Divergence and Redundancy of Transport and Metabolic Rate-Yield Strategies in a Single Escherichia coli Population

The energetic efficiency of nutrient uptake and conversion intobiomass is a key factor in the ecological behavior of microorganisms.The constraints shaping the metabolic rate-yield trade-off inbacteria are not well understood. To examine whether metabolicrate-yield settings and physiological strategies evolve towarda particular optimum in a constant environment, we studied multipleEscherichia coli isolates evolving in a glucose-limited chemostatpopulation. A major divergence in transport and metabolic strategieswas observed, and the isolates included inefficient rate strategists(polluters or cheaters) and yield strategists (conservationists),as well as various hybrid rate-yield strategists and alternativeecotypes (dropouts). Sugar transport assays, strain comparisonsbased on metabolomics, and Biolog profiling revealed varianceto the point of individuality within an evolving population.Only 68 of 177 metabolites assayed were not affected in 10 clonallyrelated strains. The parallel enrichment of rate and yield strategistsand the divergence in metabolic phylogenies indicate that bacteriado not converge on a particular rate-yield balance or uniqueevolutionary solutions. Redundancies in transport and metabolicpathways are proposed to have laid the framework for the multiplicityof bacterial adaptations.

INTRODUCTION

Top

Introduction

Microorganisms exhibit an immense range of metabolic capabilities,and there are extensive metabolic differences between membersof a single bacterial species (29). Metabolic capabilities andthe efficiency of metabolism are important in the ecologicalspecialization of all organisms (1). Well-known nutritionalstrategies involving r- and K-adapted organisms differ in termsof their metabolic rate-yield balances (24). Metabolic efficiencyis affected by the energetics of nutrient uptake (31) and isalso thought to be governed by the trade-off between the rateand yield of energy metabolism (35). Shifts in this trade-offare poorly documented. Information on what determines metabolicefficiency would help to underpin analyses of the complex architectureof metabolic networks and to determine why, for example, systemsbiology models of extant organisms exhibit multiple metabolicflux distributions that are equally optimal (26, 38). Advancesin modeling the systems biology of bacteria are being made (16),but many important questions still remain. Does metabolic plasticityextend to the rate-yield balance of organisms? How do the environmentand evolutionary selection determine the overall efficiencyof metabolism? How is the yield-rate trade-off shifted withprolonged nutrient limitation? In this study we examined thesequestions after perturbing metabolism through continuous cultureof Escherichia coli in a glucose-limited environment.

Highly relevant to the questions posed above is that, perhapsparadoxically, one characteristic of bacteria is that microbialgrowth yields are often 50% less than the optimal yield (48).It is not clear why free-living bacteria like E. coli, whichoften grow and compete in nutrient-limited environments, evolvedto have suboptimal growth yields. In our experiments, a singlepopulation of E. coli was subjected to prolonged resource limitation.When a bacterium faces a nutrient-limited environment, the intuitiveexpectation is that the organism adapts by increasing its efficiencyfor using the limiting substrate, or "K selection" (24). Onthe other hand, bacteria in populations may not have the luxuryof evolving in this way, and maximization of the growth yieldor metabolic efficiency does not appear to have been a dominantfactor in the evolution of bacteria like E. coli (48). In situationswhere there is kinetic growth competition among members of thesame culture, perhaps there is more selection for rate (r) thanfor yield (17). In a recent study MacLean and Gudelj lookedat competition between energetically efficient and inefficientyeasts with temporal and spatial variation (25), but researchershave not exhaustively studied how prolonged nutrient limitationin an unstructured environment with a single resource or nicheinfluences metabolic strategies, how the yield-rate trade-offis shifted with long-term selection, and whether redundancyin fitness solutions occurs.

Metabolic individuality in species such as E. coli (29) is,of course, only one facet of the significant intraspecies variationseen in bacteria (11). In most discussions researchers suggestthat diversity arises from adaptive radiation into the myriadof structured environmental and biotic niches or from selectionfor specialization on alternate resources (21, 25, 41). It hasalso been suggested that the trade-off between the rate andyield of energy metabolism is influenced particularly by howenvironments are structured, with multiple niches or resources(23, 35). Investigation of metabolic adaptation in simple environmentslimited by a single resource has been limited to a small numberof isolates from long-term cultures, like nutrient-limited chemostatcultures (39, 47). Long-term coexistence of phenotypically distinctclones has been demonstrated (10, 39, 40), and metabolic variancehas been described for the creation of cross-feeding consortiapartitioning metabolism in a long-term experimental population(36, 43). In general, population heterogeneity determined throughphenotypic characterization was found in other studies of experimentalevolution with E. coli (15, 18, 32, 34, 42). Recent work hasadded another dimension to the complexity of unstructured populationsand has shown that there is no ecological or periodic selectionlimitation to phenotypic and genotypic divergence in singlebacterial cultures (27). Here we describe the metabolic andenergetic consequences of an adaptive radiation in a constantunstructured environment with a single resource.

MATERIALS AND METHODS

Top

Materials and Methods

Bacterial strains and growth conditions. For the evolution experiment, E. coli K-12 strain BW2952, anMC4100 derivative, was propagated for approximately 600 h (or90 generations) at a growth rate of 0.1 h^–1 in a glucose-limitedchemostat exactly as described previously (33). Individualswere isolated from the culture each day, and the nine isolatesanalyzed were obtained by streaking a 600-h sample on nutrientagar and randomly picking colonies (27). Bacterial isolateschosen for metabolic assays and metabolome analyses were grownin glucose-limited chemostats with minimal medium A (MMA) ata dilution rate of 0.1 h^–1 as described previously (33).

Uptake of glucose and oxygen. The transport of glucose was measured by using the initial rateof uptake of 0.5 µM [U-¹⁴C]glucose added to bacteria withdrawnfrom the chemostat culture. The contributions of the Mgl andPtsG systems were determined from the uptake of 1 µM [U-¹⁴C]galactoseand 10 µM methyl- ${alpha}$ -D-[U-¹⁴C]glucoside, respectively. A50-µl sample was taken directly from the chemostat andmixed with 50 µl MMA containing 1 µM [U-¹⁴C]glucose,2 µM [U-¹⁴C]galactose, or 20 µM methyl- ${alpha}$ -D-[U-¹⁴C]glucoside.Samples (20 µl) were removed at 10-, 20-, 30-, and 40-sintervals and immediately filtered through 0.45-µm cellulosenitrate membrane filters and washed with 10 ml MMA without acarbon source. The filters were counted in liquid scintillantby using a Beckman scintillation counter (LS6500; Beckman, UnitedStates), and the initial rate of transport was calculated. Therate of transport was expressed in picomoles of sugar transportedper minute per 10⁸ bacteria.

Oxygen uptake was measured with a Clarke-type oxygen electrode(Ranks Brothers Ltd., England) as described in the manufacturer'sguide. Briefly, 3 ml of culture from a 24-h-old chemostat wastransferred into the oxygen electrode chamber at 37°C. Afterdetermination of the oxygen uptake rate due to oxidation ofendogenous metabolites, a glucose solution was added (finalconcentration, 15 µM) using a fine syringe needle. Theglucose-dependent rates were based on a solubility of O₂ of6.7 mg/liter at 37°C and atmospheric pressure.

Metabolism of glucose to acetate or CO₂. To analyze the extracellular metabolites of wild-type and chemostat-adaptedclones, 5-ml samples from glucose-limited chemostats were immediatelyfilter sterilized by suction through a 0.22-µm MillexGPfilter unit (Millipore, United States). Metabolites in the filtrateswere analyzed by determination of ¹H chemical shifts and theirassociated correlations using the comprehensive magnetic resonancespectroscopy methods described previously (2). For magneticresonance spectroscopy, a 550-µl cell-free culture wasmixed with 50 µl deuterium oxide and analyzed as describedpreviously (2). Acetate in samples was quantified using an aceticacid UV method assay kit (Boehringer Mannheim/R-Biopharm, Germany).

Rates of complete oxidation of glucose by wild-type and evolvedisolates were determined from the conversion of [U-¹⁴C]glucoseto ¹⁴CO₂. One milliliter of a culture whose optical densitywas known was removed from a 24-h-old glucose-limited chemostatculture (growth rate, 0.1 h^–1) and transferred to a 5-mlflask with a separate well (diameter, 7 mm; height, 10 mm) containing50 µl of 0.1 M NaOH. Sixty microliters of glucose (0.65mM), including 5 µl of [U-¹⁴C]glucose (0.65 mM; 306 mCi/mmol),was added to the culture, which was incubated for 5, 10, 15,or 30 min in the sealed flask. The ¹⁴CO₂ produced was trappedin a strip of Whatman chromatographic paper (thickness, 0.18mm; width, 5 mm; length, 10 mm) soaked in the 50 µl of0.1 M NaOH in the well. The strips of paper were transferredinto vials containing 4 ml of scintillation fluid (ACS; Amersham,England), and the ¹⁴CO₂ was measured using a Beckman scintillationcounter (LS6500; Beckman, United States).

Physiological parameters. In the chemostat culture, the dilution rate and thus the specificgrowth rate were constant (0.1 h^–1). Hence, specific glucoseconsumption and acetate production rates were determined fromthe difference between the concentration in the feed mediumand the concentration in the culture effluent, normalized tothe steady-state biomass concentration in each culture.

Biolog assay for substrate utilization screening. The catabolic activities of the starting strain and the chemostatisolates with 95 substrates were determined using the commerciallyavailable Biolog GN2 (Biolog, Hayward, CA) exactly accordingto the protocol provided by the vendor. Color changes were measuredat 600 nm using a microplate reader (Multiskan RC; Labsystem,Finland) after 24 h of incubation at 37°C and were correctedfor the no-substrate control. The optical densities at 600 nmwere used for cluster analysis after scoring for each substrateas follows: 0, no catabolism; 1, catabolism of the substrate.The cutoff point between negative results and positive resultswas an optical density at 600 nm of 0.2. Substrates giving valuesclose to the cutoff point were considered positive or negativeif there was at least a fourfold difference in replicate readingsbetween the ancestor and isolate. Substrates that gave variableresults or substrates whose values hovered around the cutoffpoint in duplicate assays were considered noninformative (5)and were not used for cluster analysis. There were three suchsubstrates.

Metabolome analysis. The ¹⁴C-labeling and extraction conditions used for bacterialcultures have been described in detail elsewhere (28). For phylogeneticcomparisons (29), the extracts were resolved in two dimensionson high-performance thin-layer chromatography (HPTLC) plates(Silica Gel 60; Merck, Germany) using solvent systems A andB (28, 44). Metabolite spots were detected on plates using aMolecular Dynamics PhosphorImager (Typhoon 8600; Amersham Biotech,United States) after 3 to 4 days in contact with a MolecularDynamics storage phosphor screen. PhosphorImager files wereprocessed using the Molecular Dynamics ImageQuant and Phoretixsoftware (30). The Phoretix parameter settings used for spotdetection were as follows: sensitivity, 200; operator size,31; noise factor, 5; and background, 0. The intensity of eachspot was normalized as a percentage of the total pixels on theplate. Data from spot matching with each isolate were used forcluster analysis after they were scored as follows: 1, spotpresent; 0, spot absent. For quantitative scoring, we used thefollowing scale: 1, least intense spot used in the analysis;2, two- to threefold difference in intensity from the leastintense spot; 3, three- to sixfold difference in intensity fromthe least intense spot; and 4, more-than-sixfold differencein intensity from the least intense spot. Spots with an intensitythat was <0.05% of the total pixel count in the analysisand spots which showed variable results in replicate analysiswere considered noninformative and not included in the clusteranalysis. At least three independent determinations were usedfor generation of the metabolome tree. A total of 177 spotswere detected in the ancestral strain and the isolates analyzedfrom the chemostat culture. These spots did not include low-abundancemetabolites, so the data were underestimates of the total metabolome(28, 44) but provided information that was more than sufficientfor cluster analysis and strain differentiation within a species(29).

Cluster analysis. We related differences in the metabolome and Biolog profilesof different E. coli isolates, including chemostat isolates,by using the neighbor-joining method for tree construction basedon distances estimated by the two-parameter method (37). Thereliability of the trees was confirmed by bootstrap analysisof 1,000 replications.

RESULTS

Top

Results

We examined metabolic adaptations to glucose limitation in apreviously characterized, rapidly evolving population with highendogenous mutation rates (27, 33), as shown in Table 1. Thenine strains included in Table 1 coexisted after 90 generationsor 26 days of glucose limitation but belonged to separate phylogeneticclusters within the population when 11 informative phenotypiccharacteristics were examined (Fig. 1a) (27). Metabolic andenergetic characterizations were performed in four distinctways: first, by analysis of metabolic inputs (changes in glucoseand O₂ uptake rates); second, by analysis of the end productsof energy metabolism (biomass yields, CO₂, and acetate); third,by metabolome analysis of intracellular metabolite pools; andfourth, by Biolog screening for changes in the metabolism of95 extracellular substrates.

View this table:
[in this window]
[in a new window]

TABLE 1. Properties of the ancestor (BW2952) and isolates coadapted to glucose limitation^a

View larger version (18K):
[in this window]
[in a new window]

FIG. 1. Phylogenetic relationships among E. coli isolates that evolved in a single population. The sympatric divergence of phenotypic, genotypic, and metabolic characteristics of evolved clones was analyzed by the neighbor-joining method rooted with the ancestral strain, BW2952, as described by Pupo et al. (37). Cluster A contains rpoS mutants, and cluster B strains are low-ppGpp strains (27). The dendrogram (a) was based on the 11 characteristics of 41 isolates described in reference 27. The metabolome tree (b) was based on spot matching and quantitation as described in reference 29. The number of informative spots compared for all isolates was 177. The Biolog tree (c) was based on the 23 informative substrates differentially utilized by the isolates and ancestor shown in Table 2, including all members of cluster B. The bootstrap values at nodes are percentages based on 1,000 replications.

Both the maximal growth rate and the growth yield (or efficiencyof glucose conversion into biomass) were distinctly heterogeneousin the isolates under glucose limitation conditions (Table 1),consistent with previously observed differences between individualisolates in other studies (8, 19). Three strains (BW4001, BW4005,and BW4006) were notable in that the yields of all three werestrongly reduced and two of them had reduced growth rates athigh glucose levels. These strains appeared as "small-colonyvariants" on agar plates and were previously observed in otherexperimental evolution studies (19, 45). In addition to thispolymorphism, Table 1 shows that the nine isolates did not convergeon a single optimum for glucose and O₂ uptake rates or in termsof the production of waste products of energy metabolism (CO₂and acetate).

The uptake of a limiting nutrient was assayed using low concentrationsof ¹⁴C-labeled substrates. Perhaps counterintuitively, the prolongedselection under glucose limitation conditions did not resultin uniformly high rates of uptake of glucose in evolved strains.There was variation in both the overall rates and the relativecontributions of the two major routes of glucose transport intoE. coli (the Mgl and PtsG transport systems [13]) (Table 1).The Mgl and PtsG systems have distinct energetic costs, anddifferential usage of these systems results in altered ATP balances(13). BW4005 and BW4036 exhibited, if anything, a lower rateof glucose uptake than the ancestral strain. These isolatesmost likely diverged into distinct ecotypes and evolved to competefor resources other than glucose in the long-term chemostat.The results of experiments examining competition between isolatesand between isolates and the ancestor are consistent with thisconclusion (Table 1) (27). The other isolates exhibited slightlyto very elevated glucose uptake rates; BW4002 and BW4006 weremore reliant on PtsG, whereas BW3767 and BW4004 were heavilyreliant on MglBAC. The large errors in some of the reportedvalues are due to instability between replicate cultures, probablybecause of unstable genomic rearrangements affecting transportergenes (unpublished results). Evidently, there were multiplefitness solutions for glucose transport, and the ancestral redundancyin glucose uptake pathways provided the bacteria with alternativeevolutionary solutions to the same selection conditions.

Oxygen uptake rates also evolved and were dissimilar in differentisolates (Table 1). Except for one strain (BW4029), chemostat-evolvedbacteria had elevated respiration rates with glucose as an oxidizablesubstrate. The low-yield bacteria had particularly elevatedrespiration rates, raising the possibility of faster energymetabolism and increased glucose throughput to CO₂. The elevatedrespiration rates suggested that there was an emphasis on ratesolutions rather than efficiency solutions for competition undernutrient limitation conditions. Consistent with this notion,the rates of CO₂ production from glucose were significantlydifferent for different strains, as also shown in Table 1. Inthe low-yield isolates, such as BW4001 and BW4006, there wassignificantly elevated conversion of glucose into CO₂. Furtheranalysis of culture media using proton nuclear magnetic resonancespectroscopy of fermentation products (2) showed that acetatewas a prominent peak in the medium of one strain (BW4005). Nomajor peaks corresponding to soluble carbon products were detectedin the media of other isolates. The acetate production by BW4005was confirmed by an enzymatic assay (Table 1). Interestingly,acetate production was not found in the phenotypically mostsimilar strain, BW4003 (Fig. 1a).

The changes in metabolism adopted by each isolate are shownschematically in Fig. 2; an ecological interpretation is shownin Table 1. The low-yield isolates (BW4001 and BW4006) exhibitedsignificantly elevated levels of respiration and conversionof glucose into CO₂; in these strains there was acceleratedenergy metabolism, but at the cost of a reduced biomass yield(>25% reduction). Despite the stable environment, these strainsare evidently inefficient r strategists. BW4005 appears to besimilar and produces acetate as well as CO₂, but in contrastto BW4001 and BW4006, it was not competitive against the ancestorin reconstituted glucose-limited chemostats (27). This suggestedthat the evolutionary success of BW4005 was due to its behavioras a new ecotype, dependent on unidentified resources in theevolving population and not necessarily on being effective inglucose utilization. BW4003 was also difficult to assess interms of ecological strategy, as it had moderately elevatedrates of transport, respiration, and fitness but the yield efficiencyon glucose was unchanged. More clear-cut was BW4029's solutionto fitness; this strain had a >20%-higher growth efficiencythan the ancestor and exhibited a decreased loss of resourcesthrough respiration. Hence, a K strategy also resulted in competitivefitness under glucose limitation conditions. Mutational changesin the redundant but energetically different respiratory chainsof E. coli can increase growth yields (3), so mutations in thesecomponents may explain altered respiration. BW4036, like BW4029,exhibited greatly decreased CO₂ production, but the transportor growth yields of BW4036 were not particularly altered, soagain it was unique. In several other high-biomass strains (BW3767,BW4002, and BW4004) yet another hybrid approach was observed;there were significantly elevated levels of transport, respiration,and CO₂ production but distinct 10 to 20% increases in the growthyields. Most strikingly, evolution with at least six differentmetabolic rate-yield strategies occurred side by side with asingle limiting resource. In addition, an additional strategy,the acetate cross-feeding polymorphism noted by Treves et al.(43), could be used if acetate producers such as BW4005 eventuallybecome more common in a population.

View larger version (30K):
[in this window]
[in a new window]

FIG. 2. Heterogeneity of metabolic processes in coevolved bacteria. The thickness of the arrows indicates the magnitude of the measured fluxes in the ancestral strain (light box) and isolates (shaded boxes). The rates of glucose and oxygen uptake and the rates of production of biomass, CO₂, and acetate are based on the data in Table 1. The numbers in the ellipses are the last two digits of the strain numbers in Table 1. Dashed lines indicate undetectable rates.

Additional evidence for metabolic divergence was obtained fromthe analysis of the intracellular metabolite pools in isolates.Metabolome analysis by two-dimensional HPTLC provides a powerfultool for typing bacteria at the subspecies level (29) and, asshown in Fig. 3, also resolves clonally evolved differences.Analysis of intracellular metabolites present in the startingstrain and the nine representative isolates (Fig. 4) revealedmajor differences in metabolism; none of the isolates testedin the various clusters shown in Fig. 1a had a pattern of metabolites(presence or absence of spots or differences in quantity) (Fig.3a and b and 4) identical to the pattern of another strain orto the pattern of the starting strain when the organisms weregrown individually in a glucose-limited chemostat. The changesin the evolved strains were far greater than the changes obtainedby repeated analysis of the ancestor strain grown under theselection conditions used (29). Use of the data for 177 metabolitespots reproducibly detected for all the strains to constructa dendrogram (Fig. 1b) showed that the metabolome profiles werevery distinctive for the isolates tested. The differences betweenisolates included both increases and decreases in the amountsof different classes of metabolites, such as UDP-sugars, citricacid cycle intermediates, or trehalose, compared to the amountsin the starting strain (Fig. 3g). After selection, there wasa >10-fold difference in the concentrations of common centralmetabolites, like fumarate, between isolates. The clusteringin the metabolome-based dendrogram shown in Fig. 1b was generallyconsistent with the phenotypic division of the strains shownin Fig. 1a, with both clusters A and B identified by the independentapproaches. The other, more dispersed strains in Fig. 1a wererelated somewhat differently as determined by metabolomics.Metabolomic differences thus largely paralleled phenotypic divergence,but isolates with very similar phenotypic traits (e.g., BW4002and BW4004 or BW4003 and BW4005) were still differentiated inmetabolomic profiles, and there were 34 spot differences (20differences in the presence of a spot; 14 differences in thequantity of a spot) between BW4002 and BW4004.

View larger version (57K):
[in this window]
[in a new window]

FIG. 3. Metabolome profiles for chemostat-evolved and ancestral clones. The metabolites were extracted (28) from E. coli K-12 strain BW2952 (a and d), the ancestral strain in chemostat experiments (33), and BW3767, a cluster A isolate obtained after 90 generations (b and e). The images are PhosphorImager outputs for ¹⁴C-labeled metabolites separated by two-dimensional HPTLC using pairs of solvent systems. In panels a and b the compounds were separated using solvent system A, and panels d and e were obtained using solvent system B (28, 44). Resolved spots in metabolome fingerprints were matched using the Phoretix software (30). (c and f) Results of an analysis of matched spots in the evolved and ancestral metabolomes with solvent systems A and B, respectively. ${triangledown}$ , metabolite spots that were down-regulated; ${triangleup}$ , metabolite spots that were up-regulated; –, metabolite spots that were not present; +, metabolite spots that were unique in the evolved clone. Spots with no symbol were the metabolites whose proportions were the same in the two metabolomes. (g) Variation in the metabolome proportions for four metabolite spots in the isolates. The spots identified as glutamate, UDP-hexose, and fumarate using methods described previously (28, 44) were quantitated in four replicate metabolome estimations. The error bars indicate standard deviations.

View larger version (90K):
[in this window]
[in a new window]

FIG. 4. Metabolome profiles of ancestral and evolved clones. The images are the phosphorimages of ¹⁴C-labeled metabolites of ancestral strain BW2952 and isolates BW3767, BW4001 to BW4006, BW4029, and BW4036 that coevolved in the 90-generation chemostat population. Extracts were prepared and separated by two-dimensional HPTLC using solvent systems A (bottom 10 images) and B (top 10 images) as described in the legend to Fig. 3.

The metabolome analysis revealed an even more surprising result:only 68 of the 177 metabolite spots studied were conserved inall 10 strains, indicating that there was major restructuringof metabolism in different isolates. These results demonstratethe huge metabolic plasticity in E. coli, already suggestedby the low number of conserved metabolites in diverse membersof the species E. coli (29).

Testing the transformation of exogenous substrates providedanother perspective on the metabolic variance within the chemostatculture, which was analyzed using the compounds present in 96-wellBiolog plates (5). As shown in Table 2 and Fig. 1c, Biolog GN2profiling strongly differentiated the chemostat isolates fromthe ancestor. All isolates gained and lost abilities to utilizevarious substrates during the 90 generations of glucose-limitedgrowth (Table 2), consistent with numerous mutations in thelifetime of the population. Gains in metabolic capabilitiesdue to pleiotropic mutations in rpoS, leading to improved metabolismof poor substrates (22), was evident with rpoS mutants BW3767,BW4002, and BW4004 (27). Altogether, the changes in the 23 informativesubstrates permitted construction of the dendrogram shown inFig. 1c. In most respects, including the separate grouping ofthe rpoS bacteria (cluster A), evolved differences in the metabolismof external substrates gave the same phylogenetic characteristicsas phenotypic and metabolomic typing. However, the trees obtainedby metabolomic and Biolog analyses were not entirely congruent,suggesting that intracellular metabolite pools and extracellularsubstrate utilization capabilities do not reflect the same mutationaladaptations, particularly in the rpoS⁺ bacteria that do notbelong to cluster A. Biolog profiling was extended to a moredetailed analysis of all members of one phenotypic cluster (clusterB), as shown in Fig. 1a. All of the isolates characterized bya reduced growth yield were closely related, but they exhibitedindividual differences. As shown in Fig. 1c, all nine isolatesin cluster B were differentiated. Together with the phenotypicdivergence (27), these results are indicative of concurrentand on-going variegation in the culture within a much shortertime scale than previously tested in experimental populations(21, 34, 39). In yeast cultures metabolic properties also changewithin 250 generations during glucose-limited growth (12), butthe extent of intrapopulation divergence and the rapidity ofaccumulation of the metabolic changes were not suspected previously.

View this table:
[in this window]
[in a new window]

TABLE 2. Informative substrates from Biolog analysis of evolved isolates^a

DISCUSSION

Top

Discussion

In this study, we used a domesticated laboratory strain of E.coli and subjected it to prolonged growth in an artificial (chemostat)environment. The population size also may not represent thesize of an E. coli population in a natural habitat. Despitethese artificial aspects, the extent of biological change observedwith a domesticated strain can still tell us a lot about theunderlying flexibility of bacterial responses to a new environment.Although one population was analyzed, seven other chemostatpopulations exhibited the same level of divergence, althoughthey did not always exhibit the same combination of changes(S. Seeto and T. Ferenci, submitted for publication).

Our bioenergetic and metabolomic results are indicative of rapid,multiple metabolic adaptations not previously recognized inexperimental bacterial populations (16, 39). The scope of sympatricdivergence and the rapidity of accumulation of such changeswere also not recognized in yeast cultures during glucose-limitedgrowth (7). Extrapolated to an ecological context, these resultsobviate the postulated need for multiple niches for bacterialdiversification (21) or for adapting energetic efficiency (35).Contrary to predictions (35), derivatives of a single parentalstrain in a constant environment diverge to take both sidesof what was proposed previously to be a fundamental rate-yieldtrade-off affecting growth energetics. The great plasticityof the yield-rate balance suggests that these characteristics,as well as other metabolic properties (14), are frequently readjustedin bacterial evolution and are not fundamental properties oforganisms. The robustness of bacterial evolution is underpinnedby the capacity for metabolic adaptive radiation, even in ahomogeneous, nearly constant environment. Rapid metabolic divergencein bacteria under suboptimal nutritional conditions, includinglosses and gains in the utilization of substrates, is undoubtedlya source of diversity and intraspecies heterogeneity (11).

The glucose-limited E. coli population partitioned into populationswith several approaches to resource use. As summarized in Table1, some isolates adopted energetically expensive glucose transportmechanisms (the Mgl system), whereas others adopted the moreeconomical PtsG system (31). Metabolically, some evolved isolateswere inefficient, with rapid respiration and production of largeamounts of CO₂. Another way of looking at this class is thatthey were "cheaters" depriving other organisms of availableresources (25). Concurrently, the conservationist strategy ofa minority of isolates involved increased energetic efficiencyand reduced waste. Hybrid strategies were particularly common,and these strategies involved various increases in energeticefficiency along with increased rates of substrate transformation.Alternative ecotypes (4) or dropouts no longer competing forthe original resource used alternative energy sources not usedby the ancestor population. In extending the conclusions fromthe nine isolates studied to the population as a whole, it isworth noting that of the 41 coevolving isolates studied in previousexperiments (27), 9 were wasters, 18 exhibited the oppositeadaptation and increased growth yields (mostly with hybrid strategies),and 12 maintained the same yield-rate balance as the parentalstrain. Only a minority, including the acetate producer, appearedto adapt into an alternative ecotype, and likewise, only oneisolate was an outright conservationist. Of course, these proportionsare representative of a particular time in a particular culture,and an additional week of selection resulted in an increasein the proportion of isolates derived from polluters (to about50% of the population) (results not shown).

The diversity in metabolic properties defined in this studymust reflect the availability of alternative evolutionary pathwaysto bacteria for adaptation to a glucose-limited environment.Of course, not all the changes have been proven to contributeto fitness. Some of the changes (e.g., loss of utilization ofsome substrates or loss of metabolite spots) may have been neutralunder the selection conditions used, but other changes wereclearly adaptive. Recent studies have suggested that prolongedselection with a single carbon source results in convergentmetabolic phenotypes resulting from different transcriptionalchanges (16). These and other previous studies completely missedthe potential for rapid and extensive metabolic diversificationin bacterial populations. Metabolomic data and Biolog profilingboth point to metabolic divergence to the point of individualityin populations. Bacteria adapt to the same metabolic problem(glucose limitation) by taking both sides of the rate-yielddivide and clonally evolve to maintain a very small shared coreof metabolite pool characteristics.

The inherent robustness of bacterial processes extends to evolutionaryoutcomes. This conclusion is consistent with the current viewthat bacterial metabolic pathways are robust, because many loss-of-functionmutations do not eliminate metabolic network capabilities andthe ability to grow (9). The same robustness and redundancyof functions probably contribute to the ability of bacteriato adapt their metabolism through alternative beneficial mutationsin challenging environments. Our evidence suggests that, asin the case of transport changes, robustness in this systemis due to redundancy, although the possibility of distributedrobustness cannot be excluded (46). Not only does metabolicand physiological redundancy stabilize already evolved organismsso that they can withstand immediate environmental challenges,but our results also suggest that robustness lays the foundationfor providing alternative evolutionary solutions in the presenceof prolonged stress.

Finally, what does the seemingly limitless flexibility of metabolicadaptation say about the modeling of metabolism in the systemsbiology of organisms? There is a growing expectation that computationalapproaches can explain biological systems (6, 20). It remainsto be seen whether the challenge of the explosive radiationrepresented by our results can be satisfactorily met by in silicomodels.

ACKNOWLEDGMENTS

We thank Bill Bubb for performing the nuclear magnetic resonanceanalysis.

We thank ARC for past funding support.

FOOTNOTES

* Corresponding author. Mailing address: School of Molecular and Microbial Biosciences G08, The University of Sydney, Sydney, NSW 2006, Australia. Phone: (61 2) 93514277. Fax: (61 2) 93514571. E-mail: tferenci{at}mail.usyd.edu.au.

Published ahead of print on 8 December 2006.

Present address: Department of Microbiology, 601 S. GoodwinAve., University of Illinois at Urbana-Champaign, Urbana, IL61801.

REFERENCES

Top