Chapter 147

Evolutionary Genetics of Salmonella enterica

ROBERT K. SELANDER, JIA LI, and KIMBERLYN NELSON

INTRODUCTION

The discovery of the salmonellae (family Enterobacteriaceae) and their recognition as agents of disease date to the first observation of the typhoid bacillus by Eberth in 1880 (39). For the greater part of a century, the primary basis for classification and identification of these bacteria has been a serological scheme initiated by White (199) and elaborated by Kauffmann (78) and others (43, 90, 91). Currently, the salmonellae are assigned to 2,324 serovars (139), of which the agent of human typhoid fever (serovar Typhi) is but one. Owing to Kauffmann’s (78) idiosyncratic view that serovars are species, many of them were originally designated by italicized Latin binomials (e.g., Salmonella typhimurium and S. panama); but in accordance with compelling molecular genetic evidence of relatively close relationships among all of the serovars, it is now widely agreed that all of the salmonellae are members of a single species, which has been formally designated as Salmonella enterica (92). In current practice, the original species names are retained for the serovars of subspecies I but are used as vernaculars, and the serovars of the other subspecies are indicated by antigenic formula alone. Thus, S. typhimurium is referred to as S. enterica serovar Typhimurium or, more simply, as Typhimurium.

Although S. enterica is a highly diverse species presenting excellent opportunities for comparative studies that can yield insights regarding the origin, function, and evolutionary elaboration of both structures and processes, most of what is known of the cellular and molecular biology of the salmonellae has been derived from studies of Typhimurium laboratory strain LT2 and its derivatives. This chapter reviews the recent findings of research on the genetic diversity, population structure, and evolutionary relationships of the salmonellae as a whole, with emphasis on information obtained by the comparative nucleotide sequencing of structural genes (119).

SEROTYPING

The primary antigens assayed in the Kauffmann-White scheme are those of the polysaccharide domain of the cell surface lipopolysaccharide (O antigens) and the flagellin proteins that form the filaments of phase 1 and phase 2 flagella (H1 and H2 antigens, respectively). O-antigen epitopes are determined by the type, arrangement, and condition of sugar residues in repeated oligosaccharide subunits, and more than 50 major types have been recognized (139, 144). Variation in the O antigen is normally mediated by chromosomal genes of the rfb cluster, but at least one type, O54, is determined by plasmid-borne rfb genes (79, 138). Some O serotypes are subject to lysogenic conversion, in which existing antigenic factors are modified or additional factors are expressed as a result of infection with certain phages (43, 105, 123). With respect to flagella, most serovars are diphasic, alternately expressing two genes (fliC and fljB) that encode flagellins of different antigenic character, but many are monophasic, and two of them (Gallinarum and Pullorum) lack flagella. For the phase 1 flagellin (fliC gene), 52 antigenic factors and 61 serotypes (single factors or combinations of factors) have been distinguished (139).

Each distinctive combination of O, H1, and H2 antigens is given formal recognition as a serovar, with the provision that strains of the same serotype combination (antigen formula) belonging to different subspecies are treated as separate serovars. Minor variant antigenic factors are often ignored, and two or three different serovar names are sometimes applied to strains with the same antigenic profile because of differences in host range or disease syndrome or because of biochemical differences or minor antigenic variation not detected in routine serotyping.

PATHOGENICITY AND EPIDEMIOLOGY

Disease Syndromes

S. enterica is a close relative of Escherichia coli (4, 20, 21, 83, 145), but unlike that species, most strains of which are commensals of mammals and birds, the salmonellae are intracellular pathogens and infect reptiles as well as warm-blooded vertebrates (147). Four disease syndromes are produced in humans and other mammals (67, 189). Acute enterocolitis occurs when bacteria invade the epithelium of the small intestine and multiply within mucosal cells, stimulating massive fluid secretion. The resulting diarrhea may progress to bacteremia, which may in turn lead to osteomyelitis or other focal infections. Finally, some strains cause the serious invasive diseases known as enteric fevers. In human typhoid fever, bacteria penetrate the intestinal mucosa and are taken up by mononuclear phagocytes in Peyer’s patches in the ileum, within which they multiply and, subsequently, reach the mesenteric lymph nodes, liver, and spleen. Then, escaping into the blood stream, they cause a sustained secondary bacteremia.

Frequency of Disease

With improvements in sanitation and the treatment of chronic carriers, typhoid fever has steadily declined in frequency in the United States since the early 1900s to a current level of fewer than 500 cases per year (27), most of which can be traced to international travel (148); but the disease is endemic and epidemic in many Third World countries (40, 68), where an increasing incidence of multiantibiotic-resistant strains has made treatment and control difficult (51, 57).

Since emerging as a public health problem in the United States after World War II, nontyphoidal salmonellosis in humans has steadily increased in frequency to a current estimated level of three million cases per year (27, 136). Domesticated animals are the primary reservoir of nontyphoidal salmonellae infecting humans, and contaminated poultry and poultry products are responsible for a large proportion of all outbreaks and epidemics. Since the mid-1980s, egg-borne infections caused by strains of serovar Enteritidis have become a serious health problem in the United States, Europe, and elsewhere (89, 109, 194). Among the factors responsible for the higher incidence of food-borne disease are increased consumption of poultry and fresh produce, growth in the number of commercial food service establishments, and changing methods of food production and distribution (63, 136).

RESEARCH REFERENCE COLLECTIONS

Three collections of strains are available from the Salmonella Genetic Stock Center (contact K. E. Sanderson, Department of Biological Sciences, University of Calgary, 200 University Drive NW, Calgary, Alberta, Canada T2N 1N4) for use by microbiologists and molecular biologists concerned with genetic or phenotypic variation in natural populations of S. enterica. Salmonella Reference Collection A (SARA) (13) consists of 72 strains of serovars Heidelberg, Muenchen, Paratyphi B, Saintpaul, and Typhimurium recovered from a variety of hosts and environmental sources; collection B (SARB) (19) is composed of 72 strains of 37 subspecies I serovars; and collection C (SARC) consists of the two strains of each of the eight subspecies that have formed the core sample used in nucleotide sequence studies of several housekeeping genes (18, 118, 120, 121). For each collection, genomic evolutionary relationships among strains have been estimated by multilocus enzyme electrophoresis (MLEE).

SUBSPECIFIC RELATIONSHIPS

For the numerous serovars of S. enterica, seven subspecies (designated I, II, IIIa, IIIb, IV, V, and VI) have been formally recognized on the basis of variation in biochemical characters (biotype) and the results of a limited number of genomic DNA hybridization experiments (32, 93, 94, 181). Sixty percent of the serotypes belong to subspecies I, including those of strains that are responsible for >99% of cases of human salmonellosis.

Evidence from Enzyme Polymorphism

In application of the MLEE method of assessing allelic variation in multiple genes in large samples of isolates (157, 158, 160, 162), electromorphs (allozymes) of an enzyme are equated with alleles of the corresponding structural gene, and distinctive allele profiles (multilocus enzyme genotypes) are designated as electrophoretic types (ETs). For the salmonellae, as well as other bacteria, MLEE has been a useful method for assessing the overall genetic relationships of strains within subspecies and serovars (12, 13, 19, 97, 155, 156, 163).

Estimates of the genetic relationships among 80 ETs representing the subspecies of S. enterica, based on MLEE (24 enzyme loci), are indicated in the tree shown in Fig. 1. All except four of the ETs fall into seven clusters that correspond precisely to the seven subspecies defined by biotyping and DNA hybridization; but ETs 56 to 59, which are represented by isolates that had been assigned to subspecies IV on the basis of biotype, form a distinctive eighth cluster, designated as subspecies VII (154, 160). Given the considerable level of genetic differentiation between subspecies IV and VII indicated by both MLEE and sequence analysis of several housekeeping genes (see below), it is surprising that strains of these groups are closely similar in biochemical traits. However, a comparable situation occurs in Legionella species, in which biotype characters are strongly conserved among species that differ markedly in genomic character (22).

Fig. 1Neighbor-joining tree (81, 149) for 80 ET's of the eight subspecies of S.enterica, constructed from a matric of Nei's (115) standard genetic distance between pairs of ET's, based on the proportion of 24 enyzme loci at which dissimilar alleles occur (mis-matches). Subspecies are designated by roman numerals.

Source:

In agreement with evidence from gene sequencing and genomic DNA hybridization, MLEE analysis indicates that strains of subspecies V are the most divergent forms of the salmonellae; and on that basis, subspecies V was recently elevated to species status, as S. bongori, by Reeves et al. (142). However, statistical analysis of 1,000 computer-generated trees ("bootstrapping") indicates that the evolutionary relationships among the other subspecies are not reliably indexed by the topology of the MLEE tree (Fig. 1), the branches of which diverge from one another within a very narrow range of genetic distance values. At the relatively deep level of divergence of the S. enterica subspecies, most of the loci assayed are polymorphic for multiple electromorphs, with the consequence that the detailed branching order of the subspecies depends to a large extent on variation at just a few loci. The accuracy of estimates of genetic distance is further limited by the occurrence of multiple amino acid substitutions in individual proteins and the frequent convergence of electromorphs resulting from different amino acid substitutions that produce similar effects on electrophoretic mobility.

The total genetic diversity of S. enterica is considerably greater than that of E. coli. For the 80 ETs shown in Fig. 1, the mean total genetic diversity per locus (H_T), calculated from allele frequencies at the 24 enzyme loci assayed (115), was 0.627, which may be compared with a value of 0.343 reported (158) for 62 ETs of the E. coli reference collection (64, 125). The mean diversity per locus within subspecies (H_S) was 0.271, and the amount of diversity among subspecies (G_ST), calculated as (H_T – H_S)/H_T, was 0.593, or slightly more than twice that within subspecies.

Evidence from Nucleotide Sequence Variation

In the past few years, the technology of rapid nucleotide sequenceing of DNA amplified by PCR has opened a new era for bacterial population genetics by providing full resolutino of the allelic diversity of genes among multiple isolates and by permitting the recognition of DNA segments that have been acquired by strains through horizontal transfer and recombination

Estimates of the genetic relationships among strains of the eight subspecies, as indexed by variation in the combined nucleotide sequences of five housekeeping genes, proline permease (putP) (118), glyceraldehyde-3-phosphate dehydrogenase (gapA) (121), malate dehydrogenase (mdh) (18), 6-phosphogluconate dehydrogenase (gnd) (120), and isocitrate dehydrogenase kinase/phosphatase (aceK), are shown in Fig. 2A. The topology of the tree is fully consistent with evidence from DNA hybridization experiments (93) and may, therefore, be considered broadly indicative of the actual evolutionary (phylogenetic) relationships of the subspecies, notwithstanding the occurrence of a low level of horizontal transfer of gene segments among them (see below). The fact that subspecies I, II, VI, and IIIb—the serovars of which are exclusively or predominantly diphasic in flagellar expression—cluster apart from the monophasic subspecies IIIa, IV, VII, and V provides a basis for the following hypothetical evolutionary scenario (J. Li, H. Ochman, E. A. Groisman, K. Nelson, and R. K. Selander, Proc. Natl. Acad. Sci. USA, in press).

Fig. 2(A) Neighbor-joining tree for 16 strains of the eight subspecies of S.Enterica, based on variation in the combined coding sequences of five housekeeping genes, putP,gapA, mdh, gnd, and aceK. The tree was constructed from a matrix of pairwise genetic distance estimated from the number of substitutions per site (73, 116). Subspecies are designated by roman numerals, and the number of serovars assigned to each subspecies (139) is indicated in parentheses. Bootstrap values based on 1,000 computer-generated trees are indicated at the nodes. (B) Neighbor-joining trees for four invastion genes, based on variation in the combined coding sequences.

Source: (Adapted from references 96a).

Following the divergence of S. enterica and E. coli from a common ancestor 120 to 160 million years ago, coincident with the origin of the mammals (127), E. coli evolved as a commensal and opportunistic pathogen of mammals and birds. Perhaps 80 million years ago, E. coli produced the lineages of the four nominal species of Shigella (190), which, notwithstanding their taxonomic classification, are actually clonal lineages of E. coli (20, 75, 126, 180, 193). Meanwhile, the lineage of the salmonellae remained associated with reptiles (which are still the primary hosts of the monophasic subspecies) and evolved as intracellular pathogens through acquisition of the invasion (inv/spa) genes (49, 53) and other genes that mediate invasion of host epithelial cells and otherwise distinguish S. enterica from E. coli (83, 145). Subsequently, by providing an increased ability to circumvent host immune systems, the invention of the mechanism of flagellar antigen phase shifting (diphasic condition) (167) in the S. enterica lineage ancestral to subspecies I, II, IIIb, and VI may have been a critical factor permitting an expansion of ecological range to mammals and birds, but as a pathogen rather than a commensal—a niche already long occupied by E. coli. Subspecies I became highly specialized for mammals and birds, with some serovars adapting to single host species (e.g., Typhi for humans). Secondarily (and inexplicably, given the adaptive advantage that phase shifting would seem to provide), 10% of the serovars of subspecies I and II have reverted to the monophasic condition (139), usually by loss of expression of phase 2 flagella; and at least in the lineage of Enteritidis, the phase 2 flagellin gene (fljB) has been deleted rather than merely silenced.

EVOLUTION OF HOUSEKEEPING GENES

In the absence of horizontal exchange of genes or gene segments among strains, bifurcating evolutionary trees based on variation in nucleotide sequences may be interpreted as estimates of phylogenetic relationships; but even when recombination has occurred and strict phylogenetic interpretation is, therefore, precluded, tree construction provides a useful method for presenting information on degrees of genetic distance among strains (12, 37, 121, 160). Discordant features of the topologies of trees for different loci may provide information on the frequency and extent of genetic exchange, which is a major focus of current research in bacterial population genetics because of the important effects of recombination on the genetic structure and mode of evolution of natural populations (108, 200).

Levels of Sequence Diversity

For a sample of 16 strains representing the subspecies of S. enterica, data on sequence variation in five genes encoding housekeeping proteins are presented in Table 1, together with comparable data for four invasion genes. In the table, d _S is the estimated mean number of synonymous substitutions per 100 synonymous sites that have occurred between the sequences of pairs of strains, and d _N is the comparable estimate for nonsynonymous (replacement) sites (116).

Table 1Sequence variation in nine genes among 16 strains of S. entericaa

For the five housekeeping genes and their products, on average about 16% of nucleotides and 5% of amino acids are polymorphic, and the ratio of synonymous to nonsynonymous nucleotide substitutions is about 29:1. For each of the loci, variation within a subspecies is only a small fraction of that among the subspecies, reflecting a great depth of genetic structure of populations at this taxonomic level.

With the exception of gnd (see below), the level of sequence diversity in housekeeping genes is greater in S. enterica than in E. coli, with much of the difference being attributable to the unusually strong divergence of subspecies V from the other subspecies. For the 5 genes, the mean difference in nucleotide sequence between the two species is about 15%, which is close to the average reported for 67 genes sequenced in single laboratory strains (mostly LT2 and K-12) (164).

Evidence of Recombination

Comparisons of individual trees based on the nucleotide sequences of putP, gapA, mdh, and gnd (Fig. 3) have revealed several cases in which the order of branching of lineages is discordant. Some of these topological differences are attributable to intragenic recombination events, whereas others reflect the exchange of entire genes (assortative recombination). The DNA segments involved in intragenic recombination, which have been identified with the guidance of statistical tests for nonrandom clustering of polymorphic sites (151, 179), varied in length from 6 to 1,073 bp, but several of them are in the 200- to 400-bp range, which is the size reported for recombined segments of the phoA alkaline phosphatase gene in E. coli (36).

Fig. 3Neighbor-joining trees for four genes encoding housekeeping proteins, based on variation at synonymous nucleotide sites. Subspecies are indicated by roman numerals, and bootstrap values based on 1,000 computer-generated are indicated at the nodes.

Source: (Adapted from Boyd et al. [18], Nelson et al. [121], and Nelson and Selander [118, 120]).

putP, gapA , and mdh.

In the putP and mdh trees (Fig. 3A and C), subspecies V clusters with the other seven subspecies, but in the gapA tree (Fig. 3B), it forms a branch apart from both the other salmonellae and E. coli as a consequence of the presence of a segment of the gene that is almost identical in sequence to that of Klebsiella pneumoniae (118). Except for this feature, the topologies of the putP and gapA trees are generally similar, with subspecies I, II, IIIb, and VI in the same relationships; but the positions of the branch leading to IV and VII and that of IIIa are reversed in the two trees. This difference in branching order is attributable to the occurrence of a group of 25 unique polymorphic sites that defines a large segment in the central part of the putP sequence in strains of subspecies VII.

The only distinctive feature of the topology of the mdh tree (Fig. 3C) is that subspecies I is more similar to VI than to II, whereas I and II cluster together in the trees for putP and gapA. This difference reflects a recombination event, which was detected by the nonrandom clustering of polymorphic sites in the 5' region of the mdh gene.

That gene segments of substantial length can be clearly identified as recombinant elements in contemporary lineages of S. enterica indicates that episodes of intragenic recombination involving these three loci are rare. The simplest explanation for the observation that strains of a given subspecies share the same recombined segments is that these events antedate the time of divergence of the contemporary cell lineages within the subspecies.

gnd .

Because 6-phosphogluconate dehydrogenase functions in an essential metabolic pathway and the gnd locus exhibits a moderate degree of codon bias, the expectation from evolutionary theory is that both nucleotide and amino acid sequences should be relatively conserved. However, in S. enterica, the number of allozyme alleles is larger than expected from estimates of total genic diversity (120).

For most of 36 strains studied, evolutionary relationships deduced from gnd sequences (Fig. 3D) are similar to those indicated by enzyme-encoding and other housekeeping genes (120). Thus, subspecies V is the most divergent, followed by IIIa. However, assortative recombination has occurred between strains of several subspecies (notably I and VI), and intragenic recombination has also produced positional changes for some strains. For example, the gnd sequences indicate a close relationship between Paratyphi A (s4993) and Typhi (s3333), which is inconsistent with evidence from MLEE (19) and the sequences of other genes.

In E. coli, recombination at the gnd locus—both intragenic and assortative (8, 16, 37)—has been much more extensive than in S. enterica (120), with the result that allozyme diversity in 6-phosphogluconate dehydrogenase is three time greater than that expected on the basis of its molecular size (201). Some strains have acquired alleles from species of Citrobacter and Klebsiella, and recombination has occurred so frequently that a tree based on gnd sequences bears little resemblance to other gene trees or to an MLEE tree (120).

Linkage Interactions between gnd and rfb

Because it is most unlikely that the unusually high effective rate of recombination in gnd can be attributed to the direct action of diversifying selection on the locus itself, other explanations have been sought. The most plausible one is its close linkage to loci that determine the structure of cell surface macromolecules, including genes of the rfb cluster that mediate biosynthesis of the highly antigenic polysaccharide domain (O antigen) of the cell surface lipopolysaccharide (16, 37, 112, 120, 201). Genes of the rfb region are believed to be subject to strong frequency-dependent selection (143, 144), and there is evidence that recombination occurs frequently in the rfb region and that some or all of these genes in S. enterica and E. coli have been recruited from other, distantly related bacteria (144, 180, 205, 206).

The segment of the E. coli and S. enterica chromosomes in which gnd is located has the following gene order: the cps family (determinants of capsular polysaccharide structure), galF, the rfb cluster, gnd, rol, and the his operon (10, 11). The distance from the end of the rfb region to gnd is short, although variable, and there are several open reading frames between gnd and the his operon, including the rol (or cld) locus that encodes a product regulating modal chain length of the O antigen, which is strain specific and may be important in interactions with host cell membranes and immune systems (144). The inference is that the frequency of recombination at gnd is increased by the action of selection for allelic diversity at neighboring loci, as gnd sequences occasionally hitchhike with adaptive recombinants of rfb and, perhaps, other genes.

The anomalous position of subspecies IV strains s3010, s3009, and s3031 (Fig. 3D), which are all but identical in gnd sequence and also share O antigen 50, may be explained by the cotransfer of gnd and rfb from a strain of IIIb such as s2978, which also expresses O50. The presence of nearly identical gnd and rfb sequences in Typhi and Paratyphi A (120, 144) clearly indicates cotransfer. A possible mechanism for the dissemination of O factors among strains was recently identified by the discovery that isolates of serovar Borreze carry a plasmid that bears rfb genes encoding O54 (79). For strains of E. coli, several examples of the apparent horizontal cotransfer of rfb and gnd alleles have been identified (158, 201).

Thampapapillai et al. (185) have recently reported the results of an extensive study of sequence diversity in gnd among strains of S. enterica in which multiple recombination events, some involving cotransfer of parts or all of gnd and the rfb region, were identified. Several of these events appear to have been mediated by chi-like sequences (197) located near recombination junctions.

EVOLUTION OF INVASION GENES

The invasion of host cells by S. enterica is mediated by the products of a large number of genes that map to several chromosomal locations. Homologs of some of these genes occur in E. coli (54); but there is a 40-kb segment near 59 min on the S. enterica chromosome that is not present in E. coli K-12 and contains 15 or more loci, the inv/spa genes, whose products are required for the invasion of epithelial cells (for a review, see reference 108a). Homologs of these genes, which apparently are involved in the secretion of antigens that promote cell entry, have been identified in a variety of animal and plant pathogens (14, 41, 50, 53).

Sequences of four invasion genes—invA, spaO, spaP, and spaQ—have recently been obtained for multiple strains of S. enterica (96a). Levels of sequence diversity are shown in Table 1. The range of variation in d _S and d _N among the invasion genes exceeds that shown by housekeeping genes. The SpaO protein is hypervariable, with 21% of its amino acid positions polymorphic, which is consistent with evidence that it is an exported antigen. However, unlike the flagellin fliC gene (see below), the sequence diversity in the spaO gene appears to have been generated almost entirely by point mutation, rather than by intragenic recombination. In contrast, the SpaQ protein is unusually well conserved, with only 2.3% of its amino acids polymorphic. Levels of variation in invA and spaP are relatively "normal," although both d _S and d _N are slightly inflated in invA by the presence of a large recombinant segment, imported from an unidentified source, in the sequences of subspecies IV and VII.

The topology of a tree constructed from the combined sequences of the four inv/spa genes (Fig. 2B) is generally similar to a comparable tree for five housekeeping genes (Fig. 2A), with the exception of the position of subspecies II relative to subspecies IIIb and VI and an absence of substantial differentiation between strains of subspecies IV and VII. Because the overall degree of diversification of the invasion genes among the subspecies of S. enterica is roughly equivalent to that of the housekeeping genes, the inference is that they were already present in the ancestral form of the species. Hence, there is no need to invoke horizontal transfer of homologous genes from Yersinia spp. (5) or other outside sources into any of the contemporary lineages of S. enterica. However, the lack of strong differentiation between subspecies IV and VII points to the occurrence of at least one intersubspecific exchange of part or all of the invasion gene segment. The chromosome of subspecies VII apparently is a mosaic of large segments, some similar in sequence to those of subspecies IV and others highly distinctive in character.

EVOLUTION OF FLAGELLAR FILAMENT GENES

Serotype Evolution by Mutation

Laboratory studies of serological mutants of strains of Typhimurium (72) and Muenchen (62, 122) have shown that one or a few amino acid replacements are sufficient to alter flagellar antigenic character. The diversification of flagellins by point mutations in natural populations, leading to the production of new antigenic types, has been demonstrated by an analysis of sequence variation in the phase 1 flagellin gene (fliC) among 23 strains representing 19 ETs of Antarctica, Blegdam, Dublin, Enteritidis, Moscow, Naestved, and Rostock (J. Li, Ph.D. thesis, Pennsylvania State University, University Park, 1994). These strains have O serotype 1,9,12 or 9,12, but they exhibit nine combinations of eight flagellar antigenic factors (f, g, m, p, q, s, u, and z₆₃) of the g complex (43, 207). With two exceptions (En 2 and En 18), all of the ETs are closely related, and their fliC sequences differ from one another by only one or a few replacement substitutions or, in the case of Antarctica, by a six-codon duplication. These findings are consistent with Bruner’s (23) demonstration that strains of Blegdam (g,m,q) could be converted to Moscow (g,q), Enteritidis (g,m), or Dublin (g,p) by growth in medium containing the appropriate antiserum. The evolutionary relationships among the serovars indicated by MLEE and fliC sequence analyses are fully consistent with the results of restriction fragment length polymorphism (RFLP) typing with cloned chromosomal DNA, rRNA genes, and insertion sequences as probes (178).

Recombinational Basis of Serovar Diversity

The expression of many of the flagellin and polysaccharide serotypes by distantly related strains, even those belonging to different subspecies (Table 2), theoretically could reflect the retention of alleles from ancestral populations, convergence in epitope structure, or recombination of horizontally transferred fliC, fljB, or rfb genes. On the basis of the discovery, by MLEE analysis, that Enteritidis, Derby, Newport, and some other serovars are polyphyletic assemblages of distantly related strains, horizontal transfer and recombination events involving these genes were postulated to be relatively frequent (12, 155, 156); and for fliC, this hypothesis subsequently was supported by partial sequencing of the gene in strains of Typhimurium (172) and several other serovars (171). For example, between the diphasic serovars Typhimurium and Heidelberg, which are nearly identical in MLEE genotype and are distinguished serologically only by expression of phase 1 flagellar antigens i and r, respectively, there is a 19% nucleotide sequence difference in the central part of their fliC genes (171). Since this segment is invariant among strains of either serovar, it is apparent that the rate of evolution of fliC by mutation is not high enough to have generated this degree of difference; and, consequently, it is necessary to invoke the horizontal transfer of a fliC allele to create Heidelberg (r) from Typhimurium (i) or vice versa. This interpretation is strengthened by the discovery of a plasmid-borne fliC-like gene (flpA) in a triphasic strain of a normally diphasic serovar (173).

Table 2Distribution of serovars with phase 1 (fliC) serotypes of the g complex among the subspecies of S. entericaa

To test the generality of the horizontal transfer/recombination hypothesis, Li et al. (96) obtained the complete fliC sequences of 15 strains of several serovars of subspecies I, II, IV, and VII that express seven combinations of six phase 1 flagellar antigenic factors of the g complex (f, g, m, s, t, and z_{51). In} S. enterica, as in other bacteria (204), the terminal regions of the flagellin molecule (C1 and C2), which are involved in secretion and polymerization, are strongly conserved in both length and amino acid sequence, whereas the central region (V), which is the site of the epitopic variation assayed in serotyping (71, 96, 122, 128), is hypervariable (Fig. 4).

Fig. 4Variation in frequency of synonymous and nonsynonymous nucleotide substitutions along the length of the fliC gene among 15 strains of S.enterica. For sequential sets of 10 codons, the proportions of synonymous and nonsynonymous nucleotide sites at which each pair of strains differed were determined, and the means (PS and PN) for all pairwise comparisons are plotted. Limits of the C1, V, and C2 regions of the gene are indicated, together with the corresponding segments designated by Wei and Joys (196).

Source: (From Li et al [96].)

Individual evolutionary trees based on MLEE (indexing the overall genomic relatedness of the strains), the nucleotide sequence of the combined C1 and C2 regions of fliC, and the sequence of the V region of the gene are shown in Fig. 5. If the evolution of fliC has involved little or no recombination, all three trees should be topologically similar. In contrast, horizontal exchange of the V region or the entire fliC gene among strains would be indicated by a clustering of sequences specifying the same flagellin serotype, regardless of the overall genetic relatedness of the strains in which they occur (37, 191).

Fig. 5Evolutionary trees for 15 strains of S.enterica. For each strain, the phase 1 flagellin serotype is indicated. (A) MLEE tree (based on 27 enzyme loci). Bootstrap values greater than 50% are shown at the nodes. (B) Conserved (C1 + C2) regions of the fliC gene. Clusters of ETs of subspecies I and II are labeled. Bootstrap values for all nodes are greater than 50%. (C) Central variable (V) region of the fliC gene. Note that the scale is approximately half that of the conserved-sequence tree (B). Subspecies I serovar abbreviations are Ba, Banana; Be, Berta; De, Derby; En, Enteritidis; Nm, Newmexico; Or, Oranienburg; Ot, Othmarschen; and Pe, Pensacola. Roman numerals identify ETs of other subspecies.

Source: (From Li et al. [96].)

Although strains of Enteritidis (En 1) and Othmarschen (Ot 1), which express serotype g,m, are divergent in chromosomal character (Fig. 5A), their fliC sequences are nearly identical (Fig. 5B and C).

Strains Nm 1, IV 1, and VII 1 (serotype g,z₅₁) represent three subspecies, but both their V and C1 + C2 sequences cluster together (Fig. 5B and C).

ETs Pe 1, Ba 1, Or 1, and II 5 are of serotype m,t. The three subspecies I ETs are not closely allied, and all three are distantly related to II 5 (Fig. 5A). Their C1 + C2 sequences reflect these relationships, but all four ETs form a distinct, tight cluster in the V region tree. This region and part of C2 have been exchanged among the four ETs; and part of this distinctive sequence has also been transferred to ETs Be 1 and II 4.

In sum, the occurrence of each of three flagellin serotypes of the g complex in distantly related strains is clearly attributable to horizontal exchange rather than to convergence in epitope-determining amino acid sequences or retention of ancestral sequences. If antigens of this complex are typical of flagellar antigens in general—and there is no reason to believe that they are not—interstrain exchange and recombination are clearly major evolutionary mechanisms generating both allelic variation in fliC and serovar diversity in natural populations of the salmonellae.

Sources of fliC Sequence Recruitment

The occurrence of highly divergent families of fliC alleles in S. enterica (Fig. 6) suggests that sequences have occasionally been recruited from the flagellin genes of other species or even from other types of genes (52). This notion is supported by the observation that the G+C content of the central part of the gene is unusually low (Li, Ph.D. thesis). Additionally, gene conversion, which has been demonstrated experimentally for S. enterica flagellin genes (128), may explain why some antigenic factors that are normally intrinsic to the phase 2 flagellin (fljB gene) may also be encoded by fliC genes (139).

Fig. 6Evolutionary relationships of fliC alleles of S. enterica, E. coli, and Serratia marcescens, based on nucleotide sequences of the conserved (C1 and C2) terminal of the gene. Serotype formulae, specifying O, H1, and H2 antigens, are indicated. Bootstrap values are indicated at the nodes. O antigen is not expressed by laboratory strain K-12 or its derivatives, owing to the presence of mutations in the rfb genes; but it has been experimentally resurrected as O16 (100). (Adapted from Li, Ph.D. thesis.)

Source:

Evolutionary Loss of Flagella

Serovars Gallinarum and Pullorum, which are invasive pathogens of fowl (66), are obligatorily nonflagellate. Because strains of Gallinarum and Pullorum, which cause fowl typhoid and dysentery (pullorum disease), respectively, cannot be distinguished by routine serology, they are currently classified as biotypes (or bioserovars) of the same serovar (31, 43, 139).

MLEE analysis has identified Enteritidis as a close relative of the nonmotile salmonellae (97). The sharing of distinctive allozyme alleles at three metabolic enzyme loci (Fig. 7) and a premature stop codon in flgK (see below) by all ETs of Gallinarum and Pullorum indicates that they are monophyletic, an inference supported by similarities in fimbrial type (30) and in virulence plasmid size and structure (9, 29, 134). Moreover, this evidence indicates that their most recent common ancestor was nonmotile.

Fig. 7Hypothetical evolutionary scenario for 13 ET's represented by 133 strains of the nonmotile salmonellae. Numbers indicate allelic changes in MLEE genotype (29 loci assayed) (A) and nucleotide substitutions in the fliC gene (B). AO is the postulated most recent common ancestor of the serover Enteritidis and the nonmotile salmonellae. A1 is the most recent common ancestor of the three lineages of nonmotile salmonellae, represented by ET's of Gallinarum (Ga), Pullorum (Pu), and Gallinarum/Pullorum (Ga/Pu 1), which is intermediate between the Gallinarum and Pullorum lineages in both multilocus genotype and biotype. A2 is the most recent common ancestor of Pu 2 and Pu 3. Ga 2 is the globally predominant ET of Gallinarum, and Pu 3 is the common ET of Pullorum; Ets that are distinguished from Ga 2 or Pu 3 solely by the possession of one or two null enzyme loci alleles are indicated by lowercase letters.

Source: (From Li et al. [97].)

Confirming the results of an early transduction experiment (88), Li et al. (97) found that intact, but silenced, fliC genes are present in the nonmotile salmonellae (see also reference 80). The genes of strains of ETs Ga 3, Ga 4, and Ga/Pu 1 (Fig. 7) are identical in sequence to that of Enteritidis, but a nucleotide substitution has created a shared premature stop codon in strains of ETs Ga 1, Ga 2, and Ga 2a. The fliC sequences of the Pullorum ETs differ from the standard Enteritidis sequence in having nonsynonymous substitutions in two or three codons in the V region (Fig. 7).

The only available information concerning the condition of other genes of the flagellar regulon in the nonmotile salmonellae comes from sequence analysis of flgK (which encodes the first hook-filament junction protein) in isolates of Ga 2, Pu 3, and Ga/Pu 1. The flgK genes of Ga 2 and Pu 3 are intact and similar in sequence to the gene of En 1, but the sequences of all three ETs share a premature stop codon at position 125; and that of Ga/Pu 1 also has a 336-bp deletion.

Taken together, these findings indicate that loss of motility in the avian-adapted salmonellae occurred relatively recently in evolutionary time. The nonsense mutations in fliC and flgK presumably accumulated after gene inactivation, when, as a consequence of a mutation in another gene of the flagellar regulon (86, 102) that inhibited flagellar biosynthesis, they were no longer subject to purifying selection.

Cryptic genes are defined as silenced genes that can be reactivated by mutation (58, 98), as opposed to classical pseudogenes, which have degenerated beyond a point where there is any reasonable chance of restoration by mutation (99). Because reactivation of the intact fliC genes of Gallinarum or Pullorum would require restoration of flgK and the numerous other genes of the flagellar regulon that have undoubtedly accumulated molecular defects, they are in effect nondegenerated pseudogenes.

Many isolates of Dublin recovered in recent years from invasive infections in cattle have been nonmotile, although the condition apparently is reversible (163). Like Gallinarum and Pullorum, Dublin is strongly host adapted and is also a close relative of Enteritidis. It is also of interest that the strongly host-adapted shigellae, as well as most enteroinvasive strains of E. coli (170), are typically nonmotile, although at least two of the four Shigella species carry intact fliC genes (190). Modulation of flagellar biosynthesis may be a common strategy of invasive bacterial pathogens, with motility being advantageous under certain environmental conditions (95) and during colonization but disadvantageous in later stages of infection. Flagella are not expressed in vivo by strains of Enteritidis infecting chickens (28), and their permanent loss presumably evolved in the Enteritidis-like common ancestor of Gallinarum and Pullorum because motility was for some reason no longer required as the lineage became increasingly restricted in host range to fowl. Nonmotility of isolates of Pseudomonas aeruginosa recovered from chronically colonized cystic fibrosis patients has been attributed to an adaptive advantage in resisting phagocytosis by macrophages (103).

DISTRIBUTION OF Vi CAPSULAR ANTIGEN

The virulence antigen (Vi antigen), a capsular surface polysaccharide consisting of a variably acetylated homopolymer of galactosaminuronic acid (183), is known to be expressed only by S. enterica serovars Typhi, Paratyphi C, and Dublin and by some strains of Citrobacter freundii (156, 163, 174, 175). Multiple structural genes mediating the synthesis of Vi antigen and its transport to the cell surface are located in the ViaB region, a 14-kb chromosomal segment which has been sequenced and characterized in Typhi (61, 195).

Several horizontal transfer events involving the ViaB region are required to account for the unusual taxonomic distribution of the Vi antigen. Its presence in virtually all ETs of Typhi and Paratyphi C suggests that it has been in these lineages for long periods of time (156); but in the case of Dublin, acquisition of the ViaB region presumably occurred very recently, for it is confined to a single minority variant ET (Du 3) of the predominant, globally distributed ET (Du 1), which, moreover, has a limited geographic range, occurring only in France and Britain (163).

GENETIC STRUCTURE OF POPULATIONS

The Concept of Clonal Structure

Although the widespread geographic distribution of strains of S. enterica Typhimurium carrying specific mutations affecting biotype and fimbriation (129) and the association of particular serobiotypes of enterotoxigenic E. coli with independent outbreaks of specific diarrheal disease syndromes (132, 133) had earlier suggested the existence of clonal lineages, genetic evidence of a clonal structure for bacterial populations was first provided in 1980 by the recovery of strains of E. coli of identical multilocus enzyme genotypes at different times and places in North America (159). Soon thereafter, the concept of a clonal structure for the species E. coli as a whole was firmly established (2, 3, 25, 124, 126, 202, 203). Extended to S. enterica and other species, this concept has had a major impact on the fields of bacterial epidemiology and pathogenesis (1, 26, 56, 75, 76, 84, 113, 114, 130, 161, 162) and has recently been applied to parasitic protozoans as well (188). Indeed, it has sometimes been regarded as a paradigm for all bacteria (70), but linkage disequilibrium analyses of MLEE data do not support a long-term clonal structure for the pathogenic species Neisseria gonorrhoeae (108, 131, 162, 192) and P. aeruginosa (33, 162) or for major phylogenetic divisions of the soil bacterium Rhizobium meliloti (38). It is unlikely that any bacterial species either is strictly clonal or consists of freely recombining ("panmictic") populations.

For the salmonellae, a basically clonal population structure is evidenced by the presence of strong linkage disequilibrium among alleles at enzyme loci, the association of specific O and H serotypes with only one or a small number of multilocus enzyme genotypes, and the global distribution of certain genotypes (12, 108, 142, 155, 156, 163). S. enterica populations are clonal in the sense that the effective (realized) rates of recombination for most chromosomal genes are sufficiently low to permit the mutational diversification of cell lineages in terms of biochemical characteristics and ecological niche relationships, including host distribution, disease specificity, and virulence, and the long-term, if not permanent, maintenance of differentially adapted, widely distributed chromosomal genotypes in populations (160). Clonality explains why serotyping is a powerful marker system for recognizing groups of strains with distinct host ranges and pathogenicities, including those of closely related populations such as Dublin and Enteritidis (163). At the level of population structure resolved by MLEE, most of the serovars, including most of the common pathogens of humans and domesticated animals, are single ETs or families of closely related ETs; and for many of the medically important serovars, only a single ET is globally predominant at any one time (Table 3).

Table 3Relative abundance of the commonest ET of S. enterica serovarsa

An evolutionary genetic rationale for the prevalence of clonal population structure among pathogenic bacteria has been noted by Falkow (44). Because pathogenesis is a complex and multifactorial process involving the coordinated action of a large assembly of virulence-associated loci and genes that make it possible for the pathogen to survive in a variety of habitats, including those that enable it to recognize its hosts and then to avoid, subvert, or nullify their defense systems, there could be little adaptive radiation among pathogenic bacteria if indiscriminate horizontal genetic transfer and recombination occurred for genes throughout the genome.

Clones of Serovars That Cause Human Enteric Fever

Clonal aspects of the genetic structure of S. enterica are well illustrated by the populations of the serovars that are agents of enteric fever in humans, Paratyphi A, Paratyphi B, Paratyphi C, Sendai, and Typhi. With the singular exception of Paratyphi A and Sendai, MLEE analysis has demonstrated no close relationships among them (156), which implies independent evolutionary derivation. In the following discussion, ETs are equated with clones, unless otherwise noted.

The several clones of Paratyphi C and Choleraesuis are closely related and were derived from a common ancestor without serological modification, except for the acquisition of the Vi antigen by those of the Paratyphi C lineage. The ancestral form presumably was adapted to swine, as are the extant clones of Choleraesuis.

Paratyphi B is a large, heterogeneous group of lineages that are very closely related to those of Typhimurium and Saintpaul. The ability to cause enteric fever in humans has evolved only in the globally distributed clone Pb 1 (42, 155), and very recently, since it is only weakly differentiated.

Sendai and Paratyphi A differ in serotype and biotype but are genotypically almost identical. Clones of both serovars are related, albeit distantly, to those of the broad-host-range, noninvasive serovar Panama. Although the serovar Miami has the same antigenic formula as Sendai, it is closely related to Panama. Miami is polyphyletic, with only one group of its clones being human-adapted agents of gastroenteritis.

Both biochemically and genotypically, Typhi is an unusually distinctive and homogeneous serovar. By MLEE analysis, Selander et al. (156) found that 82% of 334 worldwide isolates were of one genotype (ET 1), with a second clone (ET 2) being represented by 16% of strains, all from western Africa. Additional evidence of genetic uniformity or low levels of variation in Typhi has been provided by studies of chromosomal DNA fingerprints (104), membrane protein profiles (45, 47), IS200 profiles (187), and RFLP patterns of plasmids conferring resistance to chloramphenicol (184). Moshitch et al. (111) found a uniform multiband RFLP pattern of EcoRI- or PstI-digested genomic DNA hybridized with a REP sequence oligoprobe. The RFLP pattern of EcoRI-digested DNA hybridized with a probe of genes coding for rRNA was also invariant; but variable multiband patterns were obtained following PstI digestion, as also reported by others (6, 135). Recently, Thong et al. (186) were able to detect several subclonal lineages involved in outbreaks of typhoid fever in Malaysia by the application of pulsed-field gel electrophoresis.

Throughout most of the world, Typhi is monomorphic for the d allele at the fliC locus encoding the phase 1 flagellin protein. But Indonesian populations are polymorphic for the d allele and a variant j allele (47, 104, 176), which was derived from d by an intragenic recombination event in which the pairing of a directly repeated 11-bp sequence resulted in the deletion of a 261-bp segment in the central part of the fliC gene (48). Some Indonesian strains of either phase 1 d or j also uniquely express a z₆₆ flagellar antigen (55), which presumably is encoded by a phase 2 locus (48), although they lack the hin and fljA genes that normally mediate phase shifting in S. enterica (111). Indonesian isolates, including those carrying the j allele and expressing the z₆₆ factor, are invariant and indistinguishable from other isolates of the globally distributed clone by both MLEE (156) and Southern blot hybridization of genomic DNA with a poly(GTG)₅ probe(34); but j allele strains reportedly have a distinctive HindIII-digested DNA fingerprint (47).

Frankel et al. (48) have proposed that Typhi originally evolved as a specialized human pathogen in Indonesia and initially was diphasic (d:z₆₆); mutations subsequently produced both the monophasic condition (d:-) and the j allele at the phase 1 locus, and a single subclone of the monophasic d type then spread worldwide from Indonesia. However, another explanation for the unique occurrence of flagellar polymorphism in Indonesia may be suggested, based on the fact that flagellar antigens are the major factors eliciting immune responses after infection with Typhi or vaccine administration (62).

Inasmuch as the specific deletion that changes d to j may be experimentally produced when Typhi is grown in anti-d serum (77, 173), it is likely that it occurs repeatedly in natural populations, where, however, it may normally be disadvantageous because motility is decreased (48). If so, the problem is to explain why it has managed to reach a frequency as high as 16% (176) in the Indonesian Typhi population. It is probably not coincidental that the frequency of typhoid fever in Indonesia is the highest in the world, with as many as 1,300,000 cases and 20,000 deaths per year (168). Moreover, it is primarily a disease of children, which suggests that the Indonesian population acquires immunity through exposure (182). Flagellar antigen diversity in Indonesia may well be an example of the maintenance of polymorphism by balancing selection for avoidance of immune responses in the host population.

Inter-Locus Variation in Recombination Rate

A generalization emerging from the analysis of gene sequences in populations of S. enterica and other bacteria is that the contribution made by recombination to allelic diversity varies markedly among loci encoding proteins of different functional types (1, 118, 121, 160, 166, 192), as well as among species and subdivisions of species. As shown in this review, the effective rate of recombination in S. enterica is low for genes encoding most metabolic enzymes and other types of housekeeping proteins, as well as for the inv/spa virulence genes. In contrast, horizontal transfer and recombination are a major source of allelic diversity for both the highly polymorphic flagellin fliC locus and the rfb genes that determine O-antigen structure. Entire fliC genes and parts or all of the epitope-determining central region have been frequently exchanged within and between subspecies, and some, if not most, alleles are mosaics of segments derived from several sources. Because flagellin is highly antigenic and interacts directly with the external environment, recombinant alleles may confer an immediate adaptive advantage to a bacterial cell and, as a consequence, be brought to high frequency in local populations by natural selection and then transferred to other lineages. The prevailing view is that the extensive flagellar antigenic polymorphism in S. enterica is adaptive in permitting the reinfection of hosts (24), and Reeves (143) has suggested that antigenic variation in both flagellin and the cell surface polysaccharide is subject to "niche-specific selection." The observation that sensitivity to flagellotropic bacteriophages may be serotype dependent (69) suggests another possible adaptive basis for flagellin polymorphism; and Skurnik and Toivanen (169) have proposed that resistance to bacteriophages is also an adaptive basis for O-antigen polymorphism.

Similar explanations in terms of environmental adaptation may apply to a number of genes in other bacteria for which evidence of horizontal transfer and recombination is available (Table 4). Genes that experience frequent horizontal transfer and recombination almost without exception encode or mediate the expression of products for which there would seem to be a premium on structural diversity or which confer adaptive traits such as antibiotic resistance (160). For housekeeping or virulence genes that encode polypeptides for which there is no premium on diversity in amino acid sequence per se, it is unlikely that either intragenic or assortative recombination would result in a selective advantage to the recipient cell. The probable fate of deleterious or selectively neutral recombinants is loss from the population through purifying selection and genetic drift (82). By comparing the degree of amino acid divergence in 179 homologous proteins in S. enterica and E. coli, Whittam (200) has shown that cell surface proteins, such as flagellins, porins, and pilins, are evolving at nearly 3 times the rate of cytoplasmic enzymes and proteins that function in transport, DNA replication, and the regulation of transcription and 10 times faster than ribosomal proteins.

Table 4Proteins and polysaccharides for which there is evidence that the encoding or mediating genes are subject to frequent horizontal transfer and recombinationa

Among the double-stranded DNA bacteriophages of enteric bacteria, horizontal transfer, mediated by site-specific recombination enzymes (invertases), has been a major factor in the evolution of tail fiber genes, with relatively frequent exchange occurring between otherwise unrelated phage groups (quasispecies) (150). Tail fiber genes of various phages and certain defective prophages share homologous segments, and individual genes may have a mosaic structure. This locus-specific exchange is adaptive in generating diversity in host range determinants in the face of strong selective pressure on phages to adapt to mutated host surfaces, which, in turn, are under selective pressure to avoid phage infection. Given the great variation in rate of recombination among the genes of S. enterica and other bacteria, it would not be surprising to find that there are comparable mechanisms that function to increase the frequency of exchange of certain genes or segments of the chromosome for which selection for diversity is strong.

In sum, comparative nucleotide sequencing has yielded increasing evidence that differing modes and strengths of natural selection among loci encoding or mediating the synthesis of products of different functional type influence the effective rate of recombination within and among populations of S. enterica and other bacteria. Because it encodes a typical metabolic enzyme, the elevated recombination rate of gnd in S. enterica is unexpected, but it may be attributed to the location of the gene in a segment of the chromosome where there is strong selection for allelic diversity at several neighboring loci. This has resulted in the near randomization of gnd sequences among strains of E. coli, but in S. enterica it has produced only a moderate increase in the frequency of recombination compared with that of other metabolic enzymes (120). The difference may be due in part, at least, to ecological differences that determine opportunities for genetic exchange within species and with other enterobacteria. As a basically commensal species, E. coli is a common element of the rich intestinal flora of higher vertebrates, whereas S. enterica is a specialized intracellular pathogen.

ACKNOWLEDGMENTS

Research leading to this review was made possible through the cooperation of L. Le Minor, M. Popoff, and J.-F. Vieu of the Institut Pasteur, Paris, France, and J. J. Farmer III and K. Wachsmuth of the Centers for Disease Control, Atlanta, Ga., in providing access to the large collections of isolates under their supervision. Important contributions were made by K. Ferris and A. McWhorter-Murlin in serotyping isolates. Research in our laboratory is supported by grant AI22144 from the National Institutes of Health and by funds from the endowment of the Eberly Professorship in Biology.