The Response to 2-Aminoacrylate Differs in Escherichia coli and Salmonella enterica, despite Shared Metabolic Components

ABSTRACT

The metabolic network of an organism includes the sum total of the biochemical reactions present. In microbes, this network has an impeccable ability to sense and respond to perturbations caused by internal or external stimuli. The metabolic potential (i.e., network structure) of an organism is often drawn from the genome sequence, based on the presence of enzymes deemed to indicate specific pathways. Escherichia coli and Salmonella enterica are members of the Enterobacteriaceae family of Gram-negative bacteria that share the majority of their metabolic components and regulatory machinery as the “core genome.” In S. enterica, the ability of the enamine intermediate 2-aminoacrylate (2AA) to inactivate a number of pyridoxal 5′-phosphate (PLP)-dependent enzymes has been established in vivo. In this study, 2AA metabolism and the consequences of its accumulation were investigated in E. coli. The data showed that despite the conservation of all relevant enzymes, S. enterica and E. coli differed in both the generation and detrimental consequences of 2AA. In total, these findings suggest that the structure of the metabolic network surrounding the generation and response to endogenous 2AA stress differs between S. enterica and E. coli.

IMPORTANCE This work compared the metabolic networks surrounding the endogenous stressor 2-aminoacrylate in two closely related members of the Enterobacteriaceae. The data showed that despite the conservation of all relevant enzymes in this metabolic node, the two closely related organisms diverged in their metabolic network structures. This work highlights how a set of conserved components can generate distinct network architectures and how this can impact the physiology of an organism. This work defines a model to expand our understanding of the 2-aminoacrylate stress response and the differences in metabolic structures and cellular milieus between S. enterica and E. coli.

INTRODUCTION

The metabolic networks of microbes are characterized by robustness and redundancy that ensure their ability to absorb perturbations caused by internal and external stress (1–4). Our ability to predict the metabolic architecture of an organism and how it will respond to perturbations is poorly developed. Despite the fact that metabolic components encoded by the genome can be defined, it is not yet clear how accurately the metabolic capacity and robustness of an organism are predicted by the mere presence of these metabolic components. In some cases, the status of central and linear pathways that are predicted from genome annotation can be tested by simple growth phenotypes. For instance, the lack of a single histidine biosynthetic enzyme (i.e., ATP phosphoribosyltransferase [HisG]; EC 2.4.2.17) predicts that there will be a requirement for exogenous histidine. In contrast, growth phenotypes resulting from differential regulation can prove more difficult to predict by using genomic sequences (5). Similarly, phenotypes caused by network perturbations involving flux diversion and/or pathway recruitment are not often evident through genome analysis. The latter point is illustrated by the differences in thiamine biosynthesis between Salmonella enterica and Escherichia coli, despite the similarity of all relevant enzyme components (6). In addition to modulating the structure of single pathways, microbes often reconfigure their global metabolic networks to absorb stress and preserve metabolic function (7–9). The capacity to modify the network structure and the mechanism(s) used to do so introduce another layer of complexity that is difficult to define based on genome sequences. Historically, phenotypic analysis of model organisms has been instrumental in identifying and dissecting these complexities.

Systems biology approaches to metabolism often depend on genome annotation to generate metabolic models of the relevant organism(s) (10). The models typically describe the metabolic potential of an organism by assuming that networks with shared component parts are configured and respond to perturbations similarly. In this scenario, differences in metabolic potential indicate a different catalog of component parts. Examples of such metabolic divergence abound. For instance, the absence of the lac operon in Salmonella enterica relative to Escherichia coli explains their distinct metabolic capacities in this area (11). Organisms can also modulate the expression of conserved orthologs to generate a distinct network architecture and accommodate differences in lifestyles (12). In some instances, sequence analysis alone can predict these distinct expression patterns, based on knowledge of regulators and binding sites in combination with global analyses. Recent work outlining the distinct transcriptional responses produced by E. coli and S. enterica during triclosan exposure provides an example of this scenario (13).

The study described here addressed the responses of two organisms with similar metabolic components to the endogenously generated reactive metabolite 2-aminoacrylate (2AA). Pyridoxal 5′-phosphate (PLP)-dependent enzymes catalyzing β-elimination reactions proceed through reactive enamine intermediates, one of which is 2AA (14, 15). A subset of 2AA-generating enzymes releases this reaction intermediate from the active site into the cellular milieu (16–18). The RidA protein was demonstrated to be an enamine/imine deaminase, conserved in all domains of life. Previous work with S. enterica showed that in the absence of ridA, 2AA accumulates endogenously and inactivates enzymes by covalently modifying the PLP coordinated in the active site (16, 17, 19, 20). In total, the resulting damage led to metabolic disruptions that were defined through biochemical and phenotypic analyses of S. enterica. Those analyses revealed that the consequence of 2AA accumulation in S. enterica is a partial inactivation of at least the PLP-dependent enzymes serine hydroxymethyltransferase (GlyA; EC 2.1.2.1), branched-chain amino acid aminotransferase (IlvE; EC 2.6.1.42), and alanine racemase (Alr/DadX; EC 5.1.1.1). A summary of the RidA paradigm of 2AA stress in S. enterica is depicted in Fig. 1.

FIG 1.

RidA paradigm in S. enterica. Various fold type II PLP-dependent enzymes catalyze β-elimination reactions of 3-carbon (C₃) alpha amino acids with favorable leaving groups on the beta-carbon (1). These reactions proceed through the reactive enamine intermediate 2AA (2). 2AA can tautomerize to its imine form, producing 2-iminiopropanoate (3), which then is nonenzymatically hydrolyzed into the more stable keto acid pyruvate (4). A subset of these enzymes releases 2AA from the active site. RidA catalyzes the deamination of 2AA and other enamines in vivo. It has been demonstrated that the elimination of RidA leads to accumulated free 2AA, which can covalently modify and inactivate a number of PLP-dependent enzymes that belong to the fold type I, III, and IV families.

Collectively, data from S. enterica showed that the primary source of free 2AA was serine/threonine dehydratase (IlvA; EC 4.3.1.19), and the critical target that resulted in growth defects was GlyA (19, 21). The elimination of RidA homologs from organisms including Lycopersicon esculentum, Arabidopsis thaliana, Zea mays, and Saccharomyces cerevisiae has been linked to a range of metabolic and growth perturbations (22–25). The growing list of metabolic defects and growth phenotypes associated with organisms lacking RidA function is consistent with an ancient stress system that is modulated by RidA proteins. Growth defects caused by the accumulation of 2AA reflect the response of the metabolic network to multiple perturbations and provide a means to understand metabolic network architecture. The enzymes of the 2AA stress paradigm in S. enterica, both the generators and targets, are present in E. coli. Despite the conservation of components, the results here showed that the metabolic networks of the two organisms have dramatically different abilities to generate and respond to 2AA.

RESULTS

S. enterica and E. coli respond differently to the loss of ridA.

A number of nutritional and biochemical phenotypes caused by the accumulation of 2AA have been described for mutants of S. enterica lacking RidA. None of these growth defects were manifest in an E. coli K-12 strain lacking RidA. Besides RidA (b4243), E. coli contains two additional Rid superfamily members, which are not present in S. enterica, TdcF (b3113) and RutC (b1010). An E. coli strain lacking these three RidA family members (ridA tdcF rutC) similarly failed to show any growth defects associated with 2AA accumulation (Fig. 2). In addition to growth phenotypes, the activity of branched-chain amino acid aminotransferase (IlvE; EC 2.6.1.42) is used as a proxy for 2AA accumulation in vivo, since 2AA inactivates this enzyme (17, 22, 26). A ridA mutation decreased the activity of IlvE 4-fold in S. enterica (73 ± 9 nmol 2-ketomethylvalerate [2-KMV] min⁻¹ mg⁻¹ protein versus 18 ± 5 nmol 2-KMV min⁻¹ mg⁻¹ protein [26]), yet the IlvE activity in an E. coli triple mutant (ridA tdcF rutC) was indistinguishable from that of the wild-type strain (180 ± 7 nmol 2-KMV min⁻¹ mg⁻¹ protein versus 179 ± 2 nmol 2-KMV min⁻¹ mg⁻¹ protein) when cells were grown in minimal glucose medium. Together, these results showed that there was a significant difference in 2AA metabolism between S. enterica and E. coli.

FIG 2.

S. enterica and E. coli respond distinctly to the loss of RidA proteins. Growth of S. enterica (squares) and E. coli (triangles) strains is shown. Strains are the wild-type (filled symbols) and ridA (S. enterica) or ridA tdcB rutC (E. coli) (open symbols) strains grown in glucose (11 mM) medium with serine (5 mM) (A), pyruvate (50 mM) medium (B), and glucose medium with l-alanine (15 mM) (C) as the sole nitrogen source. Media in panels A and B included ammonia as the nitrogen source. All strains were grown at 37°C, and data are the means of results from three biological replicates.

There are two simple possibilities for the phenotypic difference between S. enterica and E. coli lacking RidA proteins: differences in metabolic network architecture or different component compositions. To distinguish between these possibilities, a hybrid genome was constructed. The glyA and ilvE loci in S. enterica were replaced with the respective open reading frames (ORFs) from E. coli using linear recombination techniques (27). S. enterica strains carrying each of the E. coli loci were generated and confirmed. A ridA mutation was then introduced into each of the strains, generating two pairs of strains isogenic at the ridA locus. Without exception, the ridA derivative strains had the 2AA-associated phenotypes characterized for S. enterica (19, 28). Specifically, these strains were unable to grow in the presence of serine or on pyruvate as a carbon source unless glycine or isoleucine was provided (data not shown). Furthermore, the activity of IlvE was assayed in S. enterica strains carrying the E. coli ilvE locus. The strain lacking ridA had a 4-fold decrease in IlvE activity relative to that of the wild type (11 ± 1 versus 47 ± 9 nmol 2-KMV/min⁻¹ mg⁻¹), as is typical of ridA mutants of S. enterica with a complete complement of Salmonella genes (26). These data supported a role for metabolic networks, not component composition, in determining the phenotypic outcomes in each organism.

An inducible serine/threonine dehydratase system modulates 2-aminoacrylate production.

A simple explanation for the lack of a RidA-like phenotype in E. coli was that less 2AA was produced and/or free in the cell. A construct that allowed the inducible production of 2AA was generated to test this possibility. The ilvA219 allele was cloned under the control of the araBAD promoter (P_BAD) and inserted into the chromosome, replacing the araBAD locus in relevant strains of both S. enterica and E. coli. The ilvA219 allele of S. enterica encodes a variant (IlvA_L447F) that is insensitive to inhibition by isoleucine (29). This construct (ABcc1) allowed the manipulation of both the timing and relative amount of 2AA production in vivo. Control experiments with S. enterica showed that without the induction of the construct, strains lacking RidA had a minor growth defect characteristic of ridA mutants. This growth defect was eliminated when exogenous isoleucine was provided, due to the well-characterized allosteric inhibition of native IlvA. As expected, the induction of ilvA219 expression resulted in a growth defect that persisted despite the addition of isoleucine (Fig. 3). Importantly, glycine restored full growth to the ridA strain, inducing the expression of ilvA219, confirming that growth limitation was caused by an increased accumulation of 2AA (see Fig. 5A) (19).

FIG 3.

The level of endogenously generated 2AA is increased with the inducible IlvA system. S. enterica strains DM15035 (ABcc1) and DM15036 (ridA ABcc1) (boxes and open triangles, respectively) were grown in 20 mM glycerol. Strain DM15036 was also grown in 20 mM glycerol plus 1 mM isoleucine (closed triangles), plus 0.2% l-arabinose (open diamonds), and plus 0.2% l-arabinose and 1 mM isoleucine (closed diamonds). Data are the means of results from experiments performed in biological triplicate.

FIG 5.

Aspartate addition overcomes the 2AA-dependent growth defect in E. coli. (A) Growth of S. enterica strains DM15035 (ABcc1) (shaded symbols) and DM15036 (ridA ABcc1) (open symbols) in minimal glycerol medium containing 0.2% l-arabinose (triangles), 0.2% l-arabinose and 1 mM aspartate (squares), or 0.2% l-arabinose and 1 mM glycine (diamonds). (B) Growth of E. coli strains DM14931 (ABcc1) (shaded symbols) and DM15055 (ridA tdcF ABcc1) (open symbols) in minimal glycerol medium with 0.2% l-arabinose. Medium was supplemented with 5 mM serine (triangles), 5 mM serine and 1 mM glycine (diamonds), or 5 mM serine and 1 mM aspartate (squares).

The increase in threonine dehydratase activity caused by inducing its expression from ABcc1 in E. coli was quantified. The addition of 0.2% (wt/vol) arabinose to the growth medium of DM14931 (ABcc1) increased the IlvA enzyme activity in crude extracts more than 5-fold (1,071 ± 10 versus 184 ± 57 nmol 2-ketobutyrate [2-KB] min⁻¹ mg⁻¹). Additionally, when the cells were grown without arabinose, 100% of the activity was sensitive to isoleucine, consistent with the source of activity being native IlvA. In contrast, with arabinose in the growth medium, >75% of the activity was retained in the presence of isoleucine, consistent with the majority of the activity being from the IlvA_L447F variant encoded by ABcc1. These data confirmed that the induction of ilvA219 increased the serine/threonine dehydratase activity in E. coli and allowed the conclusion that the 2AA level was similarly increased.

RidA or TdcF is sufficient to protect E. coli from 2AA damage.

E. coli strains carrying the ABcc1 construct and lacking the three RidA homologs (ridA tdcF rutC) or wild type at these loci were assessed for growth phenotypes in minimal glycerol medium. In all cases, the expression of ilvA219 was induced with arabinose. The data in Fig. 4 show that the induction of ilvA219, and the resulting increase in 2AA levels, compromised the growth of DM14994 (ridA tdcF rutC). The exogenous addition of serine (5 mM) further compromised growth (Fig. 4), while the addition of l-threonine (1 mM) restored growth comparable to that of the uninduced control (data not shown). This growth pattern was reminiscent of a ridA mutant of S. enterica (30, 31) and suggested that when sufficient 2AA was present, E. coli was susceptible to growth inhibition, as was S. enterica.

FIG 4.

RidA and TdcF can quench 2AA in E. coli. E. coli strains DM14931 (ABcc1) (boxes), DM15055 (ridA tdcF ABcc1) (open triangles), and DM14994 (ridA tdcF rutC ABcc1) (diamonds) were grown in 20 mM glycerol plus 0.2% l-arabinose. Strain DM14994 (ridA tdcF rutC ABcc1) was also grown in 20 mM glycerol plus 0.2% l-arabinose and 5 mM serine (filled triangles). Data are the means of results from experiments performed in biological quadruplicate.

Strains containing the inducible ilvA219 allele (ABcc1) and all combinations of ridA, tdcF, and rutC single and double gene deletions were generated and assessed by using glycerol medium. Strain DM15055, the ridA tdcF double mutant, had a growth defect similar to that of strain DM14944, the ridA tdcF rutC triple mutant (Fig. 4). The presence of neither the ridA, rutC, or tdcF single mutation nor the ridA rutC or tdcF rutC double mutation had any impact on the growth of the strain with the ABcc1 construct (data not shown). The expression of either ridA or tdcF in trans eliminated the growth defect of ridA tdcF mutant strain DM15055 (data not shown). These data allowed the conclusion that RidA and TdcF were at least partially redundant under the growth conditions tested.

Branched-chain amino acid aminotransferase (IlvE) activity was measured in strains with the ABcc1 construct as a quantitative proxy for the accumulation of 2AA. When the cells reached the mid-logarithmic growth phase (optical density at 650 nm [OD₆₅₀] of ∼0.6), l-arabinose was added, and the cells were grown for an additional hour before IlvE activity was assayed. The data showed that IlvE activities in the ridA, tdcF, and wild-type strains were not significantly different (P < 0.05) (189 ± 8, 183 ± 20, and 212 ± 1 nmol 2-KMV min⁻¹ mg⁻¹). However, for the ridA tdcF double mutant, aminotransferase activity was decreased by ∼35% (140 ± 18 versus 212 ± 1 nmol 2-KMV min⁻¹ mg⁻¹). In the absence of arabinose, this decrease in activity did not occur (data not shown). Consistent with the growth results shown in Fig. 4, the assay results indicated that 2AA had metabolic consequences in E. coli only when both RidA and TdcF were absent. The lack of 2AA accumulation in the absence of the inducible ilvA219 construct suggested that E. coli might naturally have lower serine/threonine dehydratase activity than S. enterica. However, when wild-type strains of S. enterica and E. coli were grown in minimal glucose medium, serine/threonine dehydratase activities measured in crude extracts were not significantly different (300 ± 70 versus 290 ± 39 nmol 2-ketobutyrate/min/mg). Based on these data, a more complex explanation for the lack of 2AA damage in E. coli tdcF ridA strains was warranted.

2AA generates distinct phenotypic signatures in S. enterica and E. coli.

To better understand the metabolic effect of 2AA accumulation on E. coli, metabolites were screened for the ability to rescue the growth of DM15055 (ABcc1 ridA tdcF). The data in Fig. 5 summarize the results. In contrast to the results with the S. enterica ridA mutants (19), glycine had no effect on the growth defect caused by 2AA in the E. coli ABcc1 ridA tdcF strain. Exogenous aspartate (1 mM) restored the growth of the E. coli strain but had no effect on the growth of the S. enterica ridA strain.

Aspartate aminotransferase (AspC; EC 2.6.1.1) is a PLP-dependent enzyme that catalyzes the formation of aspartate from oxaloacetate. AspC shares characteristics of the three PLP-dependent enzymes that are targeted by free 2AA in S. enterica, GlyA, IlvE, and Alr (21, 26, 32). Specifically, each of these four enzymes has reduced specific activity when challenged in vitro with 3-chloroalanine (3CA) (33–35), due to the generation of 2AA in the active site, which covalently modifies the PLP cofactor in the active site and renders the enzyme inactive (36). The effect of aspartate on the growth phenotype of the ridA tdcF strain expressing ilvA219 suggested that the cells could be limited for aspartate. AspC activity was measured in crude extracts of DM15035 (S. enterica [ABcc1]), DM15036 (S. enterica [ridA ABcc1]), DM15055 (E. coli [ridA tdcF ABcc1]), and DM14931 (E. coli [ABcc1]) (Fig. 6). AspC activities were not statistically different (P < 0.01) between the S. enterica ridA mutant and the wild type. However, when IlvA_L447F expression was induced with arabinose in the growth medium, AspC activity in DM15036 (ridA ABcc1) was significantly decreased compared to that of the isogenic wild-type strain (327 ± 21 versus 493 ± 25 nmol NAD⁺ min⁻¹ mg⁻¹) (Fig. 6A). A similar result was observed with E. coli strains DM14931 (ABcc1) and DM15055 (ridA tdcF ABcc1), where the AspC activities in the two strains were indistinguishable in the absence of arabinose. As with the S. enterica strains, when arabinose was present in the growth medium, the ridA tdcF mutant strain (DM15055) had significantly less AspC activity than did the wild-type parental strain (DM14931) (877 ± 43 versus 1,385 ± 40 nmol NAD⁺ min⁻¹ mg⁻¹, respectively) (Fig. 6B).

FIG 6.

2AA decreases the activity of aspartate aminotransferase in vivo. (A) Aspartate transaminase (AspC) activity was measured following growth of S. enterica strains DM15035 (ABcc1) (gray bars) and DM15036 (ridA ABcc1) (white bars) to late logarithmic phase (OD₆₅₀ of 0.9 to 1.0) in minimal glycerol (20 mM) medium containing 0.67 mM glycine (solid bars) or minimal glycerol medium containing glycine and 0.2% l-arabinose (hatched bars). (B) AspC activity was also measured following growth of E. coli strains DM14931 (ABcc1) (gray bars) and DM15055 (ridA ABcc1) (white bars) to the late logarithmic phase (OD₆₅₀ of 0.9 to 1.0) in minimal glycerol (20 mM) medium (solid bars) or minimal glycerol medium containing 0.2% l-arabinose (hatched bars). In all cases, aminotransferase activity was measured in crude extracts by coupling oxaloacetate formation to malate dehydrogenase activity and monitoring NADH oxidation at 340 nm from three biological replicates. Error bars represent the standard errors of the means from replicates; asterisks denote statistically significant (P < 0.01) variation between samples.

Collectively, the above-described data indicated that the accumulation of 2AA resulted in damage to AspC in vivo. Two results suggested that low AspC activity was not responsible for the reduced growth of an E. coli strain lacking RidA and TdcF. First, the addition of tyrosine did not exacerbate this growth defect (data not shown). The products of both aspC and tyrB catalyze the conversion of oxaloacetate to aspartate, and thus, aspartate aminotransferase activity would be further reduced by tyrosine due to the allosteric inhibition of TyrB (37). Second, while aspC expressed in trans restored the growth of an E. coli aspC mutant, it failed to restore the growth of the ridA tdcF mutant (data not shown).

Aspartate and adenine ameliorate the consequences, not the source, of 2AA stress.

It was found serendipitously that the exogenous addition of the purine adenine (0.4 mM), guanine (0.133 mM), or hypoxanthine (0.4 mM) restored growth to an E. coli ridA tdcF ABcc1 mutant strain under 2AA stress (Fig. 7 and data not shown). The addition of pyrimidines failed to rescue growth (data not shown). The ability of adenine, and other purines, to restore growth was dependent on functional PurR (Fig. 7). Significantly, a purR mutation had no effect on growth stimulation by aspartate (data not shown). PurR is the main transcriptional regulator involved in de novo purine synthesis and also participates in the regulation of pyrimidine biosynthesis and one-carbon metabolism (38). The data are consistent with a general model in which a gene in the PurR regulon interferes with the growth of the ridA tdcF ABcc1 mutant strain by a direct or indirect mechanism. As with aspartate, exogenous purines did not affect the growth of the S. enterica ridA mutant, a further indication of the difference in the 2AA metabolic networks between the two organisms (data not shown).

FIG 7.

Purines restore growth via a PurR-dependent mechanism. E. coli strains DM14931 (ABcc1) (black symbols), DM15055 (ridA tdcF ABcc1) (white symbols), and DM15832 (ridA tdcF ABcc1 purR) (gray symbols) in minimal glycerol medium with 0.2% l-arabinose. Medium was supplemented with 5 mM serine (triangles) or 5 mM serine and 0.4 mM adenine (circles). All strains were grown at 37°C, and data are presented as the means of results from experiments performed in biological triplicate.

In principle, the restored growth of strains compromised by 2AA stress can reflect a decreased production of 2AA or a metabolic restructuring that ameliorates the consequences of 2AA damage. IlvE activity was used as a proxy for 2AA accumulation to distinguish these possibilities. IlvE activity was measured in strains DM14931 (ABcc1) and DM15055 (ABcc1 ridA tdcF) after growth in medium containing 0.2% arabinose alone or with the addition of threonine, aspartate, or adenine. The data are shown in Table 1. As expected, a strain lacking ridA and tdcF had an ∼33% reduction in IlvE activity when grown in minimal medium with arabinose. The addition of threonine, aspartate, or adenine restored growth but had disparate effects on the activity of IlvE. In the case of threonine, IlvE activity was restored. As characterized for S. enterica, this effect is due to threonine competing with serine for the active site of IlvA and reducing 2AA production. Significantly, IlvE activity was not increased by the presence of either aspartate or adenine in the growth medium. If anything, these additions appeared to increase the discrepancy between the strains with and those without RidA activity. These data indicated that the latter two supplements restored growth by circumventing the consequences of 2AA damage and not by reducing the damage. In total, these data suggested that purines (via PurR) and aspartate supported metabolic architectures that allow growth in the presence of 2AA.

TABLE 1.

Aspartate and adenine restore growth without reducing 2AA accumulation^b

Supplement	Strain	Relevant genotypea	Mean IlvE activity (nmol 2-KMV min−1 mg−1) ± SD	Rid−/Rid+ ratio
None	DM14931	Wild type	148 ± 23	0.70
	DM15055	ridA tdcF	102 ± 5
Threonine	DM14931	Wild type	279 ± 26	0.96
	DM15055	ridA tdcF	269 ± 13
Aspartate	DM14931	Wild type	174 ± 18	0.63
	DM15055	ridA tdcF	109 ± 2
Adenine	DM14931	Wild type	183 ± 18	0.54
	DM15055	ridA tdcF	98 ± 8

^aAll strains include the chromosomal construct ABcc1 that has the ilvA219 allele under the control of the arabinose-inducible P_BAD promoter.

^bThe indicated strains were grown in minimal glycerol medium containing the indicated supplements at a 1 mM concentration to an OD₆₅₀ of ∼0.6. Arabinose was added to a final concentration of 0.2%, and cultures were grown for an hour before pelleting of cells. Branched-chain amino acid aminotransferase activity (IlvE) was measured in crude extracts. Data represent the means and standard deviations of results from experiments performed in biological triplicate, measuring the rate of formation of 2-KMV. The Rid⁻/Rid⁺ ratio reflects the ratio of IlvE activity in strains without RidA and TdcF to that in the wild-type parent and is a measure of 2AA accumulation.

DISCUSSION

A comparison of the S. enterica and E. coli K-12 genomes showed that ∼80% of the protein-coding genes in S. enterica have a highly conserved (>90% amino acid identity) orthologous partner in E. coli and defined their shared core genome (39, 40). Based on this genomic similarity, and supported by numerous studies over decades, these organisms were presumed to share significant metabolic potential and organization. Recent phenotypic studies have highlighted the need to consider this assumption more carefully, potentially on a case-by-case basis (5, 6, 12, 41–43). This study was initiated to evaluate the conservation of metabolic network structures surrounding 2AA metabolism (Fig. 1) between S. enterica and E. coli. Importantly, each of the enzymatic components in the 2AA metabolic network in S. enterica had an orthologous partner in E. coli that shared >85% identity at the amino acid level (Table 2). The paradigm defined in S. enterica that articulated the causes and consequences of 2AA stress provided a means to assess the conservation of a higher-order metabolic network architecture that was impacted by and responded to stress generated by endogenous metabolites.

TABLE 2.

The 2AA subnetwork components are conserved between S. enterica and E. coli

Function	Protein	% amino acid identitya
2AA generator	IlvA	95.5
2AA quencher	RidA	93.8
RidA homologs (absent in S. enterica)	TdcF	73.5 (to S. enterica RidA)
	RutC	28.8 (to S. enterica RidA)
2AA targets	IlvE	97.1
	GlyA	93.3
	Alr	91.4
	DadX	85.7

^aAmino acid sequences were obtained from the KEGG database (57), and protein alignments and percent amino acid identity calculations were performed by using the EMBL-EBI Clustal Omega online analysis tool (59).

The results of this study showed that S. enterica and E. coli have distinct metabolic networks associated with 2AA, despite sharing component enzymes. Specifically, the networks differ in (i) the generation and/or accumulation of 2AA and (ii) the consequence that 2AA accumulation has on the function of cellular metabolism. The data showed that an E. coli strain, made to resemble an S. enterica ridA mutant through the elimination of RidA, TdcF, and RutC, had no growth defects that paralleled those defined for the S. enterica ridA mutant. This result allowed the conclusion that the native network of E. coli did not normally generate damaging levels of 2AA, even in the presence of the 2AA precursor serine. However, when the E. coli strain was manipulated to increase 2AA production 5-fold, a significant growth defect and damage to isoleucine transaminase (IlvE) were generated. Despite being qualitatively similar, the growth defects in S. enterica and E. coli were shown to be mechanistically distinct.

Significantly, the generation of a 2AA-dependent growth defect in E. coli required not only increased IlvA activity but also the absence of both RidA and TdcF. The latter result showed that the roles of these two proteins in the cell are at least partially redundant. A TdcF-dependent phenotype was unexpected since the tdc operon was shown to be transcriptionally silent under the growth conditions used (44, 45). This result suggests that there may be an unappreciated role for this operon in the cell.

There are several possible explanations for the difference in free 2AA generated by the native metabolic networks of E. coli and S. enterica. Like many PLP-dependent enzymes, IlvA is promiscuous and can use either threonine or serine as a substrate. The former is the “true” substrate and leads to isoleucine synthesis with aminocrotonate as the reaction intermediate. Aminocrotonate is not as reactive as serine-derived 2AA and has not been shown to be detrimental in vivo or in vitro. In contrast, 2AA can damage at least 4 cellular enzymes in vivo in the absence of RidA (21, 26, 32). The IlvA proteins from E. coli and S. enterica are 95.5% identical, differing by 23/514 amino acids, making it unlikely that the specificity of one is dramatically different from that of the other. Rather, we favor a scenario where a property of the cellular milieu differs between the organisms. Possibilities that could account for the different generation/accumulation of 2AA include the serine/threonine ratio and an increased internal pH leading to a longer half-life of 2AA, among others. For instance, increased relative levels of threonine in E. coli could minimize the occupation of the active site by serine, leading to less 2AA production. Such a possibility is consistent with previous work by Bazurto et al. Those authors showed that when comparing S. enterica and E. coli, the latter had a metabolic capability that suggested an increased level of aminocrotonate in E. coli compared to S. enterica (6).

The distinct growth responses of E. coli and S. enterica to 2AA accumulation raise additional provocative questions about the cellular environment and how metabolic effects are modulated. The enzymes targeted by 2AA are >90% identical between the organisms, and their functions in vitro are often indistinguishable. Interestingly, even when the impact on these enzymes was made equal, the consequences on the growth and function of the metabolic network remained different. By using IlvE as a proxy for 2AA, the general level of 2AA can be equated in the two organisms; however, the metabolic additions that overcome the growth defect were distinct.

Previous work showed that the critical target of 2AA for causing growth defects in S. enterica was serine hydroxymethyltransferase (GlyA) (19, 21), while the work here showed that GlyA is not the critical 2AA target in E. coli. Furthermore, the data indicate that the AspC enzymes from both organisms can be damaged by 2AA. In a catalog of PLP-dependent enzymes, AspC and GlyA were the only EC-classified PLP-dependent enzymes found in the genomes of all free-living organisms (46). Rid family members are also conserved across all domains of life (47). Collectively, the biochemical results here for S. enterica and E. coli showed that both of the most widely conserved EC-classified activities (GlyA and AspC) could serve as important targets of free 2AA, consistent with the idea that RidA family members are part of an ancient stress response system.

In addition to the initial description of the RidA stress response in E. coli, this work highlighted the subtleties involved in defining the metabolic network based on the presence of component enzymes. The data here suggest that the 2AA stress system provides a model to explore the causes and consequences of subtle differences in the cellular environment. They also provide a system that is amenable to mathematical modeling to define the impact that small changes in substrate concentrations and/or enzyme kinetics can have on a defined complex system. Ultimately, this work contributes to our ability to access the next level of information from genome sequences and predict network structure and flexibility.

MATERIALS AND METHODS

Bacterial strains, media, and chemicals.

The strains used in this work are derivatives of Salmonella enterica serovar Typhimurium LT2 or Escherichia coli K-12 unless otherwise noted and are listed in Table 3. Minimal medium was no-carbon E (NCE) medium supplemented with 1 mM MgSO₄ (48); trace minerals (49); and 11 mM d-glucose, 50 mM pyruvate, or 20 mM glycerol as the sole carbon source. Luria-Bertani (LB) broth and Difco nutrient broth (NB) (8 g/liter) containing NaCl (5 g/liter) were used as rich medium. Difco BiTek agar (15 g/liter) was added for solid medium. The following antibiotics were used for rich (or minimal) medium: kanamycin at 50 μg/ml (or 12.5 μg/ml), chloramphenicol at 20 μg/ml (or 5 μg/ml), and ampicillin at 150 μg/ml (or 7.5 μg/ml). All chemicals were obtained from Sigma (St. Louis, MO).

TABLE 3.

Bacterial strains and plasmids

Strain or plasmid	Genotype or description	Source or reference
Strains
DM14520	E. coli wild type (K-12 MG1655 F− λ− rph-1)	Laboratory collection
DM14581	E. coli ilvA723::kan	This study
DM14597	E. coli ridA790::kan	This study
DM14601	E. coli DH5α/pAB1	This study
DM14705	E. coli ΔridA890 tdcF13::cat rutC754::kan	This study
DM14712	E. coli DH5α/pAB2	This study
DM14930	E. coli ilvA723::kan ara::PBAD-ilvA219 catb	This study
DM14931	E. coli ara::PBAD-ilvA219 cat	This study
DM14949	E. coli ΔridA890 ara::PBAD-ilvA219 cat	This study
DM14994	E. coli ΔridA890 ΔtdcF14 ΔrutC854 ara::PBAD-ilvA219 cat	This study
DM15049	E. coli rutC754::kan ara::PBAD-ilvA219 cat	This study
DM15050	E. coli ΔtdcF14 ΔrutC854 ara::PBAD-ilvA219 cat	This study
DM15055	E. coli ΔridA890 ΔtdcF14 ara::PBAD-ilvA219 cat	This study
DM15077	E. coli tdcF12::kan ara::PBAD-ilvA219 cat	This study
DM15078	E. coli ΔridA890 rutC754::kan ara::PBAD-ilvA219 cat	This study
DM15832	E. coli ΔridA890 ΔtdcF14 ara::PBAD-ilvA219 cat purR746::kan	This study
DM9404	S. enterica LT2 (wild-type)	Laboratory collection
DM15035	S. enterica ara::PBAD-ilvA219 cat	This study
DM3480	S. enterica ridA3::MudJa	Laboratory collection
DM15036	S. enterica ridA3::MudJ ara::PBAD-ilvA219 cat	This study
Plasmids
pAB1	pBAD24-IlvAL447F	This study
pAB2	pBAD24-IlvAL447F Cat	This study

^aMudJ refers to the MudJ1734 transposon (58). Plasmid constructs used pBAD24 (52) and are described in Materials and Methods.

^bFor simplicity, this construct is designated ABcc1 in the text.

Genetic techniques.

Transductional crosses in E. coli were performed with bacteriophage P1vir according to previously described protocols (6, 50). In general, E. coli strains were constructed by transducing the appropriate mutations from the Keio collection into the appropriate strain background, followed by phenotypic confirmation (51). The tdcF mutation was made by using the λ-Red recombinase system described previously by Datsenko and Wanner (27) and transduced into relevant genetic backgrounds by selecting for kanamycin resistance. When multiple mutations were present in the same genetic background, the kanamycin cassette was resolved by using a previously described method (27).

Construction of a chromosomal construct for inducing 2AA formation.

The ilvA219 allele, encoding the feedback-resistant variant IlvA_L447F, was amplified from S. enterica by PCR with Q5 high-fidelity DNA polymerase (New England BioLabs) using primers AB1 and AB2 (Table 4). The amplification product was purified, digested with EcoRI and PstI (New England BioLabs), and ligated into predigested pBAD24 (52) to generate pAB1. The cat gene, encoding chloramphenicol acetyltransferase, including the promoter, was amplified from pSU18 (53) by PCR using Q5 high-fidelity DNA polymerase (New England BioLabs) and primers AB3 and AB4. The product was digested with HindIII (New England BioLabs) and ligated into predigested pAB1. The ligation mixture was transformed into E. coli DH5α, and chloramphenicol-resistant (Cm^r) colonies were selected. Plasmids from the Cm^r clones were isolated, the insert was confirmed by PCR, and the resulting plasmid was named pAB2. The P_BAD-ilvA219 cat insert from pAB2 was amplified by using primer pair AB5/AB6 or AB5/AB7 and replaced the araBAD genes from E. coli K-12 and S. enterica LT2 by using the λ-Red recombinase system (27). The chromosomal construct was confirmed by using a combination of PCR amplification and phenotypic analysis. The latter demanded chloramphenicol-resistant colonies that were able to grow on minimal medium containing arabinose but not on minimal medium lacking arabinose in a strain where the construct encoded the only source of IlvA in the cell.

TABLE 4.

Primers

Primer	Sequence
AB1	GAGAGAATTCATGGCGGAATCTCAACCTCT
AB2	GAGACTGCAGTTAACCCGCCAGAAAG
AB3	TAGAAGCTTGATCGGCACGTAAGAGG
AB4	TAGAAGCTTACGCCCCGCCCTGCCA
AB5	AGCCATGACAAAAACGCGTAAC
AB6	AGCCTGGTTTCGTTTGATTGGCTGTGGTTTTATACAGTCACCGCCAAAACAGCCAAGCT
AB7	TTCATCAACGCGCCCCCCATGGGACGTTTTTAGAGGCACCGCCAAAACAGCCAAGCT

Quantification of growth.

A 2-μl aliquot of cells from an LB culture grown overnight was used to inoculate 198 μl of growth medium. Ninety-six-well plates were incubated at 37°C in a microplate reader (model EL808; Bio-Tek Instruments) with shaking, and growth was monitored as the change in the OD₆₅₀ over time. Unless stated otherwise, growth experiments were performed with biological triplicates. Results were plotted by using GraphPad Prism 6.0b to generate curves (log₁₀ format), represented as a composite of the averages and standard deviations for the replicates.

Threonine dehydratase (IlvA) assays.

LB broth cultures grown overnight were inoculated into 200 ml minimal medium (with the stated additions) and grown at 37°C to late logarithmic phase (OD₆₅₀ of ∼0.8) before being stored as pellets at −80°C until use. Cell pellets were resuspended and pelleted twice, using 5 ml of 50 mM KPO₄ (pH 7.2), before being resuspended in 5 ml of 50 mM KPO₄ (pH 7.2) with 0.4 mM dithiothreitol. Cell extracts were prepared with a French pressure cell (1,500 lb/in²) and clarified by centrifugation (1 h at 40,000 × g at 4°C). Threonine dehydratase assays were carried out as described previously (22, 54), and results are reported as nanomoles of 2-ketobutyrate formed per milligram of protein per minute. Protein concentrations were estimated by using the Bradford method (55). Each experiment included three biological replicates, and the averages with standard deviations are reported.

Branched-chain amino acid aminotransferase (IlvE) assays.

Three LB cultures grown overnight were used to inoculate (1% inoculum) 5 ml of minimal glycerol medium with the indicated additions. Following growth, cells were pelleted, resuspended twice by using NCE medium, pelleted, and frozen at −20°C until use. Cell pellets were thawed and assayed as described previously (22). Briefly, cells were permeabilized in 50 mM potassium phosphate (pH 8) containing 50 μM PLP, 10 mM 2-ketoglutarate, and 10% PopCulture reagent (Novagen). The reaction was initiated by the addition of 20 mM l-isoleucine. The product 2-KMV was derivatized by 2,4-dinitrophenylhydrazine (DNPH) to yield a chromophore absorbing at 540 nm. The 2-KMV level was quantified with a standard curve generated from known quantities of 2-KMV similarly derivatized with DNPH and normalized to the total protein content estimated by the Bradford method (55). The significances of IlvE activity differences were determined by analyzing the data using two-way analysis of variance (ANOVA) in GraphPad Prism 6.0b and demanding a P value of <0.01.

Aspartate aminotransferase (AspC) assays.

One-hundred-milliliter cultures were grown in minimal glucose medium (with the indicated additions) to early stationary phase (OD₆₅₀, ∼1.2) and stored as cell pellets at −80°C. Cell pellets were resuspended in 7 ml of a solution containing 25 mM KPO₄ (pH 7.5), 0.1 mM EDTA, 0.2 mM pyridoxal 5′-phosphate, 0.2 mM dithiothreitol, and 5% glycerol. Cell extracts were prepared from the resuspended cell pellets by using a French pressure cell (1,500 lb/in²) and clarified by centrifugation (1 h at 40,000 × g at 4°C). Aminotransferase activity was measured in the reverse direction by using a coupled assay in which the oxaloacetate formed following the addition of aspartate was reduced by malic dehydrogenase to malate using NADH as an electron acceptor. NADH oxidation was monitored at 340 nm (ε₃₄₀ = 6,200 M⁻¹ cm⁻¹) (56). Protein concentrations were determined by using the bicinchoninic acid (BCA) assay (Pierce). AspC activity is reported as nanomoles of NADH oxidized per milligram of protein per minute. The data are presented as the averages from three independent experiments, with standard deviations. Statistical significance (P < 0.01) was analyzed by performing one-way ANOVA and Tukey's post hoc test in GraphPad Prism 6.0b.