research articles transcription, translation, and the

RESEARCH ARTICLES

Transcription, Translation, and the Evolution of Specialists and Generalists

Shaobin Zhong,*� Stephen P. Miller,� Daniel E. Dykhuizen,� and Antony M. Dean�*Department of Plant Pathology, North Dakota State University; �Department of Ecology, Evolution, and Behavior, BiotechnologyInstitute, University of Minnesota; and �Department of Ecology and Evolution, State University of New York at Stony Brook

We used DNA microarrays to measure transcription and iTRAQ 2D liquid chromatography-mass spectrometry/massspectrometry (a mass-tag labeling proteomic technique) to measure protein expression in 14 strains of Escherichia coliadapted for hundreds of generations to growth-limiting concentrations of either lactulose, methylgalactoside, or a 72:28mixture of the two. The two ancestors, TD2 and TD10, differ only in their lac operons and have similar transcription andprotein expression profiles. Changes in transcription and protein expression are observed at 30–250 genes depending onthe evolved strain. Lactulose specialists carry duplications of the lac operon and show increased transcription andtranslation at lac. Methylgalactoside specialists are galS– and so constitutively transcribe and translate mgl, whichencodes a transporter of methylgalactoside. However, there are two strains that carry lac duplications, are galS–, andshow increased transcription and translation at both operons. One is a generalist, the other a lactulose specialist. Thegeneralist fails to sweep to fixation because its lacþ, galSþ competitor expresses the csg adhesin and sticks to thechemostat wall, thereby preventing complete washout. Transcription and translation are sometimes decoupled. Lactulose-adapted strains show increased protein expression at fru, a fructose transporter, without evidence of increasedtranscription. This suggests that fructose, produced by the action of b-galactosidase on lactulose, may leach from cellsbefore being recouped. Reduced expression, at ‘‘late’’ flagella genes and the constitutive gat operon, is an adaptation tostarvation. A comparison with two other long-term evolution experiments suggests that certain aspects of adaptation arepredictable, some are characteristic of an experimental system, whereas others seem erratic.

Introduction

One obvious product of evolution is life’s diversity—trees, birds, fish, germs, etc. Much of this diversity reflectsecological specialization. Trade-offs are commonly as-sumed to cause ecological specialization, yet rarely aredemonstrated in rigorous controlled experiments. Othermechanisms can also produce specialists. Adaptation in dif-ferent environments leads to independent specializations.Escherichia coli adapting to high temperatures do not losefitness at low temperatures; E. coli adapting to low temper-atures do not lose fitness at high temperatures (Bennett andLenski 1993, 2007). Accumulation of mutations that areneutral in the current environment, but deleterious in otherenvironments, such as temperature-sensitive mutations(Bennett and Lenski 1993) can produce specialists (Cooperand Lenski 2000). Independent specialization and mutationaccumulation produce specialists without trade-offs.

Trade-offs, independent specializations, and mutationaccumulation are not mutually exclusive mechanisms. Ef-forts to quantify their relative contributions in long-term ad-aptation experiments remain inconclusive (Elena andLenski 2003; MacLean and Bell 2003). The role oftrade-offs in ecological specialization remains elusive.

We are exploring the evolution of specialists and gen-eralists using laboratory populations ofE. coli competing fortwo limiting sugars, lactulose (galactosylfructose) andmeth-ylgalactoside, either singly or as a72:28mixture (DykhuizenandDean 2004; Zhong et al. 2004). Theory predicts, and ex-periments demonstrate, that two specialists may coexistwhenever differential resource consumption generates stabi-

lizing frequency-dependent selection (Lunzer et al. 2002).Small changes in fitness are predicted to destabilize the poly-morphism, resulting in a selective sweep. Nevertheless, bothinitial specialists are usually retained for extendedperiods. Inthose cultures where one is lost, two newly evolved special-ists derived from the remaining strain can be isolated. Notonly do polymorphisms of specialists routinely evolve inlong-term chemostat cultures but, even more remarkably,strains commonly switch resource specializations. General-ists are rare. The repeated independent evolution of resourcespecialists and resource switching strongly suggests, but isnot definitive proof of, a role for trade-offs.

Genomic analysis of evolved strains (Zhong et al. 2004)reveals a major role for insertion sequences (ISs) in adapta-tion. IS-inducedduplications of the lacoperon are associatedwith specialization toward lactulose. IS insertions in galSdisrupt the repressor of the mgl transport system and areassociated with specialization toward methylgalactoside.Only one of 42 strains analyzed carries both a lac duplicationand a galS disruption and appears to be a generalist. Other ISmutations, deletions extending into the fli operon (part of theflagellar regulon), and two deletions of the constitutivelyexpressed gat operon, presumably confer their benefits byeliminating unnecessary gene expression. Regulation andgene expression are evidently major targets of adaptation.

RNA transcription profiling has been used to quantifygene expression changes during experimental evolution.Ferea et al. (1999) found parallel changes in transcript lev-els at genes responsible for glucose uptake and utilization inthree lines of yeast evolved in glucose-limited cultures.Cooper et al. (2003) compared transcript profiles of twostrains evolved for 20,000 generations by serial transferin glucose minimal medium and showed parallel changesin 59 genes. Riehle et al. (2005) identified gene expressionchanges, some of which might be related to the evolutionof the temperature niche. With the exception of the in-creased expression of transport systems during prolongedstarvation (Novick and Horiuchi 1961; Horiuchi et al.

Key words: evolution, specialists, generalists, protonomics, E. coli,lac, chemostats.

E-mail: [email protected].

Mol. Biol. Evol. 26(12):2661–2678. 2009doi:10.1093/molbev/msp187Advance Access publication August 25, 2009

� The Author 2009. Published by Oxford University Press on behalf ofthe Society for Molecular Biology and Evolution. All rights reserved.For permissions, please e-mail: [email protected]

Dow

nloaded from https://academ

ic.oup.com/m

be/article/26/12/2661/1532118 by guest on 26 Novem

ber 2021

1962, 1963; Dean 1989; Sonti and Roth 1989; Ferea et al.

1999; Zhong et al. 2004), the deletion of unneeded consti-

tutively expressed genes (Zhong et al. 2004), and the in-

creased expression of enzymes involved in cross-feedingin an evolved commensalistic community derived froma single clone of E. coli (Rosenzweig et al. 1994), the phys-iological mechanisms underpinning adaptation to novel en-vironments are not well understood, even when parallelbeneficial mutations have been identified.

Several studies have shown that changes in transcriptionneed not bematched by similar changes in protein expression(Griffin et al. 2002; Corbin et al. 2003; Greenbaum et al.2003). This decoupling of transcription from translationmeans that microarray data alone are insufficient to makestrong inferences with respect to organismal phenotypes,fitnesses, and adaptation. Although essential tomany studies,microarray data should form only part of a more comprehen-sive approach to the study of adaptive evolution.

Here, we explore the evolution of gene regulation us-ing microarray RNA transcript profiling and iTRAQ two-dimensional liquid chromatography mass spectrometry/mass spectrometry (2D LC–MS/MS) (Ross et al. 2004) pro-tein expression profiling. Our goal is to identify changes intranscript levels and protein expression associated withspecialization toward lactulose and methylgalactoside, toidentify other changes in expression associated with adap-tation to slow growth chemostats in general, and to deter-mine whether these changes are always associated with ISmobilization or whether other mutations (e.g., base substi-tutions) also contribute.

Materials and MethodsBacterial Strains

Escherichia coli strains TD2 and TD10 are the ances-tral strains used to initiate all long-term experiments (table 1).They carry different lac operons but are otherwise geneti-cally identical. Strains designated R (e.g., TD10R) carrya selectively neutral genetic marker, fhuA, that confers re-

sistance to the bacteriophage, T5. TD10(R) and TD2(R)form a frequency-dependent balanced polymorphism ona 72:28 lactulose:methylgalactoside mixture (Lunzer et al.2002). Long-term chemostat cultures were initiated witha pair of strains, one T5 sensitive and one T5 resistant(e.g., TD10 with TD10R or TD2R and TD10, etc.). Strainsdesignated DD (e.g., DD2261) are isolates from long-termchemostat cultures with lactulose, methylgalactoside, ora 72:28 mixture of both as limiting resources (Dykhuizenand Dean 2004). After hundreds of generations of evolu-tion, one sensitive and one resistant isolate were randomlychosen from each chemostat. Isolates from chemostats fedmixed sugars were shown capable of reestablishing theequilibrium T5R frequency seen in each evolved popula-tion at the time of sampling (and which had evolved awayfrom the initial equilibrium). This demonstrates that the iso-lates are representative of the dominant lineages in eachchemostat (Dykhuizen and Dean 2004).

The fitness of TD10R with respect to TD2 is 0.91 onlactulose and 1.31 on methygalactoside (table 1). Similarly,for example, the fitness of DD2255 with respect to DD2253is 1.11 on lactulose and 0.78 on methylgalactoside. Special-ists are defined by their fitnesses: DD2255 is a lactulosespecialist and DD2253 is a methylgalactoside specialist.Fitnesses are not given for strains isolated from chemostatslimited by single sugars. Partitioning the contribution tofitness made by specialization toward the limiting sugarfrom the contribution made by adaptation to the chemostatenvironment in general will require extensive genetic ma-nipulations that, though planned, are beyond the scope ofthis study.

Transcript Profiling

RNA Extraction

Strainswere grown at 37 �C in 30-ml chemostats at a di-lution rateof0.3h�1 inminimalmedium(DavisSalts;40mMK2HPO4,15mMKH2PO4,7.6mM(NH4)2SO4,1.7mMNa3-Citrate, 1 mMMgCl2 at pH 7.3) (Miller 1972) with 0.2 g/l of

Table 1Evolved Strains (Dykhuizen and Dean 2004)

Chemostat Sugars Ancestor Strain Specializationa Generation Isolated Fitnessb on LU Fitnessb on MG

TD2 LU (ancestor)TD10R MG (ancestor) 0.91 ± 0.004 1.31 ± 0.004

1 LU TD2 DD2459 5981 LU TD2 DD2460 59810 MG TD2 DD2557 33210 MG TD2 DD2558 33219 Mix TD10 DD2268 Both 471 1.23 ± 0.02 1.22 ± 0.0119 Mix TD10R DD2269R 47120 Mix TD2R DD2253R MG 26020 Mix TD10 DD2255 LU 260 1.11 ± 0.01 0.78 ± 0.0121 Mix TD10R DD2302R LU 610 1.26 ± 0.01 0.90 ± 0.0121 Mix TD2 DD2304 MG 61022 Mix TD10R DD2261R LU 335 1.36 ± 0.02 0.48 ± 0.0122 Mix TD2 DD2262 MG 33523 Mix TD10 DD2266 MG 47123 Mix TD10R DD2267R LU 471 1.67 ± 0.001 0.39 ± 0.01

LU, lactulose; MG, methylgalactoside; Mix, 72:28 mixture of laculose:methylgalactoside; and R, resistance to phage T5.a Specialization as determined from relative fitnesses on pure sugars.b Fitness of evolved strain relative to its evolved partner.

2662 Zhong et al.

Dow


ic.oup.com/m


ber 2021

either lactulose or methylgalactoside as the sole limiting re-source and 10 lM isopropyl-b-D-thiogalactopyranoside(IPTG) (Dykhuizen and Dean 2004). After the populationdensity (OD600) had stabilized (18–20 h after inoculation),the entire culture from each chemostat was poured intoa 50-ml conical tube containing 3ml of ice-cold ethanol/phe-nol stop solution (5% water-saturated phenol, pH , 7.0, inethanol), centrifuged (4,000 rpm for 10min at 4 �C), and thecell pellets stored at�80 �C. Total RNAwas extracted usinga Master RNA purification kit (Epicentre, Madison, WI) ac-cording to the manufacturer’s instructions.

Microarrays and Data Analysis

Gene expression changes were identified by using par-allel 2-color cDNA hybridizations to whole-genome E. coliMG1655 spotted DNA microarrays, which were designedand printed as described and contained 98.8% of all anno-tated open reading frames (Khodursky et al. 2000, 2003).Cy3 dUTP or Cy5 dUTP (Amersham Pharmacia) was in-corporated into cDNA made from 15 to 20 g of total RNAusing Scribe First-strand cDNA labeling Kit (AmershamPharmacia). The labeled cDNA was purified using a Micro-con-30 (Millipore). Replicate experiments were performedwith a dye swap using two RNA samples from independentreplicate populations (the number of RNA preparations washigher for strains TD2 and TD10 because 1) they were usedin every hybridization and so more RNA needed to be ex-tracted and 2) some preparations were thrown out becausethey had already increased levels of expression at lac orgalS, both of which are strongly favored). Following datanormalization, outliers and missing data were replaced bymean log2 ratio values (evolved strain:ancestor) from theother replicates. A fixed model analysis of variance (AN-OVA) was performed for transcript level ratios of eachevolved strain with respect to its ancestor. No significantGenes � Dyes interaction effects were detected and dyesswaps were treated as four independent replicates. Eachanalysis therefore had Genes and Sugars as main effectsand a Genes � Sugars interaction effect. The Sugars maineffect is zero because the data were normalized. Significantchanges in transcript levels were identified using a 1% falsediscovery rate (Benjamini and Hochberg 1995).

iTRAQ Proteomics

Growth of Strains

Evolved strains were grown at 37 �C in 100-mlchemostats at a dilution rate of 0.3 h�1 using the same me-dium to which they had been adapted (minimal mediumwith 0.2 g/l of either lactulose, methylgalactoside, ora 72:28 mixture of both as the sole limiting resource and10 lM IPTG). The ancestral strains, TD2 and TD10, weregrown with lactose as the sole carbon/energy source toavoid selected changes in expression at lac and galS thatarise rapidly. After reaching steady state (22–24 h after in-oculation), the entire culture from each chemostat was har-vested by centrifugation (4,000 rpm for 10 min at 4 �C),quick frozen, and stored at �80 �C. Each pair of evolved

strains and their ancestors were grown in sets of four par-allel chemostats.

Protein Extraction and iTRAQ Labeling

For the proteomic studies, iTRAQ Reagents (AppliedBiosystems, Foster City, CA) were used to label proteinsamples, which enabled the simultaneous identificationand quantification of peptides/proteins from four differentsamples in a single experiment. For each experiment, thegrowth, protein extraction, and iTRAQ labeling of theevolved strains and their ancestral strain were done in par-allel sets of four.

Samples from four chemostats, including the ancestralstrain, were prepared and labeled in parallel for analysis by2D LC–MS/MS. The cell pellet from each chemostat wasfirst resuspended in 500 ll of 0.5 M triethylammonium bi-carbonate (pH 8.5) containing 0.1% CHAPS and 0.05% so-dium dodecylsulfate as denaturants. The suspensions weresonicated three times for 20 s each on ice with a BransonDigital Sonifier (Model 250) equipped with a microtip at25% amplitude. The suspensions were allowed to coolon ice between sonications. The samples were then centri-fuged at 16,000 � g for 10 min at 4 �C. The supernatants(cell extracts) were divided into aliquots and stored at�80 �C. Protein concentrations were determined usingthe Bio-Rad Protein Assay Reagent with IgG as a standard.

The four samples were labeled in parallel, each witha separate iTRAQ Reagent, as recommended by the man-ufacturer (Applied Biosystems). Briefly, each extract wasdiluted in sonication buffer to 5 mg/ml protein, a 100 lgof protein reduced, and all cysteines blocked with the re-agents supplied by the manufacturer. Proteins were digestedovernight at 37 �C with 4 lg trypsin and the resulting pep-tides covalently labeled with an isobaric iTRAQ reagent atlysine side chains, N-terminal groups, and tyrosines. It isimportant to note that a labeled peptide displays the samemass and chromatographic properties whether labeled withiTRAQ Reagent 114, 115, 116, or 117.

Strong Cation Exchange Chromatography ofDifferentially iTRAQ-Labeled Sample Mixture

The strong cation exchange (SCX) chromatography ofthe labeled peptide mixture was conducted by the Center forMass Spectrometry and Proteomics (University of Minne-sota, Minneapolis, MN). The four differentially labeledsample digests were combined, the mixture dried in vacuo,resuspended in 0.1% trifluroacetic acid (TFA), and appliedto a Sep-Pak C18 cartridge (Waters Corp., Milford, MA).After washing with 0.1% TFA, the peptide mixture waseluted with 80% acetonitrile (ACN) in 0.1% TFA. Theeluted peptide mixture was dried in vacuo, reconstitutedin 350 ll of Solution A (20% ACN, 5 mM KH2PO4 atpH 3.2), and subjected to SCX chromatography. The chro-matography was performed on a Magic 2002 high-performance liquid chromatography system (MichromBioresources, Auburn, CA) using a polysulfoethyl A col-umn (1.0 mm ID � 150 mm; 5-lm particles with 300 Apores; Poly LC, Columbia, MD). Peptides were eluted at

Specialists and Generalists 2663

Dow


ic.oup.com/m


ber 2021

a flow rate of approximately 33 ll/min with Solution A andSolution B (Solution A containing 500mMKCl) over a gra-dient of 0–20% B in 40 min, 20–100% B in 20 min, andconstant 100% for 10 min. The absorbance of the eluentwas monitored at 214 nm (peptide bonds) and 280 nm (ar-omatic residues) with fractions collected at 3-min intervals.Typically, the peptides eluted in 17 fractions, whichshowed mAU280 values .2. These fractions were driedin vacuo and then analyzed by reversed-phase LC–MS/MS.

Reversed-Phase LC–MS/MS Analysis

LC–MS/MS analysis of SCX fractionated peptideswasperformed by the Center forMass Spectrometry and Proteo-mics (University of Minnesota, Minneapolis, MN) ona QSTAR Pulsar i mass spectrometer (Applied Biosystems)with an online Dionex/LC Packings (LCP, Sunnyvale, CA)C18 capillary liquid chromatography system as describedpreviously (Nelsestuen et al. 2005). Each of the dried SCXfractions was reconstituted in 30 ll of the reversed-phaseloading solution (2% ACN, 0.1% formic acid). The entirevolume of fractions having an mAU280 value , 10 wasloaded onto the LCP C18 precolumn (0.3 mm ID �5 mm). Fractions having an mAU280 value . 10 (typically12 fractions) were loaded and run twice. Half of the samplewas loadedand the tandemmass spectrometry (MS/MS)datawere collected as described below. The remaining portion ofthe fraction was then loaded and data collected using anexclusion list containing the acquired precursormass/chargeratios (m/z) values from the first run.

After loading onto the C18 precolumn, each samplewas washed with the loading solution for 17 min at a flowrate of 35 ll/min. Peptides were then eluted at the same flowrate onto an analytical capillary C18 column (75 lm ID)with solvents A (5% ACN, 0.1% formic acid) and solventB (95% ACN, 0.1% formic acid) over a gradient of 0–35%B in 40 min, 35–80% B in 5 min, and 80–100% B in 2 min.Product ion spectra were collected in an information-depen-dent acquisition mode with continuous cycles of one fullscan from 400 to 1100 m/z per 1.5 s followed by four prod-uct ion scans from 50 to 2000 m/z at 3 s each. The fourprecursor m/z values with the highest intensities were au-tomatically selected for MS/MS fragmentation by theAnalyst QS software (ABI) from the MS scan during acqui-sition. Collision energy was increased 20% for fragmenta-tion of iTRAQ peptides. Neutral loss of the iTRAQReagentbalancer groups during MS/MS fragmentation producesfour reporter group ions in the 113–119 m/z region that canbe used to quantify peptide expression. Unlabeled b andy ions are also generated and used for peptide identification.

Data Processing

The identification and quantification of the relativeabundance of proteins was determined from the MS/MSdata using the ProteinPilot 2.0 software (Applied Biosys-tems). All of the MS/MS data files obtained from the pep-tide fractions of a single SCX chromatography weresearched together against the NCBI E. coli K12 protein da-tabase (RefSeq NC000913 and AC000091, Riley et al.

2006) which contains 8,772 proteins. The ‘‘thorough searcheffort’’ algorithm was used with the threshold for proteinidentification at 95% confidence. Common biological mod-ifications and amino acid substitutions were automaticallyincluded in every search. The ProteinPilot software deter-mines the relative abundance of each peptide in the evolvedstrain verses its ancestor by calculating the ratio of the peakareas of their iTRAQ reporter ions. These results are thencompiled into protein groups based on the database search.Peptides shared between distinct proteins are not used inquantification. The average of the peptide iTRAQ ratiosis calculated for each protein. This average ratio includesonly those peptides for which all four iTRAQ reporter ionswere detected. Data are normalized assuming that the ma-jority of proteins do not show differential expression andthat the median iTRAQ ratio of all the proteins is 1.

All proteins discussed were identified with a minimumof three peptides for quantification. The MS and MS/MSspectra of each was inspected manually and curated basedon the precursor ion spectra.

Standard Polymerase Chain Reaction (PCR) and DNASequencing

Primers, used to amplify galS (the repressor of the mgloperon) and gat (the galactitol operon was amplified in twosections), were designed using the genomic sequence of E.coli K12 strain MG1665. Herculase DNA polymerase(Stratagene) was used with the following cycling condi-tions: 95 �C for 15 min, 35 cycles of 94 �C for 30 s,60 �C for 90 s, 72 �C for 2 min, and a final extension stepat 72 �C for 10 min. Amplicons were sized by agarose elec-trophoresis with a 1-kb DNA ladder as a standard. GalSfragments were purified using StrataPCR purification col-umns and sequenced by the Advanced Genetic AnalysisCenter at the University of Minnesota.

Curli Expression

Expression of the extracellular adhesin protein Curliwas confirmed by the intense red staining of colonieson minimal glucose plates containing 0.1 g/l congo red(Hammar et al. 1995).

ResultsAncestral Strains TD2 and TD10

Ancestral strains TD2 and TD10 are genetically identi-cal derivatives of E. coli strain K12 that carry different lacoperons (Lunzer et al. 2002). Thus, except for the expressionof the lactose operon, we expect the transcription profile andthe protein expression for these two strains to be the same re-gardless of the environment. The next two sections test thisassumption. But they also test the ability to use transcriptionprofiling and proteomics to determine evolutionary change.If we findmany genes that are expressed differently betweenthese strains, we would question the reliability of thesemethods to distinguish evolutionary differences.

2664 Zhong et al.

Dow


ic.oup.com/m


ber 2021

Transcription in Ancestral Strains TD2 and TD10

Overall, transcript profiles are similar for each strainregardless of which sugar, lactulose or methylgalactoside,was the limiting resource (i.e., all TD2LU:TD2MG transcriptratios� 1 and all TD10LU:TD10MG transcript ratios� 1, sothat changing the chemostat sugar does not greatly affecttranscription). An ANOVA of lactulose:methylgalactosidetranscript ratios (table 2) produces no significant SxG(Strains by Genes) interaction term, indicating transcriptionresponds similarly in both strains to the change in resource.Significance of the Genes term indicates transcriptiondiffers during growth on lactulose and on methylgalacto-side. Seven genes (table 3) were identified with a 1% falsediscovery rate (Benjamini and Hochberg 1995). However,the changes in transcription are modest (generally lessthan 1.5-fold) and, with the possible exception of galP(a transporter that might be induced to scavenge galactose),there are no obvious reasons why any should be differen-tially expressed on the two sugars. We suspect thesechanges in transcript levels are either incidental or spuri-ous. Data normalization eliminates any difference in themean lactulose:methylgalactoside transcript ratios betweenthe strains.

When the ancestoral strains are compared directly, anANOVA of TD2LU:TD10LU and TD2MG:TD10MG tran-script ratios (table 4) on each of the two resources producesno significant R � G (Resources by Genes) interactionterm. This supports our contention that transcription re-sponds similarly to the change in resource in both strains.Significance of the Genes term indicates transcription atsome genes differs between the strains. Twenty genes withmodest changes in transcription were identified using a 1%false discovery rate (table 5). Again, there are no obviousreasons why any should be differentially expressed on thetwo resources. Other genes cotranscribed with the fourgenes, flgM, flgF, rpsD, and metE, show no evidence ofchanged expression. We suspect these changes in transcript

levels are spurious. Data normalization eliminates any dif-ference in the mean TD2:TD10 transcript ratios between thestrains. We conclude that transcription profiles of the twoancestral strains are very similar—if they differ at all, theydo so in minor ways.

Protein Expression in Ancestral Strains TD2 and TD10

Protein expression was explored using trypsin-digested samples labeled with iTRAQ, mixed samples beingfractionated by strong cation high-pressure liquid chroma-tography and separated by reversed-phase high-pressureliquid chromatography immediately before entering anelectrospray ion trap mass spectrometer for peptide identi-fication and quantification. The sample preparation used isbiased against hydrophobic (e.g., membrane) proteins.

An example of iTRAQ 2D LC–MS/MS proteomicsdata (fig. 1) illustrates how a single peptide can be isolated,fragmented, sequenced, and expression quantified in fourstrains simultaneously. Mass spectrometry is extraordi-narily precise in determining mass charge (m/z) ratios sosequencing errors are rare. Intensities, being subject to va-garies in sample preparation, are less precisely determined.Often, as in this example, numerous peptides from the sameprotein can be identified. Averaging them improves expres-sion estimates. iTRAQ proteomics routinely identifies pep-tides from more than 450 proteins and can be used toconfirm changes in transcription rates and, where system-atic discrepancies arise, point to the possibility of transla-tional regulation of specific mRNAs.

Strains TD2 and TD10 showed very similar patterns ofprotein expression. TD2 showed significantly higher ex-pression of LacA than TD10 (log2(TD2/TD10) 5 3.45 ±0.75). No other significant differences in expression weredetected among the 740 proteins quantified. We concludethat translation profiles of the two ancestral strains are verysimilar.

Table 2ANOVAs of Variance of Lactulose:Methylgalactoside Transcript Ratios in Ancestors TD2 and TD10

Item df SS MS F P

Strains 1 0.0 0.0 0.0 1Genes 4,141 563.4 0.136 1.39 ,0.0001S � G 4,141 312.2 0.754 0.77 1Error 10,612 1,039.6 0.979Total 18,895 1,915.1

df, degrees of freedom; SS, sum of squares.

Table 3Seven Genes Differentially Expressed on Different Sugars in Ancestors TD2 and TD10

b-Number log2 LU/MG SE Gene Function

b0719 0.45 0.11 ybgD Predicted fimbrial-like adhesinb0879 �0.51 0.11 macB Subunit of MacAB–TolC macrolide effluxb1639 0.66 0.16 mliC Inhibitor of c-type lysozymeb2943 0.52 0.11 galP Galactose transporterb3287 0.36 0.08 def Peptide deformylaseb3306 0.66 0.16 rpsH 30S ribosomal subunit protein S8b3508 0.48 0.11 yhiD Predicted Mgþþ transport ATPase

SE, standard error.


Dow


ic.oup.com/m


ber 2021

Transcription in Evolved Strains

We analyzed transcript levels in seven pairs of evolvedstrains that had evolved together: one pair adapted to 100%lactulose, one pair adapted to 100% methylgalactoside, andfive pairs adapted to a 72:28 lactulose:methylgalactosidemixture. Unlike the other five strains, the first two pairscould have a common ancestor within the experiment. Eachevolved strain was grown in a chemostat on a pure resource,RNA isolated, reverse transcribed, and competitive hybrid-ization used to assess any changes in transcript levels be-tween the ancestor and the descendent. ANOVAs ofevolved strain:ancestor transcript Log2 ratios are stronglysignificant with respect to differences in gene expression(many are greater than 4-fold). The frequency distributionof 906 effects determined to be significant (a 5 0.01) byANOVAs for the 14 evolved strains analyzed (fig. 2) showsthat transcript levels of certain genes are routinely changedduring evolution. In contrast, R � G interactions are rareand more nearly binomially distributed. One possibilityis that R�G interactions are erratic in their evolution. Moreprobably they are artifacts of sophisticated experimentalprocedures. Most are weakly significant (the overwhelmingmajority of which are less than 1.5-fold) and the one notableexception described below is indeed an artifact.

Given the huge number of comparisons made, keepinga significance level at a 5 0.01 is likely to produce a con-siderable number of unique false positives. We thereforeused a 1% false discovery rate (Benjamini and Hochberg

1995) to identify significant changes in expression in indi-vidual strains. By this criterion, 661 transcript levels (ratherthan 906) of 4,141 genes were identified as changed usinga 1% false discovery rate.

Each strain had a unique pattern of transcriptionalchange. However, an overall impression of the pattern oftranscriptional evolution can be obtained by summingthe number of significant increases and decreases in tran-scription for each gene (fig. 3). Increased transcript levelsare common at lac, mgl, and galP. Increased transcription atthe maltose regulon (malEKM, malP, and lamB) is also ob-served in some strains. Lower transcript levels are evident atgat. Lower transcript levels are routinely observed at fliE-R,fliC-T, and other genes involved with motility (flgK-M, trg,che, aer, yhjH, and tsr). However, some genes involvedwith motility often show increased levels of expression(flgA-J, flhA, and fliAZY). Changes in gene transcriptionof outer membrane proteins are common, with increasesat yraJ and fec and reductions at ompA, ompC, fadL,and fimA.

Protein Expression in Evolved Strains

With between 450 and 600 proteins quantified perstrain, coverage of protein expression is not as extensiveas that for transcription. Changes in protein expression ra-tios (evolvant:ancestor) correlate with changes in transcriptratios (fig. 4). A linear regression, with different slopes for

Table 4ANOVA of TD2:TD10 Transcript Ratios on Lactulose and Methylgalactoside

Item df SS MS F P

Resources 1 0.0 0.0 0.0 1Genes 4,141 484.1 0.117 1.62 ,0.0001R � G 4,141 309.5 0.075 1.04 0.0763Error 29,508 2,133.0 0.072Total 37,791 2,926.6

Table 5Twenty Genes Differentially Expressed in Ancestors TD2 and TD10 on Lactulose and Methylgalactoside

b-Number log2 LU/MG SE Gene Function

b0308 0.29 0.07 ykgG Predicted transporterb0473 0.33 0.07 htpG HSP90 chaperone subunitb0532 0.46 0.10 sfmD Predicted outer membrane proteinb1023 �0.42. 0.10 pgaB Predicted esteraseb1071 �0.32. 0.07 flgM Anti-sigma factor for fliAb1077 �0.31. 0.07 flgF Flagellar proteinb1241 0.43 0.10 adhE Alcohol dehydrogenaseb1272 0.20 0.05 sohB Predicted peptidaseb1348 0.45 0.10 lar Prophage geneb1972 0.38 0.10 yedZ Inner membrane proteinb1973 0.63 0.10 yedY Reductaseb2742 0.41 0.10 nlpD Predicted outer membrane lipoproteinb2761 �0.19. 0.05 ygcB Predicted proteinb3296 0.40 0.10 rpsD 30S ribosomal subunit protein S4b3591 0.48 0.10 secG Protein secretion complexb3829 0.46 0.10 metE Methionine biosynthesisb3890 �0.44. 0.10 yiiF Predicted proteinb4104 0.32 0.07 phnE Organophosphate ester transportb4286 �0.38. 0.10 b4286 Predicted proteinb4347 �0.38. 0.10 symE Predicted toxin

2666 Zhong et al.

Dow


ic.oup.com/m


ber 2021

different evolved strains, yields an r25 0.32 (r2 range from0.01 to 0.49 for individual strains). By definition, r25 0 forregressions of an ancestor on itself (in the absence ofchanges all ratios are unity and all deviations are experi-mental errors). Consequently, the correlations depend onthe number and magnitude of evolved changes in expres-sion that have arisen. Removing ratios not significantly dif-ferent from unity for both RNA and protein improves the fitwith an r2 5 0.57 (r2 range from 0.04 to 0.68 for individualstrains). One expects an r2 5 1 if transcription and trans-

lation are tightly coupled. Some of the scatter is undoubt-edly caused by experimental error. However, some changesdo not appear coupled. In particular, a number of highlyexpressed proteins show no evidence of similar changesin RNA transcript levels. Even when the analysis is re-stricted to significant changes in both protein expressionand transcription levels, 24% of observations involve a sig-nificant increase in one and a significant decrease in theother (table 6). These results suggest that a significantamount of posttranscriptional regulation occurs in E. coli.

FIG. 1.—iTRAQ data for a b-galactosidase peptide from ancestors TD10 and TD10R and the evolved specialists DD2266 (methylgalactoside) andDD2267R (lactulose). (A) The precursor ion spectrum of the peptide DWENPGVTQLNR—intensity plotted against m/z ratio. The three peaks representdifferences in stable isotope composition. The mass of the dominant peak differs from expected by 1 in 105. (B) The MS/MS (tandem massspectrometry) spectrum of the DWENPGVTQLNR peptide showing the b and y ions produced by random fragmentation of its peptide bonds. Peptidesequences are determined from the differences in the m/z (mass/charge) ratios of sequential ions. Although many b and y ions (red and green) are notobserved because the DWENPGVTQLNR is only partially fragmented, a sufficient number (blue) match the expected to allow unambiguousidentification and quantification. (C) The reporter ion spectrum reveals that only the lactulose specialist (DD2267R) has increased expression: 114.1TD10 (ratio 1:1), 115.1 TD10R (ratio 0.92:1), 116.1 DD2266 (ratio 0.81:1), 117.1 DD2267R (ratio 3.54:1). (D) Peptide coverage of b-galactosidase.Identification confidence: .95%, green; .50%, yellow; ,50%, red; no match, gray.


Dow


ic.oup.com/m


ber 2021

Notable Changes in Expression

Below we describe changes in expression, either intranscription or translation, that either occur repeatedlyacross replicate chemostat experiments or unique eventsthat are supported by additional experimental evidence.

Increases in lac Expression

Though they vary greatly in size and number of re-peats, every IS-generated lac duplication spans a small re-gion extending from prpC through the entire operon intolacI (Zhong et al. 2004). Every lac duplication shows in-creased transcription (fig. 5). Adjacent genes codAB (cyto-sine salvage pathway), cynX (putative cyanate transporter),and mhpR (transcriptional regulator of the 3-hydroxyphe-nylpropionate degradation pathway) routinely produce in-creased transcript levels when duplicated, though whetheror not these affect fitness is not known. Not all duplicatedgenes show significant changes in transcription, however(fig. 5). iTRAQ proteomics, with less coverage than micro-arrays, nevertheless confirms increased protein expressionof CodA, LacA, and LacZ and, in some lac-duplicatedstrains, of PrpC and MhpR (table 7).

Strong R � G interactions were detected at lac. Tran-script levels were consistently higher in cells harvestedfrom lactulose-limited chemostats than in cells of the samestrain harvested from methylgalactoside-limited chemo-stats. Duplications at lac have never been observed duringadaptation to methylgalactoside (Dykhuizen and Dean2004). Not only are tandem duplications highly unstable(Bergthorsson et al. 2007), but also overexpression oflac proteins is strongly deleterious during starvation in che-mostats (Stoebel et al. 2008). We consider the lower tran-script levels seen in strains grown on methylgalactoside anartifact generated by selection favoring newly arisen line-ages with contracted lac duplications.

Increases in fruBKA Expression

iTRAQ protein expression data shows many evolvedstrains have increased protein expression of FruA, FruB,and FruK, albeit without evidence of increased transcriptionat fruBKA (table 8). However, there is a real concern that the

FIG. 2.—Plot showing the frequency distribution of effects de-termined to be significant (a 5 0.01) by ANOVAs for each of 16 strainsanalyzed. The line denotes the frequencies expected from a binomialdistribution: (a þ (1 � a))n, with significance level a 5 0.01 and n 5 14.A large excess of Gene effects (dots) indicates that transcript levels ofcertain genes routinely change during evolution. The R � G interaction(squares) effects are rare and more nearly binomially distributed. EitherR � G interaction effects are erratic in their evolution or they are artifactsof sophisticated experimental procedures. A 1% false discovery rateyields a similar plot in which the number of strains with a uniquelysignificant effect is reduced.

FIG. 3.—Repeatability of experimental evolution across the Escherichia coli chromosome for genes with at least three significant changes. For eachgene, the number of significant increases is summed and the number of significant decreases subtracted. Genes within the dashed lines have only threeor four significant changes. Genes within boxes are genes of a single operon. Red, sugar transport and metabolism; blue, motility; brown, cell wall;black, known functions; and white circles, unknown functions.

2668 Zhong et al.

Dow


ic.oup.com/m


ber 2021

increased protein expression of FruA, FruB, and FruK seenin the evolved strains grown on the mixed sugars might bean artifact of having to grow the ancestors on lactose wherethere is no fructose (a product of lactulose hydrolysis) avail-able to induce transcription at fruBKA.

Evolved strains grown on methylgalactoside have thesame fruBKA transcript levels as the ancestors grown onmethylgalactoside and evolved strains grown on lactulosehave the same fruBAK transcript levels as the ancestorsgrown on lactulose (table 8). Crucially, ancestors grownon methylgalactoside have the same fruBAK transcript lev-els as ancestors grown on lactulose (table 9). Therefore,transcript levels at fruBAK are unchanged in all strainsin all environments. The increased protein expression ofFruA, FruB, and FruK seen in the evolved strains occurswithout change in fruBAK transcript levels, althoughwhether this is a physiological or an evolved response isyet to be determined.

Fructose, liberated during the hydrolysis of lactuloseby b-galactosidase, might leach from cells only to be re-couped by the PTS transport system’s fructose-specificcomponents FruA and FruB. The fructose, now phosphor-ylated and less membrane permeable, is converted by FruK(1-phosphofructokinase) into fructose 1,6 bisphosphate,which enters central metabolism. Increased protein expres-

sion of FruA, FruB, and FruK is sometimes accompaniedby increased protein expression of PTS transport systemcomponents (table 7) including Hpr and PtsI and PtsG (glu-cose specific) and ManX (hexose, including fructose, spe-cific). This supports the idea that hexoses are beingrecouped. Although sometimes confirmed by increasedtranscript levels at manX, the result is not confirmed by in-creased cotranscription of manY and manZ. However, thesource of phosphoenolpyruvate used by the PTS system,the ppsA-encoded phosphoenolpyruvate synthetase, is gen-erally less abundant than in the ancestral strains with noapparent change in rates of transcription. Thus, the dataare consistent with cross-feeding, although the extent to

FIG. 4.—The correlation between changes in protein expression and changes in transcript levels for 14 evolved strains. Correlated changesinvolved with resource use (red circles) include duplications at lac and lowered constitutive expression at, or deletion of, gat. Correlated changesinvolved with motility (blue circles) are mostly deletions involving the fli and lowered expression at che. Other correlated changes (black circles)include changes at modA and pps and unique events. Increased protein expression is not always associated with changes in transcript levels. Increases inprotein abundances for fructose metabolism (fru, red dots) and motility (flgA, flgM, and fliA, blue dots) repeatedly occur without change in transcriptlevels. The other proteins that increase uniquely in one culture are shown as black dots.

Table 6Contingency Table of Significant Changes in TranscriptLevels and Protein Abundance of Evolved Strains Relative toAncestors

Protein

– þRNA– 148 36þ 35 80


Dow


ic.oup.com/m


ber 2021

which the physiological changes are attributable to an evo-lutionary adaptive response is yet to be determined.

Increases in mgl Expression

IS inserts in galS disrupt the repressor of the mgl op-eron that encodes a glucose–galactose transport systemwith a serendipitous affinity for methylgalactoside. Allstrains with IS inserts in galS have greatly increased mgltranscript levels. Four strains, DD2253, DD2261,DD2262, and DD2304 do not have IS inserts in galSyet also display increased mgl transcript levels. Sequencingreveals each galS allele carries either a nonsense mutationor a missense mutation. Evidently, increases in expressionat mgl are adaptations to growth on methylgalactosidewhatever their provenance. iTRAQ proteomics confirms in-creased protein expression at mgl.

Increases in gal Expression

Increased transcription at galP is common and sug-gests that galactose, liberated by the intracellular actionof b-galactosidase, may leach from cells only to be re-couped by the high affinity Hþ-coupled galactose symport-er. iTRAQ proteomics failed to identify GalP peptides.Transcription at galETKM is significantly higher in strainsDD2261 and DD2262 and iTRAQ proteomics results, al-though not significant, confirm the trend. StrainsDD2558, DD2253, and DD2304 show no evidence of in-creased transcription yet iTRAQ proteomics consistentlyshows increased Gal protein expression.

Changes in mal Expression

Increased transcription at the maltose regulon (mal-EKM, malP, and lamB) is found in strains DD2557 and

DD2261 but iTRAQ proteomics finds no evidence ofchanged protein levels. iTRAQ proteomics reveals reducedprotein expression in many strains (DD2459, DD2460,DD2268, DD2269, DD2255, and DD2302) but withoutany evident reductions in transcript levels. Only inDD2267 are reduced transcript levels matched by reductionin protein expression.

Reductions in gat Expression

Wild-type E. coli K12 has an IS3 inserted in gatR thatdisrupts the repressor of the galactitol operon. Manyevolved strains display lowered transcript levels at gat.In strains DD2459 and DD2460, the IS3 transposed intoyegW to delete the entire gat operon (Zhong et al. 2004).Other evolved strains with lower gat transcript levels retainthe operon (confirmed by PCR). iTRAQ proteomics con-firms that protein expression at gat is reduced in thesestrains.

Changes in Motility Gene Expression

Transcription of fliE-R, fliC-T, and many other motilitygenes (flgK-M,ycgR, trg,che,aer,yhjH, and tsr) is reduced inmany strains. In all but one case, an IS1-d inserted at yedX intheancestorsTD2andTD10transposes into thefliE-Roperon(Zhong et al. 2004). Crucially, the deletions either remove ordisrupt fliR, a component of the flagellar export apparatusneeded, among other things, to export the flgM encodedanti-r28 factor. Transcription from class III flagellar pro-moters requires thefliA-encodedr28.Unable to export FlgM,r28-dependent transcription of the ‘‘late’’ flagellar genes issuppressed. Deletion or reduced transcription of fliT, an in-hibitor of transcription from class II promotors, probably ac-counts for the increased transcription of the ‘‘middle’’flagellar genes in theflg,flh, andfliAZY operons. iTRAQpro-teomics confirms that protein expression is reduced for FliG,

FIG. 5.—The effect of gene dosage on transcription. Duplications (horizontal lines with strain numbers) center on the lac operon. Increasedtranscription at linked genes is partly attributable to gene dosage (note the decline as one moves away from lac) and partly to other causes (thecorrelation is not perfect), including experimental error.

2670 Zhong et al.

Dow


ic.oup.com/m


ber 2021

FliC, CheZ, CheY, Tar, CheW, CheA, Tsr, Aer, Trg, andYcgR and increased for FlgH, FlgM, and FliA.

These deletions necessarily remove dsrB (unknownfunction), dcm, vsr, rcsA, and dsrA, which lie between yedXand fliR. Low levels of expression make changes in tran-scription and protein expression difficult to identify (occa-sionally dsrB and dcm display significantly reduced levels

of transcription). Dcm is a DNA cytosine methylase thatmethylates the second C in 5#-CCWGG sequences (Lieband Bhagwat 1996). Vsr is the very short patch repair mis-match endonuclease that nicks DNA in 5#-CTWGG se-quences following deamination of the 5-methylcytosineto thymine in the second position. Deleting both geneseliminates both mutator and repair system and is anticipated

Table 7Changes in Transcription and Translation

DD Strain 2459 2460 2557 2558 2268 2269 2253 2255 2302 2304 2261 2262 2266 2267

Ancestor TD2 TD2 TD2 TD2 TD10 TD10 TD2 TD10 TD10 TD2 TD10 TD2 TD10 TD10Specialty LU LU MG MG Both — MG LU LU MG LU MG MG LUMedium LU MG Mix Mix Mix Mix Mix

Specialist Resource UsagelacZYA

DNA dup dup — — dup — — dup — — dup — — dupRNA [ [ — — [ — — [ [ — [ — — [Protein [ [ — — [ — — [ [ — [ — — [

mglBACgalS — — — IS IS — mis — — mis mis mis IS —RNA — — — [ [ — [ — — [ [ [ [ —Protein — — — [ [ — [ — — [ [ [ [ —

Cross-Feeding or RecoupingfruBKA

DNA — — — — — — — — — — — — — —RNA — — — — — — — — — — — — — —Protein [ [ — — [ [ [ [ [ [ [ — [ [

galDNA — — — — — — — — — — — — — —RNA — — — — — — — — — — — — — —Protein — — — [ — — [ — — [ — — — —

manXYZDNA — — — — — — — — — — — — — —RNA — — Y — — — [ — — Y [ — — —Protein [ [ Y — — [ [ [ — — — [ — —

aceBKADNA — — — — — — — — — — — — — —RNA — — [ — — — — — — — — — — —Protein Y Y [ — — — Y — Y Y — Y Y Y

ChemostatmalGFE/malKL(lamB)M

DNA — — — — — — — — — — — — — —RNA — — — — — — — — [ — [ — — YProtein Y Y — Y Y Y — Y Y — — — Y Y

gatDNA — D — — — — — — — — — — — —RNA Y Y Y — — — — Y Y Y Y Y — YProtein Y Y Y — — — — Y Y Y Y Y — Y

cheZYBRtar, cheWA,motBA, tsrDNA — — — — — — — — — — — — — —RNA Y Y — — Y Y — Y Y Y Y — Y YProtein Y Y — — Y Y — Y Y Y Y — Y Y

fliF-K, fliL-RDNA D D — — D — — D D D D — — —RNA Y Y — — Y Y — Y Y Y Y — — —Protein Y Y — — Y Y — Y Y Y Y — — —

flgB-LDNA — — — — — — — — — — — — — —RNA [ — [ — [ Y [ [ [ [ [ — — [Protein [ — — — [ Y — [ [ [ [ — — [

fliZYA/flgMDNA — — — — — — — — — — — — — —RNA [ [ — — — — — — [ [ [ — — [Protein [ [ — — [ Y — [ [ [ [ — — [

LU lactulose, MG methylgalactoside, dup duplication, D deletion, and mis missense mutation, [ increased expression, Y decreased expression.


Dow


ic.oup.com/m


ber 2021

to produce a modest reduction in the mutation rate (Lieb1991). RcsA is a transcriptional regulator of flagella, cap-sular polysaccharide, and curli synthesis (Vianney et al.2005). However, we find no evidence that loss of RcsAhas any impact on transcription in the capsular polysaccha-ride (wza, wzb, wzc, and wcaAB) and curli (csg) operons.dsrA is a small antisense RNA that stimulates translation ofthe alternative sigma factor RpoS (Majdalani et al. 2005).

Changes in Expression of Outer Membrane Proteins

We find significant changes in the transcription of out-er membrane proteins; increases at yraJ and fec and reduc-tions at ompA, ompC, fadL, and fimA. iTRAQ proteomicsfailed to detect some hydrophobic proteins.

Changes in the Expression of the csg Operon

Strain DD2269 was observed to stick to chemostatglass walls. There was a significant increase in transcriptionof the csg operon in strain DD2269. Proteomic anaylsesfailed to identify csg-encoded proteins as upregulated.However, protein sample preparations bias against hydro-phobic proteins. Instead, the hypothesis that expression ofcurli adhesins increased was confirmed by staining cellswith congo red (Hammar et al. 1995).

Other Changes in Transcription and Translation

A number of genes show significantly increased tran-scription in some strains and significantly reduced tran-scription in others. For example, cspD shows increasedtranscription in three strains and decreased transcriptionin three others. Other changes are difficult to rationalizewithout extensive biochemical and physiological investiga-tions. Reduced transcription of two tricarboxylic acid cycle

enzymes (sdh and suc) and increased transcription of thedicarboxylate transporter (dctA) are just a few of many ex-amples. Hierarchical clustering (Eisen et al. 1998) failed toresolve these transcriptional changes into known regulons(fig. 6). Occasional increases and decreases in the transcrip-tion of individual genes may reflect experimental noise,particularly when other cotranscribed genes evince nochanges.

iTRAQ proteomics confirms changes in transcriptionat many genes, for example, with reduced expression of themolybdate ABC-type transporter (ModA) and increased ex-pression of the ATP-driven copper transporter (CopA). Inother instances, iTRAQ proteomics fails to confirm thechanges anticipated from microarray studies. For example,lower transcript levels at melA appear as increases in proteinexpression of the encoded a-galactosidase. As with FruA,FruB, and FruK, iTRAQ proteomics sometimes detectschanges in protein expression where no changes in tran-script levels are apparent.

Discussion

As FruBKA dramatically demonstrates, reproduciblechanges in protein expression can be produced withoutchanges in transcript levels. There is indirect evidence inSalmonella that fruBKA protein expression is regulatedat the level of translation (Sittka et al. 2008). Deletion ofhfq, which encodes a global posttranslational small RNAdependent regulator, results in higher expression of fruBamong many other proteins. Hfq is not implicated in otherdiscrepancies apparent in table 7. Regardless of the mech-anism, our parallel data sets show that changes in transcrip-tion and translation need not move in parallel.

A number of experimental evolution experiments andpopulation genetic surveys have made use of genomewidetranscript profiling (e.g., Gilad et al. 2006; Agudelo-Romeroet al. 2008; Genissel et al. 2008; Le Gac et al. 2008; St-Cyr

Table 8ANOVA of Transcript Ratios at fruBKA in Evolved Strains

Item df SS MS F P

Strain 9 0.879 0.098 1.01 0.43Resource 1 0.019 0.019 0.19 0.66Gene 2 0.106 0.053 0.55 0.58S � R 9 1.279 0.142 1.48 0.16S � G 18 1.869 0.104 1.08 0.38R � G 2 0.494 0.247 2.57 0.08S � R � G 18 0.790 0.044 0.45 0.97Error 123 11.839 0.096Total 182 17.361

Table 9ANOVA of Lactulose:Methylgalactoside Transcript Ratios at fruBKA in Ancestors TD2 and TD10

Item df SS MS F P

Strains 1 0.071 0.071 0.51 0.50Genes 2 0.192 0.096 0.68 0.54S � G 2 0.176 0.088 0.62 0.57Error 6 0.845 0.141Total 11 1.284

2672 Zhong et al.

Dow


ic.oup.com/m


ber 2021

et al. 2008; Vijayendran et al. 2008). Patterns of transcrip-tion differ among individuals, populations, ancestors, de-scendents, and closely related species. Correlations withother patterns of variation can be shown to be statisticallysignificant. But as to why they differ is rarely exploredfurther and hardly ever subjected to definitive testing.

Our approach has been to devise a structured experi-ment to explore a ubiquitous bipartite phenomenon—theevolution of specialists and generalists. Specialists can arisethrough any of three mechanisms (Elena and Lenski 2003):passively accumulated neutral mutations that prove deleteri-ous in another environment (mutation accumulation), bene-ficial mutations that prove selectively neutral in anotherenvironment (independent specialization) and, lastly, bene-ficial mutations that are deleterious in another environment(antagonistic pleiotropy). Populations growing on pure sug-ars can specialize by any or all of these three mechanisms.Those growing on mixed sugars cannot specialize by muta-tion accumulation because any mutation selectively neutralforonesugaranddeleteriousfor thesecondmustbepurgedbyselection. Independent specialization is expected to producegeneralists. Antagonistic pleiotropy forces specialization.

Specialists and Generalists

Duplicating lac is an adaptation to growth on lactulose(Zhong et al. 2004). Constitutive mgl expression (galS�) is

an adaptation to growth on methylgalactoside (Zhong et al.2004). In both cases, benefits are predicted to derive fromthe increased transport of limiting nutrients, although in-creased rates of hydrolysis by b-galactosidase will makea small contribution in the case of lac duplications (Dean1989, 1995). Evolution in a limiting mixture of 72:28 lac-tulose:methylgalactoside usually produces a balanced poly-morphism of two specialists: one ecotype carries a lacduplication and is fittest on lactulose, the other ecotypeis mgl constitutive and is fittest on methylgalactoside. Onlyone of 26 strains isolated from 13 long-term evolution ex-periments on the mixed sugars is a galS�,lacdup generalist(Dykhuizen and Dean 2004). Although these observationssuggest specialization through antagonistic pleiotropy, thepresence of just one generalist shows that antagonistic plei-otropy is not ubiquitous.

Invading a New Niche

How is the galSþ,lacþ strain DD2269 maintained inthe presence of the galS�,lacdup generalist DD2268?DD2269 manages to persist at low frequency (ca. 1%) eventhough transcription and protein expression at mgl and lacare no different from wild type. This is achieved by in-creased expression of csg, the curli operons, which produceand export a fibrous surface protein (the presence of whichwas confirmed by staining with congo red, Hammar et al.

FIG. 6.—Hierarchical clustering (Eisen et al. 1998) identifies changes at the mgl, lac, gat, and motility operons but fails to resolve othertranscriptional changes into known regulons.


Dow


ic.oup.com/m


ber 2021

1995) that enables cells to stick to the glass wall of the che-mostat, thereby preventing complete washout. Perhaps inresponse to increased curli expression transcription of clpB,which encodes a protein that resolubilizes aggregated pro-teins, is also increased. Similar to the adaptive radiation ofPseudomonas in static microcosms (Rainey and Travisano1998), the spatial heterogeneity provided by a chemostatwall represents a new niche to be exploited.

Genotype versus Phenotype

Another galS�,lacdup strain, DD2261, is not a general-ist but a lactulose specialist; DD2261 is fitter on lactuloseand less fit on methylgalactoside than its galS� competitorDD2262 (Dykhuizen and Dean 2004). The suggestion thatlacdup might reduce galS� fitness on methylgalactosidepoints to a role for ‘‘antagonistic pleiotropy’’ in the evolu-tion of specialists and generalists. Certainly, antagonisticpleiotropy between lacdup and galS� alleles providesa handy explanation for why specialists are commonplace,but it does not explain why generalists exist. Perhapsrare—and as yet unidentified—background mutationsnegate the proposed antagonistic pleiotropy.

Cross-Feeding or Recouping?

Cross-feeding among evolved strains can also main-tain diversity (Rosenzweig et al. 1994). Increased proteinexpression of fruBKA is evident following adaptation topure lactulose (hydrolysis of which releases galactoseand fructose) but not methylgalactoside (hydrolysis ofwhich releases galactose and methanol). Similar increasesin protein expression are seen in strains adapted to mixedsugars, including methylgalactoside specialist. Whether ex-tensive cross-feeding between competitors occurs is by nomeans certain, for the majority of fructose leached into theperiplasm of either strain might be recouped before it everhas a chance to diffuse into the environment. Moreover, thepotential impact of cross-feeding in maintaining diversitymight possibly be negated by the fact that all strains in-crease expression of fruBKA in the presence of lactulose.

Increased protein expression, but not increased tran-scription, is sometimes evident at galETKM following ad-aptation to methylgalactoside. Similar increases seenduring adaptation to mixed sugars are found in severalmethylgalactoside specialists but not in the six lactulosespecialists. It is not obvious why increases in gal proteinexpression should be so restricted—galactose is releasedupon the hydrolysis of either sugar. Increased transcriptionat galP, the low affinity galactose transporter, is evident inmany strains but iTRAQ proteomics protocol did not detectthis hydrophobic membrane protein. Again, whether or notextensive cross-feeding between competitors occurs is notknown.

Strains DD2557 andDD2558 were isolated from a cul-ture adapted to pure methylgalactoside. DD2558 was, asexpected, galS� with the expected strong increases in tran-scription and translation at the mgl operon. By contrast,DD2557 is galSþ and showed no evidence of increasesin transcription and translation atmgl, whereas transcription

and translation at gat is greatly reduced. How can DD2557persist in the face of a superior methylgalactoside special-ist? Increases in transcription and translation of crp increasetranscription and translation of a number of cataboliterepressed genes. In particular, increased transcription andtranslation at aceBAK, which encodes enzymes of thegyloxylate bypass, suggests DD2557 may utilize the acetatefermented by DD2558 as a major source of carbon andenergy (see Fischer and Sauer 2003 for an alternativepossibility).

Three Long-Term Evolution Experiments

A detailed series of experiments dissecting adaptationby E. coli strain MC4100 to glucose-limited chemostats(i.e, Death and Ferenci 1994, reviewed by Ferenci 2008)reveal that beneficial increased rates of glucose transportare caused by mutations that increase activities at mgl (ahigh affinity glucose/galactose transporter), ptsG (the glu-cose-specific PTS permease), and increase expression of thenonspecific OmpC and OmpF porins and the glucose-/malt-dextran-specific LamB porin of the outer cell wall.

In our experiments, adaptation to methylgalactosidealso favors high transport rates at mgl. We also find in-creased expression at ptsG. Unlike Ferenci, we cannot ac-count for this directly because our strains were not grownon limiting glucose. Instead, we suggest selection favorsincreased expression of the cotranscribed ptsH and ptsIcomponents common to all PTS systems. These might in-crease transport of both galactose and fructose. Indeed, in-creased translation at ptsH and ptsI is seen only in strainsdisplaying increased translation at gal and/or fru, whereasthe converse is not necessarily true (table 7). Nor do Ferenciand coworkers (Ferenci 2008) report the increased expres-sion we detect in the nonspecific hexose PTS transporterManX. The extent to which these are involved with galac-tose and fructose cross-feeding is yet to be ascertained. Itcould very well be minor, with increased expression beingprimarily associated with recouping leached monosacchar-ides. Transcription at ompC is increased in three strains, re-duced in seven more, and unchanged in four others. Nochanges in OmpC protein levels were detected. No changesin transcription were observed at ompF although one strainshows increased protein expression and three others re-duced expression. Lastly, increases in transcription atmal are not matched by increases in translation—moststrains show lower mal protein expression (table 7). Thissupports the contention that our strains are unlikely to betransporting glucose.

Strains, such as MC4100, that express high levels ofthe alternative sigma factor RpoS benefit greatly from itsloss during very slow growth in chemostats (Ferenci2003, 2005). All yedX-fliR deletions necessarily removedsrA, a small antisense RNA that stimulates translationof rpoS (Soper and Woodson 2008). Loss of dsrA maytherefore confer an advantage in addition to reduced motil-ity gene expression. However, RpoS levels vary evenamong E. coli K12 strains (e.g., MG1655 has lower levels,Spira et al. 2008) and so it is not altogether surprising thatwe should find attenuated rpoS transcription in only threestrains and no changes in the others.

2674 Zhong et al.

Dow


ic.oup.com/m


ber 2021

Rosenzweig et al. (1994) discovered that adaptation byE. coli strain JA122 to a glucose-limited chemostat pro-duced a community of three specialists, one fermentingthe primary resource to acetate and glycerol, with eachof these products consumed by one of two other specialists.The acetate specialist had increased acetyl-CoA synthetaseactivity (Acs) and the glycerol specialist had increasedglycerol kinase activity (GlpK). Treves et al. (1998) showedthat similar acetate specialists appeared in 6 of 12 subse-quent long-term evolution experiments. Studies of adapta-tion by E. coli strain MC4100 to a glucose-limitedchemostat provide no evidence of cross-feeding (reviewedby Ferenci 2008). We find weak evidence for it at best withthe DD320 background. Of the 14 strains studied, onlyDD2266 shows increased transcription at acs—yet withoutevidence of increased translation. DD2557 overexpressesaceBAK (needed to assimilate acetate and possibly attribut-able to overexpression of Crp) instead of acs. However, thismight instead reflect glucose catabolism via the novelphosphoenolpyruvate–glyoxylate cycle (Fischer and Sauer2003). If true, the polymorphism is transient (pure scramblecompetition for a single limiting resource—glucose—is in-capable of maintaining both strains). Our failure to find ev-idence of glycerol cross-feeding is likely attributable to anIS30 insertion in glpF, the glycerol channel (Zhong et al.2004).

How Idiosyncratic Is Evolution?

Improved uptake of a limiting resource is a commonadaptive response to starvation, one frequently achievedthrough increased protein expression of specific transport-ers (Horiuchi et al. 1962, 1963; Dean 1989; Sonti and Roth1989; Ferea et al. 1999; Zhong et al. 2004; Ferenci 2008).Another, attaching to the side of the chemostat, establishesa subpopulation immune to washout (so-called wallgrowth). Loss of motility is possibly the most commonadaptive phenotype during routine laboratory culture, re-peatedly appearing in otherwise identical E. coli K12strains (Macnab 1992).

Other changes are strain specific. Transcription at gatis routinely downregulated in TD2 and TD10 because it justhappens to be constitutive in these strains. MC4100 usuallyreduces its high endogenous RpoS expression as it adapts tostarvation. Similar changes were detected in only 3 of 14cases with TD2 and TD10. During starvation on glucose,JA122 often evolves into a community, whereasMC4100 never does. Other changes are environment spe-cific. Lac duplications are specific to limitation by certaingalactosides but not methylgalactoside. Increased proteinexpression at fru is dependent on the presence of fructose.The benefits conferred by rpoS mutants are higher at lowgrowth rates.

Perspective

Genomics and proteomics produce so much data thatone can easily lose sight of the trees, let alone the forest andend up concentrating only on a large pile of leaves. In thissection, we will try to provide a framework for these data,

trying to look at the ‘‘forest’’ again, by assessing the resultsin terms of the major question that motivated this work. Wewant to understand the causes of natural selection; how en-vironment and genetic variation produce natural selection.The main experimental model used is to grow E. coli ina chemostat and investigate genetic changes promoted bynatural selection. Enough work has now been done thatwe can suggest certain generalizations.

1. There is selection for constitutive expression inregulated systems involved with the uptake anddissemination into metabolism of the limiting nutrient,up to the point that the substrates cannot escape fromthe cell. We have selected for constitutive mutations inall sugars tested (10), which includes maltose, rham-nose, and fructose, used as the limiting resource inchemostats (unpublished data). The general principle isthat there will be selection for any genetic change thatincreases the gradient of the concentration of thelimiting nutrient between the outside and inside of thecell.

2. Metabolic systems with constitutive mutations not usedby cells in selective environments, such as gat in ourstrains, are eliminated. This principle of selectionagainst the production of unneeded proteins is quitegeneral (Stoebel et al. 2008). We predict that if ourchemostats had been run for a longer time, gat wouldbe eliminated in all cultures, just as rbs is eliminated inall long-term cultures of E. coli B (Cooper et al. 2001).The selection coefficient on a repressed operon issufficiently small that deletions that remove the operonare effectively neutral (Stoebel et al 2008). However,the rbs operon was eliminated by deletion even thoughit was regulated (Cooper et al. 2001). The basal level ofrbs proteins was evidently high because there wasa significant decrease in expression in the strains wherethe operon was deleted (Cooper et al. 2003). Pre-sumably, this was because of regulatory interactionswith a constitutive duplication of most of the rbsoperon unique to E. coli B (Lin 1996).

3. Metabolic systems induced in the environment used,but not required, are selectively eliminated or down-regulated. Two obvious examples are the flagellargenes (flagella are useless in a homogeneous, rapidlymixed liquid environment as is the chemostat) and theRpoS system. Again, unneeded proteins are selectedagainst. However, the RpoS system includes a largenumber of proteins, some of which might be advanta-geous and others detrimental, so the outcome ofselection can vary from culture to culture. Points 2and 3 are changes caused by the same mechanism,selection against unneeded proteins, but are different inthat the constitutives in point 2 are strain specific andare present either because of chance or history, whereasthe induction in point 3 will be a natural inductionspecific to the environment and found across moststrains.

4. When there aremultiple replaceable limiting resources, itis likely that some of the resources are at a concentrationwhere only part of the population selects for upregulationof the uptake and initial metabolism, because of the


Dow


ic.oup.com/m


ber 2021

trade-off between selection for uptake and the selection

against overexpression. For example, in mixed lactose–

maltose chemostats not all the cells are selected for

lactose constitutivity when 10% of the sugar is lactose,

but all are when 25% of the sugar is lactose (Dykhuizen

andDavis 1980). The low concentration resources can be

waste products from fermentation as ethanol or acetate orcan be an intermediate compound that can diffuse out ofthe cell as fructose seemingly does in the experimentreported here.

These four generalizations involve only two causes ofselection: Selection to increase the concentration gradientof the limiting nutrient across the cell membrane and selec-tion against unnecessary protein synthesis. Except forRpoS�, these changes happen regularly and quickly in mostcultures of an experiment. Thus, we can conclude that thesetwo causes explain the initial adaptation of bacteria to a che-mostat environment. Are there other causes for adaptationin these experiments? If not, then competition for resourcesis likely to be a minor part of evolution, becuase there are somany other genes performing so many other functions.However, numerous changes are found in only a minorityof replicate chemostat experiments. These are often ingenes that are not obviously related to the functions givenabove. We might expect that these represent other causes ofnatural selection. The reasoning for this statement is givenin the next paragraph.

The initial temptation is to assume these uncommonchanges are chance events where selectively neutral muta-tions hitchhike with advantageous mutations. This is un-likely, because the mutation rate of 10�10 per nucleotideper generation for E. coli is so much smaller than the inverseof the number of nucleotides in the E. coli genome (5� 106

nt). This means that only 1 in 2 � 103 cells has a new mu-tation in each generation. After 500 generations, only one offour cells will have a nucleotide change. In a chemostat af-ter 500 generations, if there is no selection, we expect only25% of cells to differ by a single nucleotide from the an-cestral cell. Thus, for most cells, there will be no neutralmutations present. However, because selection is present,all the cells will differ from the ancestoral cell by a numberof changes. This implies almost all the phenotypic changeswe see involve a genetic change that was selected. On theother hand, there are about 3� 109 cells in the chemostat sothat nearly every nucleotide change appears multiple timesin the first 10 generations. This could give rise to consider-able clonal interference.

The selection coefficients for upregulation of uptakesystems and against the production of unnecessary proteinswill be large and these mutations will be fixed initially. Af-ter these are fixed, genetic changes with lower selection co-efficients will be selected. There may be many of these withsmall, nearly equal, selection coefficients, giving consider-able clonal interference. Which one is present in any cellwill be a matter of chance and a diversity of these changesshould be found in a population. For example, some of thecells will increase metabolic efficiency, others uptake effi-ciency (Ferenci 2008; MacLean 2008). Thus, we might ex-pect a much richer array of causes at this second level ofselection.

Supplementary Material

Supplementary material is available at MolecularBiology and Evolution online (http://www.mbe.oxfordjournals.org/).

Acknowledgments

We thank Arkady Khodursky for use of his microarrayfacilities and Kyeong Jeong for conducting the hierarchicalclustering analysis. This work was supported by a PublicHealth Service Grant (GM06380) to A.M.D. and D.E.D.

Literature Cited

Agudelo-Romero P, Carbonell P, Perez-Amador MA, Elena SF.2008. Virus adaptation by manipulation of host’s geneexpression. PLoS ONE. 3:e2397.

Benjamini Y, Hochberg Y. 1995. Controlling the false discoveryrate: a practical and powerful approach to multiple testing. JRoy Statist Soc B. 57:289–300.

Bennett AF, Lenski RE. 1993. Evolutionary adaptation totemperature. II. Thermal niches of experimental lines ofEscherichia coli. Evolution. 47:1–12.

Bennett AF, Lenski RE. 2007. An experimental test ofevolutionary trade-offs during temperature adaptation. ProcNatl Acad Sci USA. 104:8649–8654.

Bergthorsson U, Andersson DI, Roth JR. 2007. Ohno’s dilemma:evolution of new genes under continuous selection. Proc NatlAcad Sci USA. 104:17004–17009.

Button DK. 1985. Kinetics of nutrient-limited transport andmicrobial growth. Microbiol Rev. 49:270–297.

Cooper TF, Rozen DE, Lenski RE. 2003. Parallel changes ingene expression after 20,000 generations of evolution in E.coli. Proc Natl Acad Sci USA. 100:1072–1077.

Cooper VS, Lenski RE. 2000. The population genetics ofecological specialization in evolving E. coli populations.Nature. 407:736–739.

Cooper VS, Schneider D, Blot M, Lenski RE. 2001. Mechanismscausing rapid and parallel losses of ribose catabolism in evolvingpopulations of Escherichia coli B. J Bacteriol. 183:2834–2841.

Corbin RW, Paliy O, Yang F, et al. (11 co-authors). 2003. Towarda protein profile of Escherichia coli: comparison to itstranscription profile. Proc Natl Acad Sci USA. 100:9232–9237.

Dean AM. 1989. Selection and neutrality in lactose operons ofEscherichia coli. Genetics. 123:441–454.

Dean AM. 1995. A molecular investigation of genotype byenvironment interactions. Genetics. 139:19–33.

Death A, Ferenci T. 1994. Between feast and famine:endogenousinducer synthesis in the adaptation of Escherichia coli togrowth with limiting carbohydrates. J Bacteriol.176:5101–5107.

Dykhuizen D, Davies M. 1980. An experimental model: bacterialspecialists and generalists competing in chemostats. Ecology.61:1213–1227.

Dykhuizen DE, Dean AM. 2004. Evolution of specialists in anexperimental microcosm. Genetics. 167:2015–2026.

Eisen MB, Spellman PT, Brown PO, Botstein D. 1998. Clusteranalysis and display of genome-wide expression patterns.Proc Natl Acad Sci USA. 95:14863–14868.

Elena SF, Lenski RE. 2003. Evolution experiments withmicroorganisms: the dynamics and genetic bases of adapta-tion. Nat Rev Genet. 4:457–469.

Ferea TL, Botstein D, Brown PO, Rosenzweig RF. 1999.Systematic changes in gene expression patterns following

2676 Zhong et al.

Dow


ic.oup.com/m


ber 2021

Supplementary material

http://www.mbe.oxfordjournals.org/

http://www.mbe.oxfordjournals.org/

adaptive evolution in yeast. Proc Natl Acad Sci USA.96:9721–9726.

Ferenci T. 2003. What is driving the acquisition of mutS andrpoS polymorphisms in Escherichia coli. Trends Microbiol.11:457–461.

Ferenci T. 2005. Maintaining a healthy SPANC balance throughregulatory and mutational adaptation. Mol. Microbiol. 57:1–8.

Ferenci T. 2008. Bacterial physiology, regulation and mutationaladaptation in a chemostat environment. Adv Microb Physiol.53:169–229.

Fischer E, Sauer U. 2003. A novel metabolic cycle catalyzesglucose oxidation and anaplerosis in hungry Escherichia coli.J Biol Chem. 278:46446–46451.

Genissel A, McIntyre LM, Wayne ML, Nuzhdin SV. 2008. Cisand trans regulatory effects contribute to natural variation intranscriptome of Drosophila melanogaster. Mol Biol Evol.25:101–110.

Gilad Y, Oshlack A, Smyth GK, Speed TP, White KP. 2006.Expression profiling in primates reveals a rapid evolution ofhuman transcription factors. Nature. 440:242–245.

Greenbaum D, Colangelo C, Williams K, Gerstein M. 2003.Comparing protein abundance and mRNA expression levelson a genomic scale. Genome Biol. 4:117.

Griffin TJ, Gygi SP, Ideker T, Rist B, Eng J, Hood L,Aebersold R. 2002. Complementary profiling of geneexpression at the transcriptome and proteome levels inSaccharomyces cerevisiae. Mol Cell Proteomics. 4:323–333.

Hammar M, Arnqvist A, Bian Z, Olsen A, Normark S. 1995.Expression of two csg operons is required for production offibronectin- and congo red-binding curli polymers inEscherichia coli K-12. Mol Microbiol. 18:661–670.

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

Horiuchi T, Tomizawa J, Novick A. 1962. Isolation andproperties of bacteria capable of high rates of b-galactosidasesynthesis. Biochem Biophys Acta. 55:152.

Khodursky AB, Bernstein JA, Peter BJ, Rhodius V,Wendisch VF, Zimmer DP. 2003. Escherichia coli spotteddouble-strand DNA microarrays: RNA extraction, labeling,hybridization, quality control, and data management. MethodsMol Biol. 224:61–78.

Khodursky AB, Peter BJ, Cozzarelli NR, Botstein D, Brown PO,Yanofsky C. 2000. DNA microarray analysis of geneexpression in response to physiological and genetic changesthat affect tryptophan metabolism in Escherichia coli. ProcNatl Acad Sci USA. 97:12170–12175.

Le Gac M, Brazas MD, Bertrand M, Tyerman JG, Spencer CC,Hancock REW, Doebeli M. 2008. Metabolic changesassociated with adaptive diversification in Escherichia coli.Genetics. 178:1049–1060.

Lieb M. 1991. Spontaneous mutation at a 5-methylcytosinehotspot is prevented by very short patch (VSP) mismatchrepair. Genetics. 128:23–27.

Lieb M, Bhagwat AS. 1996. Very short patch repair: reducing thecost of cytosine methylation. Mol. Microbiol. 20:467–473.

Lin ECC. 1996. Dissimilatory pathways for sugars, polyols, andcarboxylates. In: Neidhardt FC, Curtiss R III, Ingram JL, LinECC, Low KB, Magasanik B, Reznifoff WS, Riley M,Schaechter M, Umbarger HE, editors. Escherichia coli andSalmonella: cellular and molecular biology, 2nd ed.Washington (DC): ASM Press. p. 307–342.

Lunzer M, Natarajan A, Dykhuizen DE, Dean AM. 2002.Enzyme kinetics, substitutable resources and competition:from biochemistry to frequency-dependent selection in lac.Genetics. 162:485–499.

MacLean RC. 2008. The tragedy of the commons in microbialpopulations: insights from theoretical, comparative andexperimental studies. Heredity. 100:471–477.

MacLean RC, Bell G. 2003. Divergent evolution during anexperimental adaptive radiation. Proc Biol Sci.270:1645–1650.

Macnab RM. 1992. Genetics and biogenesis of bacterial flagella.Annu Rev Genet. 26:131–158.

Majdalani N, Vanderpool CK, Gottesman S. 2005. Bacterialsmall RNA regulators. Crit Rev Biochem Mol Biol.40:93–113.

Miller J. 1972. Experiments in molecular genetics. Cold SpringHarbor (NY): Cold Spring Harbor Laboratory Press.

Nelsestuen GL, Zhang Y, Martinez MB, Key NS, Jilma B,Verneris M, Sinaiko A, Kasthuri RS. 2005. Plasma proteinprofiling: unique and stable features of individuals. Proteo-mics. 5:4012–4024.

Notley-McRobb L, Ferenci T. 1999. The generation of multipleco-existing mal-regulatory mutations through polygenicevolution in glucose-limited populations of Escherichia coli.Env Microbiol. 1:45–52.

Novick A, Horiuchi T. 1961. Hyper-production of b-galactosidaseby Escherichia coli bacteria. Cold Spring Harbor Symp QuantBiol. 26:239–245.

Rainey PB, Travisano M. 1998. Adaptive radiation in a hetero-geneous environment. Nature. 394:69–72.

Riehle MM, Bennett AF, Long AD. 2005. Changes in geneexpression following high temperature adaptation in experi-mentally evolved populations of E. coli. Physiological andBiochemical Zoology. 78:299–315.

Riley M, Abe T, Arnaud MB, et al. (18 co-authors). 2006.Escherichia coli K-12: a cooperatively developed annotationsnapshot –2005. Nucleic Acids Res. 34:1–9.

Rosenzweig RF, Sharp RR, Treves DS, Adams J. 1994.Microbial evolution in a simple unstructured environment:genetic differentiation in Escherichia coli. Genetics.137:903–917.

Ross PL, Huang YN, Marchese JN, et al. (16 co-authors). 2004.Multiplexed protein quantitation in Saccharomyces cerevisiaeusing amine-reactive isobaric tagging reagents. Mol CellProteomics. 3:1154–1169.

Sittka A, Lucchini S, Papenfort K, Sharma CM, Rolle K,Binnewies TT, Hinton JCD, Vogel J. 2008. Deep sequencinganalysis of small noncoding RNA and mRNA targets of theglobal post-transcriptional regulator, Hfq. PLOS Genet.4:1–20.

Sonti RV, Roth JR. 1989. Role of gene duplications in theadaptation of Salmonella typhimurium to growth on limitingcarbon sources. Genetics. 123:19–28.

Soper TJ, Woodson SA. 2008. The rpoS mRNA leader recruitsHfq to facilitate annealing with DsrA sRNA. RNA.14:1907–1917.

Spira B, Hu X, Ferenci T. 2008. Strain variation in ppGppconcentration and RpoS levels in laboratory strains ofEscherichia coli K-12. Microbiology. 154:2887–2895.

Stoebel DM, Dean AM, Dykhuizen DE. 2008. The cost ofexpression of Escherichia coli lac operon proteins is inthe process, not in the products. Genetics. 178:1653–1660.

St-Cyr J, Derome N, Bernatchez L. 2008. The transcriptomics oflife-history trade-offs in whitefish species pairs (Coregonussp.). Mol Ecol. 17:1850–1870.

Treves DS, Manning S, Adams J. 1998. Repeated evolution of anacetate-crossfeeding polymorphism in long-term populationsof Escherichia coli. Mol Biol Evol. 15:789–797.

Vianney A, Jubelin G, Renault S, Dorel C, Lejeune P,Lazzaroni JC. 2005. Escherichia coli tol and rcs genes


Dow


ic.oup.com/m


ber 2021

participate in the complex network affecting curli synthesis.Microbiology. 151:2487–2497.

Vijayendran C, Barsch A, Friehs K, Niehaus K, Becker A,Flaschel E. 2008. Perceiving molecular evolution processes inEscherichia coli by comprehensive metabolite and geneexpression profiling. Genome Biol. 9:R72.

Zieske LR. 2006. A perspective on the use of iTRAQ reagenttechnology for protein complex and profiling studies. J ExpBot. 57:1501–1508.

Zhong S, Khodursky A, Dykhuizen DE, Dean AM. 2004.

Evolutionary genomics of ecological specialization. Proc Natl

Acad Sci USA. 101:11719–11724.

Jennifer Wernegreen, Associate Editor

Accepted August 3, 2009

2678 Zhong et al.

Dow


ic.oup.com/m


ber 2021

research articles transcription, translation, and the

Documents