Lethality of MalE-LacZ hybrid protein shares mechanistic attributes with oxidative component of antibiotic lethality

Significance

Understanding the molecular basis of the lethality of antibiotics and certain other stresses is complicated because cell death can result from direct inhibition of a critical biological process as well as from reactive oxygen species (ROS) generated by events metabolically downstream of the direct interaction of the agent with its target. Prior evidence has indicated that the ROS-dependent component of antibiotic lethality is due in part to lethal DNA problems resulting from the incorporation of oxidized nucleotides into DNA and incomplete DNA repair. Our observations unexpectedly indicate that the predominant mechanism of lethality from a hybrid protein that jams the machinery that translocates proteins across the cytoplasmic membrane shares attributes with the ROS-dependent component of antibiotic lethality.

Abstract

Downstream metabolic events can contribute to the lethality of drugs or agents that interact with a primary cellular target. In bacteria, the production of reactive oxygen species (ROS) has been associated with the lethal effects of a variety of stresses including bactericidal antibiotics, but the relative contribution of this oxidative component to cell death depends on a variety of factors. Experimental evidence has suggested that unresolvable DNA problems caused by incorporation of oxidized nucleotides into nascent DNA followed by incomplete base excision repair contribute to the ROS-dependent component of antibiotic lethality. Expression of the chimeric periplasmic-cytoplasmic MalE-LacZ_72–47 protein is an historically important lethal stress originally identified during seminal genetic experiments that defined the SecY-dependent protein translocation system. Multiple, independent lines of evidence presented here indicate that the predominant mechanism for MalE-LacZ lethality shares attributes with the ROS-dependent component of antibiotic lethality. MalE-LacZ lethality requires molecular oxygen, and its expression induces ROS production. The increased susceptibility of mutants sensitive to oxidative stress to MalE-LacZ lethality indicates that ROS contribute causally to cell death rather than simply being produced by dying cells. Observations that support the proposed mechanism of cell death include MalE-LacZ expression being bacteriostatic rather than bactericidal in cells that overexpress MutT, a nucleotide sanitizer that hydrolyzes 8-oxo-dGTP to the monophosphate, or that lack MutM and MutY, DNA glycosylases that process base pairs involving 8-oxo-dGTP. Our studies suggest stress-induced physiological changes that favor this mode of ROS-dependent death.

Agents or conditions that inhibit important biological processes can kill cells. However, physiological processes metabolically downstream of the initial inhibition can also contribute to cell death (1). For example, β-lactam antibiotics not only inhibit penicillin-binding proteins leading to lysis; they also induce a futile cycle of cell wall synthesis and degradation that contributes to their killing (2). In another example, lethal attacks on Escherichia coli mediated by the type VI secretion system P1vir phage and the antimicrobial peptide polymyxin B elicit the production of reactive oxygen species (ROS) that contribute to cell death (3). In bacteria, ROS production has been associated with the lethal effects of diverse stresses (4). In most cases, the detailed mechanisms responsible for ROS production are poorly understood; however, a variety of futile metabolic cycles can elicit H₂O₂ production, illustrating the breadth of possible metabolic perturbations that could potentially induce oxidative stress (5).

Despite widespread evidence that endogenous ROS produced as a consequence of metabolic stress can be lethal to bacteria and eukaryotes, the application of this concept to antibiotic lethality has been complicated. Evidence from multiple investigators using a variety of antibiotics to study different bacteria indicated that ROS generated metabolically downstream of the interaction of the antibiotics with their primary cellular targets contribute to drug lethality (1, 6, 7). This conclusion was challenged by critiques (8, 9) that focused particularly on an earlier paper that had carried out a systems-level analysis of the lethality of multiple classes of bactericidal antibiotics and proposed a model to account for the experimental observations (10). Substantial new evidence has subsequently been published that addresses concerns that were raised and strongly supports a contributing role for ROS in antibiotic lethality (11, 12). Recent reviews have discussed how the apparently contradictory results can be explained in part by the specifics of the experimental setup and the technical details of the assays used (1, 6, 7).

Notably, the challenges to the involvement of ROS in antibiotic lethality (8, 9, 13) did not consider the implications of prior genetic and physiological evidence that oxidized nucleotides, especially 8-oxo-dGTP, contribute to cell death by bacterial antibiotics (14). Superoxide (O₂⁻) and hydrogen peroxide (H₂O₂) have only limited abilities to react with cellular constituents (15, 16). The oxidation of nucleotides and nucleic acids instead requires the highly reactive hydroxyl radicals and intermediates produced when H₂O₂ undergoes Fenton chemistry, collectively referred to as “Fenton oxidants” (17). It is the amount and target of Fenton oxidants that are relevant for antibiotic lethality, not the intracellular levels of endogenously produced O₂⁻ or H₂O₂. These and other results suggest that the incorporation of oxidized nucleotides into nascent DNA followed by incomplete base excision repair (BER) is an important molecular mechanism that contributes to the ROS-dependent component of antibiotic lethality (11, 12, 14, 18). To date, no alternative interpretation has been suggested for this body of experimental observations.

Since a mode of cell death involving nucleotide oxidation is common to three different classes of bactericidal antibiotics (β-lactams, quinolones, and aminoglycosides), it is plausible that other types of cellular stress induce death by a similar mechanism. We analyzed an historically important lethal stress originally identified by Jon Beckwith and his colleagues during their seminal genetic experiments that defined the highly conserved SecY-dependent protein translocation system (19). We show the lethality caused by the expression of a chimeric MalE-LacZ protein consisting of the NH₂-terminal sequences of MalE (periplasmic maltose-binding protein) joined to a modestly NH₂-terminally truncated LacZ (β-galactosidase) (20) shares attributes with the oxidative component of antibiotic lethality. Our analyses suggest additional physiological parameters that are important for this mode of cell death besides the levels of ROS. The induced changes in gene expression suggest that exposing bacteria to H₂O₂ does not fully recapitulate the intracellular environment caused by antibiotics and other stresses in which this mode of DNA-based oxidative cell death occurs, thereby helping explain some of the past confusion regarding the role of ROS-dependent death in antibiotic lethality.

Results

Induction of the MalE-LacZ Hybrid Protein Kills Cells but Does Not Elicit SecY Degradation.

The MalE-LacZ_72–47 fusion protein encoded by strain PB72-47/MM18 (21) (Table S1), hereafter referred to as “MalE-LacZ,” was isolated using a bacteriophage Mu-based genetic strategy (21). The hybrid protein consists of the first N-terminal 212 amino acids of MalE (53.5% of MalE), followed by a short linker of 11 amino acids derived from Mu, followed by an N-terminally truncated LacZ missing its first 41 amino acids (Fig. 1A). As previously reported (21), induction of the MalE-LacZ fusion by maltose addition to minimal-glycerol medium results in cell killing that begins ≈2 h later (Fig. S1A). Continued incubation of the culture eventually results in resumed cell growth (21) due to the accumulation of variants that have lost the λ transducing phage or have accumulated suppressor mutations. Induction of a LamB-LacZ fusion protein elicits an FtsH-dependent degradation of SecY proposed to contribute to its lethality (22), but induction of MalE-LacZ does not (Fig. S1B), suggesting a different mechanism underlies its lethality.

Table S1.

Strains and plasmids used in this study

Name	Genotype or description	Source
Strains
PB72-47/MM18	MC4100 with λ phage containing MalE-LacZ	(20)
CG10	MM18 ΔmutM	This study
CG11	MM18 ΔmutY	This study
CG12	MM18 ΔmutMΔmutY	This study
BNT338	MM18 ΔrecA	This study
CG13	MM18 ΔrecBΔrecF	This study
CG14	MM18 ΔkatE	This study
CG15	MM18 ΔkatG	This study
CG16	MM18 ΔahpC	This study
CG17	MM18 ΔahpF	This study
CG18	MM18 ΔoxyR	This study
CG19	MM18 ΔsoxR	This study
CG20	MM18 ΔsoxS	This study
CG21	MM18 ΔsodAΔsodB	This study
CG22	MM18 pCA24N	This study
CG23	MM18 pMutT	This study
CG24	MM18 pMutS	This study
CG25	MM18 pAPX	This study
Plasmids
pCA24N	Empty vector used in ASKA collection	(31)
pMutT	ASKA collection MutT	(31)
pMutS	ASKA collection MutS	(31)
pAPX	Ascorbate peroxidase in pZE21	(12)

Fig. 1.

Expression of MalE-LacZ increases expression of an oxidative response gene and requires oxygen for killing. (A) MalE-LacZ fusion protein. N terminus of periplasmic MalE is fused to a truncated but enzymatically active LacZ by a 19-amino acid linker derived from the bacteriophage Mu. (B) Expression of MalE-LacZ by the addition of 0.2% maltose increases the expression of the oxidative response gene soxS as determined by qPCR. (C and D) The induction of MalE-LacZ expression causes bacterial cell death after 2 h when grown aerobically in LB medium (C), whereas under strict anaerobic conditions MalE-LacZ expression is bacteriostatic (D). Data shown represent the mean ± SD with at least three biological replicates. Significant values are *P ≤ 0.05; **P ≤ 0.01.

Fig. S1.

MalE-LacZ expression results in bacterial cell death. (A) Induction of MalE-LacZ by the addition of 0.2% maltose in M63 minimal medium results in loss of cfus beginning after 2 h. (B) The addition of maltose greatly increases the expression of the native MalE protein as well as the MalE-LacZ fusion protein. In contrast to the related fusion LamB-LacZ protein, levels of the transporter protein SecY do not decrease due to expression of the fusion protein. (C) Microscopy images of cells expressing MalE-LacZ. DNA is stained with DAPI. Cells exposed to maltose begin to expand in size and form long chains of partially divided cells at 3 h, becoming more pronounced at 5 h. (D and E) Live-dead staining of MalE-LacZ–expressing bacteria. (E) Dead cells begin to be present at 4 h with almost all cells being stained at 5 h. (D) The bar graph shows quantification of images. Any propidium iodide staining is interpreted as cell death. (Magnification: 1,000×.)

MalE-LacZ Induction Does Not Cause Death Predominantly by Cell Lysis but Induces soxS.

DAPI staining revealed that the elongated cells reported by Bassford et al. (21) upon MalE-LacZ induction are actually short chains of cells in which DNA partitioning and at least partial septum formation appear to have occurred but the cells have not separated (Fig. S1C). Using a live-dead stain to examine the timing of cell death after maltose induction, we observed that although the commitment to cell death and loss of colony-forming ability was evident at 3 h after induction, cells did not begin to exhibit substantial staining until 4 h. Importantly, most of the cell death occurred without cell lysis, suggesting that the lethality resulted from another cause (Fig. S1 D and E). Expression of the MalE-LacZ fusion protein jams translocation machinery, so the precursors of normally secreted proteins still containing their signal peptides accumulate after maltose addition, with some forming cytoplasmic aggregates (23). However, the normal localization of a fraction of translocated proteins (23) suggested that some other type of mechanism might be responsible for cell death. Similar to stress caused by the type VI secretion system, P1vir phage, and polymyxin B (3), we found that induction of the MalE-LacZ protein increases the expression of soxS, a marker of oxidative stress (Fig. 1B).

MalE-LacZ Induction Is Not Lethal in Anaerobic Conditions but Leads to ROS Production in Aerobic Conditions.

To test whether oxidative stress might underlie MalE-LacZ–dependent cell death, we first compared the effect of maltose addition on cell survival under aerobic and strict anaerobic conditions. We used LB medium in these experiments since E. coli cannot grow anaerobically on the glycerol medium used previously (Fig. S1A) (21). Induction of MalE-LacZ under anaerobic conditions is bacteriostatic rather than bactericidal, indicating that molecular oxygen is required for cell death (Fig. 1 C and D).

Using four independent methods, we then tested whether expression of MalE-LacZ under aerobic conditions results in ROS production. First, to directly measure the intracellular concentration of H₂O₂ induced by MalE-LacZ, we used the recently described APX system, which uses a cytoplasmic-expressed variant of ascorbate peroxidase to convert Amplex Red into a readily detectable fluorescent product in an H₂O₂-dependent manner (12). We observed that the intracellular H₂O₂ levels increase over time following MalE-LacZ induction (Fig. 2A). Second, to directly measure the dynamic production of O₂⁻ after maltose addition, we used a miniaturized cytochrome c-based biosensor that has previously been used to detect the release of O₂⁻ from cells treated with bactericidal antibiotics (24). Using this technique, we found that expression of MalE-LacZ results in the release of O₂⁻ as early as 40 min after induction (Fig. 2B). Third, we used an ELISA to show that MalE-LacZ induction leads to increased amounts of 8-oxo-guanine (8-oxo-7,8-dihydroguanine) in DNA (Fig. 2C), a well-established biomarker for oxidative stress (11) that is relevant to experiments described below. Fourth, we used two dyes based on different chemistries that have been widely used for ROS detection in previous studies (3, 12, 25): 3′-(p-hydroxyphenyl) fluorescein (HPF) and 5/6-chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate (CM-H2DCFDA), whose fluorescence has been shown to correlate with O₂⁻ production determined by direct measurement (24). For both dyes, we observed statistically significant increases in fluorescence compared with controls for maltose-induced autofluorescence and morphology changes (26, 27) in the absence of a dye (Fig. 2D). Microscopic images demonstrated this increase in fluorescence was not a flow cytometry artifact (4) caused by autofluorescence or altered cell shape (Fig. S2).

Fig. 2.

MalE-LacZ expression causes the production of ROS. (A) Expression of MalE-LacZ induces the production of H₂O₂ detected by the intracellular enzymatic sensor APX, using Amplex Red fluorescence as an output. (B) Induction of MalE-LacZ leads to an increase in amperometric responses of the cytochrome c superoxide sensor, indicating an increase in the rate of superoxide release. (C) 8-Oxo-dG content of the total DNA determined by an ELISA-based assay. Expression of MalE-LacZ induces a significant increase in 8-oxo-dG content 1 h after induction, decreasing at later time points. Cells were treated with 10 mM H₂O₂ for 60 min as a positive control. (D) Expression of MalE-LacZ induces ROS detected by two different fluorescent dyes, CM and HPF. A one-way ANOVA was performed to determine statistical significance against the no-dye autofluorescence control. Data shown represent the mean ± SD with at least three biological replicates for A and C, representative data from three independent experiments for B, and three technical replicates for D. Significant values are *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001.

Fig. S2.

MalE-LacZ expression increases the fluorescence of cells exposed to CM-H2DCFDA independent of cell size and shape or autofluorescence. Shown are fluorescent and light microscopy images of MalE-LacZ–expressing cells with and without the dye CM-H2DCFDA (CM). All fluorescent images use the same exposure time and the same minima and maxima. Fluorescence is visible only in the presence of the dye and maltose after 2 h. There is no detectable autofluorescence, and the signal is independent of cell size or shape. (Magnification: 1,000×.)

Genetic Evidence That Oxidative Stress Plays a Causal Role in Cell Death Caused by MalE-LacZ Induction.

If ROS contribute causally to cell death, as opposed to simply being generated by dying cells, then mutants sensitive to oxidative stress should be more sensitive to killing by MalE-LacZ induction. Consistent with H₂O₂ playing a causal role in lethality, we observed that null mutants lacking either of E. coli’s cytoplasmic catalases, KatE or KatG, were more sensitive to killing by maltose induction of the MalE-LacZ hybrid protein (Fig. 3 A and B), as were mutants lacking either subunit of E. coli’s ahpCF-encoded alkyl hydroperoxidase, which reduces H₂O₂ to water (Fig. 3 C and D). Moreover, the addition of exogenous catalase to the growth medium provided some protection (Fig. 3E). Since the O₂⁻ released into the medium cannot cross the membrane, and the exogenous catalase is external, some of the extracellular O₂⁻ produced upon MalE-LacZ induction may be converted to H₂O₂ that then enters the cells and contributes to lethality. Interestingly, a ΔoxyR mutant did not display increased sensitivity to killing (Fig. 3F). katE, whose deletion causes the strongest phenotype, is not regulated by OxyR, but katG and ahpCF are, indicating that basal levels of KatG and AhpCF play a physiologically significant role in protecting cells against endogenous H₂O₂ produced by MalE-LacZ induction. soxR- and soxS-null mutants exhibited an increased susceptibility to killing after maltose induction (Fig. 3 G and H), but a sodA sodB double mutant lacking both cytoplasmic superoxide dismutases was much more resistant to killing (Fig. 3I), which is similar the result reported for bactericidal antibiotics and has been attributed to the impaired ability of the strains to convert O₂⁻ to H₂O₂ (28).

Fig. 3.

ROS contributes causally to MalE-LacZ–induced cell death. (A–D) Knockouts of the catalases KatE (A) and KatG (B) and either subunit of the alkyl hydroperoxidase AhpC (C) or AhpF (D) are more sensitive to killing by MalE-LacZ. (E) Addition of exogenous catalase to the medium provides a protective effect. (F) Knockout of the major oxidative response sensor, OxyR, has no effect on killing. (G and H) Knockouts of either component of the other major oxidative sensor, SoxRS, are sensitive to MalE-LacZ killing. (I) A double knockout of both cytoplasmic superoxide dismutases sodA and sodB is insensitive to killing by MalE-LacZ. Black squares are wild type, red are wild type plus maltose, blue are mutant, and green are mutant plus maltose. Data shown represent the mean ± SD with at least three biological replicates. Significant values are *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001.

Genetic Evidence That DNA Problems Related to the Processing of Incorporated Oxidized Nucleotides Contribute Causally to the Lethality of MalE-LacZ Induction.

Our analyses of the lethality of DinB overexpression and bactericidal antibiotics led us to propose that oxidation of nucleotides, particularly those containing guanine, the most easily oxidized nucleic acid base (29), contributes causally to cell death (14). This model accounts for our observations that (i) an increased level of the nucleotide sanitizer MutT, which hydrolyzes nucleoside triphosphates containing 8-oxo-dG to their respective monophosphates, reduces killing (14); (ii) an increased level of the mismatch repair protein MutS that only modestly affects growth rate reduces killing (12); (iii) deletion of mutM and mutY, which encode BER enzymes involved in the processing of 8-oxo-dG lesions, reduces killing (14); and (iv) deletion of recA sensitizes cells to killing (14). The model suggests that incorporation of 8-oxo-dG into nascent DNA creates a situation in which incomplete MutM/MutY-dependent BER generates lethal DNA problems that can be partially ameliorated by RecA-dependent homologous recombination. We carried out parallel experiments to test whether this same mechanism might be responsible for MalE-LacZ–induced cell death.

Strikingly, increasing the expression of MutT resulted in MalE-LacZ induction being bacteriostatic rather than bactericidal (Fig. 4A), an observation suggesting that oxidation of dGTP to its corresponding 8-oxo-guanine derivative contributes to cell killing, as did increasing MutS expression (Fig. 4B). Together these two observations suggest that the lethality stems in part from 8-oxo-dGTP or some other potentially miscoding oxidized nucleotide being incorporated into nascent DNA in a fashion that leaves it potentially repairable by mismatch repair, as previously suggested for mutagenesis induced by subinhibitory levels of antibiotics (30) and cell killing by higher levels of bactericidal antibiotics (12). While multicopy suppression is a widely used technique in microbial genetics, we note that, although cellular MutT and MutS levels were increased by introducing plasmids from the ASKA collection (31), we did not add the inducer isopropyl β-d-1-thiogalactopyranoside (IPTG), thereby avoiding the concern (4) that induction of extremely high levels of protein might possibly impair overall bacterial growth and metabolism. The maltose-induced levels of the MalE-LacZ protein in the strains carrying the mutT and mutS plasmids were indistinguishable from those in the parental strain (Fig. S3A).

Fig. 4.

Cell death due to MalE-LacZ results from attempted BER of oxidative DNA damage. (A and B) Overexpression of the 8-oxo-dGTP sanitizer MutT (A) or the mismatch-repair protein MutS (B) protects cells from MalE-LacZ expression. (C) Double knockout of MutM and MutY DNA glycosylases involved in BER of 8-oxo-dG lesions is resistant to killing by MalE-LacZ. (D) The ΔrecA strain is more susceptible to killing by MalE-LacZ expression, with cell death beginning earlier. (E) Extra sensitivity of ΔrecA to MalE-LacZ killing can be partially suppressed in a ΔmutM ΔmutY background. (F) The ΔrecB ΔrecF strain is more sensitive to killing by MalE-LacZ. Data shown represent the mean ± SD with at least three biological replicates. Black squares are wild type, red are wild type plus maltose, blue are mutant, and green are mutant plus maltose. Significant values are *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001.

Fig. S3.

Overexpression and gene knockouts do not affect expression levels of MalE-LacZ. Expression levels of MalE-LacZ were unaffected in in strains overexpressing mutT or mutS (A) or in ΔmutM, ΔmutY, or ΔrecA strains (B).

Similarly, we found that deleting mutM and mutY also resulted in MalE-LacZ induction being bacteriostatic rather than bactericidal (Fig. 4C), an observation suggesting that the action of these BER DNA glycosylases involved in processing 8-oxo-dG and oxidized purine lesions contributes causally to cell death by generating BER intermediates (14). The level of the MalE-LacZ protein was not affected by loss of mutM or mutY (Fig. S3B).

Deleting the recA gene not only increased sensitivity to killing by MalE-LacZ induction; loss of colony-forming ability also could be observed as early as the 1-h time point (Fig. 4D). Loss of recA has no effect on the levels of the MalE-LacZ protein (Fig. S3B), but the increased sensitivity to killing could be largely suppressed by deleting mutM and mutY (Fig. 4E). These observations suggest that potentially lethal DNA problems involving strand breaks caused by MutM and MutY incisions at oxidized nucleotides begin to arise earlier after MalE-LacZ induction than we had suspected but are initially kept in check by a RecA-dependent process. A ΔrecB ΔrecF double mutant had a similar phenotype, suggesting that RecA’s recombinational functions, rather than its SOS regulatory roles, are important (Fig. 4F).

Time-Resolved Microarray and Network Analysis of the Consequences of MalE-LacZ Induction.

Additional mechanistic insights into the lethal physiological events triggered by MalE-LacZ induction were gained by a time-resolved microarray study (SI Materials and Methods) with qPCR verification of representative genes of interest (Fig. S4 and Table S2) and Network Component Analysis (NCA), a method that uses the connectivity from gene-regulatory networks to infer the activity of transcriptional regulators (SI Materials and Methods and Dataset S1) (32, 33). Many changes in gene expression were observed, including the rapid and robust induction of the mal operon as well as the previously reported (34) robust induction of heat-shock genes (Figs. S5 and S6). In addition to MalT and RpoH (Fig. 5 A and B), our NCA analysis detected significant changes in transcriptional activity for 14 transcription factors out of 131 analyzed (Dataset S2).

Fig. S4.

qPCR validation of representative transcriptional changes. (A–C) Expression of MalE-LacZ greatly increases expression of mutM at all time points (A) and increases the expression of mutY (B) and mutT (C) after 5 h. (D–F) MalE-LacZ also increases the expression of the catalase katE (D) but not katG (E) or aphF (F). Cells were treated with 10 mM H₂O₂ as a control. Data shown represent the mean ± SD with at least three biological replicates. Significant values are *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001.

Table S2.

Primers used in this study

Name	Sequence
rrsA F	ACCTTACCTACTCTTGACATCCA
rrsA R	CCCAACATTTCACAACACGAG
rpoA F	TAGAACAGCGTACCGACCTG
rpoA R	CGAAAGCTTCCAGTTGTTCA
soxS F	TACTTGCAACGAATGTTCCG
soxS R	ACATAACCCAGGTCCATTGC
mutM F	ACCATGTGGATTTGGTGATG
mutM R	ACATTATGCCCTTCCAGCTC
mutT F	TGCGGTAGGTATTATTCGCA
mutT R	TCCAGTTTATTCGCCATGTG
mutY F	CGCTTTCTCTGGGTAAGCAC
mutY R	TACAAATCATCGCACCCAAA
katE F	GCCTACAAACACCTTAAACCG
katE R	TCCATAAAACTACCGTCAGCG
katG F	CTGGCGGATATCATAGTGCTG
katG R	CATGAATGCTCAAACCTGCG
ahpF F	GCATGGGCGAGATTATCATTG
ahpF R	ATGATCTGCTTGTACGGAACC

F, forward; R, reverse.

Fig. S5.

Time-resolved microarrays reveal the induction of noncanonical oxidative stress responses. (A and B) malE (A) and lamB (B) expression increases following maltose induction of MalE-LacZ. (C) NCA reveals cell envelope stress by changes in CpxR activity. (D and E) Loss of ftsZ (D) and increases in sulA expression (E) coincided with the morphological changes observed by microscopy (Fig. S2). (F and G) NCA revealed induction of a noncanonical oxidative stress response with MalE-LacZ–induced SoxS (F) and MarA (G) activity. Black squares are control, red are maltose-induced. Data shown represent the mean ± SD with three biological replicates.

Fig. S6.

MalE-LacZ does not induce the SOS response but induces genes associated with DNA repair. (A and B) NCA reveals that LexA activity is unaltered (A) and recA expression is unchanged (B) by MalE-LacZ induction. (C–H) However, the expression of several other DNA-repair genes is induced, including dinG (C), mfd (D), nusA (E), recF (F), recJ (G), and recQ (H). Black squares are control, red are maltose-induced. Data shown represent the mean ± SD with three biological replicates.

Fig. 5.

Time-resolved microarrays reveal lack of induction of the OxyR oxidative stress-response regulon but increased expression of oxidized nucleotide-repair genes. (A and B) NCA shows strong induction of the MalT maltose regulon (A) and RpoH heat-shock response (B). (C) In contrast, the OxyR oxidative stress-response regulon is not activated. (D–H) Maltose addition induces expression of the katE (D) and sodB (E) oxidative stress-response genes and all three major components of oxidized nucleotide repair, mutM (F), mutY (G), and mutT (H). Black squares are control, red are maltose-induced. Data shown represent the mean ± SD with three biological replicates.

Several observations were particularly relevant. First, the canonical oxidative stress-response factor OxyR (35) did not exhibit statistically significant changes in activity following MalE-LacZ induction (Fig. 5C), a result consistent with the lack of an effect of a ΔoxyR mutation on MalE-LacZ–dependent killing. Second, of enzymes that scavenge ROS, we found transient but statistically significant induction of only katE and sodB (Fig. 5 D and E). Third, we observed significant induction at a minimum of three different time points of three genes, mutM, mutY, and mutT, whose products are associated with processing of 8-oxo-guanine and also affect the sensitivity of cells to MalE-LacZ induction (Fig. 5 F–H). Fourth, NCA did not indicate significant LexA activation, but we observed significant induction of several genes (dinG, mfd, nusA, recF, recJ, and recQ) whose products participate in DNA-repair processes (Fig. S6). Interestingly, cells overexpressing DinB (14) exhibited changes in gene expression very similar to changes in cells expressing MalE-LacZ, including this noncanonical oxidative stress response (Fig. S7A) and induction of mutM, mutY, and other DNA-repair genes (Fig. S7B).

Fig. S7.

DinB overexpression induces changes in gene expression similar to those in maltose-induced MalE-LacZ cells. Reanalysis of original microarray data from Foti et al. (14) revealed (A) similar induction of noncanonical oxidative stress responses and (B) similar induction of mutM, mutY, and other DNA-repair genes. Data shown represent the fold-change in MG1655 pINIII:DinB cells following DinB induction by IPTG over the uninduced control. n = 1 biological replicate collected in the original work.

SI Materials and Methods

Strains.

The strains used in this study are listed in Table S1. Alleles were transduced from the Keio collection into the MM18 background via standard P1vir transduction and resolved using the plasmid pCP20 as necessary. The MutT and MutS plasmids used were from the ASKA overexpressing plasmid library. MM18 and its derivative strains were grown in 0.4% glucose in LB during maintenance, transductions, and transformations to reduce MalE-LacZ expression and the potential generation of suppressor mutants.

MalE-LacZ Induction.

MM18 and derivatives were streaked onto LB agar containing X-Gal and grown ON at 37 °C. Blue, average-sized colonies were picked and grown ON in M63 medium (20) or LB with any appropriate antibiotics. The cultures were then diluted 1:100 in fresh medium, grown to OD₆₀₀ 0.1, and split into two populations with one receiving 0.2% maltose (Sigma) and grown at 37 °C on a rotating shaker. For the exogenous catalase experiment, 1,000 units of bovine catalase (Sigma) were added to each culture at time 0. Samples were taken every hour to determine cfus or to harvest cells for protein, DNA, or RNA extraction. The resulting colonies were counted on a ProtoCOL 3 colony counter (Synbiosis). Student’s t test was used to determine statistical differences between samples.

Microscopy.

MM18 was grown as previously described. At the indicated time points cultures were concentrated by centrifugation, washed in PBS, and then fixed in 4% paraformaldehyde. Cells were then washed, stained with DAPI (Thermo Scientific), and washed again. Bacteria were plated on slide-mounted agarose pads and visualized with a Nikon Eclipse Ni microscope.

Live-Dead Staining.

The LIVE/DEAD BacLight Bacterial Viability Kit (Thermo Fisher) was used to determine bacterial viability. Cultures were grown and MalE-LacZ expression was induced as previously described. At each time point cells were collected and stained according to the manufacturer’s instructions. The stained cells were then plated on slide-mounted agarose pads and visualized with a Nikon Eclipse Ni microscope. Collected images were analyzed using ImageJ (57). The number of live and dead cells was quantified for each time point. The presence of any propidium iodide staining was interpreted as a dead cell.

Microarray Analysis.

Cultures were grown in M63, and MalE-LacZ was induced as before. At the appropriate time points aliquots were removed, and RNA Protect Bacteria Reagent (Qiagen) was applied. RNA was extracted using the RNeasy Bacteria Mini Kit (Qiagen) and Turbo DNA-free (Life Technologies) DNase treatment, according to the manufacturers’ instructions. cDNA preparation and hybridization to Affymetrix GeneChip E. coli 2.0 microarrays were performed as previously described (12). CEL files for the resulting expression profiles were background adjusted and normalized using Robust Multiarray Averaging as previously described (12). In addition, CEL files corresponding to the microarray data from Foti et al. (14) were similarly reanalyzed. Microarray data collected in this study are available for download on the Gene Expression Omnibus (GEO).

NCA.

A compendium of E. coli gene-expression profiles was assembled by downloading all CEL files corresponding to microarray datasets for E. coli K12 analyzed by Affymetrix E. coli 2.0 and available for download on GEO at the time of analysis. This yielded a compendium of 851 arrays (Dataset S1). CEL files from this compendium were Robust Multi-array Averaged (RMA) normalized with the CEL files corresponding to the MM18 RNA samples collected for this study. E. coli gene-regulatory networks corresponding to transcription factor- and σ factor-based regulation were assembled by downloading the transcription factor–gene and σ–gene interaction files from RegulonDB v. 8.8. The transcription factor regulatory network was made full rank by removing transcription factors that were not linearly independent from other transcription factors and which regulated the least number of genes. This yielded a final gene regulatory network of 131 transcription factors and 1,656 genes. Transcription-factor and σ-factor activities were estimated using the ROBNCA (33) implementation in MATLAB (MathWorks). All activity estimates were normalized to the time 0, noninduced control samples.

qPCR.

Cultures were grown in M63, and MalE-LacZ was induced as before. RNA was prepared using the RNA Protect Bacteria Reagent (Qiagen) and RNeasy Bacteria Mini Kit (Qiagen) according to the manufacturers’ instructions. cDNA was created from extracted RNA using M-MLV reverse transcriptase enzyme (Invitrogen) according to the manufacturer’s protocol. qPCR was then performed using PowerUp SYBR Green (Life Technologies). The 16S ribosomal protein rrsA and the rpoA subunit of the RNA polymerase were used as endogenous controls. All primers used are listed in Table S2. Student’s t test was used to determine statistical differences between samples.

8-Oxo-dG ELISA.

Cultures were grown expressing MalE-LacZ as described previously to the desired time points, as well as positive controls treated with 10 mM H₂O₂ for 1 h. DNA was extracted using the GenElute Bacterial DNA Extraction Kit (Sigma) according to the manufacturer’s instructions. 8-Oxo-dG was quantified using the OxiSelect Oxidative DNA Damage ELISA kit (Cell Biolabs). Samples were assayed in biological triplicate.

Intracellular H₂O₂ Measurement.

MM18 was transformed with the pZE21-APX plasmid (12). Cultures were grown overnight in LB, diluted 1:100 in LB, and grown to an OD₆₀₀ of ∼0.2. Then 30 ng/µL of anhydrotetracycline was added, and they were allowed to grow to an OD₆₀₀ of ∼1.0. Cultures then were diluted back to 0.1, and maltose was added to one set. At the desired time points cultures were collected, normalized by OD, and concentrated ∼10-fold. Pellets were resuspended in freshly prepared 50 µM Amplex UltraRed (Life Technologies) and plated into a 96-well opaque black plate. Fluorescent measurements were taken with a Tecan Spark 10 M using an emission and excitation of 530 and 590 nm, respectively.

Superoxide.

Superoxide release was measured using an electrochemical cytochrome c biosensor. The cytochrome c biosensor used in this study was fabricated and calibrated using the procedure described previously (24, 58). In this study, the biosensor was fabricated on a bare gold wire 1 cm mm in length and 0.5 mm in diameter.

The calibration of the sensor was carried out in LB Lennox broth, the medium used for MM18 culture. A linear calibration curve in LB Lennox medium with a linear range from 0.47 to 1.47 μM and a detection limit of 62 nM superoxide was established and calculated as described previously (59, 60). Continuous measurement with the sensor in the absence of enzymatically produced superoxide showed no change in signal over time, indicating no interferences on response from the medium.

The experimental procedure used for measurements in MM18 cultures is as follows: 250 µL MM18 stock culture was added to 25 mL LB Lennox broth. The culture was grown to ∼0.1 OD. Optical density was monitored using a Nanodrop 2000C (Thermo Scientific). A CHI 1232b electrochemical analyzer (CH Instruments, Inc.) with a three-electrode system (Ag/AgCl reference, Pt wire counter, and the cytochrome c-modified gold wire as a working electrode) immersed in the cell culture was used for all electrochemical measurements in this study. The superoxide sensor was placed in 20 mL of MM18 culture. Then, maltose solution was added after ≈60 or 90 min after the electrochemical measurement started in the culture to a final concentration of 0.2%. Amperograms were recorded for both maltose treated and untreated cultures.

Dyes.

Two fluorescent dyes were used to quantify the production of ROS in MalE-LacZ–expressing cells: CM-H2DCFDA and HPF (Life Technologies). Cells were grown as before ON in M63 medium diluted 1:100 and grown to an OD of 0.1. Then 10 µM of either dye or a no-dye control were added, and cultures were grown in light-protected tubes in a rolling shaker at 37 °C. At each time point samples were taken and diluted into 1× PBS and run on a BD FACS Aria II. Statistical significance was determined using one-way ANOVA, with each sample being compared with the no-dye autofluorescence control. For the microscopic visualization, cells grown with CM-H2DCFDA or no dye were collected at each time point, washed in saline, and plated on slide-mounted agarose pads. They were then visualized on a Nikon Eclipse Ni microscope. A 100-ms exposure time was used for all samples. The images were then analyzed using ImageJ, with all fluorescent images being standardized to the same minima and maxima settings.

Western Blot.

Western blots were used to determine the expression level of MalE-LacZ and SecY. The desired strains were grown and MalE-LacZ expression was induced as described before. At the desired time points cultures were pelleted, washed in 1× PBS, and frozen at −80 °C ON. The bacteria were then lysed in B-Per reagent (Thermo Fisher). Protein levels in the lysates were quantified by Bradford assay and normalized. The proteins were separated on a 12% SDS/PAGE gel and transferred to a PVDF membrane, which was then blocked by the addition of 5% nonfat milk in Tris-buffered saline with Tween 20 (TBST) ON. The primary antibodies α-MalE (Abcam) and α-SecY (a gift from Tom Rapoport, Harvard Medical School, Boston) were both used at 1:10,000. After washing in TBST, the secondary α-rabbit antibody (Thermo Scientific) was added at 1:25,000. After washing, ECL was added to the membrane, and proteins were visualized on Hyblot film (Denville Scientific).

Time-Resolved Microarray and Network Analysis.

To understand the physiological events triggered by MalE-LacZ that lead to cell death, we performed time-resolved microarray analysis to quantify induced stress responses and performed NCA (32, 33) to infer transcriptional regulators that may drive cellular death processes. As expected, microarray analysis revealed rapid and robust induction of malE and lamB (Fig. S5 A and B) by maltose addition to the MM18 cells and robust MalT activation (Fig. 5A). In addition, we observed significant induction of CpxR activity by 1 h (Fig. S5C), revealing expected cell envelope stress induced by jamming of Sec transporters by MalE-LacZ. Consistent with our observations that MM18 cells exhibited elongated morphologies 3–4 h after MalE-LacZ induction (Fig. S2), we detected modest ftsZ depletion and sulA enrichment (Fig. S5 D and E) 3–4 h after maltose addition, relative to control cells. We also saw robust induction of the heat-shock response σ factor RpoH (Fig. 5B), as is consistent with the previously noted induction of heat-shock genes (34).

We next examined gene-regulatory responses following MalE-LacZ induction to infer processes that may contribute to lethality. Of the 131 transcription factors analyzed, only 16 exhibited statistically significant differences in activity (P ≤ 0.01) at more than one time point before commitment to cell death (at times 0, 1, or 2 h) between maltose-treated and control MM18 cells (Dataset S2). Most of these were directly regulated processes involved in cell metabolism (AgaR, AscG, CRP, GalS, MalT, and Nac), which we suspected to be a direct consequence of the use of maltose as an inducer. We also observed significant changes in the activities of regulators of membrane-associated genes, including two-component systems (CpxR, DhaR, OmpR, QseB, and ZraR) and transporters (MarA and NanR). Finally, we detected significant changes in other stress-response regulators (GadE: acid stress; PspF: phage shock; SoxS: oxidative stress). Interestingly, OmpR, MarA, and ZraR also respond to cell stresses (osmotic stress, oxidative stress, and zinc stress, respectively).

While NCA revealed that MalE-LacZ induced de-repression of SoxS and activation of MarA (Fig. S5 F and G), both of which respond to oxidative stress, the canonical oxidative stress-response factor OxyR (35) did not significantly change in activity (Fig. 5C). Of enzymes that scavenge ROS, we found transient but statistically significant induction of only katE and sodB (Fig. 5 D and E). In addition, although sulA was induced at later time points, NCA did not reveal significant changes in LexA activity (Fig. S6A), nor did we observe significant changes in canonical SOS-response genes such as recA (Fig. S6B). However, importantly, we did observe significant induction at at least three different time points of three genes, mutM, mutY, and mutT (Fig. 5 F–H), whose products are associated with the processing of 8-oxo-guanine damage and affect the sensitivity of cells to MalE-LacZ induction. We also observed significant induction of several genes (dinG, mfd, nusA, recF, recJ, and recQ) whose products participate in DNA-repair processes (Fig. S6 C–H). Collectively, these data reveal induction of a noncanonical oxidative stress response following MalE-LacZ induction and sensed oxidized DNA damage.

Interestingly, very similar changes in gene expression were observed in microarray data from cells overexpressing DinB, which experience lethality due to 8-oxo-deoxyguanosine incorporation (14). Like the MM18 cells in this present study, cells overexpressing DinB exhibited a noncanonical oxidative stress response following DinB induction (Fig. S7A), with negligible changes in oxyR or soxR expression and only modest induction of katE, katG, and sodA. Additionally, DinB-overexpressing cells also exhibited modest induction of mutM, mutY, dinG, nusA, recF, recJ, and recQ (Fig. S7B), similar to the maltose-induced MalE-LacZ cells. Taken together, these data indicate that noncanonical oxidative stress responses are not unique to cell death following antibiotic treatment (12) and may be common across multiple modes of bacterial lethality.

Discussion

The MalE-LacZ periplasmic-cytoplasmic fusion protein is historically important because of its use in the elucidation of SecY-dependent protein translocation (21). However, our observations suggest that cell death caused by MalE-LacZ induction does not result directly from jamming of protein translocation or SecY degradation. Rather our findings support a model in which physiological/metabolic stress from MalE-LacZ expression increases the production of low levels of O₂⁻ and H₂O₂. H₂O₂ does not accumulate to high intracellular levels because cellular conditions favor its participation in Fenton chemistry, resulting in the oxidization of nucleotides, including 8-oxo-dGTP, which are subsequently incorporated into DNA. Cellular death does not result directly from the incorporation of oxidized nucleotides but rather from lethal DNA problems caused by intermediates of MutM/MutY-dependent BER. Initially these potentially lethal DNA problems can be ameliorated by RecA-dependent processes or by a combination of RecB- and RecF-dependent processes. Multiple, independent lines of evidence support the physiological relevance of each element of this model for cell death.

These physiological and mechanistic attributes of the lethality caused by MalE-LacZ expression are strikingly similar to those of the oxidative component of antibiotic lethality (1, 6, 7, 12, 14). In the case of MalE-LacZ, this mode of cell death involving oxidative DNA damage is the dominant cause of the lethality. The situation is more complicated for antibiotics because they can also cause cell death by their direct effect on their cellular targets and associated cellular processes. The relative contribution of this oxidative component to antibiotic-induced cell death can vary depending on factors such as the experimental conditions, the antibiotic, and the metabolic state of the bacterium (1, 6, 12).

Two observations related to the H₂O₂-inducible OxyR regulon might appear to be inconsistent with this model: (i) MalE-LacZ expression does not induce the OxyR regulon, and (ii) an oxyR mutant does not display increased sensitivity to MalE-LacZ lethality. However, these observations can be understood by (i) recognizing that the amount and targeting of Fenton oxidants are critical rather than the levels of endogenously produced O₂⁻ or H₂O₂, (ii) appreciating that only a small increase in oxidative damage in DNA can prove lethal, and (iii) considering the chemical and enzymatic requirements of the DNA-based cellular death pathway indicated by our experimental observations.

Our observations implicating 8-oxo-dG and potentially other oxidized nucleotides in cell death imply that Fenton chemistry is involved in the lethality since neither O₂⁻ nor H₂O₂ reacts significantly with nucleic acids or nucleotides (15, 16, 36). It has been argued that, since it is implausible for intracellular levels of endogenously produced H₂O₂ to rise to lethal levels, stresses would have to do more than accelerate H₂O₂ formation (4). One specific effect that stresses such as MalE-LacZ and antibiotics could induce that would additionally increase oxidative damage would be to accelerate the rate of the Fenton reaction, which would also prevent H₂O₂ from accumulating to sufficient levels to induce the OxyR regulon. Fenton oxidation can occur so rapidly relative to diffusion that H₂O₂ produced by histone demethylation introduces 8-oxo-dG lesions into the surrounding DNA that are then exploited to control local gene expression (37). Various physiological factors such as pH (38), cysteine levels (39), and anionic ligands (38) accelerate the Fenton reaction and could be affected by stresses such as MalE-LacZ and antibiotics. Interestingly, H₂S, the product of cysteine metabolism, protects bacteria from antibiotic killing (40). It is particularly relevant that nucleoside triphosphates and nucleic acids are anionic ligands that complex with Fe⁺² and promote the Fenton reaction to an extent comparable to EDTA and nitrilotriacetate (38, 41). Furthermore, the proximity of the nucleic acid base of a dNTP or NTP to an Fe⁺² complexed by its phosphates (42) favors its reaction with highly reactive Fenton oxidants (11, 16). Since the diffusion distance for a hydroxyl radical is only one carbon bond length (43), nucleotides and nucleic acids must be at higher risk of damage from Fenton oxidants than many other biomolecules because of their ability to complex Fe⁺² and promote local production of Fenton oxidants.

Bacteria live on a knife’s edge with respect to their ability to tolerate oxidative stress (4), so that only a small increase in oxidative damage in DNA can have mutagenic or lethal consequences. In the case of aerobically grown E. coli, the threat posed by endogenous ROS is narrowly balanced by the titers of scavenging enzymes (15). Minor oxidative reactions that impart a gain of function to the target can be particularly important even though they represent a negligible component of the overall reactions (16); 8-oxo-dG is such a gain-of-function oxidation product because it pairs with dA as well as dC. With respect to tolerating 8-oxo-dG, aerobically grown E. coli live so close to the edge that decreasing the level of MutT by only a factor of two increases the mutation rate (44). Moreover, for aerobically grown E. coli, simply increasing the frequency of initiation of DNA replication (45) or increasing DNA polymerase IV, which has a propensity to use 8-oxo-dGTP (14, 18), is enough to cause cell death because BER cannot be completed before encountering the next replication fork (45). An advantage that bacteria gain by living so close to this threshold is that they can mutate in response to stress by increasing the incorporation of oxidized deoxynucleotides into DNA while simultaneously suppressing mismatch repair (30). Thus, MalE-LacZ expression or bactericidal antibiotic stress seems to exaggerate a type of potential oxidative toxicity that lurks just below the threshold in unstressed wildtype cells.

A factor that likely contributes to cell death is that BER of oxidative damage can be slow to complete because the lyase activities associated with MutM and MutY catalyze β or β,δ eliminations, thereby generating ends that require further processing to expose the 3′-OH needed for DNA polymerases (46). This DNA-based death mechanism, which initially can be counteracted by RecA-dependent or RecB/RecF-dependent processes, is likely complex. Since it takes only a single unrepaired double-stand break (DSB) to kill a bacterial cell (47), one problem could be DSBs caused by MutM and MutY incisions at closely spaced lesions (14) or by replication forks encountering unrepaired BER intermediates (45). Other potentially lethal DNA problems include interstrand crosslinks mediated by abasic sites (48), LigA/MutY-dependent futile cycles of ligation/incision (49), and unresolvable collisions caused by stalled transcription or replicative complexes (50).

Importantly, key genes whose functions affect this mode of DNA-based oxidative cell death are not regulated by H₂O₂-inducible OxyR or by SoxRS but rather by other stress regulators: e.g., mutM (RpoH) (51), mutY (Fur/ArcA/Fnr, normally down-regulated upon oxidative stress but up-regulated upon MalE-LacZ induction) (52), and mutT (CpxA/CpxR in an operon with secM and secA) (53), and dinB/recA (SOS) (54). Thus, the intracellular environment of cells generating endogenous H₂O₂ as a consequence of a stress can be very different from the intracellular environment of cells treated with exogenous H₂O₂. For example, the increased levels of MutM glycosylase due to the powerful heat-shock response elicited by MalE-LacZ would result in more frequent initiations of BER and hence in a greater probability of a BER intermediate with a strand break being encountered by a replication fork before repair can be completed. Some of the past confusion in this research area may have arisen because the mode of oxidative death occurring in bacterial cells undergoing stress from agents such as bactericidal antibiotics or MalE-LacZ does not conform to expectations based on studies of cells treated with exogenous H₂O₂, as these other stresses induce additional key non–OxyR-regulated proteins that affect lethality.

Our evidence that a completely different stress besides bactericidal antibiotics causes ROS-based lethality resulting from nucleotide oxidation and incomplete BER shows that this type of cell death is not unique to antibiotics and suggests that it likely contributes to death from other stresses as well. As previously discussed (14), oxidation of ribonucleotides could also contribute to lethality by other mechanisms. Our results suggest that exploiting the oxidative component of antibiotic lethality is a plausible strategy to improve the efficacy of existing antibiotics or to identify new ones (55, 56).

Materials and Methods

Detailed materials and methods can be found in SI Materials and Methods.

MalE-LacZ Induction.

MM18 and derivatives were streaked onto LB agar containing X-Gal and grown overnight (ON) at 37 °C. Blue, average-sized colonies were picked and grown ON in M63 medium or LB with any appropriate antibiotics. The cultures were then diluted 1:100 in fresh medium, grown to OD₆₀₀ 0.1, and then split into two populations with one receiving 0.2% maltose and grown at 37 °C on a rotating shaker. For the exogenous catalase experiment, 1,000 units of bovine catalase (Sigma) were added to each culture at time 0. Samples were taken every hour to determine cfus or to harvest cells for protein, DNA, or RNA extraction. The resulting colonies were counted on a ProtoCOL 3 colony counter. Student’s t test was used to determine statistical differences between samples.

Microscopy.

MM18 was grown as previously described. At the indicated time points, cultures were concentrated by centrifugation, washed in PBS, and then fixed in 4% paraformaldehyde. Cells were then washed, stained with DAPI, and washed again. Bacteria were plated on slide-mounted agarose pads and visualized with a Nikon Eclipse Ni microscope.