Emergence of antibiotic resistance from multinucleatedbacterial filamentsJulia Bosa, Qiucen Zhangb, Saurabh Vyawaharea, Elizabeth Rogersc, Susan M. Rosenbergc, and Robert H. Austina,1

aDepartment of Physics, Princeton University, Princeton, NJ 08544-0708; bDepartment of Physics, University of Illinois at Urbana–Champaign, Urbana, IL 61801-3080; and cDepartments of Molecular and Human Genetics, Biochemistry and Molecular Biology, and Molecular Virology and Microbiology, and theDan L. Duncan Cancer Center, Baylor College of Medicine, Houston, TX 77030

Contributed by Robert H. Austin, November 3, 2014 (sent for review August 26, 2014)

Bacteria can rapidly evolve resistance to antibiotics via the SOSresponse, a state of high-activity DNA repair and mutagenesis. Weexplore here the first steps of this evolution in the bacteriumEscherichia coli. Induction of the SOS response by the genotoxicantibiotic ciprofloxacin changes the E. coli rod shape into multichro-mosome-containing filaments. We show that at subminimal inhib-itory concentrations of ciprofloxacin the bacterial filament dividesasymmetrically repeatedly at the tip. Chromosome-containing budsare made that, if resistant, propagate nonfilamenting progeny withenhanced resistance to ciprofloxacin as the parent filament dies. Wepropose that the multinucleated filament creates an environmentalniche where evolution can proceed via generation of improvedmutant chromosomes due to the mutagenic SOS response and pos-sible recombination of the new alleles between chromosomes. Ourdata provide a better understanding of the processes underlyingthe origin of resistance at the single-cell level and suggest an anal-ogous role to the eukaryotic aneuploidy condition in cancer.

antibiotic resistance | SOS response | filamentation | mutation | evolution

Bacteria can evolve remarkably quickly to be resistant toantibiotics under certain conditions, as we showed in an

earlier paper (1) using microfabrication techniques to createa structured environment of bacteria in a gradient of the genotoxicantibiotic ciprofloxacin (cipro), which perturbs chromosome repli-cation (2). The need for the structured environment in which steepgradients facilitate the rapid evolution of resistance by multiplemutations confirmed a prediction from recent modeling (3). Ourearlier work on the rapid emergence of cipro resistance in amicrofabricated complex ecology did not delineate the stages in theemergence of resistance but only presented the ultimate, resistantcells (1). However, high-resolution imaging of the cells within thedevice before the full emergence of resistance showed that the cellswent through a transient filamentation phase. Fig. 1 shows imagesfrom our microecology that Escherichia coli bacteria have changedtheir shape into long filamentous structures shortly after exposureto the cipro antibiotic. The filamentation response is not perma-nent, because after 4 h some cells have reverted to a normal phe-notype (Fig. 1B) and have become resistant to high levels of theantibiotic by de novo mutation in genes encoding gyrase A top-oisomerase and efflux pumps (1). We wished to explore the rolefilamentation played in the initial stages of antibiotic resistance.Many microorganisms adopt a filamentous shape caused by

cell-division arrest yet continue cell-volume growth in responseto a variety of stressful environments, including nutrient de-ficiency (4), extensive DNA damage through the SOS responsepathway (5, 6), host innate immune responses (7), desiccation(8), high pressure (9), and antibiotic treatment (10–12). Al-though filamentation has long been considered an overstressedand/or sick phenotype or solely for completion of DNA repli-cation before division (13), it has only recently been describedas a more general survival strategy (7, 8, 14). For example, if thestress is relieved quickly enough, division of the filamentous cellresumes synchronously and rapidly at regular intervals along theentire length of the filament, resulting in multiple viable normal-

sized offspring cells (8). However, the mechanisms underlying theevolution of filamentous bacteria when the selective pressure ismaintained have been poorly described, in part because most studieshave used stress conditions that were evolutionarily unfavorable andinstead led to irreversible bacterial growth arrest and/or rapid death.Bacterial cell-division arrest resulting from inhibition of DNA

replication after cipro exposure (2) is caused by the induction ofa division inhibitor (e.g., SulA in E. coli) (13, 15). sulA gene in-duction is part of a set of over 30 induced genes involved in DNArepair pathways and error-prone mechanisms, collectively knownas the SOS response (5, 16, 17), a sequential transcriptionalprocess that is tightly controlled by two master regulators, LexAand RecA proteins (18). When the replication forks are stalled,LexA autoproteolytic cleavage facilitated by RecA protein ini-tiates the SOS stress pathway until the chromosome damage hasbeen repaired and replication of the chromosome resumes.However, when the SOS response is fully engaged, the inductionof low-fidelity replication polymerases known as error-pronepolymerases increases the rate of mutation μ during DNA rep-lication from the normally low value of 10−9 mutations per basepair per generation up to a high rate of 10−5 mutations per basepair per generation (19), so that a single bacterial chromosomesuch as the 4.6-Mb E. coli chromosome can be expected to havemultiple mutations with each round of replication. Current re-search aims at targeting the error-prone polymerase pathway toslow down the evolution of antibiotic resistance and thus im-prove the long-term viability of some antibiotic drugs (20).Here we address how filamentation is probably the first step in

the evolution of resistance to mutagenic antibiotic exposure,internally generating mutant chromosomes via the SOS responseuntil a solution is obtained: a mutant chromosome with re-sistance to the stress. We examine how individual bacteria exitfrom the filamentous state caused by subminimal inhibitoryconcentrations of the genotoxic antibiotic ciprofloxacin. Weshow that when cipro is maintained at low levels, E. coli filaments

Significance

Understanding how bacteria rapidly evolve under antibiotic se-lective pressure is crucial to controlling the development of re-sistant organisms. We show that initial resistance emerges fromsuccessful segregation of mutant chromosomes at the tips offilaments followed by budding of resistant progeny. We pro-pose that the first stages of emergence of resistance occur viathe generation of multiple chromosomes within the filamentand are achieved by mutation and possibly recombination be-tween the chromosomes.

Author contributions: J.B., Q.Z., S.V., E.R., S.M.R., and R.H.A. designed research; J.B., Q.Z.,and R.H.A. performed research; J.B., E.R., and S.M.R. contributed new reagents/analytictools; J.B., Q.Z., S.M.R., and R.H.A. analyzed data; and J.B., S.M.R., and R.H.A. wrotethe paper.

The authors declare no conflict of interest.1To whom correspondence should be addressed. Email: austin@princeton.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1420702111/-/DCSupplemental.

178–183 | PNAS | January 6, 2015 | vol. 112 | no. 1 www.pnas.org/cgi/doi/10.1073/pnas.1420702111

Dow

nloa

ded

by g

uest

on

June

16,

202

0

undergo asymmetric divisions at the filament tips, giving rise tobudded resistant offspring cells. The generation of multiple mutantchromosomes that are successfully segregated toward the fila-ments tips favors the birth of antibiotic-resistant cells. We discusshow this filamentation step is a precursor to the final highly re-sistant bacteria in a complex ecology.

ResultsWe focused on the role that the SOS mutational network, trig-gered by the presence of stalled replication forks, plays in thefilamentation process and bacterial evolution of drug resistance.The transient filamentation events were studied on agar platescontaining cipro at varying levels. The primary experiment in-volves spreading a low-density population of less than 103 permm2 of E. coli bacteria in the presence of 0.125 of the minimalinhibitory antibiotic concentration (0.125× MIC) of cipro. Time-lapse imaging revealed that 99.5% of the bacteria formed longfilaments (up to 200 μm) and expanded their length exponen-tially (Fig. 2A; see Movies S1 and S2 for videos of the growth offilaments). The remaining 0.5% of the bacteria that did notelongate or divide in the presence of the antibiotic may be per-sister cells (21) or dead cells. A lexA3 mutant strain that cannotactivate the SOS response and a recB mutant strain, whichreduces SOS response induction, general recombination, andDNA repair, were not able to filament or grow at these low ciproconcentrations, verifying that both filamentation and growth relyon the induction of the LexA-dependent SOS response andpossibly on recombination and repair pathways (SI Appendix,Fig. S1A shows the growth of filaments over a 3-h time periodand the failure of the lexA3 and recB mutant strains to growunder the same conditions). We observed that the growth ofa filament is transiently interrupted by asymmetric divisionevents (Fig. 2B and Movies S1 and S2) that occur repeatably andpredominantly at the tips of the filament (Fig. 2C) and give riseto short offspring cells that often resumed normal division (Fig.2B and Movies S1 and S2). This budding process was inherentlydistinct from the synchronous, symmetrical division recoveryprofile of the filament obtained shortly after drug removal(SI Appendix, Fig. S1).Interestingly, some degree of heterogeneity was observed in

the bud fate at this low concentration (0.125× MIC) (Fig. 3A):35% of budded cells divide symmetrically without further fila-mentation, indicating that they are now resistant to 0.125× MICcipro; 35% cease division and growth; 10% of the buds showabnormal asymmetrical division (minicell formation) with littlepropagation; and 20% resume filamentation, indicating that theydo not show evidence of cipro resistance. Buds that fail to divideand propagate may either contain a damaged chromosome ormay not contain a chromosome at all [we found that 11% of thebuds do not contain a chromosome, by using 4′,6-diamidino-2-phenylindole (DAPI) staining]. Conversely, the fate of the

filaments after prolonged growth on low cipro showed less het-erogeneity than that of the buds, since the bulk of parental fil-aments (70%) undergo death as evidenced by their inability toexclude the propidium iodide (PI) dye, reflecting loss of possiblyboth membrane integrity and efflux mechanisms (Fig. 3B). Thesefindings suggest that filamentation at low cipro is a complexphenomenon that gives rise to short progeny and eliminates theunfit filamentous parent cell.We were most interested by the buds that divide symmetrically

without further filamentation and examined whether these budsshow a lower level of SOS response. We used a sulA-gfp tran-scriptional reporter in a nongenic chromosomal location in cellswith an intact native sulA gene (22) to quantitatively measureinduction of the SOS response within the bacterial filament. Weobserved a decrease in the intensity of the gfp expression withinbuds, indicating that buds had decreased expression of the SOSregulon, as would be expected from the observed increase incipro resistance in the buds and subsequent suppression of theSOS response [Fig. 3 shows that the buds have on average about25% less expression of PsulA-gfp than the filaments, whereas theintensity of constitutively expressed RFP protein (lac-mrfp1)shows identical levels both in the buds and in the filaments, in-dicating that protein expression levels are similar in the buds andwithin the filaments].Because most buds do contain a chromosome, although cipro

directly affects DNA replication by targeting DNA gyrases (23)and thus perturbs chromosome dynamics (2), we wished to ex-amine both the content and the spatial arrangement of thechromosomes in filamentous cells to understand how low dosesof cipro allow the generation of short viable nucleated offspring.We used standard DAPI staining and live imaging of a specificchromosomal locus, R2 (17), as a reporter of an individualchromosomal copy (SI Appendix, Fig. S2) to enhance chromo-some counting. R2 loci were visualized by the binding of fluo-rescent TetR-CFP fusions to the tetO operator array located21 kb upstream of the lacZ gene (24) (SI Appendix, Fig. S2).Individual chromosomes localize as discrete CFP foci. We found

A B

Fig. 1. (A) E. coli bacteria form filaments in microhabitats after 3 h of ex-posure to a ciprofloxacin gradient. (B) Cipro-resistant bacteria that havereverted to a normal-length phenotype (arrows) in one of the microhabitats.

A C

B

Fig. 2. (A) Semilog plot of the contour length of a collection of filaments(n = 20) growing on agar containing 0.125× MIC of cipro vs. time. (B) Exampleof a budding process indicated by the white arrows at either end of a filament.Yellow arrows show budded cells resuming normal division. (Scale bar, 5 μm.)(C) Distribution of budding positions along the filaments (n = 95 cells, averagelength 42 ± 16 μm, average time at division 332 ± 71 min).

Bos et al. PNAS | January 6, 2015 | vol. 112 | no. 1 | 179

EVOLU

TION

Dow

nloa

ded

by g

uest

on

June

16,

202

0

that the filaments contain multiple chromosomes (Fig. 4 and SIAppendix, Fig. S2) that are partially or completely segregated,with a linear relation between filament length and chromosomenumber (Fig. 4A), indicating that as the filament grows it repli-cates multiple chromosomes. The presence of multinucleatedfilaments occurred only at a low level of cipro (0.125× MIC) (SIAppendix, Fig. S2C). At levels of cipro above 0.125× MIC, thechromosome morphology was severely impaired (SI Appendix,Fig. S2C) and events of budding at the filament tips were rare ornonexistent. At low levels of cipro, we found that some chro-mosomes localize near the tips (three-quarter positions of celllength) (Fig. 4B), coinciding with the position of the division sitesat the tips (Fig. 1C). Time-lapse experiments showed heteroge-neous dynamics of the chromosomes within the filaments wheresome chromosomes move tipward as the filaments grow whereasother remain essentially stationary in the midcell regions (Fig. 4C and D; SI Appendix, Fig. S3; and Movie S3). In addition, twoindividual chromosomes were eventually observed to alternatebetween overlapping in a single cluster and then separating intotwo distinct foci (Fig. 4 C and D and SI Appendix, Fig. S3). Wealso noticed that the chromosomes within the same filamentundergo replication or separation of foci in an asynchronous way(Fig. 4 C and D). All together, the chromosome dynamics incipro-stressed filaments are complex and do not indicate simplereplication, but nevertheless the filamentous cell succeeds incoordinating chromosome placement and asymmetric division atthe tips, leading to the birth of resistant daughter cells.We verified that the offspring evolved resistance in that

normal-sized and dividing bacterial buds arose at the tips of thefilaments under continued low cipro treatment, and SOS ex-pression levels, as determined by the sulA-gfp reporter, were low.Wild-type bacteria were cultured in low cipro and filtered usinga sterile 5-μm syringe filter to isolate normal-sized progeny fromfilamentous parent cells (SI Appendix, Fig. S4). About 70 ± 11%

of the isolated progeny survived when reexposed to low cipro(0.125× MIC). The frequency of progeny resistant to high levelsof cipro (1× MIC) was determined and defined as the number ofcipro-resistant colonies that arose after 24 h on solid mediacontaining 1× MIC cipro per viable cell. Progeny showed a 250-fold increase in resistance frequency [2.8 (±0.3) × 10−6] at 1×MIC cipro compared with that of the parent cells [7.2 (±3) × 10−8]that had not been preexposed to low cipro. This result indi-cates that a subpopulation of budded cells has indeed evolvedresistance during the 24-h exposure to low cipro. We assumedthat the frequency of persisters was kept low in the progenypopulation, even after filtration, because progeny either formedfilaments or resumed normal division when exposed over againto low cipro. Instead, persisters would have resumed division andgrowth only once the antibiotic is removed (21).Resistant clones were isolated by selection using subinhibitory

(0.5× MIC) or inhibitory (1× MIC) doses of cipro, and a subsetof genomic sequences of known resistant genes was analyzedby using targeted genome sequencing (Table 1 and SI Appendix,Table S2). The results showed that the level of resistance wasa function of the cipro concentration at selection. Resistance atthe bottom of the selection window was associated with over-expression of AcrAB efflux pumps, in that mutations in marRwere found and none of the other genes sequenced had muta-tions. Deletion(s) of 1 or 11 nt in the marR gene encoding theMarR efflux pump regulator were identified in 37% of the se-quenced clones. Conversely, at the top of the selection window,the resistant clones isolated were mainly mutated in the GyrAgyrase subunit, and 87% of the sequenced clones had a single

BA

C D E

Fig. 3. (A) Distribution of the different fates of the buds: normal symmet-rical division (red), asymmetrical division (yellow), no division and no growth(black), and filamentation (gray). (B) Death of long filaments after budding,imaged after 5 h (Top) and 22 h (Bottom) of low cipro exposure using themembrane-impermeable dye propidium iodide. Bright-field (BF), GFP, andRFP (PI) images are shown. (Scale bar, 2 μm.) (C) Microscopic images of fil-aments and buds expressing sulA-gfp (Top) and constitutively expressed rfp(Bottom) in the same cells after 4 h of growth in low cipro. (Scale bar, 2 μm.)(D) Histogram of average sulA-gfp expression within filaments (n = 98 cells).The red line represents a GFP mean intensity of 528 ± 122 of the population.(E) Histogram of average sulA-gfp expression within buds (n = 110 cells). Thered line represents a GFP mean intensity of 393 ± 107 of the population.

A

C D

B

Fig. 4. (A) Chromosome count as a function of length as measured by DAPIstain (red crosses) and R2-CFP loci (blue crosses). (B) Distribution of individualchromosome (R2-CFP) positions along the filaments after 4 h of growth with0.125× MIC cipro (n = 50 cells, mean R2 locus counts per cell 4.5 ± 1.9, av-erage filament length 20 ± 4 μm). (C) Time-lapse images of R2-CFP showingcomplex movement of individual chromosomes along the filament length.(Scale bar, 1 μm.) Each red circle surrounding an R2 locus denotes an individualchromosome and results from automated tracking analysis software (Mate-rials and Methods). White arrowheads indicate events of merging/splittingbetween two chromosomes. The yellow arrowhead indicates a chromosomeduplication/separation event. Dashed lines show the cell contours. (D) Distri-bution of R2-CFP locus positions in individual cells shown in C. Traces wereobtained by using a MATLAB-based particle tracking methodology; auto-mated (filled circle) or manual tracking (open circles). (Materials and Meth-ods). The black arrow indicates a replication/separation event near the tip.Gray arrows show merging and splitting events between two chromosomes.

180 | www.pnas.org/cgi/doi/10.1073/pnas.1420702111 Bos et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

0

mutation in gyrA (GyrA S83A), which is located in the quinoloneresistance-determining region, a hot spot region for mutations(25). Mutations in gyrA or marR are likely to be sufficient topromote intermediate-level resistance of the tested clones (Fig.5). The rest of the cipro-resistant clones presumably carry a re-sistance-conferring mutation(s) at a place not sequenced.The MIC of resistant progeny was 2.5× increased (10 clones

analyzed) but remained five times lower than that of the high-level resistant strain QZ030, which evolved four mutations bothin gyrA and the efflux pumps when cultured in a complex envi-ronment (1). Although the budded progeny show enhanced re-sistance to cipro, they do not show the full resistance observed inthe cells that grow in the microfabricated devices, indicatingthat the budded progeny are part-way along the path to a fullyenhanced resistance (Fig. 5 compares the resistance of filteredbud progeny to cipro with that of the strain of bacteria QZ030retrieved from our previous experiment, where evolution to re-sistance to cipro emerged in a complex gradient in a meta-population, as opposed to strains extracted from this experiment,where evolution occurs in a fixed concentration of cipro). In all,the data reveal that a low level of adaptive evolution to low ciprocan occur in a few hours within a single filament.

DiscussionOur data show that there are five fundamental features thatcharacterize the growth of the bacterial filaments during exposureto subminimal inhibitory concentrations of cipro (Fig. 6): (i) Ex-posure to 0.125×MIC cipro initiates the SOS response (Fig. 3C andFig. S1); (ii) the filament contour length L(t) increases exponen-tially with time (Fig. 2A); (iii) the filaments contain multiple copiesof the bacterial chromosome (Fig. 4); (iv) the filamenting process istransiently interrupted by a budding process that generates resistantprogeny at the tips (Fig. 1 and Movies S1 and S2); and (v) mostfilaments die, but not some of the budded progeny (Fig. 3B).Such an evolutionary path to antibiotic resistance encom-

passes two remarkable events that occur during the filamentationprocess: the presumably local decision to transiently exit fromSOS and bud a resistant daughter cell from the end of the fila-ment, and the generation of multiple chromosomes that evolveresistance via mutagenesis and possibly recombination withina single filament.The budding events are presumably triggered by the local

cessation of the SOS response in the vicinity of an evolved re-sistant chromosome with the mutations found in Table 1, be-cause the SOS response is responsible for the filamentationphenotype caused by a lack of septation. It could be either thatbetter-adapted chromosomes are generated near the tips or thatbetter-adapted chromosomes generated internally migrate toward

the tips, where they are then budded off by formation of a singleseptum. We can guess whether the cessation of the SOS responseis a global or local event in the cell by estimating how far the SOSdivision-inhibitor SulA can diffuse during the replication time ofa chromosome, about 30 min in our case. SulA has a molecularmass of about 18 kDa and measurements of the diffusion co-efficient of similar-size GFP proteins in E. coli have yielded anaverage value for the diffusion constant of about 7 μm2/s (26),and hence we would expect that in 30 min the SulA proteinshould diffuse about 150 μm, easily the length of a filament.However, the budding process is local and not evenly distributedalong the length of the filament, indicating that either SulA doesnot diffuse freely in the cytoplasm or other players such as theMin division control proteins are determining the local decisionto bud. Identifying the molecular actors that govern the decisionof local budding will be of valuable interest to answer thequestion of why asymmetric division is important to generatingresistant progeny.This view differs from current paradigms for cell-cycle

checkpoint control in bacteria and other organisms. Cell fila-mentation during SOS is thought to reflect the need to completeDNA replication before cell division to protect chromosomes afterDNA damage (13). Similarly, filamentation allows repair of DNAdamage by recombination between chromosomes by keepingmore than one chromosome per cell until repair is completed(27). Similar DNA-protection roles underlie eukaryotic cell-cyclecheckpoint control (28). We suggest that in addition to DNApreservation, cell filamentation may accelerate evolution byadding recombination, which is possible only with more than onechromosome per cell, to the increased mutagenesis documentedduring SOS. Recombination between chromosomes can potentiallyseparate beneficial mutations from linked deleterious mutations orunrepaired DNA damage. The dramatic acceleration of evolutionby recombination accompanying mutagenesis, relative to muta-genesis without recombination, is supported by modeling (29, 30).Coupling of mutagenesis with recombination is observed

throughout life, including being predicted in cancer genomes(31, 32), and observed in the immune system (33), yeast (34, 35)and bacteria (17, 36–38). Because increased mutagenesis cre-ates both deleterious and beneficial mutations, recombinationthat unlinks these can be beneficial.

Table 1. Targeted-sequencing results showing genes mutatedin resistant progeny selected at 0.5× MIC and 1× MIC cipro

Gene

Cipro atselection,ng/mL

No. of mutants(base substitution)

No. of mutants(base deletion)

gyrA 20 0/16 0/16gyrA 40 7/8 0/8marR 20 0/8 3/8 (marR Δ 1, Δ 11)marR 40 0/8 0/8acrR 20 0/8 0/8acrR 40 0/8 0/8rbsA 20 0/8 0/8rbsA 40 0/8 0/8sulA 20 0/4 0/4

The number of mutants over the number of sequenced clones (e.g., 4, 8,or 16) is indicated.

Fig. 5. Resistance of the wild-type strain, a QZ030 hyperresistant isolate (1),and three different progeny clones to cipro as measured by growth curves.The cipro concentration at which growth was monitored is indicated.

Bos et al. PNAS | January 6, 2015 | vol. 112 | no. 1 | 181

EVOLU

TION

Dow

nloa

ded

by g

uest

on

June

16,

202

0

Further, the chromosomes display a distinct dynamics patternin the same filament. Events of merging/splitting between twochromosomes were frequently observed during the experiment(Fig. 4C; SI Appendix, Fig. S3; and Movie S2), possibly reflectingthe intimate interplay between DNA replication and geneticinformation exchange during cipro exposure. In addition, repli-cation of sister loci did not occur synchronously, as shown inFig. 4C, where one chromosome out of six was captured whilereplicating, and suggests potential differential DNA damageload between the chromosomes (39). Although we found theseobservations captivating, it remains unclear to us why there issuch heterogeneity in the chromosome dynamics. One canspeculate that the nonstationary chromosomes are the chromo-somes with increased fitness (probably containing mutationsconferring cipro resistance) that would not be eliminated butchosen for replication. Instead, stationary and/or nonreplicatingchromosomes that are likely to contain neutral or deleteriousmutations would eventually be eliminated in the dying filamentor would recombine to separate beneficial mutations from linkeddeleterious damage and hence be chosen for survival (Fig. 3B).Therefore, the multiple chromosomes that are generated in thefilament could enhance the chances of producing resistantchromosomes as well as allow repair. The ultimate decision torelease the progeny would be done by pinching off the successfulmutant chromosome that is resistant to the stress by local loss ofthe SOS response. Further in-depth mechanistic studies of theinterplay between processes of stress-induced mutagenesis, re-combination, chromosome partitioning, and local budding areneeded to deepen our understanding of how antibiotic resistancestems from a single multinucleated bacterium. The chromosomedynamics in an elongating filament might possibly resemblea form of Turing machine (40), where the “computation” of fit-ness is like the writing of mutations on a chromosome and erasureof marks would correspond to the elimination of chromosomesthat were less fit and to recombination between chromosomesreverting deleterious new mutations, and the halting of theTuring machine upon solution of the problems would be repre-sented by the pinching off of the mutant, more fit chromosome.However, this is highly speculative at this point in our work.Perhaps a phenomenon similar to bacterial multichromosome

resistance occurs in aneuploid cells. For instance, cancer cellstypically contain 60–90 chromosomes (41). Aneuploidy is aubiquitous feature of cancer that may contribute to a more rapidaccumulation of mutations, increasing survival and proliferation(42–45). Therefore, the strategy of generating multiple mutantchromosomes within a single cell may represent a widespreadand conserved mechanism for the rapid evolution of genomechange in response to unfavorable environments (i.e., chemo-therapy drugs and antibiotics).

Materials and MethodsImaging. Long time-lapse movies (images recorded every 2 min for 10 h) oflow-density bacteria [JEK1036 (gfp+)] growing on LB agar pads containing5 ng/mL cipro and isopropyl β-D-1-thiogalactopyranoside (IPTG) were madeusing a Nikon 90i microscope equipped with a Nikon Plan Apo 60× objectiveand a Rolera-XR cooled CCD camera and processed with Micro-Manager(NIH, www.micro-manager.org) and ImageJ (NIH) software. All other imageswere collected on a Nikon Eclipse 90 microscope equipped with a Nikon PlanApo 100× objective and an Andor Neo camera and processed with NikonNIS-Elements software. In all experiments, cells were transferred from liquidculture to 1.4% agarose-padded slides containing LB medium (or 0.5× LB forstrain AB1157), with low ciprofloxacin and IPTG when required. A coverslipwas placed on the pad and then sealed with valap (1 vol vaseline/1 vol lanolin/1 vol paraffin). Filament death was visualized by fluorescence microscopyusing propidium iodide, a membrane-impermeable nucleic acid stain com-monly used as a cell-death marker. Nucleoids in the filaments were visualizedby adding DAPI (live-cell DNA staining) directly to the agarose pads. R2 lociwere visualized in E. coli AB1157 cells placed on 1.4% agarose-padded slidescontaining 0.5× diluted LB medium (to reduce fluorescence background).Images were recorded every 2 min right after the cells were placed on theagarose pad in the no-cipro conditions, or every 5 min after the cells wereallowed 2 h of growth on the agarose pad in the low-cipro conditions. See SIAppendix, Materials and Methods for details of experimental protocols.

Bacterial Strains and Growth. E. coli strains were cultured at 37 °C in liquid LBmedium and on an LB agar (1.7%) plate (solid), unless noted. In all experi-ments, cipro antibiotic was added at a subminimal inhibitory concentrationof 5 ng/mL (0.125× MIC) referred to as “low cipro,” unless indicated (seeSI Appendix, Materials and Methods for details of strain growth). Strainsand primers used in this study are listed in SI Appendix, Tables S1 andS2, respectively.

Minimum Inhibitory Concentration Determination. The MIC defined as thelowest concentration of ciprofloxacin that prevents any growth of thebacteria is known to be 40 ng/mL. For each strain, a volume of 3 mL ofovernight culture was used to inoculate, in triplicate, 150 μL of LB mediumcontaining increasing concentrations of ciprofloxacin (0, 5, 10, 15, 20, 25,30, and 40 ng/mL) using 96-well flat-bottom plates. Growth was moni-tored by reading the absorbance at 630 nm every 20 min for 10 h of in-cubation at 37 °C with shaking in a ChroMate 4300 Microplate Reader(Awareness Technology). MICs for ciprofloxacin-resistant offspring clones(10 clones analyzed) were determined using the same conditions describedabove; in addition, wells contained LB plus ciprofloxacin at 50, 100, 200,and 500 ng/mL.

Determination of Cipro-Resistant Mutant Frequency. The frequency of thecipro-resistant (ciproR) mutant was determined and defined as the number ofciproR colonies that arose after 24 h of growth on solid media containing 1×MIC cipro, per viable cell. Three independent cultures of wild-type strainswere subcultured at 1:100 dilution from an overnight starter culture andgrown to midexponential phase. A volume of 0.1 mL of cell culture wasplated on LB agar containing low cipro antibiotic and incubated at 37 °C for24 h. The bacterial lawn was resuspended in 5 mL LB using a sterile scraper.The resulting cell culture contained a mixed population of filaments andshort bud offspring. When needed, the cell culture was recovered by fil-tration through a 5-μm sterile syringe filter to harvest the progeny andremove the filamentous cells. All cultures (mixed-population or filteredprogeny) were serially diluted, and 0.1-mL samples of appropriate dilutionswere plated onto nonselective plates to determine total viable cell counts, aswell as on plates containing cipro 1× MIC (40 ng/mL) to score high-levelciproR mutants. CiproR frequencies were calculated by dividing the numberof high-level cipro-resistant mutants by the total number of viable bacteria.A total of three experiments was conducted independently. Means of mu-tant frequency are given in the text and the ”±” values represent SD.

Sequencing. Cipro-resistant colonies were selected on LB plates containingeither cipro 0.5× MIC (20 ng/mL) or 1× MIC (40 ng/mL), after 24 h incubationat 37 °C. See SI Appendix, Materials and Methods for details of the ex-perimental protocol.

ACKNOWLEDGMENTS. We thank Thomas Silhavy for helpful comments onthe manuscript and Douglas Austin for helping in writing the customMATLAB code. This project was supported by National Cancer InstituteGrant U54CA143803 (to R.H.A.) and National Institutes of Health Grant R01-GM53158 (to S.M.R.).

Fig. 6. Basic model for the initial stages of resistance in filamentous cells.Light gray cells are initial sensitive cells, dark gray cells have double-strandbreaks and stalled replication forks, green cells have evolved moderate an-tibiotic resistance, and red cells are undergoing death. Chromosomes areshown within the cells.

182 | www.pnas.org/cgi/doi/10.1073/pnas.1420702111 Bos et al.

Dow

nloa

ded

by g

uest

on

June

16,

202

0

1. Zhang Q, et al. (2011) Acceleration of emergence of bacterial antibiotic resistance inconnected microenvironments. Science 333(6050):1764–1767.

2. Diver JM (1989) Quinolone uptake by bacteria and bacterial killing. Rev Infect Dis11(Suppl 5):S941–S946.

3. Hermsen R, Hwa T (2010) Sources and sinks: A stochastic model of evolution in het-erogeneous environments. Phys Rev Lett 105(24):248104.

4. Pine L, Boone CJ (1967) Comparative cell wall analyses of morphological forms withinthe genus Actinomyces. J Bacteriol 94(4):875–883.

5. Radman M (1975) SOS repair hypothesis: Phenomenology of an inducible DNA repairwhich is accompanied by mutagenesis. Basic Life Sci 5A:355–367.

6. Bos J, Yakhnina AA, Gitai Z (2012) BapE DNA endonuclease induces an apoptotic-likeresponse to DNA damage in Caulobacter. Proc Natl Acad Sci USA 109(44):18096–18101.

7. Justice SS, Hunstad DA, Seed PC, Hultgren SJ (2006) Filamentation by Escherichia colisubverts innate defenses during urinary tract infection. Proc Natl Acad Sci USA103(52):19884–19889.

8. Stackhouse RR, Faith NG, Kaspar CW, Czuprynski CJ, Wong AC (2012) Survival andvirulence of Salmonella enterica serovar Enteritidis filaments induced by reducedwater activity. Appl Environ Microbiol 78(7):2213–2220.

9. Zobell CE, Cobet AB (1964) Filament formation by Escherichia coli at increased hy-drostatic pressures. J Bacteriol 87(3):710–719.

10. Rolinson GN (1980) Effect of beta-lactam antibiotics on bacterial cell growth rate.J Gen Microbiol 120(2):317–323.

11. Ryan DM, Monsey D (1981) Bacterial filamentation and in vivo efficacy: A comparisonof several cephalosporins. J Antimicrob Chemother 7(1):57–63.

12. Miller C, et al. (2004) SOS response induction by beta-lactams and bacterial defenseagainst antibiotic lethality. Science 305(5690):1629–1631.

13. Huisman O, D’Ari R (1981) An inducible DNA replication-cell division couplingmechanism in E. coli. Nature 290(5809):797–799.

14. Justice SS, Hunstad DA, Cegelski L, Hultgren SJ (2008) Morphological plasticity asa bacterial survival strategy. Nat Rev Microbiol 6(2):162–168.

15. Bi E, Lutkenhaus J (1993) Cell division inhibitors SulA and MinCD prevent formation ofthe FtsZ ring. J Bacteriol 175(4):1118–1125.

16. Michel B (2005) After 30 years of study, the bacterial SOS response still surprises us.PLoS Biol 3(7):e255.

17. Al Mamun AA, et al. (2012) Identity and function of a large gene network underlyingmutagenic repair of DNA breaks. Science 338(6112):1344–1348.

18. Butala M, Zgur-Bertok D, Busby SJ (2009) The bacterial LexA transcriptional repressor.Cell Mol Life Sci 66(1):82–93.

19. Cirz RT, et al. (2005) Inhibition of mutation and combating the evolution of antibioticresistance. PLoS Biol 3(6):e176.

20. Lee AM, Ross CT, Zeng BB, Singleton SF (2005) A molecular target for suppression ofthe evolution of antibiotic resistance: Inhibition of the Escherichia coli RecA proteinby N(6)-(1-naphthyl)-ADP. J Med Chem 48(17):5408–5411.

21. Lewis K (2010) Persister cells. Annu Rev Microbiol 64:357–372.22. Pennington JM, Rosenberg SM (2007) Spontaneous DNA breakage in single living

Escherichia coli cells. Nat Genet 39(6):797–802.23. Hooper DC (1999) Mechanisms of fluoroquinolone resistance. Drug Resist Updat 2(1):

38–55.

24. Wang X, Lesterlin C, Reyes-Lamothe R, Ball G, Sherratt DJ (2011) Replication andsegregation of an Escherichia coli chromosome with two replication origins. Proc NatlAcad Sci USA 108(26):E243–E250.

25. Yoshida H, Bogaki M, Nakamura M, Nakamura S (1990) Quinolone resistance-determining region in the DNA gyrase gyrA gene of Escherichia coli. AntimicrobAgents Chemother 34(6):1271–1272.

26. Parry BR, et al. (2014) The bacterial cytoplasm has glass-like properties and is fluidizedby metabolic activity. Cell 156(1-2):183–194.

27. Recchia GD, Aroyo M, Wolf D, Blakely G, Sherratt DJ (1999) FtsK-dependent and-independent pathways of Xer site-specific recombination. EMBO J 18(20):5724–5734.

28. Paulovich AG, Toczyski DP, Hartwell LH (1997) When checkpoints fail. Cell 88(3):315–321.

29. Hadany L, Beker T (2003) On the evolutionary advantage of fitness-associated re-combination. Genetics 165(4):2167–2179.

30. Stemmer WP (1994) DNA shuffling by random fragmentation and reassembly: In vitrorecombination for molecular evolution. Proc Natl Acad Sci USA 91(22):10747–10751.

31. Nik-Zainal S, et al.; Breast Cancer Working Group of the International Cancer GenomeConsortium (2012) Mutational processes molding the genomes of 21 breast cancers.Cell 149(5):979–993.

32. Roberts SA, et al. (2012) Clustered mutations in yeast and in human cancers can arisefrom damaged long single-strand DNA regions. Mol Cell 46(4):424–435.

33. Di Noia JM, Neuberger MS (2007) Molecular mechanisms of antibody somatic hy-permutation. Annu Rev Biochem 76:1–22.

34. Magni GE, Von Borstel RC (1962) Different rates of spontaneous mutation duringmitosis and meiosis in yeast. Genetics 47(8):1097–1108.

35. Strathern JN, Shafer BK, McGill CB (1995) DNA synthesis errors associated with dou-ble-strand-break repair. Genetics 140(3):965–972.

36. Sung HM, Yasbin RE (2002) Adaptive, or stationary-phase, mutagenesis, a componentof bacterial differentiation in Bacillus subtilis. J Bacteriol 184(20):5641–5653.

37. Harris RS, Longerich S, Rosenberg SM (1994) Recombination in adaptive mutation.Science 264(5156):258–260.

38. Rosenberg SM, Longerich S, Gee P, Harris RS (1994) Adaptive mutation by deletions insmall mononucleotide repeats. Science 265(5170):405–407.

39. Donley N, Thayer MJ (2013) DNA replication timing, genome stability and cancer: Lateand/or delayed DNA replication timing is associated with increased genomic in-stability. Semin Cancer Biol 23(2):80–89.

40. Turing AM (1995) Lecture to the London Mathematical Society on 20 February 1947.MD Comput 12(5):390–397.

41. Rajagopalan H, Nowak MA, Vogelstein B, Lengauer C (2003) The significance of un-stable chromosomes in colorectal cancer. Nat Rev Cancer 3(9):695–701.

42. Gordon DJ, Resio B, Pellman D (2012) Causes and consequences of aneuploidy incancer. Nat Rev Genet 13(3):189–203.

43. Selmecki AM, Dulmage K, Cowen LE, Anderson JB, Berman J (2009) Acquisition ofaneuploidy provides increased fitness during the evolution of antifungal drug re-sistance. PLoS Genet 5(10):e1000705.

44. Duesberg P, Stindl R, Hehlmann R (2000) Explaining the high mutation rates of cancercells to drug and multidrug resistance by chromosome reassortments that are cata-lyzed by aneuploidy. Proc Natl Acad Sci USA 97(26):14295–14300.

45. Merlo LM, Wang LS, Pepper JW, Rabinovitch PS, Maley CC (2010) Polyploidy, aneu-ploidy and the evolution of cancer. Adv Exp Med Biol 676:1–13.

Bos et al. PNAS | January 6, 2015 | vol. 112 | no. 1 | 183

EVOLU

TION

Dow

nloa

ded

by g

uest

on

June

16,

202

0