Formate dehydrogenase, ubiquinone, and cytochrome bd-I are required for peptidoglycan recognition protein-induced oxidative stress and killing in Escherichia coli

Mammalian Peptidoglycan Recognition Proteins (PGRPs) kill bacteria through induction of synergistic oxidative, thiol, and metal stress. PGRPs induce oxidative stress in bacteria through a block in the respiratory chain, which results in decreased respiration and incomplete reduction of oxygen (O2) to hydrogen peroxide (H2O2). In this study we identify the site of PGRP-induced generation of H2O2 in Escherichia coli. Tn-seq screening of E. coli Tn10 insertion library revealed that mutants in formate dehydrogenase (FDH) genes had the highest survival following PGRP treatment. Mutants lacking functional FDH-O had abolished PGRP-induced H2O2 production and the highest resistance to PGRP-induced killing, and formate enhanced PGRP-induced killing and H2O2 production in an FDH-dependent manner. Mutants in ubiquinone synthesis (but not menaquinone and demethylmenaquinone) and cytochrome bd-I (but not cytochromes bo3 and bd-II) also had completely abolished PGRP-induced H2O2 production and high resistance to PGRP-induced killing. Because electrons in the respiratory chain flow from dehydrogenases’ substrates through quinones and then cytochromes to O2, these results imply that the site of PGRP-induced incomplete reduction of O2 to H2O2 is downstream from dehydrogenases and ubiquinone at the level of cytochrome bd-I, which results in oxidative stress. These results reveal several essential steps in PGRP-induced bacterial killing.


Results
Tn-seq identifies formate dehydrogenase mutants with increased survival in PGRP-treated cultures. We used a highly saturated mini-Tn10 insertion library in E. coli MG1655 containing ~200,000 unique mutants, with each mutant containing one transposon randomly inserted into the chromosome 15 . The library was exposed for 3 hrs to BSA as a control or to a bactericidal concentration of PGRP that reduced the numbers of viable bacteria in 3 hrs by >99%, or to a sub-bactericidal (bacteriostatic) concentration of PGRP that reduced viable bacteria in 3 hrs by 50%. We selected 3 hrs for the assay to allow for the selection of mutants resistant to PGRP killing and the depletion of mutants more sensitive to PGRP killing. We used human recombinant PGLYRP4 as a representative bactericidal PGRP, as we have previously shown that all human bactericidal PGRPs had similar activity and mechanism of bacterial killing 4,6,10,11 . We used BSA-treated bacteria as a control (rather than the initial inoculum) to eliminate any possible differences in the frequencies of mutants that could be due to different growth rates of mutants during the 3-hr incubation period. We plated surviving bacteria for single colonies and identified mutated genes by Illumina Hi-seq. For each gene we calculated the survival index (SI), which reflects a change in the frequency of each Tn insertion mutant in PGRP-treated versus control (BSA) cultures. A neutral mutation with no effect on survival in the presence of PGRP has SI = 1, whereas SI > 1 denotes mutation that makes bacteria more resistant to PGRP (indicating that the product of the gene enhances PGRP killing), and SI < 1 denotes mutation that makes bacteria more sensitive to PGRP (indicating that the product of the gene protects bacteria from PGRP killing). In this study, we focused on the mutants with significantly increased SI to identify genes that participate in PGRP-induced killing.
Five out of the top six Tn-seq mutants with the most increased frequency in the Tn-seq library treated with bactericidal concentration of PGRP were in formate dehydrogenase (FDH) genes or in genes required for the assembly and activity of formate dehydrogenases (Figs. 1 and S1). The frequency of 11 Tn-seq mutants in formate dehydrogenases or in genes required for the assembly and activity of formate dehydrogenases was also significantly increased in bacteria treated with the bacteriostatic concentration of PGRP (Fig. 1).
E. coli has two respiratory formate dehydrogenases, FDH-O and FDH-N (coded by fdoGHI and fdnGHI, respectively), which transfer electrons from formate to the quinone pool in the respiratory chain. E. coli also has a third formate dehydrogenase, FDH-H (coded by fdhF), which is a subunit of formate hydrogenlyase, an enzyme complex that disproportionates formate to H 2 and CO 2 13,16,17 . All three FDH contain selenocysteine coordinated to molybdenum of molybdopterin guanine dinucleotide cofactor, which are both required for the FDH enzymatic activity. The mutants with increased frequency in PGRP-treated cultures were for the following genes ( formate dehydrogenase is required for pGRp-induced killing and H 2 o 2 production. Increased frequency of Tn mutants in FDH genes and other genes required for the activity of FDH in PGRP-treated cultures suggested that FDH may participate in PGRP-induced killing of E. coli. To test this hypothesis and to identify which FDH participate in PGRP-induced killing, we constructed deletion mutants in key FDH genes or genes required for FDH activity and tested their sensitivity to PGRP-induced killing. Mutants lacking FDH-O, FDH-N, and FDH-H (ΔfdhD), or FDH-O and FDH-N (ΔfdhE), or FDH-O (ΔfdoG) had significantly increased survival after 3-hr PGRP treatment compared with the parental MG1655 strain (with 51-, 66-, or 50-fold higher numbers of colonies, respectively) (Fig. 2a). Mutants lacking FDH-N (ΔfdnG), or FDH-H (ΔfdhF), or formate channel (ΔfocA) also had significantly increased survival after 3-hr PGRP treatment, but much lower than the FDH-O mutants (7-, 18-, or 18-fold higher numbers of colonies than parental MG1655 strain, respectively) (Fig. 2a). Mutants lacking molybdopterin molybdenum transferase (ΔmoeA), or molybdate transporter subunit (ΔmodA), or molybdochelatase (Δmog), or selenocysteinyl-tRNA-specific translation factor (ΔselB), which are all required for the activity of all FDH, also had significantly increased survival after 3-hr PGRP treatment (30-, 48-, 40-, or 26-fold higher numbers of colonies than parental MG1655 strain, respectively) (Fig. 2a).
We further confirmed the requirement for formate dehydrogenases in PGRP-induced killing by showing increased resistance to PGRP killing of ΔfdhD, ΔfdhE, ΔfdoG, ΔfdnG, ΔfdhF, ΔfocA, ΔmoeA, ΔmodA, Δmog, ΔmoaE, and ΔselB deletion mutants in another strain of E. coli, BW25113 (Fig. 2b). These results indicate that FDH-O is required for PGRP-induced killing, and that FDH-N and FDH-H have a small, but still significant contribution.
We next hypothesized that formate may be a significant electron donor for the respiratory chain required for PGRP-induced H 2 O 2 production and that formate dehydrogenases are required for PGRP-induced H 2 O 2 production, because: (i) increased H 2 O 2 production is required for PGRP-induced killing 11 ; (ii) FDH transfer electrons from formate to the quinone pool in the respiratory chain 13,16 ; and (iii) a block in the electron transfer through the respiratory chain was previously identified as a source of PGRP-induced H 2 O 2 12 .
Deleting FDH-O, FDH-N, and FDH-H (ΔfdhD), or FDH-O and FDH-N (ΔfdhE), or FDH-O (ΔfdoG) completely abolished PGRP-induced H 2 O 2 production, whereas mutants lacking FDH-N (ΔfdnG) or formate channel (ΔfocA) had still significant PGRP-induced H 2 O 2 production, although lower than the parental MG1655 strain. Tn10 insertion library was treated for 3 hrs with BSA as control or with PGRP at bactericidal (<1% survival) or sub-bactericidal (50% survival) concentration, and the survival index (SI) with PGRP, relative to BSA, for individual Tn-mutants was determined by Tn-seq. The results are means from 3 independent experiments (biological replicates) and each dot represents a single gene deletion mutant. The Tn insertion sites and the numbers of reads for the representative top FDH genes are shown in Supplementary Fig. S1. The complete Tnseq results have been deposited in NCBI SRA with accession number PRJNA549505 (https://www.ncbi.nlm.nih. gov/sra/PRJNA549505).
ΔfdhF mutant lacking FDH-H (which does not directly contribute electrons to the respiratory chain) produced similar amounts of H 2 O 2 in response to PGRP as the parental MG1655 strain (Fig. 3a). Mutants lacking molybdopterin molybdenum transferase (ΔmoeA), or molybdate transporter subunit (ΔmodA), or molybdochelatase (Δmog), or selenocysteinyl-tRNA-specific translation factor (ΔselB) also did not produce H 2 O 2 in response to PGRP.
We further confirmed the requirement for FDH-O in PGRP-induced H 2 O 2 production by showing no increase in PGRP-induced H 2 O 2 production in ΔfdhD, ΔfdhE, ΔfdoG, ΔmoeA, ΔmodA, Δmog, ΔmoaE, and ΔselB deletion mutants in E. coli BW25113 (Fig. 3b). Formate channel ΔfocA BW25113 mutant behaved similarly to the ΔfocA MG1655 mutant and showed significant, but lesser contribution, and FDH-N ΔfdnG and FDH-H ΔfdhF BW25113 mutants showed only smaller and variable contribution of FDH-N and FDH-H to PGRP-induced H 2 O 2 production (Fig. 3b). Altogether, these results in MG1655 and BW25113 strains consistently indicate that FDH-O is required for PGRP-induced H 2 O 2 production, whereas FDH-N and FDH-H have small and variable contributions.
The requirement for FDH-O in H 2 O 2 production was selective for PGRP, because paraquat induced significant increases in H 2 O 2 in both parental strains and all the mutants with non-functional FDH enzymes (Fig. 3). These results indicate that formate dehydrogenases are not required for generation of H 2 O 2 by paraquat and are consistent with the known mechanism of H 2 O 2 generation by paraquat, in which paraquat is reduced by respiratory NADH dehydrogenase to a radical cation, which then reacts with O 2 18,19 .
To further determine the role of FDH in PGRP-induced killing and H 2 O 2 production, we tested the hypothesis that adding exogenous formate should enhance PGRP-induced killing and H 2 O 2 production. As predicted, adding 0.25 mM sodium formate significantly increased PGRP-induced killing and H 2 O 2 production in E. coli MG1655 parental strain, but not in ΔfdhD, ΔfdhE, ΔfdoG mutants, showing that this effect of formate was dependent on FDH-O (Fig. 4). Exogenous formate had no effect on the bacterial viability and H 2 O 2 concentrations in BSA-treated cultures showing the selectivity of the effect of formate in PGRP-treated cells.
To verify the presence of FDH-O, FDH-N, and FDH-H, we compared the expression of these FDH and other genes required for FDH assembly and activity in control and PGRP-treated cells. We analyzed whole genome expression arrays in BSA-(control) and PGRP-treated E. coli, which we previously deposited in NCBI GEO (accession number GSE44211) and at that time analyzed for the oxidative, thiol, and metal stress response genes 11 , but not for FDH-related genes. Control (BSA-treated) cells had high expression of fdhD and fdhE (required for the www.nature.com/scientificreports www.nature.com/scientificreports/ activity of FDH-O, FDH-N, and FDH-H, or FDH-O and FDH-N, respectively), fdoG, fdoH, and fdoI (coding for FDH-O subunits), and many genes required for the synthesis of the molybdopterin guanine dinucleotide cofactor and selenocysteine (required for the activity of all three FDH enzymes) ( Supplementary Fig. S2). Control cells had low expression of FDH-N (fdnG, fdnH, and fdnI) and FDH-H (fdhF) genes. PGRP treatment significantly increased the expression of fdoG, fdoH, fdoI, fdnH, and fdnI, and also of several molybdopterin and selenocysteine synthesis genes ( Supplementary Fig. S2). These results are consistent with aerobic expression of FDH-O 16 and our results showing its role in PGRP-induced killing and H 2 O 2 production, and also show that expression  . Exogenous formate enhances PGRP-induced killing and H 2 O 2 production in E. coli and this enhancement is FDH-dependent. 0.25 mM sodium formate was added as indicated to the cultures of the parental strain MG1655 or FDH deletion mutants that were treated with BSA or PGRP and assayed for viability by colony counts after 3 hrs (a) or total amounts of H 2 O 2 after 15 min (b) of incubation. The results are means ± SEM from 3 experiments (biological replicates); *P < 0.05, **P < 0.001 PGRP vs BSA; ^P < 0.05, ^^P < 0.001 mutant vs parental strain; # P < 0.05 without vs with formate; NS P > 0.05.
Ubiquinone is required for pGRp-induced killing and H 2 o 2 production. FDH-O and FDH-N transfer electrons from formate to the quinone pool in the cytoplasmic membrane, and these electrons are then transferred from quinones to the cytochromes, which in the presence of oxygen are the terminal oxygen reductases that reduce O 2 to H 2 O. The requirement for FDH-O in PGRP-induced H 2 O 2 production (Fig. 3) and simultaneous PGRP-induced decrease in O 2 consumption 12 indicated a block in the respiratory chain and either premature diversion of electrons from the respiratory chain or its malfunction, leading to incomplete reduction of O 2 to H 2 O 2 . This PGRP-induced block in electron transfer and incomplete reduction of O 2 to H 2 O 2 could occur at the level of dehydrogenases (FDH and/or NDH), quinones, or cytochromes. If the diversion of electrons occurred from the dehydrogenases and not from quinones (located downstream in the electron transfer chain), deleting quinones should not abolish the PGRP-induced H 2 O 2 production. But if the diversion of electrons occurred at the level of quinones or downstream from quinones, deleting quinones should abolish or decrease PGRP-induced H 2 O 2 production. To discriminate between these possibilities and to identify which quinones (if any) are involved in PGRP-induced H 2 O 2 production and PGRP-induced killing, we next tested the PGRP-induced killing and H 2 O 2 production in quinone deletion mutants.
E. coli has 3 quinones: ubiquinone (UQ), menaquinone (MK), and demethylmenaquinone (DMK), which differ in their redox potential and thus affinity for electrons 20 . We constructed mutants deficient in the synthesis of UQ, MK, and DMK in both MG1655 and BW25113 strains and tested their sensitivity to PGRP-induced killing and H 2 O 2 production. Mutants deficient in the synthesis of UQ (ΔubiCA) or UQ and MK (ΔubiE) had significantly increased survival after 3-hr PGRP treatment (with 65-and 72-fold higher numbers of colonies than in the parental MG1655 strain, and 111-and 77-fold higher in BW25113 strain, respectively) ( Fig. 5a). By contrast, mutants deficient in the synthesis of both MK and DMK (ΔmenA) had unchanged survival, compared with the parental MG1655 and BW25113 strains (Fig. 5a). Thus, UQ, but not MK and DMK, is required for PGRP-induced killing.
In mutants lacking UQ (ΔubiCA) or UQ and MK (ΔubiE) PGRP did not induce any H 2 O 2 production, whereas mutants lacking both MK and DMK (ΔmenA) had significant PGRP-induced H 2 O 2 production that did not significantly differ from H 2 O 2 production in the parental strains (Fig. 5b). Mutants lacking UQ (but not MK and DMK) also had similarly abolished H 2 O 2 production in response to paraquat, which shows that UQ is required for paraquat-generated H 2 O 2 production, consistent with the proposed ability of quinones to participate in the redox cycling of paraquat 21 . Similar results were obtained in mutants independently generated in both MG1655 and BW25113 strains, validating that these effects were indeed due to the deletion of these quinone synthesis genes. These results indicate that UQ (but not MK and DMK) is required for PGRP-induced H 2 O 2 production. Thus, the site of PGRP-induced H 2 O 2 production in the respiratory electron transport chain is at the level of UQ or downstream from UQ.
If ubiquinol (UQ-H 2 ) is the source of electrons for H 2 O 2 production in PGRP-treated cells, the outcome of increased transfer of electrons to O 2 should be an increase in oxidized UQ at the time of peak H 2 O 2 production. To test this hypothesis we measured the amounts of oxidized and reduced quinones after 15 min of PGRP treatment, which is the peak of PGRP-induced H 2 O 2 production 12 . We focused on UQ, because UQ, but not MK and DMK, was required for PGRP-induced H 2 O 2 production and killing (Fig. 5a,b), and because UQ is the main quinone in aerobically growing E. coli 20,22 . Indeed, we detected a total of approximately 140 ng UQ 8 , 2.0 ng UQ 9 , 3.3 ng MK 8 , and <0.1 ng DMK 8 (below the level of detection) per culture (oxidized plus reduced forms).
A 15-min treatment with PGRP resulted in significant increases in both UQ 8 /UQ 8 -H 2 and UQ 9 /UQ 9 -H 2 (oxidized/reduced) ratios, compared with untreated and BSA-treated cells, indicating a significant shift from reduced UQ-H 2 to oxidized UQ in PGRP-treated cells (Fig. 5c). By contrast, these ratios were similar in untreated and BSA-treated cells. These results are consistent with our hypothesis of increased loss of electrons from reduced UQ-H 2 resulting in an increase in the oxidized UQ/reduced UQ-H 2 ratio in PGRP-treated cells at the time of the maximum H 2 O 2 production. These electrons could be diverted to O 2 directly from UQ-H 2 to generate H 2 O 2 , or transferred to cytochromes, or both.
To further verify the presence of quinones, we compared the expression of quinone synthesis genes in control and PGRP-treated cells, by analyzing whole genome expression arrays, which we previously deposited in NCBI GEO (accession number GSE44211) 11 , but did not analyze for quinone synthesis. Control (BSA-treated) cells had high expression of genes for the synthesis of UQ (ubiA, ubiC, and ubiE), and expression of ubiA and ubiC (required for the synthesis of UQ) was further increased by PGRP treatment (Supplementary Fig. S3a). Expression of menA, required for the synthesis of both MK and DMK, was lower and significantly decreased by PGRP treatment. These results are consistent with the dominant role of UQ in PGRP-induced killing and H 2 O 2 production.
cytochrome bd-i is required for pGRp-induced killing and H 2 o 2 production. Our results showing the requirement for UQ in PGRP-induced killing and H 2 O 2 production motivated us to determine the role of cytochromes in PGRP-induced killing and H 2 O 2 production, because in the respiratory chain UQ-H 2 transfers electrons to cytochromes. Thus, we next tested whether the PGRP-induced block in the respiratory chain and the site of incomplete reduction of O 2 to H 2 O 2 occurred down-stream from the quinones, i.e., at the cytochromes. E. coli has 3 terminal quinol/oxygen oxidoreductases: cytochromes bo 3 , bd-I, and bd-II 13,14 . We constructed mutants deficient in each cytochrome in E. coli MG1655 and tested the sensitivity of cytochrome mutants in both MG1655 and BW25113 strains to PGRP-induced killing and H 2 O 2 production.
Mutants lacking cytochrome bd-I (ΔcydB) had significantly increased survival after 3-hr PGRP treatment (with 92-and 61-fold higher numbers of colonies than in the parental MG1655 and BW25113 strains, www.nature.com/scientificreports www.nature.com/scientificreports/ respectively) (Fig. 6a). Single deletion mutants lacking cytochrome bo 3 (ΔcyoB) or cytochrome bd-II (ΔappB) had significantly lower survival than the ΔcydB mutant, but they still had somewhat higher survival than the parental strains (with 6-and 7-, or 8-and 18-fold higher numbers of colonies than in the parental MG1655 and BW25113 strains, respectively). Double deletion ΔcyoBΔcydB and ΔcyoBΔappB mutants also had significantly increased survival after 3-hr PGRP treatment (with 71-or 7-fold higher numbers of colonies than in the parental MG1655 strain), similar to the survival of the single ΔcydB and ΔcyoB mutants, respectively. These results indicate that cytochrome bd-I is required for PGRP-induced killing, and that cytochromes bo 3 and bd-II play a small role.
PGRP-induced H 2 O 2 production was completely abolished in both MG1655 and BW25113 mutants lacking cytochrome bd-I (ΔcydB), and also in double deletion ΔcyoBΔcydB mutant, compared with the parental strains (Fig. 6b). By contrast, PGRP-induced H 2 O 2 production in both MG1655 and BW25113 mutants lacking cytochrome bo 3 (ΔcyoB) or cytochrome bd-II (ΔappB), or in a double deletion mutant lacking both cytochromes bo 3 and bd-II (ΔcyoBΔappB), were similar to the H 2 O 2 production in the parental strains. These results show that cytochrome bd-I is required for PGRP-induced H 2 O 2 production and that cytochromes bo 3 and bd-II are not required. These results imply that the primary site of PGRP-induced production of H 2 O 2 is cytochrome bd-I.
The requirement for cytochrome bd-I in H 2 O 2 production was selective for PGRP, because paraquat induced significant increases in H 2 O 2 in both parental strains and all the cytochrome mutants, including mutants lacking cytochrome bd-I (Fig. 6b). These results indicate no selective requirement for any of the E. coli cytochromes for paraquat-induced H 2 O 2 production (in contrast to PGRP) and are consistent with the known direct reduction of O 2 by the paraquat radical cation 18,19 . The results are mean ratios ± SEM of UQ 8 /UQ 8 -H 2 and UQ 9 /UQ 9 -H 2 from 4 experiments (biological replicates); # P < 0.05 PGRP vs untreated, *P < 0.05 PGRP vs BSA (paired t-test). The total amounts of UQ + UQ-H 2 in untreated, BSA-treated, and PGRP-treated cultures were: 107 ± 5, 161 ± 19, and 149 ± 15 ng/culture for UQ 8 , and 1.7 ± 0.05, 2.0 ± 0.4, and 2.2 ± 0.2 ng/culture for UQ 9 , respectively (means ± SEM), with no significant differences between BSA-and PGRP-treated groups. www.nature.com/scientificreports www.nature.com/scientificreports/ Both double deletion mutants (ΔcyoBΔcydB and ΔcyoBΔappB) had higher H 2 O 2 production in control (BSA-treated) cultures than the parental strain (Fig. 6b), which is consistent with the previous report for the untreated ΔcyoΔcyd mutant 23 . Because PGRP-induced H 2 O 2 production was completely abolished in ΔcydB and ΔcyoBΔcydB mutants, these results indicate that the main source of PGRP-induced H 2 O 2 is different from the sources of the low-level H 2 O 2 produced in control cells growing without PGRP.
To verify the presence of cytochromes, we compared the expression of cytochrome genes in control and PGRP-treated cells, by analyzing our whole genome expression arrays (NCBI GEO accession number GSE44211) 11 . Control (BSA-treated) cells had high expression of cytochrome bo 3 cyoA-D genes, most of which were further increased by PGRP treatment, and very high expression of cytochrome bd-I cydABX genes. However, the expression of cytochrome bd-I cydA gene was severely decreased by PGRP treatment ( Supplementary  Fig. S3b). Cytochrome bd-II appB and appC genes had very low expression, which was not changed by PGRP treatment. These results are consistent with the role of cytochrome bd-I in PGRP-induced killing and H 2 O 2 production and suggest that maybe PGRP-treated cells decrease cytochrome bd-I expression as a defense mechanism to limit its contribution to PGRP-induced H 2 O 2 production.
Role of membrane depolarization. We next considered the possibility that PGRP induces membrane depolarization in E. coli because membrane depolarization could be either the cause or the consequence of respiratory chain malfunction, oxidative stress, and decline in respiration, and because we previously showed that PGRP treatment results in membrane depolarization in B. subtilis 10 . Treatment of E. coli with PGRP for 15 min (the time of maximum PGRP-induced H 2 O 2 production 12 ) resulted in significant, but low-level membrane depolarization, similar to membrane depolarization induced by 100 µM KCN, a cytochrome bo 3 inhibitor 14 , as measured by a decrease in red fluorescence of membrane potential sensitive probe DiOC 2 (3) (Fig. 7a,b). KCN-induced membrane depolarization is consistent with inhibition of cytochrome bo 3 and its proton-pumping ability 13,14 . CCCP (carbonyl cyanide 3-chlorophenylhydrazone), a proton ionophore that dissipates both membrane potential www.nature.com/scientificreports www.nature.com/scientificreports/ and proton gradient, induced strong membrane depolarization (as expected), 3-fold (significantly) higher than PGRP and KCN (Fig. 7a,b).
We next tested whether membrane depolarization with CCCP or cytochrome inhibition with KCN would induce H 2 O 2 production and whether the amount of H 2 O 2 would correlate with the extent of membrane depolarization. Both CCCP and KCN induced similar and significant increases in H 2 O 2 production compared with BSA-treated bacteria, but the amounts of CCCP-and KCN-induced H 2 O 2 were significantly lower than in PGRP-treated bacteria (Fig. 7c). These results reveal that membrane depolarization with CCCP or cytochrome bo 3 inhibition with KCN induce significant H 2 O 2 production, but also show that there is no correlation between the extent of membrane depolarization and the amount of H 2 O 2 produced, as PGRP induced low-level membrane depolarization and high amount of H 2 O 2 , whereas CCCP induced much higher level of membrane depolarization but lower amount of H 2 O 2 than PGRP.

Discussion
E. coli has a complex branched respiratory chain, which allows the bacterium to quickly adapt to changes in the oxygen supply and other environmental conditions 13 . However, these multiple and essential functions of E. coli respiratory chain are also a target for human bactericidal innate immunity proteins, PGRPs, which induce a block in E. coli respiratory chain 12 . The results of this study show that PGRP-induced H 2 O 2 production and bacterial killing require FDH-O, UQ, and cytochrome bd-I, in addition to the previously shown requirement for NDH-1 and NDH-2 12 . PGRP-treated E. coli up-regulates its metabolism by increasing cAMP-Crp-controlled TCA cycle and supply of NADH, by increasing NADH oxidation by NDH-1 and NDH-2 12 , and also by increasing oxidation of formate by FDH-O, perhaps in an attempt to overcome this block in the respiratory chain by increasing influx of electrons. The central and critical components of this proposed model are UQ and cytochrome bd-I, because increased function of all three dehydrogenases (FDH-O, NDH-1, and NDH-2) supplies electrons to UQ and reduces it to ubiquinol (UQ-H 2 ), which then transfers electrons to cytochromes (Fig. 8). We propose that cytochrome bd-I is the site of PGRP-induced H 2 O 2 production and not the dehydrogenases (FDH and NDH) and UQ, because deleting UQ (downstream from FDH and NDH) or cytochrome bd-I (downstream from UQ) completely abolished PGRP-induced H 2 O 2 production. If FDH, NDH or UQ were the sites of H 2 O 2 production due to electron backup, deleting downstream components of the respiratory chain (UQ or cytochrome bd-I) would have had no effect or would have enhanced H 2 O 2 production. Thus, our data showing that deleting dehydrogenases 12 , or UQ, or cytochrome bd-I all abolished PGRP-induced H 2 O 2 production suggest that electrons are being transferred all the way down to cytochrome bd-I and that incomplete reduction of O 2 to H 2 O 2 is at the level of cytochrome bd-I, possibly due to incomplete electron transfer from UQ-H 2 to cytochrome bd-I or malfunction of cytochrome bd-I (Fig. 8).
Consistent with this proposed model, there was a significant increase in UQ/UQ-H 2 ratio at the peak of PGRP-induced H 2 O 2 production (15 min), suggesting that the increased electron transfer from UQ-H 2 to cytochrome bd-I at that time had already occurred and resulted in UQ-H 2 oxidation to UQ. This high UQ/UQ-H 2 ratio is maintained, because in PGRP-treated cells there is only a short-lived spike in the metabolism (5 to 10 min) followed by a decrease 12 , which limits further supply of electrons to UQ.
Decreasing O 2 consumption at the time of PGRP-induced H 2 O 2 production 12 suggests that this H 2 O 2 production is not a small byproduct of the usual O 2 reduction to H 2 O by the cytochromes, because this would require a very large PGRP-induced increase in O 2 consumption to produce such a large amount of H 2 O 2 (as a byproduct). Therefore, the diminished O 2 consumption 12 concomitant with a high increase in H 2 O 2 production can be best explained by incomplete reduction of the majority of O 2 to H 2 O 2 and thus even a decreasing O 2 consumption could still produce the amounts of H 2 O 2 that we detected. With continued PGRP treatment the function of the respiratory chain progressively deteriorates, eventually leading to almost complete loss of O 2 consumption and a decline in PGRP-induced H 2 O 2 production at 30 min 12 , which coincides with the cessation of all biosynthetic reactions in the cell 10 .
Our proposed model is energetically favorable for efficient transfer of electrons (exergonic) 13 14 , and thus it is not energetically favorable for cytochrome bo 3 to reduce O 2 to H 2 O 2 , which may be the reason why deleting cytochrome bo 3 has no effect on PGRP-induced H 2 O 2 production.
How do PGRPs induce the respiratory chain malfunction that leads to H 2 O 2 production? PGRPs have a peptidoglycan-binding groove specific for disaccharide-peptide fragments of peptidoglycan. In Gram-positive bacteria PGRPs selectively bind to the separation sites of the newly formed daughter cells, created by dedicated peptidoglycan-lytic endopeptidases, which separate daughter cells after cell division 10 . In Gram-negative bacteria, PGRPs bind to the entire outer membrane 10 through a separate binding site specific for LPS 5,9 . This binding to the outer membrane is required for PGRP killing, because exogenous LPS blocks PGRP binding and PGRP killing of www.nature.com/scientificreports www.nature.com/scientificreports/ E. coli 10 . PGRPs do not enter the cytoplasm 10 and induce the respiratory chain malfunction from the extracellular site, but it is not known whether after binding to the outer membrane they translocate to peptidoglycan in the periplasmic space and/or to the cytoplasmic membrane. Therefore, it is not known whether the effects of PGRPs on the respiratory chain are indirect (more likely) or direct.
We considered that PGRP-induced membrane damage or depolarization could be the cause of the respiratory chain malfunction and H 2 O 2 production. However, cytoplasmic membrane is not permeabilized in PGRP-treated E. coli for up to 6 hrs 6 and PGRP induced only low-level membrane depolarization. Moreover, there was no correlation between the extent of membrane depolarization and the amount of H 2 O 2 produced, i.e., PGRP induced low membrane depolarization and high H 2 O 2 production, whereas CCCP induced high membrane depolarization and lower H 2 O 2 production (Fig. 7). These results are consistent with our previous analysis of gene expression 11 , which showed high induction of oxidative stress response genes in PGRP-treated E. coli, such as oxyS, ahpF, katG, soxS, and soxR (indicating high production of H 2 O 2 and O 2 − ), but no increase in CCCP-treated cells in the expression in oxyS, ahpF, and katG, and moderate increase in the expression of soxS and soxR (indicating production of some O 2 − but no or low amount of H 2 O 2 ) and thus showing very limited oxidative stress response to CCCP. Altogether, these results suggest that PGRP-induced membrane depolarization may be the consequence of the decreased respiration rather than the primary cause of the respiratory chain malfunction and H 2 O 2 production. However, other possibilities cannot be excluded, such as membrane depolarization as the initial event, or other effects of PGRP on the cytoplasmic membrane that interfere with the respiratory chain.
The subsequent events that follow PGRP interaction with bacteria and production of H 2 O 2 and contribute to the eventual bacterial death include: (i) depletion of cellular thiols (peaks in 30 min); (ii) generation of HO • from H 2 O 2 , which then damage bacterial DNA, membrane, and proteins (30 min-2 hrs); (iii) increase in intracellular labile Zn 2+ (peaks in 1 hr); and (iv) cessation of all biosynthetic reactions (30 min-2 hrs) 10-12 . Our results also identify a new function for FDH-O. FDH-O and FDH-N are transmembrane dehydrogenases (formate-quinone oxidoreductases) with their catalytic sites facing the periplasmic space. They oxidize periplasmic formate to CO 2 and reduce membrane quinones to quinols, which then transfer electrons to the terminal respiratory chain quinol oxidoreductases. Whereas much is known about the structure and function of FDH-N in anaerobic respiration on nitrite and nitrate 13,16,17 25 showing the ability of Enterobacteria to use formate oxidation by FDH-O and FDH-N to shuttle electrons to the respiratory chain cytochrome bd-I when oxygen becomes available. Hughes et al. results also suggests that O 2 is the final electron acceptor for formate oxidation by both FDH-N and FDH-O (which is consistent with our results) and show that this mechanism gives Enterobacteria a competitive advantage in the inflamed gut, where oxygen is supplied by blood. Our results suggest that PGRPs can exploit this mechanism to kill bacteria.
Our results show that multiple dehydrogenases -FDH-O (this study) and NDH-1 and NDH-2 12 -are all required for PGRP-induced killing and H 2 O 2 production. Moreover, elimination of even a single dehydrogenase results in a significantly increased resistance to PGRP-induced killing and in a drastic loss of PGRP-induced H 2 O 2 production, rather than in a small incremental effect of each deletion. One possible explanation for this finding could be a threshold effect, in which simultaneous action of all the dehydrogenases is required to produce the amount of H 2 O 2 that exceeds the maximum level that the cell can detoxify, and deletion of any single dehydrogenase decreases the amount of H 2 O 2 below this threshold level.
Another possible explanation could be the formation of supercomplexes by the respiratory dehydrogenases and cytochromes, in which the function of each component is dependent on the presence and/or function of the other components in the supercomplex. Formation of such supercomplexes is well accepted in mitochondria 26,27 . In E. coli formation of three supercomplexes was reported: FDH-O:cytochrome bo 3 :cytochrome bd-I supercomplex, NDH-1:NDH-2 supercomplex, and SDH:cytochrome bd-II supercomplex, with each component of the complex required for the assembly and full enzymatic activity of the supercomplex 26,28-30 . Super-resolution microscopy and quantification revealed the presence of complexes of respiratory chain enzymes in E. coli membrane, but did not show co-localization of NDH-1, cytochrome bo 3 , cytochrome bd-I, SDH, and F o F 1 ATPase in these complexes 27 . These results do not support the presence of the above-mentioned supercomplexes and make this possibility a less likely explanation of our results, although co-localization of FDH-O and NDH-2 with other respiratory enzymes was not studied in this report 27 and is still a possibility.
Cytochromes usually serve as electron sinks and under normal growth conditions protect bacteria from inadvertent production of H 2 O 2 and oxidative stress, because their high affinity for O 2 allows them to efficiently transfer 4 electrons to O 2 for its complete reduction to H 2 O 13,14,31 . However, our results suggest that in PGRP-treated cells cytochrome bd-I malfunctions, which results in an incomplete 2-electron reduction of O 2 to H 2 O 2 . Thus, the pathways and the final electron donors (cytochromes) for PGRP-induced H 2 O 2 are likely different from the sources of the low-level H 2 O 2 produced in control cells growing without PGRP, because deleting NDH-1 and NDH-2, or UQ, or cytochromes bo 3 and bd-I increase the basal levels of H 2 O 2 production in cells growing in a medium 23 , whereas, these deletions in PGRP-treated cells resulted in a decrease of PGRP-induced H 2 O 2 production (ref. 12 and Figs. 5 and 6).
Purified cytochrome bd-I has peroxidase activity and can reduce H 2 O 2 to H 2 O using quinol as an electron donor 32 . It is not known how this activity functions in intact cells, but its membrane localization suggests a role in scavenging periplasmic H 2 O 2 and protection against environmental H 2 O 2 . This peroxidase activity, however, had negligible effect in our experiments, because deletion of cytochrome bd-I abolished (rather than enhanced) PGRP-induced H 2 O 2 production.

Scientific RepoRtS |
(2020) 10:1993 | https://doi.org/10.1038/s41598-020-58302-1 www.nature.com/scientificreports www.nature.com/scientificreports/ Cytochrome bd-I is important for bacterial growth and virulence, for example in E. coli urinary tract infections 33 or colitis 25 . Cytochrome bd-I is also an attractive target for antibacterial agents, because it is only present in prokaryotes, but not in eukaryotic mitochondria. Cytochrome bd-I protects mycobacteria from killing by drugs targeting cytochrome c [34][35][36][37] , which is the opposite effect to the requirement for cytochrome bd-I in PGRP-induced killing reported in this study. However, similar to our results with PGRP in E. coli, combining cytochrome c inhibition with oxidative stress-inducing drug (clofazimine) in mycobacteria induces cytochrome bd-I-dependent enhancement of killing and production of reactive oxygen species 38 .
Some antibacterial peptides, namely aurachin D, gramicidin S, and microcin J25, selectively inhibit E. coli cytochrome bd-I redox activity [39][40][41][42][43] , and microcin J25 also induces cytochrome bd-I-dependent production of O 2 -41-43 . However, these effects have been only shown in isolated membranes or proteins, but not in cells in situ, and the sensitivity to microcin J25-induced killing was similar in cytochrome bd-I-and bo 3 -deficient mutants. Also, gramicidin S and microcin J25 have additional targets in bacteria that are most likely responsible for their antibacterial effect: gramicidin S destroys cytoplasmic membrane integrity 40 and microcin J25 inhibits RNA polymerase 41 , and, thus, their effects on cytochrome bd-I appear secondary. Human antibacterial peptide, LL-37, was also shown to induce a cytochrome bd-I-dependent burst of O 2 − and HO • production in E. coli cells before permeabilizing their cytoplasmic membranes, which is the main mechanism of LL-37-induced killing 44 . However, again, this O 2 − and HO • production did not significantly contribute to LL-37-induced killing, which was the same in cytochrome bd-I-and bo 3 -deficient mutants. Thus, these antibacterial peptides often have intracellular targets or permeabilize cytoplasmic membrane, their effects on cytochrome bd-I in cells in situ are negligible or unknown, and they are often transported into the periplasm and cytoplasm by specific transporters 41,43 . By contrast, PGRPs are large proteins, stay bound to the bacterial envelope, do not enter the cytoplasm 10 , do not permeabilize cytoplasmic membranes 4,6,10 , and induce large amounts of peroxide readily detectable in PGRP-treated cells 11,12 , which is dependent on cytochrome bd-I and correlates with the increased resistance of cytochrome bd-I-deficient mutants to PGRP-induced killing. Thus, the mechanisms of bacterial killing by small antibacterial peptides and PGRP are different.
PGRPs are conserved in evolution from insects to mammals and have retained their antimicrobial effectiveness for millions of years with no frequent emergence of fully resistant strains. The multiple factors responsible for this continued antimicrobial effectiveness of PGRPs include their ability to: (i) bind to multiple components of bacterial envelope (peptidoglycan, lipoteichoic acid, and LPS); (ii) simultaneously induce oxidative, thiol, and metal stress responses in bacteria, which individually are bacteriostatic, but in combination are bactericidal; (iii) induce oxidative, thiol, and metal stress responses in bacteria through three independent pathways; (iv) affect multiple metabolic pathways; and (v) synergize with other innate immune molecules, such as antimicrobial peptides (reviewed in 45 ). Thus, emergence of PGRP resistance and disabling these multiple mechanisms would require simultaneous acquisition of resistance to these multiple separate antimicrobial mechanisms -a very low probability event. Our current results further elucidate this intricate ability of PGRPs to control bacterial growth and survival.
In conclusion, our results identify the sequence of events in PGRP-induced bacterial killing and uncover a bactericidal mechanism that can be exploited in the future for the development of new approaches to enhance resistance to infections and increase effectiveness of antibacterial therapy.

Materials and Methods
Materials. Human recombinant PGLYRP4 (used as a representative bactericidal PGRP) was expressed in S2 cells and purified as previously described 4,6 in a buffer containing 10 mM TRIS (pH 7.6), with 150 mM NaCl, 10 µM ZnSO 4 , and 10% glycerol, and used at 100 µg/ml (0.87 µM, as PGLYRP4 is a 115 kDa disulfide-linked dimer 4 ) final concentration, unless otherwise indicated. Paraquat (methyl viologen) was from Acros Organics and CCCP (carbonyl cyanide 3-chlorophenylhydrazone) from Molecular Probes (dissolved in DMSO, with equivalent amount of DMSO added to other groups). 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. E. coli strains used in this study are listed in Supplementary Table S1 46 .
Bacteria were grown aerobically overnight at 37 °C in an orbital shaker (250 rpm) in LB, diluted to OD 660 = 0.02 into fresh LB and grown to OD 660 = 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 6,10-12 . This Assay Medium contained (final concentrations): 5 mM TRIS (pH 7.6) with 2.5% glycerol, 150 mM NaCl, 5 µM ZnSO 4 , and 1% 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 12 . 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, H 2 O 2 production, and killing 12 .
Screening of tn-mutant library. A highly saturated mini-Tn10 (Tn) insertion library in E. coli MG1655 containing ~200,000 unique mutants, generated and characterized previously 15 , was used in this study. Each mutant contained one transposon randomly inserted into the chromosome, with one insert per 20 bp on average and ~25 different transposon mutants for each gene/intergenic region, with nearly random distribution of insertions and the absence of large gaps or hot spots 15 . For each experiment, 5 µl of freshly thawed glycerol stock of the Tn library containing ~10 7 CFU/5 µl was inoculated into 5 ml of LB and grown aerobically at 37 °C in an orbital shaker (250 rpm) for 3 generations. The bacteria were sedimented by centrifugation and suspended at ~2.5 × 10 6 bacteria/ml in 50 µl of fresh warm Assay Medium with addition of 100 µg/ml PGRP, or 220 µg/ml PGRP, or 200 µg/ml BSA (control) and incubated at 37 °C aerobically with 250 rpm shaking for 3 hrs. These PGRP treatments resulted in ~50% and >99% reduction in CFU, respectively (higher concentration of PGRP than in other (2020) 10:1993 | https://doi.org/10.1038/s41598-020-58302-1 www.nature.com/scientificreports www.nature.com/scientificreports/ assays was required because of the high number of bacteria used in these experiments); BSA treatment resulted in ~40-fold increase in CFU in 3 hrs, compared with the initial inoculum. After 3-hr incubation each entire 50 µl culture was diluted with saline and plated on LB agar for single colonies on multiple 15 cm plates, which were then incubated at 37 °C aerobically overnight. Approximately 100,000, 50,000, and 1,000 colonies from BSA-and PGRP-treated groups, respectively, were then scraped from the plates and thoroughly mixed. The lower numbers of colonies in PGRP-treated groups represented all the surviving mutants (more resistant to PGRP), with ~50% and >99% of mutants (most sensitive to PGRP) killed by 100 and 220 µg/ml PGRP, respectively. For each treatment group, genomic DNA was extracted from 1-ml aliquot of pooled colonies containing ~5 × 10 9 CFU using a Gentra Puregene Yeast/Bacteria Kit (Qiagen) and subjected to deep sequencing analysis. The entire experiment was repeated 3 times (3 biological replicates).
Briefly, genomic DNA was sonicated to produce 200-to 600-bp fragments. A poly(C) tail was added to DNA fragments by using terminal deoxynucleotidyl transferase. A first round of PCR with one transposon-specific primer (olj363) and one oligo (dG) primer (olj376) (Supplementary Table S2) was conducted to amplify the fragments. A second round of PCR with a nested primer (olj385) was conducted to add adaptors and bar codes for Illumina sequencing (with BSA, PGRP 50% survival, and PGRP <1% survival groups primers, bar codes are underlined in Supplementary Table S2). The sequencing was conducted on Illumina HiSeq 2000 at The Tufts University Core Facility (TUCF), Boston, MA, using olj386 custom sequencing primer and a standard Illumina index sequencing primer. The complete Tn-seq results have been deposited in NCBI SRA with accession number PRJNA549505 (samples SAMN12090400 -SAMN12090408) (https://www.ncbi.nlm.nih.gov/sra/PRJNA549505). tn-seq data analysis. Read mapping and calculations were carried out on the Tufts University Galaxy server as described previously 15,47 . Briefly, Illumina reads were aligned to the E. coli MG1655 genome sequence using Bowtie 48 . The Tufts Galaxy server custom script was applied to enumerate read number for each insertion site. The insertion sites were then aggregated by annotated genes. In order to fully cover all genes and promoter regions, a GenBank file containing all intergenic regions was generated to allow insertion sites to aggregate to intergenic regions. The predicted number of reads for each gene was calculated based on the length of the gene and the total number of reads in the library. We identified 2,540 genes and 570 intergenic regions that contained 3 or more Tn insertion sites and we used these genes for further analysis, because >3 insertion sites are needed for reliable identification of mutants with changed frequency in the library 15 . For each gene in both BSA-and PGRP-treated groups we calculated Dval, which represents the predicted frequency of each gene in the library based on the length of the gene, the total length of the genome, and the total number of reads in the library; Dval = the number of actual reads for the gene/predicted number of reads for the gene. For each gene we then calculated the survival index, SI = Dval in PGRP-treated group/Dval in control BSA-treated group. The SI reflects a change in the frequency of each gene (Tn insertion mutant) following PGRP treatment. A neutral mutation with no effect on survival in the presence of PGRP has SI = 1, whereas SI > 1 denotes mutation that makes bacteria more resistant to PGRP, and SI < 1 denotes mutation that makes bacteria more sensitive to PGRP. Table S1) by replacing the entire coding sequence of the genes (in frame) with kanamycin resistance cassette using homologous recombination, as previously described 12,49 . We generated PCR fragments using primers (Supplementary Table S2) and individual mutants from the Keio collection 46 , digested the fragments with DpnI, gel purified, and electroporated into E. coli MG1655 or BW25113 carrying the Red helper plasmid pKD46 (CGSC #7669). Transformed cells were recovered in SOC media with ampicillin and L-arabinose at 30 °C with shaking overnight, and kanamycin (30 µg/ml) resistant colonies were picked and verified by colony PCR with kanamycin-specific and gene-specific primers (Supplementary Table S2). The gene deletions in each mutant were then further verified by capillary sequencing using gene-specific primers. To create double deletion mutants, the kanamycin resistance gene was first removed from a single mutant using pCP20 plasmid. This plasmid has ampicillin resistance, shows temperature-sensitive replication, and thermal induction of Flp recombinase 50 . Mutants were transformed with pCP20, and ampicillin-resistant transformants were selected at 30 °C, further colony purified nonselectively at 43 °C, and then tested for loss of ampicillin as well as kanamycin resistance. Then, the second mutation was introduced by transforming pKD46 into the mutant whose kanamycin cassette had been removed, followed by the gene disruption protocol described above for generation of single deletion mutants. pGRp killing assay. E. coli were grown as described above in "Bacteria, growth, and media" section and suspended at 0.25 × 10 6 bacteria/ml in 50 µl of fresh warm Assay Medium with addition of BSA or PGRP (100 µg/ ml, except for Fig. 4a, where 50 µg/ml was used) and incubated at 37 °C aerobically with 250 rpm shaking for 3 hrs. The numbers of bacteria were then determined by colony counts 10-12 . H 2 o 2 assay. H 2 O 2 production was measured as previously described 12 using Amplex Red Hydrogen Peroxide/ Peroxidase Assay Kit (InVitrogen/Molecular Probes), based on enzymatic detection of H 2 O 2 by horseradish peroxidase. E. coli were grown and suspended in 50 µl of Assay Medium at OD 660 = 0.15 (~50 × 10 6 bacteria/ml) as described above in "Bacteria, growth, and media" section, with addition of BSA (100 µg/ml), PGRP (100 µg/ml), paraquat (100 µM), CCCP (20 µM), or KCN (100 µM). 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 H 2 O 2 was determined by measuring fluorescence 12 . (2020) 10:1993 | https://doi.org/10.1038/s41598-020-58302-1 www.nature.com/scientificreports www.nature.com/scientificreports/ Statistical analyses. Quantitative results are presented as arithmetic or geometric means ± SEM, with statistical significance of the differences between groups determined by the two-sample two-tailed Student's t-test or paired Student's t-test using Microsoft Excel; P ≤ 0.05 was considered significant. The n and P values are indicated in the figures.

Data availability
All data generated or analyzed during this study are included in this published article (and its Supplementary Information) or have been deposited in NCBI as cited in the text under the accession numbers PRJNA549505 and GSE44211.