The Cost of Expression of Escherichia coli lac Operon Proteins Is in the Process, Not in the Products
Daniel M. Stoebel, Antony M. Dean, Daniel E. Dykhuizen


Transcriptional regulatory networks allow bacteria to express proteins only when they are needed. Adaptive hypotheses explaining the evolution of regulatory networks assume that unneeded expression is costly and therefore decreases fitness, but the proximate cause of this cost is not clear. We show that the cost in fitness to Escherichia coli strains constitutively expressing the lactose operon when lactose is absent is associated with the process of making the lac gene products, i.e., associated with the acts of transcription and/or translation. These results reject the hypotheses that regulation exists to prevent the waste of amino acids in useless protein or the detrimental activity of unnecessary proteins. While the cost of the process of protein expression occurs in all of the environments that we tested, the expression of the lactose permease could be costly or beneficial, depending on the environment. Our results identify the basis of a single selective pressure likely acting across the entire E. coli transcriptome.

A central tenet of evolutionary biology is that trade-offs arise as organisms allocate limited resources to various competing traits. For example, the differential allocation of intermediary metabolites drives a trade-off between investment in structures that increase reproduction and structures that increase survival (Harshman and Zera 2007). The allocation of limiting amounts of time shapes patterns of animal behavior (Stephens and Krebs 1986). Differences in the availability of nutrients to plants drive a trade-off between a fast-growing, poorly defended strategy and a slow-growing, well-defended strategy (Coley et al. 1985). In all of these cases, trade-offs shape patterns of morphological and behavioral diversity.

Trade-offs occur not only for morphology and behavior, but also for cellular processes such as gene expression. Regulatory networks to control gene expression in response to specific environmental cues have presumably been selected to manage these trade-offs. Experimental evolution can be used to measure the costs of unnecessary expression and to explore the proximate causes of this selection. Zamenhoff and Eichhorn (1967) demonstrated that it is costly for Bacillus subtlis to produce the proteins for trytophan biosynthesis when they are not needed, but did not determine why. Dykhuizen (1978) demonstrated that the cost for Escherichia coli of expressing the same proteins could not be explained by a simple energy conservation argument. Dykhuizen was, however, unable to determine the source of the cost. Several authors (Novick and Weiner 1957; Andrews and Hegeman 1976; Dykhuizen and Davies 1980; Koch 1983) have shown that constitutive expression of the lac operon in E. coli lowers fitness when there is no lactose present in the environment. While various costs have been suggested, no experiments have directly tested the competing hypotheses. The molecular basis of the cost of unnecessary expression, a potentially major force acting on regulatory networks, remains undiscovered.

The lactose operon provides an ideal model for addressing this question because it combines our ability to measure the fitness effects of specific mutations (Dykhuizen and Davies 1980; Lunzer et al. 2002) with a simple regulatory system that is thoroughly understood and that can be extensively manipulated. Expression of the lactose operon is required for E. coli to metabolize lactose (milk sugar) for carbon and energy. A specific repressor prevents transcription of the operon unless lactose is present in the growth medium (Miller and Reznikoff 1978). Strains with nonfunctional repressors (called constitutive and denoted lacI) maximally express the three lactose operon proteins (lacZ encodes the β-galactosidase, lacY encodes the lactose permease, and lacA encodes the transacetylase) under all of our experimental conditions. Understanding this simple mechanism of regulation enables us to form testable hypotheses about the costs associated with wasteful protein synthesis.

Why is it costly to express the lac operon when there is no lactose to metabolize? We aim to do more than invoke “energetic” or “metabolic” costs of expression, as all aspects of producing proteins require energy and metabolites. Instead, each step in the production and use of these proteins can be hypothesized to be costly:

  • Transcription could be costly because it uses nucleotides that could be incorporated into other RNAs.

  • Transcription occupies RNA polymerases (Ferenci 2005) that might be better used to transcribe genes whose products increase fitness.

  • Translation wastes charged tRNAs and occupies free ribosomes (Vind et al. 1993).

  • The proteins produced by translation tie up amino acids that might be better incorporated into other beneficial proteins.

  • Costly activities of the proteins, e.g., insertion of the permease in the membrane, might allow protons to leak into the cytoplasm, thereby partially dissipating the proton motive force. In addition, insertion might affect membrane fluidity and/or occupy space needed for other membrane proteins (Dykhuizen and Davies 1980).

We tested this series of hypotheses using a combination of molecular genetics and competition experiments. We began by making mutations that abolished one or a few of these potential costs. For example, we deleted lacY from a constitutive strain so that it would not have any of the permease-specific costs outlined above. To determine the fitness effect of this mutation, we competed the mutant strain against the constitutive parental strain during growth in a chemostat (Feldgarden et al. 2003). If the deletion of lacY ameliorated some of the cost of constitutive expression, then this mutant strain should grow faster than a constitutive strain, causing it to rise in frequency in the population at the expense of its competitor. The rate of change of the frequency is our estimate of the fitness difference between the two strains.



Minimal Davis salts (MD) is 7 g K2HPO4, 2 g KH2PO4, 1 g (NH4)2SO4, 0.5 g trisodium citrate, and 0.2 g MgSO4·7H2O in 1 liter of distilled deionized water. Chemostats were supplemented with 5 μm FeSO4 (from a 5 mm FeSO4 + 7.5 mm Na2EDTA stock). Carbon sources were at a concentration of 0.1 g/liter for chemostats and 2 g/liter for flasks and plates. For minimal media plates, Bacto agar was added to 15 g/liter after autoclaving. MacConkey agar was prepared according to the manufacturer's specifications. LB agar was 10 g tryptone, 5 g yeast extract, 10 g NaCl, and 13 g Meer agar. Antibiotics were used at 15 mg/liter tetracycline, 50 mg/liter kanamycin, 100 mg/liter ampicillin, and 20 mg/liter chloramphenicol.

Strains and mutations:

All strains are listed in Table 1. The common genetic background for these experiments is DMS265, which is wild type except for a small deletion of the lac operon. It is derived from DD320 (Dykhuizen and Davies 1980; Dykhuizen and Dean 1994; Lunzer et al. 2002) by P1 transduction of glpF+ from strain MG1655, allowing DMS265 to grow on glycerol.

Mutant sequences were constructed by overlap-extension PCR (Sambrook and Russell 2001) and cloned into plasmid pCR 2.1 (Invitrogen, San Diego). All primers are listed in supplemental Table 1 at The mutation was recombined into the chromosome via gene gorging (Herring et al. 2003). Cells were then plated onto lactose MacConkey agar, where pale colonies (some mutant colonies had pink centers) were picked and purified again on the same media. P1 transduction was used to move the mutant lac operon into a fresh DMS265 background. Transduction of lacI+, lacI, lacI lacZ·ssrA, and lacI ΔRBS was onto MD lactose, lacI ΔlacZ was onto MD melibiose, and lacI ΔlacY was onto MD methyl galactoside. All strains were stored at −80° in 20% glycerol, and a fresh aliquot was used to start all competition experiments. The entire lac operon of all transductants was sequenced to confirm that there were no secondary mutations.

The lacZ·ssrA showed no phenotype on lactose MacConkey plates, so a two-step process was required to insert it. The lacZ·ssrA construct was amplified and cloned into pCR2.1 to create pDMS20. tetR was amplified from pACYC184 with primers tetR+BlpI and tetRBstBI. This PCR product and pDMS20 were digested with restriction enzymes BlpI and BstBI, and tetR was ligated into the plasmid. This plasmid, pDMS26, has the 3′-end of lacZ and the 5′-end of lacY replaced with tetR. tetR was recombined into the chromosome of DMS269, resulting in pale colonies on lactose MacConkey/tetracycline plates creating DMS1383. The lacZ·ssrA construct was introduced to DMS1383 on pDMS20, and recombinants were selected on MD lactose.

Spontaneous mutants resistant to the bacteriophage T5 (caused by a mutation in fhuA) were isolated from soft LB agar supplemented with 5 mm CaCl2 in the presence of excess T5 phage. Resistant mutants were purified by streaking for single colonies on LB plates.


Chemostats were run and monitored as previously described (Lunzer et al. 2002).

Enzyme and protein assays:

Amounts of active lac proteins were determined enzymatically. Lactose permease and β-galactosidase activity (Dean 1989) and transacetylase activity (Alpers et al. 1965) were measured as described, with 100-μl volumes in a Molecular Devices SpectraMax Plus 384-plate reader. Measures of lactose permease activity in the ΔlacZ strain relied on the fact that the α-galactosides are transported by lactose permease and that the α-galactoside permease (encoded by melB) is inactive at 37° (Putzrath and Wilson 1979). Protein concentration was measured with the Bio-Rad (Hercules, CA) protein reagent with a BSA standard. Enzyme and protein levels were measured from three replicate chemostats.

Real-time quantitative PCR:

Real-time quantitative PCR (QPCR) was used to measure lacZ mRNA abundance. Total mRNA was isolated from 10 ml of chemostat culture using a RNeasy mini kit (QIAGEN, Valencia, CA), following the manufacturer's instructions, and was treated with RNase-free DNase (QIAGEN). QPCR was performed with a Stratagene (La Jolla, CA) Brilliant SYBR Green QRT-PCR master mix kit on a Stratagene Mx3000P. Following an initial 10 min denaturing, amplification was 40 cycles of 20 sec at 95°, 20 sec at 55°, and 20 sec at 72°. RNA concentration was quantified as cycles to a threshold level, using an adaptive baseline.

RNA was isolated from four replicate chemostats each of DMS267, DMS269, and DMS1360 and assayed for the level of lacZ mRNA in each sample in duplicate on two separate occasions. Thus, each sample was assayed four times, and these measures were averaged.

Data analysis:

The selection coefficient per hour was determined by regressing ln(fhuA/fhuA+) against time. The rationale for this was given by Dykhuizen and Hartl (1983b). Conversion of selection coefficients per hour to selection per generation is complicated by the fact the definition of a generation is not trivial in a continuously growing population. In some articles (Dykhuizen and Hartl 1983a), generations are defined in terms of a population doubling, which is given by multiplying the per-hour selection coefficients by ln(2)/D. Other articles, including those involving work on lac (Dean 1989; Dykhuizen and Dean 2004), have acknowledged the continuous nature of population growth in chemostats and defined generations as ln(e)/D. So that the coefficients reported here are compatible with the other work on lac, we used the latter definition of a generation.

Resistance to phage T5 did not result in a measurable effect on fitness, so replicate experiments that differed only in the marker status of strains were pooled. (For example, experiments competing lacI+ vs. lacI fhuA and lacI+ fhuA vs. lacI were pooled.)

To determine the cost of each gene's expression on glucose, we determined both the selective difference between a pair of strains (Table 2) and the difference in expression of each of the three lac proteins between for these strains (Table 3). ONPG (o-nitrophenyl β-d-galactopyranoside) translocation could not be used to measure the amount of LacY activity of DMS1346, because this strain is ΔlacZ, so there is no ONP production to monitor. Instead, we used the rate of translocation of α-p-nitrophenyl α-d-galactopyranoside (PNPG), for which the concentration of LacY is rate limiting at low substrate concentrations (Tsuchiya et al. 1978; Putzrath and Wilson 1979). In addition to the fitness results with ΔlacZ and ΔlacY, we also included the experiments competing DMS1360 with both DMS267 and DMS269 in this analysis.

We partitioned the cost with the following model:Math

In addition to a standard analysis, there are several possible alternative ways of performing this analysis. Because the selection difference between a pair of strains is a mean, we tried weighting these values by the number of replicates used to determine this mean (Faraway 2002). The difference in LacA activity between DMS1333 and DMS269 was not significantly different, so we also ran the analysis with no difference in LacA activity for these two strains. Both of these variants changed the predicted cost of LacZ by <1%, the cost of LacY by <5%, and the cost of LacA by <25%.

To compare these estimated costs with what might be expected if the cost was proportional to the amount of transcription, we multiplied each gene's proportion of the operon by 11.17%, which is the sum of the estimated costs. This calculation ignores untranslated regions of the mRNA, which are ∼4% of its length. To make an a priori prediction of the cost of amino acids, we assumed that there are 6 mol of LacZ for each mole of LacA (Beckwith 1987) and 2 mol of LacY for each mole of LacA (Jones and Kennedy 1969). If the differences in protein amounts are due to differences in rates of synthesis rather than in rates of degradation, then these predictions should also be the cost of translation.

All statistical analysis was performed with the software package R version 2.2.1 (R Development Core Team 2005).


The strains used in this study (Table 1) are all isogenic except for the specific mutations of interest at the lac operon. During competition in a chemostat, strains are distinguished by their resistance or sensitivity to the phage T5, which is caused by a mutation (fhuA) that does not affect fitness (Lunzer et al. 2002). The rate of change of the marker frequency is estimated by regression and is our measure of the fitness difference between two strains. The small but significant heterogeneity among replicates means that variation is calculated among rather than within replicates (Table 2).

View this table:

Strains used in this study

View this table:

Results of selection experiments

Competitions between otherwise isogenic strains of E. coli K-12 carrying regulated and constitutive lac operons confirm earlier work (Novick and Weiner 1957; Andrews and Hegeman 1976; Dykhuizen and Davies 1980; Koch 1983) that a regulated strain (DMS267) is fitter than a constitutive strain (DMS269) when growing on a carbon source other than lactose (Table 2). To confirm that the fitness difference was caused by the difference in the regulation of lac, and not caused by mutations elsewhere in the genome, we competed the regulated strain and the constitutive strain in the presence of 50 μm IPTG, which is sufficient to fully induce expression of lac in the regulated strain. As expected, the selection was eliminated (Table 2). These experiments demonstrate that the cost of expressing the lac operon is 10.28%/generation when growing on glucose.

We next turned to testing the various hypotheses to explain this cost. To determine if there were specific costs of expressing the lacY-encoded permease, we separately deleted lacZ (DMS1346) and lacY (DMS1333) from the constitutive strain and completed each against a strain carrying either a regulated or a constitutive operon. Enzyme assays showed that the deletions affected the levels of expression of the other proteins (Table 3), complicating the interpretation of these competition experiments. To estimate the cost of expressing each protein, we used a multiple regression model where the fitness difference between two strains was explained by the difference in expression of each of the three proteins. Because levels of the LacA protein vary among the strains, the cost of LacA is estimated by the multiple regression analysis without a ΔlacA strain. The regression shows that by far the largest cost (67.7% of the total) is attributable to expression of lacZ (Table 4).

View this table:

Activity of the three lac operon proteins

View this table:

The cost of each lac protein from multiple regression and as predicted from the proportion of lac DNA and amino acids

What might we expect the cost to be? If the cost were due to transcription, we would expect the cost of each gene to be proportional to the size of each gene. If the cost were due to the act of translation or to tying up amino acids in useless proteins, we would expect the cost to be proportional to the amount of each protein produced. (This is not the same as the amount of DNA due to unequal translation of the three genes; see materials and methods.) We find that the fitness cost of expressing each of the three genes can be predicted from the amount of protein expressed or from the size of each gene (Table 4). Hence, lacY-specific effects—on membrane fluidity, space, and proton leakage—are not major costs to fitness during growth on glucose.

The observation that the cost of expression is proportional to the amount of each protein produced is consistent with the energy conservation hypothesis, which proposes that amino acid synthesis to make up for amino acids tied up in useless protein is the major energetic cost. Returning amino acids to the free amino acid pool for use in other proteins should relieve most of the cost if the energy conservation hypothesis is correct. We added the ssrA tag (AANDENYALAA) to the end of lacZ, which targets the protein for degradation by the ClpXP and ClpAP proteases (Gottesman et al. 1998). These proteases degrade the protein into short polypeptides whose individual amino acids can then be recycled (Sauer et al. 2004; Yu and Houry 2007). The constitutive lacZ·ssrA mutant (DMS1397) has the same fitness as the constitutive strain with wild-type lacZ (DMS269) (Table 2 and Figure 2A), despite a 38.9% reduction in the level of LacZ protein (Table 3). [As a control, the fitness difference between the regulated (DMS267) and constitutive lacZ·ssrA (DMS1397) strains is not different from the difference between regulated and constitutive strains (t-test, P = 0.99).] Returning amino acids to the free amino acid pool for use in other proteins has no effect on the cost of expression. Sequestering amino acids in useless protein is not a major cause of selection against constitutive expression.

Figure 1.—

Rate of α-PNPG translocation by the constitutive strain (DMS269, “+”), the constitutive ΔlacZ strain (DMS1346, circles), and a strain with at least five copies of the lac operon (DD2460, triangles). At low substrate concentrations, the activities of DMS269 and DMS1346 are identical, indicating that the amount of LacY is the same in the two strains. As a control, DD2460 has much higher rates of transport at low substrate concentrations, as would be expected if the availability of LacY were rate limiting for transport.

The preceding results indicate that the cost of expression of the lac operon when growing on limiting glucose is not due to the product of expression, but rather to the process. We attempted to distinguish between the hypotheses that (1) the cost is due to tying up RNA polymerase or nucleotides (transcription) and that (2) the cost is due to tying up ribosomes and tRNAs (translation). To do this, we deleted the ribosome-binding site (RBS) from the leader sequence in front of lacZ. The ΔRBS mutation reduces β-galactosidase expression to 1.38% of normal constitutive levels (Table 3). Competing this constitutive ΔRBS strain (DMS1360) with a normal constitutive strain (DMS269) produced a mean fitness difference of 11.89% (Table 2 and Figure 2B), which is not significantly different from the difference between strains carrying constitutive and regulated operons (t-test, P = 0.33). Further, the constitutive ΔRBS strain (DMS1360) is selectively neutral when placed in competition with a regulated strain (DMS267) (Table 2 and Figure 2B). Deleting the RBS ameliorates all of the cost of unnecessary expression.

To determine if deleting the RBS affected transcription, we used RT–QPCR to assay the level of lacZ message in these cells. We found that the ΔRBS strain (DMS1360) had lacZ mRNA levels 61-fold lower than constitutive (DMS269), but 26 times that of regulated (DMS267). Yarchuk et al. (1992) found that point mutants abolishing translation both lowered the rate of transcription of the lac operon and destabilized the mRNA that was produced. If the lower level of lac mRNA in the ΔRBS strain was due solely to lower stability of the lac mRNA, then the cost of expression is due to the act of translation. If the lower levels of lac mRNA were due to lower rates of transcription, then the cost could be due to either transcription or translation. While it may not be possible to make a mutant that distinguishes between the costs of transcription and translation, our results do confirm that the cost constitutivity is in the process, not the product, of gene expression.

To explore the generality of the cost of expression, we competed regulated (DMS267) and constitutive (DMS269) strains during growth with succinate or maltose as sole carbon sources. Succinate was chosen because it is structurally distinct from glucose (it is a dicarboxylic acid rather than a sugar), while maltose was chosen precisely because of its similarity to glucose (it is a dimer of glucose).

The cost of expression on succinate (9.69%/generation) (Table 2) is not different from that on glucose (t-test, P = 0.61). Partitioning costs by multiple regression (Table 4) shows that the cost of LacZ is quite similar to that on glucose, that LacY is over twice as costly (a nonsignificant result after Bonferroni correction), and that LacA has a cost not significantly different from zero. We conclude that the cost of expression during growth on succinate and glucose are similar.

Surprisingly, the cost of expression on maltose (1.60%/generation) (Table 2) is far lower than expected. The structural similarity of maltose and lactose allows the lacY-encoded permease to transport maltose, albeit very inefficiently (Brooker and Wilson 1985). This ability to transport maltose could ameliorate much of the cost of constitutive lac expression during starvation on maltose. Maltose is unable to induce lac expression. The regulated strain, lac operon firmly repressed, can derive no benefit from its lacY during growth on maltose.

The hypothesis that maltose transport by LacY modifies the cost of constitutivity predicts conflicting selection pressures on expression of different lac proteins. Production of the lacZ encoded β-galactosidase and the lacA-encoded transacetylase should impose the same cost of translation during growth on maltose as during growth on glucose. In contrast, production of the lacY-encoded permease should be advantageous during growth on maltose and deleterious during growth on glucose. Both predictions proved correct. The regulated strain (DMS267) is fitter than the constitutive ΔlacY strain (DMS1333) by the same amount on maltose as on glucose (Table 2 and Figure 3A) (t-test, P = 0.49), indicating that the cost of translating lacZ and lacA is the same on both sugars. The constitutive ΔlacZ strain (DMS1346) is fitter than the regulated strain (DMS267) on maltose (Table 2 and Figure 3B) and less fit on glucose. Hence, expressing lacY increases fitness on maltose, while expressing lacZ and lacA decreases fitness.

Figure 2.—

Competition experiments to test for the cost of amino acid usage and cost of production on glucose. A constitutive ssrA-tagged strain (DMS1397) competed against a constitutive strain (DMS269, open circles) and against a regulated strain (DMS267, solid circles). (B) The constitutive ΔRBS strain (DMS1360) competed against the constitutive strain (DMS269, open circles) and regulated strain (DMS267, solid circles). In A and B, the line is the selection difference between regulated and constitutive, 10.28%/generation. (A) The ssrA tag, which recycles protein, does not alleviate the cost of expression. (B) Abolishing the act of production alleviates all of the cost of expression.

Figure 3.—

Competition experiments to determine why constitutivity is more costly on glucose (open circles and dotted lines) than maltose (solid circles and solid lines). Competition of a regulated strain (DMS267) and a constitutive ΔlacY strain (DMS1333) on glucose and maltose shows that the cost of LacZ and LacA is the same on both carbon sources (6.39 and 6.09%/generation, respectively). (B) A regulated strain (DMS267) is more fit than a constitutive ΔlacZ strain (DMS1346) on glucose (3.43%/generation), but ΔlacZ is more fit than the same regulated strain on maltose (2.54%/generation), indicating that LacY is costly on glucose but beneficial on maltose.


Our experiments identify the process of gene expression as the cost of constitutive expression. In contrast to the explicit or implicit expectations of other authors, there are no measurable costs specific to the lactose permease or to sequestering amino acids in useless protein. Other authors have implicated the availability of free ribosomes (Sarubbi et al. 1988; Vind et al. 1993; Cooper et al. 2003) or RNA polymerase (Farewell et al. 1998; Notley-McRobb et al. 2002) as crucial for the control of growth rate, although their data could be confounded by effects arising from the correlated use of ribonucleotides and charged tRNAs. Unnecessary translation of one protein can reduce rates of translation of other proteins (Vind et al. 1993), and unnecessary transcription of one set of genes can reduce the level of transcription of other genes (Farewell et al. 1998). However, we are the first to have directly manipulated costs to exclude potential costs of the products of gene expression.

The rooting of the cost of constitutivity of lac in the process of gene expression provides encouragement that the cost of lac expression is likely to be general. If the cost had proven to be due to LacY-specific effects, then the cost would not be general. At a minimum, such a result would mean that selection would act quite differently on operons that did and did not contain membrane proteins.

The maltose and succinate results highlight the fact that, while the cost of translation occurs in all environments, the cost of permease expression depends on the carbon source. Dramatically, in maltose the expression of the permease becomes beneficial while expression of the other genes remains costly. This is likely to be a general phenomenon, as catabolic enzymes organized into a single operon often have varying specificities for various compounds. Moreover, the costs and benefits of expressing biosynthetic enzymes organized into a single operon might also vary with the availability of different biosynthetic precursors in the environment. Selection should favor the fragmentation of an operon during prolonged growth on a substrate metabolized only by a subset of its gene products. In cases where maintenance of an operon is selected, differing substrate specificities among gene products or precursor availabilities might select for the evolution of internal promoters.

The cost of the process of gene expression provides the selective advantage for the evolution of repression. Our results thus provide a general explanation for why regulatory systems have been, and are being, evolved.


We thank Mark Lunzer for help with chemostats and Walt Eanes, John Flowers, Michael Last, John True, and two anonymous reviewers for comments. This work was funded by a National Science Foundation (NSF) predoctoral fellowship and a NSF Doctoral Dissertation Improvement Grant to D.M.S. and by Public Health Service grant no. GM060731 to D.E.D. and no. GM060611 to A.M.D. This is contribution 1161 in Ecology and Evolution from Stony Brook University.


  • Communicating editor: S. Yokoyama

  • Received December 9, 2007.
  • Accepted January 9, 2008.


View Abstract