Integrase-Mediated Recombination of the veb1 Gene Cassette Encoding an Extended-Spectrum β-Lactamase

The veb1 gene cassette encodes the extended spectrum β-lactamase, VEB-1 that is increasingly isolated from worldwide Gram-negative rods. Veb1 is commonly inserted into the variable region of different class 1 integrons in which it is always associated with a downstream-located aadB gene cassette encoding an aminoglycoside adenylyltransferase. In Pseudomonas aeruginosa, the majority of veb1-containing integrons also carry an insertion sequence, IS1999 that is inserted upstream of the veb1 gene cassette and disrupts the integron specific recombination site, attI1. Investigation of the recombination properties of the sites surrounding veb1 revealed that insertion of IS1999 reduces significantly the recombination frequency of attI1 and that veb1 attC is not efficient for recombination in contrast to aadB attC. Subsequent sequence optimisation of veb1 attC by mutagenesis, into a more consensual attC site resembling aadB attC, successfully improved recombination efficiency. Overall, this work gives some insights into the organisation of veb1-containing integrons. We propose that IS1999 and the nature of veb1 attC stabilize the veb1 gene cassette environment likely by impairing recombination events upstream or downstream of veb1, respectively.


Introduction
Class 1 integrons are increasingly reported as a reservoir for antibiotic resistance genes in Gram-negative rods [1,2]. These structures possess two conserved regions located on each side of a variable region consisting of integrated gene cassettes [1][2][3] ( Figure 1A). The 59 conserved segment (59-CS) classically includes a gene encoding a site-specific recombinase of the DNA integrase family, intI1, the cassette integration site, attI1, and the promoter Pc, which is oriented toward the integration point of the gene cassettes and is responsible for gene cassette expression [2][3][4][5][6]. Class 1 integrons may not always contain the entire 39 conserved segment (39-CS), which typically includes along with an open reading frame of unknown function (orf5), the truncated disinfectant (qacED1) and the sulfonamide (sul1) resistance determinants [1][2][3]. Gene cassettes from the variable region are composed of a gene, usually an antibiotic resistance gene, and a downstream recombination site known as attC site or 59-base element (59-be) [1][2][3]. Gene cassettes are independent units that are most often promoter-less. Consequently, gene expression levels depend on the cassette position in the integron. The further the cassette is relative to Pc, the lower its expression will be [4].
Gene cassettes are non-replicative mobile elements that exist under a free circular or integrated linear form [3,7]. Site-specific recombination leading to gene cassette excision or integration is catalyzed by the integrase IntI1, which recognizes two structurally distinct sites, attI1 and attC [7,8]. The attI1 site is particularly conserved and includes four integrase binding domains ( Figure 1B). A pair of inversely oriented binding sites is located at the core site and two other integrase binding sites in direct repeat (DR1, DR2) are located further upstream [2,9,10]. A full attI1 site containing four integrase binding domains is required for optimal recombination with an attC site [11,12]. The attC sites that are associated with the gene cassettes are more complex and weakly related to each other [2]. They differ greatly in sequence and length but contain two pairs of inversely oriented integrase-binding domains (1L-2L and 2R-1R) [13,14] ( Figure 1C and 1D). Recombination mediated by IntI1 involves recognition of the bottom strand of the attC site [15]. Upon folding into a hairpin structure, single-stranded attC sites present an almost canonical core site consisting of 2L-2R and 1L-1R duplexes separated by a bulged area [2,[16][17][18][19][20]. However attC site recognition and proper interaction with the integrase are not dependent on canonical DNA but on the position of two extrahelical bases that interact with the integrase and originate from symmetrical folding of the bottom strand of attC [19].
Integrase-mediated recombinations between two attC sites, between two attI1 sites, and between an attI1 site and an attC site have been documented, the latter being the most efficient [8,21]. During recombination the crossover point is located between the G and TT of the 7-bp core site motif, GTTRRRY, found at the 39 end of the recombination sites [2,14].
Among the antibiotic resistance genes that are integron-located, the extended spectrum b-lactamase (ESBL) bla VEB-1 gene has been identified in a series of Gram-negative rods that are scattered worldwide [22][23][24][25][26]. At least seven different types of veb1containing integrons were identified based on cassette content [22,23,26]. It is more than likely that the different veb1-containing integrons evolved from a common ancestor, however they have maintained interesting characteristics. In all cases, veb1 is associated with a downstream-located aadB cassette encoding an aminoglycoside acetyltransferase ( Figure 1A) [26]. An insertion sequence (IS1999) is inserted in the majority of the veb1-containing integrons characterized in P. aeruginosa [26,27]. Upon insertion, IS1999 disrupts the integron-specific recombination site, attI1, but provides an outward-directed promoter P out , which increases bla VEB-1 expression in P. aeruginosa (Figure 2A) [28]. Moreover, veb1 has always been reported as the first cassette within the variable region of IS1999-containing integrons carrying different cassette arrays [26,27].
The stability of the veb1 environment was puzzling given the fact that the variable region of an integron is normally in constant evolution since it is subject to gene cassette rearrangement, loss and acquisition [3]. Cointegration assays were performed to determine the relative recombination efficiency of the different recombination sites present in the veb1 vicinity (i.e. attI1, disrupted attI1, veb1 attC and aadB attC). This work revealed that veb1 attC and the disrupted attI1 site of IS1999-containing integrons are not efficient for recombination and consequently might preserve the associations veb1-aadB and IS1999-veb1 from being disrupted, respectively.

Results
The veb1 attC site is not efficient for recombination Integrase-mediated recombination involving integrons located on multicopy plasmids can generate different recombination products including: (i) free circular DNA molecules comprising one or more gene cassettes, resulting from recombination between the gene cassette attC and either attI1 or another attC site from another gene cassette [7]; (ii) cointegrates, resulting from recombination between two copies of the same plasmid. Different cointegrates can be formed depending on the sites that are available for recombination; (iii) gene duplications, which can arise by either insertion of a second gene copy encoded on a previously excised gene cassette or by formation and resolution of cointegrates ( Figure S2). While the abundance of circular intermediates is very low, plasmid cointegrates and gene duplications are predominantly formed during recombination when intI1 is overexpressed [29]. Moreover, under these conditions gene  [26]. (A) veb1-containing integron from P. aeruginosa 14. The 59 and the 39-CSs are underlined. ORFs are shown as boxes with an arrow indicating the orientation of the coding sequence. The promoter P c is indicated by a broken arrow. The black diamond represents attI1 and circles represent attC sites. (B-D) Recombination sites. Sequences related to the 7-bp core site are boxed; their relative orientations are indicated with arrows. The crossover points are marked by vertical arrows. The region derived from the downstream-located cassette at the recombination point is in lower case. The extra-helical bases (EHBs) as defined by Bouvier et al. [18] are marked with an asterisk. (B) attI1: the nucleotides belonging to attI1 are indicated in white on a black background. The experimentally determined strong and weak IntI1-binding sites and the pair of direct repeats, DR1 and DR2, as well as the location of the IS1999 insertion in attI1 are indicated. (C) veb1 attC and (D) aadB attC: 7-bp putative core sites (1L, 2L, 2R and 1R) related to the core site consensus as defined by Stokes et al. [14] are shown. The aadB gene cassette is boxed in grey. doi:10.1371/journal.pone.0051602.g001 cassettes that are unnecessary for bacterial growth are often excised and lost from the variable region [21].
The relative recombination efficiency of the sites present within different veb1-containing constructs was tested (Table 1 and Figure S1). Upon induction of integrase expression, circular DNA molecules (including plasmids, cointegrates and excised genes cassettes) were purified, digested with BspEI and separated according to their size on an agarose gel. The bottom part of the gel containing only small DNA molecules (,2-kb) was analysed and veb1-containing products were detected by hybridization using a bla VEB-1 specific probe ( Figure 3A). Our experimental conditions specifically allow for detection of recombination events that involved an attI1 and an attC site but do not distinguish whether the product is an excised cassette, a gene cassette duplication or a cointegrate ( Figure S2).
Plasmid pAttI.veb contains a full attI1 site, the veb1 cassette and a truncated aadB cassette, while pAttI.veb.aadB contains, in addition to attI1 and veb1, a full-length aadB cassette. While attI1 and veb1 attC sites are theoretically the only sites available for recombination in pAttI.veb, pAttI.veb.aadB offers an additional recombination site, aadB attC. Recombination between attI1 and veb1 attC leading to either veb1 cassette excision, duplication or plasmid cointegration should produce, after BspE1 digestion, a veb1-containing DNA fragment of 1.1-kb. (Figure S2 and S3). However, no such product was observed with pAttI.veb and pAttI.veb.aadB, suggesting that veb1 attC is not efficient for recombination ( Figure 3A lanes 1 and 2). Instead, a 1.6-kb product was clearly detected using circular DNA extracts from E. coli (pAttI.veb.aadB) ( Figure 3A lane 2). Additional experiments confirmed that the 1.6-kb product was integrase-mediated since it was only detected when IntI1 was expressed ( Figure 3A lanes 2  and 9).
A more sensitive approach using PCR amplification was used. Circular DNA molecules were purified, digested and separated on an agarose gel as before but agarose gel slices were cut at the expected migration of the 1.1-kb and 1.6-kb DNA fragments. Gel extracted DNAs were then subjected to PCR using outwarddirected bla VEB-1 specific primers. These primers are located on each side of the BspE1 restriction site and specifically allow for amplification of veb1-containing DNA fragments that were linearized (excised gene cassettes) or released upon BspE1 restriction from cointegrates and plasmids with veb1 duplication ( Figure S2 and S3).
PCR amplification using DNA extracted from the 1.1-kb gel slices from E. coli (pAttI.veb) and E. coli (pAttI.veb.aadB) yielded a 0.7-kb PCR product ( Figure 3B lanes 1L and 2L). Sequencing indicated that it corresponded to veb1 and further analysis of the recombinant junction confirmed that recombination occurred precisely between attI1 and veb1 attC ( Figure 1B, 1C and 4).  Similarly, PCR amplifications using the 1.6-kb product from E. coli (pAttI.veb.aadB) yielded a 1.3-kb fragment ( Figure 3B lane 2H). Sequencing revealed that it contained both veb1 and aadB gene cassettes and that recombination occurred precisely between attI1 and aadB attC ( Figure 1B, 1D and 4). The predominant formation of the 1.6-kb product also indicated that aadB remained mostly adjacent to veb1 after recombination.
These results underline the weak activity of veb1 attC for recombination and the stability of the association veb1-aadB. Both veb1 attC and aadB attC can recombine with attI1. However in contrast to veb1 attC, aadB attC is highly efficient for recombination as it is favored over veb1 attC during integrase-mediated recombination.
Plasmids pVeb and pVeb.aadB containing a truncated attI1 site with only the 7-bp core site motif (GTTAGCG) at the junction with the veb1 gene cassette were tested for recombination ( Figure 1B and S1). Surprisingly, recombination products were detected in circular DNA extracts from E. coli (pVeb.aadB) albeit at low levels ( Figure 3A lane 8 and 3B lane 8H). Sequencing revealed that recombination occurred between aadB attC and a secondary site (1a) instead of the expected GTTAGCG motif of the veb1 gene cassette ( Figure 4). This secondary recombination site (GTTAAGT) is homologous to a consensus core site GTTRRRY and is located 32-bp downstream of the translation initiation codon of bla VEB-1 . Thus, recombination resulted in a truncated veb1 gene cassette associated to aadB ( Figure 4). Despite several attempts, we were not able to detect by PCR any product containing the truncated veb1 cassette alone ( Figure 3B lanes 7L and 8L). As expected, the 7-bp core site is not sufficient to support precise recombination with attC.

Sequence optimisation of veb1 attC improves recombination
The aadB attC site fits closely to the consensus sequence of an attC site [2,14]. It is 60-bp long and made of two nearly perfect inverted repeats, which are bounded by sequences matching precisely the consensus RYYYAAC/GTTRRRY ( Figure 1D). Moreover, the bottom strand of aadB attC contains the two extrahelical bases (T-N 6 -G) found in the most easily excisable attC sites [30] ( Figure 5). Therefore it is not surprising that aadB attC worked efficiently in our recombination assays. Folding of the bottom strand of veb1 attC revealed a characteristic secondary structure with the three structural elements common to attC sites [2] ( Figure 5). However, veb1 attC presents two striking differences as compared to aadB attC: (i) the veb1 attC site (133-bp) has a longer variable terminal structure (VTS) and (ii) the two putative integrase binding sites 2L and 2R diverge from the attC site consensus sequence ( Figure 1C). The DNA segment from veb1 attC that is located between the 1L and 1R binding sites was replaced with the one from aadB attC, thus reducing the size and restoring the 2L and 2R consensus binding sites in veb1 attC (veb1 attC*) ( Figure 5D). In this configuration, ca. 1.1-kb products were strongly detected with pAttI.veb* or pAttI.veb*.aadB ( Figure 3A lane 5 and 6), which indicate that recombination between attI1 and veb1 attC* occurred. Moreover, in cells harboring pAttI.veb*.aadB, veb1and veb1-aadBcontaining recombination products were detected at similar levels ( Figure 3A lane 6) suggesting that veb attC* and aadB attC have a similar recombination efficiency. Sequencing confirmed that precise recombination had taken place and involved either the veb1 attC* site or the aadB attC site (Figure 4).
These results indicate that the sequence located between the 1L and 1R sites from veb attC is not optimal for recombination. The veb1 attC site was further modified to determine whether the long intermediate region located between the 2L and 2R integrase binding sites or the 2L and 2R sequences that diverged from an attC consensus sequence was responsible for the low recombination efficiency of veb attC. Two different veb1 attC sites named veb1 attC D and veb1 attC' were generated. The DNA segment located between the 2L and 2R binding sites from veb1 attC was reduced to 20-bp as found in aadB attC giving rise to veb1 attC D ( Figure 5E). In veb1 attC', the 2L and 2R sequences were modified to match the 2L and 2R sequences found in aadB attC ( Figure 5F).
Plasmids pAttI.veb D and pAttI.veb' were tested for recombination ( Figure 6). In contrast to veb1 attC*, veb1 attC D and veb1 attC' did not allow for the detection of veb1-containing recombination products by hybridization ( Figure 6A lanes 1-2 and 4). However, veb1-containing recombination products were detected after PCR amplification indicating that veb1 attC D and veb1 attC' were functional ( Figure 6B lanes 1-2). Overall, veb1 attC D , veb1 attC' and wild type veb1 attC had similar activities indicating that modifications made to reduce the VTS or change the 2L and 2R sequences did not improve the recombination efficiency.

IS1999 insertion decreases attI1 recombination efficiency
In P. aeruginosa, the attI1 site of many veb1-containing integrons is disrupted by IS1999 in such a way that only the last 34-bp of the site (containing three integrase binding sites) remain adjacent to veb1 ( Figure 2B). Using circular DNA isolated from E. coli (pAttI.IS.veb.aadB), veb1-containing recombination products that resulted from recombination between the disrupted attI1 and aadB attC were detected by hybridization ( Figure 3A lane 4). Recombination products involving the disrupted attI1 and veb1 attC were only detected after PCR amplification ( Figure 3B lanes 3L and  4L).
These results suggest that the attI1 site disrupted by IS1999 insertion is still functional for recombination. Nevertheless, based on signal intensity, the recombination efficiency of the disrupted attI1 site seemed significantly lowered in comparison to a fulllength attI1 site (Figure 3A lanes 2 and 4, 3B lanes 1L and 3L).

Recombination frequencies
Activities of the sites present within different veb1-containing constructs (Table 1 and Figure S1) were assayed in vivo by measuring the frequency of recovery of cointegrates formed between the test plasmids and plasmid R388. The self-conjugative plasmid R388 (TmpR, Tra+) includes an integron (In3) that contains the dfrB2 cassette conferring resistance to trimethoprim (Tmp) followed by an open reading frame, orfA, of unknown function.
Veb1-containing constructs (conferring resistance to ceftazidime, Caz) were introduced into a rifampin sensitive E. coli containing the integrase expressing plasmid and plasmid R388. Upon induction of integrase expression, cointegrates resulting from recombination between the plasmid-located veb-1 integrons and the recipient integron In3 located on plasmid R388 were predominantly formed. Plasmid R388 and cointegrates were transferred by conjugation into E. coli DH10B-Rif (rifampin resistant) and cointegration frequencies were measured as ratios of cointegrates (CazR-RifR) to total R388 transconjugants (TmpR-RifR) ( Table 2). Plasmid cointegration was IntI1 mediated since CazR-RifR cointegrates were only recovered from strains over-expressing the integrase. In plasmid pVeb, veb1 attC is the only site available for recombination with either attI1 or an attC site from In3. Cointegrates (pVeb::R388) were selected at a very low frequency (4.41610 25 ) close to the limit of detection of our assay, thus reflecting the inefficiency of veb1 attC for recombination. As compared to pVeb, the presence of aadB attC in pVeb.aadB, led to a 50-fold increase in the cointegration frequency, confirming that aadB attC is more efficient for recombination than veb1 attC. Comparison between (pVeb::R388) and (pAttI.IS.veb::R388) revealed that the disrupted attI1 site is 3-times more efficient for recombination than the 7-bp core site from pVeb. However, insertion of IS1999 into attI1 (pAttI.IS.veb) led to a 6-fold decrease in the cointegration frequency, as compared to plasmid pAttI.veb. Sequence optimisation of veb1 attC into veb1 attC* led to an almost 20-fold increase in the cointegration frequency of pAttI.veb* as compared to pAttI.veb. Similar cointegration frequencies were obtained with plasmids pVeb.aadB, pAttI.veb.aadB, pAttI.IS.veb.aadB and pAttI.veb*.aadB. All these constructs contain the highly efficient recombination site aadB attC that was likely mainly involved in cointegration.
CazR-RifR transconjugants were further screened for tetracycline susceptibility to discriminate cointegrates from precise veb1 insertion into In3. In contrast to cointegration, which results from one recombination event; cassette insertion results from two recombination events (cassette excision-integration or plasmid cointegration-resolution) and occurs at a lower frequency. Despite most colonies (.95%) were resistant to tetracycline, few CazR-RifR-TetS transconjugants were identified. CazR-RifR-TetS transconjugants were recovered only from donor cells containing pAttI.veb.aadB, pAttI.veb* and pAttI.veb*.aadB plasmids with similar frequencies ( Table 2). Each of these constructs has the particularity to carry at least two efficient sites that were used during cassette insertion. More than 95% of the CazR-RifR-TetS transconjugants recovered from cells containing pAttI.veb.aadB had integrated veb1 along with aadB, demonstrating that veb1 is comobilized with aadB. Using plasmid pAttI.veb*.aadB, both veb1 and veb1-aadB insertions were found, however veb1 inserts were mostly recovered indicating that veb1 recombines efficiently and independently of aadB when it contains veb1 attC*.

Discussion
Recombination activities of the sites surrounding the veb1 gene cassette were investigated by using two independent recombina-  tion assays. It is noteworthy that experiments were performed in E. coli DH10B, which contains an inactivated form of RecA (RecA1) excluding recombination between homologous sequences. Thus, despite plasmids share some sequence identity, cointegration is unlikely to have occurred by homologous recombination but is truly the result of site-specific recombination mediated by the integrase.
Dissemination and acquisition of a single gene cassette within the variable region of integrons is the result of integrase-mediated recombination between two different sites, one of which is the gene cassette attC [7]. We demonstrate here that veb1 attC found within the veb1 gene cassette, is not efficient for recombination implying that veb1 alone is not highly mobilizable. Our cointegration experiments also revealed that aadB attC is favored over veb1 attC as it is almost exclusively involved in recombination when both gene cassettes are present on a plasmid. Despite no selection, aadB remained associated to wild type veb1 in all the cointegrates tested, while gene cassettes that are unnecessary for bacterial growth, are often unstable and lost from the variable region when IntI1 is overexpressed [21]. Gene cassettes are often considered as independent units but their excision depends greatly on the recombination sites that are flanking them. Therefore, it was anticipated that aadB would be rarely excised since aadB excision require the involvement of veb1 attC. Stability of the association veb1-aadB within integrons is likely preserved by the nature of veb1 attC, which impairs not only aadB excision but also new insertions of gene cassettes in between them. Our cassette integration assay in plasmid R388 showed that the association veb1-aadB is to the benefit of veb1, which is mobilized along with aadB thanks to the high recombination efficiency of aadB attC.
The aadB attC site is 50 times more efficient for recombination than veb1 attC. Sequence optimisation of veb1 attC into a more consensual attC site resembling aadB attC (veb1 attC*) improved drastically the recombination efficiency and allowed independent excision of aadB from veb1-containing integrons. This experiment revealed also that the features responsible for the weak activity of veb1 attC reside within the DNA sequence located between the 1L and 1R integrase binding sites. Two major differences between veb1 attC and aadB attC were found in this region and concerned: (i) the 2L and 2R integrase binding sites, which sequences in veb1 attC diverged from the consensus and (ii) the length of the variable terminal structure (VTS). However, modification of one or the other in veb1 attC was not sufficient to improve recombination activity. While VTS seem to have a minor role in the attC recombination efficiency, other structural features including the unpaired central spacer (UCS) bulge shape and two extrahelical bases (EHBs) appear to be important for recognition and recombination rate [18]. In contrast to aadB attC, veb1 attC presents slightly larger UCS. Also, while the EHBs in aadB attC are 6 nucleotides apart (T-N 6 -G) as found in the attC sites flanking cassettes that are efficiently excised by IntI1 [30], veb1 attC contains only one EHB (G). It has been recently demonstrated that the identity and spacing of the EHBs in the attC sites has a pronounced effect on the efficiency of cassette excision [30]. The relevance of these structural differences in the activity of veb1 attC remains to be established. It is also possible that the recombination efficiency of veb1 attC is not affected by only one of these features but by several of them.
Several studies have shown that a full attI1 site containing the four integrase binding sites was required for high efficiency recombination with attC and that progressive 59 deletions reaching closer to the 7-bp core site of attI1 leads to decreased recombination rates [11,12,31]. Insertion of IS1999 into attI1 displaces only the weak integrase binding site located in DR2. Accordingly, the disrupted attI1 is still sufficient to support recombination albeit at a lower frequency than a full attI1 site. Analysis of the bases from IS1999 replacing part of attI1 did not reveal any obvious 7-bp core site sequence, which could have compensated for the loss of the fourth integrase binding site. Also, by reducing the recombination rate of attI1, IS1999 likely increases the stability of veb1 and consequently aadB at the first positions within the variable region by impairing integration of new gene cassettes upstream of veb1 and excision of veb1-aadB. However, since the disrupted attI1 still supports recombination, new gene cassettes could be inserted at the first position and benefit from the additional promoter P out located in IS1999 for their expression [28].
Several studies have demonstrated that integrase is also able to catalyse recombination between one specific site (attI1 or attC) and non-specific secondary sites conforming to the consensus GNT at a very low frequency [12,21,31]. Moreover, limitation of attI1 to the 7-bp core site is insufficient for determining recombination specificity alone and leads to the formation of cointegrates owing to recombination with secondary sites [12]. Using plasmid pVeb.aadB, we demonstrated upon PCR amplification that aadB attC is able to recombine with the secondary site (GTTAAGT) located within the bla VEB-1 gene. This event resulting in a truncated veb1 gene cassette is extremely rare since it has been detected only once in our repeats. The region that is truncated in bla VEB-1 encodes the first eleven amino acids of the pre-enzyme VEB-1 and includes the translation initiation codon and most of the signal peptide. Even upon insertion in an environment providing transcription and translation signals, it is unlikely that the protein encoded by the truncated veb1 cassette would be functional in vivo since it would not be properly targeted to the periplasm.
Overall, this work gives some insights into the organisation of veb1-containing integrons that are widespread among Gramnegative bacteria. It is more than likely that the different veb1containing integrons evolved from a common ancestor presenting an early association veb1-aadB. It is also possible that aadB is at the origin of the veb1 gene cassette recruitment and of the comobilization of veb1-aadB into class 1 integrons. Even though veb1containing integrons are still subject to gene cassette rearrangements, we propose that IS1999 and the nature of veb1 attC stabilize the veb1 gene cassette environment likely by impairing recombination events upstream or downstream of veb1, respectively.

Bacterial strains, plasmids and culture conditions
Bacterial strains and plasmids used in this study are listed in Table 1. The clinical strains P. aeruginosa 14 and 15 carrying different veb1-containing class 1 integrons were from the Siriraj Hospital, Bangkok, Thailand [26]. The recombination deficient strain E. coli DH10B (Life Technologies, Eragny, France) was used as bacterial host in electroporation experiments. The conjugative plasmid R388 (TmpR, Tra+) includes an integron (In3) that contains the dfrB2 cassette conferring resistance to trimethoprim (Tmp) followed by an open reading frame, orfA, of unknown function [32]. Plasmid p112 (a pTRC99A derivative) was a gift from D. Mazel [33]. This plasmid contains the intI1 gene under the control of an IPTG-inducible synthetic P trc promoter. E. coli DH10B harboring various plasmids and E. coli DH10B-Rif (rifampin resistant) were used for conjugation experiments. The low-copy number cloning vector, pBBR1MCS.3 was used for cloning experiments [34]. Bacterial cells were grown in Trypticase Soy (TS) broth or onto TS agar plates (Sanofi Diagnostics Pasteur, Marnes-La-Coquette, France) with antibiotics when needed.

Nucleic acid extractions
Circular DNA molecules were extracted using Plasmid Maxi Kits (Qiagen, Courtaboeuf, France) according to the instructions of the manufacturer. Extractions of whole-cell DNAs were done as described elsewhere [35].

PCR experiments
Taq and Pfu DNA polymerases were from Roche Diagnostics (Meylan, France) and Promega Corporation (Madison, Wis.), respectively. PCR experiments [36] were performed using the series of primers listed in Table S1. The PCR products were purified using Qiaquick columns (Qiagen). To determine the insertion sites of the veb1 cassette or derivatives into the In3 integron, plasmids from independent E. coli DH10B-Rif (R388::veb1) transconjugants were extracted and amplifications using combination of primers VEBINV2-TMPB, VEBINV2-ORFAB, VEBINV2-QACEB, and TMPA-VEBINV3 were performed. The aadB cassette was detected by PCR amplification using the AADBF and AADBB primers.
For each construct tested, five cointegrates resulting from cointegration between R388 and veb1-containing plasmids were analyzed. Amplifications using combination of primers T3-59CS, T7 promoter-39CS and VEBINV2-39CS were performed in order to determine which recombination sites were involved in the formation of the cointegrates.

Cloning experiments and electroporation
T4 DNA ligase, and restriction endonucleases were used according to the recommendations of the manufacturer (Amersham Biosciences, Orsay, France). The plasmid p112.Kan was constructed by inserting a HindIII-digested omega fragment (VKm) from plasmid pHP45V-Km [37], made of a kanamycin resistance gene (aph(39)-IIa) flanked by transcriptional and translational termination sequences, into the HindIII site from the multiple cloning site of p112 plasmid (pTRC99A::intI1) [33]. The pTRC99A.Kan plasmid was constructed by removing the 1.2-kb EcoRI-BamHI fragment containing the intI1 gene from p112.Kan, filling in its ends with Pfu DNA polymerase and followed by self ligation. The inserts of the recombinant plasmids pVeb and pVeb.aadB, corresponded to fragments of 1.6-kb (containing the veb1 cassette) and 1.7-kb (containing the veb1 and aadB cassettes) that were amplified with the pairs of primers VEBCASF/AADBB and VEBCASF/39CS, respectively and genomic DNA of P. aeruginosa 14 as template ( Figure S1). The inserts of the recombinant plasmids pAttI.veb and pAttI.veb.aadB corresponded to fragments of 2-kb (containing an entire attI1 site and the veb1 cassette) and 2.1-kb (containing an entire attI1 site and the veb1 and aadB cassettes) that were amplified with the pairs of primers INTIN/AADBB and INTIN/39CS, respectively and genomic DNA of P. aeruginosa 14 as template ( Figure S1). The inserts of the recombinant plasmids pAttI.IS.veb and pAttI.IS.veb.aadB corresponded to fragments of 3.4-kb (containing a disrupted attI1 site, IS1999, and the veb1 cassette) and 3.5-kb (containing a disrupted attI1 site, IS1999, and the veb1 and aadB cassettes) that were amplified with the pairs of primers INTIN/AADBB and INTIN/ 39CS, respectively and genomic DNA of P. aeruginosa 15 as template ( Figure S1). PCR products were purified prior to cloning into the SmaI-restricted pBBR1MCS.3 vector.
Three different modifications of the veb1 attC site (veb1 attC*, veb1 attC D , and veb1 attC') were generated. Creation of the veb1 attC* site was performed by using the attCVEB1 and attCVEB2 primers that anneal to the beginning of the veb1 attC site up to the inverse core site and to the beginning of the aadB gene cassette (core site), respectively. These primers have floating 59 ends, each corresponding to a half of the aadB attC site, and have in common the portion containing the BsaHI restriction site ( Figure S1). The recombinant plasmid pAttI.veb* containing veb1 attC* site was constructed as follows: a 1.5-kb fragment amplified with the pair of primers T7 promoter-attCVEB1 and pAttI.veb as template was digested with SpeI-BsaHI. A 0.7-kb fragment was amplified with the primers attCVEB2-T3 and pAttI.veb as template and was digested with BsaHI-PstI. Then, the digested inserts were purified and mixed in a three-way ligation with the PstI-SpeI-restricted pBBR1MCS.3 vector generating pAttI.veb* ( Figure S1). The pAttI.veb.aadB and pAttI.veb* recombinant plasmids were digested with SacI endonucleases. The SacI insert of pAttI.veb* was cloned into the SacI-restricted pAttI.veb.aadB plasmid generating pAttI.veb*.aadB ( Figure S1).
Similarly, pAttI.veb D (containing veb1 attC D ) was constructed as follows: a 1.5-kb fragment amplified with the pair of primers T7 promoter-Shortattc1 and pAttI.veb as template was digested with SacI. A 0.6-kb fragment was amplified with the primers Shortattc2-T3 and pAttI.veb as template and was digested with KpnI. Then, the digested inserts were purified and mixed in a three-way ligation with the SacI-KpnI-restricted pBBR1MCS.3 vector generating pAttI.veb D ( Figure S1). The recombinant plasmid pAttI.veb' (containing veb1 attC') was constructed as follows: a 1.6-kb fragment amplified with the pair of primers T7 promoter-attc2L and pAttI.veb as template was digested with NdeI. The resulting 0.9-kb and 0.7-kb fragments were purified and subsequently digested with AvaI and SacI, respectively. A 0.3-kb fragment was amplified with the primers attc2R-T3 and pAttI.veb as template and was digested with AvaI-XhoI. Then, the digested inserts were purified and mixed before ligation with the SacI-XhoI-restricted pAttI.veb plasmid generating pAttI.veb' ( Figure S1).

Sequencing
Sequencing was performed on both strands using laboratorydesigned primers on an ABI PRISM 3100 automated sequencer (Applied Biosystems, Les Ullis, France).

Detection of veb1-containing recombination products by hybridization
This assay allows for the detection of veb-1 containing molecules that are the result of a recombination event between attI1 (or disrupted attI1) and an attC site and gives an insight into their recombination efficiency. Total circular DNA content was extracted from E. coli DH10B (p112.Kan) strains harboring recombinant plasmids pVeb, pVeb.aadB, pAttI.veb, pAttI.veb.aadB, pAttI.IS.veb, pAttI.IS.veb.aadB, pAttI.veb*, pAttI.veb*.aadB, pAttI.veb D , and pAttI.veb' and E. coli DH10B (pTRC99A.Kan) harboring pAttI.veb.aadB after IPTG-induction. Five mg of circular DNA extracts were digested in duplicate with 10 units of BspEI, which cleaves at a unique site located in the veb1 cassettes ( Figure S1). Digested DNA samples were then loaded on two agarose gels (25 cm, 0.7%) and electrophoresed at 45 V for 16 h using Tris-borate-EDTA running buffer [36]. The duplicates that were loaded on the second gel were spaced two wells apart.
The first agarose gel was used for hybridization experiments as follows: The bottom section of the gel that contained small BspEI excision products (,2-kb), was cut and transferred onto a N + Hybond nylon membrane (Amersham Biosciences). Southern blot hybridizations [36] were performed under high-stringency conditions using the ECL nonradioactive labeling and detection kit (Amersham Biosciences). The probe consisted of a PCR-generated fragment internal to bla VEB-1 that was amplified using primers VEB1A/VEB1B and whole-cell DNA of P. aeruginosa 14 as template.
The second agarose gel was used to extract BspEI digested veb1containing fragments. Since DNA amounts were very low and could not be detected visually after ethidium bromide staining, DNA location was spotted by superposing the autoradiography film obtained from the first gel after hybridization. Gel slices were cut with a separate disposable scalpel to avoid sample crosscontamination. DNA was extracted using Qiaquick Gel extraction kit (Qiagen) and subjected to PCR amplification using the VEBINV3-VEBINV2 outward-directed primers. These primers are located on each side of the BspE1 restriction site, and thus allowed amplification of the recombinant junction that was created upon recombination ( Figure S2 and S3).
For E. coli DH10B (R388; p112.Kan; pBBR1MCS.3) the cointegration frequency was calculated by dividing the number of tetracycline and rifampin resistant (TetR-RifR) transconjugants by the number of TmpR-RifR transconjugants. Figure S1 Schematic representation of the plasmid constructs used in this study. All constructs were cloned into the multiple cloning site of the pBBR1MCS.3 shuttle-vector represented with a solid line. The coding regions are shown as boxes with an arrow indicating the orientation of their transcription. Dashed lines indicate truncated genes. The black diamond, white, grey, white with black dots, and black with white dots circles represent attI1, veb1 attC, aadB attC, veb1 attC D , and veb1 attC', respectively. The veb1 attC* site is highly similar to the aadB attC site and is also represented by a grey circle. The IS1999 inverted repeats are shown by empty triangles. The broken arrows indicate the P c , P out and P lac promoters. Restriction sites used for cloning are indicated on the pAttI.veb.aadB representation. Small arrows (1 and 2) located on each side of the BspEI restriction site represent the positions of the VEBINV3 and VEBINV2 primers, respectively. (TIF)