Bacterial Phylogenetic Clusters Revealed by Genome Structure

Abstract

Current bacterial taxonomy is mostly based on phenotypic criteria, which may yield misleading interpretations in classification and identification. As a result, bacteria not closely related may be grouped together as a genus or species. For pathogenic bacteria, incorrect classification or misidentification could be disastrous. There is therefore an urgent need for appropriate methodologies to classify bacteria according to phylogeny and corresponding new approaches that permit their rapid and accurate identification. For this purpose, we have devised a strategy enabling us to resolve phylogenetic clusters of bacteria by comparing their genome structures. These structures were revealed by cleaving genomic DNA with the endonuclease I-CeuI, which cuts within the 23S ribosomal DNA (rDNA) sequences, and by mapping the resulting large DNA fragments with pulsed-field gel electrophoresis. We tested this experimental system on two representative bacterial genera: Salmonella and Pasteurella. Among Salmonella spp., I-CeuI mapping revealed virtually indistinguishable genome structures, demonstrating a high degree of structural conservation. Consistent with this, 16S rDNA sequences are also highly conserved among the Salmonella spp. In marked contrast, the Pasteurella strains have very different genome structures among and even within individual species. The divergence of Pasteurella was also reflected in 16S rDNA sequences and far exceeded that seen between Escherichia and Salmonella. Based on this diversity, the Pasteurella haemolytica strains we analyzed could be divided into 14 phylogenetic groups and the Pasteurella multocida strains could be divided into 9 groups. If criteria for defining bacterial species or genera similar to those used for Salmonella and Escherichia coli were applied, the striking phylogenetic diversity would allow bacteria in the currently recognized species of P. multocida and P. haemolytica to be divided into different species, genera, or even higher ranks. On the other hand, strains of Pasteurella ureae and Pasteurella pneumotropica are very similar to those of P. multocida in both genome structure and 16S rDNA sequence and should be regarded as strains within this species. We conclude that large-scale genome structure can be a sensitive indicator of phylogenetic relationships and that, therefore, I-CeuI-based genomic mapping is an efficient tool for probing the phylogenetic status of bacteria.

Bacterial genera or species, as defined by current taxonomic approaches (18, 44), may include bacteria that in fact are not closely related phylogenetically. For example, the DNA G+C content of accepted Bacillus species varies from 32 to 69 mol% (8), although species of a genus are not normally expected to differ from one another by more than 10 mol%. This situation results largely from the lack of appropriate methodologies for classifying bacteria on a phylogenetic basis. Consequent errors in bacterial identification could be quite misleading both in basic research and in practical applications. Therefore, the ability to efficiently distinguish phylogenetic groups of bacteria within an apparent taxonomic species is desirable, especially for the correct diagnosis and timely treatment of infectious diseases caused by bacterial pathogens. Molecular techniques such as RNA and DNA or protein sequencing have been extremely useful in establishing the phylogenetic relationships among organisms and in discriminating bacteria by phylogeny. However, such sequence analyses provide information for only a tiny (though important) portion of the genome, and different choices of sequences for analysis can lead to remarkably variable conclusions (10, 21). Furthermore, it can be difficult to distinguish recently diverged bacteria solely on the basis of limited DNA or RNA sequence analysis. For example, 16S rRNA (or ribosomal DNA [rDNA]) sequencing (16), currently the most frequently used technique for phylogenetic studies of bacteria, is efficient only in resolving genera and higher taxa; for ranks lower than genus it loses resolution.

The physical structure of bacterial genomes as revealed by endonuclease mapping, on the other hand, could provide an alternative parameter to be employed in phylogenetic studies (26, 29). The mapping approach involves the use of endonucleases such as I-CeuI (17), XbaI, AvrII, et al. (24, 25). The genomic DNA fragments generated are then separated according to size by pulsed-field gel electrophoresis (PFGE). The maps thus established reveal the size and geometry of the genome, the copy number and genomic distribution of rrn operons, and major genomic events, e.g., insertions, duplications, deletions, inversions, and translocations of DNA segments (29–31). Closely related bacteria can have highly similar genome structures; the similarity decreases and eventually vanishes as the phylogenetic relationship becomes more distant.

In this study, we examine the relatedness of bacterial populations within taxonomic species by revealing and comparing their genome structures. We focus on two genera of pathogenic bacteria: Salmonella, which consists of more than 2,300 known species, or serovars (38, 39), that display a highly conserved genetic background (9, 19, 22, 23), and Pasteurella, which consists of about 25 species of both genetically and biologically more diverse bacteria (3, 7, 11, 36, 40). Our data indicate that the analysis of genome structure readily provides information useful to the elucidation and identification of phylogenetic clusters.

MATERIALS AND METHODS

Bacterial strains.

The Salmonella strains were obtained from J. Lederberg (LT2) (45), R. K. Selander (RKS strains) (1, 5, 6), and M. Y. Popoff (CIP8231 and 156-87) and are listed in Table 1. Pasteurella strains were obtained from the American Type Culture Collection, W. Albritton, S. Lundberg, and R. Lo and are listed in Table 2.

TABLE 1.

Salmonella strains

Strain no.	Original no.	Subgenus (species)	Antigenic formula	Source
LT2	LT2	Salmonella I (S. typhimurium)	1,4,[5],12:i:1,2
RKS4993	ATCC 9150	Salmonella I (S. paratyphi A)	1,2,12:a:−
RKS3222	DMS155/76	Salmonella I (S. paratyphi B)	1,4,[5],12:b:1,2	Human
RKS3614	IP87/87	Salmonella I (S. java)	1,4,[5],12:b:1,2	Cow
RKS5078		Salmonella I (S. gallinarum)	1,9,12:−:−
RKS53	CDCSSU7998	Salmonella I (S. enteritidis)	1,9,12:g,m:−
RKS2985	CDC151-85	Salmonella II	58:d:z6	Human
RKS2993	CDC3472-64	Salmonella II	42:f:g,t:−
RKS2990	3940-62	Salmonella II	47:b:1,5	Human
RKS3007	42-87	Salmonella II	60:g,m,t:z6
RKS2980	CDC346-86	Salmonella IIIa	62:z4,z23:−	Corn snake
RKS2983	CDC409-85	Salmonella IIIa	62:z36:−	Human
RKS2982	187-87	Salmonella IIIa	48:z4,z24:−	Human
RKS2984	98-84	Salmonella IIIa	18:z4,z32:−	Human
RKS2978	CDC156-87	Salmonella IIIb	501,2,3:k:z	Human
CIP8231	CIP8231	Salmonella IIIb
156-87	156-87	Salmonella IIIb
RKS2975	128-87	Salmonella IIIb	48:i:z	Human
RKS3027	CDC287-86	Salmonella IV	16:z4,z32:−	Human
RKS3022	484-85	Salmonella IV	48:g,z51:−	Human
RKS3025	145-86	Salmonella IV	43:z4,z23:−	Vacuum cleaner
RKS3026	208-86	Salmonella IV	16:z4,z32:−	Iguana
RKS3041	CDC750-72	Salmonella V (S. bongori)	66:z41:−	Frog
RKS3044	CDC2703-76	Salmonella V (S. bongori)	48:z41:−	Parakeet
RKS3040	327-80	Salmonella V (S. bongori)	66:z65:−	Fishmeal
RKS3042	1308-83	Salmonella V (S. bongori)	48:a:−	Human
RKS2995	CDC1363-65	Salmonella VI	45:a:e,n,x
RKS3052	4603-68	Salmonella VI	11:b:1,7
RKS3053	1411-60	Salmonella VI	1,6,14,25:a:e,n,x	Coconut
RKS3055	2131-71	Salmonella VI	41:b:1,7	Opossum
RKS3014	CDC5039-68	Salmonella VII	40:z4,z24:[z39]	Human
RKS3804	513/69	Salmonella VII	41:z4,z23:−

TABLE 2.

Pasteurella strains

Species and strain no.	Alternative no.	Host	Serotype	Genome group	GenBank nucleotide sequence accession no.
P. haemolytica
H099	ATCC 33367	Sheep	T3	Pha1	AF176279
H100	ATCC 33368	Sheep	T4	Pha1
H106	ATCC 33374	Sheep		Pha1	AF176280
H156		Sheep		Pha1
H157		Sheep		Pha1
H158		Sheep		Pha1
H160		Sheep		Pha1	AF176281
H161		Bighorn		Pha1
H162		Bighorn		Pha1
H163		Bighorn		Pha1
H164		Bighorn		Pha1
H159		Sheep		Pha1
H059		Cattle		Pha2	AF176282
H044		Cattle	A1	Pha3
H046		Cattle	A1	Pha3
H102	ATCC 33371	Sheep	A7	Pha4	AF176283
H103	ATCC 33372	Sheep	A8	Pha5	AF176284
H104	ATCC 33373	Sheep	A9	Pha6
H094		Cattle	A1	Pha6
H105	ATCC 33369	Sheep	A5	Pha7
H093		Cattle	A1	Pha8
H095		Cattle	A1	Pha9
H096		Cattle	A1	Pha10
H196		Cattle	A1	Pha10
H045		Cattle	A1	Pha11	AF176285
H187		Cattle		Pha12	AF176286
H188		Cattle		Pha13	AF176287
H060		Sheep		Pha14	AF176288
P. multocida
H135	NCTC10322			Pmu1	AF176289
H134		Human		Pmu1
H048		Bovine		Pmu1
H063		Chicken		Pmu1
H064		Pig		Pmu1
H119		Bovine		Pmu1
H121		Bovine		Pmu1
H122		Bovine		Pmu1
H125		Bovine		Pmu1
H126		Bovine		Pmu1
H127		Bovine		Pmu1
H236		Pig		Pmu1	AF176290
H229		Pig		Pmu2	AF176291
H061		Mallard duck		Pmu2
H062		Turkey		Pmu2
H065		Mallard duck		Pmu2
H118		Bovine		Pmu2
H130		Human		Pmu2
H140	ATCC 15742	Turkey		Pmu2
H047		Pig		Pmu2
H139	ATCC 11039	Domestic fowl		Pmu2	AF176292
H232		Pig		Pmu2
H233		Pig		Pmu2
H234		Pig		Pmu2
H116		Bovine		Pmu3	AF176293
H117		Bovine		Pmu3	AF176294
H120		Bovine		Pmu3
H239		Pig		Pmu4	AF176295
H235		Pig		Pmu5	AF176296
H231		Pig		Pmu6	AF176297
H129		Bovine		Pmu7	AF176298
H123		Bovine		Pmu7	AF176299
H154		Bovine		Pmu8	AF176300
P. pneumotropica
M10				Pmu1
P. ureae
M14				Pmu9	AF176301

Enzymes and chemicals.

I-CeuI was purchased from New England BioLabs; proteinase K was from Boehringer Mannheim. Most other chemicals were from the Sigma Chemical Co.

PFGE methods and genomic mapping.

Preparation of intact genomic DNA, endonuclease cleaving of DNA in agarose blocks, and separation of the DNA fragments by PFGE were as described previously (27, 29). PFGE was performed with the Bio-Rad contour-clamped homogeneous electric field (CHEF) mapper, Bio-Rad CHEF DRII, or Hoefer Hula electrophoresis system. Genomic mapping methods with I-CeuI were as described and modified previously (26, 33).

16S rDNA sequencing and data analysis.

Partial nucleotide sequencing of 16S rDNA, from nucleotides 800 to 1450 (Escherichia coli K-12 numbering), was carried out for 23 representative Pasteurella strains. Genomic DNA was used directly as templates for the sequencing after isolation with proteinase K, fragmentation by sonication, and purification with the Qiagen kit. The following primers were used: Pas1, 5′TACGG(C/T)TACCTTGTTACGACT3′ (for nucleotides 1130 to 1450), and Pas2, 5′TCTCCTTTGAGTTCCCGA3′ (for nucleotides 800 to 1130). The generated 16S rDNA sequences and some reference sequences obtained from GenBank were analyzed with the PHYLIP programs (13, 14).

Nucleotide sequence accession numbers.

The GenBank accession numbers for the 23 16S rDNA segments sequenced in this study are given in Table 2.

RESULTS

Construction of genome maps by complete and incomplete I-CeuI cleavages of genomic DNA.

To explore the possible utility of structural analysis in the comparison of diverse bacterial genomes, we elected first to validate the method with a group of closely related bacteria and then to apply it to bacteria for which greater diversity might be anticipated. Two very special features of I-CeuI make it an excellent endonuclease for bacterial genomic mapping: (i) it directly reveals the copy number of rrn operons because it cleaves exclusively within the 23S rRNA gene (26, 35) and (ii) it easily generates incomplete cleavage products at a wide range of concentrations. The incomplete cleavage bands on a PFGE gel help determine the neighboring relationships among completely cleaved DNA fragments. Figure 1A shows the I-CeuI cleavage pattern of genomic DNA of Salmonella typhimurium LT2 on a PFGE gel, with the incomplete as well as the complete cleavage bands indicated; Fig. 1B shows the genome map based on the data from Fig. 1A.

FIG. 1.

Construction of an I-CeuI genome map. (A) I-CeuI cleavage pattern of genomic DNA of S. typhimurium LT2, with bands and their sizes indicated (complete cleavage products labeled with larger letters than the incomplete ones). Lane MW, molecular size markers. (B) Circular genome map based on complete as well as incomplete cleavage data in panel A. PFGE: CHEF Mapper, 30 s ramping to 120 s, 6 V/cm, 120°, 18 h.

Genome structure of Salmonella strains.

Based on the extraordinary similarity in genome structure among Salmonella species within subgenus I (29, 32) and even between Salmonella and E. coli (26, 42), a comparable level of similarity was expected for Salmonella strains in all eight subgenera. Figure 2A shows the PFGE patterns of I-CeuI-cleaved genomic DNA of representative strains of the eight Salmonella subgenera, and Fig. 2B shows the maps. There is substantial similarity in the sizes and map positions of the seven I-CeuI fragments among the 32 strains shown, although subtle but unique differences among the subgenera exist. For example, a small fragment C (450 kb) is a unique feature of all subgenus IIIa strains (Fig. 2A and B). Occasionally, unusual fragment sizes were detected in strains of any of the eight subgenera, probably as a result of genomic reorganizations (e.g., the larger B or G fragments in lanes 9, 15, 17, 18, and 31). Because similarities in genome structure among species of the same subgenera are already too great to distinguish readily among them and because our earlier data had demonstrated identical genome structure among strains of the same species, e.g., S. typhimurium (29), S. paratyphi A (28), S. paratyphi B, S. java, and many other Salmonella species (23a), we chose not to present data for strains of individual Salmonella species here. Complete 16S rRNA sequences from ca. 30 Salmonella strains covering more than 20 Salmonella species and from E. coli K-12, obtained from public databases, have more than 98% identity, both among the Salmonella subgenera and between Salmonella and E. coli (data not shown), demonstrating the genetic homogeneity of Salmonella as well as the great similarity between Salmonella and E. coli. These data therefore suggest that not only is sequence divergence between Salmonella and E. coli modest but also that large-scale organizational patterns of the genomes are well conserved (Fig. 2B).

FIG. 2.

Genome structure of Salmonella spp. (A) I-CeuI cleavage patterns of representative Salmonella spp. of all eight subgenera; (B) genome maps based on data in panel A. The circular maps are presented here as linear for more convenient comparisons among them, with only a part of fragment A shown. PFGE: CHEF DRII, 30 s ramping to 120 s, 170 V, 20 h.

Genome structure of Pasteurella strains.

To extend our analysis, we next considered members of Pasteurella spp., for which taxonomy has been more problematic (3). We focused on two Pasteurella species, Pasteurella haemolytica and Pasteurella multocida, the leading causative agents of pasteurellosis in livestock, including hemorrhagic septicemia in cattle and pneumonic infections in sheep and goats. We also included examples of Pasteurella ureae and Pasteurella pneumotropica for comparison. I-CeuI cleavage of genomic DNA generated six bands in all Pasteurella strains except H103, which has seven, possibly resulting from a duplication of the 60-kb fragment (see Fig. 4). All Pasteurella strains have circular genomes, as exemplified by P. haemolytica H099 in Fig. 3, with linear maps presented in Fig. 4B for more convenient comparisons among them. P. haemolytica and P. multocida have very different overall I-CeuI cleavage patterns (Fig. 4A) and genome maps (Fig. 4B). Even within species, differences in I-CeuI cleavage pattern and genome map among the strains are also very obvious. Some strains look very similar when the complete cleavage bands are compared; however, their different incomplete cleavage bands indicate different arrangements of the DNA segments (Fig. 4). Altogether, we made genome maps for 63 Pasteurella strains, including 28 identified as P. haemolytica, 33 as P. multocida, and one each as P. ureae and P. pneumotropica (Table 2). We then tried to divide the Pasteurella strains into groups according to their similarity in genome structure, so that genomes with the same or very similar complete as well as incomplete I-CeuI cleavage patterns were grouped together (Fig. 4 and Table 2).

FIG. 4.

Genome structure of Pasteurella spp. (A) I-CeuI cleavage patterns of representative Pasteurella strains. Fragment designations and sizes are given for strain H059 in lane 1. Note that fragments C (60 kb) and F (40 kb) of H059 are shown as four bands on this PFGE gel. This reflects artifactual retardation of some bands that is occasionally seen near the margins of some gels, especially with smaller fragments. (B) Genome maps based on data in panel A. The circular maps are presented here as linear ones for more convenient comparisons. PFGE: CHEF DRII, 10 s ramping to 100 s, 160 V, 20 h; 50 s ramping to 60 s, 160 V, 6 h; 10 s ramping to 20 s, 160 V, 3 h.

FIG. 3.

I-CeuI genome map of P. haemolytica H099. (A) PFGE pattern of genomic DNA after I-CeuI cleavage; (B) circular genome map based on the data in panel A. The map was established based on several PFGE runs under different pulsing conditions to resolve all size ranges. In this particular PFGE gel, for example, the resolution is good for fragments of ca. 500 kb and smaller but not for fragment A and its combinations with fragment B or F, which are resolved under other conditions. PFGE: CHEF DRII, 10 s ramping to 90 s, 160 V, 20 h; 10 s ramping to 30 s, 160 V, 6 h; 80 s ramping to 120 s, 120 V, 14 h.

The P. haemolytica complex.

The 28 P. haemolytica strains were divided into 14 genome groups (Pha groups). They all have a large I-CeuI fragment, 1.1 Mb or larger. Group 1 contains 12 strains that have nearly identical genome maps that are very different from those of all the other P. haemolytica strains. They are biotype T strains, and it has recently been suggested that they be elevated to a new species, Pasteurella trehalosi (43). Group 2 contains only one strain in our collection, and it has a genome structure totally different from that of strains of either group 1 or groups 3 to 14. Groups 3 to 14 each contain one or two strains which have very similar I-CeuI cleavage patterns, indicating similar genetic contents, but different genome maps. H103 in group 5 is a very special case: it has an extra 60-kb fragment probably resulting from a genomic duplication; otherwise it is similar to the rest of the cluster comprising Pha groups 3 to 14.

As shown in Fig. 4B, the Pha groups differ not only in genome structure but also in genome size: Pha groups 1 and 2 have small genomes, 2,130 and 2,190 kb, respectively, and Pha groups 3 to 14 have larger genomes, around 2,600 kb.

The P. multocida complex.

To determine whether the diversity seen in the P. haemolytica complex might also exist in other Pasteurella spp., we undertook comparable analyses of P. multocida. The 33 P. multocida strains we analyzed were also quite diverse and were divided into eight groups (Pmu groups). Group 1 has 12 strains and contains the type strain, NCTC10322 (H135). Groups 1, 2 (12 strains), 3 (3 strains), and 4 to 6 (1 strain each) have somewhat similar I-CeuI cleavage patterns, although the lengths of some homologous fragments could be variable (Fig. 4). In contrast, groups 7 (two strains) and 8 (one strain) are dramatically different in genome structure from the majority of P. multocida strains. Interestingly, the P. pneumotropica strain, M10, has a genome structure that is very similar to those of group 1 P. multocida strains and therefore was included with them. The P. ureae strain, M14, was also placed in the P. multocida complex as the sole strain of group 9 because of its great similarity in genome structure to most P. multocida strains.

The striking diversity in genome structure among strains of the same species (e.g., Pha group 1 or 2 versus Pha groups 3 to 14 and Pmu groups 7 and 8 versus groups 1 to 6) and contrastingly the great similarity among some strains of P. multocida, P. pneumotropica, and P. ureae make it critical to establish the phylogenetic relationships among these bacteria. Molecular methods such as 16S rDNA sequencing would provide data that may either support or challenge our hypothesis that phylogenetically closely related bacteria usually have similar genome structures.

16S rDNA sequence analysis of Pasteurella strains.

We therefore determined portions of the 16S rDNA sequences for 23 representative Pasteurella strains and compared them. We chose bases 1100 to 1450 (E. coli numbering) because this segment is one of the divergent regions of the molecule among different bacteria but not the most divergent region (23a). We tried to avoid sequencing and comparing the most constant as well as the most divergent regions of the 16S rDNA molecules to estimate phylogenetic relationships among these bacteria at a certain level. For example, the overall sequence identity of the chosen segment between the P. haemolytica and P. multocida complexes is about 95%, whereas this same segment is completely identical between E. coli and S. typhimurium (and among many other Salmonella species; data not shown). These data support the notion deriving from the I-CeuI analyses that Pasteurella species are indeed quite diverse phylogenetically. Figure 5 presents 240 bases that contain most of the divergent parts of the compared sequences.

FIG. 5.

Comparison of 16S rDNA sequences among representative Pasteurella strains with that of E. coli K-12 included as a reference. Bases 1151 to 1390 (E. coli numbering) are compared here.

Within the P. haemolytica complex, the compared 16S rDNA sequences are up to 4% divergent among Pha groups 1, 2, and 3 to 14, with Pha groups 3 to 14 having an identical sequence in this segment of 16S rDNA. The Pha group 1 strains form a very tight cluster and stand out from all other P. haemolytica strains, supporting their phylogenetic status revealed by genome structure and their recent reclassification as a novel Pasteurella species, P. trehalosi (43) (see Fig. 6). The P. multocida complex is also strikingly divergent among the strains: H154 of Pmu group 8 is about 5% divergent from both the P. haemolytica complex and the other genome groups of the P. multocida complex, with all the other strains in this complex being 2% different or less. Pmu groups 1, 2, 4, and 7 are identical in this segment of 16S rDNA, indicating very close phylogenetic relationships among them. Groups 3, 5, and 6 are all about 2% different from the cluster comprising groups 1, 2, 4, and 7. Interestingly, M14, the P. ureae strain, is only 1% different from groups 1, 2, 4, and 7. M10, the P. pneumotropica strain, is in Pmu group 1 and has the same 16S rDNA sequence as other strains of group 1. In the phylogenetic tree in Fig. 6, which was constructed based on the 16S rDNA data, E. coli and S. typhimurium are included as an outgroup. In the tree, two Pasteurella clusters are formed, one for the P. haemolytica complex and one for the P. multocida complex, with the two being connected by H154, the Pmu group 8 strain.

FIG. 6.

Phylogenetic tree of the Pasteurella genome groups based on their 16S rDNA data, with E. coli and Salmonella included as an outgroup.

DISCUSSION

To resolve the phylogenetic status of bacterial populations in the same taxonomic groupings, we tested our genome structural analysis methodology on two bacterial genera, Salmonella and Pasteurella. With Salmonella, we wished to confirm and extend our previous observations that bacterial genome structure can be stable over long evolutionary times (26, 29) and that closely related bacteria tend to have similar genome structures. Several lines of evidence indicate that E. coli and Salmonella may have diverged over 100 million years ago (12, 15, 37); some of the eight Salmonella subgenera may also have diverged for similarly long evolutionary times. Even so, their rDNA sequences and genome structures are hardly distinguishable, with very rare exceptions (30, 31). Thus, it would appear that large-scale genomic organization as we display it here evolves very slowly in at least some bacteria.

Pasteurella is one of the four recognized genera of the bacterial family Pasteurellaceae, the taxonomy of which has been a very confusing area of study for several decades. Specific and generic names have been changed back and forth frequently for many of the taxa in this family, a situation resulting mainly from the lack of a methodology for assessing the phylogenetic status of these bacteria reliably, efficiently, and conveniently. The situation within individual Pasteurella species is equally confusing: it is not known how phylogenetically diverse the bacteria within a taxonomic grouping might be. Our genome structure analysis technology thus appears to provide a method of resolving members of a bacterial taxon into distinct groups on a phylogenetic basis by a simple comparison of rrn-delimited fragment sizes and a relative ordering of these fragments along the genome (Fig. 4). Nevertheless, it is possible that some unusual genomic rearrangements, especially those not mediated by rrn operons as seen in Pmu group 7 strains, may obscure the phylogenetic groupings. In these cases, PFGE analysis should be used in conjunction with rDNA data to obtain more robust interpretations. While our method is insensitive to the accumulation of point mutations in the genome, it easily reveals large genomic insertions, deletions, and other recombinational events that may occur during the evolutionary diversification of the bacteria.

In close correspondence with our evidence of structural diversity, the multiple genome groups that were revealed in Pasteurella by I-CeuI mapping are also highly divergent in 16S rDNA sequence (Fig. 5). If the same criteria for defining bacterial species as those used for E. coli and Salmonella taxonomy are applied, then on the basis of 16S rRNA sequence information and other types of information including genome structure, at least some of these genome groups, e.g., Pha groups 1 and 2 and Pmu group 8 (Fig. 4 and 6), should have the taxonomic status of genus or higher. Some other genome groups, such as Pha groups 3 to 14, however, are more closely related to one another; in fact, their behavior is similar to that of S. typhi strains (31) in that they have nearly identical genetic contents though with occasional differences in the ordering of fragments, presumably reflecting recombination events arising at rrn operons (31). The situations for both Pasteurella Pha groups 3 to 14 and S. typhi are consistent with the adopt-adapt model of bacterial speciation (34), which hypothesizes that bacteria speciate from ancestors by acquisition of novel genetic material (adopt) followed by adaptive genomic reorganization (adapt). By this hypothesis, Pha groups 3 to 14 might be the products of independent genomic reorganizations in individual lineages of a P. haemolytica ancestor through homologous recombinations between rrn operons in the process of genomic rebalancing (30). In this sense, strains of Pha groups 3 to 14 may still be members of the same phylogenetic species, having identical biological properties and dwelling in the same or similar niches.

The P. multocida complex has a similar situation. Except for Pmu groups 7 and 8, all P. multocida strains have similar I-CeuI cleavage patterns, with most homologous bands among the strains recognizable by size only (and confirmed by DNA hybridization; data not shown). However, strains of Pmu groups 1 to 6 are much more diverse in terms of the lengths of both individual I-CeuI fragments and whole genomes than are strains of Pha groups 3 to 14. Difference in genome size means different genetic contents; therefore, strains of Pmu groups 1 to 6 may no longer be members of the same phylogenetic species, although they could still be very closely related phylogenetically. Strains of Pmu groups 1 to 6 could be the products of recent evolutionary events, e.g., gain or loss of genetic material, with homologous I-CeuI fragments being organized in different ways in the Pmu groups as a mechanism of compensation for the genomic imbalance caused by the gain or loss of DNA (34). Pmu group 7 is a special case: it has a PFGE pattern and a genome map that seemingly do not resemble those of any other P. multocida strains. However, its 16S rDNA is identical to those of the majority of P. multocida strains, a situation that might be explained by genomic rearrangement through several crossovers. The hypothesized genomic evolution of P. multocida groups 1 to 7 is shown in Fig. 7.

FIG. 7.

Hypothesized genomic evolution of P. multocida genome groups. Recombinations rearrange the genomes, but the starting events are assumed to be gain or loss of genomic DNA (see text for details of the hypothesis). This drawing considers only the “net” results, ignoring other possible genomic reorganizations that might have occurred in between. In these genomic reorganizations, at least one rule might have to be observed: transcription of rrn genes has to be in the directions away from, not toward, the origin of replication (30).

Of the known mechanisms by which bacteria diverge, differences in genome size might be of the greatest significance, as shown in the comparisons among the P. haemolytica strains: their genomes could be up to 20% different (compare strains of Pha group 1 or 2 with those of Pha groups 3 to 14). Consistent with these observations, recent studies show that various natural isolates of E. coli may also differ by up to 20% in genome size (2, 20, 41). Horizontal transfer of novel genes or gene clusters that contribute to pathogenicity likely accounts for much of this variation (41), with clustered insertions distributed at preferred genomic sites (2). Thus, these data suggest that macrovariation in genome structure may be a sensitive marker of both phylogeny and phenotype.

Genomic change over time, therefore, may not be a smooth process. Rather, it could be a quite saltatory process, depending on whether the change involves vertical or horizontal inheritance, because stochastic events can significantly change bacterial genome structure (size and geometry), as well as phenotypic properties, rapidly. For example, the acquisition of specific loci in pathogenicity islands of various E. coli isolates may be important contributors to the divergence of these bacteria in both genome structure and the specific diseases they cause (4). In this sense, our genome structure analysis techniques can better resolve phylogenetic groupings of bacteria whose changes in genome structure have resulted from horizontal acquisition of genetic material than those whose changes have resulted from a vertical mode. Our data for Salmonella, Pasteurella, and other bacteria (30) imply that the horizontal mode dominates in nature in the divergence and speciation of bacteria.

Little information about P. pneumotropica and P. ureae as species is available. The P. pneumotropica strain has a 16S rDNA sequence and a genome map that are identical to those of P. multocida genome group 1 strains, implying either (i) that this strain is a misidentification, (ii) that P. pneumotropica should be combined with P. multocida, or (iii) that P. pneumotropica may have recently speciated, not having sufficient time to permit divergence in 16S rDNA. The P. ureae strain, M14, is interesting: it may have just begun diverging from P. multocida both in 16S rDNA (99% similarity) and in genome structure (a 1-Mb fragment which is not seen in P. multocida strains).

H154 was identified as a P. multocida strain, but phylogenetically it is equally far away from P. haemolytica and P. multocida, as judged by 16S rDNA sequence. Its genome structure is also different from those of P. multocida. In fact, it looks like an ancestral strain from which both P. haemolytica and P. multocida might have diverged.

I-CeuI is the only endonuclease so far known to directly reveal these conservative aspects of the bacterial genome. A PFGE gel can contain as many as 60 lanes and, typically, a gel run can be finished in 2 or 3 days. Therefore, theoretically, this technique allows one to construct more than 100 genome maps within a week with one PFGE machine, making it possible to carry out systematic genomic comparisons involving thousands of bacterial strains. Unlike 16S rDNA and rRNA sequences, which change continuously among bacteria, genome structure has a clear-cut nature among different phylogenetic groups of bacteria (compare, for example, Pha groups 1, 2, and 3 to 14). Therefore, the identification of phylogenetic taxa of bacteria may eventually be established based on their genome structure. Finally, the genome structure methodology may help elucidate mechanisms of bacterial evolution. Quantitative measures of genomic organization and structural relatedness will further establish the utility of this methodology in bacterial phylogenetics and taxonomy.