Bactericidal peptidoglycan recognition protein induces oxidative stress in Escherichia coli through a block in respiratory chain and increase in central carbon catabolism

Summary

Mammalian Peptidoglycan Recognition Proteins (PGRPs) kill both Gram-positive and Gram-negative bacteria through simultaneous induction of oxidative, thiol, and metal stress responses in bacteria. However, metabolic pathways through which PGRPs induce these bactericidal stress responses are unknown. We screened Keio collection of Escherichia coli deletion mutants and revealed that deleting genes for respiratory chain flavoproteins or for tricarboxylic acid (TCA) cycle resulted in increased resistance of E. coli to PGRP killing. PGRP-induced killing depended on the production of hydrogen peroxide, which required increased supply of NADH for respiratory chain oxidoreductases from central carbon catabolism (glycolysis and TCA cycle), and was controlled by cAMP-Crp. Bactericidal PGRP induced a rapid decrease in respiration, which suggested that the main source of increased production of hydrogen peroxide was a block in respiratory chain and diversion of electrons from NADH oxidoreductases to oxygen. CpxRA two-component system was a negative regulator of PGRP-induced oxidative stress. By contrast, PGRP-induced thiol stress (depletion of thiols) and metal stress (increase in intracellular free Zn²⁺ through influx of extracellular Zn²⁺) were mostly independent of oxidative stress. Thus, manipulating pathways that induce oxidative, thiol, and metal stress in bacteria could be a useful strategy to design new approaches to antibacterial therapy.

Abbreviated Summary

Bactericidal PGRP induces oxidative stress through a block in respiratory chain, which diverts electrons from NADH oxidoreductases to O₂, generates H₂O₂, and results in decreased respiration. Production of H₂O₂ depends on increased supply of NADH from cAMP-Crp-controlled glycolysis and TCA cycle. CpxRA is a negative regulator of PGRP-induced oxidative stress. PGRP-induced thiol stress (depletion of thiols) and metal stress (increase in intracellular free Zn²⁺ through influx of extracellular Zn²⁺) are mostly independent of oxidative stress.

Introduction

Peptidoglycan Recognition Proteins (PGRPs or PGLYRPs) are evolutionarily conserved innate immunity molecules homologous to bacteriophage type 2 amidases, present in animals and humans (Royet and Dziarski, 2007; Dziarski et al., 2016). Mammals have four PGRP genes, PGLYRP1, PGLYRP2, PGLYRP3, and PGLYRP4, which code for soluble secreted proteins. PGLYRP1, PGLYRP3, and PGLYRP4 are directly bactericidal for both Gram-positive and Gram-negative bacteria (Tydell et al., 2002, 2006; Lu et al., 2006; Wang et al., 2007), whereas PGLYRP2 is an enzyme, peptidoglycan amidohydrolase (Gelius et al., 2003; Wang et al., 2003). All PGRPs have one or two PGRP domains with a binding site specific for muramyl-penta-, tetra-, or tri-peptide fragment of bacterial peptidoglycan (Royet and Dziarski, 2007; Dziarski et al., 2016). In addition, mammalian PGRPs have another binding site specific for bacterial lipopolysaccharide (LPS) located outside the peptidoglycan-binding groove (Tydell et al., 2006; Sharma et al., 2011).

Bacterial killing is initiated by binding of PGRP to bacteria. In Gram-positive bacteria, PGRPs preferentially bind to muramyl peptides exposed by peptidoglycan-lytic endopeptidases at the separation sites of the newly formed daughter cells (Kashyap et al., 2011). In Gram-negative bacteria, PGRPs bind uniformly to the entire outer membrane (Kashyap et al., 2011), because of the ability of PGRPs to bind LPS (Tydell et al., 2006; Sharma et al., 2011).

PGRPs do not enter the cytoplasm, but rather exert their bactericidal effect through induction of exaggerated stress responses in bacteria (Kashyap et al., 2011). Efficient PGRP killing of both Gram-positive and Gram-negative bacteria requires simultaneous induction by PGRPs of oxidative, thiol, and metal stress responses in bacteria (Kashyap et al., 2014).

PGRP-induced oxidative stress is manifested by increased production of hydrogen peroxide (H₂O₂) and hydroxyl radicals (HO^•) and high induction of oxidative stress response genes, including OxyR and SoxR regulons in Escherichia coli and PerR regulon in Bacillus subtilis (Kashyap et al., 2011; 2014). Production of H₂O₂ and HO^• is required for PGRP-induced killing, because under anaerobic conditions, when H₂O₂ and HO^• are not produced, PGRPs are not bactericidal, but only bacteriostatic (Kashyap et al., 2014). The requirement for H₂O₂ and HO^• in PGRP-induced killing is further supported by inhibition of PGRP-induced killing by dipyridyl, which inhibits HO^• production (Kashyap et al., 2011), and by a greater sensitivity to PGRP killing of ΔrecA E. coli mutants (deficient in DNA repair) and Hpx⁻ E. coli and B. subtilis mutants (deficient in catalases and hydroxyperoxidases) (Kashyap et al., 2014).

PGRP-induced thiol (disulfide) stress in both E. coli and B. subtilis is manifested by depletion of over 90% of intracellular thiols and by a great increase in the expression of thiol stress response genes, including many genes for chaperones and protein quality control (Kashyap et al., 2014). Depletion of thiols is required for PGRP-induced killing, because thiourea (which protects from thiol depletion) diminishes PGRP-induced bacterial killing (Kashyap et al., 2011; 2014).

PGRP-induced metal stress is manifested by increases in intracellular free (labile) Zn²⁺ in E. coli and both Zn²⁺ and Cu⁺ in B. subtilis, and also by a great increase in the expression of metal efflux and metal detoxification genes (Kashyap et al., 2014). PGRP-induced metal stress is required for PGRP killing of bacteria, because selective chelation of Zn²⁺ or Cu⁺ completely abolishes PGRP killing in both E. coli and B. subtilis (Wang et al., 2007; Kashyap et al., 2014).

Although induction of each of these stress responses is required for PGRP-induced killing, each stress response individually is not sufficient for bacterial killing and has only bacteriostatic effect. Efficient killing requires simultaneous induction of all threes stress responses, i.e., oxidative, thiol, and metal stress (Kashyap et al., 2014). Similar killing can be recapitulated by the simultaneous treatment of bacteria with agents that induce H₂O₂ production, depletion of thiols, and elevation of intracellular metal concentrations. PGRP-induced stress responses lead to cessation of all biosynthetic reactions in bacteria, including protein, RNA, DNA, and intracellular peptidoglycan synthesis, membrane depolarization, and bacterial death (Kashyap et al., 2011).

The mechanism of induction of oxidative, thiol, and metal stress by PGRPs in bacteria, the sequence of their induction, and their relation to each other, are unknown. This knowledge is important not only for greater understanding how these innate immunity proteins kill bacteria, but also for future applications to rationally design more efficient antibacterial agents, and for resolving some of the recent controversies regarding the role of oxidative stress in bacterial killing (Kohanski et al., 2007, 2008; Brynildsen et al., 2013; Ezraty et al., 2013; Liu and Imlay, 2013; Keren et al., 2013; Mahoney and Silhavy, 2013; Dwyer et al., 2014, 2015; Lobritz et al., 2015; Imlay, 2015).

We performed this study to identify metabolic pathways through which PGRP induces oxidative, thiol, and metal stress responses in bacteria and to determine whether these stress responses are induced through the same single pathway or different independent pathways. We show that in E. coli PGRP induced oxidative stress through a block in respiratory chain and likely premature diversion of electrons to O₂, resulting in increased H₂O₂ production and decreased respiration. Production of H₂O₂ depended on increased supply of NADH from cAMP-Crp-controlled central carbon catabolism. CpxRA two-component system was a negative regulator of PGRP-induced oxidative stress. By contrast, PGRP-induced thiol stress (depletion of thiols) and metal stress (increase in intracellular free Zn²⁺) were both mostly independent of oxidative stress and CpxRA.

Results

Deletion of genes for respiratory chain flavoproteins, TCA cycle, Crp, or cAMP increases resistance to PGRP killing

We screened the entire E. coli Keio library of 3884 single gene deletion mutants to find mutants with increased resistance to PGRP killing (Fig. S1A). We did not find any mutants completely resistant to PGRP, if resistance is defined as growth in the presence of PGRP concentrations that are bactericidal for parental bacteria. However, we identified 14 partially resistant mutants with statistically significant higher survival than the parental strain and the ratio of survival of PGRP-treated Δ mutant versus parental strain of 10 to 100 (Fig. 1A and Table S1). 13 of these mutants were in genes coding for proteins involved in flavin-catalyzed redox reactions, including 5 mutants in genes coding for proteins that participate in oxidation of NADH in the respiratory chain (nuoE, nuoG, nuoJ, nuoK, and ndh), 5 for enzymes from the TCA cycle (sucD, sucB, lpdA, icd, and sdhD), and 3 for other oxidoreductases (frdC, betA, and ybdK) (Fig. 1A, Table S1).

Fig. 1. E. coli mutants with increased resistance to PGRP killing revealed by screening of Keio library.

(A) Mutants with increased resistance to PGRP killing were identified by three-stage screening of the entire Keio collection of single gene deletion mutants (Fig. S1 and Table S1) and here the survival of the parental strain (BW25113) and mutants following 3-hr incubation with 200 µg/ml of BSA (as a control) or PGRP is shown. Gene products, their functions, and numerical data are shown in Table S1. (B) Mutants for the key genes for the respiratory chain and TCA cycle, and their regulators (cyaA and crp) were constructed in MG1655 and their sensitivity to killing by 100 µg/ml of PGRP was similarly tested. The results are means of 3 experiments, expressed as percent of initial inoculum (100%) + SEM; ^ P≤0.05, ^^ P<0.001, numbers of viable bacteria (colony forming units) of Δ mutants versus parental strain (t-test).

To verify these results, we generated mutants in several key genes (nuoEF, ndh, sucB, and sucD) in another E. coli strain, MG1655, and we tested their sensitivity to PGRP killing. All four mutants constructed by us were significantly more resistant to PGRP killing compared with the parental MG1655 strain (Fig. 1B), which verifies the findings with Keio mutants. These results suggested that the reactions performed by the products of these genes are critical steps in PGRP killing.

Eleven genes out of the top 14 genes, whose deletion increased resistance to PGRP killing, are under the positive control of the cAMP receptor protein (Crp), eight directly and three indirectly: two through catabolite repressor activator (Cra), which is a positive regulator of Crp, and one through Fis, which is regulated by Crp (Table S1). cAMP-Crp complex is a transcriptional dual regulator of central carbon catabolism and energy production that senses insufficient generation of energy sources in the cells and up-regulates expression of genes for energy production (Supplementary text). Therefore, we next tested whether Crp and cAMP are required for PGRP-induced killing. Indeed, deletion of crp or cyaA completely abolished bactericidal effect of PGRP for E. coli (Fig. 1B). cyaA codes for adenylate cyclase, an enzyme that generates cAMP, and cAMP complexes with Crp to form the active transcription factor. Because crp or cyaA deletion mutants grew very slowly, it could be argued that their slow growth was the reason for their increased resistance to PGRP killing. However, in our screening there were also some other mutants that grew equally slowly but were efficiently killed by PGRP, which suggests that the resistance effect of crp or cyaA deletion was selective for these mutations, and not solely caused by the slow growth. Thus, these results indicate that both Crp and cAMP are required for bactericidal effect of PGRP.

PGRP-induced generation of H₂O₂ requires respiratory chain, TCA cycle, Crp, and cAMP

We next tested the role of the above-identified genes and the reactions carried out by their products in the PGRP-induced generation of H₂O₂, which we have previously shown to be required for PGRP killing of bacteria (Kashyap et al., 2014). PGRP treatment of aerobically growing E. coli induced high level of H₂O₂ production that peaked at 15 min and declined by 30 min (Fig. 2A). Treatment with paraquat as a positive control (which generates O₂⁻ and H₂O₂ directly after being reduced to radical cations by complex I) (Cochemé and Murphy, 2008; Imlay, 2013) induced a similar increase in H₂O₂ production that also peaked at 15 min, but did not decline at 30 min. E. coli similarly treated with BSA as a negative control had low levels of H₂O₂ throughout the incubation period, as expected.

Fig. 2. H₂O₂ production in PGRP-treated E. coli.

(A) Time kinetics of changes of H₂O₂ in E. coli MG1655 treated with 100 µg/ml BSA or PGRP, or with 100 µM paraquat. (B) H₂O₂ in E. coli BW25113 and in the indicated deletion mutants from Keio collection treated with 100 µg/ml BSA or PGRP for 15 min. (C) H₂O₂ in E. coli MG1655 and in the indicated deletion mutants constructed in our laboratory treated with 100 µg/ml BSA or PGRP for 15 min. The results are means of 3–4 experiments ± SEM (SEM were within symbols if not visible); * P<0.05, ** P<0.001, PGRP vs BSA; ⁺P<0.05, ⁺⁺P<0.001, paraquat vs BSA; ^ P<0.05, ^^ P<0.001, PGRP-treated mutant vs parental strain.

PGRP-induced increase in H₂O₂ was mostly abolished in all Keio mutants with deletions of genes for either respiratory chain (nuo and ndh) or TCA cycle (sucB, sucD, lpdA, and icd), and was decreased by about half in two other redox flavoprotein mutants, fumarate reductase (frdC) and choline dehydrogenase (betA) (Fig. 2B). Similarly, PGRP-induced H₂O₂ production was abolished in nuoEF, ndh, sucB, and sucD deletion mutants that we constructed in E. coli MG1655 (Fig. 2C), which verifies the results obtained with the Keio mutants. These results are consistent with the higher resistance of these mutants to PGRP killing, and with greater resistance of a double nuoA ndh deletion mutant to PGRP killing (Fig. 1A, Table S1).

We next tested the role of Crp and cAMP in PGRP-induced generation of H₂O₂. Treatment of crp or cyaA deletion mutants with PGRP did not result in an increase in H₂O₂ production (Fig. 2C). These results show that Crp and cyaA (cAMP) are required for PGRP-induced generation of H₂O₂.

E. coli mutants with deletions of genes for the respiratory chain (nuo and ndh), the TCA cycle (sucB, sucD, icd, and lpdA), crp and cyaA, or two other redox proteins (frdC and betA), when grown with BSA (as a control), had similar or somewhat higher levels of H₂O₂ compared with the parental BW25113 and MG1655 strains (Fig. 2B,C), consistent with previous reports for nuo, ndh, and frd mutants (Seaver and Imlay, 2004; Korshunov and Imlay, 2010). Because these mutants treated with PGRP had drastically lower levels of H₂O₂ than the parental strains, these results indicate that the main sources of PGRP-induced H₂O₂ are different from the sources of the low-level H₂O₂ produced in control cells growing without PGRP.

PGRP induces increased expression of Crp-controlled genes for central carbon catabolism and respiratory oxidoreductases

The requirement for Crp, cAMP, and cAMP-Crp-controlled genes in PGRP-induced killing suggested that treatment of E. coli with PGRP may increase the expression of Crp-controlled genes, including the genes for utilization of glycerol, glycolysis, TCA cycle, and respiratory chain oxidoreductases. We tested this prediction by comparing expression of these genes using whole genome expression arrays in BSA- (control) and PGRP-treated E. coli under the conditions of severe PGRP-induced oxidative, thiol, and metal stress (Kashyap et al., 2014). Analysis of these arrays (NCBI GEO accession number GSE44211) revealed significant increases in expression of five genes that participate in glycerol utilization in lower glycolysis (glpK, glpD, gapA, gpmM, and eno, as glycerol was the energy source in our medium) and 17 TCA cycle genes (aceE, lpdA, gltA, acnA, acnB, icd, sucA, sucB, sucC, sucD, sdhA, sdhB, fumA, fumB, fumC, mdh, and mqo) in PGRP-treated compared with BSA-treated E. coli (Fig. 3 and Table S2). The reactions catalyzed by the products of these genes are shown in Fig. S2. All of these TCA cycle genes and several glycerol utilization genes are positively controlled by cAMP-Crp (Fig. 3 and Table S2) (Karp et al., 2017; Shimada et al., 2011). These results suggest greater utilization of glycerol and increased glycolysis and TCA cycle in PGRP-treated cells.

Fig. 3. PGRP treatment of E. coli up-regulates genes for central carbon catabolism and respiratory chain.

E. coli MG1655 was treated with 100 µg/ml BSA or PGRP for 30 min and the fold increase in gene expression was determined by whole genome expression arrays. The results are means of 3 experiments and the quantitative expression data, the significance of differences, the gene functions are shown in Table S2, and the reactions catalyzed by gene products for glycerol usage in glycolysis and TCA cycle are shown in Fig. S2. The entire genome expression array data were deposited in NCBI GEO (accession number GSE44211).

We also determined whether PGRP-treated E. coli also had increased expression of genes for respiratory oxidoreductases, which were required for full sensitivity to PGRP-killing (Fig. 1 and Table S1) and for PGRP-induced increase in H₂O₂ (Fig. 2B,C). PGRP treatment significantly increased expression of ndh and 8 out of 13 nuo genes after 30 min (Fig. 3 and Table S2) or 15 min (Fig. S3). The nuo operon is positively regulated by cAMP-Crp, and both nuo operon and ndh are positively regulated by Fis, which is controlled by Crp (Karp et al., 2017; Shimada et al., 2011). A high (17-fold) increase in ndh expression in PGRP-treated bacteria indicates a higher demand for NDH-2 than NDH-1 to oxidize the excess of NADH generated in glycolysis and TCA cycle. PGRP treatment also significantly increased expression of genes for two other oxidoreductases, fumarate reductase (frdC) and choline dehydrogenase (betA) (Fig. 3 and Table S2), consistent with higher resistance of frdC and betA deletion mutants to PGRP killing (Fig. 1A and Table S1) and with significantly diminished PGRP-induced H₂O₂ production in these mutants (Fig. 2B).

PGRP induces an increase in NADH, phosphoenolpyruvate, and cAMP

To identify the sources and the pathways that result in PGRP-induced generation of H₂O₂, and also in thiol and metal stress, we next performed epistatic and biochemical analysis of the role of central carbon catabolism in PGRP-induced stress responses.

We first tested whether treatment with PGRP changes the concentrations of NADH and NAD⁺ in E. coli, because several mutants with increased resistance to PGRP killing involved genes whose products participate in generation of NADH in the TCA cycle or oxidation of NADH in the respiratory chain (Fig. 1, Table S1), and because NADH could be the source of reducing power for generation of H₂O₂. Under aerobic conditions bacteria produce NADH in energy-generating reactions (such as glycolysis and TCA cycle, Fig. S2) and then oxidize it to NAD⁺ in the respiratory chain using NADH:ubiquinone oxidoreductases I and II (NDH-1 coded in E. coli by nuoA-N, and NDH-2 coded by ndh).

Treatment of E. coli with PGRP induced rapid increase in NADH level and NADH/NAD⁺ ratio, which peaked in 15 min and were not observed in control BSA-treated cells, yielding significantly higher NADH levels and NADH/NAD⁺ ratio in PGRP-treated than in control cells at 5, 15, and 30 min (Fig. 4A). The level of NAD⁺ did not change in both PGRP- and BSA-treated bacteria throughout the 60 min incubation period. This PGRP-induced increase in NADH levels and NADH/NAD⁺ ratio correlated with and preceded the PGRP-induced increase in H₂O₂ (Fig. 2A). Thus, oxidation of NADH to NAD⁺ by the respiratory chain could be the source of reducing power for PGRP-induced generation of H₂O₂.

Fig. 4. Time kinetics of changes in NADH, NAD⁺ and NADH/NAD⁺ ratio (A), phosphoenolpyruvate (PEP), pyruvate, and PEP/pyruvate ratio (B), and cAMP (C) in PGRP-treated E. coli.

E. coli MG1655 was treated with 100 µg/ml BSA or PGRP. The results are means of 3–4 experiments ± SEM (SEM were within symbols if not visible); * P<0.05, ** P<0.001, PGRP vs BSA.

The two main sources of NADH in E. coli grown in our experiments aerobically with glycerol as the energy source are components of central carbon catabolism, lower glycolysis and TCA cycle (Fig. S2 and Supplemental text). Thus, we then tested whether PGRP induces an increase in central carbon catabolism, by measuring phosphoenolpyruvate (PEP) and PEP/pyruvate ratio. PGRP treatment of E. coli resulted in a rapid and significant increase in PEP concentration and PEP/pyruvate ratio with a peak at 5 min (Fig. 4B), which indicates increased glycolysis.

We then tested production of cAMP, because cAMP is required for the activity of Crp (Supplemental text). PGRP treatment of E. coli resulted in a significant increase in intracellular cAMP, which peaked in 5 min (Fig. 4C), which indicates the predicted activation of adenylate cyclase. Thus, the peaks of PGRP-induced increases in the PEP/pyruvate ratio and in cAMP (5 min) preceded and thus were up-stream from the PGRP-induced increases in NADH and H₂O₂, which peaked at 15 min (Figs. 2A and 4A).

PGRP-induced changes in NADH/NAD⁺ and PEP/pyruvate ratios in PGRP-resistant mutants

We next tested the role of NADH-oxidizing respiratory chain enzymes (NDH-1 and NDH-2), TCA cycle enzymes, and their regulators (cAMP and Crp) in PGRP-induced increased production of NADH, PEP, and cAMP by epistatic analysis using deletion mutants that were more resistant to PGRP killing.

The NADH/NAD⁺ ratios were significantly decreased in PGRP-treated sucB and sucD deletion mutants (both Keio and ours) and also in our crp and cyaA mutants, compared with the NADH/NAD⁺ ratios in PGRP-treated parental strains (Fig. 5A). These results are consistent with the predicted accumulation of NAD⁺ and the decreased capacity of these mutants to reduce it to NADH (Fig. S2). Lower NADH/NAD⁺ ratios in these mutants are also consistent with decreased PGRP-induced production of H₂O₂ in these mutants due to lower availability of NADH for the respiratory chain. Both ndh mutants had increased NADH/NAD⁺ ratios in PGRP-treated cells compared with parental strains (Fig. 5A), which reflects significantly decreased ability of ndh mutants to oxidize NADH to NAD⁺. These results suggest that NDH-2 is the main NADH-oxidizing enzyme in PGRP-treated cells, consistent with a high PGRP-induced increase in ndh expression (17-fold, Fig. 3). By contrast, the NADH/NAD⁺ ratios were significantly decreased in PGRP-treated nuoE, nuoG, and nuoEF deletion mutants compared with PGRP-treated parental strains (Fig. 5A), suggesting no significant role of NDH-1 in NADH oxidation in PGRP-treated cells and possibly diminished ability of nuo mutants to generate NADH. Thus, diminished H₂O₂ production and killing in PGRP-treated nuo mutants could be due to decreased NADH/NAD⁺ ratio, similar to the TCA cycle mutants. Changes in the NADH/NAD⁺ ratios correlated better with the PGRP resistant phenotype of the mutants than the absolute levels of NADH, which was especially evident in crp and cyaA mutants (Fig. 5A), because of increased baseline levels of NADH.

Fig. 5. NADH, NAD⁺ and NADH/NAD⁺ ratio (A), phosphoenolpyruvate (PEP), pyruvate, and PEP/pyruvate ratio (B), and cAMP (C) in PGRP-treated E. coli mutants.

Parental E. coli or the indicated deletion mutants from Keio collection (BW25113) or from our laboratory (MG1655) were treated with 100 µg/ml BSA or PGRP for 15 min (A) or 5 min (B and C). The results are means of 3–4 experiments + SEM; * P<0.05, ** P<0.001, PGRP vs BSA; ^ P<0.05, ^^ P<0.001, mutant vs parental strain.

PGRP did not induce an increase in PEP and PEP/pyruvate ratios in nuoE, nuoEF, nuoG, ndh, sucB, and sucD deletion mutants (both Keio and ours), whereas in crp and cyaA mutants there was a decrease in PEP and PEP/pyruvate ratios, in contrast to significant increases in both PEP and PEP/pyruvate ratios in PGRP-treated parental strains (Fig. 5B). These results indicate lower metabolic activity and inability to increase glycolysis following PGRP treatment in all these mutants, consistent with their increased resistance to PGRP killing and decreased PGRP-induced H₂O₂ production (Figs. 1 and 2).

We also tested PGRP-induced induction of cAMP production in PGRP-resistant mutants, because cAMP is required for activation of Crp, which would show whether the defect in these mutants is up-stream or down-stream from Crp. PGRP treatment of nuoE, nuoEF, nuoG, ndh, sucB, and sucD mutants (both Keio and ours), as well as their parental strains, induced similar increases in cAMP, whereas, as expected, cyaA mutant (deficient in adenylate kinase) had very low level of cAMP (Fig. 5C). The crp deletion mutant also had very low level of cAMP, which is unexpected, because Crp is a negative regulator of cyaA. These results confirm that control of cAMP production is up-stream from the reactions catalyzed by nuo, ndh, and suc gene products, consistent with the findings that in E. coli growing on glycerol (which is the energy source in our medium) activation of adenylate cyclase is controlled by glycerol-3-phosphate (Eppler et al., 2002).

PGRP induces a block in respiratory chain

Based on the requirement for the TCA cycle and respiratory NADH oxidoreductases for PGRP-induced generation of H₂O₂, we considered two hypotheses that could explain how H₂O₂ is generated in PGRP-treated E. coli. The first hypothesis postulated that H₂O₂ is produced as a by-product of increased rate of respiration and increased flux of electrons through the respiratory chain to the cytochromes, due to increased supply of NADH and increased oxidation of NADH by the respiratory chain NADH oxidoreductases. This hypothesis predicted that PGRP should induce an increase in O₂ consumption, because a large amount of O₂ would have to be reduced to H₂O to generate an accompanying large increase in H₂O₂ as a by-product of incomplete O₂ reduction (which is an infrequent event).

The second hypothesis postulated that H₂O₂ is produced by diversion of electrons from NADH oxidoreductases directly to O₂ due to a block of electron transfer from NADH oxidoreductases to the cytochromes. This hypothesis predicted that PGRP would cause a decrease in O₂ consumption, because a block in electron transfer to the cytochromes would result in a decrease in cytochrome-catalyzed reduction of O₂ to H₂O.

Thus, we next tested these hypotheses by measuring in real time PGRP-induced changes in O₂ consumption rate with the Seahorse XFp analyzer, which uses label-free solid-state sensor cartridges in a microplate format (Lobritz et al., 2015; Saini et al., 2016). Treatment of E. coli with PGRP induced rapid decrease in O₂ consumption rate, which was already significantly lower at 5 min and continued to decrease to near base-line level for 30 min (Fig. 6A). By contrast, BSA-treated control cells maintained stable O₂ consumption rate throughout the entire 60-min incubation period (Fig. 6A). These results support our second hypothesis and indicate that PGRP causes a block in the respiratory chain, likely leading to electron backup on NADH oxidoreductases.

Fig. 6. Oxygen consumption and extracellular acidification rates in PGRP-treated E. coli.

(A) O₂ consumption rate and (B) extracellular acidification rate were measured simultaneously in real time with the Seahorse XFp analyzer in E. coli MG1655 treated with 100 µg/ml BSA or PGRP. (C) A phenogram of O₂ consumption rate plotted versus extracellular acidification rate illustrates the total cell energy phenotype. The results are means of 6 experiments ± SEM (SEM were within symbols if not visible); * P<0.05, ** P<0.0001, PGRP vs BSA.

Simultaneously with O₂ consumption rate, we also measured in real time the extracellular acidification rate (ECAR, using Seahorse XFp analyzer), which reflects total H⁺ production from carbon catabolism. BSA-treated control cells steadily increased the ECAR throughout the entire 60-min incubation period, which reflected healthy cell growth. By contrast, in PGRP-treated cells, ECAR increased only during the first 5 min of incubation, indicating initial acceleration of metabolism, and then declined and remained low during the remaining incubation (Fig. 6B). Plotting O₂ consumption rate versus ECAR illustrates the total cell energy phenotype and shows high-energy state (respiration and metabolism) for the control cells incubated with BSA, and low-energy phenotype for the PGRP-treated cells (Fig. 6C), consistent with eventual energy depletion and impending death.

PGRP-induced depletion of thiols is mostly independent of respiratory chain, TCA cycle, and generation of H₂O₂

PGRP also induces severe depletion of intracellular thiols in bacteria (referred to as thiol stress), and this thiol depletion is required for efficient bacterial killing by PGRP (Kashyap et al., 2014). However, the mechanism of PGRP-induced depletion of thiols is unknown. Because PGRP does not enter the cytoplasm (Kashyap et al., 2011) and also because PGRP by itself is not a thiol-oxidizing electrophile, the simplest hypothesis to explain PGRP-induced thiol depletion would be through induction of intracellular reactive oxygen species (ROS), as thiol stress is often considered a consequence and a component of oxidative stress (Leichert et al., 2003; Müller et al., 2013). Here we tested this hypothesis for PGRP-induced depletion of thiols, by using deletion mutants for respiratory chain, TCA cycle, their regulators, and other flavoproteins that had severely diminished PGRP-induced H₂O₂ production (Fig. 2B,C).

PGRP treatment of parental strains resulted in 91–93% depletion of thiols. Similar treatment of mutants deficient in respiratory chain genes (nuo and ndh), or in TCA cycle genes (sucB, sucD, icd, and lpdA), or in genes for other redox flavoproteins (frdC and betA), on average resulted in 85 ± 2% depletion of intracellular thiols, which was still significant and only 10% less efficient than in parental strains (Fig. 7). Therefore, 90% of overall thiol depletion was still present in mutants for respiratory chain (nuo and ndh), TCA cycle (sucB, sucD, icd, and lpdA), and other redox flavoproteins (frdC and betA) genes, and thus was mostly independent of these genes. Therefore, this thiol depletion was also mostly independent of PGRP-induced H₂O₂ production, which was abolished or severely diminished in these mutants.

Fig. 7. Depletion of thiols in PGRP-treated E. coli.

E. coli and the indicated deletion mutants from Keio collection (BW25113) or from our laboratory (MG1655) were treated with 100 µg/ml BSA or PGRP for 30 min. The results are means of 3 experiments + SEM; * P<0.05, ** P<0.001, PGRP vs BSA; ^ P<0.05, mutant vs parental strain.

In mutants for the regulators of central carbon metabolism (crp and cyaA), PGRP induced 62% to 54% depletion of thiols, which was still significant, but shows that about 30–40% of PGRP-induced thiol depletion is dependent on cAMP-Crp (Fig. 7).

PGRP-induced increase in intracellular Zn²⁺ is independent of respiratory chain, TCA cycle, and generation of H₂O₂

In addition to strong oxidative and thiol stress, PGRP also induces metal stress, manifested in E. coli by an increase in intracellular free Zn²⁺ and strong up-regulation of metal efflux and detoxification mechanisms (Kashyap et al., 2014). To investigate the role of oxidative stress and respiratory chain, TCA cycle, and other redox flavoproteins in PGRP-induced increase in intracellular free Zn²⁺, we tested induction of intracellular free Zn²⁺ by PGRP in several mutants with deleted genes required for PGRP-induced oxidative stress.

PGRP treatment in mutants deficient in respiratory chain genes (nuo and ndh), in TCA cycle genes (sucB, sucD, icd, and lpdA), in regulators of central carbon metabolism (crp and cyaA), or in genes for other redox flavoproteins (frdC and betA) resulted in similar increases in intracellular free Zn²⁺ as in parental bacteria (Fig. 8A). These results show that PGRP-induced increase in intracellular free Zn²⁺ is independent of the respiratory chain, TCA cycle, cAMP-Crp, and two redox flavoproteins (frdC and betA), and also independent of PGRP-induced production of H₂O₂, which was abolished or severely diminished in these mutants (Figs 2B,C).

Fig. 8. Increase in intracellular free Zn²⁺ in PGRP-treated E. coli and increased resistance of Zn²⁺ importer mutants to PGRP killing.

(A, B) E. coli BW25113 and MG1655 and the indicated deletion mutants were treated with 100 µg/ml BSA or PGRP for 2 hrs and intracellular free Zn²⁺ concentration was measured by flow cytometry with Zinpyr-1 probe. The results are shown as mean fluorescence intensity (MFI) of 3 experiments + SEM. (C) E. coli MG1655 or the indicated deletion mutants for Zn²⁺ importer genes were treated with 100 µg/ml of BSA or PGRP and the numbers of surviving bacteria were determined by colony counts. The results are expressed as percent of initial inoculum (100%) and are means of 3 experiments ± SEM (SEM were within symbols if not visible). * P<0.05, ** P<0.001, PGRP vs BSA; ^ P<0.05, PGRP-treated mutant vs parental strain.

Our previous results also suggested that the import of extracellular Zn²⁺ may be the source of PGRP-induced increase in intracellular free Zn²⁺, because chelating extracellular Zn²⁺ abolished the PGRP-induced increase in intracellular free Zn²⁺ and prevented PGRP-induced killing (Kashyap et al., 2011, 2014). If this was the case, we predicted that deleting Zn²⁺ importers in bacteria would diminish PGRP-induced increase in intracellular free Zn²⁺. Indeed, PGRP-induced increase in intracellular free Zn²⁺ was significantly diminished in mutants lacking genes for Zn²⁺ importers, znuABC or zupT, and also in a double mutant lacking both Zn²⁺ importers (znuABC and zupT), in which PGRP-induced Zn²⁺ increase was reduced by 60% (Fig. 8B). By contrast, deleting other non-specific Zn²⁺ importers (zinT, mntH, or pitA) had no or little effect. These results indicate that PGRP-induced increase in intracellular free Zn²⁺ to a large extent requires import of extracellular Zn²⁺, and that ZnuABC and ZupT importers have the greatest contribution to the PGRP-induced Zn²⁺ import. These PGRP-induced increases in Zn²⁺ were not completely abolished in znuABC and zupT mutants, because nonspecific metal importers, such as MntH and PitA, and Zn²⁺-binding protein ZinT (Kershaw et al., 2007) are able to import Zn²⁺ and may to some extent compensate in the znuABC and zupT double mutant. Because Zn²⁺ is essential for growth, mutants completely lacking all Zn²⁺ importers are not viable and cannot be obtained.

To further determine the role of Zn²⁺ import in PGRP-induced killing, we next determined the sensitivity to PGPR killing of mutants lacking the specific Zn²⁺ importers (znuABC and zupT). Mutants lacking znuABC or zupT and a double mutant lacking both znuABC and zupT were all significantly more resistant to PGRP killing (Fig. 8C), which shows that Zn²⁺ import is important for both PGRP-induced metal stress and killing.

CpxRA is a negative regulator of PGRP-induced oxidative stress, but not thiol and metal stress

We previously proposed that in E. coli CpxRA two-component system positively controls PGRP-induced bacterial killing, based on activation of CpxRA in PGRP-treated cells and higher resistance of cpxA and cpxR deletion mutants to PGRP killing compared with the parental MG1655 strain (Kashyap et al., 2011). Therefore, here we tested whether some or all of the PGRP-induced pathways that result in oxidative, thiol, and metal stress are regulated by CpxRA.

To verify the role of CpxRA in resistance to PGRP killing, we first compared the sensitivity to PGRP killing of cpxA and cpxR deletion mutants from four sources. All four cpxA deletion mutants had significantly greater resistance to PGRP killing than the parental strains (Fig. 9A). By contrast, three cpxR deletion mutants and also our cpxRA mutant, all had similar sensitivity to PGRP killing as their parental strains. Only one cpxR deletion mutant (from Boston University) was significantly more resistant to PGRP killing than its parental strain (Fig. 9A), similar to our previously reported results for the same mutant (Kashyap et al., 2011). We conclude that cpxA, but not cpxR mutants are more resistant to PGRP killing.

Fig. 9. CpxRA is a negative regulator of PGRP-induced killing and oxidative stress, but not thiol and metal stress.

(A) Parental E. coli or cpxA, cpxR, and cpxRA deletion mutants from the indicated sources were treated with BSA or PGRP (at 100 µg/ml for MG1655 and BW25113, or 300 µg/ml for MC4100) and the numbers of surviving bacteria were determined by colony counts. The parental strains had different sensitivity to PGRP killing (Fig. S4), which was the reason for using different concentrations of PGRP to obtain similar killing of the parental strains. The results are expressed as percent of initial inoculum (100%) and are means of 4 experiments ± SEM (SEM were within symbols if not visible). (B–H) E. coli MG1655 and cpxA, cpxR, and cpxRA deletion mutants from our laboratory were treated with 100 µg/ml BSA or PGRP and assayed for H₂O₂ (at 15 min, B), or NADH and NADH/NAD⁺ ratios (at 15 min, C), or phosphoenolpyruvate (PEP) and PEP/pyruvate ratios (at 5 min, D), or cAMP (at 5 min, E), or thiols (at 30 min, F), or intracellular free Zn²⁺ (at 2 hrs, G), or cpxP expression (at 15 min, H). The results are means of 3–4 experiments + SEM. ^ P<0.05, ^^ P<0.001, mutant vs parental strain; * P<0.05, ** P<0.001, PGRP vs BSA.

We next determined the role of CpxRA in PGRP-induced oxidative, thiol, and metal stress, and also in the reactions leading to the oxidative stress, using our cpxA, cpxR, and cpxRA mutants.

PGRP-induced increase in H₂O₂ production was abolished in the cpxA mutant, but not in the cpxR and cpxRA mutants, although it was somewhat lower in the cpxRA mutant than in the parental strain (Fig. 9B). PGRP treatment also did not induce an increase in NADH level and NADH/NAD⁺ ratio in the cpxA mutant, relative to control BSA treatment, because NADH level and NADH/NAD⁺ ratio were already high in BSA-treated cpxA mutant. By contrast, PGRP treatment induced similar increases in NADH level and NADH/NAD⁺ ratio in cpxR and cpxRA mutants as in the parental strain (Fig. 9C). PGRP treatment also did not increase cAMP concentration in the cpxA mutant, relative to control BSA treatment, because cAMP concentration was also already high in BSA-treated cpxA mutant, whereas again PGRP treatment induced similar increases in cAMP concentration in cpxR and cpxRA mutants as in the parental strain (Fig. 9E). However, PEP and PEP/pyruvate ratio were only increased by PGRP treatment in the parental strain, but not in cpxA, cpxR, and cpxRA mutants (Fig. 9D). Thus, our results with H₂O₂, NADH, and cAMP production correlate well with the increased resistance to PGRP killing of cpxA mutant and similar sensitivity to PGRP killing of the parental strain and cpxR and cpxRA mutants. They also further confirm the requirement for oxidative stress in PGRP killing.

PGRP treatment resulted in similar 90% depletion of thiols in the parental strain and both cpxR and cpxRA mutants, and 66% depletion of thiols in cpxA mutant (Fig. 9F). Thus, PGRP induced significant depletion of thiols in all cpx mutants, although this depletion was 25% less severe in cpxA mutant than in the parental strain and cpxR and cpxRA mutants. These results again correlate with the increased resistance of cpxA mutant to PGRP killing. Deletion of cpxA, cpxR, or cpxRA had no effect on PGRP-induced increase in the intracellular free Zn²⁺ concentration (Fig. 9G).

Based on the reported constitutive activation of CpxR in the cpxA deletion mutant (Mahoney and Silhavy, 2013), our results suggested that activation of CpxR could significantly attenuate PGRP-induced killing and oxidative stress (H₂O₂ production and associated increase in NADH and cAMP), and partially attenuate thiol stress (depletion of thiols), but not metal stress (intracellular Zn²⁺ increase).

To verify this hypothesis, we next determined whether CpxR was indeed constitutively activated in our cpxA mutant, by assaying the expression of cpxP, a member of the CpxR regulon solely controlled by CpxR. Indeed, cpxA mutant had approximately 10-fold higher level of expression of cpxP relative to the parental strain (which was statistically significant) (Fig. 9H). Thus, these results show high constitutive activation of CpxR, regardless of the treatment with BSA (unstimulated negative control), PGRP, or gentamicin (positive control).

We have previously shown that CpxRA was activated after 60 min of PGRP treatment (Kashyap et al., 2011). Here, we tested whether CpxRA was activated in PGRP-treated cells at the time of the maximum metabolic and oxidative stress response (15 min). In parental MG1655 cells, 15-min PGRP treatment increased expression of cpxP 2.7-fold relative to BSA treatment, which was significant, but very low, compared with over 30-fold increase in cpxP expression induced by gentamicin and 10-fold higher constitutive CpxR activation in the cpxA mutant (Fig. 9H). These results indicate only very low activation of CpxRA by PGRP at this early time point. Of note, in both cpxR and cpxRA mutants cpxP was not expressed, confirming the requirement for CpxR in cpxP expression and validating the use of cpxP expression as an assay for CpxR activation.

In conclusion, high constitutive activation of CpxR and no PGRP-induced production of H₂O₂ in the cpxA mutant indicate that CpxR is a negative regulator of PGRP-induced oxidative stress, which is the likely reason for the increased resistance of cpxA mutant to PGRP killing.

Discussion

Our results indicate that in E. coli PGRP induces increased production of H₂O₂ and oxidative stress through a block in respiratory chain, and that PGRP-induced generation of H₂O₂ depends on increased supply of NADH from cAMP-Crp-controlled glycolysis and TCA cycle (Fig. 10). Efficient respiration in growing cells with quick flow of electrons through the respiratory chain generates only very small amounts H₂O₂, because cytochromes have high affinity for O₂ and efficiently reduce O₂ to H₂O (Imlay, 2013; Karp et al., 2017). Therefore, in an intact electron transport chain, a large increase in H₂O₂ production would require an equally large increase in respiration. However, PGRP induces decreased respiration with a large simultaneous increase in H₂O₂ production. How can a decreased O₂ consumption rate result in increased H₂O₂ production? To explain these results we propose that PGRP treatment causes a block in electron flow through the respiratory chain from NADH oxidoreductases to the cytochromes and O₂, leading to a backup of electrons on solvent-exposed flavin cofactors of NADH oxidoreductases. We further propose that the reduced flavins on these enzymes then auto-oxidize through adventitious contact with O₂, but because they have low affinity for O₂ and transfer only one or two electrons to O₂, O₂ is incompletely reduced to O₂⁻ or H₂O₂, followed by rapid dissociation of H₂O₂ from the enzyme (Messner and Imlay, 1999; Seaver and Imlay, 2004; Esterhazy et al., 2008; Imlay, 2013). Because most of the electrons backed up on the flavins are prematurely diverted to O₂, a large increase in H₂O₂ is accompanied by a decrease in total O₂ consumption. The electrons for O₂ reduction are supplied by the NADH generated from a transient increase in carbon catabolism, which could be an adaptive response of bacteria to the PGRP-induced respiratory block (Fig. 10).

Fig. 10. Proposed pathways leading to PGRP-induced oxidative, thiol, and metal stress in E. coli.

Bactericidal PGRP induces oxidative stress through a block in respiratory chain, which likely diverts electrons from respiratory chain NADH oxidoreductases to O₂, generates H₂O₂, and results in decreased respiration. Production of H₂O₂ depends on increased supply of NADH from cAMP-Crp-controlled glycolysis and TCA cycle. CpxRA two-component system is a negative regulator of PGRP-induced oxidative stress. PGRP-induced thiol stress (depletion of thiols) and metal stress (increase in intracellular free Zn²⁺ through influx of extracellular Zn²⁺) are mostly independent of oxidative stress. This proposed sequence of events was determined by epistatic and biochemical analyses and does not necessarily reflect direct interactions. Broken line indicates a partial effect.

Single deletions of genes for either one of the respiratory NADH oxidoreductases (nuo or ndh), or for the TCA cycle enzymes (sucB, sucD, icd, or lpdA), or crp or cyaA (positive regulators of carbon catabolism), all have similar effects of abolishing PGRP-induced H₂O₂ production (Fig. 2B,C) and decreasing bacterial killing (Fig. 1). These results suggest that NDH-2 could be the main enzyme that accepts electrons from NADH in PGRP-treated cells, consistent with high induction of ndh expression by PGRP treatment, and with increased NADH/NAD⁺ ratio only in PGRP-treated ndh mutant. Thus, PGRP-treated ndh mutant could have a deficient generation of H₂O₂, because it lacks this main enzyme that accepts electrons from NADH used for reduction of O₂ to H₂O₂. By contrast, both TCA cycle mutants and nuo mutants have decreased NADH/NAD⁺ ratios in PGRP-treated cells (Fig. 5A). This is consistent with the expected lower conversion of NAD⁺ to NADH in TCA cycle mutants. However, we have no explanation for the decreased NADH/NAD⁺ ratios in PGRP-treated nuo mutants. Regardless of the regulatory mechanism, this lower supply of NADH in TCA cycle and nuo mutants would result in a decreased transfer of electrons from NADH to NADH oxidoreductases and their lower capacity to reduce O₂ to H₂O₂. This explanation would account for deficient H₂O₂ production in ndh, nuo, and TCA cycle mutants and offer one possibility why nuo and ndh do not compensate for each other.

We also identified two additional flavoprotein sources of PGRP-induced H₂O₂: fumarate reductase (Frd) and choline dehydrogenase (BetA/CDH), as deletion of their genes also causes increased resistance to PGRP killing (Fig. 1A and Table S1) and diminished PGRP-induced production of H₂O₂, although significantly less than deletion of nuo or ndh (Fig. 2B). Fumarate reductase (Frd) functions mostly in anaerobic respiration, but may be a significant source of H₂O₂ under aerobic conditions (Korshunov and Imlay, 2010), as is the case here in PGRP-treated cells. BetA is another oxidoreductase linked to electron transfer and is oxygen-dependent (Landfald and Strøm, 1986). However, it is unclear how Frd and BetA function in PGRP-induced generation of H₂O₂.

In addition to oxidative stress, effective PGRP killing also requires induction of thiol and metal stress in bacteria (Kashyap et al., 2014). Therefore, in this study, we also analyzed whether thiol and metal stress are induced by PGRP through the same mechanism as the oxidative stress.

Thiols maintain the redox state in the cells and protect cells from oxidative damage. Depletion of intracellular thiols, known as thiol (or disulfide) stress, can be induced by oxidative stress through ROS-mediated oxidation of thiols, or by electrophiles, such as quinones, aldehydes, and azo- or amido-compounds. These electrophiles can also induce ROS or can directly modify thiols by oxidation, alkylation, or arylation (Liebeke et al., 2008; Pöther et al., 2009; Müller et al., 2013; Leichert et al., 2003; Faulkner and Helmann, 2011).

Our results show that PGRP-induced depletion of thiols is not caused by and is mostly independent of PGRP-induced oxidative stress, because PGRP treatment of mutants with deleted respiratory chain oxidoreductases (nuo and ndh), TCA cycle enzymes (sucB, sucD, icd, and lpdA), or other oxidoreductases (frd and betA), in which PGRP-induced production of H₂O₂ was abolished or decreased, still resulted in severe depletion of intracellular thiols (Fig. 7). Oxidative-stress independence of PGRP-induced thiol stress is also supported by our previous results showing that the amount of intracellular H₂O₂ induced by PGRP in bacteria is not sufficient to significantly deplete intracellular thiols (Kashyap et al., 2014). PGRP-induced depletion of thiols, however, is partially dependent on cAMP-Crp (Fig. 7), which is not surprising, because cAMP-Crp regulates over 300 genes (Görke and Stülke, 2008; Karp et al., 2017; Shimada et al., 2011), many of which may not be involved in oxidative stress. Because PGRP by itself is not a thiol-reactive electrophile and also because it does not enter the cytoplasm (Kashyap et al., 2011), our results suggest that PGRP triggers thiol depletion through additional currently unidentified pathways. One possibility is that PGRP induces excessive production of endogenous electrophiles, such as quinones, which would then directly deplete thiols.

It is generally accepted that oxidative and thiol stress result in release of metals from proteins and/or increased metal import, which increase intracellular concentration of free metals, including Zn²⁺, increases metal toxicity, and induce expression of genes for metal detoxification and efflux (Leichert et al., 2003; Harrison et al., 2009; Faulkner and Helmann, 2011). Our previous results were consistent with this notion, because PGRP induced strong oxidative, thiol, and metal stress, accompanied by a dramatic increase in intracellular free (labile) Zn²⁺ and expression of genes for the efflux and detoxification of zinc, copper, arsenite, and other metals controlled by metal-responsive or stress-responsive regulators (CueR, ArsR, SoxR, RcnR) (Kashyap et al., 2014).

Thus, our original hypothesis was that PGRP-induced metal stress was a consequence of PGRP-induced oxidative and/or thiol stress. However, our subsequent and current results do not support this hypothesis and indicate that PGRP-induced increase in intracellular free Zn²⁺ is independent of PGRP-induced oxidative and thiol stress. This conclusion is based on the following results. First, neither paraquat (a direct inducer of O₂⁻ and H₂O₂) nor diamide (a thiol-depleting electrophile) induce any increases in intracellular Zn²⁺ in E. coli and B. subtilis, despite inducing a large increase in intracellular H₂O₂ or severe depletion of thiols (Kashyap et al., 2014). And second, PGRP induces similar increases in intracellular Zn²⁺ in parental strains and in all deletion mutants lacking respiratory chain oxidoreductases (nuo and ndh), or TCA cycle enzymes (sucB, sucD, icd, and lpdA), or regulators of central carbon catabolism (crp and cyaA), or other oxidoreductases (frd and betA) (Fig. 8A), in which PGRP-induced production of H₂O₂ was abolished or decreased (Fig. 2B,C).

Our current and previous results also indicate that PGRP-induced increase in intracellular Zn²⁺ is mostly due to an influx of extracellular Zn²⁺ into the cells, rather than a release from intracellular sources, because this increase was significantly reduced in deletion mutants lacking the ZnuABC and ZupT Zn²⁺ importers (Fig. 8B). Moreover, as already indicated, increasing intracellular H₂O₂ with paraquat or depleting intracellular thiols with diamide do not induce an increase in intracellular Zn²⁺, which would be expected if oxidative or thiol stress were causing release of Zn²⁺ from intracellular sources. Furthermore, chelating extracellular Zn²⁺ with EGTA (whose log stability constant for Zn²⁺ is 12.9) completely prevents PGRP-induced increase in intracellular Zn²⁺ and abolishes bactericidal activity of PGRP (Wang et al., 2007; Kashyap et al., 2014). The requirement for influx of extracellular Zn²⁺ for PGRP killing is further demonstrated by increased resistance to PGRP killing of mutants deficient in Zn²⁺ importers (znuABC and zupT) (Fig. 8C).

CpxRA two-component system is a transmembrane sensor of envelope stress and is activated by a variety of insults, including misfolded proteins in the periplasm and in the cytoplasmic and outer membranes, alterations in membrane proteins, lipids, or peptidoglycan, and also by many other stresses, such as pH, metals, osmolarity, adhesion, EDTA, indole, ethanol, other metabolites, and PGRPs (Zhou et al., 2011; Hunke et al., 2012; Raivio, 2014; Evans et al., 2013; Bernal-Cabas et al., 2015; Keller et al., 2015; Delhaye et al., 2016; Kashyap et al., 2011). However, the exact mechanism of CpxRA activation by such a large variety of stimuli still remains elusive (Raivio, 2014; Delhaye et al., 2016).

Here we show that CpxR, constitutively activated in the cpxA mutant, is a negative regulator of PGRP-induced increases in H₂O₂, NADH, NADH/NAD⁺ ratio, and cAMP. By contrast, PGRP-induced thiol stress is mostly insensitive, and PGRP-induced increase in intracellular Zn²⁺ is completely insensitive to this negative regulation by CpxR. Although we previously reported that PGRP activates CpxRA, this activation occurs late, at 60 min of PGRP exposure (Kashyap et al., 2011), which is much too late to prevent PGRP-induced H₂O₂ production that peaks at 15 min of PGRP exposure (Fig. 2A). For this reason CpxRA does not protect parental cells from PGRP-induced oxidative stress and killing (unless CpxR is artificially pre-activated by cpxA deletion).

Our results suggest that PGRP induces oxidative, thiol, and metal stress through three independent pathways. These results are consistent with our previous and current data showing that inhibiting one stress (oxidative, thiol, or metal) with chemical inhibitors makes PGRPs only bacteriostatic or less bactericidal (Kashyap et al., 2011, 2014), and that single deletion mutants can only be partially, but not fully, resistant to PGRP killing. The reason for this partial resistance is that these inhibitors or single mutations can only disable one stress pathway, which only reduces, but does not completely abolish antibacterial activity of PGRP, as each stress on its own is bacteriostatic, and combining two and three stresses progressively amplifies their antibacterial effect (Kashyap et al., 2014). This strategy assures that these innate immune proteins, conserved in evolution from insects to mammals, do not easily lose their antibacterial activity.

The proposed subsequent steps that lead to PGRP-induced bacterial death after induction of oxidative, thiol, and metal stress, include: (i) generation of toxic HO^• from H₂O₂, which starts at 30 min and peaks in 2–3 hrs of PGRP exposure (likely involving Fe²⁺ from Fe-S clusters as an electron donor) and damages bacterial DNA (Kashyap et al., 2011); (ii) cessation of all biosynthetic reactions in bacteria, which starts at 30 min of PGRP exposure (Kashyap et al., 2011), likely due to the lack of energy following a decrease in respiration rate and catabolism, which happen within 30 min of PGRP exposure (Fig. 6); and (iii) membrane depolarization, which starts at 1 hr and peaks at 2–3 hrs of PGRP exposure (Kashyap et al., 2011), likely due to decrease in proton translocation by respiratory chain and energy depletion. These events are irreversible, because diluting bacteria into a fresh medium at 2–3 hrs does not allow for their growth, although these bacteria are still not dead even at 6 hrs of PGRP exposure, because their membranes are still not permeabilized (Wang et al., 2007; Kashyap et al., 2011).

Manipulations of bacterial metabolism and ROS production are often studied with a goal to enhance effectiveness of antibiotics. Increased carbon catabolism and respiration are associated with bacterial killing by bactericidal antibiotics (Dwyer et al., 2014; Belenky et al., 2015; Lobritz et al., 2015). Similar to our results with PGRP, E. coli deletion mutants in TCA cycle genes are more resistant to antibiotic killing than parental strains (Kohanski et al., 2007). However, opposite effects of metabolism on the sensitivity to antibiotics were also reported, as deleting respiratory chain or TCA cycle genes enhanced sensitivity to exogenous oxidants and β-lactam and fluoroquinolone antibiotics (Brynildsen et al., 2013). By contrast, in our experiments similar mutations enhanced resistance to PGRP, because respiratory chain and TCA cycle are required to generate PGRP-induced oxidative stress. Increased bacterial killing by aminoglycoside antibiotics in bacteria with enhanced TCA cycle and respiration was due a greater uptake of aminoglycosides driven by increased proton motive force (Allison et al., 2011; Peng et al., 2015). Furthermore, treatment of E. coli with bactericidal antibiotics induced a transient 5-fold decrease in the NADH/NAD⁺ ratio (Kohanski et al., 2007), whereas PGRP induced a transient increase in NADH/NAD⁺ ratio (Fig. 4A). Also, bactericidal antibiotics increased O₂ consumption rate (Lobritz et al., 2015), whereas PGRP, which is also bactericidal, decreased it (Fig. 6). Moreover, many effects of antibiotics on the metabolism seem to be the consequences or indirect responses to the primary effects of antibiotics, as they happen late in antibiotic-induced killing (30 to 90 min) (Kohanski et al., 2007; Belenky et al., 2015). By contrast, the kinetics of PGRP-induced changes in metabolism and respiration reported here are much faster (5 min). These changes are the first detectable effects of PGRP, which argues for the block in respiratory chain and increased catabolism as the primary effects of PGRP on bacteria. There are many additional differences between bacterial killing by PGRP and antibiotics, which were discussed previously (Kashyap et al., 2011, 2014).

In conclusion, in E. coli bactericidal PGRP induces severe oxidative stress through a block in respiratory chain, which likely diverts electrons from respiratory chain NADH oxidoreductases to O₂, generates H₂O₂, and results in decreased respiration. Production of H₂O₂ depends on increased supply of NADH from cAMP-Crp-controlled glycolysis and TCA cycle (Fig. 10). CpxRA two-component system is a negative regulator of PGRP-induced oxidative stress. PGRP also induces thiol stress and metal stress, which are mostly independent of oxidative stress and CpxRA. Manipulating these pathways that induce oxidative, thiol, and metal stress in bacteria could be a useful strategy to design new approaches to antibacterial therapy.

Experimental Procedures

Materials

Human recombinant PGLYRP4 was used as a representative bactericidal PGRP, as we have previously shown that all human bactericidal PGRPs had similar activity and the mechanism of bacterial killing (Lu et al., 2006; Wang et al., 2007; Kashyap et al., 2011, 2014). PGLYRP4 was expressed in S2 cells and purified as previously described (Lu et al., 2006; Wang et al., 2007) in a buffer containing 10 mM TRIS (pH 7.6), with 150 mM NaCl, 10 µM ZnSO₄, and 10% glycerol. Paraquat (methyl viologen) was from Acros Organics. Fatty acids-free purified bovine serum albumin (BSA, used as a negative control), and other reagents were from Sigma-Aldrich, unless otherwise indicated.

Bacteria, growth, and media

The entire Keio collection of 3884 E. coli K-12 in-frame single-gene knockout mutants was obtained from the National BioResource Project, National Institute of Genetics, Japan (Baba et al., 2006). Individual E. coli strains used in this study are listed in Table S3 (Baba et al., 2006; Yun et al., 2005; Kohanski et al., 2008; Sabri et al., 2009; Mahoney and Silhavy, 2013). Bacteria were grown aerobically overnight at 37°C in an orbital shaker (250 rpm) in LB, diluted to OD₆₆₀ = 0.02 into fresh LB and grown to OD₆₆₀ = 0.6. Bacteria were centrifuged and suspended in the “Assay Medium”, which was the same for all the killing or metabolic experiments here and in our previous killing experiments on E. coli (Wang et al., 2007; Kashyap et al., 2011, 2014). This Assay Medium contained (final concentrations): 5 mM TRIS (pH 7.6) with 2.5% glycerol, 150 mM NaCl, 5 µM ZnSO₄, and 2% of 100% LB. E. coli MG1655 generation time in this assay medium at 37°C was 30 min. The change in the medium from LB to the Assay Medium did not create any significant stress for bacteria, as they immediately (within <1 min) resume their metabolic activity, as measured by respiration and extracellular acidification rates (Fig. 6A,B), and placing the cells in this medium did not induce oxidative, thiol, or metal stress. Also, pre-incubation of cells in this medium for 30 min did not change their response to PGRP, as measured by respiration, extracellular acidification rate, H₂O₂ production, and killing.

Screening of Keio collection for mutants resistant to PGRP killing

To screen the entire E. coli Keio library of 3884 single gene deletion mutants to find mutants with increased resistance to PGRP killing, we grew each mutant aerobically at 37°C on a shaker in 100 µl of LB in 96-well plates to an average approximate OD₆₆₀ of 0.12. Then we pooled equal volumes of cultures of 10–12 mutants, and treated 20 µl of each pool of mutants diluted 1:2000 times into the Assay Medium in a 96-well plate with 100 µg/ml human PGRP aerobically for 3 hrs at 37°C, which is sufficient to kill over 99.5% of E. coli K-12 strain. We then added 180 µl of fresh LB into each well with the treated pools of mutants, re-incubated them at 37°C aerobically, and measured the growth of surviving bacteria by following changes in optical density (OD₆₆₀) every 10 min for 16 hrs. This method yielded growth curves for each pool of mutants and the curves with accelerated growth reflected either more bacteria that survived PGRP killing or presence of faster growing mutants in this pool (Fig. S1A). We then selected 24 pools of mutants with the most accelerated growth curves for the second round of screening, which was performed the same as the first screening described above, but in this screening to prepare the cultures of pooled mutants, the same numbers of bacteria for each mutant added to each pool were used, calculated based on OD₆₆₀ of each mutant culture. This assured starting PGRP treatment with the same numbers of each mutant in each pool and minimized the effect of differences in the rate of growth of mutants. Then at the end of 3 hrs of PGRP treatment, the numbers of bacteria in each pool were determined by colony counts and compared with the numbers of surviving bacteria in similarly prepared and treated isogenic BW25113 parental strain (Fig. S1B). From this second screening we then selected three pools of mutants with the highest numbers of surviving bacteria with mutant pool/parental strain ratios of 20, 15, and 13 (pools 14, 7, and 13 in Fig. S1B). We then compared PGRP killing of all mutants from these three pools to the killing of the BW25113 parental strain, using colony counts for each mutant grown individually after 3 hrs of treatment with 200 µg/ml PGRP. We also screened by colony counts individual mutants from six additional mutant pools that had the ratios of surviving mutants to parental strain between 8 and 10 (pools 9, 17, 18, 20, 22, and 24 in Fig. S1B), concentrating on mutants for respiratory chain enzymes, TCA cycle enzymes, and several other oxidoreductases, because most of the more resistant mutants belonged to these categories. Out of all the mutants individually tested by colony counts for increased resistance to PGRP killing, there were 14 mutants with statistically significant higher survival than the parental strain and the ratio of survival of PGRP-treated Δ mutant versus parental strain of >10. These mutants are shown in Fig. 1A and Table S1.

PGRP killing assay

E. coli were grown as described above in “Bacteria, growth, and media” section and suspended at ~10⁶ bacteria/ml in 50 µl of fresh warm Assay Medium (5 mM TRIS, pH 7.6, with 150 mM NaCl, 5 µM ZnSO₄, 2.5% glycerol, with 2% of 100% LB) with addition of 100 µg/ml BSA or PGRP and incubated at 37°C aerobically with 250 rpm shaking for 3 hrs. The numbers of bacteria were then determined by colony counts (Kashyap et al., 2011, 2014). Bactericidal activity is defined as an at least 100-fold decrease in the number of inoculated bacteria in 3 hrs.

Construction and verification of deletion mutants

We created gene deletion mutants (Table S3) by replacing the entire coding sequence of the genes with kanamycin resistance cassette using homologous recombination. First, we generated PCR fragments using primers (Table S4) and methodology described by Baba et al. (2006). Briefly, the N-terminal deletion primers had a 50-nt 5’ extension including the gene initiation codon (H1) and a 20-nt sequence 5’-ATTCCGGGGATCCGTCGACC-3’ (P1). The C-terminal deletion primers consisted of 21 nt for the C-terminal region, the termination codon, and 29-nt downstream (H2), and the 20-nt sequence 5’-TGTAGGCTGGAGCTGCTTCG-3’ (P2). PCR reactions were run for 30 cycles in 50 µl volumes using Advantage 2 polymerase (Clontech), 1 ng DNA from individual mutants from the Keio collection, 0.4 µM of each primer, and 200 mM dNTPs. PCR products were digested with DpnI, gel purified, and suspended in nuclease free water. Gene disruption was carried out by the method of Datsenko and Wanner (2000), with slight modifications. Briefly, E. coli MG1655 carrying the Red helper plasmid pKD46 (CGSC # 7669) was grown in 5-ml SOB culture with ampicillin and L-arabinose at 30°C overnight. Next day, the culture was diluted to OD₆₆₀ = 0.02 in fresh SOB medium and allowed to grow to OD₆₀₀ = 0.6. Electro-competent cells were made by concentrating the cells 100-fold and washing three times with ice-cold 10% glycerol. Electroporation was done in Gen Pulser Xcell™ electroporation system (Bio-Rad) according to the manufacturer’s instructions by using 50 µl of cells and 10–100 ng of DpnI-treated gel-purified PCR product. Transformed cells were recovered in SOC media with ampicillin and L-arabinose at 30°C while shaking, overnight. The next morning, the transformation mix was spread on LB agar plates with 30 µg/ml kanamycin. Resistant colonies were picked after overnight incubation at 37°C. Removal of the pKD46 plasmid was confirmed by the lack of growth on LB/Amp plates at 30°C. The correct chromosomal deletions were first verified by colony PCR. A freshly isolated colony was suspended in 10 µl nuclease free water and 5 µl of this bacterial suspension was used for two 25 µl PCRs following a 4-min hot-start at 94°C with kanamycin-specific primers kan-F or kan-R and gene-specific primers F or R (Table S4) as described by Datsenko and Wanner (2000). The gene deletions in each mutant were then further verified by capillary sequencing using gene-specific F and R primers.

H₂O₂ assay

H₂O₂ production was measured by a modification of our previous assay using Amplex Red Hydrogen Peroxide/Peroxidase Assay Kit (InVitrogen/Molecular Probes), based on enzymatic detection of H₂O₂ by horseradish peroxidase (Kashyap et al., 2014). E. coli were grown and suspended in 50 µl of Assay Medium at OD₆₆₀ = 0.15 as described above in “Bacteria, growth, and media” section, with addition of 100 µg/ml BSA (control) or PGRP, or 100 µM paraquat. Bacteria were then incubated aerobically at 37°C on a shaker for 15 min (unless otherwise indicated). The reactions were then stopped by lysing bacteria with 40 mM HEPES, 4 mM EDTA, 400 mM KCl, 0.2% Triton X-100, pH 8.1 in a bath sonicator on ice for 5 min, and total amount of H₂O₂ was determined by measuring fluorescence according to the manufacturer’s instructions. In this procedure, detection of H₂O₂ after quick lysis of bacteria does not require the use of Hpx⁻ mutant E. coli strain, which we used previously (Kashyap et al., 2014). H₂O₂ concentration was calculated from a standard curve and reported as µM (µmol/liter) for the original bacterial incubation mixture.

NADH and NAD⁺ assays

E. coli were grown and suspended in 150 µl of Assay Medium at OD₆₆₀ = 0.3 as described above in “Bacteria, growth, and media” section, with addition of 100 µg/ml BSA (control) or PGRP. Bacteria were incubated aerobically at 37°C on a shaker for 15 min (unless otherwise indicated), 100 µl was used for NADH and 50 µl for NAD⁺ assay. The reactions were stopped by addition of 100 mM NaOH for NADH assay or 100 mM HCl for NAD⁺ assay in 40 mM HEPES, 4 mM EDTA, 400 mM KCl, 0.2% Triton X-100, and the cells were sonicated and heated at 60°C for 5 min. Then, NADH or NAD⁺ concentrations were determined using EnzyChrom NAD⁺/NADH Assay Kit (BioAssay Systems) according to the manufacturer’s instructions. The assay is based on lactate dehydrogenase cycling reaction, in which NADH reduces a formazan reagent. NADH and NAD⁺ concentrations were calculated from a standard curve and reported as µM (µmol/liter) for the original bacterial incubation mixture.

Phosphoenolpyruvate (PEP) and pyruvate assays

E. coli were grown and suspended in 100 µl of Assay Medium at OD₆₆₀ = 0.15 as described above in “Bacteria, growth, and media” section, with addition of 100 µg/ml BSA (control) or PGRP. Bacteria were incubated aerobically at 37°C on a shaker for 5 min (unless otherwise indicated). The reactions were then stopped by adding ice-cold 3 M HClO₄ and vortexing, followed by neutralization with 3 M KHCO₃. The concentrations of extracted PEP and pyruvate were then measured in the cell lysates using PEP Fluorometric Assay Kit and Pyruvate Fluorometric Assay Kit (BioVision), respectively, according to the manufacturer’s instructions. PEP and pyruvate concentrations were calculated from a standard curve and reported as µM (µmol/liter) for the original bacterial incubation mixture.

cAMP assay

E. coli were grown and suspended in 50 µl of Assay Medium at OD₆₆₀ = 0.15 as described above in “Bacteria, growth, and media” section, with addition of 100 µg/ml BSA (control) or PGRP. Bacteria were incubated aerobically at 37°C on a shaker for 5 min (unless otherwise indicated). The reactions were then stopped by adding lysis buffer (40 mM HEPES, 4 mM EDTA, 400 mM KCl, 0.2% Triton X-100) and boiling for 5 min. Then, cAMP concentrations were determined using luminescent cAMP-Glo™ Assay Kit (Promega) according to the manufacturer’s instructions. The assay is based on specific activation of protein kinase A by cAMP. cAMP concentration was calculated from a standard curve and reported as nM (nmol/liter) for the original bacterial incubation mixture.

Thiols assay

Depletion of thiols (all reduced sulfhydryl groups) was measured as previously described (Kashyap et al., 2014). E. coli were grown and suspended in 50 µl of Assay Medium at OD₆₆₀ = 0.15 as described above in “Bacteria, growth, and media” section, with addition of 100 µg/ml BSA (control) or PGRP. Bacteria were incubated aerobically at 37°C on a shaker for 30 min. The reactions were then stopped by lysing bacteria with 40 mM HEPES, 4 mM EDTA, 400 mM KCl, 0.2% Triton X-100, pH 8.1 in a bath sonicator on ice for 5 min, and total amounts of reduced thiols were determined using fluorescent Measure-iT Thiol Assay Kit (InVitrogen/Molecular Probes) according to the manufacturer’s instructions. Total concentration of thiols was calculated from a standard curve using glutathione and reported as µM (µmol/liter) for the original bacterial incubation mixture.

Zn²⁺ assay

Intracellular free (labile) Zn²⁺ was measured as previously described (Kashyap et al., 2014). E. coli were grown and suspended in 50 µl of Assay Medium to ~2×10⁷ E. coli/ml as described above in “Bacteria, growth, and media” section, with addition of 100 µg/ml BSA (control) or PGRP. Bacteria were incubated aerobically at 37°C on a shaker for 2 hrs. Bacteria were then washed and incubated with Zinpyr-1 (from Santa Cruz, dissolved in DMSO with 20% Pluronic F-127, 5 µM final concentration) for 15 min at 37°C. Bacteria were then washed and analyzed by flow cytometry using MACSQuant (Miltenyi) flow cytometer and FITC excitation and emission settings. The results are reported as mean fluorescence intensity (MFI) ± SEM, normalized for the bacterial cell size, measured by forward scatter (FSC-A), as described below. We performed several controls to verify that our results truly reflect changes in intracellular free Zn²⁺ concentration. (i) We previously confirmed the linearity of Zn²⁺ detection with Zinpyr-1 by spiking our bacteria-PGRP mixtures with different ratios of free Zn²⁺ and EDTA (a divalent chelator with high affinity for Zn²⁺ with 16.6 log dissociation constant), and assaying Zn²⁺ concentrations using the Zinpyr-1 probe and fluorescence spectroscopy (with Molecular Devices Gemini EM Spectrofluorimeter). The detected Zn²⁺ concentrations were in good agreement with the calculated free Zn²⁺ concentrations (Kashyap et al., 2014). (ii) We excluded the possibility that increased intracellular Zinpyr-1 accumulation was due to an increase in membrane permeability following PGRP treatment, because we previously reported that treatment with bactericidal concentration of PGRP did not increase membrane permeability for at least 6 hrs in E. coli (Wang et al., 2007) or in Gram-positive bacteria (Lu et al., 2006; Kashyap et al., 2011). (iii) We previously confirmed that the fluorescent signal is due to increase in intracellular Zn²⁺ that enters the cells from outside, using membrane-permeable and membrane non-permeable Zn²⁺ chelators and adding extracellular Zn²⁺ (Kashyap et al., 2014). (iv) Bacterial cell filamentation or other increase in cell volume, often caused by many antibiotics, can result in increased fluorophore signal per cell when measured by flow cytometry (Paulander et al., 2014). For this reason, here we verified by dark field microscopy that treatment of E. coli with 100 µg/ml of PGRP for 2 hrs at 37°C had no effect on the bacterial shape and size, compared with BSA-treated E. coli (Fig. S5). Increased cell volume could be also due to osmotic swelling that precedes osmotic lysis, but PGRPs do not induce osmotic lysis of bacteria either (Lu et al., 2006; Wang et al., 2007; Kashyap et al., 2011). Nevertheless, to correct for any possible changes in the cell volume in PGRP-treated bacteria, here we normalized the Zinpyr-1 fluorescence signals to the cell size, as measured by FSC-A, by multiplying fluorescence signal for PGRP-treated cells by the ratio of median FSC-A for BSA-treated cells to median FSC-A for PGRP-treated cells.

Gene expression arrays

Preparation of the whole genome expression arrays was described previously (Kashyap et al., 2014) and the entire data for all the arrays were deposited in NCBI GEO with accession number GSE44211 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE44211). Here, from these arrays, we present and analyze the expression in PGRP-treated E. coli of up-regulated and down-regulated genes for the central carbon catabolism, which are the subject of this study and were not analyzed previously. Briefly, exponentially growing E. coli MG1655 were suspended in the Assay Medium at OD₆₆₀ = 0.3 as in “Bacteria, growth, and media” section, and incubated aerobically at 37°C with 100 µg/ml albumin (control) or PGRP (PGLYRP4) with 250 rpm shaking. RNA was then extracted using Ambion RiboPure-bacteria RNA extraction kit, cDNA was synthesized with random hexamer primers, fragmented, labeled with terminal transferase and biotin, and hybridized to whole genome Affymetrix E. coli Genome 2.0 Array GPL3154 using Affymetrix Hybridization Oven 640 and Affymetrix GeneChip Fluidics Station 450 and protocols provided by Affymetrix GeneChip Technical Manual. Scanning and data extraction were done using Affymetrix GeneChip Scanner 3000 and protocols provided by Affymetrix GeneChip Technical Manual. cDNA synthesis, labeling, hybridization, and scanning were performed at the Genomic and RNA Profiling Core facility, Baylor College of Medicine, Houston, TX. The entire experiment was repeated 3 times. Hybridization intensity data signals were normalized and analyzed as described (Kashyap et al., 2014) using Affymetrix GeneChip Command Console Software. Signal intensities from 3 experiments were used to calculate fold increases or decreases in gene expression between treated and control groups, and the fold changes in gene expression were calculated using the formula: intensity in treated group/geometric mean of intensity in control (albumin) groups, and reported as means ± SEM. Transformed Ln(signal intensity) values were used for direct statistical comparisons of expression signals between treated and control (albumin) groups.

qRT-PCR

The amounts of mRNA for individual genes were measured using quantitative reverse transcription real-time PCR (qRT-PCR) as previously described (Kashyap et al., 2014). E. coli (300 µl), grown and prepared as in “Bacteria, growth, and media” section, were incubated in the Assay Medium with 100 µg/ml BSA (control), 100 µg/ml PGRP, or 5 µg/ml gentamicin aerobically at 37°C on a shaker for 15 min, and RNA was extracted as described above for gene expression arrays. cDNA was synthesized from 100 ng of RNA using RT² PCR Array First Strand Kit (Qiagen). Gene expression was quantified by qRT-PCR using the ABI 7900 Sequence Detection System with 1 cycle 10-min at 95°C and 40 cycles 15 sec at 95°C and 1 min at 60°C using Qiagen SYBR Green Master Mix and the gene-specific primers (listed in Table S4) or common primers for 16S rRNA from all Eubacteria (ACTCCTACGGGAGGCAGCAGT and ATTACCGCGGCTGCTGGC) as a housekeeping gene. For each gene, ΔCt was calculated followed by normalization to the housekeeping gene, followed by calculation of ΔΔCt for each gene: ΔΔCt = ΔCt1 − ΔCt2, where ΔCt1 is the PGRP-treated bacteria and ΔCt2 is BSA-treated bacteria. This calculation gives the fold increase in expression of each gene in PGRP-treated bacteria versus BSA-treated bacteria. For expression of cpxP, ΔCt1 was the measured sample, and ΔCt2 was the background without RNA, which reflects absolute amount of cpxP RNA in arbitrary units. Each entire experiment was repeated 4 times.

Annotation of gene functions and regulation

The functions of genes and gene regulons were annotated using the following web databases: EcoCyc: http://biocyc.org/ECOLI/organism-summary, and RegulonDB: http://regulondb.ccg.unam.mx/index.jsp.

Oxygen consumption and extracellular acidification

O₂ consumption rate (OCR, which reflects total aerobic respiration) and extracellular acidification rate (ECAR, which reflects total H⁺ production from carbon catabolism) were measured simultaneously in real time with the Seahorse XFp analyzer (Agilent Technologies), which uses label-free, solid-state sensor cartridges in a microplate format (Lobritz et al., 2015; Saini et al., 2016). First, to attach bacteria to the XFp cell culture microplates, the plates were coated for 20 minutes at room temperature with 25 µl/well of BD Cell-Tak Cell and Tissue Adhesive™ (BD Biosciences, prepared by diluting 46 µl of Cell-Tak into 3 ml of 0.1 M NaHCO₃ and then neutralizing it with ½ volume 1 N HCl), followed by rinsing the wells twice with 200 µl of sterile distilled H₂O and then drying the wells at room temperature for 20 min. E. coli MG1655 were grown and diluted to OD_660nm = 0.0125 with warm Assay Medium (5 mM TRIS pH 7.6 with 2.5% glycerol, 150 mM NaCl, 5 µM ZnSO₄, and 2% of 100% LB) as described in “Bacteria, growth, and media” section, and 40 µl/well of bacteria was dispensed into the Cell-Tak-coated wells (2×10⁶ bacteria per well). The bacteria were then sedimented by centrifugation at 2270 g for 2 min, BSA (control) or PGRP were added to a final concentration of 100 µg/ml, and incubation at 37°C and OCR and ECAR measurements were started immediately. The micro-chamber wells are automatically re-equilibrated and oxygenated between measurements through the up and down mixing by the probes of the XFp sensor cartridge. The results are reported as mean ± SEM OCR (in pmol/min) and ECAR (in mpH/min) for the original bacterial incubation mixture, and also as phenograms of OCR plotted versus ECAR that illustrate the total cell energy phenotype.

Statistical analyses

Quantitative results are presented as means ± SEM, with statistical significance of the differences between groups determined by the two-sample one-tailed Student’s t-test or one-sample Student’s t-test using Microsoft Excel; P≤0.05 was considered significant. The n and P values are indicated in the figures and tables. Gene expression results are presented as heat maps generated using Java TreeView and represent mean fold changes in gene expression in PGRP-treated relative to BSA-treated bacteria, after converting <1 ratios to negative fold difference using the formula: (−1)/ratio. For microarray data statistical significance of differences in gene expression was also analyzed by calculating P values using two-sample two-tailed Student’s t-test, followed by calculation of π₀(λ) and then FDR (false discovery rate) q values, with significance threshold of q≤0.05, as described (Storey and Tibshirani, 2003).