Genome features and comparative analysis

The general features of the Bordetella genomes are shown in Table 1 and Figure 1. The genome of B. petrii has a similar size (5,287,950 bp) and coding-capacity (5,034 CDSs) to that of B. bronchiseptica (5,339,179 bp; 5,019 CDSs). The genomes of B. pertussis, B. parapertussis and B. avium are much smaller, showing massive gene loss during host-adaptation. The overall GC content (65.5%) of B. petrii is intermediate between that of B. bronchiseptica (68.08%) and B. avium (61.58%). B. petrii has 125 pseudogenes, less than B. pertussis (358) and B. parapertussis (220), but more than B. bronchiseptica (18) and B. avium (68).

	B. pertussis	B. parapertussis	B. bronchiseptica	B. avium	B. petrii
Size (bp)	4,086,186	4,773,551	5,338,400	3,732,255	5,287,950
GC content (%)	67.72	68.10	68.07	61.58	65.48
Coding sequences	3,816	4,404	5,007	3,417	5,034
Pseudogenes	358	220	18	68	125
Coding density	91.6%	92.2%	92.0%	88.6%	90,6%
Average gene size (bp)	978	987	978	972	957
rRNA operons	3	3	3	3	3
tRNA	51	53	55	61	51
IS481	238	0	0	0	0
IS1001	0	22	0	0	1 (truncated)
IS1002	6	90	0	0	0
IS1663	17	0	0	0	6
IS3-family and others	0	0	0	0	98

The genomes of B. petrii and B. avium show lower similarity and synteny than those of mammalian-adapted Bordetellae, suggesting that they are most distantly related. Reciprocal BLASTP analysis with other sequenced Bordetellae identified coding sequences unique to B. petrii (Figure 2) [4, 5]. B. petrii, B. bronchiseptica and B. avium share 2,049 CDSs, which represent the Bordetella core gene set, which was likely inherited from a common ancestor. B. petrii shares more CDSs with B. bronchiseptica (2,884) than with B. avium (2,229). Interestingly, B. petrii has 1,825 CDSs that are not present in either B. bronchiseptica or B. avium. A large number of these B. petrii-unique CDSs (1,157 CDSs and 188 genes encoding transposases) are located on mobile elements and encode accessory metabolic functions (Table 2).

Mobile genetic elements – a mosaic genome

The most prominent feature that distinguishes B. petrii from other Bordetellae is the presence of seven laterally acquired genomic islands (GI1-GI7). The locations of these GIs on the chromosome correlate with atypical GC-content and GC-bias (Figure 1). These GIs have the characteristic features of mobile elements collectively known as integrative and conjugative elements (ICE). Four of the B. petrii GIs (GI1, GI2, GI3 and GI6) are related to the clc element of Pseudomonas knackmussii B13, which codes for proteins involved in the biodegradation of chloroaromatics [19]. The similarity between these four GIs and the clc element is mostly confined to the conjugal transfer region and varies substantially (42%–98%) between individual genes within this region [see Additional file 2]. However, among these GIs, GI3 is the most similar (78%–98%) to the clc element and also carries all the chloroaromatic degredation genes found in the clc element. Like the clc element in B13, these four GIs are integrated into the 3'-end of tRNA^Gly genes, are flanked by a 15–18 bp direct repeat generated by the duplication of the 3'-end of tRNA, and each encodes an integrase highly similar (88%, 69%, 100% and 52% amino acid identity) to that of the clc element (IntB13). GI1, GI2 and GI3 are located in close proximity within a 510 kb region (Figures 1, 3). The clc-type mobile elements are self transmissible and B. petrii may have obtained them from environmental Pseudomonads, with which it shares habitat [20]. During the conjugal transfer process, ICEs are excised from the chromosome and form a circular intermediate [21]. Interestingly, during in vitro culture the whole 510 kb region encompassing GI1, GI2 and GI3 is deleted in a fraction of the bacterial population. Sequence analysis of the deletion point revealed a regenerated 18 bp integration site (Figure 3), indicating that these GIs are still active in terms of excision from the chromosome. In fact, circular intermediates were detected for all GIs in B. petrii cells during normal growth (Lechner and Gross, unpublished results).

The remaining GIs (GI4, GI5 and GI7) encode a conjugal transfer system similar to the type IV secretion systems associated with plasmids. GI4 is highly similar to Tn4371 of Ralstonia eutropha A5/Cupriavidus oxalaticus, a 55-kb conjugative transposon that carries genes involved in the degradation of biphenyl or 4-chlorobiphenyl [22]. The features of all GIs are summarized in Table 2.

It has been previously reported that the chromosomes of B. pertussis and B. parapertussis carry a large number of IS elements (261 and 112, respectively). These IS elements are responsible for the extensive shuffling and deletions in their genomes [4]. In contrast, B. bronchiseptica and B. avium carry none. B. petrii carries 105 IS elements, most of which belong to the IS3-family. Interestingly, there are six copies of IS1663 which was found in B. pertussis and one truncated copy of IS1001 (Bpet1198) which was found in B. parapertussis. The vast majority (66%) of the B. petrii IS are located within GIs, whereas 34% are randomly dispersed around the chromosome. B. petrii also carries two intact prophages (phage 1, Bpet0942-1008; phage 2, Bpet4384-4432) (Figure 1), and phage-like particles were indeed observed by electron microscopy of the sequenced strain [7]. Phage 1 is inserted at the 3'end of a gene encoding a tRNA^cys, while phage 2 is related to bacteriophage Mu. In addition, there are several regions on the B. petrii genome which may encode prophage remnants (Figure 1). In contrast, B. avium carries three and B. bronchiseptica carries four prophages, however, none of them is related to the B. petrii phages.

The GIs (GI1-GI7), IS elements and prophages represent 22% of the B. petrii genome. This highly mosaic genome structure reflects extensive DNA exchange and provides first evidence for a considerable horizontal gene transfer-driven evolution in the genus Bordetella.

Metabolism of B. petrii – genomic basis for metabolic versatility

The central metabolic pathways are remarkably similar between B. petrii and the pathogenic Bordetellae. For instance, the glycolytic pathway is incomplete, lacking the determinants for glucokinase and phosphofructokinase. B petrii encodes all the enzymes needed for the TCA cycle and for the synthesis of all essential nucleotides, cofactors, amino acids and fatty acids. However, the in silico analysis of the genome sequence of B. petrii suggests the presence of a set of auxiliary pathways for the use of alternative nutrients, such as gluconate, degraded plant products, cyanate (as N-source), and various aromatic compounds (see below), which may contribute to the ability of B. petrii to thrive in different environments (Figure 4).

Gluconate is an intermediate in the degradation of sugar acids and sugar derivatives, which are components of more complex molecules, such as plant materials and secondary metabolites. The genome analysis of B. petrii suggests that this microorganism can also utilize gluconate via a variant of the Entner-Doudoroff (ED) pathway, which is only found in few bacteria, such as Rhodobacter sphaeroides, but absent in all the other sequenced Bordetellae [23]. In this pathway, gluconate is first transformed by a gluconate dehydratase (Bpet3300) to 2-dehydro-3-deoxy-D-gluconate, which is then imported into the cytoplasm via 2-dehydro-3-deoxy-D-gluconate permease (Bpet1985) and phosphorylated to 2-dehydro-3-deoxy-D-gluconate-6-phosphate by a 2-dehydro-3-deoxy-D-gluconate kinase (KdgK, Bpet0876). This compound is then converted to glyceraldehyde 3-phosphate and pyruvate by the action of 2-dehydro-3-deoxy-D-gluconate-6-phosphate aldolase (KdgA, Bpet0875). A IclR family transcription regulator (Bpet0874) is present immediately upstream of the kdgA and kdgK genes suggesting that this variant ED pathway may be inducible by this regulator depending on the availability of the external carbon source. In this regard, it is surprising that although a periplasmic glucose dehydrogenase (gcd, Bpet4644), which converts glucose to gluconate, is encoded in the genome, B. petrii is unable to metabolize glucose [7].

Interestingly, B. petrii encodes a protein (Bpet0241) with significant similarity to pectate lyases, extracellular enzymes secreted by phytopathogens, such as Erwinia carotovora, which catalyze the cleavage of pectate [24–26]. The reaction products are channeled into the variant ED pathway [27]. In agreement with this, the determinants for two other important enzymes of these pathways, 2-deoxy-D-gluconate 3-dehydrogenase (Bpet0046, Bpet1132) and altronate hydrolase (Bpet0414) are present, indicating that B. petrii uses plant cell-wall constituents as nutrient sources, which is in line with the isolation of a B. petrii strain from a plant root consortium [14]. The presence of a cyanate transporter (Bpet3878) and a cyanate lyase (Bpet3621) catalyzing the decomposition of cyanate into ammonia and bicarbonate, suggests that B. petrii is not only able to cope with the toxicity of environmental cyanate, but like E. coli may use it as a nitrogen source [28]. A striking difference between B. petrii and the other Bordetellae is that it has a facultative anaerobic energy metabolism and can utilize nitrate as the terminal electron acceptor during anaerobic respiration [7]. Accordingly, on its core genome B. petrii encodes the factors (nitrate-, nitrite-, and nitrous oxide reductases) necessary to carry out complete denitrification from nitrate to N₂. Interestingly, only some of the respective genes are present in the pathogenic Bordetellae. B. bronchiseptica appears to encode a functional nitrate reductase, while B. parapertussis has a truncated napD gene probably leading to a non-functional nitrate reductase, and B. pertussis does not encode genes involved in denitrification at all.

Metabolic talents of B. petrii – degradation of aromatic compounds

The in silico analysis of the B. petrii genome revealed an unusual metabolic capability regarding the degradation of aromatic compounds summarized in Figure 5. For the metabolisation of some of these compounds including benzoate, phthalate, 3-hydroxybenzoate, 3-hydroxyphenylacetate and 4-hydroxyphenylacetate we have obtained experimental evidence (data not shown).

A prominent feature is the presence of a surprisingly large number of genes coding for enzymes of the chloroaromatic metabolism. One complete set of chlorocatechol pathway genes (Bpet1533-1538) is localized on GI3 highly related to the clc element of Pseudomonas knackmussii B13. A second set of chlorocatechol pathway genes (Bpet3747-3752) is located on GI5 and shows high similarity to genes identified in the 1,2,4-trichlorobenzene degrading Pseudomonas strain P51 [29]. A third possible chlorocatechol gene cluster encoding proteins only distantly related to those previously characterized is also located on GI5. This multiplicity of chlorocatechol gene clusters probably endows this strain with the ability to effectively metabolize a broad range of differently substituted chlorocatechols, providing thereby a valuable competitive advantage to thrive in polluted environments [30–32].

On GI5, chlorocatechol genes are preceded by a gene cluster (Bpet3738-3742) encoding a chlorobenzene dioxygenase and chlorobenzene dihydrodiol dehydrogenase, highly homologous to those of the 1,2,4-trichlorobenzene degraders Cupriavidus strain PS12 and Pseudomonas strain P51 [33, 34]. The absence of any linked extradiol dioxygenase encoding gene suggests that similarly to what is observed in the strains PS12 and P51, this cluster has specifically evolved for the transformation of chlorobenzenes [29, 30, 35].

B. petrii also harbors four different central pathways for aromatic metabolism via dihydroxylated central intermediates, the gentisate (Bpet0516-0517 and Bpet1429-1430, the latter located on GI2), the homogentisate (Bpet 4016–4017) and the catechol and protocatechuate branch of the 3-oxoadipate pathway (see below). Such versatility is similar to that observed in bacteria well known for their high metabolic capabilities, such as P. putida KT2440 and Cupriavidus necator [36]. Only genes encoding enzymes of the homogentisate pathway were detected in all other sequenced Bordetellae, and those involved in the gentisate pathway are only present in B. avium.

Various natural aromatic compounds are known to be degraded via either the catechol or the protocatechuate branch of the 3-oxoadipate pathway, which thus plays a key role in bacterial aromatic catabolism. This pathway is widespread in Proteobacteria, however, all previously sequenced Bordetella strains are devoid of complete pathways and only a catIJCD cluster (Bpet4528-4532) is present as a remnant. In B. petrii, complete pathways are formed by genes of the protocatechuate branch (Bpet0260-0254) located in a genomic region with a quite low GC content, but lacking other typical features of a genomic island (named GI in Table 2 and in Additional file 3), and two clusters localized on GI2, encoding the enzymes of the catechol branch. We believe this to be the first example of natural horizontal transfer of this central aromatic metabolic pathway. Notably, also peripheral pathways for funneling benzoate (Bpet1396-1399) and benzylalcohol (Bpet1387-1388) to catechol are located on GI2. A peripheral pathway for channeling phthalate to protocatechuate (Bpet0248-0252) is located on GI1. The low number of peripheral routes contrasts with the situation in other Proteobacteria, such as B. xenovorans LB400, where five substrates are degraded via protocatechuate [37].

A strategy of aerobic degradation of aromatics alternative to the metabolism via dihydroxylated intermediates is initiated by CoA ligases [38, 39]. B. petrii carries gene clusters required for the metabolism of all three substrates, which are known to be degraded by such a strategy (Bpet0549-0554 for anthranilate metabolism and Bpet1924-1935 for phenylacetate metabolism, which are present in other Bordetellae, and Bpet3568-3575 for benzoate metabolism). As degradation via CoA derivatives needs less oxygen molecules compared to degradation via dihydroxylated intermediates, this may provide B. petrii with a selective advantage to also thrive in oxygen limited environments. For an overview about the aromatic degradation capabilities of B. petrii see Figure 5 and Additional file 3.

"Virulence factors" of B. petrii and host-restricted Bordetellae

Pathogenic Bordetellae produce many surface associated or secreted virulence factors [1]. B. petrii does not code for the well-characterized toxins of classical Bordetellae (i.e., pertussis toxin, adenylate cyclase toxin and dermonecrotic toxin). However, it produces a low amount (22–53 nM/OD) of the non-protein tracheal cytotoxin (TCT), a spontaneously released muramylpeptide that disrupts epithelia [40]. This expression range is comparable to the expression levels observed for the broad host range pathogen B. bronchiseptica.

Autotransporter or type V secretion proteins play a major role in Bordetella virulence [1]. B. bronchiseptica encodes 21 autotransporter proteins, whereas B. petrii encodes only four (Bpet0190, Bpet0193, Bpet3980, Bpet4156) [4]. Two of them (Bpet0190 and Bpet0193) are paralogous and show significant homology to the putative serine protease autotransporter BB2301 of B. bronchiseptica, whereas the other two do not show significant similarity to any database entry. Pathogenic Bordetellae also encode a two-partner secretion system which allows secretion of the colonization factor filamentous hemagglutinin (FhaB) [41]. This protein requires the FhaC protein for translocation across the outer membrane. B. petrii encodes proteins related to FhaB and FhaC (Bpet600-601). B. petrii FhaB has 3039 amino acids and its N-terminus, which is required for export, is well-conserved with respect to pathogenic Bordetellae. However, the overall similarity with FhaB of pathogenic Bordetellae is very low, being more similar to putative secreted proteins of plant pathogens (e.g., Ralstonia, Xylella and Burkholderia). This suggests that the fha genes found in B. petrii and the pathogenic Bordetellae may not be orthologous.

Other protein secretion systems are differentially distributed among Bordetellae. In contrast to the mammalian pathogens, B. petrii and B. avium lack type I and III protein secretion systems, but encode a type II secretion system (Bpet2855-2866 and BAV0331-0343). Type IV secretion systems are either engaged in protein transport to the extracellular medium or into eukaryotic cells, or they are involved in DNA-transfer. There are three gene clusters in B. petrii, which are associated with GIs and appear to encode complete type IV secretion systems (Bpet2204-2215, Bpet3701-3719, Bpet4615-4626). None of these systems appears to be related to the pertussis toxin exporting type IV system of B. pertussis, being probably involved in conjugal DNA transfer.

The B. petrii genome carries three fimbrial gene clusters (Bpet0111-0114, Bpet4129-4132 and Bpet4464-4468), each encoding a major fimbrial subunit, a fimbrial assembly chaperone, an outer membrane usher protein and a fimbrial adhesin. The chaperone and usher proteins have significant sequence similarities with their counterparts in the pathogenic Bordetellae, with e-values ranging from 6e-23 to 3e-35 for the chaperones and e-values ranging from 3e-71 to 6e-103 for the usher proteins. In contrast, the homology of the major structural proteins and fimbrial adhesins with the proteins of the pathogenic Bordetellae is very low (e-values > 1e-5). Similarly to B. avium, B. petrii lacks the genes for the biosynthesis of siderophores. However, it specifies 12 TonB-dependent iron scavenging receptors, which may be important for iron acquisition in soil and plant-associated environments.

Most virulence factors are coordinately regulated in pathogenic Bordetellae by a single master regulator, the BvgAS two-component system [42, 43]. This system consists of two proteins, a response regulator (BvgA) and a multidomain histidine kinase (BvgS), consisting of a periplasmic solute-binding domain, and, on the cytoplasmic side, of a PAS, a transmitter, a receiver and an HPt domain. Although the B. petrii BvgA (Bpet 4471) is highly homologous to the BvgA proteins of the pathogenic Bordetellae (65% amino acid similarity), the structure of the B. petrii equivalents of BvgS is more complex. First of all, there are two histidine kinases, BvgS1 (Bpet4469) and BvgS2 (Bpet4472), with significant sequence similarities to BvgS. BvgS1 contains a PAS, a transmitter and a receiver domain, whereas BvgS2 contains a transmitter and a receiver domain. In contrast to BvgS, both B. petrii proteins are devoid of a large periplasmic domain. The HPt domain, which in pathogenic Bordetellae is part of the multidomain BvgS protein, in B. petrii is an independent protein (Bpet4470) (see Additional file 4. To see legends for Additional files 1, 2 and 4 please see Additional file 5 ). This suggests that during their evolution to pathogens, the virulence regulatory system has been streamlined by the omission of one histidine kinase and incorporation of the HPt domain into the multidomain histidine kinase BvgS.

Therefore, B. petrii is endowed with several putative virulence factors for plant and animal hosts, such as adhesins (e.g., Fha, fimbriae) and a master virulence regulator (Bvg), but lacks Bordetella toxins. This suggests that pathogenic Bordetellae might have acquired their toxin genes horizontally, while adapting to an obligate pathogenic lifestyle. Alternatively, it is possible that toxin genes have been lost from the B. petrii lineage.

Whole genome shotgun sequencing

For the shotgun phase we produced 103,308 paired-end sequences derived from three pTZ18R genomic libraries (with insert sizes of 1 kb, 3 kb and 10 kb) yielding 11.72-fold coverage, using dye terminator chemistry on ABI377 and MegaBACE 1000 automated sequencers. As a scaffold we used 2,688 paired-end sequences pCC1BAC libraries with insert sizes of 17–30 kb and 40–70 kb (leading to a total coverage of 5.14-fold). Assembly of the paired-end sequences was performed by PHRAP (P. Green, unpublished data; http://www.phrap.org/). The complete genome sequence was finished by using the Staden package (GAP v4.8b1) [44]. Another 5,154 sequencing reads were generated to close the remaining gaps and to polish the sequence.

Genome analysis and annotation

Curation and annotation of the genome was done using the genome annotation system GenDB [45]. A combined gene prediction strategy was applied on the assembled sequences using GLIMMER and CRITICA [46]. Putative ribosomal binding sites and transfer RNA genes were identified with RBSFINDER and tRNAscan-SE, respectively [47, 48]. In a first step, automatic annotation was computed based on various tools: similarity searches were performed against different databases including SWISS-PROT, TrEMBL, KEGG, Pfam, TIGRFAM and InterPro. Furthermore, SignalP, helix-turn-helix and TMHMM were applied. Additionally, each gene was functionally classified by assigning of a Cluster of Orthologous Groups (COG) number and its corresponding COG category and Gene Ontology numbers [49, 50]. Finally, the annotation of each predicted gene was manually inspected and adjusted if necessary.

Genomic comparisons

For comparative analyses, the annotated genome sequences of the following bacteria were imported into the genome annotation system GenDB [45]: Bordetella avium 197N (gb;AM167904), Bordetella pertussis Tohama I (gb;NC_002929), Bordetella parapertussis 12822 (gb;NC_002928), Bordetella bronchiseptica RB50 (gb;NC_002927). Similarity searches were conducted against the genomes on the nucleotide and amino acid sequence level by using BLASTN and BLASTP, respectively [51]. Synteny plots were created with GenDB [45]. For the determination of the core genome common to all sequenced Bordetellae all the CDS were compared to each other reciprocally using BLASTP with an e-value < 1e-20 as a cut-off.

Detection of regions with atypical GC content

Genomic regions with atypical GC content were identified using the 'sliding window' technique with a window size of 3,000 bp and a step size of 1,000 bp. For this purpose, the GC content was assumed to follow a Gaussian distribution and regions with at least 2.0 standard deviation differences from the mean were determined.

Use of aromatic compounds as sole carbon source by B. petrii

Growth experiments were carried out in a minimal medium containing 10% salt solution (2.5 g NaCl, 0.5 g KH₂PO₄, 0.2 g KCl, 0.1 g MgCl₂ × H₂O, 1.5 g Tris/HCl resuspended in 100 ml distilled water), 1.8 μM CaCl₂, 36 μM FeSO₄ × H₂O, 2 mM (NH₄)SO₄. Aromatic carbon sources were added to final concentrations of 2–5 mM each.

Determination of TCT production

Tracheal Cytotoxin (TCT) was quantified as previously described [52].

Accession number

The complete, annotated genome sequence has been submitted to EBI under the accession number AM902716.