The evolved clone BW4005 was previously characterized phenotypically by comparison to its ancestor, and exhibited a lower growth yield, a higher production of acetate from glucose in chemostat culture, and a lack of obvious difference in glucose uptake 28 . At the ecological level, BW4005 was present in low proportion in its glucose-limited evolving population and displayed negative frequency-dependent fitness when competed with the ancestor 19 . By contrast, no genetic information was available to explain these phenotypic and ecological traits, even after re-sequencing the genome of a clone, called BW4005.1, recovered from the initial glycerol stock of BW4005, by using a combination of 454 and Illumina technologies. Indeed, we detected a single mutation compared to the ancestor, consisting of a short deletion in ahpC encoding the alkyl hydroperoxide subunit involved in peroxide resistance (Table 1). As shown below, this mutation did not contribute to increased fitness in the chemostats. Moreover, the genome sequence of BW4005.1 was inconsistent with previously detected changes in the RFLP profiles of another clone isolated independently from the same glycerol stock of BW4005, called BW4005.2, using the mobile genetic element IS3 as a probe 27 . Indeed, while the RFLP-IS3 profile of the ancestral strain BW2952 revealed four IS3-containing EcoRV genomic fragments of 1.7, 2.6, 4.6, and 10.6 kb, respectively, with hybridization intensities proportional to the relative molecular weight of these fragments (Figure 1a), the one of BW4005.2 showed a fifth hybridizing fragment of ~7.9 kb with a hybridization intensity stronger than expected from its molecular weight (Figure 1a). This was unlikely to be due to contamination, since both clones harbored the ahpC mutation (Table 1).

^aThe positions are given according to the genome sequence of the ancestor strain BW2952 29 . ^bThe genome sequence of BW4005.1 was first obtained and detected only the first mutation, while the one of BW4005.C6 was obtained afterwards and contained the three mutations. ^cNot applicable.

We characterized the genomic sequences adjacent to IS3 in the 7.9-kb fragment specific to BW4005.2 by inverse PCR experiments. One side (right on Figure 1b) corresponded to the gatD gene with the IS3 insertion at exactly the same position (2,061,936) as observed in the ancestor for the IS3 element contained in the 10.6-kb hybridizing fragment. This particular ancestral IS3 is inserted between positions 2,060,677 and 2,061,936. The other side corresponded to a genomic sequence (2,058,010) located 2668 bp upstream in the ancestral genome. Therefore, this fifth IS3 probably resulted from an IS3 transposition event at position 2,058,010 followed by a recombination event between this new IS3 copy and the one located in the ancestral genome between positions 2,060,677 and 2,061,936 within gatD. These events subsequently resulted in the 2668-bp deletion of the intervening fragment, affecting the two genes yegR and yegS and part of ogrK (Figure 1c). However, two features of the RFLP-IS3 pattern suggested a more complex rearrangement (Figure 1a): the two 10.6 and 7.9 kb IS3-hybridizing fragments were co-existing in BW4005.2, indicating the presence of two copies of at least gatD, and the 7.9 kb fragment presented a hybridizing intensity higher than expected from its molecular weight compared to the four other hybridizing fragments. Altogether, these data might be explained by the presence in the chromosome of BW4005.2 of a tandem duplication containing two heterozygous copies, with and without the IS3-generated deletion respectively (Figure 1c). We then performed PFGE experiments with XbaI-digested genomic DNA of BW2952 and BW4005.2 (Figure 1d). We found two extra XbaI fragments of 61.5 and 19 kb in BW4005.2 compared to BW2952, which also suggested the presence of a complex rearrangement.

These genomic rearrangements were not detected by genome sequencing in the clone BW4005.1. Since both clones BW4005.1 and BW4005.2 were derived from a single cycle of growth after plating the initial glycerol stock of BW4005 onto LB solid medium, we hypothesized that the rearrangements were present in the original evolved clone, but were characterized by a high instability. We therefore investigated the stability of the genomic rearrangements by isolating six additional clones, called BW4005.C1 to BW4005.C6, after plating the initial glycerol stock of BW4005 onto LB medium, and analyzing them by RFLP-IS3 and PFGE (Figure 1e). The RFLP-IS3 profiles were checked for the presence or absence of the 10.6 and 7.9 kb IS3-hybridizing fragments. Hence, their simultaneous presence was diagnostic of the putative heterozygous duplication while the presence of only the first or second fragment indicated the resolution of the putative duplication into the ancestral or deleted copy, respectively. Similarly, the PFGE profiles were checked for the presence or absence of the two 61.5 and 19 kb bands. The RFLP-IS3 profiles revealed that four (BW4005.C1 to BW4005.C4) of the six clones had lost either one of the putative duplicated DNA copies (Figure 1e), with clones BW4005.C1 and BW4005.C2 harboring the single deleted copy and BW4005.C3 and BW4005.C4 the ancestral copy. The two last clones, BW4005.C5 and BW4005.C6, retained the putative heterozygous duplication. Consistent with the RFLP-IS3 profiles, the PFGE analysis showed that clones BW4005.C1 and BW4005.C2 retained only the 19 kb band, while BW4005.C3 and BW4005.C4 had the ancestral profile, and BW4005.C5 and BW4005.C6 harbored the two extra fragments (Figure 1e). These results indicated the high genetic instability of the putative duplication as well as its capacity for resolution into alternative states.

Moreover, both RFLP and PFGE experiments, using genomic DNA isolated from different independent cultures of BW4005.C6, exhibited variation in the intensity of the additional bands (Figure 1 and data not shown), which further suggested the instability of the genetic rearrangement. This instability may provide an explanation for the discrepancies between the RFLP/PFGE experiments and the genome re-sequencing results and is considered in more detail below.

To address the discrepancies between the different approaches, we decided to re-sequence the genome of BW4005.C6 that harbored the complex rearrangement (Figure 1e). We now detected three mutations compared to BW2952 (Table 1): the same in-frame 3-bp deletion in ahpC, the 2668-bp deletion of ogrK-yegR-yegS, and a duplication of ~180 kb extending from genomic positions 1,993,429 to 2,172,740. The duplicated fragment is localized between two IS5 elements (Figure 2), and therefore likely resulted from a recombination event between one and the other of these two IS5 copies during DNA replication. We further validated the presence of the duplication by three additional tests: first, we PCR-amplified and sequenced the new junction sequence resulting from the duplication by using the primers FrontR and FrontF (Table 2, Figure 2). Both the size and sequence of the PCR products confirmed the duplication. Second, an additional IS5-containing EcoRV restriction fragment of 1.5 kb should be produced at the junction of the duplication (Figure 2) and should therefore be detected by RFLP-IS5. This was indeed shown to be the case (data not shown). Finally, the PFGE profiles with the additional XbaI bands were easily explained on the basis of the duplicated genome sequence (Figure 2).

^aThe positions are given according to the genome sequence of the ancestor strain BW2952 29 . ^bSequences in bold are homologous to the bla antibiotic resistance gene. ^cNot applicable; these primers are complementary to sequences inside the IS3 elements.

We next investigated the relationships between the complex chromosomal rearrangements and the phenotypes known for BW4005. We measured the growth yield, acetate production, and fitness of the three clones BW4005.C2, BW4005.C3 and BW4005.C6 compared to the ancestor BW2952. These three clones are genotypically different (Figure 1e): the duplication has been resolved into the deleted or ancestral copy in BW4005.C2 and BW4005.C3, respectively, while BW4005.C6 contains the heterozygous duplication. As shown in Table 3, BW4005.C2 and BW4005.C6 had similar evolved phenotypes (decreased fitness at a 50:50 ratio, reduced yield, and high acetate production under glucose limitation), while BW4005.C3 had phenotypic properties similar to the ancestor. These results suggested that the 2668-bp deletion was the main contributor to the BW4005 evolved phenotypes, and not the ahpC mutation. Moreover, the presence of the heterozygous duplication seemed to confer similar phenotypes compared to the 2668-bp deletion alone.

^aGrowth yields are shown relative to the ancestral strain BW2952 and were obtained from 24-hour glucose-limited chemostats by measuring the optical density of the cultures at 600 nm. Values are from three independent measurements.

^bAcetate production was tested with 24-hour chemostat cultures. +, acetate present at 2–5 mg l^-1; -, below the detection limit (< 0.15 mg l^-1). Data were from at least three independent biological repeats with less than 10% standard errors.

^cCompeting strains were mixed at 50:50 or 1:99 ratios after individual acclimation in the chemostat evolution conditions. Changes in the proportions of the competitors after 24 hours were used to calculate fitness values. Data presented are mean ± SEM from at least two independent experiments. ND, not determined.

^dSymbols are identical to the legend of Figure 3.

We next investigated the respective phenotypic contribution of each of the yegR and yegS genes, by first moving the 2668-bp deletion and second inactivating them individually into the ancestral genetic background. We found that both the 2668-bp deletion and yegS-inactivated strains exhibited the evolved phenotypes, while the yegR-inactivated strain had phenotypes indistinguishable from the ancestor (Table 3).

An important unresolved question is whether the duplication itself conferred a benefit under the chemostat selection conditions. Since evolved phenotypic attributes were not specifically provided by the duplicated state, we tested this indirectly by comparing the population dynamics of the duplication in two environments: non-selective conditions in LB batch cultures and the glucose-limited chemostat selection environment (Figures 1e and 3). The clone BW4005.C6, harboring the heterozygous duplication, was grown either in LB medium for three 24-hour growth cycles or for three days in chemostats in low-frequency competition with MC4100 to mimic the evolution experiment (Materials and methods). In each case, 32 individual clones, sampled at each day interval, were analyzed for the presence of the genomic rearrangements by colony-PCR using the primer pairs IS5F1/IS5R1, IS3F1/IS3R1, and IS3F1/IS3R2. The primer pair IS5F1/IS5R1was specifically designed to detect the new junction produced by the duplication, and IS3F1/IS3R1 and IS3F1/IS3R2 to detect the 2668-bp IS3-mediated deletion (Figure 3a-c). Under the non-selective conditions (Figure 3e), the proportion of clones harboring the heterozygous duplication decreased significantly by the third 24-hour growth cycle (t-test: P < 0.05), with the concomitant significant increase of the proportion of clones with single ancestral or deleted copies (t-test: P < 0.05). By contrast, the proportion of clones harboring the heterozygous duplication was high and stable over the three days under the original selection condition (Figure 3d). These results showed the genetic instability of the heterozygous duplication under non-selective conditions, in stark contrast to its stability over a short-term period in the glucose-limited conditions prevailing during the evolution experiment. However and despite these stability differences, we were unable to distinguish phenotypically (Table 3) the clones harboring the heterozygous duplication and the single deleted copy (even the reconstructed clone with the specific yegS inactivation in the ancestral genetic background had identical evolved phenotypes).

The 2668-bp deletion from the BW4005 evolved clone was moved into the ancestral genome by allelic exchange using the suicide plasmid pKO3 51 . Briefly, a 1100-bp PCR fragment with about 550 bp of adjacent sequences on each side of the deletion was obtained using BW4005 genomic DNA and the primers DelR and DelF (Table 2), and then cloned into pKO3. The obtained plasmid was introduced by electro-transformation into BW2952 and chloramphenicol-resistant cells, formed after chromosomal integration of the plasmid, were selected. Resistant clones were streaked onto LB+sucrose plates to select for cells that lost the plasmid. Indeed, pKO3 carries the sacB gene that makes cells carrying that plasmid susceptible to killing by sucrose. These cells were then screened by PCR and scored for the presence of the deletion allele by the size of the PCR products. We constructed yegS::bla and yegR::bla derivatives of BW2952 using homologous flanking sequences as described 50 . The primer sets used for strain constructions are listed in Table 2.

Analysis of extracellular acetate in culture media from individual chemostats inoculated with different bacterial strains was performed after filtration through a Milex® filter unit (Millpore, Ireland, 0.22 μm) by using an assay kit for acetic acid with the UV method (Boehringer Mannheim/R-Biopharm, Germany), according to the manufacturer’s guidelines. The growth yields were inferred by measuring the optical density at 600 nm of 24-hour chemostat cultures growing in the exact same conditions prevailing during the evolution experiment.

Each bacterial strain was competed against a metC::Tn10 derivative, called BW3454 (Table 4), of the BW2952 ancestral strain 29 , under the same conditions used during the chemostat evolution experiment, except that the medium was supplemented with 4 μg ml^-1 methionine. Before mixing the two competitors, each strain was independently acclimated by overnight growth in the glucose-limited chemostats to ensure identical physiological state. The fitness of the competing strains was calculated as the selection coefficient S, as described 19 .

A first single colony of BW4005, called BW4005.1, was isolated after streaking the original glycerol suspension on LB plate and incubating overnight at 37°C. The BW4005.1 clone was then inoculated into 10 ml LB and grown overnight at 37°C. This culture was used for genomic DNA extraction as described 29 . Whole genome re-sequencing of BW4005.1 was performed by 454 and single-end Solexa technologies 33 . Briefly, a total of 216,058 single-end reads with an average length of 404 bps, giving a 19.08X coverage of the genome, were generated using the GS FLX system (Roche, Welwyn Garden City, Hertfordshire, UK) and assembled to 78 contigs with 454 Newbler assembler (http://www.454.com/products/analysis-software/). A total of 8,862,831 single-end reads with an average length of 33 bps were generated using the Solexa Genome Analyzer (Illumina, Little Chesterford, Essex), giving a 63.9× coverage. All Solexa reads were mapped to the contigs generated from Newbler using MAQ 52 . The order of the contigs was determined based on BLASTn alignments with the genome sequence of BW2952 as the reference (Genbank accession number NC_012759). By analyzing these data, we found a few sequencing errors in the original BW2952 genome sequence (Additional file 1: Table S1). All 78 sequence gaps in BW4005 and 151 suspected SNP sites were filled or checked by sequencing PCR products with ABI 3730XL capillary sequencer (Applied Biosystems, Foster City, California, USA).

In our second genome re-sequencing attempt, a second colony derived independently from the same initial glycerol stock of BW4005, and called BW4005.C6 (see below, Table 4), was grown as described above. (BW4005.C6 was confirmed by our molecular analyses to harbor the heterozygous duplication). Its genomic DNA was extracted, checked and re-sequenced on the Illumina Genome Analyzer platform using single-end 36-bp reads (GATC-Biotech, Germany). Candidate mutations were identified in comparison to the ancestral BW2952 genome 29 using a recently developed computational pipeline 17 . Genome sequencing data have been deposited in the ENA Sequence Read Archive (ERP002251).

Restriction fragment length polymorphism (RFLP) using internal fragments of IS3 and IS5 as probes (RFLP-IS3 and RFLP-IS5, respectively) was performed with genomic DNA of BW2952, BW4005 and derived clones as described previously 27 . Genomic sequences adjacent to relevant IS3 elements were identified by inverse PCR experiments 27 , using primers IS3ER and IS3EF (Table 2) that are complementary to the extremities of IS3 and oriented toward its flanking regions. Briefly, after EcoRV digestion of genomic DNA from clone BW4005.2 and fractionation by agarose gel electrophoresis, the relevant DNA fragments were extracted from the gel, self-ligated and submitted to PCR using the primers IS3ER and IS3EF. The obtained PCR products were then sequenced to identify the IS3 genomic neighborhood.

The new junction fragment generated by the duplication event was detected by PCR experiments using the two primer pairs FrontR/FrontF and IS5F1/IS5R1 (Table 2). The stability of the duplication was assessed by sub-culturing the evolved clone BW4005 in two different environments: non-selective LB medium and the chemostat evolution conditions. For non-selective conditions, the initial glycerol stock of BW4005 was plated onto LB solid media and incubated overnight at 37°C. Six colonies were sampled and grown for 12 hours in LB liquid medium at 37°C, the resulting clones being called BW4005.C1 to BW4005.C6. The BW4005.C6 clone, that was shown to retain the duplication after culture, was further inoculated by a 1:100 dilution into two independent McCartney bottles, each containing 5 ml LB, and incubated for 24 hours at 37°C with shaking at 200 rpm. These dilution/growth cycles were repeated for a total of three times. At each of the three steps, 100 μl of the appropriately diluted 24-hour cultures were plated onto LB solid medium and 32 well-separated colonies, 16 deriving from each of the two independent McCartney bottles, were sampled for DNA extraction and subsequent PCR analysis to score for the presence or absence of the duplication. For selective chemostat conditions, glucose-limited chemostats of BW4005.C6 and MC4100 were mixed after 16 hours of acclimation. Owing to the negative frequency-dependent fitness and low frequency of BW4005 in the initial chemostat evolution experiment 19 , we mixed BW4005.C6 with MC4100 at a ratio of 1:99 and this co-culture was maintained for three days. Once every day, we sampled 32 colonies derived from BW4005.C6 after plating appropriate dilutions on LB plates and scoring for colonies resistant to kanamycin.

PFGE was performed using a CHEF Mapper system (Bio-Rad Laboratories, Hercules, CA, USA) according to the procedure described by the US CDC PulseNet program (http://www.cdc.gov/pulsenet/pathogens/pfge.html). XbaI-digested genomic DNAs of BW2952, BW4005 and derived clones were separated at 6.0 V/cm for 19 hours at 14°C, with increasing pulse times of 6–36 s and an angle of 120°.