Genomic islands (GEIs) are large DNA segments, present in most bacterial genomes, that are most likely acquired via horizontal gene transfer. Here, we study the self-transfer system of the integrative and conjugative element ICEclc of Pseudomonas knackmussii B13, which stands model for a larger group of ICE/GEI with syntenic core gene organization. Functional screening revealed that unlike conjugative plasmids and other ICEs ICEclc carries two separate origins of transfer, with different sequence context but containing a similar repeat motif. Conjugation experiments with GFP-labelled ICEclc variants showed that both oriTs are used for transfer and with indistinguishable efficiencies, but that having two oriTs results in an estimated fourfold increase of ICEclc transfer rates in a population compared with having a single oriT. A gene for a relaxase essential for ICEclc transfer was also identified, but in vivo strand exchange assays suggested that the relaxase processes both oriTs in a different manner. This unique dual origin of transfer system might have provided an evolutionary advantage for distribution of ICE, a hypothesis that is supported by the fact that both oriT regions are conserved in several GEIs related to ICEclc.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Recent genome sequencing efforts have revealed that bacterial chromosomes typically contain a number of large DNA regions (> 10 kb) with suspected foreign origin, possibly acquired via horizontal gene transfer (Juhas et al., 2009). Such foreign segments, frequently named genomic islands (GEIs), have been instrumental in shaping the bacterial genome and can cause ‘fast forward evolution’ of the host, because of the large size of the acquired DNA (Dobrindt et al., 2004). GEI may confer several traits to the host with potential adaptive benefits, such as antibiotic resistance (Beaber et al., 2002; Mohd-Zain et al., 2004), iron scavenging (Larbig et al., 2002), plant symbiosis (Sullivan et al., 2002) or aromatic compound degradation (Toussaint et al., 2003; Gaillard et al., 2006). Similar to other mobile genetic elements therefore there is little doubt that GEIs are an important element for bacterial evolution. GEIs do not comprise a single class of mobile DNA, with a unique mechanistically identical mode of mobility and transfer (Juhas et al., 2009). Rather, the term GEI encompasses all sizeable DNA of suspected foreign origin in a bacterial chromosome, some of which have only scars left of past mobility, whereas others are still fully functionally self-transferable. Among GEI one finds prophage-like elements, integrated plasmids, and integrative and conjugative elements (ICEs) also known as conjugative transposons.
Integrative and conjugative elements are particularly interesting model systems, because they have retained full mobility and self-transfer from a host cell to a new recipient via conjugation. The ‘lifestyle’ of ICEs has been inferred from studies on a number of them in diverse bacteria, such as ICESXT in Vibrio cholerae (Beaber et al., 2004), ICEBs1 in Bacillus subtilis (Auchtung et al., 2005), ICEMlSymR7A in Mesorhizobium loti (Ramsay et al., 2009), ICEHin1056 in Haemophilus influenza (Mohd-Zain et al., 2004) and ICEclc in Pseudomonas knackmussii B13 (Minoia et al., 2008; Sentchilo et al., 2009). ICEs, as their name implies, reside in and replicate along with the host bacterial chromosome. Under certain conditions, they can excise from the chromosome and circularize as an extrachromosomal molecule. The circular molecule is supposed to be transferred to recipient cells via conjugation upon which it reintegrates into the recipient chromosome at one or more specific target sites (typically designated as attB sites in analogy to temperate phages). Despite this mechanistically common lifestyle, ICEs come in phylogenetically unrelated families with different gene function and organization, for which many fundamental aspects have remained obscure.
The most well-studied parts of ICE mobilization include the integration/excision reactions. Integration and excision is carried out by a site-specific recombinase (integrase), in addition to other accessory proteins (Burrus and Waldor, 2003; Lesic et al., 2004; Schubert et al., 2004; Qiu et al., 2006; Ramsay et al., 2006; Lee et al., 2007; Malanowska et al., 2007; Wilde et al., 2008). By contrast, ICE transfer mechanisms have received much less attention, and current knowledge is for the most part based on similarities to plasmid conjugative systems. Plasmid conjugation starts with the proteins involved in DNA processing forming a nucleoprotein complex called a relaxosome at the cognate origin of transfer (oriT) (de la Cruz et al., 2010). At that point, the relaxase, a key enzyme of the relaxosome, catalyses site- and strand-specific cleavage at the oriT and covalently binds to the 5′ end of the cleaved single-strand DNA (ssDNA) (Guasch et al., 2003). A complete ssDNA of the whole plasmid molecule is then unwound by DNA helicase activity, and simultaneously the other strand is replicated via rolling-circle mode. The ssDNA terminates at the regenerated oriT sequence by a strand-transfer activity of the relaxase. Finally, the ssDNA-relaxase substrate is released and transferred into recipient cells through a specialized apparatus, for Gram-negative bacteria mostly referred to as type IV secretion system (T4SS) (Fronzes et al., 2009). Indeed, in some described ICEs the transfer genes have significantly similar sequence and organization to those of conjugative plasmids, suggesting the steps in the ICE transfer to be congruent. For example, the ICESXT/R391 family encodes an IncA/C plasmid-like T4SS gene cluster (Wozniak et al., 2009), hypothesized to have emerged from a conjugative plasmid. ICEMlSymR7A contains an IncP-like tra/trb gene module (Sullivan et al., 2002), whereas the ICEBs1/Tn916 family carries transfer genes similar to those of Gram-positive conjugative plasmids (De Boever et al., 2000). However, such clear homology is not recurrent for all ICEs. For example, gene clusters involved in transfer of ICEHin1056 are evolutionarily very distant from all previously described plasmid T4SS genes (Juhas et al., 2007). Also the ICE elements pKLC102 and PAPI-1 from Pseudomonas aeruginosa, and ICEclc from P. knackmussii B13 show little sequence homology to known plasmid conjugative systems (Juhas et al., 2008). As several bacterial genomes, such as Parvibaculum lavamentivorans, Xanthomonas campestris, Burkholderia xenovorans, Ralstonia sp. or Bordetella petri contain GEI belonging to the ICEHin1056/ICEclc family (Minoia et al., 2008), this rather atypical conjugative system may have played a crucial role in their wide distribution. A more precise understanding of the details that make this type of conjugative system evolutionary successful is therefore merited.
The model we deploy here to study the conjugative transfer of this GEI/ICE family (GEI inasfar as integration and conjugation has not been demonstrated in all cases) is the clc element (or ICEclc) (Ravatn et al., 1998a). ICEclc has a size of 103 kb and was first discovered in P. knackmussii strain B13, where it is present in two copies (Ravatn et al., 1998b). ICEclc can transfer itself readily among various β- and γ-proteobacteria such as Pseudomonas putida, P. aeruginosa or Cupriavidus, and confers to its host the capacity to metabolize 3-chlorobenzoate (3CBA) and 2-aminophenol (Gaillard et al., 2008; Sentchilo et al., 2009). The process of ICEclc transfer begins with activation of the promoter of the intB13 gene, encoding a P4-family tyrosine recombinase-type integrase, leading to an excised and circularized ICEclc molecule (Ravatn et al., 1998c; Sentchilo et al., 2003a,b). Integrase expression is dependent on a protein called InrR, which is located at the end of ICEclc opposite to the position of intB13, and is limited to a small subpopulation of cells primarily in stationary phase after growth on 3CBA (Minoia et al., 2008). The circularized ICEclc is supposed to be transferred to a recipient cell through its own conjugation apparatus, and can subsequently integrate into the recipient chromosome. The region of ICEclc proposed to be implicated in self-transfer extends over some ∼50 kb and contains a number of ‘core genes’ defined on the basis of comparison with ICEHin1056 (Juhas et al., 2009). However, homologies to ICEHin1056 and other T4SS are too low to infer mechanistic details. As a first attempt to unravel ICEclc transfer, we explored its reliance on oriT-like sequences and searched evidence for possible relaxosome DNA processing. A functional screening was designed to capture oriT of ICEclc, and demonstrate its functionality in P. knackmussii or P. putida using gene replacement of oriT with egfp. Interestingly, our results indicate that ICEclc bears two separate and different oriTs, both of which are functional and are used in ICEclc transfer with indistinguishable efficiencies. Bioinformatic analysis further corroborated other GEI/ICE of this family to have two oriTs as well. By using gene deletion strategies a relaxase gene was identified on ICEclc, which was essential for ICEclc transfer. Further in vivo strand exchange assays confirmed relaxase action but suggested that it must process the two oriTs in a different manner.
Functional screening for the ICEclc oriT
To isolate the oriT region from ICEclc, four pME6010-based DNA sublibraries, designated pME-libA, -libB, -libC and -libD, were constructed from four cosmids covering the whole ICEclc element (Ravatn et al., 1998a). Each sublibrary contained an average 1 kb insert size and was introduced en masse into P. putida dint1-5, which is a UWC1 derivative strain harbouring one ICEclc copy but with a defective intB13 gene. ICEclc with defective intB13 can not transfer itself from strain dint1-5 (data not shown), but we hypothesized that it might mobilize exogenous plasmids carrying oriT of ICEclc by complementation. For each of the four libraries more than 300 P. putida dint1-5 transformants were pooled and mated with another P. putida recipient, strain UWCGC. This strain is a gentamicin resistant derivative of strain UWC1 tagged with a constitutively expressed mCherry on a mini-Tn7 transposon. Transconjugants that were TcR (because of the incoming pME plasmid), GmR and red fluorescent (indicative for being UWCGC) were obtained at different frequencies (Table 1), depending on which donor library was used and under which conditions the matings were carried out [Luria–Bertani (LB) or mineral medium (MM) with 0.5 mM 3CBA, see Experimental procedures]. No transconjugants were found from matings on LB agar, but two and twelve transconjugants were obtained from matings on 3CBA agar using the pME-libC and pME-libD donor libraries respectively (Table 1). Strain P. putida dint1-5 with pME6010 alone as donor did not produce any transconjugants, demonstrating that pME6010 itself was not mobilized in trans by ICEclc (Table 1). The two transconjugants harbouring plasmids derived from pME-libC displayed overlapping inserts. The region in common between the two inserts could be mapped to the intergenic region between orf50240 and orf52324, and was designated oriT1 (Fig. 1A). Inserts from the 12 mobilizable plasmids derived from the pME-libD pool overlapped in a region that mapped to orf94175 (encoding a putative ssDNA binding protein or ssb) and inrR, and was designated oriT2 (Fig. 1B). Purified and retransformed pME-libC and pME-libD clones obtained from the initial screening, reproducibly transferred from P. putida dint1-5 to UWCGC (Fig. 1).
Table 1. Plasmid mobilization frequences of sublibraries covering ICEclc fragments.
Plasmid mobilization tested in a P. putida UWC1 background carrying one copy of ICEclc in which the intB13 gene was interrupted (P. putida dint1-5). Number of transconjugants expressed as colony-forming units(cfu) per number of cfu of the donor.
ND(< 4.1 × 10−10)
ND(< 1.4 × 10−9)
ND(< 2.5 × 10−8)
ND(< 8.8 × 10−10)
ND(< 1.1 × 10−9)
ND(< 1.1 × 10−9)
ND(< 1.0 × 10−9)
ND(< 9.8 × 10−10)
ND(< 1.0 × 10−9)
ND(< 3.6 × 10−10)
ND(< 7.1 × 10−9)
ND(< 8.3 × 10−9)
ND(< 2.9 × 10−8)
ND(< 4.0 × 10−10)
ND(< 1.5 × 10−9)
ND(< 6.2 × 10−9)
ND(< 7.8 × 10−10)
ND(< 2.0 × 10−10)
2(1.7 × 10−9)
ND(< 1.7 × 10−10)
ND(< 1.2 × 10−9)
ND(< 9.9 × 10−9)
ND(< 2.1 × 10−8)
ND(< 8.7 × 10−10)
ND(< 2.5 × 10−9)
ND(< 1.1 × 10−9)
2(2.0 × 10−7)
ND(< 6.5 × 10−10)
10(2.5 × 10−7)
Definition of the minimal oriT1 and oriT2 region
The minimal functional sizes of oriT1 and oriT2 were determined by examining in trans mobilization from P. putida dint1-5 of pME6010 derivatives carrying fragments of different length around the presumed oriTs. One of two mobilizable plasmids derived from pME-libC, named pMEC9, carried a 798 bp fragment with oriT1 (Fig. 1A). The progressive deletion of oriT1 from 798 to 446 bp did not significantly reduce observed mobilization frequencies. Further deletion of oriT1 to 371 bp resulted in a 10-fold reduction of the mobilization frequency, but shortening to 239 bp (a fragment corresponding to the intergenic region between orf50240 and orf52324) decreased the mobilization frequency essentially to that of the negative control. We thus conclude that the minimal oriT1 region is between 372 and 446 bp, and contains the intergenic region between orf50240 and orf52324 as well as part of the coding sequence of orf50240 (Fig. 1A). Concerning oriT2, one of twelve mobilizable plasmids derived from pME-libD, named pMED7, carried a 1614 bp fragment spanning the entire orf94175 (ssb) and inrR genes (Fig. 1B). Shortening the fragment to exactly the 280 bp intergenic region between orf91884 and ssb (as in plasmid pMETS285) resulted in no detectable mobilization by strain dint1-5. Subsequent deletion to a 1336 bp fragment, then to 677 bp and to 320 bp led to a 10-fold reduction of the mobilization frequency (Fig. 1B). This 320 bp fragment was the shortest fragment tested that maintained oriT functionality. We therefore conclude that the minimal essential region of oriT2 is within this 320 bp, which, surprisingly, lays within the coding sequence of inrR (Fig. 1B).
OriT1 and oriT2 sequences were analysed for specific motifs and compared with other ICEclc-like GEIs. A survey of sequence motifs in both oriTs revealed several direct and inverted repeats (Fig. S1). Eight direct repeats of 6 bp were found in oriT1 (Fig. 2A) and were highly conserved in corresponding regions of related GEIs (Fig. S2). Interestingly, this 6 bp repeat unit was also found in oriT2 although the sequence contexts of oriT1 and oriT2 were totally different (Fig. 2B). The sequences CCCTTT or GCCTTT were found 78 times on ICEclc. On average therefore one would expect to find such a sequence once per 1320 bp. In contrast, seven and four repeats were located in oriT1 (446 bp) and oriT2 (320 bp), respectively, which is thus 20.7-fold (oriT1) or 16.5-fold (oriT2) higher than expected from random distribution. Moreover, no other regions of ICEclc show such a high abundance. As it is known that repeat structures in plasmid oriTs are involved in the specific recognition by the oriT-processing proteins (Lawley and Wilkins, 2004), it might be that the repeats found here for the oriT1 and oriT2 of ICEclc have a similar function.
oriT functionality in Pseudomonas strains
The results described above suggested that ICEclc would have two oriTs, whereas most conjugative plasmids employ one oriT for transfer. To investigate the functionality of oriT1 and oriT2 for ICEclc transfer, we applied the original host strain P. knackmussii B13 that carries two ICEclc copies. Two mutants were then constructed in which either one of the oriTs was displaced by a promoter–egfp fusion. Both strains thus still have two copies of ICEclc, but in one (called B13-3207) oriT1 and in the other (named B13-3217) oriT2 are substituted by egfp on one of the two ICEclcs. Although these mutations are expected to affect the expression of adjacent genes because oriT1 and oriT2 overlap with the coding sequences of orf50240 and inrR, respectively, we expected that the second intact copy of ICEclc in these cells would complement transfer of the GFP-labelled ICEclc copy. Therefore, any transconjugants having ICEclc with egfp substituting oriT1 would be evidence for mobilization using oriT2, and vice versa.
Transconjugant colonies were indeed obtained that could grow on 3CBA and were RifR, which was indicative for successful ICEclc transfer into P. putida UWC1 as recipient (Fig. 3A). Transfer frequencies (as the ratio of colony-forming units of transconjugants per donor) were slightly lower for the egfp-labelled donor strains than for wild-type B13, e.g. 8.2 ± 1.8 × 10−3 in the case of B13-3207, 1.0 ± 0.2 × 10−3 in the case of B13-3217 and 1.8 ± 0.3 × 10−2 for B13 wild-type after 48 h mating time. As observed before, for all strains the calculated ICEclc transfer frequencies increased from 6 to 24, 48 and 72 h mating time (Fig. 3A). This indicated that the ICEclc transfer systems in both oriT-deleted strains worked as in the wild-type. Approximately 30% of all transconjugant colonies obtained from matings with the egfp-labelled donor strains were fluorescent (Fig. 3B), showing that both oriTs can be operated in trans by complementation. For all mating times tested, no significant difference was observed between the proportion of green fluorescent transconjugant colonies for strains in which oriT1 or oriT2 was substituted by egfp, e.g. 24.6% ± 5.7 in the case of B13-3207 and 29.0% ± 3.3 in the case of B13-3217 after 48 h. This demonstrated that both oriT1 and oriT2 are used for ICEclc transfer from P. knackmussii B13 with indistinguishable efficiencies.
To evaluate the effect of host background, we constructed two P. putida UWC1-derived donor strains, which each had one copy of GFP-labelled ICEclc (lacking either oriT1 or oriT2) inserted into the gene for tRNAGly-6, and one copy of an intact ICEclc inserted into the gene for tRNAGly-3. These two strains were named UWC3218 and UWC3220. After 48 h mating with P. putida UWC1G (GmR) as recipient, the transfer frequencies and the proportions of green fluorescent transconjugants were statistically indistinguishable from those when using B13 as donor background (Table 2). This demonstrated that use of either oriTs of ICEclc was donor independent and thus must have come from ICEclc located functions. Interestingly, when an ICEclc was produced that lacked both oriTs (oriT2 displaced with egfp and oriT1 deleted), transfer complementation by a second intact ICEclc copy in UWC1 (named UWC3280) was still observed, albeit at 7% of the total observed ICEclc transfer (frequency 6.7 ± 3.6 × 10−3, of which 7.0% ± 6.1 were green fluorescent, Table 2). The types of ICEclcs present in randomly chosen green transconjugants were characterized by PCR analysis using oriT-specific primers. With UWC3218 as donor (one copy of ICEclc with ΔoriT1-egfp/oriT2+ plus a second intact copy) 44% of green transconjugant colonies tested amplified an intact oriT1 fragment, indicating cotransfer of the second intact ICEclc copy from the donor (Fig. 4). Vice versa, with UWC3220 as donor, 42% of green transconjugants amplified intact oriT2 (Fig. 4). In the case of UWC3280 as a donor (one copy of ICEclc with both oriT1 deleted and oriT2 replaced by egfp, and a second intact ICEclc), still 44% of green transconjugant colonies amplified both an intact oriT1 and oriT2, indicating cotransfer of both the double-mutant and intact ICEclc. Seventeen per cent of the tested green colonies amplified an intact oriT1 but not oriT2 fragment, suggesting that these transconjugants harboured a recombinant ICEclc (oriT1+/ΔoriT2-egfp) that somehow arose during transfer. Thirty-nine per cent of the tested green transconjugant colonies did not amplify either an intact oriT1 or oriT2 fragment. From this we conclude that they harboured only the double-oriT mutant ICEclc copy, which therefore must have been able to transfer despite lacking both oriT1 and oriT2 (Fig. 4).
Table 2. Transfer of GFP-labelled ICEclc from P. knackmussii B13 and P. putida UWC1.a
All matings were performed for 48 h. Values indicate the mean and standard deviations calculated from three independent experiments.
Total transfer frequency of ICEclc expressed as the number of total transconjugants (3CBA+ and RifR or GmR) divided by the number of donors (3CBA+ and green fluorescent).
Proportion of number of green fluorescent transconjugant colonies to the total number of colonies on MM with 3CBA and Rif or Gm.
P. knackmussii B13-3207
8.2 ± 1.8 × 10−3
0.29 ± 0.034
P. knackmussii B13-3217
1.0 ± 1.8 × 10−2
0.30 ± 0.059
P. putida UWC3218
7.6 ± 5.6 × 10−3
0.30 ± 0.064
P. putida UWC3220
6.1 ± 2.7 × 10−3
0.28 ± 0.050
P. putida UWC3280
oriT2, (oriT1 deleted)
6.7 ± 3.6 × 10−3
0.070 ± 0.061
The gene product of orf50240 is required for ICEclc transfer
We explored genes possibly involved in the processing of the ICEclc oriTs and focused on two orfs that had been named orf50240 and orf91884 based on their position on ICEclc (Gaillard et al., 2006). Both orfs are located close to their respective oriT (Fig. 1), and are induced in stationary phase cells grown in MM with 3CBA; conditions, which promote ICEclc transfer (Gaillard et al., 2010). Furthermore, functional prediction suggested these gene products to be the only two candidates encoded on ICEclc displaying DNA cleavage-joining activity. The orf50240 (and orthologues from other ICEclc-like GEIs) is predicted to encode a protein with 27% amino acid identity to a relaxase of GEI in Neisseria gonorrhoeae (Salgado-Pabon et al., 2007), but bears no similarities with known plasmid-encoded relaxases. Orf91884 is a putative DNA topoisomerase III with TOP1Ac domain (cd00186) and shows 39% identity to functional topoisomerase III from Escherichia coli (Lacour et al., 2006). Thus, we constructed two deletion mutants in either of those putative genes, named UWC3012 (Δorf91884) and UWC2989 (Δorf50240), using P. putida strain clc6 that carries only one ICEclc insertion in its chromosome (in the gene for tRNAGly-6). Transconjugant colonies appeared at the same frequency using P. putida UWC3012 or P. putida clc6 (Fig. 5), suggesting that orf91884 is not directly involved in ICEclc transfer. On the other hand, no transconjugants were obtained from P. putida UWC2989 as donor (Fig. 5), showing that orf50240 is essential for ICEclc transfer. The reintroduction of orf50240 under control of the inducible Ptac promoter in single copy on the chromosome of P. putida UWC2989 restored the transfer of ICEclcΔorf50240 but in an IPTG-concentration-dependent manner. These results indicated that the orf50240 gene product is a trans-acting element required for ICEclc transfer. ICEclcΔorf50240 transfer frequencies became maximal at 1 mM IPTG, but were still twofold lower than wild-type ICEclc. This imperfect complementation might be due to the inability of ICEclc to use oriT1 as disruption of orf50240 also resulted in partial disturbance of the oriT1 sequence context.
To further confirm that both oriT regions require Orf50240 for transfer, we compared the transfer frequency of plasmids carrying the cloned oriT1 (pME680-4) or oriT2 fragment (pME285si-1N) from P. putida clc6 (carrying a wild-type ICEclc) or P. putida UWC2989 (ICEclc with deletion in orf50240). Both plasmids could indeed transfer from clc6 (frequency of ∼10−6 per donor) but not from UWC2989 (< 10−9) (Fig. S3), indicating that both oriTs are dependent on Orf50240 activity.
Orf50240 is capable of in vivo strand exchange
To evaluate whether Orf50240 is a relaxase with DNA cleavage-joining activity for oriT sequences, an in vivo strand exchange assay was designed. As shown in Fig. 6A, two oriTs were separately subcloned into two different plasmid replicons that could coexist in the same cell (i.e. pME6010 and pBBR1MCS-5). A PCR was then carried out on cultures carrying those replicons to detect junctions formed between two oriTs as a result of the Orf50240-mediated inter-molecule recombination (Fig. 6A). When E. coli DH5α was transformed with the two replicons pME680-4 and pBBR680-4, both containing oriT1, and with a third coexisting replicon (pBAD50240) carrying orf50240 under the arabinose-inducible PBAD promoter, positive PCR products were obtained corresponding to the two new junctions formed (Fig. 6A and B). The amount of junction PCR products was slightly higher when cells were induced with arabinose, suggesting that more Orf50240 promotes the strand exchange between oriT1s more efficiently. No junction products were detected in E. coli cultures transformed with pBAD50240 plus pME680-4 (oriT1) and pBBR285si-1 (oriT2), or with pBAD50240 plus pME285si-1N (oriT2) and pBBR285si-1N (oriT2). This was independent of arabinose being present or absent in the culture medium (Fig. 6B), and independent of the combination of primers used to detect such junctions (not shown). Furthermore, no specific junctions were formed when E. coli pBAD50240 was transformed with two plasmids each carrying an identical fragment of the clcAB catabolic genes (pMEAB630 and pBBRAB630), independently of addition of arabinose (Fig. 6B). These results showed that Orf50240 is likely a relaxase, the activity of which is sufficient to promote the nicking-rejoining reaction on the oriT1 sequence, but not on oriT2. Interestingly, as shown above (Fig. S3), transfer from oriT2 in P. putida clc6 also requires orf50240. The apparent incompatibility between the two results suggests that other factors on ICEclc in conjunction with the Orf50240 relaxase would be necessary for oriT2 processing.
Nicking site of oriT1 is located within orf50240
Finally, we investigated the nicking site of oriT1 by in vivo strand exchange assay. Six variants of pBBR680-4 containing a point mutation at different positions of oriT1 were separately introduced into E. coli in combination with pME680-4 and pBAD50240. Sequence analysis of PCR products of the hybrid junctions between pBBR680-4 variants and pME680-4 revealed that strand exchange occurred within a 65 bp region between nucleotides 52017 and 52082 on ICEclc, which lays within the coding region of the orf50240 gene (Table 3 and Fig. 2).
Table 3. Sequences of hybrid junctions between wild-type and point-mutated oriT1s.
The numbers are corresponding to the respective ICEclc region (GenBank Acc. No. AJ617740).
Sequences of hybrid junctions were determined from PCR products of in vivo strand exchange assay using two primer pairs; pSV1.F and M13.F, and pSV1.R and M13.R.
Although GEI/ICE are ubiquitous components of bacterial genomes typically being inferred to have been acquired via horizontal gene transfer, many essential details on their actual transfer mechanisms are lacking. In this work we showed how ICEclc, an element representative for a large syntenic group of ICE in α-, β- and γ-proteobacteria, carries two separate and functional oriTs (oriT1 and oriT2). The regions acting as oriT were identified by screening a plasmid sublibrary of ICEclc fragments for mobilization by a P. putida strain containing an excision-deficient ICEclc. In addition, by exchanging each of the oriTs by a constitutively controlled egfp in wild-type ICEclc, we demonstrated that both oriTs are used during transfer at equal frequencies in a population of cells. Such a dual oriT system is rather exceptional, because most conjugative mobile elements previously identified contain only a single oriT used for transfer. OriTs have been identified from various conjugative plasmids and analysed at the molecular level (Zechner et al., 2000; de la Cruz et al., 2010). Typically, oriT sequences are similar among closely related plasmids but different between unrelated groups. This is in contrast to the nic sites, which consist of the specific short sequences that are actually cleaved and rejoined by cognate relaxase proteins. Nic sites are more conserved than the oriT sequences and sequence alignments produce five major groups: IncP/I, IncF, IncQ, ColE1 and pMV158 (Zechner et al., 2000). The presence of the five consensus nic sequences was searched for, but could not be detected in the oriTs of ICEclc (data not shown). In addition to plasmid oriTs, to our knowledge, only three oriTs from ICE have been experimentally identified and characterized. ICESXT has an oriT located in the intergenic region between s003 and mobI (Ceccarelli et al., 2008), and ICEBs1 and Tn916 carry similar oriT regions around the nicK and orf20 genes respectively (Jaworski and Clewell, 1995; Lee and Grossman, 2007). However, no significant similarities were found among oriT sequences from ICESXT, from ICEBs1/Tn916 and the two from ICEclc, probably because the genetic backbones of the elements are too distant (data not shown). Interestingly, whereas both oriTs of ICEclc have a completely different DNA sequence, they were enriched with a similar repeat motif, and this motif is also quite conserved in the corresponding regions of GEI/ICE related to ICEclc (Fig. S2). Therefore, the dual origin of transfer system of ICEclc is different from other conjugation systems but seems to be common in this ICE family.
In terms of transfer frequency, no significant differences between oriT1 and oriT2 were found. This was concluded from in trans mobilization assays using pME6010 derivatives (Fig. 1) and from self-transfer assays using GFP-labelled ICEclcs both in P. knackmussii B13 and P. putida as donors (Fig. 3 and Table 2). On the contrary, the transfer frequencies measured with cloned oriTs on pME6010 in trans in an excision-deficient ICEclc containing donor strain (i.e. pME680-4 with oriT1 or pME285si-1N with oriT2) were 10 000-fold lower than with intact ICEclc (i.e. the ICEclcs in which either oriT1 or oriT2 was displaced by egfp). Also, the transfer frequencies of the larger cloned oriT1 and oriT2 regions in plasmids pMED9 and pME285si-1, respectively, were up to 10-fold higher than the smallest cloned regions (Fig. 1). From these observations we conclude that other cis-acting elements on ICEclc must be necessary to achieve maximal transfer rates. Such cis-acting elements may thus consist of additional sequence motifs, which enhance processing of the oriT or where additional proteins bind that assist the relaxosome formation, e.g. gene products of ssb (putative single-strand binding protein gene) or of orf52324 (putative SpoVT-AbrB superfamily transcriptional regulator gene). Another explanation for the large difference in transfer frequencies of the mobilizable plasmids with oriTs from ICEclc and of ICEclc itself might be the topological state of the DNA transferred to the recipient cell. According to studies on λ integrase, site-specific recombination confers dynamic topological changes on DNA structure (Bliska and Cozzarelli, 1987; Crisona et al., 1999). Excised ICEclc, produced via the site-specific recombination reaction catalysed by IntB13 integrase, may thus have a temporarily different topology than a covalently closed plasmid, and this might somehow facilitate its transfer.
If oriT1 and oriT2 are initiating ICEclc transfer at the same frequencies, one would expect the proportion of green fluorescent transconjugants from donor cells having two ICEclc copies, in which one oriT is disrupted, to be close to 33%. The reason for this is that three intact oriTs are available in such donors; e.g. one copy of oriT1 and two copies of oriT2 in strain B13-3207 or UWC3218. Our results of ∼30% of GFP-labelled transconjugants are in agreement to this prediction. Interestingly, 56% of such green fluorescent transconjugants contained only the single-mutant copy of ICEclc (ΔoriT1-egfp) (Fig. 4). Therefore, the bona fide transfer frequency of ICEclc (ΔoriT1-egfp) from UWC3218 would be 1.3 × 10−3[= 7.6 × 10−3 (observed frequency) × 0.30 (proportion of green transconjugants) × 0.56 (proportion of green transconjugants with true single ICEclcΔoriT1-egfp transfer)], which would be indicative for efficiency of ICEclc transfer using oriT2. Vice versa, the bona fide transfer frequency of ICEclc (ΔoriT2-egfp) from UWC3220 would be 1.0 × 10−3 (= 6.1 × 10−3 × 0.28 × 0.58), which would be representative for the efficiency of ICEclc transfer by oriT1 only. Because the transfer frequency of intact ICEclc from UWC3218 as donor is 5.3 × 10−3[= 7.6 × 10−3 × 0.70 (proportion of non-green transconjugants)], and from UWC3220 as donor is 4.4 × 10−3 (= 6.1 × 10−3 × 0.72), the transfer frequency of an ICEclc having both functional oriTs is four times higher than that of ICEclc containing a single oriT. This calculation suggests that the dual oriT system conferred an evolutionary selectable advantage on ICEclc because of its more efficient transfer. Remarkable was also the finding that an ICEclc with both oriT1 and oriT2 disrupted still could transfer from a donor carrying an additional intact ICEclc, albeit at low proprotion (∼7% of transconjugants, Table 2). As shown in Fig. 4C, this aberrant transfer was partly due to cotransfer with the second intact ICEclc present in the donor and even in a small proportion of tested transconjugants, because of a recombination between the intact ICEclc and the double-oriT mutant. However, in 39% of the tested green transconjugants the double-oriT disrupted ICEclc seemed to have transferred on itself. This suggests the existence of a third but less optimal oriT on ICEclc. There is precedent for such low frequency events, given that in vitro experiments with other systems such as Tn916 demonstrated that its relaxase protein has the potential to cleave several sites in the oriT region non-specifically (Rocco and Churchward, 2006).
Although we found the two oriTs to be functional simultaneously in a population of donor cells, this does not necessarily mean that both oriTs are active on the same ICEclc molecule in the same cell. In fact, if this would occur the excised molecule could potentially be digested twice with the consequence of a stalling relaxosome. Only two conjugative plasmids have so far been reported to carry two oriTs: plasmid R6K from E. coli and pAD1 from Enterococcus faecalis. R6K, known as a plasmid replicating in a peculiar way, contains two functional oriTs (named the α- and β-oriT) that are located in the vicinity of the α- and β-origins of replication (oriV) (Avila et al., 1996). Both oriTs of R6K consist of highly similar 96 bp palindromic sequences and completely identical nic sites but on different strands with respect to each other. Double nicking at both nic sites has indeed been detected on the same R6K molecule, indicating that two relaxosomes can in principle be pre-assembled simultaneously if they are on opposite strands. In the pheromone-inducible conjugative plasmid pAD1, two regions that confer mobility on exogenous plasmids have been identified as oriT. A main oriT leading to high efficient transfer consists of direct and inverted repeats (Francia and Clewell, 2002), and is located downstream of orf53 encoding a putative TraD/VirD4-like coupling protein (Francia et al., 2001), while a second oriT with lower efficiency is located within the repA gene (An and Clewell, 1997). However, it is still difficult to explain the mechanism of processing two oriTs by the previous models of conjugation because two simultaneously active oriTs on the same ICE molecule would lead to stalling relaxosomes. In the case of DNA-processing forks proceeding in the same direction the generation of ssDNA started from the first oriT would terminate at the second one before the entire ICEclc is processed, whereas in the case of forks proceeding in opposite directions they would stagnate when they meet. Our current working hypotheses are that either not all cells in a population of donors employ the same oriT or that there is a timing difference between using one or the other oriT, by which donor cells could transfer ICEclc twice consecutively, upon a reconstitution of the excised single-stranded molecule in the donor cell after the first round of transfer. We also identified the relaxase for ICEclc conjugation in this work as being the gene product from orf50240, and showed by knockouts that it is essential for ICEclc transfer from both oriT1 and oriT2. According to a recent review comparing the sequence of relaxases (Garcillan-Barcia et al., 2009), the relaxase of ICEclc is classified into the MOBH family, which is comprised of relaxase homologues from different GEI/ICE. The experimental information on relaxases belonging to this family is scarce, but the mechanism of DNA processing has been thought to be different from that of other well-known relaxases because of their predicted structural uniqueness. In this study the function of the relaxase of ICEclc was further proven by showing that it can catalyse the in vivo strand exchange between two plasmids both carrying oriT1. Strangely however, the relaxase could not catalyse the nicking-rejoining reaction in E. coli on plasmids carrying oriT2, suggesting that other factors in conjunction with the relaxase are necessary for oriT2 processing on ICEclc. The additional factors needed for oriT2 might thus control the timing and occasion of oriT2 processing, and as a consequence regulate ICEclc transfer using one of two oriTs to avoid the ‘stalling’ paradox of operating two oriTs simultaneously.
We can thus conclude that despite the low homology to known conjugative plasmids and related ICE conjugative systems the ICEclc family primes transfer with the use of a typical relaxase activity to nick the excised ICEclc molecule. The dual oriT use shown here for ICEclc is on the contrary unique compared with other mobile genetic elements, but may be more common in ICEs belonging to the ICEclc/ICEHin1056 family. Given the quite wide variety of GEI/ICE belonging to the ICEclc/ICEHin1056 family in several α-, β- and γ-proteobacterial genomes, the dual oriT transfer system may have been the crucial evolutionary advantage for its more effective distribution.
Strains and plasmids
Bacterial strains and plasmids used in this study are listed in Table 4. LB medium (Sambrook and Russell, 2001) was routinely used for growing E. coli and Pseudomonas cells. As a defined MM the type 21C MM (Gerhardt et al., 1981) was used, supplemented either with 5 mM 3-chlorobenzoate (3CBA) or with 10 mM fructose. P. knackmussii B13 is the original host strain of the ICEclc element of which it harbours two copies.P. putida clc6 is a transconjugant of P. putida UWC1 harbouring one copy of ICEclc. P. putida dint1-5 is an UWC1 derivative harbouring one mutant copy of ICEclc, which has pRL27-derived mini-Tn5 insertion in its intB13 gene. When required 50 µg ml−1 of ampicillin (Ap), kanamycin (Km), tetracycline (Tc), and rifampicin (Rif), 20 µg ml−1 of gentamicin (Gm), 500 µg ml−1 of carbenicillin (Cb), and/or 100 µg ml−1 of spectinomycin (Sp) were added. Solid media were prepared by the addition of 1.5% agar. Strains of E. coli and Pseudomonas were grown at 37°C and 30°C respectively.
Table 4. Bacterial strains and plasmids used in this study.
Strain or plasmid
Source or reference
3CBA+, utilizes 3CBA as a sole carbon source; MCS, multiple-cloning sites.
Preparation of plasmid and chromosomal DNAs, digestion with restriction endonucleases, DNA fragment recovery, DNA ligation and transformation of E. coli cells were carried out according to established procedures (Sambrook and Russell, 2001) or to specific recommendations by the suppliers of the molecular biology reagents (Promega and Qiagen). The transformation of bacterial cells by electroporation was performed as described previously (Miyazaki et al., 2008). PCR was performed with GoTaq DNA polymerase (Promega) or PrimeStar DNA polymerase (Takara), and the primers used are listed in Table S1. All PCR products cloned were confirmed by sequencing with the BigDye system version 3.1 (PE Applied Biosystems) and an ABI PRISM 3700 sequencer (PE Applied Biosystems). The nucleotide and protein sequences were analysed by using Genetyx 13 software (SDC, Tokyo, Japan). Homology searches were performed using blast programs, and a multiple alignment was created using clustalw program. Sequence comparison and secondary structure searches were performed using GenomeMatcher (Ohtsubo et al., 2008).
Screening of the oriT region from an ICEclc DNA library
Four overlapping cosmids (i.e. 1H5, 3C9, 3G3 and 5H11) that had been constructed previously are covering the whole ICEclc region (Ravatn et al., 1998a). To obtain plasmid-based short insert sublibraries (−1 kb), each cosmid was partially digested with Sau3AI and subcloned into the BglII site of pME6010, after which the sublibrary was introduced into E. coli DH5α by electroporation. Subsequently, the sublibrary was purified from the mixture of more than 103 transformant colonies and reintroduced into P. putida strain dint1-5 by electroporation. Approximately 103 transformant colonies were grown on solid media and then collected in 1 ml of MM by washing colony material from the plates. Simultaneously, variation of insert fragment in the sublibrary was verified by isolating plasmid from 10 independent transformants chosen randomly. The mixture was inoculated with 0.25% dilution in MM with 5 mM 3CBA and Tc, and grown for 48 h at 30°C until stationary phase (OD = 1.0). The culture was then mixed with an equal volume of P. putida UWCGC culture that had been grown overnight (OD = 1.0) in MM with 10 mM fructose and Gm. Cells in the mixture were collected by centrifugation, washed twice with 1 ml of MM, centrifuged again and resuspended in 20 µl, after which the mixture was spotted on a MM agar plate containing 0.5 mM 3CBA. After incubation at 30°C for 48 h, the cells on the agar were suspended in 1 ml of LB broth, serially diluted and plated on selective agar plates (Tc and Gm). Transconjugants of UWCGC were also visually confirmed by their red fluorescence from mCherry. Transconjugant colonies were purified and their plasmid content was examined. Inserts of the pME6010-derived purified plasmids residing in purified transconjugants were sequenced and positioned to the respective ICEclc region (GenBank Acc. No. AJ617740).
Constructions of the plasmids containing the oriT1 or oriT2 regions
Differently sized fragments of both oriT1 and oriT2 regions were produced by PCR amplification and cloned into the BglII site in the multiple-cloning sites of pME6010. For oriT1 this resulted in pME680-4, pME560-2, pME480-25, pME540-6, pME370-2 and pME240-4. A 1336 bp fragment containing oriT2 was PCR amplified and cloned in two directions in pME6010, generating pME285si-1 and pME285si-3 (Fig. 1B). Subclones pME285si-1E and pME285si-1N were generated from pME285si-1 by digestion with EcoRI or NcoI, respectively, and self-ligation. Subclones pME285si-3E and pME285si-3N were generated from pME285si-3 in the same way (Fig. 1B). A 280 bp intergenic region between orf91884 and ssb was amplified and cloned into BglII site in pME6010, generating pMETS285. As a negative control, 630 bp of the clcAB region was amplified and cloned between EcoRI and BglII sites in pME6010, generating pMEAB630. The insert fragments of pME680-4, pME285si-1N and pMEAB630 were released by XhoI-HindIII digestion and cloned into pBBR1MCS-5, generating pBBR680-4, pBBR285si-1N and pBBRAB630 respectively. All plasmids were constructed in E. coli first, and when required they were purified and then introduced into P. pudita dint1-5.
Minimal oriT region determination
For mobilization tests P. putida dint1-5 donors harbouring pME6010 derivatives were grown in MM with 5 mM 3CBA and Tc until stationary phase, and then mixed with recipient strain P. putida UWCGC as described above. Cell matings were allowed to proceed for 48 h at 30°C, after which the cells on the agar surface were suspended in 1 ml of LB broth, serially diluted and plated on selective agar plates (Tc and Gm). To control for spontaneous resistance both donors and recipient strains were mated individually.
Disruption of the orf50240, orf91884 and oriT regions
Allelic exchange mutagenesis of ICEclc was carried out using a system inspired in the I-SceI method of Posfai (Posfai et al., 1999). Detailed information on which regions were deleted is shown in Fig. S4. Approximately 750 bp regions located upstream and downstream of a target region were amplified by PCR using total DNA of strain B13 as a template. In some cases a custom synthesized Ptac-egfp cassette (DNA2.0) was inserted between the two PCR fragments. All fragments were cloned into the multiple-cloning site of pJP5603-ISceIv2 (the kind gift of E. Martinez-Garcia and Victor de Lorenzo). The resulting plasmid was purified from E. coli CC118 λpir and introduced into P. putida clc6 or P. knackmussii B13 by electroporation. Possible single recombinants in which the pJP5603-ISceIv2 derivative had integrated were selected on media with Km. Subsequently, the secondary plasmid, pSW(I-SceI) (Wong and Mekalanos, 2000), was introduced into a purified single recombinant by electroporation. For P. putida recombinations single transformants were purified by selection on Cb and Km and regrown, after which I-SceI endonuclease was induced by the addition of 15 mM 3-methylbenzoate overnight. Induced cultures were plated on solid media with Cb and CbR colonies were screened for KmS by replica plating. KmS clones were purified and the expected double-cross-over-mediated homologous recombination was verified by PCR. For P. putida about 50% of KmS clones had the expected allelic exchange, whereas the other 50% were revertants to wild-type. Double P. putida recombinants with the desired allelic exchange were cured from plasmid pSW(I-SceI) via several passages in LB without antibiotics and confirmation by PCR. In this way, five mutants, UWC2989 (orf50240 deleted), UWC3012 (orf91884 deleted), UWC3212 (oriT1 replaced by egfp), UWC3213 (oriT2 replaced by egfp) and UWC3245 (oriT1 deleted and oriT2 replaced by egfp), were constructed. To introduce a second intact copy of ICEclc, UWC3212, UWC3213 and UWC3245 were separately mated with P. knackmussii B13-Spc, in which one of two ICEclc was marked by SpcR gene inserted in the intergenic region between orf48922 and orf50240 (Sentchilo et al., 2009). After mating on MM with 0.5 mM 3CBA agar for 48 h, SpcR and RIfR transconjugants that have the second copy of ICEclc at tRNAGly-3 were isolated and verified by PCR using primers RT-tRNA3fw and 376rv, resulting in UWC3218, UWC3220 and UWC3280. Probably because of leaky I-SceI expression, in the case of P. knackmussii B13 single recombinant transformed clones with pSW(I-SceI) were selected on media with Cb alone but not with Cb and Km, and second recombination occurred without induction by 3-methylbenzoate. In such CbR and KmS clones, unexpectedly, only about 2% of suspected double recombinants in P. knackmussii B13 carried the proper allelic exchange, whereas 98% had reverted to wild-type. The pSW(I-SceI)-cured double recombinants were obtained via several passages in LB, resulting in P. knackmussii B13-3207 (ΔoriT1-egfp) and B13-3217 (ΔoriT2-egfp).
Construction of the strain containing chromosomal mini-Tn7 with orf50240
The 1867 bp fragment containing the entire orf50240 gene and its ribosome binding site was amplified by PCR and cloned into the BamHI-XhoI sites of pUC18-mini-Tn7T-LAC. The resulting plasmid pmT7LAC50240 was introduced into P. putida UWC2989 (Δorf50240) by electroporation with pUX-BF13, which contains the Tn7 transposase gene (Koch et al., 2001). GmR transformants were selected, and PCR was performed to verify if transformants carried a single chromosomal copy of the construct at the attTn7 site located downstream of glmS gene. The Flp recombinase-mediated marker excision system (Hoang et al., 1998) was then used to remove the GmR gene from the chromosomal mini-Tn7 element, generating P. putida UWC3026.
ICEclc self-transfer assay
Self-transfer assays with donors containing ICEclc were performed as described previously (Sentchilo et al., 2003a). As donors we used here P. knackmussii B13, B13-3207 (egfp insertion in oriT1) or B13-3217 (idem, in oriT2), and P. putida clc6 (one copy of ICEclc), UWC3012 (orf91884 disrupted), UWC2989 (orf50240 disrupted), UWC3026 (orf50240 disrupted but complemented with mini-Tn7), UWC3218 (two copies of ICEclc and egfp insertion in one of two oriT1s), UWC3220 (idem, in one of two oriT2s), or UWC3280 (two copies of ICEclc but oriT1 disrupted and egfp insertion in oriT2 on the same ICEclc copy). Recipients were P. putida UWC1, UWC1G (GmR derivative) or UWCGC (GmR red fluorescent derivative). All donors were grown in MM with 5 mM 3CBA for 48 h (OD = 1.0) and mixed with recipients grown in MM with 10 mM fructose for 24 h (OD = 2.0). After matings on MM with 0.5 mM 3CBA for 6, 24, 48 or 72 h, the cells were serially diluted and plated on MM agar with 3CBA and MM agar with 3CBA plus antibiotics (Rif or Gm) to select for donors and transconjugants respectively. When required 100 nM, 10 µM or 1 mM of IPTG was added in the donor cultures and the mating agar. The types of ICEclcs in transconjugants were characterized by PCR using two primer pairs; Bgl_51878.F and Bam_52323.R specific for the 446 bp oriT1 region, and 94886fw and 95204rv specific for the 319 bp oriT2 region.
In vivo strand exchange assay
The 1867 bp BamHI-XhoI fragment containing orf50240 was released from pmT7LAC50240 and cloned into pBluescriptII SK+. From here, the fragment containing orf50240 could be recovered by XbaI-XhoI and was cloned into pBAD18, generating pBAD50240. E. coli DH5α was cotransformed with the plasmid pBAD50240 and with two plasmids of which the in vivo exchange was to be tested, in the following combinations; pME680-4 (oriT1) and pBBR680-4 (oriT1), pME680-4 (oriT1) and pBBR285si-1N (oriT2), pME285si-1N (oriT2) and pBBR285si-1N (oriT2), or pMEAB630 (clcAB) and pBBRAB630 (clcAB). As a negative control, pBAD18 was used in place of pBAD50240. The cells were grown overnight in LB with Ap, Tc, Gm, and 0.2% glucose and then diluted 100-fold into fresh medium. After 3 h of growth (OD = 0.4) the cultures were divided into two equal portions, one of which was supplemented with 0.2% l-arabinose. The growth of cultures was monitored for an additional 4 h. Plasmid DNA was purified by alkaline lysis and 100 ng of it was used as template for PCR with four primer pairs: pVS1.F and either M13.F or M13.R, and pVS1.R and either M13.F or M13.R. The cycle program was set for 30 cycles as follows: 94°C for 30 s, 51.5°C for 30 s and 72°C for 60 s. PCR products were purified from 1.0% agarose gel and confirmed by sequencing.
Construction of the plasmids containing oriT1 with a point mutation
A point mutation was introduced in the oriT1 region by inverse PCR using pBBR680-4 as a template and six different primer pairs (mut51953.F and 51952.R, 52018.F and mut52017.R, 52085.F and mut52087.R, mut52137.F and 52152.R, 52232.F and mut52231.R, or 52324.F and mut52323.R). Each primer pair was designed in head to head fashion to amplify the entire pBBR680-4 from a different starting position, with one of a pair of primers containing a single nucleotide mutation. PCR products were subsequently phosphorylated by T4 polynucleotide kinase (Fermentas), self-ligated and then transformed into E. coli DH5α. Six variants of pBBR680-4, named pBBRT1mut12, pBBRT1mut34, pBBRT1mut56, pBBRT1mut78, pBBRT1mut910 and pBBRT1mut1112, were isolated and their point mutations were confirmed by sequencing. Subsequently, the six variant plasmids were tested in in vivo strand exchange assays with pME680-4 and pBAD50240, and the nicking region was deduced from sequence analysis of the junction amplicons (see above).
Significance of different treatments was examined by pairwise t-test (two-sided, P < 0.05). Error bars denote standard deviations from the mean in triplicate experiments.
The authors gratefully thank E. Martinez-Garcia and V. de Lorenzo for the kind gift of plasmid pJP5603-ISceIv2 and pSW(I-SceI); Nicolas Pradervand for preparation of P. putida dint1-5; Antonia Mayer for assisting computer analysis. R.M. was supported by a postdoctoral fellowship from the Japan Society for the Promotion of Science for Research Abroad and by a grant from the Swiss National Science Foundation (31003A-124711).