The Tandem Inversion Duplication in Salmonella enterica...

Copyright ! 2010 by the Genetics Society of AmericaDOI: 10.1534/genetics.110.114074

The Tandem Inversion Duplication in Salmonella enterica: Selection DrivesUnstable Precursors to Final Mutation Types

Elisabeth Kugelberg,*,1 Eric Kofoid,* Dan I. Andersson,† Yong Lu,‡

Joseph Mellor,‡ Frederick P. Roth‡ and John R. Roth*

*Department of Microbiology, College of Biological Sciences, University of California, Davis, California 95616, †Department of MedicalBiochemistry and Microbiology, Uppsala University, S-751 23 Uppsala, Sweden; and ‡Department of Biological Chemistry and Molecular

Pharmacology, Harvard Medical School, Boston, Massachusetts 02115

Manuscript received January 12, 2010Accepted for publication February 27, 2010

ABSTRACTDuring growth under selection, mutant types appear that are rare in unselected populations. Stress-

induced mechanisms may cause these structures or selection may favor a series of standard events thatmodify common preexisting structures. One such mutation is the short junction (SJ) duplication withlong repeats separated by short sequence elements: AB*(CD)*(CD)*E (* ! a few bases). Anothermutation type, described here, is the tandem inversion duplication (TID), where two copies of a parentsequence flank an inverse-order segment: AB(CD)(E9D9C9B9)(CD)E. Both duplication types can amplifyby unequal exchanges between direct repeats (CD), and both are rare in unselected cultures but commonafter prolonged selection for amplification. The observed TID junctions are asymmetric (aTIDs) and mayarise from a symmetrical precursor (sTID)—ABCDE(E9D9C9B9A9)ABCDE—when sequential deletionsremove each palindromic junction. Alternatively, one deletion can remove both sTID junctions togenerate an SJ duplication. It is proposed that sTID structures form frequently under all growthconditions, but are usually lost due to their instability and fitness cost. Selection for increased copynumber helps retain the sTID and favors deletions that remodel junctions, improve fitness, and allowhigher amplification. Growth improves with each step in formation of an SJ or aTID amplification,allowing selection to favor completion of the mutation process.

IN general, genetic mutations are seen as discontin-uous changes in base sequence whose origin can be

explained by a single event. This view is based onlaboratory genetics in which mutants are isolated usingstringent selection for large discontinuous phenotypicchanges or using screens that involve no growthlimitation. These standard genetic procedures oftenmiss the most common of all rearrangement types—gene copy-number changes, which may be extremelyimportant to genetic adaptation during growth underselection. Because copy number increases are deleteri-ous and unstable (Reams et al. 2010), they may oftenescape detection. However, selective conditions thatdetect copy-number increases can favor cells withsecondary changes that stabilize and reduce the costof the underlying structures. Prolonged selection canthus contribute to the formation and detection ofmutations without affecting the molecular mechanismsthat create them. This can happen without an increasein mutation rate.

Formation of mutations under selective conditionshas been extensively studied in a system developed by

Cairns and Foster (1991). The system employs abacterial tester strain whose mutant lac operon limitsthe ability to use lactose. The mutant lac allele produces2% of the b-galactosidase (LacZ) level found in re-vertant lac1 cells. During nonselective growth this lacallele reverts at a rate of 10"8/cell/division. Cells of thisstrain (108) are plated on minimal lactose medium andgive rise (over several days) to #100 Lac1 revertantcolonies that appear above a lawn of nongrowing parentcells. Since the reversion rate of the lacmutation duringnonselective growth is 10"8/cell/division, the 100 colo-nies accumulated from 108 nongrowing cells suggestedthe possibility that stress might induce in nongrowingcells a mutagenic mechanism that evolved underselection for its ability to create beneficial mutations(Hall 1998; Foster 2007; Galhardo et al. 2007). Wehave argued that a mechanism for stress-inducedgeneral mutagenesis would be maladaptive in view ofthe vast excess of deleterious over beneficial mutations(Roth et al. 2003).An alternative model, which we favor, uses selection

(without mutagenesis) to explain behavior of the Cairnssystem and other related systems (Roth et al. 2006;Andersson et al. 2010). In this model, extremelycommon duplication types that are normally deleteri-ous and unstable (Reams et al. 2010) are detected by

1Corresponding author: Centre for Molecular Microbiology and Infec-tion, Imperial College London, London SW7 2AZ, United Kingdom.E-mail: [email protected]

Genetics 185: 65–80 (May 2010)

selection for increased levels of lac expression. This ispossible because the original mutant lac allele retainssubstantial activity (2% of the revertant b-galactosidaselevel). Under selection, these copy-number variants ini-tiate colonies in which successive mutant types arise andimprove growth progressively until one dominates thecolony. Selection contributes to mutation formation byfavoring progressive growth improvement without anyincrease in mutation rate. In some clones, amplificationprovides sufficient lac target copies that a normally rarereversion event (to lac1) occurs, permitting loss ofmutant alleles and overgrowth to produce a colonypopulated primarily by stably Lac1 revertant cells (stable-rich). In other colonies, deletions arise that reduce thefitness cost of an amplified lac region, allowing higheramplification and faster growth. In these colonies,improvement is achieved by remodeling the originalduplication structure (not by point mutations). Thiscourse of events leads to colonies rich in unstable Lac1

cells with high lac copy number (Kugelberg et al. 2006).The duplications described here were found in theseunstable-rich Lac1 colonies during prolonged growthunder selection.

Two types of amplifications have been observed inthe Cairns system (Kugelberg et al. 2006). In the firsttype (short junction, or SJ), directly repeated copiesof the lac region are separated by short junction se-quences (3–12 bp) (Kugelberg et al. 2006; Slack et al.2006). We have proposed that these arise by remodel-ing of a frequent large precursor duplication formedbetween IS3 sequences (1.25 kb) (see Figure 1). Thisprecursor duplication is carried by 1/500 of the un-selected parent culture prior to plating, but seldomappears in selected unstable revertant colonies. We pro-pose that deletions remove the common IS3 junctionto reduce the repeat size and create a shorter duplica-tion whose junction point is defined by the deletion endpoints. These modifying deletions reduce fitness costand allow higher lac amplification. We will propose herethat SJ duplications can also form from the symmet-rical tandem inversion-duplication (sTID) precursorsdescribed below.

A second amplification class is the tandem inversion-duplication (TID) type described here (Kugelberget al. 2006). In this class, the underlying duplicationhas direct-order sequence repeats that flank a cen-tral inverse-order segment. For example, an asym-metric inversion-duplication (aTID) rearrangement ofthe parent sequence ABCDEF might have the formAB(CD)(E9D9C9B9)(CD)EF, where each letter designa-tes a multi-gene sequence. Note that the two junctionsbetween inverse-order repeats are not extended palin-dromes, hence the name, asymmetric or aTID. Thebasic aTID unit can amplify by unequal exchangesbetween the flanking direct repeats, generating morecopies of a basic repeat unit that includes the centralregion plus one flanking sequence. One example of this

was described previously for a revertant derived fromthe Escherichia coli version of the Cairns strain (Slacket al. 2006), and its formation was attributed to unknownmechanisms brought into play by growth limitation.

We propose that the aTID arises stepwise from acommon precursor, the symmetrical sTID, whose longinverse-order repeats extend from a quasi-palindromicjunction. These rearrangements arise frequently butare rapidly lost because they are both deleteriousand unstable. Figure 2 diagrams a proposal for fre-quent formation of the initial sTID structures at quasi-palindromic sites in the parent sequence. This modelwill be described in more detail below with supportingevidence. These sTID structures may be deleteriousdue to their copy-number increases and to the tendencyof their extended quasi-palindromic sequences to forma hairpin structure during replication. The sTIDs areexpected to be unstable due to the demonstrated fre-quent deletions of hairpin structures and to the possi-bility of simple exchanges between the flanking directrepeats.

Under selection for higher expression of some gene(e.g., lac), increased gene copy number offsets thefitness cost and instability of the sTID by a processsuggested in Figure 3. During prolonged slow growthunder selection, deletions (frequent at palindromicjunctions) render the junctions asymmetric (less un-stable) and reduce repeat size (lower fitness cost). Thusselection favors the conversion of frequent unstablestructures into more stable duplications that are easierto amplify under selection. Any amplification of theinitial sTID duplication increases the remodeling rateby providing more sites for remodeling changes.

Genetic adaptation can be extremely rapid wheneverselection detects copy-number changes, which arisefrequently under all growth conditions. The speed ofadaptation reflects both the high frequency of contrib-uting mutation types and the exponential increases inmutant frequency that occur during growth underselection. Ultimately, the growth of cells with morecopies of a particular gene enhances the frequency ofpoint mutations within repeats by providing moretargets to the expanding clone. The process describedhere seems likely to contribute to human copy-numberpolymorphisms, to evolution of malignant cells, and tothe many systems for which ‘‘stress-induced’’ mutationhas been suggested.

MATERIALS AND METHODS

Strains, media, and chemicals: Strains used are derivativesof Salmonella enterica (Serovar Typhimurium, strain LT2). Thetester strain (TT18302) has a chromosomal leu deletion and aproBTTn10 insertion and carries the E. coli-derived plasmidF9128 with a triply mutant lac region. A deletion (V) fuses thelacI and lacZ genes, a point mutation improves the lacIpromoter (IQ), and a 11 frameshift mutation (lacI33) is lo-cated within the lacI portion of the hybrid lacI-lacZ gene. The

66 E. Kugelberg et al.

minimalmediumwasno-citrateE salts (NCE), containing addednutrients as recommended previously (Slechta et al. 2002).Carbon sources were used at a concentration of 0.2% (w/v).Rich medium was nutrient broth (NB; Difco Laboratories)with 5 g/liter NaCl. The chromogenic b-galactosidase sub-strate X-gal (5-bromo-4-chloro-3-indolyl-b-d-galactopyranoside;Diagnostic Chemicals, Oxford, CT) was used at 25 mg/ml inminimal media and at 40 mg/ml in NB plates.

Isolation of strains with a selected lac amplification:Independent cultures grown in NCE glycerol were plated(23 108 cells) on minimal lactose plates containing X-gal andleucine. Scavenger cells (109) were added to consume con-taminating carbon present in the agar. Plates were monitoredfor blue lac revertants every day. New independent revertantcolonies appearing on day 5 were plugged and resuspendedin NCE and diluted and plated for single colonies on NB me-dium containing X-gal. After 3 days, unstably Lac1 cells formcolonies that are blue with many white (Lac") sectors; suchcolonies are known to carry a tandem array of lac copies(Tlsty et al. 1984; Andersson et al. 1998; Hastings et al.2000). These unstably Lac1 cells were used as conjugationaldonors and their plasmids were transferred into a recArecipient strain (DA7700) to stabilize the amplification array(Kugelberg et al. 2006).

Cloning, PCR, and sequencing: The sequences of allprimers for PCR and sequencing are available upon request.Genomic DNA was isolated using Wizard Genomic DNApurification kit (Promega). For sequencing, PCR productswere purified from solution using QIAquick purification kit(Qiagen). Purified PCR product was used as template in asequencing reaction using BigDye Terminator v3.1 cyclesequencing reaction kit (Applied Biosystems). TOPO TAcloning kit (Invitrogen) was used whenever it was necessaryto clone a PCR product.

Experiments to test the structure of an aTID: A tetracyclineresistance cassette (TetRA) was added to the TID junction ofstrain TT25771 by linear transformation (Slechta et al. 2003).Plasmid F9128 from this strain was conjugated into TT4632[pro-47(del:AB) recA1 srl1], which lacks drug markers. To deter-mine the orientation of the TetRA-marked junction sequence,a second linear transformation was performed using a chlor-amphenicol resistance (CamR) marker as donor. Four dif-ferent CamR probes were used. Two probes had a CamR

determinant flanked on one side by the recombining se-quence from the left side of the amplified region and on theother side by the TetRA sequence. Two other probes had aCamR determinant flanked on one side by the recombiningsequence from the left side of the amplified region and on theother by the TetRA sequence in one orientation or the other.The design of these four probes is described in Figure 7 andTable 2. The specificity of these crosses was demonstrated bycontrol transformations in which the recipient had a Tn10insertion in an unamplified lac operon. In these controls,inheritance of CamR depended on both the presence and theproper orientation of the Tn10 insertion. All four probes weredesigned so that inheritance of CamR would delete the TetR

determinant from the recipient sequence involved in theexchange. Transformants were selected on LB 1 Cam plates.Ten single transformants from each of the four differenttransformations were restreaked for single colonies on LB 1Cam. Ten single colonies from each transformant were thentested for tetracycline resistance by patching to LB1Cam andLB 1 Cam 1 Tet plates.

Whole-genome sequencing and mapping: Whole-genomesequencing was performedwith the IlluminaGAII SequencingSystem (Illumina). Purified genomic DNA (5 mg/sample) wasfragmented to a targeted average size of 150–250 bp with aCovaris S2 System (Covaris). The DNA fragment libraries were

end-polished, 39-adenylated, and adapter-modified accordingto standard Illumina paired-end protocols. Reamplified li-brary fragments of#200 bp were selected and purified from a3% agarose gel for subsequent paired-end sequencing. Base-calling of raw sequencing data was performed by Illuminasoftware. Sequence reads in the form of two 36-base sequences(mated end pairs) from each DNA fragment were mapped toboth the reference S. typhimurium genome from the NationalCenter for Biotechnology Information and the F9128 plasmidsequences withMAQ (Li et al. 2008). Mapping was done in twopasses. In the first pass, each read pair was mapped with therestriction that the distance between the mapped locations isconstrained to the original selected DNA fragment size.Because read pairs spanning the duplication/deletion bound-ary violate this constraint, they cannot be mapped in this pass.In the second pass, the remaining reads were mapped withoutany distance constraints. The number of reads finally mappedto each position (read-depth) was counted and used for lateranalysis.Determination of junction sequences: Determination of

junction sequences of tandem duplications has been de-scribed previously (Kugelberg et al. 2006). Briefly, 77 primersacross 150 kb of the F9128 plasmid were designed with 39 endsdirected away from lacZ. Seven pools of PCR primers wereassembled; four pools contained primers directing replicationclockwise from lac and three in a counterclockwise directionon F9128. Each of the counterclockwise pools was used incombination with each of the clockwise pools. A unique PCRproduct will be produced when a clockwise and counterclock-wise primer cross a duplication join point. One inverse-orderjunction of a TID structure was identified when primers froma single pool (clockwise or counterclockwise) generated aunique PCR fragment. Hence, at TID junctions primers in thesame orientation (in the parent sequence) yield a PCRproduct. Once a unique band was found, the PCR productwas cloned into a TOPO vector and sequenced using vector-specific primers.Numbering of base sequence from F9128 lac: In the Cairns

system, selection is imposed for reversion of a lac pointmutation. Reversion involves intermediate duplications andamplifications of the lac region, which is carried on plasmidF9128. This plasmid includes the entire F element and asegment (150 kb) of the E. coli chromosome (see Figure 4).The rearrangements described here affect the E. coli-derivedsequences in the top half of the F9128 lac as seen in Figure 4.In the text, ‘‘left’’ and ‘‘right’’ will refer to the position inrelation to lac in this figure with ‘‘left’’ meaning decreasingcoordinate numbers and ‘‘right’’ meaning increasing coordi-nate numbers.

RESULTS

Characterizing selected lac amplifications: The strainused in the Cairns selection system carries a partiallyfunctional mutant lac allele on a conjugative F9 plasmidand has no chromosomal lac region (Cairns andFoster 1991). In the original Cairns experiment, thisplasmid was carried in E. coli cells. Experiments de-scribed here are done with cells of S. enterica that harborthe same mutant F9lac plasmid and have been shown tobehave similarly with regard to lac reversion. Cells withthis plasmid (phenotypically Lac") are plated on solidminimal lactose medium to select revertants (seematerials and methods). Over the course of 6 daysnew revertant colonies appeared each day. New colonies

Tandem Inversion Duplications 67

appearing on day 5 were picked and single cells wereplated on rich medium with X-gal, where cells with a lacamplification form sectored (blue-white) colonies. Sec-tors (white) are initiated whenever some cell in the(Lac1, blue) colony experiences a segregation eventthat removes lac copies and therefore the Lac1 pheno-type. This behavior contrasts with that of haploid rever-tant lac1 cells from the same revertant colony, whichform solid blue (stably Lac1) colonies under the sameconditions. Sectored colonies were used as conjuga-tional donors to transfer the Lac1 phenotype into a re-cipient lac deletion strain with a recombination defect(recA). The RecA defect prevents copy loss and stabilizesthe amplified array (Slechta et al. 2003).

Once stabilized, these high-copy amplifications wereanalyzed by PCR, using pools of primers designed toidentify the junctions between copies repeated intandem. To identify simple junction sequences, PCRprimers were assembled into pools containing multipleprimers whose 39 ends initiate replication away from thelac locus in the same direction on the F9 lac plasmid.Primer pairs in any one pool are not expected toproduce a PCR product since their replication tracksproceed in the same direction. Combinations of primerpools that initiate synthesis in opposite directions (awayfrom lac) are not expected to generate products from awild-type template sequence because their replicationtracks diverge. However, when the template sequenceincludes tandem repeats, some pairs of normally di-vergent primers can initiate replication tracts thatconverge across a duplication junction; the sequenceof these fragments revealed the nature of the junctionfor standard tandem-repeat amplifications. The primersets used are able to detect the junction of any simple lacduplication ,175 kb (Kugelberg et al. 2006).

Ninety-nine amplification strains were isolated fromunstable-rich Lac1 revertant colonies that arose duringprolonged growth under selection. Of these 99, two-thirds (64 revertants) had a simple tandem duplicationstructure described previously (Kugelberg et al. 2006).Of these 64 strains, 48 had a 3- to 12-bp SJ, 15 hada 30-base repetitive extragenic palindromic element(Gilson et al. 1984) at the junction, and one had a copyof IS3, suggesting that it arose by replicative trans-position of the IS3C element. None of the 64 amplifi-cations arising under selection had a join point formedby an exchange between IS3A and IS3C, a duplicationtype found in .0.1% of cells in the unselected parentculture and diagrammed in the center of Figure 1. Webelieve that our PCR methods can detect junctions ofany simple tandem duplication. The remaining 35 (ofthe initial 99 amplifications) lacked a simple duplica-tion junction and appeared to be of a distinct type.

The nature of the new amplification type was firstsuggested when template DNA from one unstably Lac1

revertant yielded a PCR product generated by a singleprimer. Initially, a single pool of primers oriented in the

same direction produced a product, and later a singleprimer from this pool was shown to be sufficient. Thatis, the rearrangement included two oppositely orientedcopies of a single sequence, such that one primer typecould initiate converging replication tracks across therepeat junction. The sequence of this PCR fragmentincluded our first example of a TID join point. ThirteenTID junctions are described here.

For 29 of the unstably Lac1 strains, no junctions couldbe identified by PCR, and the nature of the amplifica-tion in these ‘‘recalcitrant’’ strains is not known. Thesemay carry new amplification types or they may be TIDamplifications whose junctions resisted detection byPCR. We suspect that many are of the TID type thatsimply eluded our PCR attempts. TID junctions wereharder to identify by PCR than SJ junctions, and onerecalcitrant junction (identified as a TID type by ge-nome sequence) has been difficult to demonstrate byPCR even when its sequence was known. To clarify themethods used to identify these join points and themodel proposed to explain their formation, the formalproperties of tandem inversion duplications are listedbelow.

Formal structure of a TID: The 13 junctions de-scribed below collectively define the TID as a newamplification type and suggest a model for its forma-tion. The formal properties of a TID are the followingand are described in Figures 2 and 3.

1. Thebasic unit has two extendeddirect-order sequencerepeats flanking a central inverted sequence, e.g.,AB(CD)(E9D9C9B9)(CD)E. Unequal recombination

Figure 1.—Formation of SJ duplications under selection.The short arrows between b–c and d–e in the parent sequenceare repeats of 3–12 bp. The boxes are the 1.25-kb elementsIS3A and IS3C that flank lac on the F9128 plasmid. The dupli-cated region between IS3 copies is #125 kb, and duplicationsof this type are carried by 0.1% of cells prior to selection. De-letion between identical short sequences leaves a smallerduplication, typically #20 kb. Shortening reduces the rever-sion rate and fitness cost of a duplication and thereby allowshigher amplification (Kugelberg et al. 2006).


exchanges between the flanking repeats (CD) of thebasic unit can generate an amplified tandemarray thatcarries repeats of the [(CD)(E9D9C9B9)] sequence withthe overall structure AB[(CD)(E9D9C9B9)]nCDE.

2. In an aTID duplication, sequences can be presentin two or three copies. The example above—-AB(CD)(E9D9C9B9)(CD)E—has two copies of B andE and three copies of C and D. This distinguishes theTID from a conventional inversion, which has no re-peated sequences.

3. A TID has two alternating asymmetric inversionjunctions (tail–tail, head–head) that together pro-vide duplication status and lead to direct-ordersequence repeats. This distinguishes the TID fromconventional tandem duplications, which have a sin-gle tail-to-head sequence junction (the D–A junctionof ABCD–ABCD).

4. The TID amplification junctions described here areaTIDs. That is, the inverse-order sequences on eitherside of the two junctions are not an extendedpalindrome and thus are not prone to form extendedhairpin structures. In the third and fourth lines ofFigure 3, the left-hand aTID junction is the asym-metric d–e9 (not the symmetric possibilities dd9 or ee9).This contrasts with the hypothetical sTIDs in thesecond line of Figure 3 or the last line of Figure 2,which have extended inverse-order sequences (b c de–e9 d9 c9 b9) and might be expected to formhairpins. Such hairpin structures have been shownto be deleterious and unstable in other systems(Leach 1994; Cromie et al. 2000), especially whenthe junction is perfectly palindromic.

5. The asymmetrical aTID structures described here areproposed to form from frequent symmetrical pre-

cursor sTID structures (bottom of Figure 2). Thesymmetry (or quasi-symmetry) proposed for the sTIDmay allow inverse-order repeats to form hairpinstructures that are deleterious when extended. Aninitial quasi-palindromic junction can be madeasymmetric by a deletion that removes unequalamounts of sequence from either side. These dele-tions leave extensive sequences between the inverse-order sequence repeats and minimize the chances ofa hairpin extension.

6. Between inverse-order repeats, a unique short se-quence element (4–31 bp) marks each asymmetricjunction (aTID). Two copies of each junction ele-ment are present in inverse order at widely separatedpositions in the parent sequence (up to 24 kb apart;red and blue short arrows in Figure 3). The deletionthat remodels a junction extends between suchrepeats and leaves one copy of the element at theremodeled TID junction. Each aTID junction se-quence element characterized here is unique: no twoTID amplifications characterized so far have thesame junction sequence. Thirteen of these TIDjunction elements are described below.

Determining the sequence of TID junction elements:To identify TID amplifications among the unstableLac1 revertants, pools of similarly directed PCR primerswere assembled. Each pool was tested for the abilityto produce a PCR product using template DNA from43 uncharacterized nonstandard amplifications—35 se-lected from the standard Cairns strain and 8 from astrain lacking IS3 (see materials and methods).Screening of these DNA PCR samples revealed 12inversion-duplication junctions—both junctions fromeach of 4 strains (8 total) and one junction from each of

Figure 2.—Proposal for the formation of sym-metrical TID structures. Short quasi-palindromicsequences are frequent and can support forma-tion of snap-back structures that initiate repairsynthesis, which can switch templates and be re-directed toward the fork. The resulting branchedstructure can be broken at arrow and repaired orcan be replicated to produce the sTID structureproposed to initiate the aTID structures de-scribed here. These events may also underlie re-combination-independent formation of simpleduplications.


4 additional strains. A 13th junction was described laterby genome sequencing.

The first six lines of Table 1 describe amplificationsfrom 6 of the original 35 revertants that showed anunstableLac1phenotypebut lackedstandardduplicationjunctions. Our PCRmethods showed that these six are allaTID types. Sequencing of PCR fragments revealed bothjunctions from 4 strains (lines 1–4: DA8097, DA7948,TT25721, TT25773) and one junction for each of twostrains (lines 5 and 6: TT25790 and TT25772). Later useof genome sequencing (see below) revealed the missingleft junction for the strain in line 5 (TT25790). Sequencenumbering conventions are diagrammed in Figure 4.

The strains described in lines 7 and 8 were isolatedfrom parents lacking IS3A and IS3C. The strain de-scribed in line 7 (TT25542) was isolated under selectionlike those above and shows an aTID amplification. Thestrain described in line 8 (TT25561) was trapped non-selectively (see below) and has an aTID duplication witha somewhat longer junction element at its one de-termined junction (a 31-bp REP element).

Verifying amplification structure by genome se-quencing: Two types of amplifications (SJ and aTID)have been inferred from the sequence of PCR frag-ments that include a junction. Amplifications basedon SJ tandem duplications were described previously(Kugelberg et al. 2006) and those based on TID du-plications are described above (Table 1). To examinethe association of junction sequences and lac ampli-fication, a full-genome sequence was determined forone previously described strain with an SJ amplifica-tion (TT24815) and a second strain inferred to carryan aTID amplification (TT25790 in Table 1).

The SJ amplification strain sequence confirmed thesingle junction element (AGGGCAGG) determinedpreviously by PCR fragment sequencing (Kugelberget al. 2006). This junction is proposed to arise when a

deletion event removes the material between two copiesof the small sequence element (see Figure 1). A deletionbetween the short elements in different duplicationcopies (Figure 1) removes the IS3 sequence at thejunction of the original duplication and generates theobserved SJ junction of a shortened duplication. Read-depth of Illumina sequencing demonstrated that theregion between the two short elements in the wild-typegenome was amplified in the selected Lac1 revertant.The original lacmutation was present in all copies in theamplification.

In the case of the TID-based amplification strain(TT25790), the right (divergent) junction had beencharacterized by sequencing a PCR product (Table 1;line 5). The genome sequence confirmed the knownjunction sequence and revealed the previously un-known left (convergent) join point, which had escapedidentification by PCR. This junction is described below.The lac amplification carried by strain TT25790 wasdemonstrated by examining the number of sequencereads (see Figure 5) mapping to each genome position(read-depth). That is, the chromosome was sequencedwith an average coverage of 3.2 reads/nucleotide (#90-fold coverage at 30 bases/read). The bulk of the F9128plasmid (outside the lac region) was sequenced at #5.2reads/base, suggesting that the plasmid was present at a1.6 excess or nearly 2 copies/cell. This low numberseems reasonable since DNA was prepared from station-ary phase cells in which plasmid copy number is ex-pected to be low. The plasmid region between TIDjunction points was read 19.3 times/nucleotide, sug-gesting an amplification of approximately sixfold vis-a-vis the chromosome (fourfold vis-a-vis the rest of theplasmid). While this complete genome sequence re-vealed the junction sequences and the amplification ofthe affected region, it cannot demonstrate the structureof the amplified array (see below).

Figure 3.—Conversion of an sTID precursorto an aTID duplication. The sTID forms and islost frequently. Asterisks in the parent sequenceindicate small palindromic sequences that pro-mote sTID formation. Prolonged selection forhigher expression (e.g., lac, the black dot) favorsretention of the sTID and sequential accumula-tion of deletions that render junctions asymmet-ric (forming the aTID). As fitness cost is reduced,higher amplification allows faster growth and ul-timately provides sufficient lac targets so thatpoint mutations can stably alter the gene underselection. Short colored arrows are repeated oli-gonucleotide sequences present in the parentthat serve as endpoints for deletions and junc-tions of the final aTID duplication. Uninvolvedrepeats are omitted after the first two lines.


TABLE1

Seq

uen

ceofinversion-duplicationjunctions

Left(convergen

t)junction

Right(d

iverge

nt)

junction

Line

Strain

Sequen

ceelem

enta

Coordinates

inparen

tbSe

quen

ceelem

enta

Coordinates

inparen

tb

Amplified

unit:central

sequen

ceplusone

repeat(kb)

Amplificationselected

instrain

withIS3A

andIS3C

1DA80

97GCGCGCCAG

156,98

5–93

CGCTACGCCA

143,37

3–82

1615

7,78

8–80

15,507

–68

2DA79

48TAGCGTCGCATCAGGCATCTGCGCACGACT

159,13

3–62

ACCTACCG-ATTG

146,50

4–11

26(a

REPelem

ent)

161,28

0–51

147,67

0–66

3TT25

771

TCCGCTGACCT

175,25

4–64

TTGGCG

134,61

2–17

2415

0,79

0–80

141,54

3–38

4TT25

773

GACCTAGACGCG

158,78

8–99

AACC-CCTTAC

145,28

0–90

4417

2,18

2–71

141,56

2–52

5TT25

790

0bases

(see

Figu

re9)

ATCGTAGCCGC

132,09

8–21

0850

134,08

5–75

6TT25

772

ATTTCCGGCACCTGTT

1841

17–3

2ND

ND

ND

1746

47–6

2

Amplificationselected

instrain

withoutIS3A

,-C

7TT25

542

ND

ND

CGGAACAGGC

131,30

0–09

ND

1408

91–8

2

Duplication(T

ID)trap

ped

afternonselectivegrowth

instrain

withoutIS3A

,-C

8TT25

561

CGGATAAGGCGTTCACGCCGCATCCGGCAGT

165,49

6–55

26ND

ND

(aREPelem

ent)

159,08

9–59

ND,notdetermined

.aBoldface

typeindicates

sequen

cedirectlyinvolved

inmakingthedeletionan

dcreatingthejunctionelem

ent.Other

sequen

cesindicateex

tended

homology

nearthe

junctionthat

mighthavesupported

theex

chan

ge.

bCoordinatenumberingisdescribed

inFigu

re4an

din

mater

ials

andmet

hods.Coordinatenumbersascendclockwiseacrosstheregionoftheplasm

id,includinglac.If

thesequen

ceelem

entsABCDarearrange

din

thisascendingorder,theinversionduplication(A

BCD9C

9B9ABCDE)has

aleftmostjunctionCD9an

darigh

tmostjunctionB9A.If

arrowsfollowascendingco

ordinatenumbers,theleftmostjoin

pointissaid

tobe‘‘convergen

t’’an

dtherigh

tmost‘‘d

iverge

nt.’’From

thesequen

cesad

jacentto

thejunction

regions,onecandeterminethesize

oftheam

plified

repeatas

thesum

ofthesize

ofthecentral

inverseregionplusthesize

ofoneofthedirectrepeats

that

flan

kit.


Even a complete genome sequence is insufficient todefine the structure of a TID amplification: Each TIDjunction carries a short sequence element at which anexchange appears to have occurred to generate theinversion (see colored short arrows in Figure 3). Theparent sequence has two copies of each junctionelement in opposite orientation. In the sTID, thesequence on opposite sides of the junction is inverted,which brings copies of the repeat element into directorientation, such that they can serve as sites for deletionformation. As seen in Figure 6, there are two ways ofusing these repeat pairs to form a deletion, and eachdeletion type leads to a different array structure.However, the local junction sequence is identical re-gardless of which pair is chosen. That is, one cannot

tell from local sequence information how that junctionis oriented in the chromosome: a C–D9 junction isindistinguishable from aD–C9 junction. As diagrammedin Figure 6, one pair of junction sequences, C–D9(orC9–D) and A–B9(or B–A9), is consistent with any offour different TID structures. The uncertainty reflectsthe fact that sequence determination requires the as-sembly of short reads. When the parent region includesrepeats of a sequence that is many kilobases in length(substantially longer than the sequence genomic frag-ment), the sequence reads cannot be unambiguouslyassembled.

A genetic method for testing the structure of aninversion duplication: One can, in principle, decidewhich of the four structures in Figure 6 is correct bydetermining the orientation of each junction sequencewith respect to an unamplified outside reference se-quence. To make this test, a known junction of an aTIDamplification was marked with an added sequence. Thisstrain was used as the recipient for a transformationcross that revealed the orientation of the marker. First,a DNA fragment encoding the tetracycline resistancedeterminant (tetRA of Tn10) was added to a sequencedaTID junction (orientation unknown) by Red-mediatedtransformation (Figure 7, middle). Orientation of theinserted tetRA sequence was then determined by a sec-ond transformation for which the constructed donorfragment included a CamR marker flanked on the leftby an outside reference sequence and on the right bya sequence from the tetRA fragment (one orientationor the other). The donated drug resistance can be in-herited only if the recombining sequences (40 bp long)on the right side (in donor and recipient) are in thesame orientation. The tetRA recombining sequence ischosen so that inheritance of the CamR marker deletespart of the TetR determinant. The orientation of therecipient tetRA sequence dictates which of the donorfragments (Ref-Cam-tetRA or Ref Cam-ARtet) allowsintroduction of CamR and removal of the recipient TetR

Figure 4.—Structure of the F9128 plasmid on which rear-rangements occur. The base numbering conventions arethe following: Bases 1–100,413 are derived from the F factor(bold arc of circle). Bases 100414–150235 are from the E. colichromosome extending from IS3C to include all of the lac op-eron up to base 150,235, which is immediately adjacent to thepromoter-proximal end of the lacZ coding sequence. Bases150,236–231,638 of the plasmid are also from the E. coli chro-mosome and extend from lacZ to beginning of F factorsequence (at base 1). The origin and structure of this plasmidhave been described in detail (Kofoid et al. 2003).

Figure 5.—Depth of Il-lumina reads from theF9128 lac plasmid of a straincarrying a TID amplifica-tion. The plasmid geno-type is diagrammed belowthe graph with the F factorplasmid sequence indi-cated by a bold line andthe E. coli material by afiner line. Strains are recAand were grown on glycerolwith no selection for lacamplification. The readdepth for chromosomalgenes (not shown) was#90-fold. The nearly 200-

fold coverage seen for the plasmid (plotted horizonal line) suggests that the strain has several copies of the F9128 plasmid.The lac region is amplified #4-fold above the rest of the plasmid.


phenotype. Control transformations using simplelacTTn10 insertions with no amplified array showedthat formation of a CamR transformant requires that therecipient has a tetRA sequence in the orientationconcordant with the donor fragment.

Figure 7 describes this cross as applied to the leftjunction of the strains diagrammed in Figure 6. Thecross reveals whether the recipient junction resemblesthat of strains I and IV (in Figure 6) or strains II and III.Learning the complete structure of the duplicationwould require determining the orientation of the rightjunction by a similar series of crosses.

The procedure described above gave unanticipatedresults. It was applied to the convergent junction of theaTID amplification strain TT25771 (see Table 1, line 3).This junction has a known sequence but an unknownorientation. A tetRA cassette was added to the junction(Figure 7, middle), and a second transformation crosswas done using two donor fragments with oppositelyoriented tetRA sequences (Figure 7, bottom). Only oneof the two donor fragments was expected to yield CamR

transformants because only one should have a TetRAsequence whose orientation matches that of the re-cipient junction. When the procedure was repeatedusing a reference sequence on the other side of theamplified region, congruent results were expected, i.e.,only one of the two donor fragments was expected toyield CamR transformants. The arrangement and resultsof the crosses are in Table 2. These crosses were done inRecA1 strains.

Contrary to expectation, all four donor fragmentsgave CamR transformants, regardless of the orientationof their tetRA recombining sequence. This suggestedthat the recipient population contained multiple tetRAsequence elements, some in each orientation. Eithersome cells have tetRA in one orientation and some in theother or multiple orientations are present within asingle amplified array. Also contrary to expectation,roughly half of the CamR transformants retained TetR

instead of losing it as predicted by the design of thecross. It had been expected that the TetR resistancedeterminant (tetRA) would be added to only onejunction and the deletion made by the second crosswould remove this TetR determinant. The result ob-tained suggested that the arrays have multiple copies ofTetRA, even though only one was introduced by trans-formation into the amplification strain. The resultssuggested that the repeats of the TID amplificationwere rearranged during growth. This is possible becausethe junctions were marked and cells for this cross weregrown in RecA1 cells. It is interesting that crosses

Figure 6.—Multiple structures are consistent with a singlepair of junction sequences. In the examples listed (I–IV), theleftmost (converging) junction is D–C9 (equivalent to C–D9)and the rightmost (diverging) junction is B9A (equivalent toA9B). If the asymmetric inversion junctions are formed by de-letions in a symmetrical precursor, then four possible struc-tures could be generated by the various combination ofdeletions described in the bottom line. The colored arrowsindicate repeated sequences present in inverse order in theparent chromosome and how they contribute to aTID forma-tion.

Figure 7.—Genetic testing of TID amplification structure.To determine whether an aTID junction is in orientation CD9or DC9, a resistance determinant (tetRA) was added to the se-quenced junction of strain TT25771 (see Table 1). The chro-mosome of the resulting strain is expected to be asdiagrammed for strain 1 or strain 2 (circled). In a secondtransformation cross, a donor CamR determinant is intro-duced with one flanking sequence matching a reference re-gion at the left and the other matching the tetRA sequence.The CamR determinant can be inherited only if the orienta-tion of the recipient junction (tetRA or ARtet) matches thatof the donor recombining sequence (right side of CamR).The donor fragments are designed such that inheritance ofCamR creates a deletion of the region between the referencesequence and the right side of the TetR determinant in therecipient. The recipient strain is recA1 and was grown nonse-lectively before transformation.


involving the ‘‘forward’’-oriented Tn10 element gavemore transformants than those involving a ‘‘reversed’’Tn10. This result suggests the presence of more for-ward-oriented Tn10 elements in the array than reverse-oriented ones.

Taken together, the transformation crosses suggestedthat the arrays were expanding and contracting suchthat the introduced TetRA sequence was subject toamplification and loss during subsequent (nonselec-tive) growth. It also suggested that the orientation ofsome of the introduced TetRA sequences changed dur-ing growth. Thus, it appears that the amplified arraysgenerated from a tandem inversion duplication have astructure that is complex. The array is being scrambledduring growth, both in terms of copy number and joinpoint orientation. A look at the sequence of an aTIDarray shows why this might have been expected (seeFigure 8). This figure diagrams the consequences of aninversion arising between inverse sequence repeats. Theend result is an amalgam of the four types diagrammedin Figure 6.

The arrays have direct and inverse-order repeats ofextensive sequence blocks that are subject to recombi-nation events that can change copy number andorientation. The copy-number changes seem reason-able (in retrospect), but were not expected to occur sofrequently. Part of the explanation may be the intenserecombination known to occur on the F9 plasmid(Seifert and Porter 1984; Syvanen et al. 1986). This

recombination may continuously scramble the arrayduring growth and thereby break down the formaldistinction between the TID types diagrammed inFigure 6, which are expected only for a simple expan-sion without any inversion events.

Regardless of the nature of the initial duplication, theamplified array is predicted to become a heterogeneousseries of segments containing, in random order, all ofthe possible inverse and direct-order sequence blocksdiagrammed in Figure 6. If this structure were notscrambled, one would expect that some TID arrayswould have no copies of lac in inverse order. Given thatscrambling occurs, all TID arrays are predicted to havesome copies of lac in inverse order once the arrayexpands and inversions occur. This may provide adiagnostic test for TID amplifications.

A model for multi-step formation of asymmetricinversion duplications: The defining properties ofaTID duplications make it difficult to explain theirorigin by a single event. If they were to form by simplebreak-and-join events, then at least three parental copieswould be required to acquire four simultaneous breaksand two rejoining events with very short junction se-quence elements. The shortness of junction sequences(4–31 bp) makes it unlikely that homologous recombi-nation is involved since the strand exchange protein(RecA) requires more extensive substrates. To makethese junctions by single-strand annealing would re-quire bothmultiple breaks and production of extremelylong (10–20 kb) single-stranded regions. It has beensuggested that junction sequence elements might ap-pear at a broken 39 end and initiate replication on asister chromosome (Hastings et al. 2009). However,formation of an aTID duplication in this way wouldrequire two simultaneous breaks, each providing an endfor RecA-mediated strand invasion and replication initi-ation at extremely short primer sequences.

The alternative model diagrammed in Figure 2 isdescribed here in more detail. The model suggests thatduplications form by multiple sequential events occur-ring at different times (in different cell generations).This requires simpler genetic events and may explainthe formation of both standard tandem (SJ) and aTIDduplications without RecA function, which has recentlybeen reported (Reams et al. 2010). It also explains whyboth duplication types are more frequent followinggrowth under selection. Each successive step in theirformation provides an additional fitness advantageunder selection. The first event is the formation of asymmetrical inversion duplication sTID with threecopies of lac. The sTID is predicted to form at a highrate, assuring that the structures are common in anypopulation despite being deleterious and unstable.When selection favors more lac copies, the inherentfitness cost and instability of the duplication is offsetleading to an increased frequency in the population(Reams et al. 2010). As cells grow under selection,

TABLE 2

Demonstrating the rearrangement of an amplified TID array

Probea

No. of CamR

transformantsbTetR

(% of CamR)cTetS

(% of CamR)c

L-Cam-Rev (#1) 12 72 28L-Cam-Fwd (#2) 386 69 31R-Cam-Fwd (#3) 200 60 40R-Cam-Rev (#4) 28 72 28

a Numbered probes are described above.b Total transformant numbers from one transformation

cross.c Percentages are based on testing 10 CamR transformants

from each cross. Since these clones are often mixed, 10 singlecolonies were isolated from each transformant clone, andthe TetR percentage reflects behavior of the resulting 100strains.


secondary deletion mutations reduce fitness cost, in-crease stability, and therefore allow higher lac amplifi-cation (see Figure 3). The underlying events arefrequent with or without selection, but the intermedi-ates in the process are usually lost by reversion andcounterselection. Selection helps maintain the unstableintermediates in the population and favors completionof the rearrangement process.

Details of the model for formation of an aTIDamplification: The symmetrical inversion duplication(sTID) serves as precursor for an aTID: The initial step informing an aTID is generation of the symmetricalprecursor (the sTID), which has extended palindromicsequences at each of its two junctions (see Figure 2).These junctions are prone to extrusion as a hairpinstructure, whose likelihood is dictated by the size of theloop at the junction, i.e., the deviation of the center ofthe palindrome from perfect symmetry. Perfect palin-dromes of several hundred base pairs are prone tohairpin formation and are virtually lethal in bacteria(Leach 1994), and even small increases in loop size areknown to increase stability (Warren and Green 1985).The proposed sTID junctions are not perfectly symmet-rical, but are sufficiently stable to be replicated andamplified (by exchanges between the flanking directrepeats). Fitness cost may result from occasional hairpinformation and also from simple increases in genedosage, which are known to impair the growth ofcopy-number variants (Reams et al. 2010).

Formation of the symmetrical precursor: It is proposedthat imperfect palindromic sequences in the normalsequence initiate sTID formation by the events de-scribed in Figure 2. Events of the type proposed herehave been described by others in a variety of organisms(Persky and Lovett 2008). After replication fork stall-ing, single strands of nascent DNA are released andcontribute to chromosome rearrangement (Lovett2004). Short (imperfect) palindromic sequences arefrequent (Stern et al. 1984; Bachellier et al. 1999;

Vasconcelos et al. 2000) and can form ‘‘snap-back’’structures in single-stranded DNA (see Figure 2).A snap-back structure can prime repair synthesis

using its own strand as template. This synthesis mustopen the template duplex and is likely to be prone totemplate switching, which has been observed in manysituations, especially for the PolA repair polymerase(Ross et al. 1979; Ahmed and Podemski 1998; Pinderet al. 1998; Lovett 2004; Lee et al. 2007). Strand ex-tension after this template switch generates a branchedstructure that can be resolved either by chromosomereplication or by repair of the three-way junction to yielda symmetrical inversion duplication (see nick indicatedin Figure 2). Formation of a symmetrical inversion du-plication in this way is predicted to be independentof RecA function since it involves no strand-exchangeevent. These series of events may provide a way ofcircumventing hairpins in the template strand withoutrequiring template breaking, homology seaches, andfork rebuilding.Secondary deletion events remodel the sTID to form an

aTID: The initial symmetrical TID is expected to form ata high rate. The sTID is expected to be deleterious dueto copy-number increase and stem formation. However,this cost can be reduced by deletions that removematerial and reduce junction symmetry to produce anaTID. Such remodeling deletions (diagrammed inFigures 3 and 6) are also expected to be frequent.There are several reasons for this. A large region isavailable on either side of the symmetrical junction thatcan be removed without fitness penalty. Deletionformation at symmetrical junctions is enhanced by theknown propensity of PolA to switch templates, whichcreates deletions as seen by Leach and coworkers inreversion of toxic perfect palindrome structures (Sharpand Leach 1996; Leach et al. 1997) and by Klecknerand co-workers in ‘‘imprecise excision’’ of Tn10 inser-tions (Foster et al. 1981). Frequent asymmetric deletionsremove the extensive (51 bp) inverse repeat near one

Figure 8.—Effects of inversions on structure ofamplified aTID array. One initial aTID structure isdepicted with the changes that are possible due toamplification followed by recombination betweeninverse-order sequence repeats.


end of transposable ‘‘Mud’’ elements (Zieg and Kolter1989); highly asymmetric deletions of this stem in theSalmonella chromosome were recently measured at10"6/cell/division (S. Quinones-Soto and J. R. Roth,unpublished results) and are made more common byprior amplification of the sTID because modifyingdeletions can have endpoints in any pair of previousrepeats.

The role of selection in aTID formation and amplification:During selection for lac amplification, the benefits con-ferred by extra lac copies can compensate for the fit-ness cost and instability of the initial sTID and therebyfavor its retention in the population. As a clone expandsunder selection, join point deletions improve fitness,increase stability, and allow higher lac amplification(as aTID). Amplifications can occur by unequal ex-changes between direct-order sequences, which havebeen found to occur at rates that approximate 10"2/cell/division (Reams et al. 2010) and thus provide afrequent source of variants that improve growth rateunder selection. In this way, selection favors rapid com-pletion of a process that involves unstable deleteriousintermediates.

A fulfilled prediction of the model for aTIDformation: If the aTID amplifications described hereform as proposed above, one might expect to findoccasional amplifications in which only one of the twojunction types has been remodeled and the otherretains the initial snap-back structure. A TID junctionrevealed by Illumina genome sequencing (Table 1, line5, left column) fits this description. A complex hairpinstructure at this junction (left side of Figure 9) is present

in the parental sequence and appears at the convergentjunction of the TID ( junction 1 in Figure 9). Thedivergent junction ( junction 2 in Figure 9) is of thedeletion-modified aTID type and was characterized byPCR analysis (and verified by the Illumina sequence).We suggest that the stem-loop structure in Figure 9supported a snap-back pairing that primed the forma-tion of a more extended inverse-order repeat, and thisjunction was not furthermodified by deletion ( junction1 in Figure 9).

The inverse-order sequence found at this junctionextends the base of the parental structure. The imper-fections of the parental snap-back structure may besufficient tomake the extended palindrome (formed byreplication from the snap-back structure) resistant tohairpin formation without further modification. Wecannot eliminate the possibility that this junction wasformed by a deletion arising in a larger sTID, but thisalternative would require a deletion that fortuitouslyended exactly at the base of an extensive quasi-palindromic stem-loop. It seems more likely that aquasi-palindromic snap-back structure initiated theformation of a structure that was not modified bydeletion. The frequency of such junctions is likely tobe revealed by genome sequences of more amplifica-tions with ‘‘recalcitrant’’ junctions (see below).

Explaining why many ‘‘recalcitrant’’ junctions weremissed by PCR: Analysis of an initial set of 99 amplifi-cation strains by PCR revealed first the simple SJ junc-tions described previously (Kugelberg et al. 2006). Theremaining junctions were then tested using PCR primersets designed to identifyTID junctions as described above

Figure 9.—A TID ampli-fication with one unmodi-fied symmetrical junction.The diagram at left andthe solid arrows at top rightindicate a parental quasi-palindromic sequence thatmight generate snap-backpairing. The events at rightdescribe how the snap-backstructure might prime re-pair synthesis, leading tothe TID junction sequen-ces found in strainTT25790 (Table 1). Thequasi-palindromic junctionis extended without modifi-cation. The second junc-tion forms by templateswitching and is later mod-ified by deletion.


in Table 1. In some strains, only one junction could beidentified. Many other strains proved to be completely‘‘recalcitrant’’ in that PCR yielded neither an SJ nor anyaTID join point. The recalcitrant junctions could haveescaped detection for mundane reasons; i.e., our primersets may have been too limited. Alternatively, the re-calcitrant junctions may reflect a third type of joinpoint—one that standard PCR fails to reveal.

The junction described in Figure 9 was recalcitrant toPCR analysis, but was shown by Illumina sequencingto be part of a TID amplification. The junction isdifferent from those found by PCR: an extensive quasi-symmetrical sequence was present in the wild-typesequence and may have been extended by repair rep-lication from a parental snap-back structure. It seemspossible that the imperfect symmetry of this junctionmakes it resistant to standard PCR. That is, this se-quence may self-anneal when made single-stranded inthe PCR process.

To test this idea, the genome sequence of this am-plification strain (TT25790) was used to design primersthat flank the symmetrical structure in Figure 9. So farwe have been unable to recover a PCR product thatincludes the suggested junction sequence. If this in-terpretation is correct, junctions of this type may becommon and genome sequencing of more strains withrecalcitrant junctions may reveal many unmodifiedamplification junctions that are missed by PCR. Moreexamples of this type would provide support for themulti-step model for aTID amplification formation.

sTID duplications may also give rise to SJ amplifica-tions: The sTIDs proposed to initiate formation of aTIDjunctions, can, in principle, also lead to SJ duplications.This idea isdiagrammed inFigure10 in thecontextof theoverall process of reversion under selection in the Cairnssystem. Each detected aTID junction is proposed to formby a separate deletion event that removes one symmet-rical junctionof theparent sTID, rendering that junctionasymmetric (right side of Figure 10). The endpoints of

each deletion dictated the junction sequence elementsfor each TID junction listed in Table 1. These separatedeletions can occur independently and may be frequentbecause palindromic sequences are known to be subjectto frequent deletion events (discussed later).However, the two symmetrical junctions of an sTID

duplication can also be removed by a single deletionthat removes the entire central repeat segment (Figure10, middle). Such a deletion would lead to an SJduplication, identical to that proposed to arise from alarge parent IS duplication in Figure 1 and Figure 10(left). An SJ duplication can be highly amplified underselection, regardless of how it formed. One dia-grammed route for SJ duplication (Figure 10, left) canbe blocked by removing IS3 repeats from the parentstrain. This allows a test of the possibility of other routesto SJ duplications.Effect of IS3 removal on the relative frequency of SJ

and aTID amplifications: Following prolonged selec-tion for improved growth on lactose, the standardCairns strain (with IS3 copies flanking lac) generatesunstably Lac1 revertants with either SJ or aTID ampli-fications. It was suggested above that some SJ types arisewhen deletions remove the join point from an IS3duplication (see Figure 1) or when a deletion removesthe entire center segment from an sTID join point(Figure 10). In either case, fitness is improved andhigher amplification is allowed, which is favored byselection. Without selection, both the SJ and aTIDduplication types were rare and the recovered duplica-tions were primarily those with IS3 junctions.Removal of IS3A and -C from the Salmonella version

of the Cairns assay strain reduced the revertant numberabout fourfold (Table 3). The amplifications recoveredfrom unstable-rich revertant colonies showed fewer SJtypes relative to aTID types. The fact that any SJ typeswere found at all suggested that their formation doesnot require IS3 duplications as a precursor (consistentwith their formation from sTID precursors). The effects

Figure 10.—The sTID can also contribute toformation of an SJ duplication. The parental se-quence is diagrammed at the top (boxed) inwhich open rectangles designate a copy of IS3and the solid circle designates the lac operon.An SJ duplication can form when a deletion re-moves the junction of an IS3 duplication (leftmiddle). An SJ duplication also forms when a sin-gle deletion removes both sTID junctions (rightmiddle). Two deletions convert an sTID to anaTID duplication. Some IS3-bounded lac duplica-tions may form through sTID intermediates.


of removing IS3 sequences suggest that 2/3 of theobserved SJ amplifications come from precursors withan IS3 duplication and 1/3 arise by another mechanism(e.g., an sTID).

A rough estimate of the breakdown of revertant typesin the Cairn system is shown in Figure 10. Most revertantcolonies (90%) contain predominantly stably Lac1 cells,reflecting a reversion event early in the history of oneof the many plated cells with an unmodified low-copyIS3 amplification (left side of Figure 10). Unstable-richrevertants (10%) have amplifications with either an SJ oran aTID junction. Seventy percent of these (7% of total)have an SJ amplification and 30% (3% of the total) havean aTID amplification. This estimate suggests that sTIDduplications generate amplifications of which 1/3 arethe SJ type and 2/3 are the aTID type. Unselected du-plications isolated from strains lacking IS3 showed pre-dominately REP element junctions among the reducednumber of duplications found. However, one aTID typewas recovered. This shows that (as for SJ types) aTID canoccur in the absence of selection.

Role of F9 plasmid in aTID formation: All of theaTID structures reported here arose on the F9 plasmid,whose transfer replication is known to stimulate RecA-dependent recombination (Seifert and Porter 1984;Syvanen et al. 1986). In our Salmonella strains, thegenes for plasmid transfer (tra) are repressed. Severallines of evidence suggest that TIDs form in the chro-mosome and that the plasmid contributes primarily tothe RecA-dependent amplification events that underliethe reversion in the Cairns experiment. Chromosomalduplications (like those on F) can form without RecAand would be explained by the model presented inFigure 10 for conversion of an sTID to an SJ duplication.

DISCUSSION

The TID rearrangements described here and the SJtandem duplications described previously are rare inunselected cultures but common among amplificationsisolated after prolonged selection for increased lac copy

number. Both duplication types are proposed to arisefrom a very common, but normally short-lived pre-cursor, the sTID. The initial rearrangement is proposedto form at a high rate but disappear rapidly from thepopulation due to fitness cost and instability. Only whencells are placed under selection for increased gene copynumber are the precursor structures maintained suchthat they can be remodeled and lead to a detectableamplification. How could these rearrangements haveescaped detection for so long?

The standard practice of bacterial genetics is to usestringent positive selection to detect mutants that ariseduring preceding nonselective growth. These restrictiveconditions have made laboratory bacterial geneticspossible. Their efficacy was demonstrated by Luria,Delbruck, and Lederberg (Luria and Delbruck 1943;Lederberg and Lederberg 1952). Although thesestringent conditions detect rare preexisting cells withlarge-effect mutations, they prevent common small-effectmutants (such as sTIDduplications) from contrib-uting to the yield of selected colonies. They systemati-cally eliminate detection of most copy-number variants.

The Cairns selection system uses less stringent con-ditions than standard laboratory positive selection. Thelevel of b-galactosidase in the parent strain is 2% of thatin revertants. Selection is just stringent enough toprevent growth of the parent strain, but allows slowgrowth of mutants with a few extra copies of the mutantlac allele. These copy-number variants are common: #1in 500 cells of the unselected plated culture carries a lacduplication. Selective conditions favor copy-numberincreases, which occur at a very high rate (10"2/cell/division)—1 million-fold higher than the reversion rateof the Cairns frameshift mutation. The nonstringentCairns selection conditions (which more closely approx-imate natural selection) detect common copy-numbervariants and ultimately reveal the SJ and aTID amplifica-tions described here. The low stringency of this selectionmay allow this system to reveal aspects of geneticadaptation that have beenmissed by stringent laboratoryselection conditions and mutant screens.

TABLE 3

Effect of removing IS3 repeats on the frequency of SJ and aTID duplications

Duplication/amplification type ( join-point type)a

Source of strainswith repeat structures SJ short join

REPjunctions

IS3A/IS3Crecombination

Transpositionof some IS aTID

ProbableaTID

Amplifications arising under prolonged selection on lactoseStandard Cairns strain 48% (48) 15% (15) 0 1% (1) 6% (6) 29% (29)Strain without IS3(4-fold fewer revertants)

18% (3) 12% (2) 0 24% (4) 6% (1) 41% (7)

Duplications trapped following nonselective growth on glycerolStandard Cairns strain 0 20% (4) 80% (20) 0 0 0Strain without IS3 0 95% (16) 0 0 5% (1) 0

a Percentages describe frequency of each duplication type among those from source strain. Parentheses enclose the actual num-ber of duplications of that type used to determine percentage.


Formation of duplications by a RecA-independentpathway: Duplications have been assumed to arise byunequal recombination between extensive sequencerepeats in sister chromosomes. Recent measurementssuggest that homologous (RecA-mediated) recombina-tion is not required for duplication formation evenwhen exchanges appear to form between substantialsequence repeats (Reams et al. 2010). The modelpresented here may explain the RecA independenceof duplication formation. That is, a common precursor(the sTID) may form in a RecA-independent mannerand then be converted to either an SJ or an aTIDduplicationbyRecA-independent deletions. These dele-tions are expected to be RecA-independent even whenthey arise between extensive sequence elements. Thusmany simple tandem duplications (including some ofthe frequent IS3-bounded duplications) may form with-out recombination by multi-step pathways initiated bysTID rearrangements (see Figure 10).

The gene amplification processes described here maybe major contributors to adaptive evolution becausethey allow extremely frequent genetic variants to bedetected by selection. Whenever cellular growth isrestricted (but not eliminated) by external conditions,copy-number differences may be the most commonsource of improvement. Copy-number variants cancontribute to subsequent sequence change by providingmore copies of the growth-limiting gene and by allowingexpansion of a subpopulation in which point mutationsarise. Later adaptive mutations may occur within theamplified genes or in unrelated genomic regions, whoselikelihood is enhanced only by growth (Roth et al. 2006;Andersson et al. 2010).

For example, selective gene amplification and sub-sequent adaptive mutation have been shown to occurduring selection for improved growth (i) on limitingcarbon sources (Horiuchi et al. 1963; Sonti and Roth1989; Andersson et al. 1998; Hendrickson et al. 2002;Slechta et al. 2003; Kugelberg et al. 2006), (ii) inthe presence of toxic compounds such as antibiotics(Sandegren and Andersson 2009; Sun et al. 2009)with defective translation machinery (Nilsson et al.2006; Lind et al. 2010), or with impaired biosyntheticpathways. The process may underlie evolution of newgenes under continuous selection (Bergthorsson et al.2007). The frequent initiation and rapid modifica-tion and expansion of gene amplifications make themchallenging to study, but suggest that their contribu-tion to genetic adaptation may have been seriouslyunderestimated.

We thank the West Quad Computing Group at Harvard MedicalSchool for computational resources and support. F.P.R. was supportedby grants from the National Institutes of Health (NIH; G00423,HG004756, HG003224 and HG0017115) and by the Canadian In-stitute for Advanced Research. J.M. was supported by an IndividualNational Research Service Award from the NIH/National HumanGenome Research Institute (HG004825). This work was supported in

part by NIH grant GM27068 ( J.R.R.) and grants from the SwedishResearch Council to D.I.A. Elisabeth Kugelberg was supported by afellowship from Wenner–Gren Foundation.

LITERATURE CITED

Ahmed, A., and L. Podemski, 1998 Observations on templateswitching during DNA replication through long inverted repeats.Gene 223: 187–194.

Andersson, D. I., E. S. Slechta and J. R. Roth, 1998 Evidence thatgene amplification underlies adaptive mutability of the bacteriallac operon. Science 282: 1133–1135.

Andersson, D., D. Hughes and J. Roth, 2010 The origin of mu-tants under selection: interactions of mutation, growth and selec-tion, in EcoSal: Escherichia coli and Salmonella, Cellular andMolecular Biology. ASM Press (in press).

Bachellier, S., J. M. Clement and M. Hofnung, 1999 Short palin-dromic repetitive DNA elements in enterobacteria: a survey. Res.Microbiol. 150: 627–639.

Bergthorsson, U., D. I. Andersson and J. R. Roth, 2007 Ohno’sdilemma: evolution of new genes under continuous selection.Proc. Natl. Acad. Sci. USA 104: 17004–17009.

Cairns, J., and P. L. Foster, 1991 Adaptive reversion of a frameshiftmutation in Escherichia coli. Genetics 128: 695–701.

Cromie, G. A., C. B. Millar, K. H. Schmidt and D. R. Leach,2000 Palindromes as substrates for multiple pathways of recom-bination in Escherichia coli. Genetics 154: 513–522.

Foster, P. L., 2007 Stress-inducedmutagenesis in bacteria. Crit. Rev.Biochem. Mol. Biol. 42: 373–397.

Foster, T. J., V. Lundblad, S. Hanley-Way, S. M. Halling and N.Kleckner, 1981 Three Tn10-associated excision events: rela-tionship to transposition and role of direct and inverted repeats.Cell 23: 215–227.

Galhardo, R. S., P. J. Hastings and S. M. Rosenberg, 2007 Mu-tation as a stress response and the regulation of evolvability. Crit.Rev. Biochem. Mol. Biol. 42: 399–435.

Gilson, E., J. M. Clement, D. Brutlag and M. Hofnung, 1984 Afamily of dispersed repetitive extragenic palindromic DNA se-quences in E. coli. EMBO J. 3: 1417–1421.

Hall, B. G., 1998 Adaptive mutagenesis: a process that generatesalmost exclusively beneficial mutations. Genetica102–103: 109–125.

Hastings, P. J., H. J. Bull, J. R. Klump and S. M. Rosenberg,2000 Adaptive amplification: an inducible chromosomal insta-bility mechanism. Cell 103: 723–731.

Hastings, P. J., J. R. Lupski, S. M. Rosenberg and G. Ira,2009 Mechanisms of change in gene copy number. Nat. Rev.Genet. 10: 551–564.

Hendrickson, H., E. S. Slechta, U. Bergthorsson, D. I. Anderssonand J. R. Roth, 2002 Amplification-mutagenesis: evidence that‘‘directed’’ adaptive mutation and general hypermutability resultfrom growth with a selected gene amplification. Proc. Natl. Acad.Sci. USA 99: 2164–2169.

Horiuchi, T., S. Horiuchi and A. Novick, 1963 The genetic basisof hyper-synthesis of b-galactosidase. Genetics 48: 157–169.

Kofoid, E., U. Bergthorsson, E. S. Slechta and J. R. Roth,2003 Formation of an F9 plasmid by recombination betweenimperfectly repeated chromosomal Rep sequences: a closer lookat an old friend (F9(128) pro lac). J. Bacteriol. 185: 660–663.

Kugelberg, E., E. Kofoid, A. B. Reams, D. I. Andersson and J. R.Roth, 2006 Multiple pathways of selected gene amplificationduring adaptive mutation. Proc. Natl. Acad. Sci. USA 103:17319–17324.

Leach, D. R., 1994 Long DNA palindromes, cruciform structures,genetic instability and secondary structure repair. Bioessays 16:893–900.

Leach, D. R., E. A. Okely and D. J. Pinder, 1997 Repair by recom-bination of DNA containing a palindromic sequence. Mol. Micro-biol. 26: 597–606.

Lederberg, J., and E. M. Lederberg, 1952 Replica plating and in-direct selection of bacterial mutants. J. Bacteriol. 63: 399–406.

Lee, J. A., C. M. Carvalho and J. R. Lupski, 2007 A DNA replicationmechanism for generating nonrecurrent rearrangements associ-ated with genomic disorders. Cell 131: 1235–1247.


Li, H., J. Ruan and R. Durbin, 2008 Mapping short DNA sequenc-ing reads and calling variants using mapping quality scores.Genome Res. 18: 1851–1858.

Lind, P., C. Tobin, O. G. Berg, C. G. Kurland and D. I. Andersson,2010 Compensatory gene amplification restores fitness afterinter-species gene replacements. Mol. Microbiol. 75: 1078–1089.

Lovett, S. T., 2004 Encoded errors: mutations and rearrangementsmediated by misalignment at repetitive DNA sequences. MolMicrobiol 52: 1243–1253.

Luria, S. E., and M. Delbruck, 1943 Mutations of bacteria fromvirus sensitivity to virus resistance. Genetics 28: 491–511.

Nilsson, A. I., A. Zorzet, A. Kanth, S. Dahlstrom, O. G. Berg et al.,2006 Reducing the fitness cost of antibiotic resistance by ampli-fication of initiator tRNA genes. Proc. Natl. Acad. Sci. USA 103:6976–6981.

Persky, N. S., and S. T. Lovett, 2008 Mechanisms of recombina-tion: lessons from E. coli. Crit. Rev. Biochem. Mol. Biol. 43:347–370.

Pinder, D. J., C. E. Blake, J. C. Lindsey and D. R. Leach,1998 Replication strand preference for deletions associatedwith DNA palindromes. Mol. Microbiol. 28: 719–727.

Reams, A. B., E. Kofoid, M. Savageau and J. R. Roth, 2010 Dupli-cation frequency in a population of Salmonella enterica rapidlyapproaches steady state with or without recombination. Genetics184: 1077–1094.

Ross, D. G., J. Swan and N. Kleckner, 1979 Nearly precise exci-sion: a new type of DNA alteration associated with the translocat-able element Tn10. Cell 16: 733–738.

Roth, J. R., E. Kofoid, F. P. Roth, O. G. Berg, J. Seger et al.,2003 Regulating general mutation rates: examination of thehypermutable state model for Cairnsian adaptive mutation.Genetics 163: 1483–1496.

Roth, J. R., E. Kugelberg, A. B. Reams, E. Kofoid andD. I. Andersson,2006 Origin of mutations under selection: the adaptive muta-tion controversy. Annu. Rev. Microbiol. 60: 477–501.

Sandegren, L., and D. I. Andersson, 2009 Bacterial gene amplifi-cation: implications for the evolution of antibiotic resistance.Nat. Rev. Microbiol. 7: 578–588.

Seifert, H. S., and R. D. Porter, 1984 Enhanced recombinationbetween lambda plac5 and mini-F-lac: the tra regulon is requiredfor recombination enhancement. Mol. Gen. Genet. 193: 269–274.

Sharp, P. M., and D. R. Leach, 1996 Palindrome-induced deletion inenterobacterial repetitive sequences. Mol. Microbiol.22: 1055–1056.

Slack, A., P. C. Thornton, D. B. Magner, S. M. Rosenberg and P. J.Hastings, 2006 On the mechanism of gene amplification in-duced under stress in Escherichia coli. PLoS Genet. 2: e48.

Slechta, E. S., J. Liu, D. I. Andersson and J. R. Roth, 2002 Evidencethat selected amplification of a bacterial lac frameshift allele stim-ulates Lac(1) reversion (adaptive mutation) with or without gen-eral hypermutability. Genetics 161: 945–956.

Slechta, E. S., K. L. Bunny, E. Kugelberg, E. Kofoid, D. I. Anderssonet al., 2003 Adaptive mutation: general mutagenesis is not aprogrammed response to stress but results from rare coamplifi-cation of dinB with lac. Proc. Natl. Acad. Sci. USA 100: 12847–12852.

Sonti, R. V., and J. R. Roth, 1989 Role of gene duplications in theadaptation of Salmonella typhimurium to growth on limiting car-bon sources. Genetics 123: 19–28.

Stern, M. J., G. F. Ames, N. H. Smith, E. C. Robinson and C. F. Higgins,1984 Repetitive extragenic palindromic sequences: a major com-ponent of the bacterial genome. Cell 37: 1015–1026.

Sun, S., O. G. Berg, J. R. Roth and D. I. Andersson,2009 Contribution of gene amplification to evolution of in-creased antibiotic resistance in Salmonella typhimurium. Genetics182: 1183–1195.

Syvanen, M., J. D. Hopkins, T. J. Griffin IV, T. Y. Liang, K. Ippen-Ihler et al., 1986 Stimulation of precise excision and recombi-nation by conjugal proficient F9plasmids. Mol. Gen. Genet. 203:1–7.

Tlsty, T. D., A. M. Albertini and J. H. Miller, 1984 Gene ampli-fication in the lac region of E. coli. Cell 37: 217–224.

Vasconcelos, A. T., M. A. Maia and D. F. de Almeida, 2000 Shortinterrupted palindromes on the extragenic DNA of Escherichiacoli K-12, Haemophilus influenzae and Neisseria meningitidis.Bioinformatics 16: 968–977.

Warren, G. J., and R. L. Green, 1985 Comparison of physical andgenetic properties of palindromic DNA sequences. J. Bacteriol.161: 1103–1111.

Zieg, J., and R. Kolter, 1989 The right end of MudI(Ap,lac). Arch.Microbiol. 153: 1–6.

Communicating editor: J. Lawrence


The Tandem Inversion Duplication in Salmonella enterica...

Documents

Transcript of The Tandem Inversion Duplication in Salmonella enterica...