We used BLAST to search for intI genes in organisms whose whole genomes have been partially or completely sequenced, and to query sequences in the nr GenBank database. Although the promoter and attI recombination site are integral parts of both integron structure and function, we chose to broaden our definition and search for just the presence of integron integrase genes because IntI can also catalyze recombination events between the attC site and secondary sites 203435. These events occur at low frequencies, but may be important for bacterial evolution because they result in the insertion of a gene cassette that is flanked by a single recombination site, significantly lowering the possibility of IntI-mediated excision 35. Thus, we use the terms "integron integrase gene" "intI", and "integron" interchangeably.

We identified a total of 103 different bacteria that contain integron integrase genes (the typically mobile integrons were excluded from this analysis but have been reviewed elsewhere 14). We found intI genes in eighteen different bacterial orders within six divisions, including the Bacteroidetes/Chlorobi group, Chloroflexi, Cyanobacteria, Planctomycetes, Proteobacteria and Spirochaetes (Additional File 1), expanding on the diversity recovered from the most recent survey 23. These phylogenetically diverse organisms are found in a variety of both oxic and anoxic environments, and their metabolisms range from heterotrophy to photoautotrophy. The diversity of integron-containing organisms suggests that either these elements are ancient, playing significant roles in shaping microbial genomes over long timescales, or that the evolutionary advantage of being able to catalyze integron-mediated gene transfer has led to the more recent, rapid dispersal of integrons among disparate bacterial lineages.

There are many biases associated with the selection of organisms for both laboratory work and for whole genome sequencing, so the lack of a particular type of organism or group of organisms in Additional File 1 should not be interpreted as evidence for the lack of an integron (with the exception of within specific, entirely sequenced genomes, see below). However, some divisions and subdivisions have been the targets of major sequencing efforts yet are notably absent from this table, including organisms within the Actinobacteria (37 complete genomes sequenced representing 3 orders), and the Firmicutes (103 complete genomes representing 7 orders). It is unknown whether integrons are missing from these lineages all together, or if they are merely absent from the specific organisms selected for sequencing projects. This uncertainty is especially interesting in light of the importance of Gram positive organisms as environmental reservoirs of class 1 integrons 15. In addition, none of the 54 α-proteobacterial genomes, representing 6 orders, contain integrons, yet these elements are found in the γ, β, δ, and ε proteobacterial subdivisions (Additional File 1). To our knowledge, no type 1 integrons have been found in the α-proteobacteria.

Nearly one third of the intI genes that we uncovered are predicted to be non-functional, containing either a stop codon, a frameshift mutation, or a major insertion or deletion that would likely render them inactive (*, Additional File 1). The Vibrios appear to be an exception to this rule, as only 4% of intI genes from this group are pseudogenes. Another interesting feature of integron distribution is that more than one-fifth of all integron-containing lineages harbor more than one intI gene (Additional File 1). This is likely an underestimate of the fraction of organisms with two or more genes: fewer than half of the isolates in Additional File 1 are completely sequenced, and the remainder may contain additional integrase genes in the unsequenced regions. Again, the Vibrios differ from the rest, as only 4% of these species with integrons contain multiple integrase genes.

We mapped the integron phenotype (putative functional genes in green, putative non-functional in purple) onto a 16S rRNA gene tree of all orders of bacteria that contain integrons (again, for simplicity, the typically mobile integrons were excluded). All species reported in Additional File 1 were included in this phylogeny as integron-positive, while all organisms whose genomes are completely sequenced and do not contain intI genes were designated integron-negative. The distribution of integron-positive lineages suggests that integron-containing organisms comprise a significant fraction of the phylogenetic diversity within these bacterial orders (Figure 1). Specifically, 67.1% of the total branch length within the 16S rRNA gene tree leads to modern genomes that contain integrons, indicating that less than a third of the phylogenetic diversity falls only within integron-negative lineages. UniFrac analysis 3637 showed significant phylogenetic clustering (P < 0.001 using unweighted UniFrac significance test and P test through the UniFrac web site), indicating that closely related lineages tend to share integron-positive or integron-negative status more than would be expected by chance.

The current phylogenetic distribution of integrons supports that loss may be an important feature of integron evolution (Figure 1). In some cases this pattern is apparent even within strains of the same species: for example, Shewanella baltica OS195 contains two intI genes, while Shewanella baltica OS155 lacks integrons. Holmes et al. 38 also reported on the spotty nature of integron distribution, finding that two of three isolates of Pseudomonas stutzeri contained intI genes. Although some of these differences may also be due to horizontal transfer of integrons between closely related species 23, the abundance of pseudogenes suggests that decay and subsequent loss are also important processes. While intra-species differences in gene cassette composition is predicted and often observed 31323940 due to the inherent integration/excision activity of integrons, these phylogenetically small-scale differences in the presence or absence of integrons is not an expected feature of a non-mobile gene element. Again, the genus Vibrio is a notable exception to this pattern, as all species examined to date harbor integrons.

We performed extensive phylogenetic analyses using a variety of optimality criteria (maximum likelihood, Bayesian inference, maximum parsimony and minimum evolution) on all intI genes and the predicted protein sequences listed in Additional File 1 as well as on the typically mobile intI genes (Figure 2). For all methods, intI genes formed a supported clade, exclusive of the XerC/XerD outgroup 1741, with a total of fifteen intI clades that were supported by nearly every analysis (Figure 2). Although our tree structure hints at the recently described soil/freshwater and marine groups 17, these clades are not supported by all of our phylogenetic analyses. Each clade of integrase genes primarily originates from organisms within a specific bacterial lineage (clades 1–15, with taxon names in Figure 2; colored boxes in Additional File 1). The vast majority of these intI genes have been identified by sequence similarity only; thus their functions remain unknown. However, the seven intI genes that have been functionally characterized in vivo are widely distributed throughout the intI phylogeny (highlighted red in Figure 2), supporting a similar role for the uncharacterized genes in the integration and excision of gene cassettes 203842434445].

The integrase gene tree suggests some similarities between integron and organismal evolutionary relationships, and in many cases the intI genes found within supported clades derive largely from a related group of bacteria (Figure 2). However, there is some evidence for incongruent intI and organismal phylogenies, and some of these inconsistencies may be due to horizontal gene transfer events. As stated above, many organisms contain multiple intI genes, and in some cases, the intI phylogeny suggests that a horizontal gene transfer event may have given rise to one of the genes (Figure 2, Additional File 1). For example, a gene from each of the deltaproteobacterial species Geobacter metallireducens and Geobacter lovleyi falls just outside of integrase clade 2, which contains genes primarily from within the Betaproteobacteria. These organisms harbor at least two intI genes (G. lovleyi is not completely sequenced yet and therefore may contain more), but the other genes fall into a supported clade with sequences from the Deltaproteobacteria (clade 15, Figure 2). This is also true for Shewanella baltica and Shewanella woodyi, which both contain two intI genes, one which falls in clade 9 and the other which falls in clade 8.

Indeed, intI mobility is a character state that is widely distributed across the integrase phylogenetic tree – found in clades 5, 7, 8, and 9 – supporting that the mobile phenotype has arisen multiple times (Figure 2, circled genes) 23. Additionally, although at first glance, integron integrase genes roughly fall into supported clades reflecting the organisms from which they originated (Figure 2), there are significant differences between integrase and 16S rRNA gene phylogenies 25. We performed likelihood-based tests for congruence 46 on the intI and 16S rRNA gene phylogenies for Deltaproteobacteria (clade 15) and Shewanella (clade 10). In all cases, there were significant differences between the two phylogenies (Figure 3). However, these data are somewhat difficult to interpret because in many cases the relationships among several "housekeeping" genes (e.g., fusA, rpoA, recA, gyrA) were also determined to be significantly different from the 16S rRNA phylogenies, and from one another (Figure 3). Assuming that the 16S rRNA gene acts as the most accurate metric of genomic evolution, several factors could lead to the lack of concordance between 16S rDNA and integrase relationships. The presence of paralogous intI genes, selection, and horizontal gene transfer events could all lead to a lack of agreement between organismal and integrase phylogenies. Paralogous genes are certainly a factor, as many species contain multiple genes, some which appear to have arisen recently within the genome from gene duplication events (Additional File 1, Figure 2). However, even when these events are ignored, and all nodes that are shared between the 16S rRNA gene tree and any paralog are identified, there is a notable lack of concordance. Genetic exchange may also provide a plausible explanation for these patterns, and the association of integrons with transposons and transposase genes provides a possible mechanism for transfer 23. Finally, differential selection on the 16S rRNA gene and the integrase gene could also contribute to a lack of agreement between phylogenetic reconstructions, and is an interesting possibility given the evidence for positive selection on IntI proteins (see below).

As mentioned above, pseudogenes make up more than one third of all integrase genes uncovered in this study. Others have found intI pseudogenes in a number of different bacteria 192439. For example, the typically mobile class 2 integron integrase, intI2, is a non-functional pseudogene, the activity of which can be recovered by suppressing the internal stop codon 19. Additionally, intI pseudogenes were prevalent in a recent survey of integrons from Xanthomonas strains 39. Pseudogenes are also abundant in molecular phylogenetic surveys of intI genes from environmental samples, comprising 4–20% of total genes in these collections (Nemergut and Schmidt, unpublished data; Rodriguez-Minguela et al., unpublished data, accession numbers DQ282376–DQ2822194). In general, pseudogenes are relatively rare features of bacterial genomes, as non-functional genes typically comprise only 2–8% of all genes in free-living bacteria 4748. Therefore, the high percentage of pseudo-integrase genes, or "pseudo-integrons", that we uncovered is surprising. Recent analyses suggest that pseudogenes are likely to be of contemporary origin, as the same pseudogenes are typically not found in the genomes of closely related bacteria 4748. Indeed, orthologs of both functional and non-functional integrases are found in different strains of the same species of Xanthomonas 39. The widespread phylogenetic distribution of pseudointegrons within the bacteria (Additional File 1, Figure 1) implies that decay is a common feature across lineages.

Likewise, the scattered distribution of pseudogenes on the intI phylogenetic tree (Figure 2), suggests that gene decay is a frequent process across different classes of integrase genes. For the most part, these appear to be randomly arrayed on the tree, supporting the hypothesis that pseudogenes decay rapidly and are quickly removed from genomes 4748. However, integrases within clades 4 and 15 are notable exceptions. Interestingly, despite the minimal evolutionary distance among the pseudogenes found in clade 4, these mutations all appear to have originated independently by unique frameshift mutations. This may be an artifact of biased sampling, or it may suggest that strong purifying selection is acting on genes in clade 4, contributing to their rapid decay. In contrast, a single gene fusion event characterizes an entire group of integrase genes within clade 15, and these proteins are an average of ~120 amino acids longer than other IntI proteins. As mentioned previously, pseudogenes are not generally conserved between closely related species 4748 as they are typically lost at rates higher than speciation. Thus, the shared gene fusion event within clade 15 may suggest that these are actually functional genes.

Pseudogenes are relatively common features of some eucaryotic genomes, particularly vertebrates 49, where they arise from either retrotransposition or DNA duplication events. In bacteria, duplication processes give rise to pseudogenes as well; however, they are also thought to develop via the decay of native single-copy genes and following failed horizontal transfers 50. Our phylogenetic analysis suggests that many of the pseudointegrons that we identified appear to have arisen through intragenomic duplication events (a nearly identical functional integrase and pseudogene within the same genome), while others appear to have entered the cell through gene transfer processes (a pseudogene from a different clade in the same genome with a functional gene from the "expected" clade), and still others appear to have been resident genes that are undergoing decay (only one pseudogene from the "expected" clade in the genome) (Figure 2, Additional File 1). In a recent multigenome analysis of bacterial and archaeal pseudogenes, Liu and coworkers revealed that while pseudogenes occur in only approximately 1–5% of total genes, the proportion of integrase pseudogenes are significantly higher, suggesting a different evolutionary dynamic for this class of genes 50. While, in some cases, the ability to integrate and excise foreign DNA may be selectively advantageous, the gain, loss or rearrangment of gene cassettes could also be deleterious, selecting against functional integrases. Thus, such a high fraction of pseudogenes may suggest that the selective impact of integrons on genomes is variable, oscillating between beneficial and deleterious, possibly depending on environmental conditions. Other interpretations are possible as well, and, as noted by one insightful reviewer, the formation of integrase pseudogenes may result when organisms inhabit environments without gene cassettes, or when recombination processes result in the deletion of gene cassettes that maintain the array (e.g., toxin genes 51).

Finally, there is mounting evidence that not all sequences which have been identified as pseudogenes are without a biological function. In some organisms, for example, pseudogenes appear to be important in the generation of variation within multigene families, including both antibody and antigen determinants 5253. These pseudogenes act as a reservoir of sequence diversity and promote the rapid diversification of gene families through intragenomic recombination events. In many of these cases, genes are under "positive selection" pressures, and mutation (i.e., diversification) is favored. A hallmark of genes undergoing positive selection is an increased ratio of non-synonymous (protein altering) to synonymous (non-protein altering) mutations 54 within groups of closely related genes.

Thus, we examined the selection dynamics across integrase proteins (Figure 4A). Heuristic dN/dS (also called K_A/K_S) pairwise comparisons were made between proteins within integrase clade 7 to the functionally characterized intI from Vibrio cholerae. This comparison reveals two excursions in the dN/dS ratios above 1, suggesting mutational hotspots ranging from codon ~1 to ~50 and from codon ~160 to ~220. However, when these regions were further examined under the best-fit evolutionary model 54, only the N-terminal domain showed evidence for positive selection (data not shown). This motif is thought to be important in the recognition of attC and attI sites 45, and it is possible that selection for mutation in this region would expand or change the consensus sequences that an integrase is able to recognize. Alternatively, as pointed out by one reviewer, this site is also involved in the expression of gene cassettes 13; thus, mutation in this region may modulate gene cassette mRNA levels in the host cell. There was a suggestion of a similar pattern for the Shewanella proteins in clade 10; however, no pairwise comparisons were significantly greater than 1 when variable evolutionary rates were considered (Figure 4B).

Finally, we characterized the integron-associated gene cassettes in two species: Methylobacillus flagellatus KT, an obligate methylotroph; and Dechloromonas aromatica RCB, a perchlorate reducing bacterium (Figure 5). These organisms are found in different orders of the subphylum β-proteobacteria and contain intI genes that group with clade 2 (Figure 2). Several of the cassettes that we identified were classified as hypothetical proteins, and pBLAST provided no further clues as to the potential functions of the proteins coded for by these genes. However, several interesting putative functions for gene cassettes were assigned, including a heavy metal (Co, Zn, Cd) pump (CAT) and a heat shock protein (HSP). Additionally, both integron arrays contained genes related to the c-type peptidyl prolyl cis-trans isomerase family, which are involved in protein folding.

The Enterobacter cloacae IntI1 protein sequence (GenBank accession ABO46012) was used to search the nr GenBank database with the PSI-BLAST algorithm and to query the NCBI microbial genome database using the tBLASTn algorithm on April 20, 2007 55. All coding region matches that contained the intI integrase additional domain 56, or intI patch 57 were selected for subsequent analysis, as were the corresponding putative translated sequences. Pseudogenes were identified as those coding regions that were either interrupted by a frameshift mutation or a stop codon, or as proteins that were at least 20% larger or smaller than the intI1 gene. For phylogenetic analyses of pseudogenes, we selected only the translated region that aligned to the IntI1 protein and the corresponding DNA sequence. For pseudogenes that contained multiple mutations, alignable regions of both DNA and proteins were concatenated so that their sequences were contiguous. It is of note that NB231_00025, an intI sequence from Nitrococcus mobilis Nb-231, may not be a pseudogene, as it was found on the end of a shotgun sequence and may be complete within the contiguous genome.

The online version of the sequence alignment program MAFFT 58 was employed to align the amino acid sequences and four outgroup sequences (XerC and XerD from Escherichia coli and Thiobacillus denitrificans 1741) using the following settings: FFT-NS-I, and the BLOSUM45 model. Amino acid alignments were manually adjusted, and any unalignable regions, particularly from within pseudogenes, were trimmed and removed from further analysis. Next, a Python script was developed to guide the insertion of gaps into the corresponding coding DNA sequence for the production of a DNA-based codon alignment (Robeson, unpublished program).

These coding DNA data sets were analyzed using the parallel version of MrBayes 59 and run anywhere from two million to eight million generations in order to achieve convergence under the GTR+I+G model of evolution, as selected via MrModelTest 60. Consensus trees from the analyses were constructed in PAUP*4.0 61 after typically removing ~10% of the burnin trees. Parsimony phylogenetic reconstructions were performed using the PHYLIP 62 software package and subjected to 100 bootstrap replicates with 10 randomizations of taxa input order for each bootstrap. Maximum Likelihood analysis used the GTRGAMMA model and was subjected to 100 boostrapped replicates in RAxML-VI-HPC 63.

We performed likelihood-based tests for congruence 46 on the intI, 16S rRNA, fusA, rpoA, recA, gyrA gene trees for Deltaproteobacteria (clade 15) and Shewanella (clade 10). For each comparison, we downloaded DNA sequences from Genbank for all available phyla, and generated trees using distance, parsimony and likelihood optimality criteria. Likelihood tests were executed in PAUP*4.0 as described previously 25.

For all organisms found to contain integrons, available 16S rRNA gene sequences were obtained from GenBank. We were unable to obtain 16S rRNA gene sequences for all species with integrons; specifically: Alteromonadales bacterium TW-7, Vibrio sp. DAT722 31, Xanthomonas sp. CIP 102397 24, Vibrio cholerae MAK 757, Vibrio cholerae MZO-3, the Xanthomonas DAR strains 39, Shewanella putrefaciens CIP 69.34 24. In addition, because of the lack of availability, several of the 16S rRNA genes that we obtained were not from the same strain that the integron was identified in, specifically: Listonella anguillarum, Listonella pelagia, Vibrio mimicus, Vibrio metschnikovii, Vibrio natriegens, Vibrio salmonicida, and Xanthomonas badrii. All told, we found integrons in eighteen different bacterial orders. 16S rRNA genes from integron containing lineages were aligned with 16S rRNA genes from completely sequenced bacteria from within these same eighteen orders that lacked integrons, as well as with five archaeal 16S rRNA gene outgroup sequences: Pyrobaculum clidifontis, Aeropyrum pernix, soil clone cren34kb, Methanobacterium thermoautotrophicum, and Methanosphaera stadtmanae. We used the online NAST-based algorithm available from the Greengenes website 64 to align the 16S rRNA genes, imported alignments into ARB and exported the alignment using lanemaskPH to remove hypervariable regions 65. MODELTEST 66 was used to estimate the best-fit model of sequence evolution for 16S rRNA gene alignments, and a minimum evolution-based phylogenetic reconstruction was generated in PAUP 4.0. Maximum parsimony and maximum likelihood methods yielded largely similar phylogenic reconstructions but were not shown.

Integrase protein alignments for sequences in clades 7 and 10 were selected for pairwise comparisons to the functionally characterized Vibrio cholerae and Shewanella oneidensis IntI proteins, respectively. Only putative functional integrase proteins that were >95% different from all others and that originated from within the genus Vibrio or Shewanella were analyzed. Amino acid alignments were generated using MAFFT with the parameters described above. Next, a codon-based DNA alignment was constructed using our Python script. DNA alignments were then formatted to compare two sequences at a time, using the Shewanella oneidensis and Vibrio cholerae intI genes as reference sequences. DNA alignments were submitted to the online program SNAP 67 which calculates the change in nonsynonymous (dN) and synonymous (dS) substitutions at each codon position. dN/dS ratios were calculated by integrating over 20 nucleotide positions, and the two-times greater frequency of synonymous sites in a codon was corrected for by dividing this value by 2. dN/dS ratios were plotted against codon position integrated over twenty nucleotide positions. Finally, we used PAML 68 to calculate the dN/dS ratio and likelihood values of a variety of evolutionary models 54.