At the time of analysis seven genomic islands were available for study. Three were individually sequenced, whereas the others were identified from completed bacterial genomic sequences (Table 1). The islands were found among H. influenzae and H. parainfluenzae as follows: ICEHin1056 in H. influenzae 1056, ICEHin299 in H. influenzae 299, ICEHin2866 in H. influenzae 2866, ICEHin028 in H. influenzae 028, ICEHinB in H. influenzae B, ICEHpa8f in H. parainfluenzae 8f, and ICEHpaT3T1 in H. parainfluenzae T3T1 (Table 1). General features of these Haemophilus spp. genomic islands and their gene content are shown in Table 2 and Figure 1, respectively. Among these, the H. influenzae island ICEHin1056 and the H. parainfluenzae island ICEHpaT3T1 are the largest and have the most open reading frames. Only a single copy of the genomic island was identified per chromosome of the respective host strain (data not shown).

All seven genomic islands exhibited extensive DNA sequence homologies and preserved gene order in most cases (Figure 1). The only major difference in gene order is represented by the region containing Tn10, in which the orientation is reversed between ICEHin1056 and ICEHpaT3T1 (Figure 2). Whereas high sequence similarities (some of 100%) were observed, especially in the putative replication and T4SS modules (Figures 1 and 2), the degree of sequence conservation between pairs of genomic islands varied more widely in the putative integration module (right-hand side of Figure 2). For example, we observed 74% to 99% nucleotide sequence similarity for int, encoding the putative integrase.

As shown in Figure 1, a total of 84 open reading frames were identified in the seven genomic islands surveyed. The putative replication module and the putative transfer module (including T4SS) comprised the majority of the 43 genes conserved across all genomic islands. These 43 genes were defined to constitute the essential core set of genes of an ICEHin1056-based subfamily of genomic islands. The remaining 41 genes, including the transposon-based genes, are likely to be accessory genes. A majority of the accessory genes, whether they were part of transposons (transposon [Tn]3 in ICEHin1056 or Tn10 in ICEHpaT3T1) or not, were located 3' of traC and 5' to the putative traI. Many of these accessory genes have unknown functions. This localized site of accessory gene presence suggests that this is a region that is particularly permissive for their insertion. Curiously, this region in the larger family of related proteobacterial genomic islands is also a nidus for accessory gene insertion 1, suggesting that this area is a 'hot spot' for recombination for the entire family. No explanation for this phenomenon is apparent, although this region is just 5' to the putative oriT of ICEHin1056.

The observed distribution between core genes and accessory genes in this subfamily of genomic islands indicates that 50% are accessory. This is remarkably similar to the distribution found in the complete H. influenzae genomes analyzed by Hogg and coworkers 10. Furthermore, excluding the antibiotic resistance transposons, the GC content of the other accessory genes is similar to that of the H. influenzae genome and to that of these genomic islands' core genes. These features suggest that this subfamily of genomic islands conforms to the distributed genome hypothesis in the same way that the host genome does. This hypothesis proposes that the full complement of genes available to a pathogenic bacterial species exists in a supragenome pool, which is not contained by any particular strain. Although individual strains carry only a subset of genes, a diverse range of genes can be available to naturally transformable bacterial strains 10. Evidence for the distributed genome hypothesis has thus far been demonstrated for several bacterial pathogens, including H. influenzae, Streptococcus pneumoniae, and P. aeruginosa 101112.

The divergence in gene content between the seven genomic islands described here is clearly illustrated in Figure 1. No two islands were identical and they differed by many genes. Even the core genes of conserved modules exhibited many variable sites, with the nucleotide sequences between several core genes differing by up to 10% or more (Figure 1). A similar level of variation is seen in the housekeeping genes of the bacterial host H. influenzae 13. These accumulations of mutations must have taken substantial time. Both the divergence in gene content between the genomic islands and the variation in nucleotide sequence of their core genes are contrary to these genomic islands having recently been acquired and disseminated by clonal expansion. Furthermore, the similarity of GC content of the genomic island core genes (38%) to the GC content of the host genomes (Table 2) suggests that this ratio for the island has ameliorated to that of the host bacteria, which is time dependent. All these features are consistent with this subfamily of genomic islands having a longstanding association with its Haemophilus host species.

None of these genomic islands contained nucleotide sequences indicating recent acquisition from other subfamilies present in H. somnus and H. ducreyi, nor from more distantly related members of this family of islands. This observation provides evidence that the ICEHin1056 subfamily is evolving by descent within its host species without recombination with other species-specific subfamilies. However, it is curious that the two genomic islands present in H. parainfluenzae had no features to distinguish them specifically from the five genomic islands present in H. influenzae. From other recent work, this ICEHin1056 subfamily of genomic islands appears to be shared only by these two human Haemophilus spp. 14. The factors that limit this genomic island to two species to the exclusion of the many other closely related human Haemophilus spp. are unclear and await further investigation.

The putative replication module of the genomic islands surveyed is composed of between 15 and 18 genes, mostly of hypothetical function, that share extensive sequence homology and are arranged in the same order. Fourteen of these genes, including homologs of parA, dnaB, ssb, topB, inrR, and radC, were present in all islands tested and would be predicted to be involved in DNA replication. The gene parA encodes the chromosome partitioning protein that divides and distributes plasmid copies upon cell division; dnaB encodes a helicase that unwinds DNA strands; ssb encodes a single-strand protein that binds the lagging DNA strand thus preventing re-annealing; topB encodes a topoisomerase that protects free DNA strands from random ligation; and radC encodes a DNA repair protein involved in the recombination repair associated with a DNA replication fork 151617. One of the other encoded conserved hypothetical proteins exhibited low similarity to the carboxyl-terminal region of Shigella flexneri OSA protein, which is known to suppress the oncogenicity activity in Agrobacterium tumefaciens, but its function in H. influenzae is unclear 18.

The most notable variation in the putative replication module was found in the H. parainfluenzae genomic island ICEHpaT3T1, where the osa homolog, which is present in all of the other islands, was missing and substituted by four ICEHpaT3T1-specific genes. Out of these four genes, two encode hypothetical proteins, whereas the others encode an ATP-binding protein that is highly conserved in variety of bacterial species and a possible filamentous phage CTX-RtsR-like repressor. The fact that ICEHpaT3T1 is missing the OSA homolog and that the four ICEHpaT3T1-specific genes are not present in other islands would suggest that only the remaining 14 genes are essential for the replication of this subfamily of genomic islands.

The putative origin of replication oriV has been identified in the replication module, based on the presence of iteron-like (multiple direct repeats of DNA bases) sequences and multiple indirect repeats in an AT-rich region 19. These features are consistent with the oriVs of the larger family of genomic islands found in a variety of other Proteobacteria, including pKLC102 of P. aeruginosa C 20, SPI-7 of S. enterica serovar Typhi 7, and PAPI of P. aeruginosa PA14 21.

Apart from the putative replication module, the T4SS module was the most conserved part of the ICEHin1056 subfamily of genomic islands, with DNA similarity ranging from 95% to 100% between islands. T4SSs are multi-subunit cell envelope spanning structures that comprise a secretion channel and often a pilus or other surface filament or protein 22. Conjugation systems, assembled by a subfamily of the T4SSs, are usually used in the process of conjugative transfer of DNA from donor to recipient cells 23. In our previous work we showed that this conserved cluster of genes encodes a novel T4SS that is responsible for formation of the conjugative pilus and the resulting conjugative transfer of the H. influenzae genomic island ICEHin1056 2.

Of the Haemophilus spp. strains harboring islands, only those that exhibited the highest conjugation transfer frequencies (ICEHin1056, ICEHin299, and ICEHpaT3T1) were shown to produce the T4SS pilus (Figure 3a-c and Table 3). It is reasonable to assume that this pilus is produced also by other genomic islands surveyed, but its presence is harder to detect because of their lower conjugation transfer frequencies (see below). This hypothesis was tested by an experiment in which different genomic islands were transferred into the same host background (H. influenzae strain Rd). The characterized H. influenzae fimbrial gene cluster (hif) is absent in strain Rd, which has not been demonstrated to produce either fimbriae or type IV pili 2425. However, electron microscopic examination of strain Rd harboring genomic islands ICEHin1056, ICEHin299, ICEHin2866, ICEHpa8f, and ICEHpaT3T1 revealed T4SS pili (Figure 3d-h), thus providing experimental evidence on the activity of T4SS encoded by these genomic islands.

Further experimental proof of the activity of the T4SS encoded by the genomic islands tested was provided by the analysis of the conjugation transfer frequencies. As shown in Table 3, genomic islands were transferred from the donors to a rec^- streptomycin-resistant recipient H. influenzae strain (Rd) by conjugation, albeit with different efficiencies ranging from 10^-5 (ICEHin1056 and ICEHin299) to 10^-9 (ICEHpa8f). Any contribution of natural transformation and homologous recombination to transfer of genomic island-borne genes is remote because the recipient was recombination deficient and previous work did not demonstrate such a process 26. Transconjugants were selected on the plates containing either streptomycin plus ampicillin or streptomycin plus tetracycline, and recipients were selected on plates containing streptomycin. The conjugation frequency of the island ICEHinB was assumed to be lower than could be detected using our method. To test whether the differences in the conjugal transfer frequencies of the ICEHin1056 subfamily of genomic islands were due either to the respective host strain or to the gene content of the island, five genomic islands (ICEHin1056, ICEHin299, ICEHin2866, ICEHpa8f, and ICEHpaT3T1) present in the streptomycin-resistant H. influenzae strain Rd background were each transferred to the same recipient: a nalidixic acid-resistant strain Rd. Although recombination proficient, previous unpublished work does not demonstrate any transfer of genomic island borne genes by transformation and homologous recombination in this setting. Transconjugants were selected on plates containing either nalidixic acid plus ampicillin or nalidixic acid plus tetracycline, and recipients were selected on plates containing nalidixic acid. As shown in Table 3, the conjugal transfer frequencies were almost constant when H. influenzae strain Rd was used as a donor, which is indicative of the different host strains introducing variations in conjugation efficiency in the initial experiment. This observation is consistent with the very high sequence homology of the T4SS module between the genomic islands tested.

The only exception to similar conjugation frequencies in strain Rd was genomic island ICEHpaT3T1, which demonstrated a significantly reduced conjugation efficiency (Table 3). An explanation for the lower frequency of transfer of ICEHpaT3T1 from strain Rd donor, when compared with the others of the ICEHin1056 subfamily, may be found in the sequence diversity of the genetfc20 in ICEHpaT3T1. The tfc20 gene constitutes the only variation in the otherwise homogenous block of genes of the conserved T4SS module. The tfc20 gene encodes a hypothetical protein with unknown function that is conserved in six genomic islands surveyed. As shown in Figure 4, the second half of tfc20 in ICEHpaT3T1 differs substantially from tfc20 identified in the rest of the genomic islands surveyed. This may be a consequence of the insertion and subsequent excision of a transposon into and out of the ancestor of ICEHpaT3T1. Because Tn3 was found to be localized between genes tfc20 and tfc21 in five genomic islands, namely ICEHin299, ICEHin2866, ICEHin028, ICEHinB, and ICEHpa8f, the sequence between genes tfc20 and tfc21 may identify a 'hot spot' for transposon insertion.

The genomic islands tested contain a range of antibiotic resistance genes, which is a crucial feature that contributes both to bacterial adaptation and to its clinical significance. Genes that confer resistance to three different antibiotics, namely ampicillin, tetracycline, and chloramphenicol, were identified. In each case, the antibiotic resistance gene determinants are carried on transposons, with ampicillin on Tn3 and tetracycline and chloramphenicol on Tn10 variants. The homologies of the transposon sequences were investigated in detail.

Tn3 is one of the most extensively studied and best understood of prokaryotic transposable elements. It consists of three genes: tnpA, which encodes a transposase that is essential for all transpositional recombination; tnpR, which encodes a resolvase that acts both as a site-specific recombinational enzyme and as a repressor of tnpA and tnpR expression; and bla, which encodes a β-lactamase precursor 27. Tn3 was identified in six of the genomic islands tested, ICEHpaT3T1 being the exception. The complete copy of Tn3 revealed a gene organization identical to the Escherichia coli derived Tn3 transposon 28 and was present in ICEHin299, ICEHin2866, ICEHin028, ICEHinB, and ICEHpa8f (Figure 5). Correspondingly, the host strains harboring genomic islands ICEHin299, ICEHin2866, ICEHinB, and ICEHpa8f were resistant to ampicillin (Table 4). Moreover, when these islands were transferred into the ampicillin-sensitive H. influenzae strain Rd, they also expressed ampicillin resistance; however, most exhibited much greater resistance than the parent strains (Table 4). A truncated Tn3 present in ICEHin1056 did not contain the transposase tnpA (Figure 5). As shown in Table 4, truncation of Tn3 in ICEHin1056 did not abolish the ampicillin resistance activity of the host strain. The reason for the truncation of Tn3 in ICEHin1056 is unknown, although such modified transposons have frequently been identified in other bacteria. Tn3 is inserted in the same specific site, between genes tfc20 and tfc21, in the five genomic islands that harbor the complete copy. Moreover, the nucleotide sequences of tnpA, tnpR, and bla genes of these genomic islands were identical. ICEHin1056 shares an identical bla gene, but it has a single variable site in its tnpR nucleotide sequence. The Tn3 found in E. coli 28 contained approximately 300 variable sites (1% to 6% difference depending on the gene) compared with those complete Tn3s found in the five Haemophilus spp. genomic islands. In addition the GC content of the Tn3 genes (50%) differed substantially from the core genes of this subfamily of the genomic islands (38%).

Tn10 is a composite bacterial transposon, consisting of indirectly repeating insertion sequences (ISs) IS10-right and IS10-left flanking genes, which are involved in tetracycline resistance 29. Tn10 was identified in only two genomic islands, ICEHin1056 and ICEHpaT3T1. In ICEHin1056 Tn10 was a variant containing additional sequences incorporating IS5-right and IS5-left, which flank a chloramphenicol acetyltransferase gene (cat) and IS30. Correspondingly, ICEHin1056 confers resistance to both chloramphenicol and tetracycline (Table 4). Moreover, ICEHin1056 Tn10 was missing the jemA/gltS homolog, encoding sodium-dependent glutamate permease 30. Absence of the jemA/gltS homolog suggests that ICEHin1056 depends on tetB as the only transport system for tetracycline. This observation may also explain the quantitative difference in the level of tetracycline resistance conferred by ICEHin1056 and jemA/gltS homolog-containing ICEHpaT3T1 (Table 4). Not only did the structure of the two Tn10 structures differ, but so did their sites of insertion and orientation. Furthermore, the GC content of the genes within Tn10 were as follows: 45% for IS10L, 52% for IS5L, 42% for cat, 55% for IS30, 51% for IS5R, 41% for ybdA, 40% for tetR, 43% for tetB, and 45% IS10R. These differed substantially from the GC content of the genomic island core genes (38%).

These data suggest that these genomic islands acquired Tn3 and Tn10 a number of times, both independently and recently. It is not possible to estimate the dates of these events. At the latest it was approximately 1972 when antibiotic resistance was first detected in H. influenzae. Additional lines of evidence indicating recent acquisition are as follows: first, the lack of sequence diversity between complete Tn3 sequences found in five genomic islands tested; and second, the different GC content compared with that for the genomic island core genes and for the host chromosome, indicating a lack of time dependent amelioration. Although only two examples of Tn10 were available for study, their distinct and higher GC content is also consistent with relatively recent acquisition. The lack of sequence diversity among the complete Tn3 sequences and their identical point of insertion are curious, because they are from bacterial isolates obtained from points separated markedly in time and geography (Table 1). This may suggest that the same variant of Tn3 is being repeatedly acquired from an independent source and inserting at a 'hot spot' in this subfamily of genomic islands. Alternatively, it may be that Tn3 was acquired as a rare event and has disseminated rapidly worldwide by conjugative transfer of the host genomic island. We are currently actively investigating this possibility. Whichever interpretation is confirmed, it can be inferred that the ICEHin1056 subfamily of genomic islands are small self-perpetuating genomes that are dependent both on a set of highly conserved core genes and on their host bacterial species. However, this subfamily is also the nidus for the accumulation of antibiotic resistance genes, making these islands vectors for the dissemination of antibiotic resistance and thus promoting survival of the bacterial host species in the presence of antibiotic selective pressures.

All bacterial strains and genomic islands used in this study are listed in Table 1. Unless stated otherwise, H. influenzae and H. parainfluenzae strains were grown on heart infusion broth (HIB) medium (Columbia agar containing nicotinamide adenine dinucleotide [15 μg/ml] and hemin [15 μg/ml]). When required, the media was supplemented with kanamycin (10 μg/ml), tetracycline (2 μg/ml), ampicillin (4 μg/ml), streptomycin (20 μg/ml), or nalidixic acid (10 μg/ml). All Haemophilus spp. plate cultures were grown for 24 to 48 hours at 37°C in an atmosphere containing 5% carbon dioxide. For liquid culture, Haemophilus spp. were grown in brain heart infusion broth (BHI) supplemented with nicotinamide adenine dinucleotide (10 μg/ml), hemin (15 μg/ml) and, where necessary, with antibiotics in the concentrations described above and incubated at 200 rpm on a rotatory shaker at 37°C.

Minimum inhibitory concentrations of ampicillin, chloramphenicol, and tetracycline were determined using E-test. The E-tests were performed using E-strips (AB Biodisk, Solna, Sweden), following the manufacturer's instructions and interpreted according to the manufacturer's guidelines.

The transfer efficiency of each genomic island was determined following a modification of the method proposed by Stuy 31. Donor or recipient cells were grown for 48 hours on HIB agar and approximately 10⁸ cells were scraped off the plate and re-suspended in 1 ml BHI broth. Ten microliters of the suspension of donor cells and 100 μl of the suspension of recipient cells were gently mixed, achieving a ratio of donor to recipient cells of approximately 1:10. This mixture was subsequently spread in the centre of antibiotic-free HIB agar plates, allowed to dry, and then incubated for 6 hours. The cells were then harvested by flooding the plate with 1 ml BHI broth and re-suspending using a spreading paddle. Serial dilutions were plated to determine the viable counts and the number of transconjugants, donors, and recipients using agar containing the appropriate selective antibiotic. In the first step (transfer of genomic islands from original host strains to a streptomycin-resistant recipient rec^-H. influenzae strain Rd, using a method previously described 26), transconjugants were selected on the plates containing either streptomycin plus ampicillin or streptomycin plus tetracycline, and recipients were selected on plates containing streptomycin. In the second step (transfer of genomic islands from streptomycin-resistant donor H. influenzae strain Rd to a nalidixic acid-resistant recipient H. influenzae strain Rd), transconjugants were selected on plates containing either nalidixic acid plus ampicillin or nalidixic acid plus tetracycline, and recipients were selected on plates containing nalidixic acid. Experiments were carried out in triplicate, and the mean values for conjugation frequency and its standard error for each strain were calculated. Conjugation frequencies were calculated as number of transconjugants divided by the number of recipients.

To visualize the T4SS-dependent pilus, Haemophilus spp. were grown overnight and then allowed to conjugate on antibiotic-free HIB agar plates for 6 hours. The conjugating cells were harvested by flooding the plate with 1 ml distilled water and subsequently re-suspended using a spreading paddle. A copper grid coated with formvar and carbon was floated in this suspension for 2 minutes. Excess fluid from the copper grid was removed, prior to negative staining with 1% methyl tungstate. The cells were examined for the presence of a pilus by transmission electron microscopy.

H. influenzae strain 299 and H. parainfluenzae strain 8f were sequenced by the method described earlier 1. DNA sequence similarity searches for genomic islands related to ICEHin1056 using the BLASTN and BLASTX algorithms 32, and position-specific iterated BLAST (PSI-BLAST) 33 were performed by interrogating databases available through the National Center for Biotechnology Information website. Haemophilus spp. genomic islands were annotated and visualized with Artemis 34, a sequence viewer and annotation tool that allows visualization of sequence features as well as the results of analyses within the context of the sequence, and its six-frame translation. Complete host genome sequences and sequences of genomic islands were compared using the TBLASTX algorithm of the Artemis comparison tool 34 to identify regions of homology. Comparison of both the nucleotide and translated protein sequences of the seven genomic islands included several different pair-wise combinations of sequences to ensure fair and unbiased comparison. Pair-wise alignment for the percentage of sequence homology between genomic islands tested was carried out using BLAST2 35.