Abstract
Duplications are often attributed to “unequal recombination” between separated, directly repeated sequence elements (>100 bp), events that leave a recombinant element at the duplication junction. However, in the bacterial chromosome, duplications form at high rates (10−3–10−5/cell/division) even without recombination (RecA). Here we describe 1800 spontaneous lac duplications trapped nonselectively on the low-copy F′128 plasmid, where lac is flanked by direct repeats of the transposable element IS3 (1258 bp) and by numerous quasipalindromic REP elements (30 bp). Duplications form at a high rate (10−4/cell/division) that is reduced only about 11-fold in the absence of RecA. With and without RecA, most duplications arise by recombination between IS3 elements (97%). Formation of these duplications is stimulated by IS3 transposase (Tnp) and plasmid transfer functions (TraI). Three duplication pathways are proposed. First, plasmid dimers form at a high rate stimulated by RecA and are then modified by deletions between IS3 elements (resolution) that leave a monomeric plasmid with an IS3-flanked lac duplication. Second, without RecA, duplications occur by single-strand annealing of DNA ends generated in different sister chromosomes after transposase nicks DNA near participating IS3 elements. The absence of RecA may stimulate annealing by allowing chromosome breaks to persist. Third, a minority of lac duplications (3%) have short (0–36 bp) junction sequences (SJ), some of which are located within REP elements. These duplication types form without RecA, Tnp, or Tra by a pathway in which the palindromic junctions of a tandem inversion duplication (TID) may stimulate deletions that leave the final duplication.
GENE duplications are among the first mutations for which a physical basis was known. The Bar–Eye mutation of Drosophila was discovered in 1914 (Tice 1914) and shown genetically and cytologically in 1936 to be a duplication (Bridges 1936; Muller 1936)—well before the Watson–Crick model for DNA. The attached-X mutation was discovered even earlier and is essentially an inversion duplication (Morgan 1925). Recently, duplications have been shown to be exceedingly common polymorphisms in human populations (Conrad et al. 2010) and somatic amplifications play an important role in origins of malignancy and resistance to cancer therapies (Lu et al. 2008; Zhang et al. 2009; Conrad et al. 2010). Gene duplications also play a major role in the evolution of new genes (Ohno 1970; Bergthorsson et al. 2007). Despite the high frequency and biological importance of duplications, the process by which they form remains uncertain.
Distinct models have been suggested for duplication formation in various genetic systems. In Drosophila, duplications often have transposable elements between copies (Green 1985; Tsubota et al. 1989). In humans, duplications most commonly arise between extensive sequence repeats (Redon et al. 2006) and are sometimes promoted by palindromic sequence elements (Conrad et al. 2010). A major question is how often duplications form by unequal-crossing-over model and how often by one of the multiple alternative pathways.
In bacteria, several properties of duplications are apparent:
Duplications of any particular gene form at extremely high rates (10−3–10−5/cell/division) (Anderson and Roth 1981; Reams et al. 2010).
In unselected populations, the frequency of duplications and higher amplifications rapidly approach a steady state due to a balance between their high formation rate and their frequent loss and fitness cost (Reams et al. 2010).
Duplications form most often between large direct sequence repeats >200 bp (Anderson and Roth 1981; Flores et al. 2000; Goldfless et al. 2006).
Duplications often arise by recombination between directly oriented preexisting copies of transposable elements (Morris and Rownd 1974; Clugston and Jessop 1991; Haack and Roth 1995; Ogawa and Miyashita 1995; Reams and Neidle 2003; Reams and Neidle 2004b; Nicoloff et al. 2007).
Even when substantial flanking repeats are involved, the rate of chromosomal duplication formation depends only weakly on RecA-mediated homologous recombination (Reams et al. 2010), which seems contrary to the idea of unequal crossing over.
Short adjacent repeats in high-copy plasmids can recombine to form deletions and duplications without RecA (Morag et al. 1999; Bzymek and Lovett 2001; Reams and Neidle 2004a; Goldfless et al. 2006; Gore et al. 2006). These rare exchanges form by reciprocal exchanges at stalled replication forks leading to formation of a plasmid dimer with a deletion in one copy and a corresponding duplication in the other (Dianov et al. 1991; Mazin et al. 1991; Lovett et al. 1993).
Following prolonged growth under selection for increased gene copy number, amplifications (more than two copies) whose structure differs from the duplication types that predominate in unselected cultures are recovered (Kugelberg et al. 2006; Kugelberg et al. 2010). Selected amplifications have tiny junction sequences (0–10 bp) that seem unlikely to form by recombination (Reams and Neidle 2004a; Kugelberg et al. 2006, 2010).
Some selected amplifications (20%) have repeated tandem inversion duplications (TID), in which a parental sequence element (e.g., ABCD) is repeated three times in alternating orientation ABCD–D′C′B′A′–ABCD and amplified by exchanges between flanking direct repeats. The parent sequence ABCD can be >10 kb and the initial TID junctions are flanked by extensive perfectly complementary sequences that can be rendered asymmetric (ABCD–C′B′A′–BCD) by junction deletions (Kugelberg et al. 2010; Lin et al. 2011).
In the experiments presented here, duplications of the lacZ gene were isolated nonselectively on a low-copy bacterial plasmid (F′128) with several features that contribute to duplication formation. These include insertion sequence (IS) elements, a conjugative replication origin, and multiple palindromic repetitive extragenic palindromic (REP) elements. Pathways of duplication formation are inferred from the frequency of duplication types that appear in various mutant backgrounds. We first describe formal aspects of gene duplication that are relevant to these results.
Mechanisms of duplication formation
These mechanisms are intended as background to the experiments described here and will be referred to in describing this work.
Unequal crossing over:
Most duplications are thought to form by a single “unequal crossover”—i.e., homologous recombination between separated direct sequence repeats. Figure 1 diagrams this event as it might occur between bacterial sister chromosomes. Recombination creates a tandem head-to-tail duplication in one sister and, if the exchange is reciprocal, a corresponding deletion in the other. In either case, a copy of the recombining sequence is left at the rearrangement join point. The idea that duplications arise by unequal homologous recombination is called into question by the fact that bacterial duplications are frequent even in recombination deficient (recA) mutant strains and often have junction sequences that seem too short to support standard recombination.
Duplication by unequal crossing over. This conventional view of duplication formation involves standard recombination events between separated sequence elements embedded in nonhomologous surrounding sequence (top). Recombination between the elements generates a duplication with a copy of the recombining element at the junction between duplicated regions and at the site of a corresponding deletion (middle). Regardless of how a duplication forms, the repeated segment can be lost or further amplified (bottom) by standard RecA-dependent recombination between nonallelic copies at any point of their extensive shared homology (see center).
Duplication formation without RecA was observed in the bacterial chromosome (Anderson and Roth 1978; Reams et al. 2010), but has been most extensively studied by selecting rare deletions between adjacent short repeats (20–700 bp) in small high-copy plasmids. These deletions were recovered in plasmid dimers—duplications of the plasmid whose second copy had a duplication or triplication of the parental repeats. This suggested that deletions and duplications form by unequal reciprocal exchanges between sister chromosomes (Dianov et al. 1991; Mazin et al. 1991; Lovett et al. 1993). The independence of RecA appears to reflect close spacing of short repeats that allows annealing between single DNA strands present in both sister chromosomes near a stalled replication fork (Lovett 2004). When repeats are farther apart, as in standard chromosomal duplication events, the process of duplication may be somewhat different, although the low dependence on RecA has been noted in the bacterial chromosome (Reams et al. 2010). The duplicated regions described here are 20–131 kb in size, most of which arise between repeats >1 kb.
Unequal crossing over is also believed to cause loss or higher amplification of a duplicated segment. These secondary events depend heavily on RecA (Lin et al. 1984; Goldberg and Mekalanos 1986; Petes and Hill 1988; Galitski and Roth 1997; Poteete 2009). See Figure 1.
Role of fork collapse in gene duplication:
In considering Figure 1, a mechanistic problem is the source of DNA nicks or breaks that might initiate an exchange between short separated sequences. Most recombination events in unperturbed cells are thought to occur during repair of spontaneously collapsed replication forks (Kuzminov 1995). Replication of a nicked DNA duplex can lead to fork collapse leaving a double-strand break in one sister chromosome. Fork stalling can also leave single-strand ends that are prone to annealing. Mechanisms have been proposed for repair of forks using the strand-exchange protein RecA or by pairing of single-strand regions as described later (Kuzminov 1995). The problem is how to obtain discontinuities near the specific sequence repeats.
Duplication by transposition:
The frequent involvement of mobile elements in duplication formation suggests a role for transposition. Duplications formed directly by replicative transposition have a copy of the transposable element at the join point, but are unique in having a novel sequence junction between one end of the element and the target sequence (indicated by a circle in Figure 2). A full replicative transposition would simultaneously generate a corresponding deletion allele with another novel junction.
Duplication by replicative transposition. Replicative transposition can directly create a duplication with a copy of the transposable element at the rearrangement join point. The transposition event generates a novel sequence (circled above) at the juxtaposition of one end of the element and the transpositional target site. This event simultaneously generates a deletion that will appear in a different cell and have a novel sequence at the other end of the junction element. Duplications formed in this way have been observed in this system, but do not explain the involvement of transposase in the duplications described here. Conservative transposition can insert an element copy at a novel site for use by recombinational duplication, but we have not observed such events.
Replicative transpositions with one novel sequence junction (Figure 2) underlie some gene amplifications in Acinetobacter (Reams and Neidle 2003; Reams and Neidle 2004a) and on the E. coli F′ plasmid (Kugelberg et al. 2006; Kugelberg et al. 2010), but they are rare and have been detected only after long-term selection for increased copy number. Extremely rare amplified duplications have been seen with novel sequence junctions at both ends of a transposable element at the duplication junction (Reams and Neidle 2003; E. Kofoid, unpublished results). It is not clear how these duplications form, but their structure suggests that two transposition events were required. Conservative transposition events can in principle generate a duplication by providing separated sequences that can then serve as homology for standard unequal recombination (described in Figure 1), but we have never seen duplications that formed by this concerted transposition/recombination process. Formation of the duplication types described here is stimulated by a transposase, but involves no act of transposition (generates no novel sequence junctions).
Single-strand annealing:
Formation of a duplication by single-strand annealing requires two simultaneous breaks in different chromosomes near repeated sequences that flank the region to be duplicated (Figure 3). Resection can reveal complementary single-strand sequences that pair to form the duplication. This general annealing process has been demonstrated in vivo and in vitro and is independent of RecA (Mortensen et al. 1996; Bzymek and Lovett 2001; Liu et al. 2011). Duplication by annealing seems unlikely because two breaks must occur in different chromosomes, each near a recombining sequence repeat. The likelihood of annealing would increase if some feature of the repeated sequence increased the frequency of adjacent nicks that can be converted to double-strand breaks. This could happen at heavily transcribed genes (e.g., rrn) if the nontemplate strand is prone to breakage. It could also happen at transposable elements, whose ends can be nicked by transposase as discussed below.
Duplication by single-strand annealing. In this diagram, breaks form at separated sites in different sister chromosomes behind a single replication fork. Breaks may form when a replication fork encounters a nick in its template or by conversion of a simple nick to a double-strand break. In the diagram, these breaks form near sequence repeats. This nicking could be caused by transposase or by fork stalling. The 5′ strands at these breaks are resected, leaving 3′-single-strand extensions that could permit pairing of the repeats. Repair of the paired structure by nucleases and synthesis can lead to a duplication. Simultaneous single strands may coexist at a single collapsed replication fork and allow duplication between nearby repeats as diagramed previously (Lovett et al. 1993) or may form adjacent to transposable elements that are subject to transposase nicking.
Strand slippage:
Strand slippage was suggested to explain addition or removal of a base or two in formation of +1 and −1 frameshift mutations (Streisinger et al. 1966). In principle, the mechanism could also explain deletions and duplications between any pair of directly repeated sequences (Farabaugh et al. 1978). Slippage could underlie the tendency of large palindromic sequences to be removed during replication (Trinh and Sinden 1991; Leach 1994; Lovett 2004). If large single-strand regions are generated during replication, any included palindromic region might form a hairpin structure in either a nascent or template strand and favor slipped pairing of nearby regions to stimulate duplication or deletion of the palindromic region. Mechanisms such as this are probably responsible for the frequent duplication and deletion of small regions with palindromes (Sinden et al. 1991; Leach 1994; Bzymek and Lovett 2001; Lovett 2004; Gore et al. 2006; Quiñones-Soto and Roth 2011), but seem unlikely to explain rearrangements of extensive chromosomal regions, as seen in the duplications described here.
Tandem inversion duplication:
This duplication type was identified in Lac+ revertants isolated after prolonged selection of a leaky lac mutant for improved growth on lactose (Kugelberg et al. 2006; Kugelberg et al. 2010; Lin et al. 2011). Roughly 20% of the Lac+ revertants with an amplification of the region ABCD had multiple tandem copies of a basic structure that could be described as: ABCD–C′B′A′–BCD. This unit can be amplified by RecA-dependent unequal exchanges between the flanking tandem repeats (BCD). Note that the observed repeat junction sequences are asymmetric (nonpalindromic). It was proposed that the asymmetric structure (aTID) arises when deletions modify an initial symmetrical structure (sTID: ABCD–D′C′B′A′–ABCD) (see Figure 4A). Formation of this structure with perfectly paired adjacent sequences at each junction can form by snap-back repair from an initial short quasipalindromic sequence (Kugelberg et al. 2010). The palindrome stimulates deletions that generate the asymmetric (aTID) and tandem duplications with short junction sequences (see Figure 4B). Duplications of the initial symmetrical sTID type have been observed in yeast following extensive growth under selection for copy-number increase (Araya et al. 2010). A TID amplification with one symmetrical and one modified junction has been reported in bacteria (Kugelberg et al. 2010).
Tandem inversion duplications and their modification. Amplifications of this type were recovered in bacteria and yeast after prolonged growth under selection for additional gene copies (Araya et al. 2010; Kugelberg et al. 2010) and are also found in certain cancer cells, especially following exposure to cancer chemotherapy. Their formation is thought to be initiated by short quasi-palindromic structures. (A) One model for TID formation that relies on initiating repair replication at snap-back structures and switching templates to produce symmetrical triplications whose junctions have extended complementary palindromic sequences. (B) Deletions that modify a sTID to generate asymmetric junctions or simple tandem duplications.
A mechanism for forming the symmetrical precursor (sTID) is described in Figure 4A (Kugelberg et al. 2010). In this model, duplication formation is initiated by a quasipalindromic sequence near a strand end that snaps back to prime repair synthesis using itself as template. A second template switch generates a branched structure that resolves to leave the symmetrical TID (sTID). A correlation between short palindromes and duplications has been noted in several systems (Mehan et al. 2004; Rattray et al. 2005; Narayanan et al. 2006). Alternative models for TID formation have been suggested (Hastings et al. 2009; Brewer et al. 2011). In bacteria, the asymmetric TID structures found after growth under selection have been attributed to secondary modifications of sTIDs that form during growth under selection (Kugelberg et al. 2010).
Additional mechanisms of duplication and amplification:
Several alternative duplication formation models have been proposed on the basis of systems in which amplifications arise under selection by a series of events. These include models in which palindromic sequences lead to snap-back replication to form dicentric chromosomes that break at cell division to produce a terminal inversion duplication (Gordenin et al. 1993; Narayanan et al. 2006). In addition, the “onion skin” model suggests amplification by nonhomologous end joining following local overreplication in chromosomes with multiple replication origins (Botchan et al. 1979; Spradling 1981). Rolling circle replication has been suggested for achieving sudden amplification (Petit et al. 1992; Roth et al. 1996; Hastings and Rosenberg 2002). These models seem unlikely to contribute to duplication formation in the system described here.
The F′128 lac plasmid
The duplications described here arise on the E. coli F′128 plasmid carried by S. enterica. Figure 5 diagrams the genome of this plasmid with some relevant sequence features (Kofoid et al. 2003). The lac operon is flanked by identical copies of the IS3 insertion sequences IS3A and IS3C, each 1258 bp separated by 131 kb. A third IS3 copy (IS3B) differs from the others by seven base substitutions (four amino acid substitutions in the transposase) that appear to eliminate transposase activity and prevent recombination between IS3B and the other IS3 copies. A plasmid operon encoding conjugational transfer functions (tra) encodes the TraI protein, which nicks plasmid DNA at the transfer origin (OriT) and displaces a 5′-ended single strand for transfer (Traxler and Minkley 1987; Frost et al. 1994). These nicks stimulate RecA-dependent recombination on the plasmid from 10- to 50-fold (Seifert and Porter 1984; Syvanen et al. 1986). The F′128 plasmid is rich in quasipalindromic REP elements (Stern et al. 1984; Gilson et al. 1990; Lupski and Weinstock 1992), which are found as clusters of paired inverse-order repeats (30 bp) as diagramed in Figure 5. Duplication of the plasmid lac region are trapped by a method that detects any cell with two copies of lacZ regardless of where the duplications might end (see Materials and Methods).
Features of the F′128 plasmid. The thick line represents the portion of the plasmid derived from the F-plasmid. The lac-containing segment between IS3A and IS3C is derived from the E. coli chromosome. Elements IS3A and IS3C are identical and differ from IS3B by seven base substitutions. REP elements are quasipalindromic repeated sequences that fall into three related families (Y, Z1, and Z2). These elements are typically found as clusters with multiple pairs of inversely oriented REP sequences.
Formation of duplications by multistep pathways on the F′ plasmid
The idea of multistep duplication formation was first suggested by the lac amplifications isolated on F′128 following prolonged growth under selection for improved growth on lactose (Andersson et al. 1998). Selected amplifications were of two types. One type had tandem repeats of a small region (10–20 kb) with short (10 bp) junction sequences. The other carried multiple copies of a TID (described above) with lac sequences in alternating orientation (Kugelberg et al. 2006). In contrast, the lac duplications trapped in unselected populations show duplicated regions that are generally larger than those amplified under selection, with most (97%) having a copy of IS3 at the junction. (See Appendix for a comparison.)
We describe a set of lac duplications that arose in several genetic backgrounds during nonselective growth. The formation rates of the several duplication types are interpreted in terms of several pathways for duplication formation, some of which have multiple steps, even though they arose with no deliberately applied selection. These pathways can be distinguished by their dependence on RecA. We propose that the RecA-dependent pathway involves formation of a plasmid dimer, which is remodeled by deletions stimulated by the transposase activity of IS3. The RecA-independent duplications may form either by annealing-mediated exchanges between sister copies of a monomeric plasmid or by remodeling an initial sTID duplication formed in a single plasmid.
Materials and Methods
Strains and media
All strains were derivatives of Salmonella enterica (Typhimurium) strain LT2 and are listed in Table 1. Rich medium was Luria broth (LB) with antibiotics as described below. The chromogenic β-galactosidase substrate, 5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside (X-gal), was obtained from Diagnostic Chemicals (Oxford, CT) and used in plates at 40 μg/ml.
Mutations that inactivated the IS3 transposase were constructed by linear transformation as described by Yu et al. (2003). To construct these strains, a cassette (Sac-Kan) encoding sucrose sensitivity (SacB) and kanamycin resistance (Npt1) was inserted into the initiation codon of the transposase gene, selecting kanamycin resistance (Lawes and Maloy 1995; Muro-Pastor and Maloy 1995). This cassette was then replaced with a mutant version of the initiation codon region with AUA in place of AUG by selection for resistance to sucrose. The Sac-Kan cassette was originally amplified from plasmid pPC217 with primers: TP2183 5′ GGCTAAGTGAGTAAACTCTCAGTCAGAGGTGACTCACATATGGCGGCCGCTCTAGAACTAG 3′ and TP2184 5′ GCTGTTTACGGGGTTTTTTACTGGTTGATACTGTTTTTGTCGACGGTATCGATAAGCTTGA 3′. Strains carrying the Sac-Kan cassette within IS3 were transformed with single-stranded DNA that included the point mutation: 5′ GTAAACTCTCAGTCAGAGGTGACTCACATAACAAAAACAGTATCAACCAGTAAAAAACCC 3′. Sucrose-resistant transformants that had lost kanamycin resistance were verified by sequencing of PCR fragments.
Trapping duplications formed in overnight cultures
Three methods for measuring duplication frequencies in unselected overnight cultures were described previously (Reams et al. 2010). These are: (1) Ka-Kan, (2) T-Recs, and (3) the standard transduction-based assays. The Ka-Kan assay was used here and its results were confirmed by both of the other two assays (Reams et al. 2010).
Duplication frequency measurements were confirmed by quantitative PCR to estimate the frequency of cells with the IS3A/C duplication joint point. One primer directed replication counterclockwise across IS3C and the other directed synthesis clockwise across IS3A as the plasmid is drawn in Figure 5. These primers were: TP1040 5′ TGTGAATATGCTGACATGCC 3′ and TP1041 5′ CGGCGAATGGCTGGGATA 3′, respectively. Since these primers direct divergent replication in the parent plasmid, a PCR product will be amplified only in strains with an IS3A/C duplication junction. To estimate the frequency of cells with an IS3A/C duplication, multiple independent cultures of various tester strains were grown overnight in LB. Each culture was diluted serially and a PCR reaction was run on each dilution. All PCR products were separated on a 0.7% agarose gel using standard electrophoresis procedures and visualized after staining with ethidium bromide. The frequency of cells with an IS3A/C duplication was estimated by determining the greatest dilution that still generated a detectable junction PCR product.
In the PCR duplication assay, it was possible that the join point fragment could be artifactually generated by hybridization between single strands of the IS3A and C sequence produced by the PCR process. This possibility was tested using as template a mixture of DNA from two strains, each with only one of the IS3 elements, IS3A or IS3C, and therefore unable to form a duplication. This template mixture did not produce the junction point PCR fragment under the PCR conditions used for our tests (<104 cells per assay). The artifactual fragment was seen only when 100-fold more cells were tested.
Construction of a plasmid that overproduces IS3 transposase
The IS3-transposase gene was PCR amplified with an accurate polymerase (Pfx) and cloned into a pBAD-TOPO TA vector (Invitrogen, San Diego, CA) using the following primers: TP2425 5′ TAACACGCGGCTAAGTGAGTAAA 3′ and TP2426 5′ GCCTAAGCGAGGTTCTTGTT 3′. This placed expression of the IS3-transposase gene under control of the arabinose-regulated pBAD promoter where it could be induced by 0.06% l-arabinose. As a negative control, the mutant IS3-transposase gene whose initiation codon is changed to ATA was inserted into pBAD-TOPO TA vector using the same primers. Overexpression of the normal transposase gene (but not the inactive mutant gene) showed an effect on duplication and deletion formation.
Determination of duplication junction sequences
Junction sequences were amplified using combinations of primer pools, as previously described (Kugelberg et al. 2006). All primers directed synthesis away from the lac operon. In some pools, all primers directed synthesis clockwise away from lac and, in other pools all primers directed synthesis counterclockwise away from lac. Combinations of a clockwise and a counterclockwise pool were used to test each duplication strain. No single pool (same direction) or pair of pools (divergent replication) is expected to prime fragment amplification from the parent template. However, cells with a lac duplication allowed divergent primer pairs to converge across a duplication join point. For each pool combination that produced a fragment, the responsible primer pair was identified, and its product was sequenced. All trapped duplications were initially screened for the most common duplication type, with an IS3A/C junction, using the primer pools. Duplications yielding a PCR product with a size consistent with that of the IS3A/C junction fragment were scored as such. In a set of 100 independent duplications for which pools generated an appropriate-sized fragment, the IS3A/C junction was confirmed using two primers most proximal to the IS3A/C junction: TP1040 and TP1041. Thus, the fragment generated by the pools is reliable evidence of an IS3A/C duplication. Other duplication types generated amplified fragments of distinct sizes that were characterized by sequencing.
Determining rates of duplication formation (kF) and loss (kL)
Rates of duplication formation were estimated after 33 generations of growth as described previously (Reams et al. 2010). In determining duplication rate, observed increases in duplication frequency with time was corrected for both duplication loss and fitness costs using a spreadsheet simulation described previously (Reams et al. 2010).
Fitness cost measurements
The fitness cost of duplications was determined by comparing the growth rate of the parent haploid strain with that of a strain harboring a lac duplication as described previously (Reams et al. 2010). Similarly, the fitness advantage of cells with a deletion on the plasmid was measured by comparing the growth rates of the parent haploid strain with a strain harboring a lac deletion. To prevent duplication loss during the growth rate tests, all strains carried a recA mutation. This recA mutation reduced all growth rates slightly, but did not affect the relative fitness cost. Similarly, the fitness advantage of strains that lost a duplication was the same in both rec+ and recA strains. The small fitness changes due to lac duplications and deletions on plasmid F′128 were determined by direct growth competition between the haploid parent strain and either a duplication or deletion strain. Relative fitness was calculated using the spreadsheet simulation method described previously (Reams et al. 2010).
Measuring deletion formation rates
The rate of recombination between separated IS3 repeats in a single plasmid can be estimated as the rate at which the parent haploid plasmid loses the segment between IS3A and IS3C that includes the lac region. Deletion rates were estimated during nonselective growth (LB) of a lac+ strain with its IS3B locus replaced by a spectinomycin-resistance determinant to assure that only deletions between IS3A and C are being scored. Parallel cultures were initiated from single blue colonies on a LB plate containing spectinomycin and X-gal and grown by serial passage of saturated cultures by 1:10 dilutions in 3 ml LB. The increasing frequency of Lac− deletion mutants was determined at three time points (e.g., 66, 99, and 132 generations) by counting the number of white (Lac−) colonies on LB plates containing spectinomycin and X-gal. Spectinomycin was included to assure the strains maintained F′ and did not form a white colony by simply losing the plasmid. The number of generations was calculated by measuring viable cells at each time point and the total dilution factor. Deletion formation rates were calculated from the change in frequency of Lac+ cells using the spreadsheet simulation method (Reams et al. 2010) to correct for fitness differences between the parent and the deletion mutant. In recA+ strains, >99% (100/100) of Lac− cells carried a deletion between IS3A and IS3C; in recA mutant strains 71% of Lac− cells carried the IS3A–C deletion, which was used to correct results in estimating deletion rates.
Results
Detection of duplications
All lac duplications were isolated in nonselectively grown populations using the Ka-Kan trapping method described previously (Reams et al. 2010). Parent cells have a promoterless kanamycin resistance determinant (Ka) replacing (in inverse order) the lacI gene of an otherwise normal lac operon. These cells are phenotypically KanS and Lac+. Duplications of the lac region are detected by introducing a single-strand fragment that provides a promoter for the Ka gene and simultaneously deletes the lacZ gene. Cells without a duplication inherit this fragment to become kanamycin resistant (KanR Lac−) and form white colonies on rich medium with kanamycin and X-gal. Cells with a lac duplication become KanR but remain Lac+ and form blue colonies because of their extra copy of the lac region. That is, one copy retains the parental Ka, lac+ allele and the other acquires KanR and a lacZ deletion. The ratio of blue (duplication) to white (haploid) colonies gives the frequency of duplications in the recipient culture. This method detects lac duplications in RecA+ or RecA− cultures because the transformation is mediated by recombination functions of phage lambda (Red) that are induced immediately before plating on selective medium. Recombination mediated by the Red enzyme does not require RecA and does not contribute to duplication formation (Reams et al. 2010).
As described previously, the frequency of duplications in an unselected population comes to a steady state dictated by the formation rate, loss rate, and fitness cost of duplications (Reams et al. 2010). After 33 generations in rich medium, the lac duplication frequency reaches 3 × 10−3 (±0.5 × 10−3 SD), about two-thirds the steady-state level. The duplication formation rates were calculated on the basis of duplication frequency at 33 generations, corrected for fitness cost and loss rate (Reams et al. 2010). The formation rate of lac duplications is 3 × 10−4/cell/generation (±0.5 × 10−4 SD). Through the initial part of this article, duplication frequency at 33 generations (rather than rate) is used to compare formation of several duplication types. These duplication frequencies closely parallel formation rates because the sampling time is still within the period of initial linear duplication accumulation.
Characterizing trapped duplications
The junction sequences of trapped duplications were characterized using pools of PCR primers that directed transcription away from lac on the F′128 plasmid either clockwise or counterclockwise (Kugelberg et al. 2006). These primer pools detect the junction of virtually any tandem lac duplication on the F′128 plasmid. Over 1800 lac duplications were trapped from the parent or several derivative strains. Duplications fell into three classes that differ by the nature of their junction sequence. These types are the IS3, REP, and SJ (short junction) duplications described in Figure 6 and Table 2. Most duplications (97%) trapped in wild-type populations are IS3 mediated in the sense that their join point has a hybrid IS3A/C element (top duplication in Figure 6). These duplications form by an exchange between the identical 1258-bp IS3A and IS3C elements, separated by 131 kb and the largest sequence repeats flanking lac in F′128. A less common IS3-mediated duplication type (4%) arose between IS3C and the third IS3 element, IS3B, which differs from the other two IS3 elements by seven base substitutions, four of which cause amino acid substitutions within the transposase (Tnp) gene. The sequence changes in IS3B are presumably sufficient to reduce its ability to recombine with the other IS3 copies and may impair function of its transposase.
Structure of the duplication types. The haploid parent (top) has a lac operon flanked by copies of IS3 and several clusters of palindromic REP elements (see Figure 5). Duplications to form between separated short sequences and leave one copy of the sequence at the join point indicated by parentheses. While these junctions appear to have formed by unequal recombination, all duplication types can form in strains lacking RecA.
All of the other duplications have shorter junction sequences (< 37 bp). Most of these duplications are the REP type (2.7% of the total duplications) whose joint point sequence is derived from two flanking REP elements. REP elements are “repeated extragenic palindromic” sequences roughly 40 bp in size that are present in several hundred copies in the genomes of S. enterica and Escherichia coli (Lupski and Weinstock 1992; Bachellier et al. 1997, 1999). Clusters of REP sequences flank the lacZ operon of F′128 (Kofoid et al. 2003). The REP duplication junctions have 4- to 36-bp sequences shared with nonallelic REP elements that flank lac. Thus, the duplications appear to have formed by exchanges between two (usually nonidentical) REP elements located in direct order on opposite sides of lac. The various REP elements found at duplication endpoints are summarized in the Appendix. The small size of REP junction sequences makes it unlikely that these duplications formed by RecA-dependent homologous recombination. The regions duplicated by REP duplications had an average size of 26 kb, with a range of 17–70 kb.
The rest of the non-IS3 duplications (0.3% of total duplications) have short junctions (0–9 bp) not derived from REP elements and an average repeat size of 65 kb, (range of 21–108 kb). Surprisingly, many SJ duplications (10/28) had one end within a particular small region between the closely spaced REP elements R27 and R28 and the other end within 90 bp of REP element, R31. The 10 duplications at this hot spot had 0- to 2-bp junction sequences and included a region 17.2–17.5 kb. While 8 of these 10 duplications were unique, two independently isolated duplications were identical, carrying the same 2-bp junction sequence.
Homologous recombination (RecA) contributes to formation of IS3 duplications but not REP or SJ types
As seen on the left side of Figure 7, elimination of RecA reduced total duplication frequency 7.8-fold (0.037% ± 0.009% SD), but did not affect formation of REP and SJ types (see red bars).
Effect of RecA and IS3 on formation of duplications. The frequency of several duplication types was determined for populations grown 33 generations without selection. RecA affects formation of duplications with IS3A/C junctions but not duplications with short junction sequences. Strains lacking RecA still produce an appreciable number of IS3A/C duplications.
The duplication rate in recA strains was still quite high (3 × 10−5/cell/division). Most of the duplications formed without RecA have a copy of IS3 at their duplication junction and thus depend in some way on extensive IS3 repeats. The short junction sequences of REP and SJ duplications make it difficult to imagine their formation by homologous recombination. Consistent with this idea, the frequency of these duplication types was unaffected by the absence of RecA (0.01% ± 0.006% SD).
Removal of IS3 elements reduces lac duplication frequency
Formation of most lac duplications (with or without RecA) involves some kind of exchange between IS3A and IS3C, the only extensive sequences that flank lac. Deleting IS3C prevents both the IS3A/C and IS3B/C duplication types and reduces the overall frequency of lac duplications by 32-fold to 0.009% (±0.003% SD). The lac duplications that form in the IS3C deletion strain were all REP or SJ types. Formation of these duplication types was unaffected by removal of the IS3C element and did not depend on RecA (see right side of Figure 7). This is consistent with the idea that REP- and SJ-mediated duplications form by a mechanism that involves neither IS3 homology nor recombination. In contrast, formation of the IS3C/A duplications depends on exchanges between IS3 elements that can be achieved either with or without RecA. The putatively defective IS3B element had very little effect on total duplication frequency. This supports the idea that the sequence changes that distinguish IS3B from A and C reduce its ability to recombine and produce active transposase (see below).
IS3 transposase stimulates IS3A/C-mediated duplications
The involvement of IS3 in duplication formation without RecA suggested that IS3 elements might contribute more than simple sequence repeats. To test the role of transposase, the IS3A and C elements were replaced by directly oriented copies of the rifR gene (rifampicin resistance). This gene is nearly the same size as IS3 (1017 bp compared to 1258 bp), but is not derived from any transposable element and does not encode transposase. As shown in Figure 8, replacing only the IS3C element with rifR left no pair of sequences flanking lac and reduced duplication frequency 32-fold. Replacing both IS3A and C elements with rifR sequences left the lac region flanked by directly oriented copies of the rifR gene, but eliminated transposase. In the rifR-rifR strain, the lac duplication frequency was 10.7-fold lower than that in the parent IS3–IS3 strains (0.288–0.027% ± 0.007% SD) suggesting that sequence homology is insufficient. Of the lac duplications recovered in the rifR-rifR strain, 80% had a rifR gene at the join point and formed by exchanges that occurred between repeated rifR sequences without benefit of transposase nicking.
Effect of IS3 transposase on lac duplication formation. All strains except the one at the far left have a deletion of the defective IS3B element. For each strain, the two duplication types were determined: IS3 (open columns) and REP plus SJ (red columns). In the third, fourth, and fifth strain, IS3A and IS3C sequences are present but transposase (Tnp) production has been eliminated from one, the other, or both elements by a base substitution in the initiation codon of the tnp gene. In the strain at the far right (shaded column), both IS3A and IS3C were individually replaced by directly oriented copies of the rifR gene, leaving lac flanked by identical sequence repeats with no encoded transposase.
A recA mutation reduces the frequency of rifR-rifR duplications 2.5-fold (from 0.027 to 0.011% ± 0.007% SD). In this recA mutant, 70% of the duplications had rifR junctions and the rest were the short junction types. Together, these results suggest that direct-order sequence repeats recombine at low frequency and can contribute to duplication formation even without RecA, but some property of the IS3 repeats stimulates the process significantly. In all of these tests, the formation rate of REP- and SJ-duplication types was not affected by the presence or absence of flanking IS3 or rifR repeats, or by a recA mutation, suggesting that these types form by an alternative mechanism.
To test contribution of IS3 transposase, expression of the IS3-transposase gene was eliminated by changing its ATG start codon to ATA. This point mutation was shown previously to eliminate expression of the IS3-transposase gene and reduce the IS3 co-integration rate by approximately 100-fold (Spielmann-Ryser et al. 1991). Thus, a single base change can eliminate transposase without reducing the size of the recombining sequence. Strains were constructed with this point mutation in one or both of the IS3 elements, IS3A and IS3C. In testing the effect of these mutations on duplication, the entire putatively defective IS3B element was removed to eliminate any potential contribution. As shown in Figure 8, removing IS3B had very little effect on total duplication frequency in an otherwise normal plasmid (second column).
Surprisingly, elimination of transposase expression from either one of the remaining IS3 elements reduced duplication formation 3.5- and 3.6-fold, respectively (to 0.080 ± 0.015% SD and 0.076% ± 0.022% SD). Silencing both transposase genes caused very little additional drop in duplication rate (to 0.067% ± 0.019% SD). This suggests that the transposases produced by both elements work together, possibly in cis on the element that encoded them, to stimulate duplication formation. This suggests that duplication may often involve simultaneous events at both elements. The double transpose defects reduced duplication frequency in both rec+ (4.3-fold) and recA (2.6-fold) strains. This suggests the transposase contributes to duplication both with and without RecA.
It should be noted that transposase contribution does not involve an act of transposition, since the scored duplications all showed an exchange between IS3 copies (as diagrammed in Figure 1) and did not generate any novel sequence junction as would be expected for replicative transposition (see Figure 2). The absence of IS3 transposases had no affect on the overall frequency of REP or SJ duplication types.
In strains defective for transposase, duplication formation rates were partially restored by overexpressing IS3 transposase in trans from a plasmid. The plasmid is present at a copy number of about 10 per cell and expresses transposase from an arabinose-inducible promoter (pBAD). Inducing this transposase gene increased the duplication rate only about twofold in strains lacking either IS3A or IS3C transposase singly, and 1.8-fold in a strain defective for both transposases (data not shown). Given the strength of the arabinose promoter and the high plasmid copy number, these results suggest that IS3 transposase acts preferentially in cis on the sequence element that produced it, as shown previously for IS3 (Sekine et al. 1999) and many other transposable elements (Jain and Kleckner 1993; Weinreich et al. 1994; Derbyshire and Grindley 1996).
Plasmid conjugational transfer functions stimulate IS3-mediated duplications
Sequences on the F′128 plasmid are subject to intense recombination in E. coli caused by constitutive expression of the plasmid’s conjugative transfer functions (tra) (Hopkins et al. 1980; Seifert and Porter 1984; Syvanen et al. 1986). The Tra functions encoded on F′128 accomplish cell-to-cell transfer by nicking F-plasmid DNA at the OriT site, displacing a single-strand 5′-end for transfer. In the Salmonella strains used here, tra expression is repressed by FinO, which is encoded by Salmonella’s resident pSLT plasmid (i.e., tra expression is “low”). Transfer ability of F′128 is eliminated by deleting traI, which normally nicks at OriT (i.e., tra activity is “none”). Conversely, transfer activity can be increased to high constitutive levels by removing the pSLT plasmid, thereby eliminating the tra repressor (i.e., tra expression is “high”). Strains with these three levels of tra expression (none, low, and high) were tested for their rate of duplication formation.
Expression of tra functions contributed substantially to formation of duplications in a recA+ strain, but its effect was reduced in absence of RecA (Figure 9A) and was eliminated in strains that lack both recA and IS3 transposase (Figure 9C). We conclude that nicking at OriT by TraI stimulates RecA-dependent exchanges on the plasmid that contribute to duplication formation. Transfer functions make no systematic contribution to the short junction duplication types (note the different scale in Figure 9B). These rates were determined in an IS3C deletion to eliminate any contribution from IS3 duplication types.
Effects of plasmid transfer functions. (A) The effect of tra expression—and thefore TraI nicking at OriT—on duplication frequency in rec+ and recA strains. (B) The effect of tra expression on the frequency of SJ plus REP duplication types in strains lacking IS3C and therefore incapable of making duplications with an IS3A/C junction. (C) The effect of tra on duplication frequency in strains with flanking IS3 elements but lacking transposase. Throughout the figure, the expression level “none” is a traI mutant with no nicking at OriT, “low” is a traI+ strain with tra operon repressed by pSLT), and “high” is a traI+ strain with derepressed tra operon (no pSLT).
Role of RecB and RecF in duplication
In the model described below, RecA acts primarily to form plasmid dimers. These dimers arise by RecA-mediated exchanges between any homologous points on two plasmids and are stimulated by TraI nicking at OriT (Seifert and Porter 1984; Syvanen et al. 1986). Previous tests of recombination between extensive sequence repeats in the bacterial chromosome (segregation of a chromosomal duplication) suggested that the exchanges required RecA plus either the RecBC or the RecF recombination pathways (Galitski and Roth 1997). The rate of duplication formation on F′128 is reduced in a recA mutant and in a recB, recF double mutant, but not in a recB or recF single mutant (Figure 10). This dependence is expected if recombination is initiated by single-strand nicks or double-strand breaks (Spies 2005) and blockage of either single pathway leaves the other functional. The model suggests that a major part of the RecA contribution to duplication is to promote exchanges between long homologies and make plasmid dimers that are precursors to duplication formation.
Effects of recB and recF mutations. For each strain, the two general duplication types were determined: IS3 (open columns) and REP plus SJ (red columns).
A model for duplication formation suggested by results
The results suggest that duplications with an IS3 join point form by both recombination-dependent and -independent pathways. The IS3 transposase contributes under both conditions without catalyzing transposition. Duplication formation is stimulated by plasmid conjugation functions that are known to stimulate homologous recombination between plasmids (Seifert and Porter 1984; Syvanen et al. 1986). Rare duplication types with short junction sequences (REP and SJ) form by a process that is independent of recombination, transposase, and conjugation functions.
Most duplications form in a plasmid dimer generated primarily by RecA-dependent exchanges between plasmid copies (Figure 11A, right) and stimulated by DNA nicks made at OriT by the conjugation function TraI. In this plasmid dimer, a deletion between nonallelic IS3 elements leaves a monomeric plasmid with a duplication. Both IS3 transposase and RecA contribute to the deletion event, which resembles the resolution step in IS3 replicative transposition. Such deletions can occur by homologous recombination or, less frequently, by single-strand annealing. We propose that IS3 transposase nicks DNA at the ends of IS3, leading to breaks that initiate deletion by either recombination or single-strand annealing.
Pathways for duplication formation and replicative transposition. (A) Three different pathways that form duplications on F′128. Each pathway can be distinguished on the basis of its dependence on RecA, IS3 transposase, and Tra functions. (B) The process of replicative transposition through a co-integrate intermediate. Note the similarity between the resolution event in B and the deletion affecting the plasmid dimer at left side of A.
Duplications can also form without a dimer by standard unequal sister-strand exchanges between nonallelic IS3 elements (IS3A and IS3C) (diagramed in Figure 1). In a recA+ strain, these events are mediated primarily by recombination, but also by single-strand annealing. Without RecA, dimers are less frequent, but duplications can arise by single-strand annealing of IS3 sequences (see Figure 3 and middle of Figure 11A). Annealing requires two coexisting breaks and is made more likely by frequent nicking at IS3 and by longer persistence of breaks in the absence of RecA.
Involvement of a plasmid dimer intermediate is one example of duplication by a multistep pathway. Another example is in Figure 4B above. The model shares general features with replicative transposition—a multistep pathway in which an IS element in one plasmid mediates co-integration with a different plasmid (Figure 11B). The co-integrate is an intermediate that breaks down by exchanges between IS copies to regenerate two separate plasmids, each now having a copy of the IS. In replicative transposition, the subsequent resolution step can be catalyzed by a dedicated resolvase or by a second activity of the transposase that catalyses exchanges between two IS elements. Resolution can also be achieved by recombination between the two elements. The deletions between IS3 elements described here can be considered a resolution activity of the IS3 transposase.
Plasmid dimers are known to form at high frequency in Rec+ strains, but are much less abundant in recA strains (Biek and Cohen 1986; Perals et al. 2000; Barre and Sherratt 2005). Some dimers clearly form at a low rate without RecA and can be detected by positive selection (Dianov et al. 1991; Mazin et al. 1991; Lovett et al. 1993). Our estimates suggest that dimers of F′128 are present in about 25% of cells in an unselected RecA+ culture, while none (<0.1%) were found in recA strains (A. B. Reams, unpublished results). Most F plasmid dimers are resolved by recombination or by the ResD enzyme of F′128, which catalzyes site-specific recombination at the rtsF resolution site (Lane et al. 1986).
The model proposes that the dimer is resolved by unequal exchanges between nonallelic IS3C and IS3A sequences (see Figure 11A left). If this exchange is reciprocal, it produces one plasmid with a tandem duplication and another with a corresponding deletion. These unequal exchanges can occur by simple recombination between any embedded sequence repeats, but are much more frequent when the recombining sequence is nicked, e.g., when tranposase nicks the ends of IS3.
Testing Predictions of the Model
In this model, duplications form by two steps. A plasmid dimer forms and is then resolved by a deletion between IS3 copies. The second step is tested below and provides evidence that deletions do form between IS3 copies and depend heavily on RecA and on IS3 transposase.
Deletions that cause loss of a duplication:
The first test examined the recombination events diagramed at the top of Figure 12. The parent IS3A/C duplication was trapped by the Ka-Kan method, which leaves one copy with a functional lac allele (Lac+, KanS) and the other with a lac deletion (Lac−, KanR). This duplication strain is phenotypically Lac+ KanR and produces haploid segregants that lose either their Lac+ or KanR phenotype with equal probability. Rates of duplication loss were determined during nonselective growth by following a loss of the Lac+ phenotype with time and correcting this loss rate for fitness effect as described in Materials and Methods. A duplication copy can be lost by exchanges between IS3 elements (dashed and solid lines) or between any other homologous sites in the two repeats (dotted lines). Loss by the second route is essentially eliminated by a recA null mutation, which decreases the formation rate of Lac− segregants 50-fold and leaves a residual RecA-independent loss rate that reflects primarily events between IS3 copies. Rates of duplication loss were compared in recA strains with and without transposase point mutations in their IS3 copies. Figure 13 shows that in recA strains IS3 transposase contributes roughly two-thirds of the duplications. Overexpressing transposase in trans from a plasmid restored the deletion rate to strains with a tnp mutation. Thus, IS3 transposase can catalyze deletions between copies of IS3 repeats in the absence of RecA.
Assays of deletion formation on F′128. The figure describes the two events assayed to assess formation of deletions between two IS3 elements. The first test, duplication loss, was performed in recA strains, where duplication loss occurs primarily by exchanges between two IS3 elements. The second test, insert loss rate, uses a monomeric F′128 plasmid and determines the rate at which the insert (lac) is lost by exchanges between nonallelic IS3 elements.
Duplication loss recombination-deficient strains depends on IS3 transposase. The duplication loss rate was measured in recA strains whose F′128 plasmid carries a lac duplication. Three such duplication strains were analyzed: (1) a strain with wild-type IS3 elements, (2) a strain in which all copies of IS3 lack transposase expression, and (3) a strain with impaired IS3 transposase expression and a plasmid that overexpresses IS3 transposase in trans. The frequency of cells without the duplication was measured at various time points during nonselective growth and the duplication loss rates were calculated after correcting for fitness effects using a spreadsheet simulation.
Deletions between IS3 copies that remove lac from a monomeric F′128 plasmid:
The second test examined deletions that remove the lac segment (IS3A–lac–IS3C) from a plasmid monomer. Loss of the Lac+ phenotype occurs primarily by deletions between nonallelic IS3 elements as diagramed at the bottom of Figure 12 and in the second step of the duplication model (Figure 11A, left). These deletions remove the entire 131-kb lac region and occur at a rate of 4 × 10−4 deletions/cell/generation (see row 1 of Table 3). PCR verified that >99% of Lac− derivatives had an IS3A/C deletion join point. Removal of the IS3C element from the parent plasmid reduced Lac− mutant formation over 1000-fold (row 5), demonstrating that the scored deletion events reflect exchanges between IS3A and IS3C, as suggested by the model.
The deletion rate dropped 300-fold in a recA mutant (Table 3, row 3). Of the reduced number of Lac− cells arising without RecA, 71% had deletions between IS3A and C, showing that IS3 copies can recombine without RecA, albeit at severely reduced rates. Reduced expression of IS3 transposase lowered deletion formation about twofold in a RecA+ strain (rows 1 and 2) and about fourfold in a recA mutant (rows 3 and 4). These results suggest that IS3 transposase stimulates (perhaps by DNA nicking) formation of deletions both by RecA-dependent and RecA-independent mechanisms. Thus, the deletions proposed by the model can form and RecA and IS3 transposase are major and minor contributors.
Comparing rates of duplication and deletion:
A surprising aspect of the duplication process was revealed by comparing the deletion rate (estimated by the insert loss) to the duplication rate (measured by the Ka-Kan assay). As seen in the top row of Table 3, the F′128 plasmid forms duplications and deletions at comparable rates in a recA+ strain, using IS3 elements as recombining sequences. A defect in transposase expression reduced duplication and deletion rates two- to fourfold in recA+ and recA strains. Unexpectedly, a recA mutation reduced the deletion rate 300-fold and duplication rate only 10-fold (compare rows 1 and 3 in Table 3). The larger effect of recA on deletion rates was also seen in strains lacking IS3 sequences (rows 5 and 6). As a result, duplications are considerably more common than deletions in the absence of RecA. This was unexpected, because duplications and deletions are generally thought to form in the same way (unequal recombination).
To validate the larger effect of RecA on deletion formation, rates were assayed by direct PCR tests of whole populations. Cultures of the parent strain with a monomeric plasmid were used as template for PCR tests that amplified join points of IS3-mediated duplications and deletions, both which are absent in the parent strain. The frequency of cells with a duplication or deletion join point was estimated from the dilution endpoint at which junction fragments could no longer be detected. In this assay, a recA mutation reduced the IS3A/C deletion frequency about 500-fold and the IS3C/A duplication frequency only 10-fold. The larger effect of RecA on deletion than on duplications is consistent with the results using the two genetic assays. Reasons for this difference are discussed later.
Finally, Table 3 shows that strains lacking the entire IS3C element (row 5) form lac deletions at about the same rate as strains having both IS3 elements but lacking RecA and IS3 transposase (row 4). In both of these strains, the contribution of IS3 is eliminated and the rate of duplication is at least 20-fold higher than that of deletions, suggesting the existence of a low level pathway for duplication that does not require RecA, transposase, or direct order sequence repeats. We propose below that these duplications (the SJ and REP types) form by a pathway initiated by a TID (see Figure 4). The ability of this pathway to generate duplications but not deletions may explain the excess of duplications in strains lacking both recombination and IS3 sequences.
Discussion
Duplications of the lac operon are frequent (3 × 10−4/per cell/division) and are of two general types. Roughly 97% arose by genetic exchanges between IS3 sequences and left a copy of IS3 at the duplication join point. Elimination of RecA, TraI endonuclease, or IS3 transposase reduced, but did not eliminate, formation of IS3-mediated duplications. Much rarer lac duplication types (3%) have short junction sequences (REP or SJ) and end at a variety of sites other than IS3. The REP and SJ duplications appear to form by an independent mechanism.
Several aspects of these findings were unexpected:
While most duplications arise by exchanges between IS3A and C elements, the duplication rate is reduced only about 11-fold in the absence of RecA and rate of duplication in recA mutants is still quite high (>10−5/cell/division). See Figure 7 and Table 3.
The RecA dependence of duplication on F′128 is unexpected since RecA makes a smaller contribution to duplication at some sites in the chromosome (Reams et al. 2010) and on small plasmids (Dianov et al. 1991; Mazin et al. 1991).
While deletions and duplications between long repeats form at similar rates in RecA+ strains, deletion formation is more heavily RecA dependent than duplication. That is, in the absence of RecA, duplications are more common than deletions. This was seen in genetic assays and was verified by PCR amplification of deletion and duplication join points in unselected cultures. This is also seen in the chromosome between 5-kb rrn repeats, where a recA mutation reduces duplication formation only a few fold (Reams et al. 2010) but reduces duplication loss (40-kb deletion) several thousand fold (A. B. Reams, unpublished results).
IS3 transposase contributes to duplication formation both with and without RecA, yet no transposition event is involved.
The conjugative (transfer) replication origin of F contributes to duplication, probably because the TraI enzyme stimulates plasmid–plasmid recombination by nicking DNA at a site (OriT), far from the recombining IS3 sequences.
The functions that contribute to formation of IS3AC duplications and deletions (RecA, TraI, and Tnp) had no effect on formation of duplications with short junction sequences (REP and SJ).
Duplications form at a high rate in many genetic systems. The traditional idea that duplications form by unequal recombination between repeated sequences is attractive, but may be true only with long repeats subject to nicks that generate recombinogenic ends. Experimental genetics may have predisposed us to thinking of recombination as frequent, because the recombination we see in eukaryotic systems has been stimulated by the DNA breaks introduced during the meiotic process. Similarly, bacterial recombination assays provide linear fragments with broken ends. Separated sequence repeats embedded in a larger chromosome may recombine rather seldom, as seen here for IS3 sequences that lack their transposase or for the repeated rifR gene sequences. In light of these considerations, it is surprising that duplications of a typical chromosomal gene with no provided repeats form at a high rate (1 × 10−5/cell/division). The high rate may result from a very large number of potential endpoints, even though the rate of each individually type is very low. High duplication rates at particular sites may reflect catalyzed events between substantial sequence repeats.
We suggest here that the high rate of IS3-mediated duplications reflects frequent plasmid dimers that are remodeled by frequent deletion events to produce the detected tandem duplication (Figure 11A). Such multistep pathways are effective because each step is catalyzed and the intermediate cell type may grow or be subject to selection between steps. The first step, F plasmid dimerization, can result from a recombination event at any point in the plasmid (231 kb), events that are highly stimulated by the nicks introduced by TraI (Seifert and Porter 1984; Syvanen et al. 1986). We estimate that one cell in four has a dimer of its F′128 plasmid in rec+ strains (A. B. Reams, unpublished results). This agrees with frequencies of RecA-dependent dimers seen for other plasmids (Biek and Cohen 1986). The second step in the proposed duplication pathway is a deletion between nonallelic IS3 elements stimulated by transposase-introduced DNA nicks near substantial (1.2-kb) sequence repeats. The IS3 transposase recognizes and cleaves the 3′-ends of the inverted repeats flanking IS3 sequences (Sekine et al. 1999). We propose that IS3 normally uses these nicks (with or without RecA) to resolve the plasmid co-integrates it produces in the course of replicative transposition. In the F′ plasmid, these nicks can generate deletions between nonallelic IS3 copies. With RecA, a break at either IS3 copy can be used to generate a duplication by homologous unequal recombination between sisters. Without RecA, double-strand ends near nonallelic IS3 elements on different sisters may be independently resected to expose single strands that can recombine by annealing. (See Figure 3.) Without catalyzed events, duplications and deletions arise at significantly lower rates.
Perhaps most surprising is the finding that RecA appears more important for deletion that for duplication. To account for these observations, we propose that the strand exchange activity of RecA contributes equally to the process of deletion and duplication. However, in the case of duplications, the reduced homologous recombination caused by lack of RecA may be compensated by an increase in single-strand annealing. This increase may occur because DNA breaks required for annealing persist longer in the absence RecA, making it more likely for two breaks to coexist. In addition, revealed single strands may anneal better when not involved in RecA protein filaments (A. B. Reams, unpublished results).
Duplications in the least frequent class (3%) have short junction sequences (SJ), some of which are within palindromic REP elements far from IS3A and C elements. The structure of these duplications is diagramed in the Appendix. Formation rate of these duplications is about 30-fold lower than that of IS3-mediated duplications and is not affected by lack of RecA, Tra, or IS3 elements. The junctions of SJ and REP duplications resemble those of duplications found in the bacterial chromosome in regions far from rrn repeats (E. Kofoid, unpublished results). Like SJ and REP duplications, the RecA dependence of chromosomal duplication is very low. We suggest that in the absence of long repeats, duplications form by a pathway initiated when short a quasipalindromic sequence generates a symmetrical inversion duplication (sTID) as in Figure 4A (Kugelberg et al. 2010). This structure is later modified by deletions that leave the observed REP and SJ duplications. The proposed pathway is supported by the amplified TID structures found following selection for increased gene copy number in E. coli, Salmonella, and yeast (Rattray et al. 2005; Narayanan et al. 2006; Kugelberg et al. 2010; Brewer et al. 2011; Lin et al. 2011).
The results described here have implications for the origin of Lac+ revertants appearing during prolonged growth under selection (Cairns and Foster 1991; Andersson et al. 2011). In the Cairns system, all revertant colonies include some cells with a high-copy lac amplification, but none of these amplifications has the IS3 junction that is most common in the absence of selection. Selected amplifications have shorter repeats, mostly with SJ junctions. The revertants appearing under selection were first attributed to cells with an IS3-mediated lac duplication that grew slowly and acquired junction deletions that enhanced growth and left the observed short junction sequences (SJ) (Kugelberg et al. 2006). Results presented here argue against this idea by showing that the formation rate of SJ and REP duplications is not affected by removal of IS3. Furthermore, the yield of Lac+ revertants under long-term selection is reduced very little by removal of IS3 (S. Maisnier-Patin and J. P. Aboubechara, unpublished results). These results suggest that the common IS3 duplications are not precursors of the short-junction duplications described here or found in selected amplifications. It seems clear that the whole range of duplication types (IS3, SJ, REP) can form during nonselective growth. We propose that when a population including such cells is placed under selection for more lac copies, only short duplications have sufficiently low fitness cost to support growth and selective amplification (Reams et al. 2010). Consistent with this idea, amplifications recovered after prolonged selection (Kugelberg et al. 2006) have generally shorter repeats than the unselected SJ duplications described here. (See Appendix.)
Acknowledgments
We thank members of the lab for their advice and suggestions during the course of this work and during preparation of the manuscript, including John Aboubechara, Natalie Duleba, Manjot Grewel, Douglas Huseby, Sophie Maisnier-Patin, Semarhy-Quiñones-Soto, and Emiko Sano. This work was supported in part by a grant from the National Institutes of Health, GM27068.
Appendix: The Size and Endpoint Distribution of Duplications with Junctions Other than IS3
The top line in Figure A1 is a genetic map of the affected region of the plasmid. Duplications with REP join points recurred repeatedly and the number of each type is presented. The REP–SJ hot spot is a small region near REPs that contains the endpoints of many SJ duplications. The junction types found in amplifications following prolonged growth under selection (Kugelberg et al. 2010) are presented at the bottom of the figure. One-third of selected amplifications had REP junctions. The most common duplication (97%) found without selection (between IS3C and IS3A) was not found among selected amplifications. The size of REP duplications is dictated by the position of REP in the plasmid and is ∼25 kb. The average size of SJ amplifications trapped nonselectively (65 kb) was larger than the average of the SJ duplications recovered after selection (34 kb). We suggest that all duplication types form prior to selection, but the lower fitness cost of shorter duplications allows them to be amplified successfully under selection.
Duplications isolated with and without selection (SJ and REP types). Duplications other that the most frequent IS3C/IS3A type are described in the top half of the figure. The junctions of SJ duplications lie outside of any REP element. These duplications all include lac but their size is not otherwise constrained by the method used to trap them. The REP duplication all arise by exchanges between REP elements lying on either side of lac. The size of these duplications are constrained by the location of REP elements in the F′128 plasmid. The amplification-bearing strains described in the bottom half of the figure were all isolated as Lac+ colonies after prolonged growth under selection on lactose.
Footnotes
Communicating editor: S. Sandler
- Received June 2, 2012.
- Accepted July 24, 2012.
- Copyright © 2012 by the Genetics Society of America
