Massive Diversification in Aging Colonies of Escherichia coli

Abstract

The evolutionary success of bacteria depends greatly on their capacity to continually generate phenotypic diversity. Structured environments are particularly favorable for diversification because of attenuated clonal interference, which renders selective sweeps nearly impossible and enhances opportunities for adaptive radiation. We examined at the microscale level the emergence and the spatial and temporal dynamics of phenotypic diversity and their underlying causes in Escherichia coli colonies. An important dynamic heterogeneity in the growth, metabolic activity, morphology, gene expression patterns, stress response induction, and death patterns among cells within colonies was observed. Genetic analysis indicated that the phenotypic variation resulted mostly from mutations and that indole production, oxidative stress, and the RpoS-regulated general stress response played an important role in the generation of diversity. We observed the emergence and persistence of phenotypic variants within single colonies that exhibited variable fitness compared to the parental strain. Some variants showed improved capacity to produce biofilms, whereas others were able to use different nutrients or to tolerate antibiotics or oxidative stress. Taken together, our data show that bacterial colonies provide an ecological opportunity for the generation and maintenance of vast phenotypic diversity, which may increase the probability of population survival in unpredictable environments.

INTRODUCTION

Bacteria have colonized practically all environmental niches on Earth. Their adaptability relies greatly on their capacity to continually generate phenotypic diversity, which allows bacterial species to explore the fitness space in a multidirectional way. The ecological and evolutionary forces that govern the generation and maintenance of phenotypic diversity in bacterial communities have undergone intense examination in the last decades. Several studies have emphasized the importance of environmental structure in shaping diversity within bacterial populations (1–3). Structured environments provide microhabitats that are physically diverse and chemically heterogeneous. In existing microhabitats, environmental parameters change continuously, and new microhabitats are created as a consequence of bacterial activity. Such spatially and temporally different selective pressures favor the emergence and coexistence of diverse variants. In contrast, in a homogeneous environment, natural selection reduces diversity by eliminating all but the fittest variants. The high genetic and phenotypic diversity observed in structured environments compared to unstructured environments could contribute to the hardiness of structured bacterial populations by increasing the range of conditions under which the population as a whole can flourish. This notion is analogous to the “insurance hypothesis,” which has been proposed to explain the robustness of ecological communities (4). According to this hypothesis, increasing biodiversity protects ecosystems against declines in functioning due to environmental fluctuations.

In nature, bacteria can be found both in planktonic populations and in spatially structured populations such as chains, mats, biofilms, and colonies (5). However, bacteria favor structured populations, probably because this mode of growth confers high resistance to various types of physical, chemical, and biotic stresses (6). In different habitats, different forms of structured populations predominate due to the presence of specific biotic and abiotic constraints. For example, soil bacteria are found mostly in colonies, while bacteria that live on riverbeds are typically found in biofilms. Bacterial biofilms have been intensively studied because they adversely affect humans in medical and industrial settings (for reviews, see references 7, 8, and 9). Throughout the history of microbiology, bacterial colonies have been used for diagnostic and enumerative ends but have rarely been objects of scientific scrutiny for their own sake (10–15). However, the available data show that there are many similarities between biofilms and colonies. In both structures, bacterial cells are embedded in an exopolymer matrix (14, 16, 17), and both structures provide spatially and temporally heterogeneous environments. For example, as in biofilms, bacterial cell growth in colonies produces gradients of nutrients, waste products, ionic strength, redox potential, pH, and oxygen (16, 18). Moreover, both in aging colonies and in biofilms, extensive diversification due to an increase in mutagenesis has been observed (19–21).

The spatial and temporal dynamics of the ecological forces that govern the generation and maintenance of phenotypic diversity in bacterial colonies at the microscale level have never been thoroughly studied. For example, it is unknown whether tradeoffs are responsible for the presence of diversification within colonies, as has been shown to be the case in Escherichia coli stationary-phase planktonic cultures (22). These planktonic cultures are highly dynamic, that is, successive waves of mutants continuously arise within them while the subpopulation of the parental wild-type strain declines. Eventually, many phenotypically distinct subpopulations evolve and coexist. Most of the mutants in these subpopulations exhibit modified expression of the genes in the general stress response pathway, which are regulated by the RNA polymerase sigma factor RpoS (σ^s) (23, 24). Larger amounts of RpoS per cell correlate with better stress resistance and lower nutritional competence, while smaller amounts of RpoS per cell result in higher nutritional competence but lower levels of resistance. Consequently, modification of the expression of the general stress response regulon enables the balancing of these two traits for adaptation to a particular selective pressure. Similar adaptive variation has been observed in E. coli chemostat cultures (2).

The goal of our study was to investigate the spatial and temporal dynamics of the generation and maintenance of diversity within E. coli colonies. E. coli not only is the best-studied model organism but is also a pathogen that can cause a broad spectrum of intestinal and extraintestinal diseases, including diarrhea, urinary tract infections, sepsis, and neonatal meningitis (25, 26). In addition, E. coli is one of the most common etiologic agents of nosocomial infections (27). E. coli infections cause considerable morbidity and significant health care costs (28). E. coli can survive in secondary environments such as water and soil for days to months depending on temperature and environmental conditions (29, 30). The capacity of E. coli populations to survive in secondary environments is a liability for humans because these bacteria represent powerful sources of contamination that can cause disease or death (31). This is particularly true for water and food contamination, as in the case of the outbreak of hemolytic uremic syndrome in Germany in 2011; this outbreak was caused by the enterohemorrhagic E. coli strain O104:H4, which survived on fenugreek sprouts and seeds (32). In primary and secondary environments, E. coli lives in structured populations (6, 26, 33, 34). For example, uropathogenic E. coli strains establish persistent microcolonies in the epithelial cell layer of the mammalian bladder, and these colonies serve as a source for recurrent acute infections (35–37). Therefore, it is of great interest to investigate the dynamics of E. coli populations in structured environments such as colonies.

We analyzed global temporal changes in colony size, number of cells, cellular physiological state, and gene expression in E. coli colonies growing on LB agar plates. We also monitored the spatiotemporal growth dynamics, viability, metabolic activity, and gene expression patterns of individual cells within these colonies using confocal laser scanning microscopy (CLSM), fluorescent dyes, and reporters of gene expression within individual cells. The roles of the alternative sigma factor RpoS, mutation rates, and indole in the generation of phenotypic diversity were investigated. Finally, the emergence and persistence of phenotypic variants, as well as their fitness, were studied. Overall, the results of our study show that spatially structured environments such as bacterial colonies provide ecological opportunities for the generation and maintenance of vast genetic and phenotypic diversity.

MATERIALS AND METHODS

Bacterial strains.

All strains used in this study are derived from E. coli strain MG1655 (38). For detailed descriptions of the individual strains, see Table S6 in the supplemental material.

Construction of strains carrying fluorescent reporter fusions.

The DNA sequence encoding the yellow fluorescent protein YFP⁺⁺ (39), without a promoter sequence and followed by a chloramphenicol resistance (Cm^r) cassette flanked by the FLP recombination target (FRT) sites for excision with the FLP recombinase (39), was inserted into the E. coli MG1655 genome by homologous recombination (40, 41) just downstream of the stop codon of the ygdI and yiaG genes, to obtain, respectively, strains pygYFP and pyiYFP. To construct strain pyiRFP, the DNA sequence encoding the red fluorescent protein (RFP) without a promoter sequence (42), followed by the FRT-flanked Cm^r cassette, was inserted just downstream of the stop codon of the yiaG gene. The Cm^r cassette was removed by using the temperature-sensitive pCP20 plasmid carrying the flp recombinase gene.

Strains MGFtsZ-GFP (where GFP is green fluorescent protein) and pyiRFPFtsZ-GFP were obtained by P1 transduction of the ftsZ-gfp fusion from strain AND101 to strains MG1655 and pyiRFPF, respectively. Strains pygYFPbssR, pygYFPtnaA, pygYFPmutS, pygYFPmutT, and pygYFPrpoS were constructed by P1 transduction of the following alleles to the pygYFP strain: ΔbssR and ΔtnaA alleles from the Keio collection (43), the ΔmutS allele from strain IB11, the ΔmutT allele from strain MGN16, and the rpoS359::Tn10 allele from strain NEC291. The dnaE486 zae502::Tn10 allele was transferred by P1 transduction from strain NEC175 to strains pyiRFP, NEC252b, pygYFP, and NEC518 to obtain strains pyiRFPdnaEts, pyiYFPdnaEts, pygYFPdnaEts, and precA-GFPdnaEts, respectively.

Media and growth conditions.

The media utilized were Luria-Bertani (LB) broth, LB agar (Difco, Lawrence, KS, USA), COS medium (bioMérieux France), and M9 (44). Bacteria were grown in agitated liquid LB medium at 37°C. When needed, the medium was supplemented with antibiotics at the following concentrations: spectinomycin (25 μg/ml), streptomycin (50 μg/ml), chloramphenicol (30 μg/ml), rifampin (100 μg/ml), and ampicillin (100 μg/ml).

For the colony analysis, 200 cells from overnight LB liquid cultures were deposited on polycarbonate filters (0.22-μm pores, 13-mm or 25-mm diameter; GE Water and Process Technologies, Trevose, PA, USA), which were laid on fresh LB agar plates (15 g/liter, 20 ml per petri dish of 85-mm diameter) and incubated at 37°C for up to 3 weeks. Evaporation was limited by wrapping the plates with aluminum foil. The filters with growing colonies can be easily handled with tweezers. For counting of CFU or flow cytometry analysis, the filters with growing colonies were immersed in 1 ml of LB or phosphate-buffered saline (PBS) and vortexed to recover all the cells from the colony. To establish the number of CFU per ml of liquid culture or per colony, aliquots of the diluted cells were spread on LB agar plates and incubated overnight at 37°C, and the CFU were counted.

Transcriptome experiments.

We performed transcriptome analysis at three time points: 4 h, 1 day (D1), and 7 days (D7) after inoculation for the wild-type strain and D1 and D7 after inoculation for the rpoS strain. The 4-h time point was chosen because 4-h-old colonies consisted primarily of exponentially growing cells. D1 and D7 were chosen because they are before and after, respectively, the breakpoint (D3/D4) observed in the growth kinetic of the colonies and before and after, respectively, the appearance of the first islands of dividing cells within the colonies.

Isolation of total RNA.

Colonies were scraped from the filters and immediately lysed with 1 ml of RNAplus2 RNA extraction solution (Quantum Appligene). All subsequent purification steps were carried out according to the RNAplus2 manual. To purify RNA from contaminating genomic DNA, samples were treated with RQ1 N-free DNase (Promega) followed by phenol-chloroform extraction and precipitation with ethanol. The recovered RNA was resuspended in RNase-free H₂O (Gibco BRL) and quantified by absorbance at 260 nm. The RNA's integrity was assessed by running aliquots on a 1% denaturing agarose gel.

Labeled cDNA synthesis.

Two micrograms of RNA was used for each reaction mixture. E. coli cDNA-labeling primers (Sigma-Genosys, France) were used to prime RNA for subsequent cDNA synthesis. For this annealing step, RNA mixed with 4 μl of primers was heated to 90°C for 2 min and the mixture was then ramped slowly to 42°C. Probe synthesis was carried out at 42°C for 2 h in a 40-μl reaction mixture containing 1× reverse transcriptase buffer, 10 mM dithiothreitol (DTT), 0.5 mM dATP, 0.5 mM dTTP, 0.5 mM dGTP, 0.05 mM ddTTP, 70 μCi [α-³³P]dCTP (1 to 3,000 Ci/mmol; Amersham, France), 40 U of RNase-OUT (Life Technologies), and 50 U of Superscript II reverse transcriptase (Life Technologies). The RNA template was then degraded by the addition of 2 μl of RNase H (10 U μl⁻¹; Life Technologies) followed by incubation for 30 min at 37°C. The labeled DNA was purified using Microspin S-25 columns (Amersham Pharmacia Biotech, France).

Macroarray hybridization.

Panorama E. coli gene arrays produced by Sigma-Genosys were used. Each DNA array consists of a 12- by 24-cm positively charged nylon membrane on which 10 ng each of a complete set of 4,290 PCR-amplified open reading frame (ORF)-specific DNA fragments of E. coli have been printed in duplicate. Prehybridization (1 h) and hybridization (4 h) were carried out by incubating the blot at 68°C in 20 ml of ExpressHyb hybridization solution (Clontech). The entire cDNA probe was added to the hybridization buffer. The blots were then washed with buffer (0.1 SSC [1× SSC is 0.15 M NaCl plus 0.015 M sodium citrate], 0.2% SDS) two times for 10 min at room temperature and four times for 30 min at 62°C. The blots were exposed to a phosphor screen for 3 to 5 days and then stripped in 500 ml of boiling 10 mM Tris (pH 7.5), 1 mM EDTA, and 1% SDS in a microwave oven using the defrost setting for 30 min.

Data acquisition.

The phosphor screens were scanned on a Model 840 Storm PhosphorImager (Molecular Dynamics, Sunnyvale, CA, USA) with a pixel size of 50 μm. The software XdotsReader commercialized by Cose (France) was used to grid the resulting tiff image and to calculate the pixel density for each spot. A background value was calculated from the expression of dots without DNA and subtracted from the intensity of each dot. Dot intensity was normalized to the mean value of the total intensities of all spots on each DNA array. The spot intensities were exported into a Microsoft Excel spreadsheet. Each hybridization experiment was repeated 4 times with independent RNA preparations. Genes with intensity values of expression of <0.1 were considered to be not expressed and were assigned a value of 0.1. The eight values for each gene (4 experiments, 2 duplicates values per experiment) were used to test the null hypothesis of equal mean expression between the two strains and two different times by calculating the P value with a t test using the Benjamini-Hochberg correction (pvBH). The null hypothesis of equal mean expression was rejected when pvBH was equal to or below 0.05. To analyze the fold change in transcript abundance between colonies of various ages (4 h, D1, and D7) or between rpoS and wild-type strains, the ratios of the mRNA intensity values were calculated for each gene.

Live/dead cell staining.

Dead cells were detected by staining with Alexa Fluor 633 hydrazide (AFH) (excitation wavelength, 630 nm; wavelength, 645 nm) or propidium iodide (PI) (excitation, 536 nm; emission, 617 nm) using flow cytometry or microscopy. Both dyes were purchased from Invitrogen (Carlsbad, CA, USA).

Assessment of metabolically active cells.

The Backlight bacterial membrane potential kit containing 3,3′-diethyloxacarbocyanine iodide [DiOC₂(3)] (excitation, 488 nm; emission, 480 nm and 615 nm) and the carbonylcyanide 3-chlorophenylhydrazone (CCCP) H⁺ ionophore were used to estimate the membrane potential of the bacterial cells. Membrane potential reflects both the state of energy metabolism and the physical integrity of the cytoplasmic membrane (45). The experiments were performed as previously described (46).

ROS detection.

Reactive oxygen species (ROS) were detected using dihydrorhodamine 123 (Invitrogen) (excitation, 488 nm; emission, 525 nm), which interacts with H₂O₂. Cells obtained from colonies were resuspended in PBS and diluted to 10⁵ to 10⁶ cells per ml. Dihydrorhodamine 123 was added to the cell suspension to a final concentration of 10 μmol/liter. This mixture was incubated for 15 min at 37°C and then analyzed by flow cytometry.

Measurement of colony areas.

Colonies were prepared as described above and photographed daily. The surface of the agar covered by the colony was calculated by analyzing photographs using ImageJ and used as a proxy for colony size.

Flow cytometry experiments.

Cell samples were analyzed and sorted using a fluorescence-activated cell sorter (FACS)-Aria instrument (Becton, Dickinson Biosciences, San Jose, CA, USA) equipped with a solid-state laser providing 25 mW at 488 nm and a HeNe air-cooled laser providing 20 mW at 633 nm. Fluorescence emissions were detected using a 530/30 bandpass filter for the green fluorescence of YFP and a 660/20 bandpass filter for Alexa Fluor 633 hydrazide. The sheath fluid was BD FACSFlow (Becton, Dickinson Biosciences). All bacterial analysis and sorting were performed at flow rate 5, and acquisitions were done over a period of 1 min. Samples were run at the rate of 2,000 events/s. Acquisition and analysis were performed with BD FACSDiva software (Becton, Dickinson Biosciences). For some experiments, we used an Accuri C6 FACS (Becton, Dickinson Biosciences) to analyze and count the bacterial cells. Blue laser light (488 nm) was used to analyze YFP and propidium iodide fluorescence (filters, 530 nm ± 30 nm and >670 nm, respectively). The 640-nm diode was used to analyze Alexa Fluor 633 hydrazide (filter, 675 nm ± 25 nm).

Microscopy.

Images were acquired using an inverted Zeiss LSM 700 confocal laser scanning microscope (Carl Zeiss, Jena, Germany). YFP (excitation, 514 nm; emission, 527 nm) and RFP (excitation, 558 nm; emission, 583 nm) images were obtained with the laser at 488 nm and 561 nm, respectively. Alexa Fluor 633 hydrazide images were taken with 633 nm laser (excitation, 630 nm; emission, 645 nm). Propidium iodide images were taken with 488-nm laser light (longpass filter 650). The following objectives were used: 10×/0.5 Fluar, 20×/0.75 Plan Apo, 40×/1.3 differential interference (DIC) Plan Apo, 63×/1.4 DIC Plan Apo.

To observe bacterial colonies by confocal laser scanning microscopy, filters with growing colonies were placed on a microscope slide. To maintain colony shape, an adhesive isolator spacer (1 mm deep; Invitrogen) was placed alongside the filter on the slide. A coverslip was placed on top of the colony. Alternatively, the colonies were first moved on agar containing dye(s) (Alexa Fluor 633 hydrazide, propidium iodide, or dihydrorhodamine 123) for staining over a period of 10 to 20 min at 37°C. Afterward, the filters containing the colonies were moved to the microscope slide. Image analyses were performed with the help of Imaris v6.1.5 software (Bitplane AG, Zurich, Switzerland) and Image J software.

Analysis of ftsZ-gfp expression.

Colonies of bacteria carrying the ftsZ-gfp fusion were grown without IPTG ((isopropyl-β-d-thiogalactopyranoside). For induction, filters containing growing colonies were removed from the agar plate. A drop of 50 mM IPTG solution was placed on the agar at the exact location from which the filter had been removed, and the filters were then replaced in their original positions. After 2 h of incubation at 37°C to allow the induction of ftsZ-gfp fusion by IPTG (47), the colonies were analyzed by microscopy.

Analysis of dnaE(Ts) colonies.

Colonies of the temperature-sensitive dnaE(Ts) strain were grown at 30°C. When required, dnaE(Ts) strain colonies were shifted to 37°C and incubated for 3 h, which is the time required for maximal cell elongation. The colonies were analyzed by microscopy. The dnaE(Ts) strain also carries a precA-gfp fusion that allows visualization of the induction of the SOS system (48). After the temperature shift, all filamentous cells expressed the precA-gfp fusion, showing that SOS was induced in all of them. Because the dnaE(Ts) colonies grew more slowly at 30°C than at 37°C, colonies were analyzed on day 2 (D2) instead of on D1.

Time-lapse experiments for observation of single colonies over time.

For preparation of a growth support, a piece of flat, solidified LB agar (15 g/liter) was transferred to a 50-mm culture dish with a hole (diameter, 14 mm) in the bottom (Matek Corporation, Ashland, MA, USA). Hot liquid LB agar was added to seal the piece of agar in the dish. When the agar had solidified, the dish was turned upside down, and a polycarbonate filter (diameter, 13 mm) was placed in the 1-mm-deep well formed by the hole. Cells from an overnight culture were deposited on the filter, as indicated above. The culture dishes were incubated in a box at 37°C. The same colonies were analyzed on several successive days using a spinning-disk confocal laser-scanning microscope (49) (Leica DMI6000 equipped with a CSU10 confocal Yokogawa scan head; Roper Scientific, Germany) and an electron microscopy charge-coupled-device (EMCCD) Quantem 512SC camera (Photometrics, Tucson, AZ, USA) equipped with a plan-apochromat 10× objective with 0.3 numerical aperture (NA). This microscope induces less photobleaching and cell toxicity than classical confocal laser scanning microscopes (49). For this reason, it is convenient for time-lapse experiments. The confocal images were analyzed using Metamorph 7 (Universal Imaging, NY, USA) and Image J software.

Indole assay.

Indole production was measured using Kovac's reagent (Sigma-Aldrich). The filters containing colonies were placed in tubes containing 1 ml of PBS, and the colonies were broken up by pipetting. The cells were removed by centrifugation, and 0.4 ml of Kovac's reagent was added to the supernatant. The reaction mixture was diluted 5-fold in an HCl-amyl alcohol solution (30 ml of HCl and 90 ml of amyl alcohol) (50), and the absorbance of the diluted reaction mixture at 540 nm was measured using a spectrophotometer (Uviline 9100, Secomam, France).

Isolates from 3-week-old colonies.

A colony of strain pygYFP was incubated for 3 weeks at 37°C. The colony was then resuspended in 1 ml of 10 mM MgSO₄ and diluted 10⁻⁴ in 10 mM MgSO₄, and 100 μl of the diluted cells was plated on LB plates, which were incubated at 37°C. After overnight incubation, single colonies (isolates) were picked and inoculated into liquid LB medium. Overnight cultures of each isolate were frozen in LB–20% glycerol for further experiments.

Competition within colonies.

The fitness of individual bacterial isolates relative to the parental strain was estimated by measuring competition in mixed colonies. Parental strain pyiRFP was used for competitions with isolates, while parental strain pygYFP was used as a control in competitions against parental strain pyiRFP. First, overnight cultures of the various isolates and the two parental strains were prepared. An aliquot of each culture was diluted, stained with propidium iodide, and analyzed in the C6 Accuri flow cytometer to assess the number of cells per ml and the percentage of dead cells. Second, the overnight cultures were diluted to obtain suspensions containing 10⁵ live cells per ml. Third, the parental pyiRFP cells and cells of a particular isolate expressing YFP were mixed in a 1:1 ratio. As a control, a 1:1 mixture of cells of parental strains pygYFP and pyiRFP was prepared using the same protocol. Fourth, 200 cells from each mixed suspension were spotted onto polycarbonate filters, which were then laid on agar plates as described above. After 7 days of incubation, mosaic images (tile scan images) of the mixed colonies were obtained with a CLSM Zeiss 700 microscope with a 10× objective. Four mixed control colonies of pygYFP and pyiRFP were analyzed. The proportion of each strain within each colony was evaluated by comparing the green and red fluorescent areas (A_pygYFP and A_pyiRFP, respectively) using Image J software and by calculating the ratio A_pygYFP/A_pyiRFP for each control colony and the mean of these ratios (CICpa). For each isolate, 4 mixed colonies in which the isolate/pyiRFP ratio was determined were analyzed. For each isolate, the competitive index in colonies (CIC) was obtained using the following formula: CIC = (A_IsYFP/A_pyiRFP)/CICpa, where A_IsYFP is the isolate area and A_pyiRFP is the parental pyiRFP area.

Competition in liquid culture.

Mixtures of the pyiRFP parental cells and cells of a particular isolate, obtained as described above, were used to inoculate LB medium (approximately 200 cells of each strain in 10 ml). As a control, mixed liquid cultures of the two parental strains pygYFP and pyiRFP were also prepared. The cultures were incubated at 37°C for 7 days. The relative number of cells of each strain in 4 mixed cultures was evaluated by analyzing an aliquot of cells spotted onto a microscope slide. The microscopic images were analyzed using Image J software. For each isolate, a competitive index in liquid (CIL) was obtained using the following formula: CIL = (N_IsYFP/N_PYIRFP)/CILpa, where N_IsYFP is the number of cells of the isolate, N_pyiRFP is the number of cells of the pyiRFP parental strain, and CILpa is the mean of the ratio N_pygYFP/N_pyiRFP for 4 mixed controlled liquid cultures of the 2 parental strains pygYFP and pyiRFP.

Biolog GN2 microplate assays.

Overnight LB cultures of different strains were used to inoculate fresh LB medium. When the cultures reached an optical density at 600 nm (OD₆₀₀) of 0.8, the cells were spun down and the pellet was resuspended in M9 medium lacking a carbon source. The bacterial suspension was diluted to an OD₆₀₀ of 0.003. One hundred fifty microliters of this suspension was pipetted into each well of the Biolog GN2 MicroPlate (Biolog, Hayward, CA, USA) (51). The microplates were incubated in a Victor3 plate reader (PerkinElmer, Waltham, MA, USA) at 37°C and read every 10 min for 24 h. For each isolate and the parental strain, 3 microplates were prepared. The mean values for each well were established for each isolate and compared to the corresponding values for the parental strain.

Swarming assay.

The swarming ability of the strains was evaluated as previously described (52). Briefly, cells from overnight cultures were placed on swarming plates prepared with 0.5% agar (Difco), 1.0% Bacto tryptone (Difco), 0.5% yeast extract (Difco), 0.5% NaCl (Sigma-Aldrich), and 0.5% d-(α)-glucose (Sigma-Aldrich) using a sterile toothpick. Each plate was inoculated with cells of one isolate at one spot and with parental cells at another spot for the comparison. The plates were photographed after incubation for 24 h at 37°C. The sizes of the swarming colonies were determined using Image J software. The size of the swarming colony of each isolate was divided by the size of the swarming colony of the parental strain growing in the same plate.

Resistance to paraquat.

Paraquat (Sigma-Aldrich) was added to aliquots of overnight LB cultures to a final concentration of 40 mM. After 2 h of incubation, treated cells and untreated (control) cells were diluted and plated on fresh LB agar plates. The number of CFU was assessed after 1 day of incubation at 37°C. The fraction of surviving cells was estimated by dividing the number of CFU of the treated cells by the number of CFU of the untreated cells for each strain.

Biofilm assay.

The capacities of the strains to generate biofilms were estimated using the Biofilm Ring test according to the manufacturer's guidelines (BioFilm Control, Saint Beauzire, France) (53). This test monitors the immobilization of inert paramagnetic beads (toner) in the culture medium during the formation of the biofilm. A magnet is used to assemble the nonimmobilized beads into a macroscopic spot. The resulting spot is quantified through specialized image algorithms. The bacterial strains were grown in COS medium. The turbidity of the bacterial suspensions, which was measured with a densitometer, was adjusted by the addition of 0.45% sterile sodium chloride solution to match that of a McFarland standard of 0.5 to 0.64. The suspensions were then diluted to an OD₆₀₀ of 0.004 in sterile brain heart infusion (BHI) medium provided by BioFilm Control, and toner was added to each suspension to a final concentration of 10 μl/ml. This mixture was homogenized by vortexing, and 100 μl was added to each well of microplates provided by BioFilm Control. The biofilm index (BFI) values corresponding to the mobility of beads in the magnetic field were measured after 6 h of incubation of each strain at 37°C.

Antibiotic susceptibility testing.

Each strain was cultured in COS medium. After adjustment of the turbidity of the bacterial suspensions as described above, 145 μl of each suspension was diluted by the addition of 3 ml 0.45% sterile sodium chloride and inoculated into Vitek GN, ASTN103, and AST-EXN8 cards. The cards were loaded and read in the Vitek 2 instrument (bioMérieux, France).

RESULTS

Global analysis of aging colonies.

We first analyzed global temporal changes in colony size, number of cells, cellular metabolic state, and gene expression in aging colonies of E. coli. Colonies were obtained from an inoculum of 200 cells (Fig. 1A). During the first several hours, the bacterial cells divided without obvious organization, as previously described (10). By 4 h, the cells had formed compact colonial masses with outwardly expanding fronts. The colony surface area increased 20-fold from day 0 (D0) to D3 (Fig. 1B), whereas from D3/D4 to D7, colony expansion was only 5-fold.

FIG 1.

Growth and gene expression patterns in aging wild-type colonies. Colonies began growing from approximately 200 cells spotted on the polycarbonate filters that were deposited on the LB agar plates on day 0. (A) Photographs of colonies at D1 (left) and at D7 (right). The diameter of the filters is 25 mm. (B) Increase in colony area in mm² over time. (C) Change in the number of cells (red curve) and in the number of CFU (blue curve) per colony over time. (D) Percentages of metabolically active cells per colony over time as detected by staining with DIOC2. In panels B, C, and D, each point shows the mean value obtained from four independent experiments; the error bars correspond to the standard error of the mean (the error bars in B and C are small and are not visible on the curve). (E) Global changes in the transcriptome profiles between 4 h and D1 and between D1 and D7. Selected metabolic pathways and genes that were upregulated or downregulated are indicated in red and blue, respectively.

We determined the number of CFU by plating and the number of cells by flow cytometry analysis. On the first day (D1), both the number of cells and the CFU per colony increased 10⁷-fold relative to the inoculum. Between D1 and D4, the number of cells increased 7- to 8-fold, while the number of CFU increased only 2-fold. Between D4 and D7, the number of cells increased 1.5- to 2-fold, while the number of CFU decreased 2-fold (Fig. 1C). The discrepancy between the number of cells and the CFU in aging colonies can be explained by the occurrence of cell death; this explanation is corroborated by our previously published observation that the number of dead cells in aging colonies increases with time (46). Staining with 3,3′-diethyloxacarbocyanine iodide (45), which estimates membrane potential, showed that the number of metabolically active bacteria decreased as the colonies aged. At D4 and at D7, 40% and 8% of bacteria, respectively, were metabolically active (Fig. 1D).

To further investigate the physiological state of cells in aging colonies, we examined the transcriptome of the wild-type strain in colonies grown for 4 h and in colonies obtained on D1 and D7. The expression of 14% of bacterial genes was altered during colony aging, i.e., differed on D1 and D7 versus the 4-h time point (Fig. 1E; for details, see Tables S1 and S2 in the supplemental material). Between 4 h and D1, genes involved in the modification of the cellular outer membrane, DNA superhelical density, the glyoxylate pathway, the mobility of the IS and rhs elements, intercellular communication, and biofilm formation were induced. Between D1 and D7, genes involved in transcription and translation, carbon and energy metabolism, and cell motility were downregulated, indicating that cells in the aging colony were reducing their energy consumption as if they were entering dormancy. As the colonies aged, an RpoS-regulated general stress response was induced (see Table S3 in the supplemental material). This is indicated by the observation that 46% of the known RpoS-regulated genes (54–56) were expressed at significantly higher levels in wild-type D1 and D7 colonies than in rpoS D1 and D7 colonies. Genes involved in other stress responses (genes associated with acidic stress, osmotic shock, oxidative stress, carbon source starvation, cold shock, heat shock, Cpx envelope stress, and phage shock responses) were also induced in aging colonies.

Taken together, global analysis of aging colonies showed that growing cells, nongrowing but metabolically active cells, and dead cells coexist inside the aging colonies and that the number and distribution of these types of cells change over time. The number of dividing and metabolically active cells diminishes, while the fraction of bacteria incapable of growth and/or dead bacteria increases with time. After D3/D4, colony expansion slows. Determination of global transcriptome profiles indicated that most cells enter dormancy as the colonies age. The induction of many stress responses indicates that aging colonies are very stressful environments for E. coli cells.

Growth of individual cells within aging colonies.

To study the spatiotemporal dynamics of bacterial growth within aging colonies, we decided to analyze the expression of two fluorescent gene reporters using confocal laser scanning microscopy, which allows the study of individual bacterial cells without destruction of the colonies. The first reporter strain that we used carries an ftsZ-gfp fusion under the control of the lac promoter (47). Cytoplasmic fluorescence indicates that the ftsZ-gfp cells are metabolically active (57, 58), while the presence of FtsZ-GFP fluorescent rings indicates that ftsZ-gfp cells are dividing (59). It should be noted that all cells having fluorescent cytoplasm are not dividing.

Most of the cells within 4-h-old ftsZ-gfp colonies contained fluorescent rings and had highly fluorescent cytoplasm (data not shown). In D1 colonies, cells with fluorescent rings were mostly localized in the 50-μm-wide colony rim (Fig. 2A), while they were rare elsewhere in the colony. In D4 colonies, only a few cells with fluorescent rings were present at the colony rim (Fig. 2B), and even fewer such cells were present at the rims of D7 colonies (Fig. 2C). However, at D4, clusters of cells with fluorescent rings and highly fluorescent cytoplasm compared to the surrounding cells began to appear inside colonies (Fig. 2D, E, and F). Clusters of cells with different phenotypes relative to the cells surrounding these clusters are referred to as “islands” in the text.

FIG 2.

Localization of dividing cells within aging colonies. (A to F) Confocal laser scanning microscopic (CLSM) images of colonies of a strain carrying an ftsZ-gfp fusion under the control of the lac promoter. The ftsZ-gfp fusion permits the detection of cells that are undergoing cell division, i.e., cells with the fluorescent FtsZ-GFP septation rings, and metabolically active cells that respond to IPTG, transcribe reporter fusion, and carry out protein synthesis, i.e., cells with fluorescent cytoplasm. (A) Edge of a D1 colony containing a rim of dividing cells within which numerous FtsZ-GFP rings are visible. (B and C) Edge of a D4 and a D7 colony, respectively, within which few FtsZ-GFP rings are visible. (D to F) Islands of cells expressing high levels of the ftsZ-gfp reporter fusion in D4 and D7 colonies. (F) Island with only a few dividing cells. The FtsZ-GFP rings are indicated with an arrow in panels A to F. (G to L) CLSM images of colonies of a strain carrying the dnaE(Ts) allele and a precA-gfp fusion; dnaE(Ts) encodes a thermosensible variant of the DNA polymerase III α subunit. In this strain, at a temperature of ≥37°C, DNA replication is blocked, which induces the SOS response (induction of precA-gfp reporter fusion) and causes filamentation of only those cells that are replicating DNA. (G) Filamentation of cells in a 4-h colony 3 h after temperature shift from 30°C to 37°C. Nearly all cells were filamenting. All filamenting cells expressed high levels of precA-gfp (in green). (H) View of the edge of a D2 colony at 30°C. (I) The same colony after temperature shift. Cells were filamenting at the edge of the D2 colony. (J) Edge of D7 colonies where few cells are filamenting. (K and L) Islands in D7 and D10 colonies. Filamentation is primarily observed within islands inside the D7 and D10 colonies after temperature shift.

The second reporter strain used in this study carries a dnaE(Ts) allele that encodes a thermosensitive variant of the DNA polymerase III α subunit (60). At nonpermissive temperatures (≥37°C), DNA replication is blocked, which causes filamentation of the cells. We used this phenotype to detect cells that are replicating their DNA. In 4-h-old dnaE(Ts) colonies, all cells were filamentous after the shift to the nonpermissive temperature, indicating that all cells were replicating DNA (Fig. 2G). In D2 colonies after the temperature shift, tight networks of filamentous cells formed a 40-to-50-μm-thick rim, while isolated filamentous cells were scattered inside the colonies (Fig. 2I). The localization of the filamentous cells in D2 dnaE(Ts) colonies after the temperature shift corresponded to the localization of cells with FtsZ-GFP rings in D1 colonies. In D7 colonies after temperature shift, only rare isolated filamentous cells could be found scattered at the colony rim (Fig. 2J), while islands of tightly packed cells composed entirely or partially of filamentous cells were embedded within the colonies (Fig. 2K and L).

Taken together, the two reporter fusions indicated that all cells in the 4-h-old colonies were replicating DNA and were actively dividing. In the D1 colonies, the dividing cells responsible for colony expansion were mostly localized in a thin layer at the rim of the colony. From D4 onward, the vast majority of the cells stop dividing. In these aging colonies, dividing cells were localized mainly within islands, while fewer and fewer dividing cells were observed within the colony rim.

Spatiotemporal phenotypic heterogeneity within aging colonies.

Transcriptome analysis showed that the RpoS regulon is induced in the aging colonies (Fig. 1E; see also Tables S1 and S3 in the supplemental material). We hypothesized that the RpoS regulon was induced mainly in nondividing cells. To test this hypothesis and to study how patterns of RpoS-dependent gene expression change in aging colonies, we used yiaG and ygdI as reporter genes. These genes were chosen because they were among the most strongly induced genes in the aging colonies and because they showed lower expression in rpoS colonies than in wild-type colonies (see Fig. S1A and S1B and Tables S1 and S3 in the supplemental material). The yiaG gene encodes an RpoS-dependent putative helix-turn-helix transcription factor (54, 55). The ygdI gene encodes a putative lipoprotein and contains a peptide signal sequence for potential localization in the outer membrane (38). We fused promoters of these two genes to the genes coding for the YFP and RFP fluorescent proteins and characterized their induction by various treatments using flow cytometry. Expression of the yiaG and ygdI genes was induced 5-fold and 4-fold, respectively, in stationary-phase liquid cultures compared to exponential-phase liquid cultures (see Fig. S1C in the supplemental material). Whereas ygdI was strongly induced by osmotic shock (>8-fold), yiaG was induced only 2-fold, suggesting that the two genes may be differently regulated under different stress conditions (see Fig. S1D in the supplemental material).

Using confocal laser scanning microscopy, we analyzed aging colonies of strains containing both the pygdI-yfp and pyiaG-rfp reporter fusions or only one of them. In D1 colonies, both reporter fusions were expressed homogeneously in the upper colony layer (see Fig. S1E and F in the supplemental material), whereas the fusions were not expressed in most of the dividing cells at the colony rim (see Fig. S1G and S2 in the supplemental material). In D4 colonies, most of the cells in the upper colony layer still expressed both genes, but islands of cells with particular expression patterns began to appear. None of these islands expressed pyiaG-rfp, whereas the expression of pygdI-yfp in the islands was variable, ranging from no expression to very strong expression (Fig. 3A and B; see also Fig. S4C in the supplemental material). In the dnaE(Ts) mutant background, we observed islands of filamentous cells in D7 colonies upon temperature shift; some of the cells in these islands expressed pygdI-yfp, but none expressed pyiaG-yfp (see Fig. S3A to D in the supplemental material).

FIG 3.

Confocal scanning laser microscopic analysis of aging colonies. (A, B, and C) Emergence of phenotypic diversity in aging colonies: islands with various pyiaG-rfp and pygdI-yfp expression levels are visible at D4 inside a colony of a strain expressing the 2 reporter fusions. The colony was stained with Alexa Fluor 633 hydrazide (AFH) to visualize dead cells. (A) pyiaG-rfp expression (red); (B) pygdI-yfp expression (green); (C) merged images of panels A and B and AFH staining (in blue). (D, E, and F) Views of islands within a D7 colony of the strain expressing the pygdI-yfp reporter. (D) pygdI-yfp expression (green); (E) staining with AFH (blue); (F) merged images of panels E and F. (G) Confocal mosaic image view of a D10 colony expressing the pygdI-yfp reporter fusion (in green). The colony was stained with propidium iodide (PI) (red) and AFH (blue) to visualize dead cells. PI binds to DNA; AFH binds to carbonylated proteins. Many clusters of cells (islands) with different phenotypes from those of the surrounding cells are visible. (H) Drawing of a colony indicating the plane within which the confocal pictures were taken (red line).

Study of the same individual pygdI-yfp colonies over an 8-day period allowed us to observe that the number of islands and their size increased considerably with time (6-fold increase in number and 2- to 20-fold increase in size between D4 and D7) (Fig. 3E; see also Fig. S4A and B, S5, and S1H and I in the supplemental material). Staining of aging colonies with Alexa Fluor 633 hydrazide and/or propidium iodide showed that some of the islands contained dead cells (Fig. 3C, F, and G). In some islands, dead cells were surrounded by living cells strongly expressing ygdI (Fig. 3G; see also Fig. S6 in the supplemental material). The observed pattern suggests that many cells that start growing within these islands die, which may provide nutrients for other cells to grow. The islands in D10 colonies showed very great diversity with respect to the expression of pygdI-yfp and cell death patterns (Fig. 3G).

Because yiaG is regulated by RpoS (54) and because it is induced in stationary-phase cultures (see Fig. S1C in the supplemental material) as well as in nongrowing cells in aging colonies (see Fig. S2 and S3 in the supplemental material), yiaG can be considered a reporter for dormant cells. The expression of this gene in most cells in aging colonies suggested that most of the cells in the aging colonies were in dormancy. Because the ygdI gene is strongly expressed in cells growing within islands in aging colonies, it can be used as a reporter for the islands. For this reason, we introduced the pygdI-yfp reporter fusion into the rpoS mutant strain. We observed that in the rpoS mutant strain 10-fold-fewer islands strongly expressed pygdI-yfp than in the wild-type strain. The fact that this reporter fusion was expressed in the rpoS strain confirms that ygdI is not exclusively regulated by the RpoS sigma factor, as is the case for many RpoS-regulated genes (54). We also tested an rssB mutant strain carrying the pygdI-yfp reporter fusion. This mutant strain has a constitutively induced RpoS regulon because RssB protein delivers RpoS to the ClpXP protease for degradation (61). We observed a large amount of phenotypic diversification at D2 in rssB colonies (data not shown). At D4, rssB colonies contained large areas of cells that did or did not express pygdI-yfp and many islands (Fig. 4C). These results strongly suggest that induction of the RpoS regulon is required for island genesis.

FIG 4.

Confocal microscopic images of D4 colonies. The parental strain (A) and its rpoS (B) rssB (C), mutS (D), mutT (E), and bssR (F) derivatives. All the strains carry a pygdI-yfp reporter fusion (green).

Role of mutations in island genesis.

Our study demonstrated that there is important spatiotemporal heterogeneity of cellular phenotypes within aging colonies, most of which is localized within islands. We hypothesized that island founders could be genetic variants. Genetic variants could already be present in the population of cells from which the colony grew. However, the fact that colonies growing from one or two single cells generated the same spatiotemporal heterogeneity of cellular phenotypes as those previously observed for colonies growing from a population of approximately 200 cells shows that this is not the case (see Fig. S7A and B in the supplemental material). Hence, the spatiotemporal heterogeneity of cellular phenotypes was generated during the growth and aging of the colony.

If mutations are at the origin of the observed phenotypic heterogeneity and in particular are the origin of island generation, then colonies of mutator strains should exhibit higher phenotypic heterogeneity and produce more islands. To test this hypothesis, we used the mutS and mutT mutant strains, which have constitutively high mutation rates. The mutation rates of these mutator strains are 100- to 1,000-fold higher than the mutation rate of the nonmutator wild-type strain. The mutS gene codes for a mismatch recognition and binding protein that plays a key role in maintaining the fidelity of DNA replication and the fidelity of recombination between nonidentical DNA sequences (62). The MutT protein hydrolyzes mutagenic 8-oxo-7,8-dihydroguanine in the nucleotide pool, thus preventing its incorporation into DNA during replication (48). In these mutator strains, islands appeared earlier and were 10-fold higher in number at D4 than in colonies of the wild-type strain (Fig. 4D and E and 5A). These results show that mutagenic events can indeed generate island founders in aging colonies. However, we cannot exclude the possibility that some or most of the island founders in the wild-type strain colonies are different phenotypic variants of isogenic cells.

FIG 5.

Island formation, death, and ROS production in aging colonies. (A) Number of islands per mm in D4 and D7 colonies of the parental (WT), bssR, tnaA, mutS, mutT, and rpoS strains. (B) Percentage of dead cells detected by Alexa Fluor hydrazide staining in D4 and D7 colonies from the parental strain (WT) and from tnaA strains growing on LB agar plates supplemented or not supplemented with 1.5 mM tryptophan (trp). (C) ROS levels in D4 and D7 colonies of the parental (WT) and tnaA strains detected by staining with DHR123 and analyzed using flow cytometry. For each chart, the bars correspond to the mean values of at least four independent experiments; the error bars indicate the standard deviations.

Role of indole in island genesis.

Transcriptome analysis showed that bssR gene expression was 19-fold and 5-fold higher in D1 and D7 colonies, respectively, than in 4-h colonies (see Table S1 in the supplemental material). In addition, D4 and D7 bssR mutant colonies had 8-fold and 2-fold-fewer islands, respectively, than D4 and D7 wild-type colonies (Fig. 4F and 5A). BssR regulates the expression of genes involved in catabolite repression, stress responses and the transport of various molecules such as indole (63). Because BssR increases the levels of indole inside bacterial cells (63), it may be that the negative impact of bssR gene inactivation on island genesis is a consequence of the lower intracellular concentration of indole in cells in which this gene is inactivated. Because indole can damage membranes and induce the production of superoxides (64, 65), we hypothesized that indole accumulation stimulates the genesis of islands by increasing cell death. Dead cells could serve as a source of nutrients for island founders.

We first verified that indole is produced in aging colonies; second, we showed that indole is responsible for ROS production, cell death, and island generation. Because indole is formed from tryptophan by TnaA tryptophanase (66), we used a tnaA mutant as a negative control. As expected, we detected indole in wild-type but not in tnaA mutant colonies at D7 (see Fig. S8A in the supplemental material). Measurements of the oxidative stress in D4 and D7 wild-type colonies using dihydrorhodamine 123 and flow cytometry showed that ROS increased significantly 1.4-fold in D7 colonies compared to D4 colonies (Fig. 5C). Wild-type colonies had approximately 2-fold-higher ROS concentration than tnaA colonies at D4 (Fig. 5C). Using Alexa Fluor 633 hydrazide and flow cytometry to quantify the number of dead cells, we found that wild-type colonies contained 50% more dead cells than the D4 and D7 tnaA colonies (Fig. 5B). We also grew colonies on LB plates supplemented with 1.5 mM tryptophan, which results in the generation of high levels of indole (67). D4 and D7 colonies grown on LB complemented with tryptophan had 30% and 10% more dead cells, respectively, than colonies grown on LB alone. D7 wild-type colonies grown with tryptophan exhibited a 5-fold-greater number of islands than colonies grown without tryptophan (see Fig. S8B in the supplemental material), whereas tnaA colonies grown with tryptophan had 8- and 4-fold-fewer islands than the wild-type colonies at D4 and D7, respectively. Taken together, our data show that indole production (i) results in death in aging colonies, probably via the generation of ROS, and (ii) plays a positive role in the formation of islands.

Isolation and characterization of phenotypic variants from a 3-week-old colony.

Cells growing in islands within aging colonies could be mutants that have been selected for fitness. If so, as colonies age, they should become enriched in mutants capable of outcompeting the parental strain. To test this hypothesis, we analyzed colonies over a 3-week period, using flow cytometry and confocal microscopy. In the 3-week colonies, only 2.6% (standard deviation [SD], 0.41%) of the cells were alive based on staining with Alexa Fluor 633 hydrazide (see Fig. S9B in the supplemental material). Most of the live cells highly expressed the pygdI-yfp reporter fusion, whereas the dead cells exhibited weak residual fluorescence or no fluorescence. This phenotype allowed us to easily localize live cells in the colonies by confocal microscopy. Highly fluorescent cells were detected only in the islands. No live fluorescent cells were detected elsewhere in the colonies, i.e., at the edges of or inside the colonies (see Fig. S9 in the supplemental material).

To isolate live cells, we resuspended a single 3-week colony and plated the resuspended cells on LB plates. From the resulting colonies, 25 isolates were randomly chosen. First, we characterized these isolates using the following phenotypic tests on Biolog GN2 microplates: growth in LB, swarming capacity, resistance to paraquat (a superoxide-generating compound), capacity to form biofilm, tolerance to 33 antibiotics, and ability to metabolize 95 carbon sources. Second, we measured the ability of each isolate to compete with the parental strain in reconstructed aging colonies as well as in liquid LB.

In LB liquid medium, 16 isolates showed growth comparable to that of the parental strain, and 9 isolates grew significantly more slowly (Table 1). The differences were noticeable for 4 of them (the ratio of the growth rate of the isolate to the growth rate of the parental strain ranged from 0.2 to 0.54) (Table 1). Seven of the isolates had a higher capacity to swarm than the parental strain (>1.5-fold), while 2 isolates had a reduced capacity to swarm (<0.5-fold) (Table 1). Nine of the isolates showed better resistance to oxidative stress than the parental strain (>1.5-fold), while 3 isolates showed lower resistance (<0.7-fold) (Table 1). Seven isolates produced thicker biofilms than the other isolates and the parental strain (Table 1).

TABLE 1.

Phenotypic features of the isolates from a 3-week-old colony^a

Isolate	Growth rate in LBb		Competitive indexc in:				Swarming capacityd		Biofilm formatione		Resistance to paraquatf		MICg
			Colony		Liquid
	Ratio	SD	Ratio	SD	Ratio	SD	Ratio	SD	Ratio	SD	Ratio	SD	Amp	Cef
y1	0.96	0.01	0.2	0.1	1.1	0.1	1.1	0.2	1.4	0.37	1.3	0.3	2	1
y2	0.98	0.04	3.0	0.1	1.0	0.1	0.9	0.3	0.9	0.03	0.8	0.4	2	1
y3	0.94	0.06	18.5	1.1	1.0	0.1	1.7	1.4	0.5	0.20	1.0	0.2	0.5	1
y4	1.02	0.03	19.4	1.2	1.0	0.1	2.0	0.6	0.6	0.07	2.3	0.7	0.5	1
y5	0.94	0.03	14.4	6.4	1.0	0.1	1.8	0.1	0.8	0.13	2.5	0.5	0.5	1
y6	0.99	0.01	5.1	1.5	1.0	0.1	1.9	0.4	1.0	0.19	0.8	0.6	2	1
y7	0.91	0.06	0.3	0.2	1.4	0.1	1.0	0.1	1.1	0.06	0.6	0.2	0.5	1
y8	0.94	0.04	5.6	1.9	0.9	0.1	1.4	0.1	0.9	0.14	1.6	0.3	1	1
y9	1.00	0.01	1.3	0.3	1.0	0.1	1.5	0.1	1.0	0.01	1.0	0.3	1	1
Y10*	0.96	0.02	3.4	1.2	1.0	0.1	1.1	0.2	0.8	0.05	0.8	0.3	1	1
Y11***	0.93	0.06	0.4	0.1	2.0	0.4	1.1	0.2	0.8	0.05	0.7	0.3	0.5	1
y12	1.02	0.15	2.2	1.2	1.0	0.1	1.5	0.4	0.9	0.19	1.5	0.4	0.5	1
y13	1.02	0.06	3.1	0.8	1.0	0.1	1.0	0.2	0.4	0.15	3.6	0.8	2	4
y14	0.90	0.04	1.1	0.3	1.0	0.1	1.0	0.2	1.0	0.02	0.8	0.2	1	1
y15	0.68	0.08	3.0	0.7	1.0	0.1	1.3	0.1	1.0	0.01	1.0	0.3	1	1
y16	1.02	0.03	3.4	0.3	1.1	0.1	1.9	0.1	1.0	0.07	1.0	0.2	2	1
y17	0.87	0.04	2.7	0.7	1.1	0.1	1.6	0.4	1.0	0.05	0.5	0.3	0.5	1
y18	0.99	0.05	9.8	2.1	1.1	0.1	2.2	0.7	0.8	0.14	0.5	0.2	2	0.5
y19	0.94	0.03	2E-03	3E-04	1.0	0.1	1.0	0.2	1.0	0.13	1.1	0.4	2	1
y20	1.00	0.06	0.7	0.1	1.0	0.1	1.4	0.2	1.0	0.04	1.5	0.3	1	1
y21	0.97	0.05	1.1	0.3	0.9	0.1	1.3	0.1	1.0	0.15	4.8	1.2	1	1
y22	0.26	0.27	0.3	2E-02	0.2	4E-02	0.3	0.1	1.4	0.19	3.4	0.6	0.5	0.25
Y23**	0.28	0.02	0.2	0.1	0.4	0.1	1.8	0.9	1.3	0.14	1.4	0.3	0.5	0.25
y24	0.23	0.15	0.3	3E-02	0.2	4E-02	1.2	0.3	1.3	0.11	2.5	0.8	0.5	0.25
Y25**	0.54	0.03	0.2	2E-02	0.7	0.1	0.3	0.2	1.4	0.13	1.7	0.3	0.5	0.25

^aAll data are ratios of the results for the test strain to those of the parental strain. Each value is the mean value from at least 4 independent experiments. Isolates in italics have better fitness than the parental strains in colonies. Underlining indicates that the corresponding P values are <0.05 (by the t test for these data between isolates and the parental strain). Asterisks: *, the isolate is a mutator; **, the isolate is resistant to rifampin; ***, the isolate is an rpoS mutant.

^bThe maximum growth rate [μ = ln OD₂ − ln OD₁/(t2 − t1)] in LB was calculated for each isolate and compared to that of the parental strain.

^cThe relative fitness in colonies and in liquid was established for each isolate compared to the parental strain (for examples, see Fig. 6). The competitive index in colonies (CIC) was calculated by analyzing fluorescent microscopic images of the mixed colonies (for examples, see Fig. 6). The competitive index in liquid (CIL) was calculated by counting fluorescent cells from mixed liquid cultures after microscopic imaging.

^dThe size of the swarming colony of each isolate was divided by the size of the swarming colony of the parental strain growing in the same plate.

^eThe capacity of different isolates to form biofilms relative to the ability of the parental strain was evaluated using BioFilm ring test technology (Biofilm Control Company).

^fCells were challenged with 40 mM paraquat. The values presented correspond to the ratio of the survival frequency of the isolate (CFU of treated cells/CFU of untreated cells) and the survival frequency of the parental strain (CFU of treated cells/CFU of untreated cells).

^gFor complete antibiogram data, see Table S6 in the supplemental material. Amp, ampicillin; Cef, cefoxitin.

The MICs of individual isolates to the antibiotics studied ranged from 0.5- to 4-fold that of the parental strain (Table 1; see also Table S4 in the supplemental material). For example, 7 isolates had a higher MIC for ampicillin than the parental strain, whereas 11 had a lower MIC. Isolate Y13 was resistant to cephalosporins (cephalothin MIC, 32 mg/liter; cefoxitin MIC, 8 mg/liter). Finally, after plating isolates on agar plates containing rifampin, we observed that isolates Y23 and Y25 were resistant to rifampin. In these two isolates, rifampin resistance resulted from different mutations in the gene rpoB, which we detected by sequencing (data not shown).

Eleven isolates had the capacity to metabolize more carbon sources or to metabolize some carbon sources more efficiently than the parental strain, whereas 10 isolates had lower capacities to metabolize carbon sources than the parental strain (see Table S5 in the supplemental material). Nine isolates exhibited 3- to 8-fold-better capacity to grow on the short-chain fatty acids (SCFA) α-ketovaleric acid than the parental strain. Most of these 9 isolates also grew better than the parental strain on each of the other SCFAs (acetic acid, α-ketobutyric acid, and propionic acid) as the sole carbon source.

We evaluated the ability of the 25 isolates to compete with the parental strain in reconstructed aging colonies. These colonies were generated by mixing equal numbers of parental and particular isolate cells. Control competitions between two differentially marked parental strains were also performed; in those experiments, the final ratio between the two strains was 0.96 (SD = 0.12). Thirteen isolates were able to outcompete the parental strain, i.e., their competitive index in colony (CIC) was between 2 and 19 (Table 1; Fig. 6A and B). The parental strain cells formed few sectors in the colony center and did not reach the edges of the colonies. Such a pattern has been shown to arise when there are significant fitness differences among the cells growing in a colony (68). The 13 isolates that outcompeted the parental strain in reconstructed aging colonies showed growth that was equal to or significantly slower than that of the parental strain in LB medium. Nine of the 25 isolates, including the rpoS mutant, performed poorly in competition with the parental strain in mixed colonies, i.e., their CICs ranged from 0.002 to 0.7 (Table 1; Fig. 6C). Finally, we also performed competitions in liquid LB medium between the parental strain and each of the 13 isolates that showed high fitness in aging colonies. All of these isolates exhibited the same fitness as the parental strain in liquid LB medium (Table 1).

FIG 6.

Competition between the parental strain and evolved variants in aging colonies. (A to C). CLSM mosaic images of D7 mixed colonies. Colonies were generated by mixing equal numbers of the parental cells (pyiRFP) expressing the red fluorescent protein and cells of different evolved variants Y6 (A), Y16 (B), and Y7 (C) previously isolated from a 3-week-old colony originated from pygYFP (expressing the yellow fluorescent protein YFP). (D) Control mixed colonies. Colonies were obtained by mixing equal cell numbers from the two parental strains pygYFP and pyiRFP. (A to D) The left column corresponds to the merged YFP and RFP images, the middle column to the YFP images and the right column to the RFP images. Y6 (A) and Y16 (B) were able to outcompete the parental strain, while Y7 (C) was not.

Finally, we searched for a correlation between phenotypic tests and the ability to outcompete the parental strain in colonies. We found a correlation between the capacity to form a thicker biofilm and CIC (R² = 0.35; P = 0.004) (see Fig. S10A and B in the supplemental material), a finding that suggests that the ability to form thick biofilms provides a selective advantage in aging colonies. We also found that, among the 95 carbon sources tested, the capacity to grow in itaconic acid correlated best with CIC (R² = 0.39; P = 0.008) (see Fig. S10A and C in the supplemental material). Taken together, the results of our analysis of 25 isolates from a 3-week colony showed that they were all different, indicating that great diversity occurred during aging of the colonies.

DISCUSSION

In this study, we examined the genetic and metabolic determinants of diversification in aging E. coli colonies. Using fluorescent reporters, fluorescent dyes, and confocal laser scanning microscopy, we were able to follow the temporal and spatial dynamics of diversification without perturbing colony architecture. On the first day of incubation, most of the cells in the colonies were actively dividing. From D1 to D4, the proportion of metabolically inactive cells in the colonies increased. The majority of dividing cells were localized in a thin layer at the edge of the colonies, where they contributed to the expansion of the colony. From D4 onward, the vast majority of the cells stopped dividing, while the number of dead cells increased. In these aging colonies, dividing cells were localized mainly within islands, while fewer and fewer dividing cells were observed within the colony rim.

We hypothesized that islands probably resulted from the generation of mutants during colony growth. The frequency of islands per wild-type colony was 8 × 10⁻⁸ and 2.6 × 10⁻⁷ (calculated as a function of the total number of cells, which was established by flow cytometry) at D4 and D7, respectively. Because the frequency of islands in the D4 mutS colony was 7.4 × 10⁻⁷, the observed frequency of islands in the wild-type colonies is compatible with the hypothesis that the island founders were mutants. We found that only 1 of 25 characterized isolates displayed a constitutive mutator phenotype. However, mutation rates can increase in aging colonies by a variety of molecular mechanisms. One possible mutagenic mechanism is ROS production. Endogenous oxidative stress was previously reported to be responsible for genetic diversification in Pseudomonas aeruginosa biofilms (21). Increased mutagenesis in aging E. coli colonies was also reported to depend on oxidative metabolism (20). In the present study, we observed that ROS accumulate in aging colonies and that genes involved in the oxidative stress response were induced in aging colonies. ROS could be generated by fumarate reductase as a consequence of the shift from anaerobic to aerobic conditions in the aging colonies (12, 69). Another source of ROS is indole, which is produced from tryptophan in aging colonies. We observed that indole production results in increased ROS production and cell death as well as in increased island formation. The role of indole could be dual: indole could promote the generation of mutant island founders, and the cells killed by indole could also provide nutrients for the island founders. Mutations could also be generated by other mechanisms, e.g., by genomic rearrangements and mobility of ISs. This possibility is supported by the fact that genes that induce transposition and genetic rearrangements, such as IS elements and rhs sequences, were overproduced in the aging colonies. However, although we showed that high mutation rates can increase the number of islands in aging colonies, we cannot exclude the possibility that some or most of the island founders in the wild-type strain colonies represent different phenotypic variants of isogenic cells.

Island formation was shown to be a genetically controlled process in that it is regulated by the RpoS sigma factor. Colonies of the rpoS-deficient strain contained many fewer islands, while the rssB strain that accumulates RpoS protein produced more numerous and more diverse islands than the wild-type strain. Among the genes induced as part of the RpoS regulon and that could be involved in island formation, the tnaA gene, which codes for the tryptophanase that generates indole from tryptophan, is the most likely candidate. Another possible candidate is RpoS-dependent mismatch repair downregulation (70, 71), which results in increased mutagenesis in aging colonies (20) and can thereby contribute to the formation of islands.

Why do islands begin to appear only after 4 days of incubation? Several nonexclusive hypotheses can be proposed to explain this observation. First, the mutants that found islands could be generated late in colony development. This idea is supported by the observation that diversification appeared earlier in colonies of mutator strains. Second, some mutants might require the presence of other mutants that scavenge toxins or provide nutrients for growth. Such symbiotic interactions have been described in biofilms (72). Third, mutants may only begin growing or may grow better when local microenvironmental conditions become favorable. For example, dead cells (20% of the total cells in D4 colonies) could provide nutrients to colony founders. The massive diversification that occurs in aging colonies of E. coli was demonstrated by the phenotypic characterization of 25 isolates from a single colony. Based on pygdI-yfp reporter expression, there is a very high probability that all of these isolates were obtained from islands. All the isolates had phenotypes that differed from that of the parental strain. All observed phenotypic differences were heritable, an observation that suggests that each isolate had one or more mutations that were responsible for the observed phenotypes. We found that 52% of isolates were able to outcompete the parental strain in reconstructed mixed colonies, while 44% had lower fitness than the parental strain in this setting. The capacity to outcompete the parental strain in mixed colonies was not accompanied by higher fitness in liquid LB culture competition or by a higher growth rate in LB liquid medium. Therefore, winning isolates did not behave as GASPers, rpoS-attenuated mutants that have been described previously (22). The fact that some isolates that had lower fitness in mixed colonies and in liquid LB, e.g., rpoB mutants, were also isolated suggests either that these isolates had a selective advantage under very local restricted conditions within aging colonies or that they were maintained in the population by genetic drift (68). This finding illustrates how a structured environment may provide low-fitness mutants with the opportunity to persist and to give rise to new mutations, some of which might be beneficial. It is therefore possible that genotypes that would not have emerged in well-mixed unstructured environments arise in aging colonies.

We found that two phenotypes, the capacity to metabolize itaconic acid and the ability to form thicker biofilms, correlated with the ability to outcompete the parental strain in mixed colonies. The capacity to grow in itaconic acid, which is degraded to succinate and acetyl coenzyme A (acetyl-CoA) (73), should be advantageous in aging colonies because acetyl-CoA accumulates during the growth of cells in rich medium, which is the case in our study, and becomes the main carbon source available when cell growth slows. This phenomenon is referred to as the “acetate switch” (74). We indeed observed that the amount of mRNA of the aceA, aceB, asnA, and asnB genes, which are known to be involved in the assimilation of acetates, increased in aging colonies (see Table S1 in the supplemental material). The capacity to form thicker biofilms should provide an advantage to the cells in the colony because by forming an outward expanding front with a thick exopolysaccharide matrix, they could mechanically exclude cells of the parental strain.

Some acquired phenotypes may provide clear advantages in environments in which bacteria may find themselves after colony dispersal. For example, 28% of isolates had a higher capacity to make biofilms, a growth mode that endows bacterial populations with greater capacity to resist various stresses. Increased capacity to form biofilms can also increase the virulence of E. coli pathogens because it can facilitate the generation of intracellular microcolonies, such as those observed in urinary tract infections (33, 36), and of the biofilms formed on small and large bowel epithelia by enteroaggregative E. coli (26). In addition, 12% of the isolates in this study became resistant to certain antibiotics, and some isolates had higher MICs to some antibiotics than the parental strain: 36% of isolates had higher tolerance to ROS, and 48% of isolates acquired the capacity to grow better on short-chain fatty acids as the sole carbon source than did the parental strain, a finding that suggests that these isolates had a more efficient glyoxylate pathway than the parental strain (75, 76). Consequently, these isolates could thrive better than the parental strain in host niches in which nutrients are limiting. For example, the glyoxylate pathway is required for the virulence of microorganisms during intracellular invasion (77–80).

In conclusion, the work presented here shows that spatial constraints in E. coli colonies increase evolutionary diversification within populations, probably by attenuating the exclusion dynamics between competing genotypes. The production of diverse offspring may allow bacterial populations to face ecological uncertainty. For example, our study demonstrates the emergence of novel metabolic capabilities that are likely to transform bacteria into more successful invaders of the host organism as well as the emergence of increased tolerance to certain stressors, including antibiotics. It is of major interest that heritable antibiotic tolerance emerged in the absence of antimicrobial agents. This may have significant practical implications for the treatment of bacterial populations in structured environments with antimicrobial agents.