Within-host evolution of bacterial pathogens

Abstract

Whole genome sequencing has opened the way to investigating the dynamics and genomic evolution of bacterial pathogens during colonization and infection of humans. The application of this technology to the longitudinal study of adaptation in the infected host — in particular, the evolution of drug resistance and host adaptation in patients chronically infected with opportunistic pathogens — has revealed remarkable patterns of convergent evolution, pointing to an inherent repeatability of evolution. In this Review, we describe how these studies have advanced our understanding of the mechanisms and principles of within-host genome evolution, and we consider the consequences of findings such as a potent adaptive potential for pathogenicity. Finally, we discuss the possibility that genomics may be used in the future to predict the clinical progression of bacterial infections, and to suggest the best treatment option.

Introduction

Until recently the extent of bacterial diversity associated with humans and the ability of these bacteria to evolve in response to drug pressures, immune responses and host environment was presumed limited compared with rapidly evolving viruses such as HIV ^1,2. Long-term estimates of bacterial evolutionary rates (over millions of years), calibrated using known dates of ecological and geological events, were of the order of 10^-10 to 10^-9 substitutions per site per year ^3–5, which is low enough to suggest that recently diverged within-host populations should be completely monomorphic. However, recent studies have shown that bacterial short-term evolutionary rates (over months or years) are of the order of 10^-7 to 10^-5 substitutions per site per year and thus much higher than the aforementioned long-term rates ^6–9 — a discrepancy of great consequence that mirrors previous observations in viruses ^10,11. Although bacterial short-term evolutionary rates remain lower than those reported for RNA viruses (10^-4 to 10^-3), bacteria have much longer genomes, which means their per genome evolutionary rates can be comparable or even higher ^9,12,13.

Given these short-term per site evolutionary rates and the short timescale of within-host evolution of a few days to a few years, molecular techniques that sample only a small subset of the genome, such as multi-locus sequence typing (MLST ¹⁴), are of limited utility to study within-host diversity . However, whereas higher resolution approaches had previously not been economically feasible for large population studies , the cost and time required to perform whole genome sequencing (WGS) of bacterial isolates has been substantially reduced over the past few years ^15,16. This has led to a flurry of WGS studies revealing previously unsuspected layers of within-host diversity that evolve measurably over time, so much so that bacteria can adapt to the specific conditions in a given host. Analysing the genome evolution of bacterial pathogens in human hosts in these studies requires a wide range of specific methods, including the collection of clinical samples, treatment for isolation and growth of pathogenic bacteria, DNA preparation, genome sequencing, sequence alignment, genome assembly, variant calling and comparative genome analysis (Box 1).

Text Box 1: methodology for within-host studies.

Isolate collection and whole genome sequencing

A first approach to capturing within-host diversity is to collect multiple clinical samples, either longitudinally or simultaneously, and either at one or several body sites. A second approach is to use a single clinical sample but sequence several separate genomes, for example by culture on a suitable medium, selection of colonies and independent further sub-culturing ⁵⁵. A third approach is to sequence only a single sample, but look for variation in the raw output before assembling the raw sequence data into a genome ¹³⁵. It is however important to bear in mind that no approach is guaranteed to fully sample the within-host pathogen population. Once each isolate has been grown long enough to yield sufficient DNA for whole-genome sequencing, the DNA is extracted and purified. The most popular approach to sequencing is currently sequencing-by-synthesis ¹³⁶, as implemented by the Illumina HiSeq and MiSeq sequencers. Sample preparation for these platforms entails fragmentation of the DNA and multiplexing allows several genomes to be sequenced at the same time by uniquely tagging every fragment from each genome; the typical output is paired-end reads corresponding to the start and ends of the genomic fragments . In the past few years, read lengths have been increasing (up to 300bp), as has the speed and affordability of sequencing protocols. This is especially true of benchtop sequencers, to the point that real-time clinical applications of whole-genome sequencing are emerging ¹⁵,¹⁶,¹³⁷,¹³⁸.

Assembly and variant calling

The most frequently used approach to assemble sequencing reads into a genome is to use a previously sequenced reference genome as a scaffold. Assuming that the reference and target genome are not too distantly related, each read can be mapped onto the reference genome ¹³⁹, creating a so-called pileup. The average number of reads mapping to each unique position of the reference genome is called the coverage or depth, and represents an important measure of sequencing reliability. At each position along the reference genome, variant calling is done by determining whether the mapped reads of the target genome are identical or different to the reference genome . If the disagreement between reads is too high, or the site coverage too low, the site may be left uncalled. If no closely related reference genome exists, lower throughput long-read technology can be used to generate one, as in the study of a Klebsiella pneumoniae outbreak ¹⁴⁰ . Alternatively, de novo assembly can be attempted, without the requirement of a reference genome ¹⁴¹, but is typically able to reconstruct only segments of the genome, so called contigs . In both approaches, repetitive regions of the genome are difficult to assemble ¹⁴², which is especially problematic for studying highly repetitive genes, such as the ones coding for surface proteins ¹⁴³.

Comparative genome analysis

The simplest method for comparative genome analysis is to count the number of positions where two genomes differ. When the two genomes are from the same host, this provides a first idea of the amount of within-host diversity ⁵⁵. When the two genomes are from different hosts, the number of differences reflects the likelihood of transmission between the two hosts¹⁴⁴, although it is difficult to define the threshold above which a transmission event can be definitively ruled out . Comparisons of more than two genomes typically involve the reconstruction of their ancestry using phylogenetic methods. When the genomes have been sampled at different times, a popular approach is to use the BEAST software tool to try to infer both the rate of evolution and a time-scaled genealogy ⁹,¹⁴⁵. Alternatively, the evolutionary rate can be estimated using multiple pairs of sequential genomes from the same hosts ⁹. Recombination can disrupt phylogenetic reconstructions, but new methods are emerging that can account for this in whole genome datasets ¹⁴⁶,¹⁴⁷.

In this Review, we discuss recent work that has advanced our understanding of the genome dynamics of bacterial pathogen populations as they evolve in human hosts. After surveying the various evolutionary processes and forces that affect these populations, we describe how within-host diversity affects epidemiological studies of transmission between individuals. Finally, we consider the contribution of within-host evolution to antibiotic resistance and heightened or reduced virulence phenotypes, as well as to adaptation to the host environment .

Within-host evolutionary dynamics

Until the recent advent of WGS, little was known about the genomic diversity of pathogenic populations of bacteria infecting human hosts. Previous genotyping methods such as pulsed-field gel eletrophoresis (PFGE), variable-number tandem repeats (VNTR), multi-locus enzyme electrophoresis (MLEE) and MLST had comparatively low resolution ^17–19, which could only reveal the presence of multiple colonizing or infecting lineages rather than the evolution of each lineage. For example, mixed colonization by several bacterial types has been shown to be relatively common amongst Staphylococcus aureus carriers using PFGE ²⁰ and VNTR ^21,22. Such mixed colonization is most likely to be the result of two or more separate transmission events. Alternatively, the different types could have been transmitted simultaneously from the same donor who was also multiply colonized. This would require a relatively loose transmission bottleneck, so that substantial diversity from the donor can be transmitted to the recipient, but little is known about the transmission bottleneck of bacterial pathogens ²³. A noteworthy exception where the within-host evolution of a given lineage can be detected by low-resolution approaches (that is, sequencing only a few genes rather than whole genomes) is the stomach pathogen Helicobacter pylori, owing to its unique combination of a long infection time and very high evolutionary rate ²⁴.

Applying WGS to multiple isolates sampled from the same host has allowed us to develop a much more complete picture of the various processes involved in within-host bacterial evolution (for example, for the hospital-associated pathogen S. aureus, see Figure 1). Point mutations provide raw material for this evolution, and studies comparing pairs of genomes sampled simultaneously or longitudinally from the same hosts have confirmed the high rate of within-host point mutation in H. pylori, with ∼30 mutations per year per genome ^25,26 . More modest rates of within-host point mutation have been reported for other bacterial pathogens, such as ∼10 mutations per year per genome for Klebsiella pneumoniae ²⁷, ∼8 for S. aureus ²⁸, ∼2 for Clostridium difficile ^6,29,30, ∼1 for Escherichia coli ³¹ and ∼0.5 in both Mycobacterium tuberculosis ^32,33 and Mycobacterium abscessus ³⁴. These differences between species are partly due to differences in the lengths of the genomes, but also to differences in the per site mutation rates, which are modulated by varying levels of efficacy of the DNA mismatch repair systems (for example, H. pylori lacks several genes involved in this process that are found in most other bacteria ³⁵). A much higher rate of point mutation, known as hypermutation, can also occur when the mismatch repair system becomes disrupted. Hypermutation has long been known to be important for adaptation to new ecological niches in the external environment ^36,37 but has recently been shown also to be important within-host, for example generating excess diversity in a Burkholderia dolosa infection of the lungs of a cystic fibrosis patient ³⁸. A related mechanism is that of phase variation, in which loci that are susceptible to high rates of mutation, for example due to slippage of repetitive motifs, can rapidly modulate the expression of phenotypes over very short timescales, often displaying variability even within a colony^39,40. For example, during long term carriage of Neisseria meningitidis, phase variation reduces the expression of certain genes that encode surface proteins in order to avoid detection by the host immune system ⁴¹.

Figure 1. Within-host evolution of Staphylococccus aureus.

A Staphylococcus aureus infection begins with the transmission of a S. aureus population (blue circles)_from a donor host to a new host. Owing to the transmission bottleneck, only a subset of the within-host diversity in the donor host is transmitted to the newly infected host. Colonization starts at a given site — typically somewhere on the skin — and the pathogen population initially increases in size, with a new mutation arising occasionally (orange circle). The infection may also spread to other sites in the host, such as from skin to nose. When adaptive mutations (red circles) arise — that is, mutations that confer a selective advantage — they can quickly become fixed in the pathogen population through a selective sweep . When a secondary infection from another donor transmits a second S. aureus population (green circles) to the host, the two infecting populations may recombine to form a new genotype (purple circles). Transmission to other hosts may occur at any stage of colonization, as can spread to the blood stream, where the pathogen can cause life-threatening septicaemia.

Another factor that can lead to faster diversification of the genome than point mutation alone is the acquisition of genetic sequences from unrelated organisms through horizontal gene transfer ⁴². One possible outcome is homologous recombination, where a fragment of the chromosomal genome is replaced with a homologous sequence from another cell. This process has an important role in the genomic evolution of many pathogen species, such as Escherichia coli ⁴³, Streptococcus pneumoniae ⁴⁴, Salmonella enterica ⁴⁵ and Campylobacter jejuni ⁴⁶. Recombination is especially potent as a factor in within-host evolution when a mixed infection is present to provide genetic material for import. For example, in H. pylori it has been estimated that when multiple strains are present within a host then homologous recombination can accelerate evolution up to a hundred fold ^25,26,47. An alternative outcome is the gain of novel, non-homologous, genetic material, which is counterbalanced by occasional loss of content through genome degradation ⁴⁸. For example, both gain and loss of genes have been described during the early stages of adaptation of Pseudomonas aeruginosa to the lungs of cystic fibrosis sufferers ⁴⁹. The genomic plasticity of gene gain and loss has an important role in adaptation to changes in selective pressures, with antibiotic resistance genes in many pathogens being found on various mobile genetic elements such as plasmids ⁵⁰, transposons ⁵¹ and bacteriophages ⁵².

Within-host diversity is shaped by random genetic drift ^53,54 (Box 2), and by both purifying selection and diversifying selection. The role of genetic drift might seem counterintuitive when the total number of cells infecting a host can be very high, which dampens drift . However, drift is amplified by several factors, such as isolation of distinct populations in different body parts and fluctuations in population size. Indeed, fluctuations in population size have been found to be an important factor for the within-host evolution of S. aureus ⁵⁵ . The effects of selection on within-host diversity is non-random and can take many forms . At longer evolutionary scales, the effect of purifying selection dominates the evolutionary landscape, as evidenced by a paucity of non-synonymous polymorphisms relative to synonymous polymorphisms (measured by the dN/dS ratio). However, purifying selection is expected to be weaker in within-host populations than in other populations , due to stronger genetic drift and little time available to purge slightly deleterious mutations ⁵⁶. This may be a factor in the seemingly higher rate of short-term evolution observed in within-host populations since proportionally more mutations will be observed that will get purged in the longer term ⁸. Diversifying selection, which favours the evolution of new variants, has been found to be important in several within-host studies ^57–59. For example, certain genes coding for the outer membrane proteins of H. pylori have been found to evolve faster than the remainder of the genome, presumably because these proteins would otherwise be targeted by the host immune system ^25,47,60. Diversifying selection can also be driven by the use of antimicrobial drugs such as antibiotics, and may result in an arms race between the bacterial pathogen and its human host ⁶¹.

Text Box 2: a primer on genetic drift.

Genetic drift is the process whereby allele frequencies fluctuate over time due to the birth and death of individuals in the population. To illustrate this concept, the evolution of a bacterial population was simulated under the assumptions of the Wright-Fisher model (see the figure). In this example, the population has an initial effective population size of 20 bacteria and each generation (horizontal row) is formed by randomly selecting parents from the previous generation (parental relationships are shown by lines connecting individuals from one row to the next in the figure ). At generation 5, the leftmost individual mutates, as illustrated in red, and the new mutation is inherited by all of its descendants . This mutation slightly increases the fitness of its carriers, so that they have a slightly higher expected number of offspring compared with the wild type bacteria shown in black. For that reason, even though its frequency was only 1 in 20 in generation 5, the red allele sweeps through the population, which means that its frequency progressively increases. At generation 14, the whole population carries the red allele, which means that it has become fixed and the black allele has become extinct. At generation 15, a new allele appears, shown in green, which again has slightly increased fitness compared to the red allele, and its frequency starts to increase progressively. However, at generations 19 and 20, the bacteria undergo a bottleneck, so that the effective population size is reduced to only 6 individuals. Bottlenecks enhance the effect of genetic drift, so that sweeping, fixation and extinction all happen at faster rates . By the the end of the bottleneck in generation 21, the green allele has swept through the population and the red allele has become extinct. Similar loss of variation happens when a new population is initiated from a small number of individuals drawn from a larger population, a phenomena known as the founder effect, which corresponds to the transmission bottleneck in the case of the transmission from a donor to a new host.

Host-to-host pathogen transmission

Using phylogenies to reconstruct transmission

WGS of bacterial pathogens has allowed the development of a new approach to tracking direct transmission between hosts, based on the simple idea that if a direct transmission event happened, genomes sampled from the donor and the recipient are expected to be highly similar ^6,62–64. Most studies of recent transmission reported to date have been based on single genomes sampled from each host, and have therefore not taken within-host diversity into account . However, depending on the duration of host carriage, incubation and infection compared with the rate of genomic evolution, within-host evolution can make direct transmission events difficult to reconstruct based on a single genome per host ²³.

For example, consider single genomes sampled from three individual hosts, A, B and C. The two genomes sampled from B and C may be more closely related to each other than to the genome sampled from A; however, perhaps counter-intuitively, this does not rule out the possibility that A infected both B and C through a lineage that is different from the one that was sampled from A (Figure 2). In other words, the genetic variants are representative of the phylogenetic relationships between the sampled genomes, but these may not directly reflect the order and timing of transmission events from donor to recipient. This problem is similar to the relationship between species trees and gene trees caused by incomplete lineage sorting ⁶⁵. Formally establishing the relationship between phylogenetic and transmission trees requires explicit modelling of the transmission bottleneck and within-host dynamics, which increases the accuracy of the results but also reveals much uncertainty in who infected whom based on single genomes per host ^2,66,67.

Figure 2. Effect of wihtin-host evloution on the reconstruction of transmission events .

This example illustrates the benefit of sampling several genomes, rather than a single genome, from each host when reconstructing a transmission network . In this example, three hosts (A, B and C) are infected with a bacterial pathogen, with host A having become infected first and having directly transmitted to B and then to C, as indicated by the vertical arrows. If only a single genome is sampled from each host (see inset), their ancestry is only weakly informative about who infected whom. In the example shown, the ancestry of the genome sampled from host A is such that the genomes sampled from host B and host C are more closely related to one another than with the genome sampled from host A, which might be mistaken as evidence for host B having infected host C (or vice versa) . This error can be avoided if within-host evolution is accounted for, but doing so with only a single genome sampled from each host results in substantial uncertainty in reconstructions of transmission . However, when several genomes are available for each host (coloured circles in the transmission tree), an ancestry analysis becomes much more informative about the transmission events . In the example shown, the three genomes sampled from host B form a clade that is closely related to a clade of genomes sampled in host A, as do the four genomes sampled from host C, correctly suggesting that A infected both B and C . Different colour circles represent genetic variation in the pathogen genome following within-host mutations .

By capturing the diversity within each host — for example, by sequencing multiple genomes per host sampled either simultaneously or longitudinally — within-host evolution can reconstruct transmission events with much more certainty than sampling single genomes . This approach greatly clarifies who infected whom; for example, in cases where host A has infected host B, the diversity of pathogen genomes sampled from B should be a clade contained within the diversity of pathogen genomes sampled from A (Figure 2), a principle that has been known and applied previously in other settings, such as in studies of HIV infection ^68,69. In a pioneering study that applied this approach to bacteria, an outbreak of B. dolosa among 14 cystic fibrosis patients over ¹⁶ years was investigated by sequencing a total of 112 isolates, clearly revealing likely donors and recipients of transmission events ⁵⁹. A similar study used 168 consecutive isolates of M. abscessus from 31 patients of an adult cystic fibrosis centre in the UK, and found that 11 patients were likely to have infected one another during hospital visits ³⁴. Deep sequencing of S. aureus was crucial to revealing the likely role of a healthcare worker in spreading infections at an English neonatal intensive care unit ⁷⁰, tracing the source of an outbreak in an adult intensive care unit in Thailand to a single individual with high within-host diversity ⁷¹ , and reconstructing the transmission network between animal patients and staff at an English veterinary hospital ⁷² .

Identifying the cause of recurrent infections

A problem related to that of reconstructing transmission networks concerns the ability to distinguish reinfection from relapse in individuals who suffer subsequent episodes of disease. For example, in a study of African patients with recurrent invasive non-typhoidal Salmonella infection, 25 out of 32 recurrent episodes were caused by identical or almost identical consecutive strains, suggesting relapse, whereas the remaining seven cases showed many differences between consecutive strains, suggesting reinfection ⁷³. The same approach was applied to recurring cases of C. difficile infection, and relapse was found to be about five times more frequent than reinfection in both the UK and Ireland ^74,75. The question of relapse versus reinfection is especially relevant for M. tuberculosis infections, which are more frequent in people living with HIV and can often recur after seemingly successful treatment. In a study focused mostly on HIV-negative patients, 33 clear cases of relapse were found compared with only three reinfections ⁷⁶. Another study based on a cohort with higher HIV infection prevalence identified 55 relapsed and 20 re-infected individuals, with the latter being more likely to be HIV-positive ⁷⁷. This study also found that lineage-2 was more likely to cause reinfections whereas lineage-3 was more likely to cause relapse, echoing differences in adaptive strategies that have also been observed when comparing transmissibility between lineages of M. tuberculosis ⁷⁸.

Evolution of antibiotic resistance

Natural selection within individual hosts

Following successful transmission, a pathogen must survive in the new host in order to transmit again and continue the life cycle. Survival within the host poses many challenges — including physical barriers to colonization and infection, competition with the native microbiota, containment by the immune system, basic sanitation and medical intervention — and the relative importance of these forces may differ from one host to the next. However, the ability of bacteria to adapt rapidly to selection pressures such as antibiotic treatment, even within a single host ^25,26–34, is highlighted by the rapid rise in antibiotic resistance, exacerbated by overuse and misuse ⁷⁹.

Evolution of antibiotic resistance within individual hosts epitomizes the evolutionary principle of a selective sweep, as the survival advantage gained from the resistance-conferring gene (or mutation within a gene) in the presence of antibiotic treatment increases the frequency of the gene (or mutation) in the within-host population . Furthermore, where resistance-conferring genes are carried by plasmids or other mobile elements, horizontal gene transfer can accelerate the rate of spread .

Until the advent of WGS, the characterization of bacterial microevolution was the purview of experimental studies ⁸⁰. WGS captures this process in extraordinary detail, and as a result has now provided valuable insights into many examples of the evolution of antibiotic resistance within-host . As a result, the occurrence and spread of individual point mutations that increase resistance and contribute to treatment failure can often be pinpointed , as in the evolution of vancomycin-resistant S. aureus in an initially vancomycin-susceptible bloodstream infection ⁸¹. Rifampicin resistance-associated mutations in rpoB, a β-lactam resistance-associated mutation in blaR1, and mutations in several genes associated with reduced vancomycin susceptibility were identified in bacteria isolated from the patient only after treatment with the respective antibiotic . Mutations progressively accumulated over time, concomitant with a stepwise decrease in vancomycin susceptibility ⁸¹.

The sheer potency of bacterial evolution was revealed by the evolution of extensively drug resistant (XDR) M. tuberculosis from a wholly drug-sensitive ancestor in a single patient over just 3.5 years ^82,83. Resistance evolved to seven drugs to which the patient was exposed: rifampicin (to which resistance was conferred by mutations arising in rpoB), isoniazid (katG), ethambutol (embB), streptomycin (rss and gid), ofloxacin (gyrB), ethionamide (mshA) and amikacin (rss). In most cases, resistance-conferring mutations arose in one or more lineages in the within-patient population only after several months of treatment with the respective antibiotic. Ultimately, in an extreme example of clonal interference, a single multidrug resistant lineage prevailed, displacing the others. The spread of advantageous resistance mutations elevated the measured substitution rate 14-fold , and propelled the substitution of numerous synonymous mutations on the same genetic background as the advantageous mutations, in an example of a process known as hitchhiking.

Some naturally evolved resistance mutations appear to be pleiotropic, leading to multifaceted changes in bacterial phenotype, including resistance to drugs to which the population has not been exposed ^81,84,85. In five patients, S. aureus with reduced vancomycin susceptibility evolved from initially susceptible strains following fewer than six weeks of vancomycin treatment. Non-synonymous mutations in the WalKR two-component regulator ⁸⁴, which has a central role in controlling cell wall metabolism⁸⁶, were found in four of these cases, as well as in six of eight unrelated isolates with reduced vancomycin susceptibility . The majority of isolates with reduced vancomycin susceptibility also demonstrated reduced daptomycin susceptibility, despite no exposure to this second antibiotic ⁸⁴. Allelic exchange experiments showed that mutations in walKR led to morphological and transcriptional remodelling, including reduced autolytic activity, increased cell wall thickness, reduced biofilm formation and attenuated virulence in a wax moth model of infection ⁸⁴. This incidental pre-adaptation to other antibiotics, a phenomenon known as cross-resistance, presents a particular cause for concern by further limiting treatment options.

Revealing the adaptive potential of bacterial pathogens

The ability to study the evolution of antibiotic resistance in forensic detail using WGS has enabled the discovery of novel mutations and mechanisms of resistance that may help predict treatment failure and guide the development of new drugs ^87–93 . These in vivo surveys of within-host evolution present a complementary approach to experimental studies . Although it is not possible to control for all variables, observing natural evolution in within-host populations has the advantage of incorporating complexities such as strain differences, the host environment and realistic fitness trade-offs, which has highlighted the formidable adaptive potential of bacterial pathogens .

Certain genes have been identified repeatedly as undergoing adaptive evolution in response to antibiotic treatment, with independently arising mutations often hitting different sites in the same gene. Common targets of selection include RNA polymerase gene rpoB in response to treatment with rifampicin (which inhibits RNA polymerase) in several species ^85,94–96, DNA gyrase gene gyrA ^38,96–99 and DNA topoisomerase IV gene grlA ¹⁰⁰ in response to fluoroquinolones (which target topoisomerases) in various species, and cell wall metabolism genes walKR and vraRS in response to treatment with vancomycin (which inhibits cell wall synthesis) in S. aureus ^81,84,101.

Reports of convergent evolution in independent patients exposed to the same drugs offer the strongest evidence for adaptation. A study investigated the evolution of resistance to a wide range of drugs in 123 M. tuberculosis strains representing transmission clusters and epidemiologically unrelated cases. Resistance evolved independently up to 20 times to the first line drugs isoniazid, pyrazinamide, ethambutol and rifampicin by substitutions in katG and inhA (isoniazid), pncA (pyrazinamide), embB (ethambutol) and rpoB (rifampicin). Identical, independent substitutions were detected at two sites in rpoB and one in embB ⁸⁷. The observation that resistance to common antibiotics has evolved not once, but many times in parallel, underlines the adaptive potential and repeatability of bacterial evolution.

Given the rapidity with which bacteria can respond to selection pressures by within-host evolution , it may seem surprising that antibiotic resistance has not spread yet more quickly. It has been suggested that the fitness costs of resistance might explain this discrepancy, because resistance-conferring substitutions in key enzymes may reduce the efficiency of replication and transcription, while specialist resistance-conferring proteins may be costly to produce ^88,102. However, compensatory mutations may arise that counteract fitness costs of antibiotic resistance (Figure 3). WGS allowed a joint investigation of naturally evolving rifampicin resistance-conferring and compensatory mutations in tuberculosis ⁸⁸. Putative compensatory mutations were detected in 38 genes, with a particular enrichment of mutations affecting the interface between RNA polymerase subunits α and β′. High in vitro fitness was demonstrated in strains bearing compensatory mutations subject to strong convergent evolution across lineages, demonstrating that a cost to resistance is not inevitable.

Figure 3. Within-host adaptive potential during antibiotic exposure .

A susceptible bacterial strain (blue circle), when exposed to an antibiotic, will be highly likely to be killed, but may occasionally survive by evolving into a resistant strain (orange circle). Resistance usually has a high fitness cost (inset), so that resistant strains disappear when not exposed to the antibiotic. However, resistant strains can evolve compensatory mutations (green circle) so that they remain resistant without the associated fitness cost. Such compensated strains pose a serious danger to public health, because they do not disappear simply as a result of antibiotic disuse. Alternatively, strains may evolve adaptability (yellow circle), allowing them to quickly switch resistance on or off and therefore avoid the associated fitness cost, presenting a similar risk to public health as that presented by compensated strains.

Even when fitness costs are inevitable, bacteria can evolve adaptability. A well-studied form of adaptability is the previously mentioned phase variation mechanism, in which frequent, reversible, genetic changes determine expression of particular genes, such as those encoding surface antigens . Rapid mutation allows expression of these genes to be quickly adapted to changing selection pressures. Another form of adaptability that is increasingly recognized is heteroresistance . For example, population analysis profiles (PAPs) reveal that, remarkably, most methicillin-resistant S. aureus (MRSA) strains are heteroresistant, meaning that the vast majority of cells have only low- or moderate-level resistance in the absence of exposure, whereas cells exhibiting several hundred-fold increased resistance are present only at very low frequency ⁹⁵. Upon exposure to methicillin, the rare, highly resistant cells can rapidly sweep through the population. Genome sequencing has found that high-level resistance can be conferred by any one of a large number of mutations across the S. aureus genome, many of which affect gene expression . Although this indicates a large target for selection, the over-representation of mutations in genes involved in transcription and stringent stress response suggests a degree of parallel evolution ⁹⁵.

Heteroresistance demonstrates the capacity for large bacterial populations in the host to mutate and harbour potentially advantageous mutations, facilitating adaptability. The phenomenon of heteroresistance in a range of antibiotics across several pathogens including S. aureus, M. tuberculosis, S. pneumoniae and enterococci ¹⁰³ indicates that within-host evolution is not the exception, but the norm, in bacteria and may be a leading contributor to monotherapy treatment failure ¹⁰⁰. As the potential for adaptability seems to vary between bacterial strains, capacity for within-host evolution might even explain differences in the success of global lineages ¹⁰⁴ .

Adaptation to the host environment

In the absence of antibiotic pressure, bacterial pathogens must still overcome a variety of challenges in the host, carving out a successful niche by securing nutrient sources and evading killing by other microorganisms and the innate and adaptive immune system. Such challenges represent opportunities for adaptation if mutations arise that confer survival or reproductive advantages. WGS is uncovering diverse signals of adaptation to the host environment, and this is shedding new light on the main forces that shape bacterial populations in the host.

Opportunities for adaptation are particularly abundant in infections of immunocompromised hosts by opportunistic pathogens that are adapted to normally living elsewhere. Infections of cystic fibrosis patients with bacteria that are not usually pathogenic in humans have been particularly well studied (Figure 4). In vitro experiments have described an initial period of rapid adaptation to new environments that slows down as the population improves in fitness ¹⁰⁵. Similar patterns of rapid adaptation have been detected in opportunistic infections of patients with cystic fibrosis in the form of signals of parallel evolution. In an outbreak of B. dolosa among patients with cystic fibrosis, signatures of selection in the form of convergent evolution were detected in genes responsible for O-antigen repeat expression in the lipopolysaccharide coat, oxygen-dependent regulation and resistance to ciprofloxacin ⁵⁹. O-antigens elicit strong immune responses and typically exhibit high variability that probably reflects selection for immune evasion. It is thought that changes in oxygen-dependent regulation may be crucial for adaptation of free-living organisms to the cystic fibrosis lung, where the build up of mucus and biofilms has reduced the availability of oxygen ⁵⁹.

Figure 4. Within-host evolution of pathogens in the lungs of a cystic fibrosis patients.

Infection of the lungs of a cystic fibrosis patient begins with transmission from the environment or the skin of a donor, and progresses with a rapid increase in the size of the pathogen population . Mutations occasionally occur, some of which may be adaptative mutations that spread through the pathogen population in a selective sweep (red circles). Another important mechanism of genome evolution for these pathogens is hypermutation, whereby a strain loses the function of its mismatch repair machinery and thus becomes a hypermutator, with a mutation rate that is increased several fold, thus greatly increasing its evolutionary potential . As the infection progresses, the pathogen colonizes all ecological niches within the cystic fibrosis lungs, and separate adaptation to each niche leads to the coexistence of differentially adapted lineages.

Patterns of convergent evolution have been used to identify pathoadaptive mutations in P. aeruginosa infections in cystic fibrosis lungs, highlighting the roles of genes encoding transcriptional regulators, lipopolysaccharide biosynthetic protein, outer membrane antigen and antibiotic resistance in pathogenicity ^58,96,99. In some cases, within-host adaptation may be facilitated by hypermutators ^96,97,99,106, which accelerate the rate of adaptation by increasing the supply of genetic novelty, a greater proportion of which is likely to be beneficial in a novel environment ¹⁰⁷. Selection for changes to transcriptional regulators such as sigma factors and two-component systems indicate that transcriptional reprogramming may be a key element of adaptation to the host environment in opportunistic pathogens ^96,99. Such changes to gene expression can cause rapid phenotypic change with little evolution in terms of genetic mutation. In general, dissecting the molecular mechanisms that drive signals of within-host adaptation is difficult because, unlike in the case of antibiotic resistance, the phenotypes are difficult to recreate in vitro, owing to the complexity of the within-host environment and the potential role for host–pathogen interactions . However, an experimental CF lung, which may negate some of these difficulties, has been used to study one candidate for adaptive transcriptional reprogramming, the iron-scavenging Pseudomonas haem utilization system. The increased expression of this system induced by promoter mutations was demonstrated to be advantageous to P. aeruginosa growth in the presence of haemoglobin ¹⁰⁸. Although such experiments are challenging, they provide useful validation of the findings of observational studies.

Within-host population structure is likely to be an important modifier of adaptation to the host environment. In homogeneous environments, drift and positive selection are expected to lead to the evolution of one prevailing lineage (Box 2). However, within-host heterogeneity can facilitate the existence of multiple coexisting lineages, either through neutral processes such as genetic isolation of subpopulations, or adaptive processes such as niche differentiation (Figure 4). An investigation of within-host population structure in five B. dolosa-infected CF patients found stable coexistence of multiple lineages for more than five years and evidence of parallel evolution even in individual patients, manifest as multiple substitutions in genes with functions in antibiotic resistance, outer-membrane synthesis, iron scavenging and oxygen sensing. These results uncover previously unsuspected population structure in the infected lung ³⁸. Sequencing of a P. aeruginosa infection sampled over a 32-year period revealed that the infecting strain had rapidly diverged into distinct sublineages, with unique functional signatures and threefold variation in the rate of adaptation ⁹⁷. Differences in the frequency of these sublineages in the sinuses, upper airways and lower airways, and their stable coexistence over several decades, suggested specialization to distinct within-host niches. Sublineages differed in growth rates, mucoidy and production of proteases, with one of the clusters displaying a phenotypic signature characteristic of chronic infections, namely longer doubling times, mucoidy and the loss of protease production ⁹⁷.

Notwithstanding the numerous examples of within-host adaptation revealed by WGS , most of the genome nevertheless remains subject to purifying selection. Even within individual hosts, the genome-wide dN/dS ratio typically lies below its neutral expectation of one and thus is indicative of purifying selection . In a B. dolosa outbreak in CF patients, singly mutated genes showed a dN/dS = 0.63, significantly below one ³⁸, similar to the dN/dS = 0.56 and 0.66 reported in P. aeroginosa outbreaks in CF patients ^96,97 and dN/dS = 0.55 reported for S. aureus nasal populations colonizing individual hosts ⁵⁵. Taken together, these figures indicate that the prevailing effect of natural selection in the host is to conserve functionality for the majority of genes, with adaptation acting only on some genomic positions in a subset of genes.

Within-host adaptation might be less frequent in common bacterial pathogens of humans that have already had many generations over which to adapt to an infectious lifestyle. A study of asymptomatic S. aureus nasal carriage in healthy adults concluded that adaptive evolution was rare, revealing only a weak signal of convergent evolution across thirteen carriers. Notably, multiple non-synonymous or protein-truncating mutations were observed in surface anchored proteins Ebh and SasC and secreted enterotoxin Seg ⁵⁵. Indeed, surface and secreted proteins are common antigens that interact directly with the adaptive immune system leading diversifying selection to favour novel epitopes that can evade recognition. In another example, a longitudinal study of asymptomatic H. pylori carriers reported excess horizontal gene transfer in the Hop family of outer membrane proteins ²⁵. It may be that most within-host adaptation in common human pathogens occurs within antigenic loci, although well-known examples of recurrent evolution in other genes, such as loss-of-function agrC and lasR mutants that knock out quorum sensing in S. aureus and P. aeruginosa, respectively, suggest this will not be the full story ¹⁰⁹ .

Selection pressures within the host are difficult to dissect. Conflicts arise not only with the host, owing to the need for nutrient acquisition, immune evasion and onward transmission, but also with other bacteria, owing to direct competition for resources. Complex selective forces may also be at work as a result of social dynamics within a bacterial community. For example, whereas loss-of-function mutations might in many cases signal host adaptation¹¹⁰, the loss of siderophore production in P. aeruginosa during long-term infections in CF patients can instead be driven by cheating behaviour¹¹¹ . In some patients, cheats that no longer synthesize the iron scavenging molecule pyoverdine still retain the pyoverdine receptor, allowing them to uptake iron scavenged by pyoverdine produced by co-operative bacteria. Only when the co-operators are lost from the population do the cheats lose the pyoverdine receptor.

Evolution of disease severity

Dose within-host evolution increase virulence?

A key question that follows from the discovery that bacteria rapidly evolve on timescales relevant to colonization and infection is what the effect of this within-host bacterial evolution is on disease manifestation. In other words, does bacterial evolution in the human body increase virulence? Dramatic phenotypic changes and global transcriptional remodelling in the host support the feasibility of such a notion ^{84,85,108,112,113}. For example, a single spontaneous mutation in relA in a persistent S. aureus infection was sufficient for permanent activation of the stringent response, which diverts cellular resources towards survival during nutrient limitation by instigating widespread regulatory changes, including upregulation of amino acid synthesis and protease production. In this pathogen, the stringent response induced multifactorial phenotypic changes including reduced growth, smaller colony size, reduced attachment to human cells and attenuated virulence in a wax moth model of infection⁸⁵ . WGS provides an unprecedented opportunity to reveal how within-host bacterial evolution is associated with the progression of disease in a wide variety of bacterial pathogens.

Many clinically important bacterial pathogens are predominantly commensals of humans, including S. aureus, N. meningitidis, S. pneumoniae, H. pylori and E. coli. As carriage is common and invasive disease is not an obligate part of the lifecycle, most transmission may be asymptomatic, with disease being caused by previously carried bacteria rather than transmission. In the case of S. aureus, 82% of serious infections of deep tissue are self-infections, in the sense that the invasive strain matches the nasally carried strain ¹¹⁴. Although the triggers for invasive disease are likely to be complex and multifactorial, and include the general health of the patient and host genetics, an important question in light of rapid within-host evolution is whether the carried bacteria can evolve to become more virulent.

In one long-term S. aureus nasal carrier who developed a severe bloodstream infection, WGS charted the genomic changes that accompanied the transition from asymptomatic carriage to invasive disease ²⁸. Over thirteen months, multiple isolates were sequenced from each of a longitudinal series of samples, revealed 30 point mutations and four insertions or deletions. The population evolved at a steady rate except for a cluster of mutations preceding the transition to disease. Eight mutations differentiated the original nasal population from the bloodstream population, of which half were protein-truncating, including a mutation in rsp, an AraC-family transcriptional regulator (AFTR) previously implicated in pathogenicity ¹¹⁵. AFTRs are regulators of carbon metabolism, stress response and virulence that respond to changing environmental conditions such as antibiotic use and oxidative stress¹¹⁵. In N. meningitidis, a loss-of-function mutation in the AFTR mpeR is associated with the hypervirulent ST 32 complex ¹¹⁶. A statistically significant excess of protein-truncating mutations accompanied the progression of S. aureus to invasive disease in this patient. However, both for this case and more generally, whether disease progression is driven by bacterial evolution (or whether, conversely, bacterial evolution is driven by the worsening health of the patient) remains an open question ^28,117.

Attenuation of virulance as an evolutionary strategy

Rather than exacerbating virulence, there is evidence that bacterial evolution in the host can attenuate virulence ^57,118. Burkholderia pseudomallei causes the potentially life-threatening disease melioidosis, and is not considered a commensal. However, one patient out of 707 survivors in a 23-year study in Darwin, Australia, developed persistent asymptomatic carriage ⁵⁷. Two isolates sampled 11.5 years apart differed by 23 point mutations, including a protein-truncating mutation in the gene encoding universal stress response sigma factor RpoS. Four deletions in chromosome two removed 221 genes during this time, including some involved in metabolism, survival outside the host and pathogenesis. Both isolates showed loss-of-function mutations in the essential virulence factor wcbR, a component of the capsular polysaccharide I locus, suggesting early attenuation of virulence may have promoted long-term persistence in this patient.

A recent study on H. pylori virulence sought to explain why two Columbian regions separated by only 200km and with similarly high prevalence of infection showed a 25-fold difference in the incidence rates of gastric cancer ¹¹⁹. In the region where disease was rare, most individuals were of African descent and colonized asymptomatically with African lineages of H. pylori, whereas in the region with high disease rate, African ancestry was low in the human population (∼3%) but higher in the bacterial population (∼20%). The highest risk of disease was found in individuals of non-African descent who were infected by H. pylori with substantial African ancestry. This suggests that, in Africa, H. pylori and its human host have coevolved towards lower virulence, which could also explain why the high prevalence of H. pylori in Africa does not seem to correspond to similarly high levels of disease ¹²⁰. According to the adaptive trade-off hypothesis, host–pathogen coevolution can lead to the evolution of disease with reduced severity ^121–123. H. pylori is often transmitted in families or communities of closely related individuals ^26,124,125, and the virulence optimum is expected to be particularly low for pathogens transmitted in this way because their long-term survival is linked with that of the host ¹²⁶.

Strains that have evolved attenuated virulence may even provide a therapeutic tool. Urinary tract infection (UTI) by E. coli is a severe and potentially life-threatening condition. However, E. coli can also be carried asymptomatically in the bladder in a state known as asymptomatic bacteriuria (ABU). ABU strains have been extensively used for therapeutic urinary bladder colonization in chronic UTI patients¹²⁷. Previous work showed that 50% of long-term ABU strains evolved from uropathogenic strains in which genome reduction and inactivating mutations attenuated virulence ¹²⁸. By studying six patients colonized with a single prototypic ABU strain, WGS revealed further evolution of the strain following therapeutic inoculation that suggested within-host adaptation specific to each individual host. In some patients who required repeated therapeutic inoculations with the prototypic strain, mutations were observed repeatedly in loci involved in iron uptake (fecIR promoter), oxidative stress response (frmR) and osmoregulated periplasmic glucan synthesis (mdoH). The repeatability of evolution within individual hosts suggested the existence of a characteristic signature of adaptation to individual hosts sometimes known as host imprinting . Ongoing loss of gene function suggested that progressive evolution towards commensalism rather than virulence was favoured in these ABU strains¹²⁷.

Conclusion

By enabling the study of microbial evolution right inside our own bodies, WGS has revealed a remarkable degree of adaptability of bacteria in the human host . Even the notoriously slow-growing and slowly evolving ³² M. tuberculosis is capable of extraordinarily rapid adaptation in response to antibiotics, with one infection evolving resistance to seven antibiotics over a 42-month period ⁸². The numerous examples in which resistance has evolved in response to antibiotic therapy (not only in M. tuberculosis ^82,88 but also S. aureus ^85,95, E. coli ¹²⁹, K. pneumoniae ^92,130 and P. aeruginosa ^96,99), together with the examples of convergent evolution in opportunistic pathogens such as P. aeruginosa ^96,97 and B. dolosa ⁵⁹, shows that rapid within-host evolution is common across disparate species. This raises some important questions. First, what are the requirements for rapid within-host adaptation? Second, is within-host adaptation of bacterial pathogens sufficiently rapid to influence disease outcome? If so, which conditions favour increased virulence in the host, and which conditions favour attenuated virulence?

Rapid within-host adaptation depends on a range of factors, notably the mutation rate of potentially beneficial mutations, the effective population size and the fitness advantage of mutants ¹³¹. The larger these quantities are, the faster is the rate of adaptation of the population as a whole. In M. tuberculosis, recent studies have shown that both the mutation rate and effective population size within the host are very small. The genome-wide mutation rate barely registers one mutation every two years ^32,33,82, and diversity is so low that isolates sampled at the same time differ by an average of only 0.5 mutations ³³. Assuming, for illustrative purposes, a generation length of 24 hours¹³², this is equivalent to an effective population size of fewer than 200 reproductively viable cells. Target sizes for selection, in terms of the number of possible beneficial mutations, differ among drugs but commonly involve around a dozen sites within a particular gene. To observe adaptation, in the presence of antibiotics, in a timescale of months then requires a thousand-fold or greater daily replicative advantage of resistant over susceptible cells. This is plausible for antibiotic-induced selection, where the advantage of resistance over susceptibility can in some cases be of this order of magnitude . However, these values suggest that M. tuberculosis is unlikely to evolve within-host adaptations during the course of a single infection unless there are many potentially beneficial mutations or unless adaptation is a matter of life or death .

For within-host adaptation to influence the outcome of a single infection, selection must therefore be relatively strong, although other pathogens are likely to be more adaptable than M. tuberculosis. For example, S. aureus exhibits approximately ten-fold higher rates of mutation and within-host diversity than M. tuberculosis, at least during asymptomatic carriage ⁴², suggesting it has greater capacity for adaptation during the course of infection — not just to antibiotic treatment, but also to the host immune response . Although clear-cut examples of adaptation to the host have come mainly from opportunistic infections of P. aeruginosa and B. dolosa, several studies of S. aureus infection show that even single point mutations can lead to widespread changes in gene expression that may radically alter phenotypes, including virulence in model systems ^84,85. These studies strongly suggest that within-host bacterial adaptation has the potential to influence disease progression. However, disentangling cause from effect is crucial for drawing firm conclusions from such studies.

Whereas the adaptive potential of bacteria in the host is striking, less clear is whether we should expect bacteria to evolve to become more or less harmful. Adaptive trade-off theory predicts that when the long-term survival of the pathogen depends on the well being of the host, the pathogen will tend to evolve reduced virulence ^121–123. A number of longitudinal studies of within-host evolution have reported the attenuation of virulence over time ^85,127. However, adaptive trade-off theory concerns selection acting on transmission at the population level, and not during colonization or infection of individual hosts. Where trade-offs exist between onward transmission and short-term survival, we should expect within-host evolution to favour immediate reward at the expense of long-term success.

The years ahead promise further insights as studies continue to investigate within-host evolution in an increasingly diverse array of major pathogens . Fully capitalizing on the potential of WGS will require the development of new analysis methods for detecting recent transmission and adaptation, untangling gene expression and elucidating phenotype-to-genotype relationships . For example, studies of transmission stand to gain in accuracy by sampling within-host diversity (Figure 2), but existing analysis methods do not currently exploit this information to its full potential. Our current knowledge of the size of transmission bottlenecks is very limited, and so this is an area where studies of diversity within donors and recipients could provide valuable new information . Predicting bacterial phenotypes from genotypes is likely to grow in importance, particularly in translational settings for genome-based antibiotic resistance prediction . Already, high levels of accuracy can be achieved ^133,134, and the quality of prediction is set to increase as we unravel the genetic architecture of bacterial phenotypes. As sequencing technologies continue to improve, real-time WGS-based diagnostics will provide clinicians with deeper insights into the pathogen population and its evolutionary potential to respond to different treatments, and therefore more information on which to base clinical decisions.

Key Points.

• Whole genome sequencing of multiple isolates from single hosts has revealed previously unsuspected within-host diversity of many bacterial pathogens.

• Within-host bacterial populations are subject to multifarious evolutionary forces including mutation, genetic drift, natural selection and fluctuating population sizes.

• Within-host evolution blurs transmission relationships based on sampling a single genome per host, conferring a benefit to sequencing multiple genomes per host.

• Resistance to some antimicrobials frequently evolves independently in multiple hosts, revealing the enormous potential of bacteria to adapt in the human body.

• Within-host adaptation plays a major role in the evolution of opportunistic infections in immunocompromised patients by otherwise free-living bacteria.

• The study of within-host genomic evolution promises to shed new light on whether pathogens tend to become more or less virulent within the host, and the underlying selective pressures.

Glossary

Adaptability

The ability to rapidly adapt to a change in selective pressure, such as antibiotic usage.

Adaptive trade-off hypothesis

Hypothesis that the long term evolutionary success of a pathogen requires a balance between duration of infection and virulence, based on the assumption that an increase in virulence decreases the average duration of infection.

Genome assembly

Bioinformatics process in which overlapping sequencing reads are combined into longer, contiguous sequences known as ‘contigs’, ideally a single contig per chromosome but usually several .

Clonal interference

Evolutionary dynamic in which selectively advantageous alleles at a given locus in one lineage outcompete advantageous alleles at other loci in other lineages , driving them to extinction. In organisms with the capacity for genome recombination , this can be avoided by combining all advantageous mutations in the same genome.

Compensatory mutations

Mutations that redress, possibly only partially, the fitness cost of mutations conferring adaptation to specific selection pressures, such as antibiotic resistance. Without compensatory mutations, adaptations that incur a fitness cost may be lost when the selection pressure is removed.

dN/dS ratio

The ratio of the number of non-synonymous substitutions, which alter the protein sequence, to the number of synonymous substitutions, which do not alter the protein sequence, normalized by the ratio expected under neutrality. dN/dS below one indicates purifying selection and above one positive selection.

Diversifying selection

A form of recurrent positive selection that favours the emergence of new alleles in a population; for example, the selective pressure of the host immune system on antigen evolution in pathogens .

Fitness trade-offs

The existence of some constraint, possibly mechanistic or genetic, that causes adaptations to one selection pressure to be disadvantageous with respect to another.

Fixation

The point at which an allele replaces all alternative alleles of the same locus in the population. Coincides with loss of the other alleles.

Heteroresistance

Varying levels of antibiotic resistance within an extremely closely related population, such as an individual colony.

Hitchhiking

Effect whereby an allele can increase in frequency even though it is not favoured by selection, only because it is found on the same genomes as other alleles of other loci that have a selective advantage.

Homologous recombination

Evolutionary event in which a segment of the genome of a recipient cell is replaced with the homologous segment of the genome from a donor cell.

Horizontal gene transfer

Uptake of genetic material by a recipient cell by a variety of mechanisms, commonly transformation of naked DNA, bacteriophage-mediated transduction or plasmid-mediated conjugation.

Hypermutators

An individual or lineage with elevated mutation rate, usually as a result of a loss of functionality of the DNA repair systems.

Incomplete lineage sorting

A phenomenon where a gene tree is discordant with the population or species tree. Occurs when lineages ancestral to several different populations split prior to, and in a different order to, the splitting of the respective populations. For within-host populations, causes discordance between phylogenies and transmission trees.

Melioidosis

Infectious disease caused by B. pseudomallei, leading to septisis and pneumonia, and endemic in South-East Asia and Australia.

Mismatch repair systems

Mechanism found in all bacteria to repair the mistakes introduced in the genome when it is replicated to allow clonal reproduction.

Evolutionary rate

The rate at which substitutions arise in a lineage, also known as the molecular clock rate. Population genetics theory predicts a constant rate in a neutrally evolving population with constant mutation rate, irrespective of changes in population size.

Mucoidy

A bacterial phenotype describing the production of glycoproteins resembling mucus.

Multi-locus enzyme electrophoresis (MLEE)

A molecular epidemiology approach in which strains are typed by the electrophoretic properties of a number of proteins.

Multi-locus sequence typing (MLST)

A molecular epidemiology approach in which strains are typed by their nucleotide sequences at a number of loci, typically 400 to 500bp long fragments of seven housekeeping genes.

Effective population size

Also known as N_e. The size of an idealized (neutrally evolving, homogeneous) population that is otherwise equivalent to an observed population. N_e is typically smaller than the number of individuals in the population, owing to population structure and variation in survival or reproductive viability.

Convergent evolution

Also known as parallel evolution. Occurrence of the same mutation at the same site in two or more independently evolving lineages, for example in separate hosts.

Pathoadaptive

An adaptation that confers pathogenicity.

Phase variation

A mechanism that bacteria use to allow rapid evolution of a specific trait in which frequently occurring, reversible mutations control gene expression.

Pleiotropic

The unexpected influence of one locus on multiple, apparently unrelated, phenotypes.

Point mutations

Mutation changing a single nucleotide.

Positive selection

Tendency for an allele that confers a survival or reproductive advantage to increase in frequency and fix at a higher rate. Advantageous alleles may nevertheless become lost, owing to random genetic drift despite positive selection.

Pre-adaptation

Also known as exaptation. Phenomenon where a previously existing trait confers an advantage in an environment to which it was not previously exposed.

Pulsed-field gel electrophoresis (PFGE)

A molecular epidemiology marker that allows strains to be typed by the lengths of the DNA molecules obtained after cutting the genome using a restriction enzyme .

Purifying selection

Tendency for an allele that incurs a survival or reproductive disadvantage to decrease in frequency and become lost. Deleterious alleles may nevertheless become fixed, owing to random genetic drift.

Quorum sensing

Mechanism by which a cell responds to changes in population size or density, classically by the secretion and detection of small peptides (also known as pheromones) .

Random genetic drift

Variations in genetic allele frequencies in a population caused by the random genetic sampling that occurs during the birth and death of individuals .

Selective sweep

Rapid increase in frequency and fixation of an advantageous allele. Selective sweeps are caused by positive selection.

Variable-number tandem repeats (VNTR)

A molecular epidemiology marker that allows strains to be typed by counting the number of copies of a specific pattern, which may be made of one or more nucleotides and is known to occur in a repeated fashion at a given location along the genome .

Variant calling

Bioinformatics process that determines the nucleotide at a given genomic site based on sequencing reads .

Virulence

The quantifiable frequency or severity of disease.

Biographies

Author information

Xavier Didelot

Xavier Didelot is a Lecturer in the Department of Infectious Disease Epidemiology at Imperial College London. He received his D.Phil. in statistical genetics from the University of Oxford, UK, in 2007. He was a research fellow at the University of Warwick, UK, for 3 years before moving back to Oxford to work on the Modernising Medical Microbiology (MMM) project. He is perhaps best known as the author of ClonalFrame, software that reconstructs the genealogy of a sample of bacteria while accounting for the confounding effect of recombination.

A. Sarah Walker

Sarah Walker is a Professor of Medical Statistics and Epidemiology at the University of Oxford and University College London. She received her PhD from University College London in 1999 and has spent the last 20 years designing, managing and analysing large randomised controlled trials and observational studies in infectious diseases. Over the last 10 years she has been a key member of the Modernising Medical Microbiology (MMM) consortia, integrating electronic health records and pathogen whole genome sequencing to investigate transmission epidemiology and improve patient management.

Tim E. Peto

Tim Peto is an infectious diseases physician and a professor of medicine at the Oxford University Hospitals, and a UK National Institute of Health Research (NIHR) senior investigator. He read for a D.Phil. and qualified in medicine at the University of Oxford. As a clinical epidemiologist and clinical trialist for over 20 years, he has been overseeing key large scale trials in HIV, malaria and tuberculosis. Currently he studies the use of whole-genome sequencing for tracking common pathogens and predicting drug resistance in both hospitals and the community.

Derrick W. Crook

Derrick Crook is a clinical microbiologist, infectious diseases physician, professor of microbiology at the Oxford University Hospitals, an NIHR senior investigator and Director of the National Infection Service at Public Health England. He qualified in medicine at the University of the Witwatersrand, Johannesburg, South Africa, studied at the London School of Tropical Medicine, UK, and trained in the Department of Medicine, University of Virginia, Charlottesville, Virginia, USA, and the Department of Infectious Diseases, New England Medical Centre, Boston, Massachusetts, USA. His research encompasses diagnostics, epidemiology, new sequencing and informatics technologies aimed at improving management of infectious diseases.

Daniel J. Wilson

Daniel Wilson is a Wellcome Trust and Royal Society Sir Henry Dale Fellow at the University of Oxford, where his laboratory investigates pathogen evolution and epidemiology via whole-genome analysis. Daniel’s work currently focuses on the identification of genetic variants in pathogen genomes that explain differences in the frequency and severity of infections, particularly hospital-associated infections. He read biological sciences and completed a doctorate in pathogen population genetics at the University of Oxford, before pursuing postdoctoral research at Lancaster University and the University of Chicago.