We compared the DNA sequences of the two chromosomes of Vibrio cholerae [4] with the sequence of the Escherichia coli chromosome [5] using a suffix tree alignment algorithm [6]. The analysis revealed a significant alignment at the DNA level between the V. cholerae large chromosome (chrI) [4] and the E. coli chromosome [5] spanning the entire length of these chromosomes (Figure 1a). Analysis of the reverse complement of V. cholerae chrI with E. coli also produced a significant alignment (Figure 1b). When superimposed, the two alignments produce a clear 'X' shape (Figure 1c) that is symmetric about the origin of replication of both genomes. This symmetry indicates that matching sequences tend to occur at the same distance from the origin but not necessarily on the same side of the origin. The X-alignment between V. cholerae and E. coli was found to be statistically significant using a test based on the number of matches found in diagonal strips in the alignment (see the Materials and methods section). Specifically, when V. cholerae chrI is aligned in the forward direction against E. coli, there are 459 maximal unique matching subsequences (MUMs; see the Materials and methods section), of which 177 occurred in a diagonal strip covering 10% of the total area (compared to the expected value of 46). The probability of observing this high a number of MUMs by chance is 4.7 × 10^-59. The alignment of V. cholerae chrI in the reverse direction against E. coli (which corresponds to the MUMs on the anti-diagonal) has a probability of 1.8 × 10^-90. As a control, we compared the genomes of distantly related species, such as E. coli and Mycobacterium tuberculosis. These do not show a significant X-alignment (Table 1).

We have found that X-alignments of whole genomes are not limited to the V. cholerae versus E. coli comparison. For example, a whole-genome comparison of two bacteria in the genus Streptococcus - S. pyogenes [7] and S. pneumoniae (H. Tettelin, personal communication) - reveals a global X-alignment similar to that of V. cholerae versus E. coli (Figure 1d) which is also statistically significant (Table 1). In addition, an X-alignment is found between two species in the genus Mycobacterium - M. tuberculosis [8] and M. leprae [9] (Figure 1e) - as well as between two strains of Helicobacter pylori (data not shown). The X-alignments observed between any two pairs of genomes are not identical in every aspect. For example, in the alignment between the two Mycobacterium species, each conserved region is much longer than in the other genome pairs. We believe this is due to different numbers of evolutionary events between the species (see below). Whole-genome X-alignments were not found between any other pairs of species, although a related pattern was seen between some of the chlamydial species (see below).

To test whether the X-alignments found in the DNA analysis could also be found at the level of whole proteomes, we conducted comparisons of homologous proteins between species (see the Materials and methods section). Figure 2a shows a scatterplot of chromosome positions of all proteins homologous between V. cholerae chrI and E. coli. The presence of many large gene families causes a great deal of noise in this comparison. This noise can be reduced by considering only the best matching homolog for each open reading frame (ORF), rather than all protein homologs (Figure 2b). This filtered protein comparison results in an X-alignment that is statistically significant (Table 2).

The finding of the X-alignment pattern between species led us to search for similar patterns within species; that is, global alignments of a genome with its own reverse complement. Of the genomes for which we found between-species X-alignments (M. tuberculosis, M. leprae, S. pyogenes, S. pneumoniae, E. coli and V. cholerae), statistically significant self-alignments are detected for all except M. tuberculosis (Figure 3; probabilities shown in Table 1). Interestingly, these self-alignments are not as strong as those between species. Proteome analysis also shows an X-alignment within species (shown for V. cholerae chrI in Figure 2d; probabilities shown in Table 2). The X-alignment of proteins within V. cholerae chrI is statistically significant only for recently duplicated-genes, but disappears when all paralogs are included. The importance of filtering for recent duplications is discussed below.

One possible explanation for an X-alignment within and between species is an ancestral inverted duplication of the whole genome, as has been suggested for E. coli [10]. The weak or missing X-alignment within species could be explained by gene loss of one of the two duplicates of many of the pairs of genes in the different lineages. Gene loss has been found to follow large chromosomal or genome duplications [11,12,13]. This gene loss is thought to stabilize large duplications by preventing recombination events between duplicate genes. If gene loss is responsible for the weak X-alignment within species, then to maintain the X-alignments between species, the member of the gene pair lost in a particular lineage should be essentially random. If an ancient inverted duplication followed by differential gene loss is the correct explanation for the observed X-alignments, one would expect the genes along one diagonal to be orthologous between species (related to each other by the speciation event), while the genes along the other diagonal should be paralogous (related to each other by the genome duplication event before the speciation of the two lineages). However, the evidence appears to contradict this model: likely orthologous gene pairs are equally distributed on each diagonal (data not shown).

A second possible explanation for the X-alignments is that an underlying mechanism allows sections of DNA to move within the genome but maintains the distance of these sections from the origin and/or terminus. There are a variety of possible mechanisms for such movement, but we believe the most likely explanation is the occurrence of large chromosomal inversions that pivot around the replication origin and/or terminus. Large chromosomal inversions, including those that occur around the replication origin and terminus, have been shown to occur in E. coli and Salmonella typhimurium in the laboratory (see, for example, [14,15,16,17,18]). The occurrence of such inversions over evolutionary time scales was first suggested by comparative analysis of the complete genomes of four strains in the genus Chlamydia [19]. In that study, we found that the major chromosomal differences between C. pneumoniae and C. trachomatis (shown in Figure 2c) were consistent with the occurrence of large inversions that pivoted around the origin and terminus (including multiple inversions of different sizes). In Figure 4 we present a hypothetical model showing how a small number of inversions centered around the origin or terminus could produce patterns very similar to those seen in the Chlamydia, Mycobacterium and Helicobacter comparisons. The continued occurrence of such inversion over longer time scales would result in an X-alignment similar to that seen in the V. cholerae versus E. coli and S. pneumoniae versus S. pyogenes comparisons. Thus the different between-species X-alignments could be the result of different numbers of inversions between particular pairs of species.

Inversions about the origin and terminus could also produce an X-alignment within species, through the splitting of tandemly duplicated sequence. Many sets of tandemly duplicated genes are found in most bacterial genomes [19,20] (also see Figure 3a,c). As tandem duplications are inherently unstable (one of the duplicates can be rapidly eliminated by slippage and/or recombination events [21]), the fact that many tandem pairs are present within each genome suggests that tandem duplications occur frequently. Thus, it is reasonable to assume that occasionally a large inversion will split a pair of tandemly duplicated genes. An inversion that pivots about the origin and also splits a tandem duplication will result in a pair of paralogous genes spaced symmetrically on opposite sides of the origin.

If our inversion model is correct, then the genes along both diagonals in the between-species alignments should be orthologous, which is the case (see above). In contrast, genes along the anti-diagonal in the within-species X-alignments should be recent tandem duplicates that have been separated by inversions. This also appears to be the case - in the within-species analysis of V. cholerae chrI ORFs, the X-alignment shows up best when only recent duplicates are analyzed (Figure 2d). The splitting of tandem duplicates by inversions may be a general mechanism to stabilize the coexistence of duplicated genes, as it will prevent their elimination by unequal crossing-over or replication slippage events.

What could cause inversions that pivot around the origin and terminus of the genome to occur more frequently than other inversions? One possibility is that many inversions occur, but there is selection against those that change the distance of a gene from the origin or terminus. Such a possibility has been suggested by experimental work in E. coli [14,15]. Additional studies have, however, suggested that there is little selective difference between inversions and that instead there may be certain regions that are more prone to inversion than others [16,17,18,22,23]. Alternatively, the inversion events could be linked to replication, as has been suggested for small local inversion events [24]. Whatever the mechanisms, the fact that we find evidence for such inversions between many pairs of species suggests that they are a common feature of bacterial evolution. Many aspects of the X-alignments require further exploration. For example, to split a tandem duplication, an inversion must fall precisely on the boundary between two duplicated genes. This would appear to be unlikely, requiring a large number of inversions in order to generate a sufficient number of split gene pairs. If the mechanisms of gene duplication are somehow related to the mechanisms of inversion, however, then this model is more plausible. The process of duplicating a gene, if it occurs during replication, might promote a recombination event within the bacterial chromosome that inverts the sequence from the origin up to that point. As with inversion events, recombination and replication have been found to be tightly coupled [25].

Complete published genome sequences were obtained from the National Center for Biotechnology Information website [26] or from the TIGR Comprehensive Microbial Resource [27]. These included Aeropyrum pernix [28], Aquifex aeolicus [29], Archaeoglobus fulgidus [30], Bacillus subtilis [31], Borrelia burgdorferi [32], Campylobacter jejuni [33], Chlamydia pneumoniae AR39 [19], Chlamydia pneumoniae CWL029 [34], Chlamydia trachomatis (D/UW-3/Cx) [35], Chlamydia trachomatis MoPn [19], Deinococcus radiodurans [36], Escherichia coli [5], Haemophilus influenzae [37], Helicobacter pylori [38], Helicobacter pylori J99 [39], Methanobacterium thermoautotrophicum [40], Methanococcus jannaschii [41], Mycobacterium tuberculosis [8], Mycoplasma genitalium [42], Mycoplasma pneumoniae [43], Neisseria meningitidis MC58 [20], Neisseria meningitidis serogroup A strain Z2491 [44], Pyrococcus horikoshii [45], Rickettsia prowazekii [46], Synechocystis sp. [47], Thermotoga maritima [48], Treponema pallidum [49], and Vibrio cholerae [4]. In addition, a few unpublished genomes were analyzed: Streptococcus pyogenes (obtained from the Oklahoma University Genome Center website [7]), Streptococcus pneumoniae (H. Tettelin, personal communication), and Mycobacterium leprae (obtained from the Sanger Centre Pathogen Sequencing Group website [9]).

DNA alignments of the complete genomic sequences of all bacteria used in this study were accomplished with the MUMmer program [6]. This program uses an efficient suffix tree construction algorithm to rapidly compute alignments of entire genomes. The algorithm identifies all exact matches of nucleotide subsequences that are contained in both input sequences; these exact matches must be longer than a specified minimum length, which was set to 20 base pairs for this comparison. To search for genome-scale alignments within species, complete bacterial and archaeal genomes (25 in total including all published genomes) were aligned with their own reverse complements. To search for between-species alignments, all genomes were aligned against all others in both orientations.

The predicted proteome of each complete genome sequence (all predicted proteins in the genome) was compared to the proteomes of all complete genome sequences (including itself) using the fasta3 program [50]. Matches with an expected score (e-value) of 10^-5 or less were considered significant.

To calculate the statistical significance of the X-alignments, the maximal unique matching subsequences (MUMs) for unrelated genomes were examined and found to be uniformly distributed [6]. With a uniform background, the expected density of MUMs in any region of an alignment plot is a simple proportion of the area of that region to the entire plot. In particular, in an alignment with N total MUMs, the probability (Pr) of observing at least m matches in a region with area p can be computed using the binomial distribution in Equation 1:

The origins of replication for the bacterial genomes have been characterized by a variety of methods. For E. coli, M. tuberculosis and M. leprae, the origins have been well-characterized by laboratory studies [51,52]. The origins and termini of C. trachomatis, C. pneumoniae and V. cholerae were identified by GC-skew [53] and by characteristic genes in the region of the origin [4,19]. GC-skew uses the function (G-C)/(G+C) computed on 2,000 bp windows across the genome, which exhibits a clear tendency in many bacterial genomes to be positive for the leading strand and negative for the lagging strand. The origin of H. pylori was determined by oligomer skew [54] and confirmed by GC-skew. The origins and termini of S. pneumoniae and S. pyogenes were determined by the authors of the present study using GC-skew analysis and the locations of characteristic genes, particularly the chromosome replication initiator gene dnaA.