The sequencing of bacterial genomes (see Glossary) has occurred against the backdrop of an established programme of research on bacterial pathogenesis. Nonetheless, it has uncovered aspects of pathogen biology that were unexpected before the genomic revolution. Here, we examine the 'creative clash' between genomic research and bacterial pathogenesis research, an encounter that has spawned new technologies and new avenues for applied research. In addition, we discuss the forces that have shaped the evolution of bacterial pathogens, and we reappraise human–pathogen interactions in the light of bacterial ecology and evolution.
Genome dynamics
At the start of the genomic era, each project to sequence a bacterial genome was viewed as equivalent in difficulty to a Moon landing. However, the cutting edge soon shifted to determining the genome sequences of multiple strains in each species1, 2, 3, 4, heralding a transformation in our view of bacterial diversity. Comparisons between the genomes of related strains and species of bacterial pathogens, across the whole range of taxonomic variation, have made it clear that a 'one size fits all' approach cannot be applied to the evolutionary dynamics of bacterial virulence5, 6, 7. Instead, different evolutionary processes predominate in different taxonomic groups.
Three main forces have been found to shape genome evolution: gene gain, gene loss and gene change (that is, any changes that affect the sequences or order of the existing genes) (Fig. 1). In the genome of some bacterial pathogens (for example, Yersinia pestis8), all three are evident. In addition, differences in the scale and the timing of these changes in different lineages of bacterial pathogens have resulted in at least three main patterns of genome dynamics. First, some genetically uniform lineages, which are also usually reproductively isolated, have emerged recently in evolutionary terms (for example, Bacillus anthracis and Mycobacterium leprae). Second, recombination can occur between closely related sequences in closely related strains; this is common in naturally competent mucosal pathogens (for example, Neisseria meningitidis, Haemophilus influenzae and Streptococcus pneumoniae). Third, widespread horizontal gene transfer, bringing in new sequences, predominates in certain pathogens (for example, many enterobacteria, and some staphylococci and streptococci).
Figure 1: Bacterial genome dynamics.
There are three main forces that shape bacterial genomes: gene gain, gene loss and gene change. All three of these can take place in a single bacterium. Some of the changes that result from the interplay of these forces are shown.
High resolution image and legend (34K)The smallest-scale variation in bacterial genomes occurs at the level of single-nucleotide polymorphisms (SNPs). SNP detection has been applied extensively to recently emerged genetically uniform pathogens, such as M. leprae, Mycobacterium tuberculosis, Y. pestis and B. anthracis (the last driven by forensic considerations after the anthrax attacks of 2001 in the United States)9, 10, 11, 12. More recently, whole-genome sequencing has been used to detect SNPs in more variable species, such as Escherichia coli and Francisella tularensis13, 14, 15. This approach to SNP detection enabled E. coli strains that had diverged for as few as 200 generations to be differentiated16 and revealed genomic changes in Burkholderia mallei after accidental human infection17. These studies indicate that the use of whole-genome sequencing could soon become a routine epidemiological tool in bacteriology, as it already is in virology (Box 1).
Genome sequencing has also confirmed that phase variation is a widespread source of intraspecific genotypic and phenotypic variation18, 19. Several mutational mechanisms are exploited by bacteria to switch gene and/or protein expression on or off. For example, in Campylobacter jejuni, the presence of several tens of homopolymeric nucleotide repeat sequences can lead to slippage during DNA replication, resulting in a varied repertoire of structures exposed on the bacterial-cell surface20. By contrast, Bacteroides fragilis uses DNA inversion to modulate more than 20 genetic loci, which contain genes that encode bacterial surface proteins, polysaccharides and components of regulatory systems21. The combinatorial mathematics of phase variation mean that a bacterium with just 20 phase-variable loci can exist in 220 (that is, more than a million) different states.
Horizontal gene transfer
The greatest surprise resulting from the application of genomics to bacteriology was the extent of genomic variability within many bacterial species. Two E. coli strains can differ by as much as a quarter of their genomes: for example, the laboratory strain E. coli K-12 is missing 1.4 megabases of DNA present in E. coli O157 (ref. 3). For many important pathogens, the genes common to all strains within a species (known as the core genome) are a minority component of the entire gene pool for that species (the pan-genome). Furthermore, a distinction can be made between closed pan-genomes and open pan-genomes. For closed pan-genomes, completing the genome sequencing of additional bacterial strains is unlikely to yield new genes. By contrast, for open pan-genomes, each new genome sequence reveals new members of the gene pool for that species22.
The genomes of some bacterial pathogens have gained genes through gene duplication, resulting in increased numbers of key gene clusters or the expansion of important protein families: for example, in M. tuberculosis, the gene families encoding acidic glycine-rich proteins and the gene clusters encoding the secreted protein ESAT6 (early secretory antigenic target 6) and its homologues have undergone extensive rounds of gene duplication23. Nonetheless, gene gain as a result of horizontal gene transfer remains the most potent source of 'innovation' and variation. However, unlike viruses, bacteria seldom acquire 'eukaryotic-like' genes from their hosts (although there seem to be some exceptions, for example, Legionella pneumophila24). Instead, horizontal gene transfer generally occurs between different strains and species of bacteria.
Horizontal gene transfer is mediated by diverse mobile genetic elements, including plasmids, bacteriophages and pathogenicity islands (Table 1). These elements often carry genes that encode factors involved in infection (often termed virulence factors) (Box 2). For example, numerous virulence factors and systems are encoded on plasmids. These virulence-associated plasmids can be large (for example, the plant pathogen Ralstonia solanacearum carries such a plasmid of more than 2 megabases25). They can also be promiscuous: that is, they can move freely between cells with markedly different chromosomal backgrounds. In the extreme case of enterotoxigenic E. coli, the association of promiscuous plasmids with diverse chromosomal lineages is all that defines the pathotype of these bacteria26.
Bacteriophages (that is, bacterial viruses) can also mediate horizontal gene transfer. Some classic virulence factors, such as diphtheria toxin, are encoded in the genomes of bacteriophages that have integrated into the bacterial chromosome (which are known as prophages)27. Genomic analyses show that prophages have a widespread role in driving the diversification of bacterial pathogens as distinct as E. coli, Streptococcus pyogenes and Staphylococcus aureus28, 29, 30. Prophages, particularly those derived from tailed bacteriophages, often carry genes that are superfluous for bacteriophage replication, and these genes are present within distinct 'passenger compartments' at one end of the prophage genome. These compartments are sometimes called morons to reflect that the associated prophage genomes encode more DNA than is necessary for bacteriophage replication alone31. The genes in these compartments are often implicated in virulence and can show a bias in base composition that sets them apart from the rest of the prophage and from the genome of the bacterial host. For example, in E. coli O157, the passenger compartments of lambdoid prophages contain genes with a low G+C composition that encode effector proteins capable of translocation into host cells by a type III secretion mechanism29.
Pathogenicity islands are another class of mobile element involved in horizontal gene transfer. The term 'pathogenicity island' originated from the study of uropathogenic E. coli but has subsequently been widely applied to bacterial pathogens32. Pathogenicity islands are usually defined by five characteristics. First, they are clusters of contiguous genes that are present in some related strains or species but not in others. Second, they are presumed to have been acquired by horizontal gene transfer. Third, they are generally associated with transfer RNA gene loci. Fourth, they typically have a G+C content that differs from that of the host bacterial genome. Fifth, they confer on the host bacterium a complex and distinctive virulence phenotype in a single step. Although some pathogenicity islands carry genes encoding integrases (enzymes that integrate the pathogenicity island into the host DNA), the mechanisms underlying the transfer of pathogenicity islands from one genome to another are unclear in many cases, as is the identity of the donor microorganisms.
Despite their mobility, pathogenicity islands are remarkably well integrated into the global regulatory network of bacterial cells. For example, numerous external factors affect the expression of genes on the locus of enterocyte effacement (LEE) pathogenicity island of E. coli, making it part of the 'genomic continent'33, 34, 35. It is also important to recognize that some 'pathogenicity islands' are deletions in one lineage rather than insertions in another36. Therefore, instead of considering the evolution of pathogens as a series of acquisitions of pathogenicity islands, a more sophisticated outlook is that genomes are 'molecular palimpsests': that is, the variable compartment of the genome bears the scars of repeated rounds of gene acquisition and erosion.
Gene loss
Bacterial genomes remain about the same size despite the pervasive effects of horizontal gene transfer, so gene gain must be balanced by gene loss37. Indeed, it is expected that any gene that is not maintained by natural selection is lost: bacterial genomes are subject to the 'use it or lose it' maxim. Genome sequencing has now provided a series of 'snapshots' that show directly the dynamic processes of gene loss and genome decay (that is, the progressive purging from the genome of unnecessary genes). For example, in many E. coli lineages, the Flag-2 and ETT2 gene clusters — which, when intact, span tens of kilobases — have been reduced to small scars occupying only a few hundred base pairs, presumably because they no longer provide any selective advantage to the organism36, 38.
The most surprising snapshots of genome decay have come from recently emerged pathogens that have changed lifestyle, usually to live in a simpler host-associated niche. For example, the genomes of M. leprae39, Y. pestis40 and Salmonella enterica serovar Typhi41 contain hundreds or even thousands of pseudogenes; in the M. leprae genome, there are nearly as many pseudogenes as functional genes39. These examples contradict the view that every gene in a bacterial genome must have a function and that bacterial genomes never contain 'junk' DNA. Instead, every genome should be viewed as a work in progress, burdened with some non-functional 'baggage of history'.
Another common feature of recently emerged genetically uniform pathogens is the 'proliferation' of transposable elements, particularly insertion sequences, in the genome42. This abundance of insertion sequences facilitates homologous recombination within the genome, a process that can result in large-scale chromosomal rearrangements that disrupt the ancestral gene order. In the case of Y. pestis, recombination between insertion sequences results in marked anomalies in GC skew (usually a marker of the direction of replication for a given region of chromosome) and in reversible chromosomal rearrangements during in vitro growth of the organism40. It is unclear whether such large-scale genomic rearrangements have functional relevance.
The most extreme form of genome decay is seen in host-associated bacteria, particularly endosymbionts that have been isolated for long periods in a static and 'undemanding' intracellular niche43. Pioneering studies by Siv Andersson and colleagues established that certain intracellular bacteria, such as Rickettsia prowazekii, have undergone considerable genomic downsizing, shedding many (or even most) of their ancestral genes44. Buchnera aphidicola, an aphid endosymbiont, is a pertinent example in that it is a close relative of E. coli but has fewer than one-tenth of the genes present in the latter45. Extreme genome decay is often accompanied by a shift towards a low G+C content: the largest known shift is in the 160-kilobase genome of the psyllid (jumping plant lice) symbiont Carsonella ruddii, which has a G+C content of only 16.5% (ref. 46). But perhaps the most extreme example of bacterial genome decay is that of human mitochondria, which belong to the 
-proteobacterial lineage and retain a tiny, 17-kilobase, genome (arguably the first bacterial genome to be sequenced47).
Less common than genome decay, but more marked in its consequences, is positive selection for gene loss. This occurs as a newly emerged pathogen adapts to its niche and forms part of a process known as pathoadaptation. Pathoadaptation can involve any changes that refine newly formed virulence mechanisms. One example is glucosylation of the surface molecule lipopolysaccharide, which optimizes the exposure of the type III secretion apparatus of Shigella flexneri48. Pathoadaptation also encompasses gene loss, although it might seem counter-intuitive that losing genes can specifically improve the fitness of a bacterium in vivo and make it more pathogenic. The best-known example occurs among the shigellae: loss of the gene cadA (which encodes the enzyme lysine decarboxylase) provides a selective advantage in the intracellular niche, because the product of lysine-decarboxylase activity, cadaverine, inhibits the plasmid-encoded virulence factors of these bacteria49. Genome sequencing has shown that the genetic mechanisms underlying loss of cadA vary between Shigella lineages, thus providing an example of convergent evolution in bacterial genomes.
Intriguingly, several recently emerged pathogens (including Bordetella pertussis, B. mallei, Y. pestis, all Shigella lineages and some E. coli O157 lineages) have independently lost flagellar motility during the transition to a new virulence-associated lifestyle. This suggests that these bacteria are subject to a common pathoadaptive selective pressure, but it is unclear whether the driving force is loss of a target (the protein flagellin) recognized by both the innate immune response and the adaptive immune response in mammals or changes in bacterial metabolism that occur concurrently50.
An 'eco–evo' perspective on host–pathogen interactions
A glance at the post-genomic landscape shows that our previous knowledge of the ecology and evolution of bacterial pathogenesis was limited. New findings mean that previous assumptions need to be questioned and terms need to be redefined. Among genetically variable bacterial species, it is now clear that a single strain rarely typifies an entire species, particularly because genomics has provided compelling evidence that commonly used laboratory strains (for example, E. coli K-12, S. enterica serovar Typhimurium LT2, Pseudomonas aeruginosa PAO1 and S. aureus COL) have undergone marked genotypic and phenotypic changes during their descent from the ancestral free-living isolate51, 52.
Similarly, the readily available bacterial genome-sequence data have challenged the simplistic views that a bacterial pathogen can be understood solely by identifying its virulence factors and that pathogens always evolve from non-pathogens by acquiring virulence genes on plasmids, bacteriophages or pathogenicity islands. Instead, genomics has helped to blur the distinction between pathogens and non-pathogens and between virulence factors and colonization factors. And it has catalysed a copernican shift in how host–pathogen interactions are viewed, a shift away from an anthropocentric focus towards a broader perspective that places interactions between eukaryophilic bacteria and eukaryotes in a wider ecological and evolutionary context (Fig. 2). Inherent in this 'eco–evo' perspective is the need to identify the selective advantages of virulence factors in the broader lifestyle of the pathogen. In addition, 'evolutionary narratives' that interweave genomic changes with ecological shifts can now be constructed. For example, genomic comparisons allow a reconstruction of how the plague bacillus, Y. pestis (a rodent and flea pathogen that is occasionally transmitted to humans), evolved from a gastrointestinal pathogen (Yersinia pseudotuberculosis) in an evolutionary blink of an eye (about 10,000 years), through the processes of gene gain, loss and rearrangement8, 53, 54.
Figure 2: The eco–evo view of bacterial pathogenomics.
a, Pathogenic bacteria and commensal bacteria often share their habitats with bacteriophages, other bacteria, amoebae, insects, nematodes, annelids (such as leeches), fungi, plants and mammals (such as humans). This mixed ecology is a considerable driving force in the evolution of these microorganisms. In this context, it is not surprising that genes encoding 'virulence factors' are found in both human pathogens and non-pathogens. b, In addition, consideration of the evolutionary history of a pathogen might be needed to explain some of the features of its genome. Within bacterial genomes, it is common to find remnants of genes or gene clusters that presumably provided an adaptive advantage in the past but are now non-functional (indicated in blue). Also, it should be considered that a microorganism that is pathogenic now might at one time have been a commensal microorganism, and vice versa (indicated by the phylogenetic tree).
High resolution image and legend (85K)A more fundamental consequence of the eco–evo view is that it is now expected that what, at first, seem to be virulence factors are encoded in the genomes of 'non-pathogens' (Table 2). There are several reasons for this. First, it is now clear, both from genomic and pathogenesis studies, that pathogens, commensal microorganisms and symbionts rely on similar strategies and molecular systems in their interactions with eukaryotic hosts (for example, phase variation)21, 55. Second, it is also understood that bacteria sometimes produce virulence factors that provided an advantage only in a previous, now non-existent, niche. Last, it has also become evident that many bacterial pathogens infect humans only incidentally and often produce virulence factors that are active against non-mammalian adversaries as diverse as plants, insects, protozoans, nematodes, predatory bacteria and bacteriophages. Inherent in this view is the realization that many bacterial virulence factors have been shaped by evolutionary forces outside the context of human–pathogen interactions, and only by studying these forces can the emergence of human infections be understood.
Enterohaemorrhagic E. coli strains, particularly E. coli O157, provide a compelling test case for the eco–evo view. E. coli O157 is a rare but devastating pathogen of humans, but it is also a common commensal microorganism of the bovine gut. Genomic comparisons have helped to explain how this pathogen has evolved from a non-pathogenic ancestor by acquiring virulence factors encoded on various mobile elements (for example, Shiga toxin, which is encoded on a bacteriophage, and the type III secretion system encoded on the LEE pathogenicity island)56. However, recent studies have shown that a pilus-adherence factor that is crucial to the virulence of E. coli O157 in humans is also carried by commensal strains of E. coli57. Also, similarly to many commensal strains of E. coli, E. coli O157 carries remnants of a gene cluster (ETT2) encoding a virulence-associated secretion system that is now thought to be inactive36. It is only through an eco–evo view that the evolution and transmission of E. coli O157, and why it produces such lethal virulence factors, might be understood. One potential explanation is that the 'virulence factors' of E. coli O157, such as Shiga toxin, help it to colonize the bovine gut. However, the evidence for this hypothesis is equivocal at best58. Instead, a recent study shows that the Shiga-toxin-encoding bacteriophage increases bacterial survival in the presence of a grazing ciliate, Tetrahymena pyriformis, indicating that interactions with non-mammalian adversaries might have driven the evolution of this virulence factor59. Similar points can be made about many other pathogens: for example, it has long been known that the pathogenesis of legionellosis in humans relies on mechanisms that legionellae use to subvert amoebae60. Clearly, in the post-genomic era, even for important human pathogens, humans can no longer be considered to be the centre of the bacterial universe.
Future challenges
The clearest challenge for future studies of bacterial pathogenomics is coping with the flood of new data unleashed by the arrival of affordable and quick, 'next-generation', sequencing technologies61. Now that the cost of sequencing bacterial genomes fits comfortably within the budget of a standard research project grant, it is set to become an integral and routine part of research programmes. Therefore, within the next decade, tens of thousands of bacterial genomes will be sequenced62. And the focus will shift from the mechanics of generating sequence data to the problems of analysing it, creating an urgent need for better ways to compare and visualize genomic data. Also, there are likely to be many more incompletely sequenced genomes, as the efficiency of the finishing stage of a genome project lags behind the rapid pace of next-generation whole-genome shotgun sequencing.
For genetically uniform species, particularly those with potential as bioterrorism agents, the resequencing of hundreds of isolates (for example, by using tiling arrays) will drive forward forensic genomics. For more variable species, such as E. coli, S. enterica and S. aureus, the focus will be on defining the extent of the pan-genome and on developing improved approaches to understanding epidemiology, particularly for those that cause hospital-acquired infections. The most important challenge will still be to add functional relevance to genome sequences, a challenge that will continue to drive the application of high-throughput 'omics' approaches to the study of virulence.
Furthermore, sequencing the genomes of environmental organisms and carrying out metagenomic surveys of diverse environments will provide not only an improved understanding of microbial biodiversity but also insight into the evolution of bacterial factors that are involved in human disease63, 64. Metagenomic surveys of eukaryote-associated bacterial communities will strengthen our understanding of the ecology of bacterial infections (for example, the micro-ecological changes that accompany antibiotic-associated diarrhoea) and help to shed light on the pathogenesis of polymicrobial infections, such as those that cause periodontal disease and bacterial vaginosis65. Similarly, studying the metagenomics of bacteriophage populations will help to unravel the connections between these mobile elements and the evolution of virulence.
In addition to the genomic technologies discussed here, evolutionary game theory will need to be applied so that the complex interactions between bacteriophages, the virulence factors that they encode, the bacteria that they infect and the eukaryotic targets of their virulence factors can be understood. Similar approaches will also be needed to solve the conundrum of invasive disease: for example, to explain why meningococci cause meningitis, despite the fact that the disease has no role in the transmission of the bacteria66. Evolutionary systems-biology approaches will also be useful for understanding the evolution and regulation of complex virulence systems, the interactions between pathogens and their host, and the co-evolution of their genomes.
The first decade of bacterial genomics has afforded unprecedented insights into the evolution of virulence. The next decade holds the promise of being even more rewarding as the new eco–evo view of host–pathogen interactions draws on ever more genome and metagenome sequences.