Strains, plasmids, and media

Experiments were done using E. coli K-12 MG1655 [23] and various derivatives of it. MDS12, MDS30, and MDS42 are reduced-genome versions of MG1655, carrying 12, 30, and 42 genomic deletions, respectively [24, 26]. T7 polymerase-expressing MG1655-T7 and MDS42-T7 strains will be described in detail elsewhere. Briefly, they carry an IPTG-inducible lac operator/T7 polymerase cassette replacing the yahA-yaiL genomic region of the parental strain. MDS42+IS1 carries a single IS1 in yeaJ. Plasmids pCTX and pCTXVP60 were originally constructed by Günther M. Keil [24] (Additional file 1). Plasmid pSG1144, carrying an F-type of replicational origin, a trfA-oriV replicon from plasmid R2 [27], a chloramphenicol resistance gene and a T7 promoter, was obtained from Scarab Genomics (Madison, WI, USA). Inducible variants of orf238 and ctxvp60 were constructed by cloning the genes in the multiple cloning site of pSG1144 (Additional file 1). Genes ctxvp60 opt and ctxvp60 dezopt were synthesized by GenScript Corporation (Piscataway, NJ, USA). Plasmids were prepared using IS-free, MDS42 host strain. Standard laboratory media LB and LB-agar plates were used during the cultivations [28]. Chloramphenicol and ampicillin were added at 25 μg/ml and 50 μg/ml final concentrations, respectively.

PCR primers

Analysis of mutations

To analyze the mutational spectrum of the ctxvp60 gene, a 2481-bp segment was amplified from the mutant cells using the primer pair T7/chi3. The amplified fragments were resolved on a 1% agarose gel and compared to a fragment generated from the wild-type template. Identical size indicated mutations affecting only one or a few nucleotides, a decrease in size or failure of amplification indicated a deletion, and an increase in size suggested an IS insertion. In case of insertion mutants, the identity and the localization of ISes were determined by PCR using IS- and plasmid-specific primer pairs. For 12 mutants, insertions sites were further analyzed by sequencing.

Growth measurements and statistics

Growth characteristics were monitored in liquid medium in 100-well Honeycomb 2 plates (Oy Growth Curves Ab, Helsinki, Finland). Growth curves were taken by following the optical densities (O.D.) at 540nm in each well using the Bioscreen C Automated Microbiology Growth Analysis System (Oy Growth Curves Ab, Helsinki, Finland). Ten parallels of resuspended colonies for each plasmid-host combination were grown to saturation in ten wells at 37°C using continuous shaking in 200 μl LB medium supplemented with 25 ug/ml chloramphenicol, if required. The orf238 containing plasmids were grown in the presence and in the absence of IPTG. IPTG was added to induce the cultures at 0.35 O.D. in 0.6 mM final concentration. When testing the effect of pCTXVP60 on the growth of strains harboring various numbers of IS elements, the median O.D. value of the ten parallel cultures corresponding to each strain (MG1655, MDS12, MDS30 and MDS42) was calculated and plotted for every time point.

RNA isolation from colonies and RT PCR

After transformation with pCTXVP60, small and large colonies were collected separately from the transformation plate. Cells from small and large colonies were resuspended into fresh minimal medium and the O.D. of the suspensions were adjusted to 1.0. RNA was isolated from the suspensions using the RNEasy RNA isolation kit with RNAProtect Bacteria Reagent (Qiagen, Düsseldorf, Germany), supplemented with lysozyme from Chicken Egg White (Sigma, St. Louis, MO, USA). DNA contamination was digested using DNAseI from Bovine Pancreas (Boehringer Mannheim, Germany) for 45 min at 37°C in the presence of 10 mM Tris (Sigma, St. Louis, MO, USA) and 2.25 mM MgCl₂ (Fermentas, Vilnius, Lithuania). The reaction was stopped by adding 2.5 mM EDTA (Merck, Darmstadt, Germany) and incubation at 65°C for 10 min. cDNA was synthesized from purified- and DNase-treated RNA samples using the Transcriptor First Strand cDNA Synthesis Kit (Roche, Mannheim, Germany) with random hexanucleotide primers, according to manufacturer's instructions. The cDNA samples were used as templates in the following PCR using ORFstart and ORFstop primers, specific to the previously identified small ORF region of the ctxvp60 transcript. Primers groL1 and groL2 were used as positive controls.

Cell imaging

Cells were incubated 30 mins. at 37°C with 10 μM nonyl acridine orange (NAO) and 10 μg/ml Ethidium bromide (EtBr). DNA staining was performed using 500 ng/ml 4',6-diamidino-2-phenylindole (DAPI) for 10 minutes after 70% ethanol fixation (30 mins) and PBS washes. Two percent agarose gel blocks (1 cm × 1 cm) were used to immobilize the cells on cover slips before microscopy observation. NAO, EtBr and DAPI were from Invitrogen, Carlsbad, CA, USA. Confocal laser scanning microscopy was performed using Olympus Fluoview FV1000 confocal laser scanning microscope (Olympus Life Science Europa GmbH, Hamburg, Germany) equipped with UPLSAPO 60× (oil, NA:1.35) objective. Using Olympus Fluoview software (version 1.7.2.2), DAPI images were pseudo colored red.

Bioinformatic analysis of E. coli IS sequences in shotgun data

Shotgun sequencing data from early genome sequencing projects (Sanger method), where clone libraries of various genomes were constructed and maintained in E. coli host, provide a large dataset for checking the frequency of host IS contamination of the cloned sequences. Sequence reads were fetched from the NCBI Trace Archive. Bacterial and archaeal genomes were analyzed when individual read sequences, assembled sequence, annotations, and experimental notes were available. We built two BLAST databases for each analyzed organism, one from the read sequences, and the other from the assembled genome. A representative set of E. coli K12 MG1655 (U00096) IS elements was compiled, and expanded by an IS10 sequence from DH10B (NC_010473). Database queries were done by nucleotide level BLAST searches. To produce a conservative estimate for the number of insertion events, stringent parameter settings (expectation value 0.1, minimum score 75, minimum run length 40 nucleotides) were used. Reads from possibly non-cloning sequencing, with "finishing", "primerwalk", "454" annotations were carefully removed.

The first pass of BLAST used IS sequence baits against shotgun read databases. This round of BLAST caught reads that contain some E. coli IS-like sequences. With these reads as baits, a second round of BLAST mapped the genomic sequence parts of the reads onto the assembled genomes. Using IS baits, genome segments with matches to E. coli IS sequences were identified next. Potentially, these segments contain endogenous IS elements (and not elements acquired in the cloning process), and were excluded from further analysis. Using the remaining segments as queries, the read databases were searched, and all the reads matching to IS-acquiring segments were identified. Finally, reads collected at this pass were BLAST-ed against the assembled genomes, to find and retain only the reciprocal best hits. This back-and-forth BLAST-ing identified genome segments from which some cloning inserts acquired E. coli IS elements during the shotgun sequencing process. Also, the procedure yielded the reads that genuinely map to these segments, including those that contain acquired E. coli IS sequences. To find the functions of the genes encoded in these segments, GenBank annotations were parsed. The relationships of reads, IS elements, genome segments and gene annotations were overlaid in simple diagrams.

Supporting online material. Nucleotide sequences of pCTXVP60, pCTXVP60frameshift, pSG1144-orf238, pSG1144-ctxvp60, pSG1144-ctxvp60opt, and pSG1144-ctxvp60dezopt, codon use of ctxvp60 opt and ctxvp60 dezopt, as well as data related to the structure prediction of ORF238 can be found at http://www.brc.hu/genome, and at the end of this article (Additional files 1, 2, 3 and 4).

Instability of pCTXVP60 clones propagated in wt E. coli MG1655

In an earlier study [24], it was found that an expression construct of a fusion gene, coding for adjuvant CTX B subunit fused to VP60, could not be stably propagated in MG1655. Upon transformation by plasmid pCTXVP60, heterogeneous colonies of transformants were obtained on agar plates (Fig. 1A). While the majority of the colonies were small, slow-growing, and translucent, a few percent showed normal growth and appearance, as compared to control pCTX clones (not shown). Upon prolonged incubation, eventually all small colonies started to develop sectors of normal growth. A preliminary analysis of the cells from normal, healthy colonies showed that the plasmids recovered had altered restriction digestion patterns, primarily due to IS element insertions in the fusion gene [24]. An extended analysis of 100 colonies by PCR spanning the fusion gene confirmed the extreme instability of the plasmid. A vast majority, 92%, of the large colonies carried plasmids with IS insertions, while 8% displayed deletions or no alteration detectable by PCR. Insertion events were due to IS1, IS3, and IS5 translocations, as detected by using IS-specific PCR primers. Insertion sites were determined by sequencing 12 clones. All IS insertions occurred in the 5' third of the fusion gene, in the ctx part or near the 5' end of vp60. Results of the analysis are summarized in Fig. 2.

Similar heterogeneity of pCTXVP60 transformants was observed in other regular E. coli hosts, including DH5α, DH10B or C600. In contrast, transformants of IS-less MDS42 [24] displayed a nearly uniform morphology (Fig. 1B) on LB plates. Colonies generally displayed moderately retarded growth, and only < 0.1% of the colonies was larger, showing normal growth. Generally, plasmid DNA recovered from cultures using MDS42 host yielded unaltered restriction digestion pattern and nucleotide sequence (data not shown).

Growth retardation effect of pCTXVP60 is due to an artificial byproduct, a Leu-rich ORF

The synthetic fusion gene ctxvp60 contains a large number of rare Arg ( 1 AGG, 21 AGA) codons. It was assumed that translation of ctxvp60 exerts a toxic effect on the cell by exhausting the rare tRNA_Arg pool and thus rendering it unavailable for the synthesis of essential proteins. To test this hypothesis, two new versions of ctxvp60 were synthesized. Version ctxvp60 opt was codon-optimized for E. coli and carried no rare codons, while ctxvp60 dezopt contained the rare AGG codons for all the 22 arginines (Additional files 2 and 3). Both versions were cloned in pSG1144 (Additional file 1), using MG1655-T7 host. Surprisingly, induced expression of both constructs, optimized and deoptimized, had only a minor negative effect on the growth of the cell, comparable to that seen with the vector plasmid (Fig. 3).

On the other hand, a frame-shift mutation of ctxvp60, introduced in a position near the 5' end of the vp60 part (Fig. 2; Additional file 1), did not relieve the growth retardation effect of the plasmid (data not shown). All these results indicate that neither the CTXVP60 protein per se, nor the supposed rare tRNA depletion phenomenon is the cause of the toxic effect of pCTXVP60.

A thorough inspection of the original ctxvp60 gene revealed the presence of a 238 aa ORF (ORF238) out of the original reading frame. ORF238 spans the joint between ctx and vp60 and extends into the 5' part of vp60 (for coordinates, see Fig. 2). Translation of ORF238 results in an extremely Leu-rich (102 Leu residues) protein (Additional file 4). The hydrophobic nature of the protein and its four putative transmembrane domains (predicted with HMMTOP [29] and DAS-TMfilter [30, 31]) strongly suggest that it integrates in the cell membranes, and might be the cause of the toxic effect (Additional file 4).

To test the hypothesis that ORF238 was the culprit of the growth defect, it was cloned as an inducible construct in plasmid pSG1144. Upon induction, expression of ORF238 resulted in a phenotype identical to that of the original ctxvp60 cloned in the same plasmid. Growth retardation of the culture, due to induction of ORF238, is demonstrated in Fig. 3.

Confocal laser scanning microscopy imaging of the cells expressing either CTXVP60 or ORF238 confirmed its inhibitory role. Nucleic acid staining (Ethidium bromide or DAPI) combined with differential interference contrast (DIC) imaging and membrane staining (nonyl acridine orange) revealed that cell division is impaired, resulting in slow-growing, abnormally long cells with aberrant nucleoids (Fig. 4B, E, F). Untransformed (MG1655) or uninduced ORF238-harboring cells, on the other hand, displayed normal phenotype and size distribution (Fig 4A, C, D).

ISes integrated in ctxvp60 block transcription of the Leu-rich ORF

All IS integrations, found in pCTXVP60, occurred either upstream or in the 5' end region of ORF238. It was assumed that ISes landing in this region relieve the cell from stress by blocking transcription of ORF238. To prove this, we showed that RNA prepared from cells harboring pCTXVP60 contained transcripts of the toxic gene detectable by RT-PCR. However, when picking a mutant with an IS1 or IS5 inserted into the 5' end of ctxvp60, transcripts of downstream sequences could not be detected (Fig. 5).

Growth dynamics and evolution of strains harboring pCTXVP60 is related to the number of genomic ISes

Growth characteristics of transformants of toxic and mutant plasmids were monitored in liquid medium. Ten parallels originating from 10 colonies were grown for each plasmid-host combination, and automatic O.D.-readings were taken in a Bioscreen C instrument (Fig. 6).

Initial growth of MG1655/pCTXVP60 stopped at ~O.D. = 0.2-0.5. However, after prolonged incubation, individual cultures resumed growth at various time points, and reached a final density comparable to the MG1655/pCTX control (Fig. 6A, B). Plasmids isolated from these outgrown cultures almost invariably contained ISes inserted in pCTXVP60. Reinoculation of the outgrown cultures into fresh medium resulted in uninterrupted, normal growth (Fig. 6C). The results were consistent with the observed growth characteristics of transformants on agar plates: i) pCTXVP60 causes growth retardation, ii) in a stochastic manner, a fraction of the cells picks up an IS insertion in ctxvp60, due to the mobility of genomic ISes, iii) cells harboring an IS-inactivated ORF238 resume normal growth and quickly become dominant in the culture.

In contrast, growth of MDS42/pCTXVP60 displayed a nearly uniform pattern. Initial growth of the cultures stopped at ~O.D. = 0.3, and only one culture of 10 parallels grew eventually to high density (Fig. 6D). Plasmid isolated from this dense culture revealed a deletion in ctxvp60. Results indicate that the primary route to inactivation of the toxic gene is via IS translocation.

Next, strains representing the various stages of the genome reduction process (MG1655, MDS12, MDS30, MDS42), and thus harboring various numbers of ISes (44, 25, 5, and 0, respectively) were transformed with pCTXVP60. The median growth curves observed for the transformants seemed to support the notion that the time needed for evolution of a non-toxic plasmid variant in the culture correlated with the number of ISes in the host genome (Fig. 6E). Even a single IS1, present in the host genome, had a marked effect on growth dynamics, and accelerated the evolution of fast-growing variants (Fig. 6F), as compared to the IS-less, essentially isogenic host (MDS42) (Fig. 6D).

Bioinformatic analysis of E. coli IS sequences in shotgun sequencing data

To test whether IS-mediated instability of cloned sequences could be more wide-spread than anticipated, a bioinformatic analysis of raw data from early genome sequencing projects was performed. Protocols for shotgun genome sequencing by the Sanger method started by the insertion of fragments from a target genome into a plasmid vector, followed by transformation into E. coli cells. Next, transformed cells were grown first on solid medium, then in liquid cultures, prior to processing. Enriched target fragments were then sequenced, and overlapping sequence reads were assembled into contigs. The middle, cell growing steps can be viewed upon as some biological test of E. coli 's tolerance to the introduction of foreign DNA, consequently, read data from the thousands of clones can be mined for signs of the host's reaction to foreign DNA segments [32]. In this study we searched the read sets for the appearance of E. coli IS elements in raw (prior to their assembly into target genome sequences) sequence reads.

A typical bacterial shotgun library for Sanger sequencing would comprise 2-4 × 10⁴ clones, which are grown for about 20 generations on solid medium, followed by another 10 generations of growth in liquid cultures, prior to processing. Considering the typical spontaneous, combined transposition rate of 10^-8/gene/generation of resident ISes [9], the chances of sequencing a spontaneously arisen insertion mutant is about 1.2 × 10^-2 (=30 × 4 × 10⁴ × 10^-8). This means that approximately one in 100 sequencing projects would yield a gene interrupted by an IS of the cloning host. Higher incidence of host ISes in the sequence reads would likely indicate selection events preferring the IS-inserted clones during cultivation.

Data from 295 shotgun genome sequencing projects were downloaded and analyzed. The sets typically contain 50-70 thousand reads. Of the ~18 million reads, a total of 22 thousand match some E. coli IS sequence within the strict expectation, score, and run length thresholds. While 166 sets contain one or more reads with IS sequences, 129 sets are free of such reads. Of the 295 analyzed organisms 62 have complete (assembled) genome sequence. The number of organisms possessing both IS- containing reads and an assembled genome (intersection of the 166 set and the 62 set) is 30. In some IS-containing reads chromosomal sequence was not detectable, while in others IS elements seemed to have cloning vector origins (e.g., are spliced with E. coli lacZ), these have been weeded out. To avoid the difficulties of distinguishing true cloning related E. coli IS insertions from genomic self-rearrangements, organisms which have their own IS elements were excluded. After these filtering steps 14 shotgun sequencing sets remain, which possess a total of 109 IS-containing reads. IS10 and IS1 are the dominating invaders (they show up 68 and 30 times, respectively), but six other IS types were also found (IS2, IS3, IS4, IS5, IS150, IS186). Examples of IS elements appearing in sequence reads are shown in Fig. 7 and Additional file 5.