic elements or the bacteria themselves. Additionally, the correspondence between resistance phenotype and genotype is not always straightforward. Alongside genetic variation, bacteria often live in polyclonal or polymicrobial communities, which leads to phenotypic variation coming from interactions with other members of the community. We label this third source of variation under the broad umbrella of ecological interactions. Finally, outside of the lab, bacterial populations are often not closed systems and a substantive amount of variation comes from the migration or translocation of new strains from different environments or parts of the host.

Figure 1:

Scales & evolutionary mechanisms driving antibiotic resistance. A) Genetic and phenotypic variations leading to antibiotic resistance occur through a range of mechanisms: de-novo innovation (mutation or structural rearrangement happening within an individual bacterial cell), horizontal gene transfer (exchange of genetic material between cells), ecological interactions within members of a bacterial community and finally migration of resistant bacteria from different environments or parts of the host. B) Antibiotic resistance evolution can occur at many scales (in-vitro, in patients or within host populations) characterized by variations in selective pressures, timeframes and different contribution of each evolutionary mechanism.

Scales of evolution

Over long timescales, these mechanisms are not independent from each other: genes transferred through horizontal transfer originally arise through mutations, and novel bacterial strains or species arise from variation acquired through mutation or horizontal gene transfer. However, these mechanisms can be distinct at the scales that we focus on to counter antibiotic resistance: evolution within a patient (carriage or infection), or evolution at level of the (human) host population (Figure 2B). Each of these scales is defined by different selection pressures and opportunities for variation, and therefore varying contribution of each evolutionary mechanism to antibiotic resistance.

Figure 2:

Evolutionary dynamics within patients and populations. A: Genetic diversity in infections fluctuates based on duration and type of infection. After an initial bottleneck which heavily constrain genetic diversity, novel diversity can arise through several mechanisms. This process is facilitated in chronic infection where environmental heterogeneity helps generate diverse niches. B. Competition drives the rise and fall of resistant genotypes during infection. Antibiotic exposure impacts bacterial populations by both reducing the population size (bottleneck) and creating selective sweeps for resistant bacteria (full circles). However, these selective sweeps rarely lead to fixation due to clonal interference between resistant mutants (clonal interference). Finally, once the antibiotic selective pressure is removed, resistant bacteria may be outcompeted by sensitive or resistant clones with higher fitness. C. Evolutionary success differs based on the scale considered. At the population level, the success of a bacterial strain is defined by its ability to replicate within a host (replicative fitness) but also its ability to move between and colonize new hosts (transmission fitness).

The study of the evolutionary mechanisms of antibiotic resistance has long been driven by experimental evolution and other experimental approaches in model systems, but the development and popularization of whole genome sequencing in recent years has opened a new door to investigate resistance evolution directly as it happens in patients and populations. Starting from experimental work (Section 1) before going to patients (Section 2) and populations (Section 3), we will review our current understanding of the evolutionary mechanisms driving antibiotic resistance at each of these scales, and how it may inform evolutionary-guided interventions to contain AMR. This review does not claim to be exhaustive, but aims to provide a framework to summarize and conceptualize our current understanding of resistance evolution. In the interest of space, we will primarily focus on antibiotic resistance defined as a heritable trait linked with an increase in antibiotic minimum inhibitory concentration, in opposition to persistence and tolerance, for which we point the reader towards additional reviews (18, 51).

2. Evolution in experimental systems

The study of antibiotic resistance evolution in the lab is nearly as old as the discovery of antibiotics themselves, with pioneering work on salvarsan resistance emergence in the 1910s (2). In this section, we will review the current understanding obtained from experimental studies of the role of mutations, horizontal gene transfer and community interactions in the evolution of antibiotic resistance.

2.1. Evolution through mutation and de-novo innovation

Emergence of resistance from mutations is frequent

Due to ease of study, experimental evolution studies have historically been focused on the mutations and de-novo innovations that arise in a clonal population. Resistance to nearly all antibiotic classes can be readily evolved in-vitro through mutations (83). The mutation rate, defined as the frequency at which detectable resistant mutants arise in a bacterial population, varies for each bacterial species and antibiotic combination, spanning from more than six orders of magnitude (68). It heavily depends on the antibiotic concentration considered: mutations that provide high levels of resistance are usually rare, while mutations that provide low levels of resistance are much more frequent (42, 172).

The mutation rate is also impacted by bacteria physiology, such as induction of the SOS response or in response to starvation (99). As such, antibiotics themselves can act on mutation rates, for example, norfloxacin increases mutation rate by an order of magnitude in E. coli (93). On the extreme is the so-called ‘hypermutator’ phenotype, where mutations in the DNA repair pathway can lead to up to 10 to 1000-fold increase in mutation rate (55). Polyploidy can also accelerate the evolution of resistance by increasing the number of targets for mutations: novel resistant mutations have been shown to occur more frequently when a gene is on a multicopy plasmid versus on the chromosome (128). These mutations may however be subject to genetic dominance, where dominant sensitive alleles can block the impact of recessive resistant mutations (126, 145).

Resistance mutations incur a fitness cost

Given the mutation rate (usually above 10⁻¹⁰ per nucleotide per generation) and the size of bacterial populations found in infections (10¹⁰ cfu), evolution of resistance by single mutation is not generally constrained by mutational supply (68).Thus, the probability of a mutation persisting in the population will be a function of the selective advantage it provides to its host. The selective advantage obtained from a mutation depends on both the benefits (eg. level of resistance) but also of the fitness cost it creates in the absence of antibiotic, as resistance mutations often occur in essential pathways and directly impair their function (102). Antibiotic resistance mutations are often pleiotropic and may also generate indirect costs, for example by restricting niche breadth (63). Another example of pleiotropy is collateral sensitivity, where resistance to one antibiotic leads to increased sensitivity to another (147). For example, modifications of the cell membrane reducing the uptake of aminoglycosides impede efflux pumps, leading to increased susceptibility to several other antibiotic classes (83).

Even when antibiotic resistance mutations are costly, these costs can be mitigated, and even eliminated, by further compensatory mutations (10) which can stabilize resistant mutations in a population even in the absence of antibiotic selection. Compensatory mutations are an example of a broader phenomenon named epistasis, under which the cost and impact of resistant mutations depend on the genomic background in which they occur (155), which in turn impacts the trajectory of resistance evolution (97).

De novo resistance evolution is not limited to mutations

Finally, resistance evolution in clonal populations can also occur through structural rearrangements (insertions, duplications, etc.), at even higher rates than point mutations (5). These rearrangements are facilitated by mobile genetic elements such integrons and insertion sequences (22, 137). Duplications in particular can lead to unstable amplification and overexpression of resistance genes, and generate transient but heritable resistance (110).

2.2. Evolution through horizontal gene transfer

Horizontal gene transfer modifies the evolutionary dynamics of resistance

Horizontal gene transfer is the exchange of genetic material between potentially unrelated organisms. Its contribution to antibiotic resistance is enormous, as it allows the movement of genes across bacterial species and ecosystems as well as the acquisition of multidrug resistance in a single event, due to the frequent co-localization of resistance determinants on the same mobile genetic element (114).

Horizontal gene transfer is a stabilizing force for the presence of resistance in bacterial populations. The cost of resistance from acquired resistance genes is usually less than the cost of resistance from mutations, potentially due to ongoing adaptation of the resistance genes as they move between hosts (156). The genetic linkage between resistance genes on mobile genetic elements also facilitates co-selection of resistance, where selection for a single resistance gene leads to maintenance of the entire multidrug resistance island. Additionally, in the presence of horizontal gene transfer, the persistence of a resistance gene becomes the product of both its ability to transmit vertically (from mother to daughter cells, similarly to chromosomal mutations) as well as horizontally (between unrelated cells). This phenomenon can lead to the persistence of resistance even in the absence of selection if horizontal gene transfer rates are high enough (94, 140), and is often reinforced by the presence of addiction systems preventing the loss of mobile genetic elements (167).

Horizontal gene transfer occurs through several mechanisms

Mechanistically, horizontal gene transfer can be mediated by mobile genetic elements (conjugation and transduction) or promoted by the bacterial host itself (natural transformation). Each of these mechanisms is characterized by varying efficiency, host range and genetic cargo capacity.

Conjugation, especially via conjugative plasmids, is the best understood mechanism of horizontal gene transfer, in which genetic material is exchanged from a donor to recipient cell through a conjugative pilus. Plasmid conjugation rates are extremely variable but can exceed mutation rate by several orders of magnitude (3). Conjugation rates have been shown to depend on donor-recipient relatedness, presence of other plasmids, as well as abiotic factors, such as planktonic or solid environments (15) (3). Similarly to resistance mutations, resistance plasmids often generate a fitness cost for their host (129). The molecular mechanisms driving the cost of plasmid carriage remain poorly understood but have been shown to be highly host dependent (44).

The evolutionary dynamics of transduction and transformation are by comparison much less studied, but both can be upregulated in response to antibiotics (8, 121).

Bacteriophages rarely carry resistance determinants as part of their genomes (with the exception of the recently described plasmid-phages (117)) but they can transfer resistance genes by mistakenly packaging DNA from their host (either chromosomal or plasmid) alongside their own in a process called transduction. Bacteriophages have a narrower host range than plasmids and the rate of transduction is thought to be much smaller than conjugation (107), but recent results point towards the potential for much higher transduction rates (25). Finally, natural transformation, the ability of bacteria to take up and integrate DNA from their environment, is an ancient trait that predates the separation between Gram negative and Gram positive bacteria (74). It is tightly regulated, making its study in the lab challenging (14). Natural transformation has been shown to allow the transfer of both resistance alleles (164) and entire mobile genetic elements, such as integrons and transposons (41) (153).

2.3. Impact of ecological interactions and complex environments on resistance evolution

Bacteria in the environment do not live in the clonal, planktonic cultures that are most often studied in experimental settings, but instead in genetically diverse communities and complex environments which can impact both variation and selection for antibiotic resistance (17).

Community composition alters variation and selection for resistance

The simplest example of the impact of community composition on drug resistance is the effect of bacterial density on antibiotic inhibition (65), with drugs showing decreased or increased inhibition at high density (77). At a fundamental level, the increase in standing genetic variation found in a polymicrobial versus clonal population can accelerate adaptation (7). Diversity in hosts may also stabilize mobile genetic elements by generating source-sink dynamics between species (61).

Interactions between members of a community can modify the impact of selection and either hamper or favor resistant strains. For example, Klümper et al. (81) embedded isogenic strains of E. coli with and without a resistance gene in a pig fecal community and showed the resident community decreased the strength of selection for resistance by either increasing the cost of the resistance gene or protecting the susceptible strain. On the other hand, the killing of a sensitive strain by antibiotic treatment can free up nutrients for the resistant strain, giving it a competitive advantage against the rest of the community (competitive release) (111). Additionally, several antibiotic resistance mechanisms are cooperative, where resistant bacteria can protect the rest of the population (154); for example through the production of antibiotic-degrading enzymes which will also protect non-producer cells (115). These social interactions between strains can lead to complex dynamics in the prevalence of resistance, such as ‘cheating’ where sensitive (‘cheater’) cells do not incur the fitness cost of resistance but benefit from the protection provided by the resistant cells, limiting the spread of resistant genes in the population (170).

Complex environments impact resistance evolution

The physical nature of the environment can heavily shape the evolution of antibiotic resistance. Bacteria often live in biofilms, complex matrices of secreted polymers. Biofilms are characterized by their heterogeneity: the matrix itself can impede the diffusion of both antibiotics and nutrients, leading different parts of the population to experience varying selection pressures (139). As such, biofilms often contain bacterial subpopulations in different physiological states (40). The spatial structure of biofilms also restricts competition between mutants, lessening the impact of selection (130).

The environment may also impact both fitness costs and rates of horizontal gene transfer. While a meta-analysis found no statistical difference between the cost of resistance in-vitro and in mice (156), several studies have shown considerable variation in the fitness and evolution of resistance between bacteria evolved in urine or synthetic sputum versus traditional rich media (67, 82). Similarly, rates of horizontal gene transfer can be variable between in-vitro and vivo systems, with many conjugative plasmids in-vitro failing to conjugate in mice (108), or, on the opposite, high rates of transduction in piglets that were impossible to replicate in-vitro (101).

3. Evolution within individual patients

The developments in whole genome sequencing have made it possible to track antibiotic resistance dynamics during infection and colonization (106). Evolution of resistance has been witnessed during both chronic and acute infection in response to treatment, as well as from bystander selection in microbiomes under antibiotic exposure. In this section, we will review the recent discoveries from genomic studies on the mechanisms driving both variation and selection for resistance within patients.

3.1. Mechanisms driving genetic and phenotypic variation of resistance in patients

Infection types differ in initial genetic variation

Chronic infections, with their large population size and long duration, often contain extensive genetic and phenotypic diversity (38) (Figure 2A). One of the most studied systems of chronic infection is the cystic fibrosis (CF) lung, characterized by P. aeruginosa or S. aureus infections spanning decades (50). Initial colonization of the lung is followed by extensive diversification in both genotypes and resistance profiles, driven by the heterogeneity of the CF environment (36, 98). Enhanced diversity is also frequent in other chronic infections, such as H. pylori where divergent lineages with diverse resistance profiles have been identified in different part of the stomach (133)), and in tuberculosis with genetic variation coming from both mixed infections and de novo variation and sublineages coexisting for years (89).

By comparison, acute infections start from a recent bottleneck, frequently leading to a nearly clonal initial population (26) (159). Recent studies however have highlighted an underappreciated genetic diversity at the onset of acute infections, coming from either mixed infections or genetic diversity within the same clone (87) and correlating with previous antibiotic exposure (26). This initial diversity can directly impact the likelihood of resistance emergence. For example, a study of acute P. aeruginosa lung infections showed antibiotic resistance evolved faster in polyclonal than single clone infections due to selection on pre-existing variation (37).

As infections are frequently seeded from the host microbiome (169) (104), carriage of resistant strains is a strong risk factor for future resistant infections (122). Microbiomes, in locations such as the skin or the gut, are complex environments acting as reservoir of both resistance strains and genes (136) (66). After colonization by a resistant strain, rapid diversification often leads to co-existence of both resistant and sensitive variants. Some examples include the rapid loss of mecA and structural variations in the SCCmec cassette have been shown in S. epidermidis as it colonizes newborns (30); while a sensitive version of the pOXA48 plasmid was found to emerge rapidly and co-exist in the gut of a patient colonized by carbapenem resistant K. pneumoniae (34).

In-patient evolution of resistance can happen through mutations and structural rearrangements

Alongside initial variation in resistance, in situ emergence of resistance during chronic infection has been repeatedly observed, especially through mutations. M. tuberculosis infections are a typical example where in situ resistance evolution is purely driven by mutations: evolution of an infection from fully susceptible to pan resistant has been observed through successive mutational steps (45). Resistance through mutations can also occur in the short time frame of an acute infection (26) (158), with a P. aeruginosa double mutant appearing in less than four days under ciprofloxacin treatment, leading to treatment failure (84).

Additionally, hypermutator strains with enhanced mutation rate have been repeatedly identified in infections, especially in cystic fibrosis (112) and urinary tract infections (35). In UTIs, it was found that hypermutators can contribute up to 50% of the overall genetic diversity (19). Evolution of resistance through mutation in patients is not limited to chromosomal targets but has also been observed on plasmids through mutations increasing plasmid copy number (34) (159). Finally, intracellular rearrangements driven by mobile genetic elements have also been reported to contribute to emergence of resistance, through rampant movement of insertion sequences in P. aeruginosa disrupting antibiotic import (134) as well as integron cassette shuffling leading to increased expression of a beta-lactamase (64).

Horizontal gene transfer drives evolution of resistance in the microbiome

Plasmid mediated resistance is the main driver of resistance in Enterobacteriaceae pathogens, with the gut microbiome acting as a hotspot for the spread of resistance plasmids through conjugation (76) (86). New techniques such as Hi-C are enabling the tracking and quantification of plasmid transfer directly from microbiome samples (79). Kent et al. identified unique networks of horizontal gene transfer within the gut of various patients, with a constant basal level of horizontal gene transfer even in the absence of antibiotic treatment (see also (59)). Otherwise, evidence of the contribution of conjugation to resistance evolution during infection remains sparse: in cystic fibrosis, P. aeruginosa was found more likely to lose DNA than acquire new elements (124) and more plasmid loss than gain was identified in recurrent urinary tract infections (53).

The overall contribution of transduction to the spread of resistance remains to be determined. Potential for transduction-mediated transfer of resistance has been demonstrated on the skin, with rampant transduction of S. aureus on the skin of piglets (101). Similarly, phages isolated from the mice gut post antibiotic treatment increase resistance prevalence when transferred to a naïve population (103). However, while studies report identification of phages containing antibiotic resistance genes from feces (123) and CF lungs (49), it has been shown that identification of resistance genes from viromes is prone to false positives (46).

Finally, examples of the impact of natural transformation on resistance evolution within infection are limited, but extensive homologous recombination has been observed in H. pylori during infection (21), and important pathogens such as S. pneumoniae, A. baumanii and N. gonorrhea are known to be highly competent (14).

Infection microenvironment alters resistance phenotype

Resistance phenotypes within an infection do not always correlate with resistance measured in standard microbroth assays, especially as many infections present as biofilms. For example, successful treatment in a pig model of implant infection by S. aureus required more than hundred times the planktonic MIC (72), and E. coli was found to survive in the bladder at concentration higher than the MIC (11). Mismatch between antibiotic susceptibility in-vitro and treatment success is of particular concern in cystic fibrosis (135), reinforced by the complex interactions between the coexisting bacterial species (16).

Migration and translocation contribute to genetic diversity during infection

In comparison to lab experiments, infections are often not closed systems. Migrations of bacteria within the infection site has been observed in both H. pylori and M. tuberculosis (1, 89). By comparison, bacteria in cystic fibrosis infection present little to no mixing between parts of the lung (75)). Movement of bacteria between body sites is also common, and recurrent translocation from the gut to the bladder of the same E. coli strain is a frequent source of recurrent UTIs (149). Migration can provide genetic diversity even during the short time frame of an acute infection, with gut to lung transmission detected during a lung infection (159). Finally, genetic diversity can be acquired by invasion of new strains from the environment, as shown by the acquisition of resistant strain in the microbiome during travel (165), or the strain replacement by epidemic strains of P. aeruginosa observed in cystic fibrosis, which was shown to provide more genetic diversity than in-situ diversification (161). Similarly, a large scale study of UTI reoccurrence showed that strain replacement, more than mutation or horizontal gene transfer, was the main driver of resistance emergence in recurrent UTIs, and was strongly impacted by antibiotic exposure history (142).

3.2. Dynamics of selection for antibiotic resistance in patients

Antibiotic treatment generates rapid evolution of bacterial populations

Antibiotic use during infection is frequently linked with rapid selective sweeps of resistant mutants. Using targeted deep amplicon sequencing of select resistance mutations, Chung et al. (26) showed in acute P. aeruginosa acute lung infections that both de-novo and pre-existing mutations can increase in frequency within days in response to antibiotic treatment. Similarly, exposure to antibiotics is often associated with an increase in prevalence of resistance mutations and resistance genes in the gut microbiome (33, 162, 168). Alongside increasing the relative frequency of resistance, antibiotic treatment often acts as a bottleneck reducing the population diversity, even in the presence of an already resistant population (1, 152).

As mentioned previously, the selection pressure created by antibiotic treatment is highly dependent on the infection environment: treatment suppressing P. aeruginosa in the lung did not impact the same bacterial population found in the gut of a patient (158). Likewise, cystic fibrosis communities have been found to be very robust to the high dose of antibiotics used during exacerbations (29, 52), potentially due to the complex community and the high prevalence of resistance genes.

Competition drives rise and fall of resistant genotypes

The selective sweeps in response to antibiotic exposure seldom lead to complete fixation of resistance (144) (45) in part due to frequent clonal interference between resistant mutants which slows fixation (88). Sensitive subpopulations can also survive antibiotic treatment (1) (159). This surviving diversity will impact the resistance dynamics once the selective pressure for resistance subsides. High resistance, low fitness mutants can be replaced by either higher fitness, lower resistance mutants (158) or sensitive strains (159). These dynamics can be very quick as well as environment dependent: (26) found that resistant mutations for antibiotics not currently administrated rapidly went extinct within days, and in (159) a meropenem resistant mutant was quickly outcompeted by a sensitive lineage in the gut but remained stable in the lung. Interestingly, how quickly resistance recedes in a bacterial community is impacted by its genomic localization (eg. chromosomal mutations vs mobile genetic elements), with chromosomal mutations decreasing in prevalence in the gut more quickly than mobilizable genes, potentially due to horizontal gene transfer (168)

The immune system shapes resistance evolution

The immune system is an important selection pressure in the host environment (38), which can also indirectly alter resistance evolution. Host immunity can strongly impact bacterial load, reducing it by a factor of 10 before antibiotic treatment even starts (158) and be the main driver leading to the resolution of an infection (158). Serial passaging of A. baumannii in immunodeficient and immunocompetent mice showed that resistance evolution is constrained by the immune system, with resistant mutants arising more frequently in immunosuppressed mice (69). On the other hand, inflammation has been shown to boost horizontal gene transfer in the gut. Salmonella-inflicted enteropathy leads to bloom of Enterobacteriaceae and reduce the prevalence of commensal bacteria in the gut, which otherwise physically block conjugation between Enterobacteriaceae (138).

Antibiotics can also act synergistically with the immune system: beta lactams have been shown to sensitize MRSA to host immune peptides as well as human-made antimicrobial peptides, such as colistin and daptomycin (127). However, the therapeutic use of antimicrobial peptides can be problematic, as resistance to antimicrobial peptides can also lead to cross-resistance to host immune peptide (60) This can result in increased virulence as bacteria resistant to antimicrobial peptides are then able to evade the immune system more efficiently (71).

4. Evolution at the host population level

While de novo emergence of resistance within patients impacts clinical outcome, the main contributor to antibiotic resistance burden remains infections by already resistant pathogens (12, 78). As such, predicting the dynamics of resistance at the population scale requires understanding both emergence and transmission of resistant bacteria.

4.1. Epidemic clones and successful mobile genetic elements drive antibiotic resistance at the population level

High risk epidemic clones are major contributor of resistant infections

Not all bacterial strains contribute equally to the burden caused by resistant infections. The use of whole genome sequencing in epidemiology has revealed that many resistant infections are driven by the clonal expansion of a few high-risk clones in both Gram positive and Gram negative bacteria (160, 163). A recent survey found that 70% of carbapenem resistant K. pneumoniae infections came from only four clonal lineages (31), while E. coli ST131 made on average 40% of all ESBL-producing E. coli isolates in Europe (109).These high risks clones often combine high fitness and transmission ability with extensive levels of antibiotic resistance. E. coli ST131 has been shown to carry ESBL plasmid at little to no fitness cost (132), and can adhere to host epithelial cells (131) as well as displace commensal E. coli even in the absence of antibiotic treatment (27). Similarly, S. aureus ST8:USA300 is resistant to methicillin through near cost-free carriage of a SCCmec element and contains virulence factors that enhance its carriage and competitive ability in-vivo (39) (151).

High risk clones arise through a combination of evolutionary mechanisms

Both mutations and horizontal gene transfer have been shown to contribute to the emergence of high risk clones. Acquisition of resistance to fluoroquinolones through mutations of gyrA, gyrB and parC has been identified several times in epidemic clones (54). The acquisition of ciprofloxacin resistance soon after its introduction in the clinic is thought to be one of the main driver of ST131 success (9). Similarly, horizontal gene transfer has been a key factor in the rise of MRSA clones: acquisition of the mobile methicillin resistance element SCCmec by horizontal gene transfer is estimated to have occurred at least 20 times within S. aureus phylogeny (125). Plasmid-borne resistance determinants are also associated with many high-risk clones, such as CTX-M beta-lactamases in E. coli ST131 (141) and carbapenemase KPC in K. pneumoniae ST258 (100).

These high risk clones are not genetically static but are instead often characterized by a high variability in the resistance determinants they carry. For example, P. aeruginosa ST235 has been identified with more than hundred different resistance elements (113) and E. coli ST131 with different variants of CTX-M (141). Finally, the dominance of specific clones is very dynamic, with now declining prevalence of S. aureus USA300 (119) and replacement of E. coli ST131 by ST1193 in certain regions (118).

Mobile genetic elements can spread resistance globally

Alongside epidemic bacterial clones, the spread of resistance across the globe has also been driven by successful mobile genetic elements. A prime example of this is the class 1 mobile integron whose spread was driven by the use of hospital disinfectants and sulfonamides in the 1930s (56). They are now found between 40 to 70% of Gram-negative bacteria, usually carrying 1 to 5 resistance genes (56). Integron presence is correlated with anthropogenic antibiotic use to such an extent that the prevalence of the integron integrase IntI1 has been suggested as marker for antibiotic contamination in soil and water bodies (57). More recently, the mobilization in the mid 2000s in China of the colistin resistance gene mcr1 on an ISApl1 transposon and IncI2 and IncX4 plasmids was then followed by its rapid global dissemination (157). The contribution of mobile genetic elements to the spread of resistance can also be seen at a local scale, with recent increased awareness of ‘plasmid outbreaks’ occurring in hospitals (48, 116).

4.2. Selection & transmission of resistance within populations

Within- versus between-host evolutionary success

The contribution of epidemic clones and successful mobile genetic elements to resistance highlights how transmission plays a crucial role in the spread of resistance at the population level. The success of a bacterial strain therefore becomes defined by its ability to replicate within a host (replicative fitness) but also its ability to move between and colonize new hosts (transmission fitness) (90) (Figure 2C). The transmission fitness of a new resistant mutant will be highly situational: high fitness mutants arising in a part of the body that does not permit onward transmission (such as the blood) can be considered an evolutionary dead-end (‘adapt-and-die’ vs ‘adapt-and-live mutations (28)) with a transmission fitness of zero. As such, not all resistance arising within patients will contribute similarly to the prevalence of resistance at the population level.

For resistance that arise in a transmissible background and environment, how much resistance will impact transmission remain an open question. We know that resistance can impair colonization and transmission of Campylobacter jujeni in comparison to a sensitive strain (95), and that increased transmissibility is associated with compensatory mutations in rifampicin resistant tuberculosis (58), but it is unclear if resistance is always linked with reduced transmission.

Transmission dynamics shape resistance prevalence

Additionally, transmission generates an additional layer of complexity to the dynamics of resistance. At the population level long-term co-existence between resistant and sensitive bacterial populations has been observed in several species (E. coli, S. pneumoniae (12, 85). This observation is puzzling, as simple ecological models suggest resistance should sweep through a population as long as antibiotic use leads to increased fitness of the resistant strain (12). It has therefore been suggested that the prevalence of resistance may be constrained by heterogenous transmission rates within the host population (13) or linkage with other traits under frequency-dependent selection such as duration of carriage (85). The latter is another great example of how transmission dynamics can impact resistance evolution, as longer duration of carriage leads to higher chance of exposure to antibiotics, increasing the fitness of resistant strains (85). Finally antibiotic use itself can impact transmission of resistant bacteria by disturbing pre-existing bacterial communities, facilitating colonization by resistant organisms which may not otherwise able to invade (91).

Transmission and selection within and between environments

Selection pressure and opportunities for transmission differ based on the environment, visible in the separation between hospital- and community-acquired strains in species such as S. aureus and K. pneumoniae. In the community, exposure to antibiotics is sparse at the level of an individual host, and it has been shown that for many bacterial species, most of the antibiotic exposure they experience is the result of treatment for a condition they did not cause (148). While inpatient antibiotic usage represents only 10% of overall antibiotic consumption (4), the high rate of antibiotic use and the increased density of susceptible hosts make hospitals hotspots for antibiotic resistance evolution (4), with hospital-associated strains usually displaying more antibiotic resistance than their community counterparts (24, 166). In hospitals, transmission of resistant bacteria is enhanced by colonization of the built environment, notably through biofilms in sink traps acting as reservoirs of resistance (143).

Selection for antibiotic resistance may happen not only in human hosts, but also in agriculture settings or in any environment polluted by antibiotic-containing effluents (62). As mobile genetic elements allow the transfer of resistance genes across bacterial species and habitat boundaries, there is a growing concern that the use of antibiotic outside of human health promotes transmission of resistance genes back into human populations, requiring a One Health approach to antibiotic resistance (62). An example of this danger is the emergence of mcr1 from agricultural settings in China, following decades of colistin use in agriculture (now banned) (92, 157). How frequently resistance genes and mobile genetic elements move between agricultural and clinical environments remain however an open question, as several studies have found little overlap in the mobile genetic elements content of agricultural and clinical bacterial populations (96) (20).

5. Perspective: towards evolution-guided interventions

Experimental work has generated not only knowledge about the fundamental mechanisms of resistance evolution, but also a wealth of potential approaches to act at the level of both variation and selection for resistance, ranging from leveraging collateral sensitivity through antibiotic cycling (80) to developing ‘anti-evolvability’ drugs targeting stress-induced mutagenesis (171). But which of these ideas and strategies will make a tangible impact on resistance evolution is much harder to determine and is limited by the many blind spots that remain in our understanding of how resistance behaves within hosts and populations. For example, while cycling and combination of antibiotics have shown potential with mutation-based resistance, their relative efficacy will also depend on the starting genetic variation (73), potential for co-selection of mobile genetic elements and influx of mutants within a population (6) at both patient and hospital levels. New strategies that take into account future evolution will require a more detailed understanding of how resistance evolves in natural environments beyond the laboratory.

The clinical use of antibiotics has effectively provided us with 80 years of experimental evolution, and we now have the technology to observe the results and begin to answer these questions. The rise of affordable and high-throughput whole genome sequencing now allows us to study the dynamics of antibiotic resistance as they occur, both within patients and populations. While we have historically focused on sequencing resistant bacteria, we now have the capacity, through large-scale sampling of both resistant and sensitive bacteria, to understand when resistance evolves as much as when it does not. Understanding what dynamics already constrain the evolution of resistance in populations will allow us to design future interventions that account for evolution and are not so readily countered by it.

Future directions.

• In-vitro: Better experimental systems reproducing both the environment and the diversity found in infections, as well as experimental systems able to investigate the parameters impacting transmission

• In-patient: Better understanding of the evolutionary parameters found in infections: population size, diversity, migration, bottlenecks and selection pressures, which could then be used to improve experimental systems.

• In-patient: Go beyond patient case studies to determine statistics (prevalence, impact on resistance) and risk factors of each evolutionary mechanisms.

• Population-level: Better understanding of the biology driving the evolution of high risk clones, to forecast future threats and guide surveillance

• Population-level: Identify and leverage additional biological mechanisms that can hinder the spread of resistance at the population level, such as phage predation or competition between mobile genetic elements.

Terms and Definitions

Mobile genetic elements

Genomic sequences able to move either within or between genomes.

Minimum inhibitory concentration (MIC)

lowest concentration of antibiotic preventing visible bacterial growth; used to quantify the resistance phenotype of a bacterial population

Tolerance

Ability of a bacterial population to transiently survive exposure to lethal doses of antibiotics

Persistence

Ability of a subpopulation of bacteria to survive antibiotic exposure, these survivors are called persistor cells

Pleiotropy

Phenomenon where a single mutation impacts several apparently unrelated phenotypic traits

Epistasis

Phenomenon where the effect of a mutation is impacted by the presence of other genes or mutations

Integron

Genetic platform allowing the capture, expression and shuffling of mobile gene cassettes often encoding for antibiotic resistance