Pyrimidine homeostasis is accomplished by directed overflow metabolism.

Cellular metabolism converts available nutrients into usable energy and biomass precursors. The process is regulated to facilitate efficient nutrient use and metabolic homeostasis. Feedback inhibition of the first committed step of a pathway by its final product is a classical means of controlling biosynthesis. In a canonical example, the first committed enzyme in the pyrimidine pathway in Escherichia coli is allosterically inhibited by cytidine triphosphate. The physiological consequences of disrupting this regulation, however, have not been previously explored. Here we identify an alternative regulatory strategy that enables precise control of pyrimidine pathway end-product levels, even in the presence of dysregulated biosynthetic flux. The mechanism involves cooperative feedback regulation of the near-terminal pathway enzyme uridine monophosphate kinase. Such feedback leads to build-up of the pathway intermediate uridine monophosphate, which is in turn degraded by a conserved phosphatase, here termed UmpH, with previously unknown physiological function. Such directed overflow metabolism allows homeostasis of uridine triphosphate and cytidine triphosphate levels at the expense of uracil excretion and slower growth during energy limitation. Disruption of the directed overflow regulatory mechanism impairs growth in pyrimidine-rich environments. Thus, pyrimidine homeostasis involves dual regulatory strategies, with classical feedback inhibition enhancing metabolic efficiency and directed overflow metabolism ensuring end-product homeostasis.

The metabolic network of E. coli consists of approximately 1,000 metabolites connected by around 2,000 enzyme-catalysed reactions[9]. Control of metabolite concentrations and fluxes occurs through the regulation of enzyme concentrations, activities and substrate occupancies. Metabolic control analysis provides a systematic framework for investigating the impact of particular enzymes on cellular metabolic activities[10-13]. Studies modulating the concentrations of enzymes suggest that control of metabolic flux is frequently distributed across multiple enzymes, with demand for end product often having a key role in controlling biosynthetic fluxes[2,14-16].

Consistent with distributed flux control, de novo pyrimidine biosynthesis has been reported to be regulated both at the first committed pathway step, catalysed by aspartate transcarbamoylase (ATCase), and the previous step, catalysed by carbamoyl phosphate synthetase (CPSase), which also feeds arginine biosynthesis. The E. coli ATCase enzyme complex, which consists of six catalytic and six regulatory subunits, is subject to feedback inhibition by the pyrimidine end products uridine triphosphate (UTP) and more strongly cytidine triphosphate (CTP), and is activated by ATP[1,5,17-19]. Its allosteric regulation provided one of the first examples of feedback inhibition[3,4]. CPSase is feedback inhibited by the pyrimidine intermediate uridine monophosphate (UMP)[20].

To explore the physiological relevance of ATCase and CPSase allostery, we created strains dysregulated for feedback control of ATCase (ΔpyrI), CPSase (carB* (carB(S948F))[21]) or both (ΔpyrI carB*) (Fig. 1a). We then analysed the metabolite concentrations in these strains by liquid chromatography–mass spectrometry (LC–MS)-based metabolomics. We anticipated that such strains would have increased levels of pyrimidine nucleotide triphosphates (NTPs)[2,11,16], the pathway’s terminal products; however, UTP and CTP levels were steady in the absence of feedback control. Instead, the only notable change we observed was markedly increased uracil levels (Fig. 1b–d). To understand the robustness of this pathway, we also conducted transcriptome analysis. Rather than compensatory downregulation of pyrimidine biosynthetic genesin the ΔpyrI carB* strain, we observed modest upregulation. In addition, we observed enhanced expression of genes involved in arginine synthesis (which also requires CPSase) and of the Rut pathway (a recently discovered uracil-degradation pathway that is induced by uracil) (Supplementary Fig. 1)[22].

Figure 1

Increased pyrimidine flux triggers overflow to uracil

a, Canonical pyrimidine regulatory schematic. Carbamoyl phosphate is an intermediate in both pyrimidine and arginine synthesis. Carbamoyl aspartate is committed to pyrimidine synthesis. Carbamoyl phosphate synthetase (carAB) is feedback inhibited by UMP. Aspartate transcarbamoylase (pyrBI) is feedback inhibited by UTP and CTP and activated by ATP. b, c, Extracted ion chromatograms showing uracil (b) and CTP (c) levels in wild-type (black) and feedback-defective (ΔpyrI carB*; red) strains. d, Metabolite fold changes relative to wild type in ΔpyrI, carB* and ΔpyrI carB* strains. Error bars denote ± standard error (n = 6). e, Metabolite fold changes at 5min after addition of orotate. Fold changes were computed relative to un-supplemented controls. Error bars denote ± standard deviation (n = 2–3). Time course data appear in Supplementary Fig. 2. f, Fraction of UMP and UTP derived from endogenous sources and from exogenously added 15N-orotate. Error bars denote ± standard deviation (n = 2). The ΔpyrI (blue) and wild-type (black) lines lie under the carB* (green) line. g, Competitive growth advantage of wild-type versus ΔpyrI carB* appears selectively under energy limitation. Competitions were performed in indicated media with lacZ− marker in wild-type and in feedback-dysregulated strain (wild-type lacZ) to control for effect of marker on growth. Calculations and experimental details are described in Methods. In brief, media were glucose-ammonia minimal media unless otherwise indicated. Alternative carbon and nitrogen sources were used in conditions K, L and O. + indicates supplements to the minimal media; +/− indicates alternating supplementation/removal of indicated nutrient every 8 h. Grey ellipses mark 95% confidence interval (n = 6–10).

To assess the homeostatic capacity of pyrimidine metabolism in response to pyrimidine intermediate addition, we switched E. coli grown on minimal media to media containing orotate or uracil. Such pyrimidine upshift was sufficient to activate feedback inhibition of pyrimidine synthesis in the wild type, as evidenced by reduced N-carbamoyl-aspartate and dihydroorotate concentrations within 5 min; this decrease did not occur in the ΔpyrI carB* strain (Fig. 1e and Supplementary Fig. 2). In addition, pyrimidine upshift led to markedly increased uracil, and in the case of orotate addition, also UMP, and these increases were observed in all genetic backgrounds (Fig. 1e). Moreover, in both the presence and absence of feedback control, there were only minor increases in UTP and CTP.

Given the end-product homeostasis in the doubly feedback-dysregulated strain, even upon pyrimidine upshift, we sought to confirm that the ΔpyrI carB* strain is indeed defective in de novo pyrimidine biosynthetic regulation. To this end, we measured the incorporation of isotopically labelled uracil or orotate into UTP and CTP. Similar to the wild-type strain, the ΔpyrI carB* strain imported the labelled intermediates and incorporated them into end products. The residual production of unlabelled end products, indicative of persistent de novo synthesis, however, was higher in the ΔpyrI carB* strain (Fig. 1f and Supplementary Fig. 2). This confirms that feedback regulation is functionally impaired.

If the feedback-regulation mechanisms facilitated superior pyrimidine homeostasis, perhaps too subtle to be detected by our LC–MS methods, we expected that the feedback-dysregulated strain would have a growth defect relative to its wild-type parent. The ΔpyrI carB* strain did not, however, exhibit impaired growth in various nutrient conditions, including rich medium, minimal media, media enriched in nucleotides or amino acids, media limited for nitrogen or phosphate, or periodic switches between these conditions (Fig. 1g). We did, however, detect a modest (~10%) growth advantage for the wild-type strain under anoxic conditions and in the presence of the uncoupler 2,4-dinitrophenol.

Slower growth only during energy limitation suggests that the feedback-dysregulated strain engages in a chronic, energy-wasting process. Given the observed uracil excretion, a likely candidate for this inefficient processis the degradation of UMP to uracil. Indeed, orotate addition results in both higher UMP and higher uracil (Fig. 1e), and when the orotate is 15N-labelled, the extent of uracil isotope labelling mimics that of UMP (Supplementary Fig. 3). To elucidate the route from UMP to uracil, we knocked out the genes known to catalyse the interconversion of UMP, uridine and uracil[17]: udk (uridine + GTP → UMP + GDP), upp (uridine + PRPP R UMP → pyrophosphate (PPi)) and udp (uridine + inorganic phosphate (Pi) → ribose-1-P + uracil). Only the udp knockout reduced uracil accumulation, decreasing it to 17% of the level normally found in the feedback-dysregulated background (Fig. 2b). We were able to eliminate uracil production more fully (to 6%) by double deletion of udp and the cytosine deaminase codA. Following orotate addition, this double-deletion mutant feedback-dysregulated strain accumulates uridine instead of uracil (Fig. 2c and Supplementary Fig. 4).

Figure 2

Pyrimidine overflow pathway is initiated by catabolism of UMP by UmpH

a, Pathway schematic. b, Uracil excretion does not depend on the canonical pyrimidine interconversion enzymes Udk and Upp, but does require Udp. Excreted uracil accounts for approximately half of total uracil. Error bars mark standard deviation (n = 2–3). c, Uracil is produced by UMP degradation following orotate addition. UMP is degraded to uridine by UmpH (also known as NagD) and UmpG (also known as SurE), and uridine to uracil by Udp. Isotopic tracing from orotate to UMP, uridine and uracil appears in Supplementary Fig. 3. Residual uracil in a Δudp ΔcodA strain indicates at least one unannotated activity also makes minor contributions to uracil production. End-product levels are not perturbed following orotate addition by deletion of either ΔumpH or ΔumpG individually, but double deletion leads to increased UTP. Error bars mark ± standard deviation (n =3). d, Inability to degrade UMP to uracil causes a growth defect upon orotate addition but increases final culture density (reproducible in two biologically independent experiments each of 6–8 technical replicates). D, attenuance.

There is no enzyme with a well-described physiological function of UMP degradation to uridine. This reflects the lack of a straightforward genetic screen for such enzymes. Although often annotated as reversible, the Udk and Upp reactions are thermodynamically unfavourable for the degradation of UMP (both ΔGUdk and ΔGUppc = +22kJ mol−1 based on ΔG°Udk = +19.5 kJ mol−1 and ΔG°Upp = +20 kJ mol−1 and [GTP] = 4.9 mM, [GDP] = 0.68 mM, [uracil] = 5.5 mM, [uridine] = 0.15 mM, [UMP] = 0.5 mM, [diphosphate] = ~1 mM and [PRPP] = 0.26 mM[23,24]), and indeed, their genetic deletion does not prevent uracil accumulation (Fig. 2b). On the basis of published in vitro enzymology and homology[17], we identified ~15 putative UMP phosphatase, hydrolase and nucleotidase genes with unknown physiological functions and screened deletion mutants for uracil levels following orotate addition. We found two genes whose deletion led to lower uracil and higher UMP levels than the wild type (Fig. 2c), indicating a possible role in UMP degradation in conditions of pyrimidine excess. We propose that these genes, nagD and surE, be renamed umpH and umpG, respectively, to highlight their newly recognized activities in UMP degradation. UmpH and UmpG belong to the haloacid dehalogenase-like phosphatase family and SurE phosphatase families, respectively, whose members span taxa from bacteria to humans, but little is known about their physiological importance[7,8,25,26]. We observed that knockout of umpH, although not altering pathway end-product levels, impaired growth of E. coli upon orotate upshift, with the double deletion showing a stronger phenotype and end-product accumulation (Fig. 2c, d and Supplementary Fig. 4). Thus, UmpH, and to a lesser extent UmpG, function in a UMP-degradation pathway that is required for optimal growth in response to environmental pyrimidine intermediates.

The UmpH and UmpG phosphatases degrade UMP to uridine, which is further degraded by Udp to uracil and ribose-1′-phosphate (cytosine similarly liberated from CMP may be deaminated by CodA to uracil). This overflow pathway dissipates three high-energy phosphate bonds (ATP equivalents) and one NADPH per uracil excreted aerobically, increasing to five ATP and inducing production of ubiquinol under anoxia, which may be problematic (Supplementary Fig. 5). The most straightforward way to control flux through this overflow pathway is through the concentration of its substrate UMP. In all conditions where we observe uracil excretion, we also observe increased intracellular UMP. The normal steady-state UMP concentration (0.052 mM) is below the Michaelis constant (Km) of UmpH for UMP (0.12 mM), but rises to eight times the Km under conditions inducing directed overflow metabolism (Fig. 2c). This suggests that UMP accumulation has a central role in triggering the overflow flux.

We sought to identify the molecular underpinnings of UMP accumulation during growth in medium enriched in pyrimidine intermediates. In analysing the regulatory architecture of the pathway, we noticed an additional potential feedback loop near the end of the pathway: UMP kinase (which catalyses the ATP-dependent phosphorylation of UMP to UDP), encoded by pyrH (Fig. 3a), is regulated in vitro by cooperative (ultrasensitive) inhibition by UTP (Hill-coefficient of 2.8, inhibition constant (Ki) of 154 μM)[6]. This inhibition can be partially overcome by GTP activation[6,27], and thus, like the regulation of ATCase by CTP and ATP, may function to achieve balance between pyrimidine and purine pools. Moreover, regulation at the end of the pathway would have the advantage of minimizing perturbations in UTP and CTP levels in response to alterations in any upstream pathway substrate (for example, not only of the de novo pathway, but also uracil, orotate, or their nucleosides).

Figure 3

Cooperative inhibition of UMP kinase by UTP maintains end-product homeostasis

Variant pyrH alleles were expressed from its native promoter on low-copy plasmid (pACYC) with genomic pyrH removed. a, Schematic of downstream regulatory events in pyrimidine metabolism. Wild-type UMP kinase (PyrH) is feedback inhibited by UTP in a switch-like manner (high degree of cooperativity). Allosteric parameters of pyrH alleles appear in shaded box, with D93A lacking the switch-like behaviour. nH, Hill coefficient. b, Altered expression or regulation of UMP kinase following orotate addition impairs growth. Defects also occur in pyrHWT/pyrH diploid strains (Supplementary Fig. 6). c, Metabolite fold changes upon orotate addition in strains with altered UMP kinase expression or allosteric regulation reveal defects in both pyrimidine and purine homeostasis. Error bars mark standard deviation (n = 3).

To test the importance of the cooperative inhibition of UMP kinase by UTP, we created strains expressing mutant forms of UMP kinase with reduced sensitivity to UTP: the N72A mutant has a higher Ki (373 μM for UTP rather than 154 μM), and the D93A has both a higher Ki (332 μM) and less cooperative inhibition (Hill coefficient 1.6 rather than 2.8). Because UMP kinase catalyses an essential reaction, we first transformed wild-type cells with low-copy plasmids (pACYC) carrying a natively expressed pyrH allele (pyrHWT, N72A or D93A) before knocking out genomic pyrH by transduction. We then assayed the growth and metabolic response of these strains to uracil and orotate. Although all three strains grew normally in the absence of pyrimidines and upon uracil addition (Supplementary Fig. 6), orotate addition inhibited growth of all three strains, particularly those with impaired inhibition of UMP kinase by UTP (Fig. 3b). The growth defect of the pyrHWT (pACYC::pyrHWT ΔpyrH) strain was presumably due to mild overexpression of wild-type UMP kinase, as inducible overexpression led to the same response (pAC24N::pyrHWT + 0.05 μM isopropyl β-D-1-thiogalactopyranoside (ITPG); Supplementary Fig. 6). Thus, proper control of UMP kinase protein levels is required for an optimal response to pyrimidine upshift. The cooperative feedback inhibition of UMP kinase by UTP and CTP is also important, as impairment of such regulation led to a profound growth defect upon orotate addition (pACYC::pyrH(D93A) ΔpyrH, Fig. 3b).

We also detected changes to the nucleotide pools of these pyrH mutants in response to orotate addition. In the wild type, orotate addition leads to UMP accumulation without major changes in NTP levels (Fig. 3c). Lack of proper UMP kinase regulation, however, results in increased UTP and decreased ATP. Thus, dysregulated pyrimidine metabolism saps purines, either by increasing their usage (for example, because higher pyrimidine NTPs leads to increased ribosomal RNA biosynthetic rates), decreasing their synthesis (for example, by depleting required substrates), or the combination of these factors.

We designate the general mechanism, whereby feedback inhibition of a downstream pathway step leads to excretion of a pathway intermediate or by-product, directed overflow metabolism (Fig. 4a). Such overflow is triggered by excessive biosynthetic pathway flux and is carried out by a degradation pathway sensitive to levels of an accumulating biosynthetic intermediate. It is analogous to overflow in central carbon metabolism, wherein excessive sugar catabolism (typically via glycolysis) leads to build-up of pyruvate[28], which may be excreted as lactate, ethanol or acetate, depending on the organism. In the case of lactate excretion in humans, inhibition of pyruvate dehydrogenase (by high NADH or inhibitory phosphorylation[29]) has a similar role to the inhibition of UMP kinase in the pyrimidine pathway: it forms a choke point that instead directs the enzyme’s normal substrate towards by-product formation and excretion (Fig. 4b). In the case of pyrimidine biosynthesis, the cooperative inhibition of UMP kinase by UTP renders the directed overflow mechanism exquisitely precise in controlling UTP and CTP levels.

Figure 4

Directed overflow metabolism in biosynthesis is analogous to central carbon overflow metabolism

a, Schematic of directed overflow metabolism as a biosynthetic regulatory mechanism. b, Schematic of overflow in central carbon metabolism, using the Warburg effect as a canonical example. PDH, pyruvate dehydrogenase; PDHK, pyruvate dehydrogenase kinase (which catalyses inhibitory phosphorylation of PDH); TCA, tricarboxylic acid.

Thus, pyrimidine homeostasis involves two strategies for regulation. The canonical feedback architecture contributes to metabolic efficiency by decreasing unnecessary de novo flux. Directed overflow metabolism provides end-product homeostasis by diverting excess flux to uracil, thereby ensuring end-product homeostasis in response to altered availability of the full range of pathway substrates and intermediates. These upstream and downstream regulatory mechanisms work in concert to balance speed, efficiency and robustness.

METHODS SUMMARY

E. coli (parent strain NCM3722 (ref. 30)) were prepared for metabolite measurement using a filter culture technique: cell-laden nitrocellulose filters were grown on agarose plates. Pyrimidine upshift was accomplished by transferring filters to plates enriched with indicated supplements. Metabolism was quenched and metabolites concomitantly extracted by placing filters into −20°C solvent (40% methanol, 40% acetonitrile and 20% water with 0.1M formic acid). This solvent mixture reliably extracts nucleotides, nucleosides and bases without their interconversion or degradation. Negative mode LC–MS and LC–MS/MS measurements were performed as previously described[23]. Deletion strains were created by P1 transduction from the Keio deletion collection. Stable strains carrying carB* were generated by electroporation and lambda Red-mediated recombination of PCR-amplified S948F into a ΔcarB::kan strain (where kan denotes kanamycin resistance) followed plating and selection for prototrophs on minimal media. Growth was assayed by absorbance at 600nm in a 96-well format using a Biotek Synergy II plate reader. Competitive growth advantage was assessed by co-culture of a marked lacZ (ΔlacZ::kan) and an unmarked lacZ strain, determination of relative cell numbers after competitive growth by plating on MacConkey agar containing 1% lactose, and regression analysis of wild type to ΔpyrI carB* ratios. Unless otherwise indicated, growth media for metabolic studies consisted of Gutnick minimal media with 0.4% glycerol and 10 mM NH4Cl with triple-washed ultrapure agarose and 0.8 mM orotate as needed.

METHODS

Strains

The wild-type E. coli K-12 strain NCM3722 was used in this study as wild type and as parent for all strains in this study because it lacks the rph-1 mutation carried by MG1655 that causes pyrimidine pseudo-auxotrophy from reduced expression of orotate phosphoribosyltransferase (pyrE)[30]. Single-gene deletion mutants of NCM3722 were generated by P1 transduction[31] of deletion alleles with kanamycin resistance (kan) cassettes from the Keio collection[32] and verified using PCR with gene-specific primers. Multiple gene deletions were similarly created serially using the FLP helper plasmid system (pCP20), which was also used to eliminate kanamycin resistance cassettes to produce scarred deletions[33].

The carB(S948F) mutant was generated using the lambda-Red recombinase system[33]. A plasmid-borne S948F mutant allele of carB[21] was amplified from plasmid by PCR (carB forward primer, 5′-CGCATAAATCCCTGTTCGAC-3′; carB reverse primer, 5′-CCATTCGGCGATTAACAAGT-3′), gel-purified and used to transform a NCM3722 ΔcarB::kan strain carrying pKD46 and induced with arabinose as described previously[32]. Recombinants were selected on minimal media for pyrimidine and arginine prototrophy and subsequently cured of pKD46. Following screening for loss of all antibiotic resistance (ampicillin and kanamycin), the carB(S948F) (carB*) mutation was confirmed by sequencing.

Strains natively expressing variant pyrH alleles in a ΔpyrH::kan background were also created using the lambda-Red recombinase system. The pACYC::pyrH is the ligation product of XbaI-digested, CAP-treated pACYC184 (all New England Biolabs) with XbaI-digested wild-type pyrH generated by PCR from NCM3722 genomic template using primer set (pyrH forward primer, 5′-ACTCTAGACCAA TGCAAAACCCGTCTAT-3′; pyrH reverse primer 5′-GATCTAGAACTTACG CGGAATCTTACCC-3′) (XbaI sites are underlined). Variant alleles were produced through site-directed mutagenesis (Genewiz) and verified by sequencing. NCM3722 carrying plasmid pACYC::pyrH and pKD46 was transformed with a kan cassette flanked by pyrH homology regions (created by PCR of a Keio collection strain with PAGE-purified primer set (forward, 5′-TTGTAAATTCAGCTAACCCTTGTGGGGCTGCGCTGAATTCCGGGGATCCGTCGACC-3′;reverse, 5′-ACCAAACTGCCTGCAACAATAACGCCTTATAACCAGTGTAGGCTGG AGCTGCTTCG-3′)) to create a genomic ΔpyrH::kan allele. To prevent suppressing mutations, the ΔpyrH::kan allele was introduced by P1 transduction to strains carrying pyrH alleles on the pACYC plasmid and thereafter strains retained plasmid born resistances in absence of antibiotic.

Strains expressing pyrH from IPTG-inducible promoter on the pAC24N were created by transformation with pAC24N::pyrH from the Aska collection after removal of green fluorescent protein by NotI digestion and self-ligation[34]. Variant alleles were produced through site-directed mutagenesis (Genewiz) and verified by sequencing. Low-level expression was induced with IPTG as indicated.

Media and bacterial culture

Gutnick glucose minimal medium (‘Glucose’, Gmin1) refers to a salts mixture[35] with a 0.4% (w/v) glucose carbon source and a 10 mM ammonium chloride nitrogen source unless otherwise stated. ‘Glycerol’ and ‘acetate’ refer to the identical Gutnick with 0.4% glycerol or 0.4% sodium acetate, respectively, substituted for glucose. Strains were also grown in media supplemented (+) with one of the following: adenine, arginine, guanosine, hypoxanthine, ornithine, uracil or 2,4-dinitrophenol (all 1 mM). Media with alternative nitrogen sources contained Gutnick salts, 0.4% glycerol and 10 mM total concentration of usable nitrogen. Phosphate-limited MOPS (TekNova) medium was prepared with 0.4% glycerol, 10 mM ammonium chloride and 0.5 mM monobasic potassium phosphate[36]. Anoxia experiments were conducted in Gmin1 media at 37°C inside a chamber containing approximately 90% nitrogen gas, 5% carbon dioxide and 5% hydrogen. Oxygen levels were maintained at 0 p.p.m. and rarely exceeded 100 p.p.m. +/− indicates alternating every 8 h between glycerol minimal media and enriched minimal media as indicated above. ‘LB, glucose’ indicates cells were alternated between LB and Gmin1 media every 8 h. Difco MacConkey agar (Becton, Dickinson) was prepared according to the manufacturer’s instructions with 1% lactose as carbon source.

Metabolite measurements

Cells were cultured and extracted at −20°C 40:40:20 methanol: acetonitrile: water with 0.1 M formic acid[37] using the filter culture method described previously[38]. Low-molecular-weight metabolites were quantitated using both LC–MS[39] and LC–MS/MS[40] at both steady state[23] and following addition of exogenous pyrimidines, including unlabelled and isotopically labelled uracil and orotate. Absolute quantification of uracil and UMP was computed as in ref. 23. LC–MS(/MS) data were analysed using MAVEN software[41]. U-[15N]-tracers (for example, uracil and orotate) were purchased from Cambridge Isotopes Laboratories.

Competition assays

The lacZ− (ΔlacZ::kan) derivatives of wild-type and ΔpyrI carB* strains were prepared as described above. After overnight culture in a 50:50 mix of medium for competition and Gmin1, lacZ+ and lacZ strains were combined to equal proportions (based on attenuance D600) and were diluted to an OD600 of 0.02 in medium for competition to create two paired cultures: wild type lacZ with ΔpyrI carB* lacZ and wild type lacZ with ΔpyrI carB* lacZ. Each culture pair was grown at 37°C and generally diluted every 8 h (12 h for media with doubling time over 120 min) into fresh medium to D600 of 0.02. Some cultures were diluted into alternating media as indicated above.

The ratio of wild type to ΔpyrI carB* of the culture was determined by regular plating on MacConkey agar plates with 1% lactose (6–10 replicates). Cultures were diluted to ensure that approximately 100 colonies formed on each plate.

Growth advantage was computed by a linear regression of the ratio of wild-type and mutant strains. The ratio of wild-type to mutant cells in time, t, can be described as R(t) = R0× 2(1/−1/), where R0 is the initial ratio, and W and M are the doubling times for the wild type and the mutant, respectively. Performing a linear regression of this equation (R software, http://cran.r-project.org) in logarithmic space allows for computation of the growth advantage.

For 2,4-dinitrophenol, advantage was similarly computed from the growth rate of each strain (lacZ+) individually in culture and plotted along the diagonal.

Microarrays

RNA purification, labelling, microarray measurement and data processing were performed as described in ref. 42. In brief, overnights of the wild-type and ΔpyrI carB* mutant were prepared as above in Gmin2 and grown to mid-exponential phase (D600 of 0.4) and then mixed with RNAprotect Bacteria Reagent according to manufacturer’s specifications (Qiagen). After 5 min at 25°C, the mixture was pelleted for 10min at 5,000g, and RNA was extracted using the Total RNA Purification kit (Norgen Biotek). Purified total RNA was treated for 30 min using E. coli poly(A) polymerase (New England Biolabs). RNA from the wild-type and ΔpyrI carB* strains were labelled with Cy5 and Cy3, respectively, using the Two-Colour Low Input Quick Amp Labelling Kit (Agilent). Samples were hybridized to an E.coli Gene Expression Microarray (Agilent). Resulting data were analysed using iPAGE[43] and R software.

Growth assays

Absorbance measurements were obtained using a Biotek Synergy II plate reader (Biotek, Winooski, VT)in 96-well formatat 600 nm using clear, flat-bottom plates with two independent biological replicates each consisting of six to eight technical replicates. Cultures included 0.1% Tween-20 to prevent clumping and were sealed with air permeable membrane. To facilitate comparisons, some curves are aligned to D600 of 0.03.