A Klebsiella pneumoniae DedA family membrane protein is required for colistin resistance and for virulence in wax moth larvae

Ineffectiveness of carbapenems against multidrug resistant pathogens led to the increased use of colistin (polymyxin E) as a last resort antibiotic. A gene belonging to the DedA family encoding conserved membrane proteins was previously identified by screening a transposon library of K. pneumoniae ST258 for sensitivity to colistin. We have renamed this gene dkcA (dedA of Klebsiella required for colistin resistance). DedA family proteins are likely membrane transporters required for viability of Escherichia coli and Burkholderia spp. at alkaline pH and for resistance to colistin in a number of bacterial species. Colistin resistance is often conferred via modification of the lipid A component of bacterial lipopolysaccharide with aminoarabinose (Ara4N) and/or phosphoethanolamine. Mass spectrometry analysis of lipid A of the ∆dkcA mutant shows a near absence of Ara4N in the lipid A, suggesting a requirement for DkcA for lipid A modification with Ara4N. Mutation of K. pneumoniae dkcA resulted in a reduction of the colistin minimal inhibitory concentration to approximately what is found with a ΔarnT strain. We also identify a requirement of DkcA for colistin resistance that is independent of lipid A modification, instead requiring maintenance of optimal membrane potential. K. pneumoniae ΔdkcA displays reduced virulence in Galleria mellonella suggesting colistin sensitivity can cause loss of virulence.

Colistin acts in a manner mechanistically similar to the CAMPS (cationic antimicrobial peptides) of the innate immune response by binding to and disrupting negatively charged bacterial membranes leading to cell lysis and death 32 . CAMPs are found in all kingdoms of life where they play roles in immune defense, predation and competition 33,34 . Gram-negative pathogens, including K. pneumoniae, can acquire resistance to colistin by modification of outer membrane lipopolysaccharides (LPS) with cationic substituents resulting in reduced affinity to the polycationic colistin 35 . Common mechanisms involve expression of enzymes, including aminoarabinose (Ara4N) transferase (ArnT) and phosphoethanolamine transferase (EptA) which transfer aminoarabinose and phosphoethanolamine, respectively, to LPS lipid A. Together, these amine-containing groups neutralize the negative charge of lipid A disrupting colistin binding 35 . The discovery of plasmid associated eptA homologues (termed mcr for "mobile colistin resistance") genes raise concern about possible spread of colistin resistance via horizontal gene transfer [36][37][38] .
Cytoplasmic synthesis and inner membrane transport of undecaprenyl phosphate-α-L-Ara4N precedes periplasmic modification of lipid A with Ara4N (1). The transport of this lipid-linked intermediate is carried out by EmrE-like transporters designated ArnE and ArnF (2). Proton motive force (PMF)-dependent transport of EmrE substrates such as methyl viologen and ethidium bromide is compromised in an E. coli mutant with deletions of two DedA family members (3). It is therefore possible that transport of undecaprenyl-P-α-L-Ara4N is inefficient in certain DedA family mutants resulting in production of lipid A that lacks Ara4N and colistin sensitivity. We have renamed K. pneumoniae dedA dkcA (dedA of Klebsiella required for colistin resistance). In this work, we analyzed a ΔdkcA mutation in virulent, multidrug resistant K. pneumoniae ST258 and demonstrate an altered lipid A structure that is almost completely lacking Ara4N. In addition, we report that DkcA is required for lipid A modification-independent colistin resistance, instead requiring DkcA for optimization of membrane potential. Finally, we demonstrate DkcA is required for virulence in Galleria mellonella suggesting that colistin sensitivity can cause loss of virulence in vivo, consistent with previous studies demonstrating the importance of membrane polarization 39 and lipid A modifications 40 for virulence. This work supports the lipid A modification pathways 4,7 and DkcA 6 as potential targets of colistin adjuvants and anti-virulence drugs.

Results
Construction of a K. pneumoniae ST258 ΔdkcA mutant and a complementing pBAD-based plasmid (engineered with an apramycin-resistance marker to allow for selection in MDR K. pneumoniae) expressing dkcA with an N-terminal His-tag, has been described previously 6 . Measurement of the colistin minimal inhibitory concentration (MIC) using E-strips demonstrates that K. pneumoniae ΔdkcA is nearly eight times more sensitive to colistin than wild type (Fig. 1A) and that expression of K. pneumoniae dkcA or E. coli dedA from a plasmid restores sensitivity of the mutant to near wild type levels ( Fig. 1B) consistent with our previous observation 6 .
While K. pneumoniae DkcA is ~ 90% identical to E. coli DedA, most of our work with the E. coli DedA family has been with YqjA and YghB proteins because of the striking phenotypes of the double mutant strain [11][12][13]15 , and the ΔyqjA strain 16 . We have also characterized dkcA homologues from B. thailandensis and B. glumae named dbcA (dedA of Burkholderia required for colistin resistance), which are required for high-level colistin resistance in these species [17][18][19] . K. pneumoniae DkcA shares only ~ 30% amino acid identity to E. coli YqjA, YghB, and Burkholderia DbcA proteins but still possesses membrane-embedded charged amino acids E40, R133 and R139 in similar locations as found in their YghB/YqjA counterparts that we have shown are critical for function (see Supplementary Fig. S1 online for amino acid alignment) 14,15 . Therefore, we sequentially changed E40, R133 and R139 each to alanine and determined if the mutant proteins were capable of complementing the colistin sensitivity of the ΔdkcA strain. We found that all three mutants were inactive in this assay, demonstrating the importance of these charged amino acids (Fig. 1C). All proteins were found to be membrane associated, ruling out misfolding of the mutant proteins (Fig. 1D). These findings suggest that K. pneumoniae DkcA may function as a proton-dependent membrane transporter, similar to other members of the DedA family.
Since lipid A modifications contribute to colistin resistance in numerous Gram-negative species including Klebsiella pneumoniae 41,42 , the lipid A structure of wild type K. pneumoniae ST258 and ΔdkcA grown in LB media was analyzed using mass spectrometry. In wild type K. pneumoniae (Fig. 2a), major species are hexa-acyl lipid A (m/z: 1825.2) and its hydroxylated derivative (m/z: 1841.3). In addition, several species containing Ara4N were observed (designated by red font in Fig. 2; m/z: 1892. 8, 1956.3, 1972.3, and the phospoethanolamine/Ara4N modified 2079.3). A similar pattern of lipid A species was observed in wild type and ΔdkcA expressing dkcA from a plasmid ( Fig. 2b and d). However, in the ΔdkcA strain harboring the control vector, Ara4N modified species of lipid A were substantially decreased by MS (Fig. 2c), similar to what was found with a Burkholderia thailandensis ΔdedA strain 18 . Levels of the dual modified Ara4N/phosphoethanolamine species m/z 2079.3 were low but detectable at similar levels in all strains, suggesting the phosphoethanolamine modified species constitutes a minor portion of the total lipid A and is unaffected by the ΔdkcA mutation. See Fig. 2e for structures of these species. This observation in combination with the marked sensitivity to colistin is noteworthy and prompted us to analyze the K. pneumoniae DkcA protein and ΔdkcA strain in more detail.
In our previous work analyzing functions of DedA family members from E. coli, B. thailandensis, and B. glumae, we observed that mutation of DedA family members causes alterations in the membrane proton motive force. These changes were somewhat inconsistent in that E. coli and B. thailandensis mutants were somewhat depolarized 12,18 , but a B. glumae mutant strain was hyperpolarized 19 . This suggests that DedA family proteins may function in different manners in optimizing membrane potential depending upon the gene and organism in which they are found. Membrane potential was measured in K. pneumoniae ST258 and ΔdkcA and we found that the ΔdkcA strain was slightly but significantly hyperpolarized (Fig. 3A). We were unable to complement the membrane hyperpolarization of ΔdkcA with pBADdkcA at various levels of arabinose induction due to the observed effect of arabinose on membrane potential even in the wild type strain (data not shown www.nature.com/scientificreports/ if this hyperpolarization can result in colistin sensitivity, cells were briefly exposed to colistin (45 min) in the presence or absence of 25 µM CCCP, a proton ionophore, to cause membrane depolarization. Cell survival for all four strains was low in the presence of colistin alone. It was found that a brief preexposure to CCCP could partially rescue cells, both wild type and ΔdkcA, from the lethal effects of colistin (Fig. 3B). This result suggests that colistin may require optimum membrane potential as part of its mechanism of action. This hypothesis was tested in a series of experiments. Magnesium influx has been reported to alleviate hyperpolarization associated with ribosomal stress in Bacillus subtilis 43 . In contrast, exposure to sublethal ribosome targeting antibiotics such as spectinomycin can increase the number of cells that exhibit membrane hyperpolarization 43 . Therefore, we tested the ability of extracellular Mg ++ to rescue K. pneumoniae strains from the lethal effects of colistin and sub-MIC concentrations spectinomycin to sensitize strains to colistin. While Mg ++ is known to protect Gram negative bacteria from the effects of colistin by competing for binding with negatively charged sites on LPS 44 , it has also been shown to contribute to polymyxin resistance independently of LPS and the PhoPQ regulon by its ability to depolarize the membrane 39 . We measured the effects of Mg ++ on membrane potential and colistin sensitivity of K. pneumoniae ST258 and ΔdkcA. Growth in the presence of 50 mM or higher of Mg ++ causes membrane depolarization of K. pneumoniae ΔdkcA (Fig. 4A). Mg ++ at 50 mM completely restores colistin resistance to this mutant (Fig. 4B) under conditions predicted to repress expression of the PhoPQ operon 45 . These results collectively show that alleviation of membrane hyperpolarization of wild type cells or ΔdkcA strain with Mg ++ results in increased colistin resistance, again supporting a role of proper regulation of membrane potential as critical for colistin resistance, likely independently of any alterations to the structure of lipid A.
We then measured the effect of spectinomycin-induced hyperpolarization on the colistin sensitivity of K. pneumoniae ST258. We first determined the spectinomycin MIC of the strain to be ~ 4000 µg/ml (see Supplementary Fig. S2 online). Exposure of wild type K. pneumoniae ST258 to a sub-MIC concentration of 500 µg/ml spectinomycin causes hyperpolarization of membrane potential (Fig. 5A). Spectinomycin treatment of ST258 also causes increased colistin sensitivity, decreasing the colistin MIC by ~ 100 fold (Fig. 5B). This indirectly shows hyperpolarization of membrane potential may be responsible for increased colistin sensitivity and again implies that maintenance of optimal membrane potential is required for colistin resistance. This is also consistent with the observation above that depolarization using a brief exposure to CCCP can partially rescue cells from the lethal effects of colistin (Fig. 3B). Growth in the presence of 30 mM or higher of Mg ++ causes membrane depolarization of K. pneumoniae ST258, even when hyperpolarized by exposure to spectinomycin (Fig. 5C). Mg ++ also causes significant eightfold increase of the colistin MIC observed in the presence of spectinomycin (Fig. 5D), again showing a strong correlation between membrane polarization and colistin sensitivity.  (Table 1) and performed similar measurements. ArnT is the Ara4N transferase required for periplasmic modification of lipid A with Ara4N and colistin resistance [46][47][48] . The ΔarnT strain is sensitive to colistin in the absence of complementation (Fig. 6A). Spectinomycin treatment of K. pneumoniae ΔarnT causes hyperpolarization of membrane potential to a similar extent as that seen in K. pneumoniae ST258 (Fig. 6B). While ΔarnT is exquisitely sensitive to colistin (MIC ~ 0.4 µg/ml), exposure to 2000 µg/ml spectinomycin decreases the MIC by as much as 2.7-fold (Fig. 6C). This observation serves as direct evidence showing hyperpolarization of membrane potential is responsible for increased colistin sensitivity and supports our hypothesis that maintenance of optimal membrane potential is required for colistin resistance independent of lipid A modification.
Loss of resistance to colistin suggests mutation of dkcA may cause loss of virulence since CAMPS of the innate immune system use the same mechanism to kill bacteria as colistin; i.e. by binding to and disrupting negatively charged bacterial membranes leading to cell lysis and death 32 . In order to measure virulence in vivo, we chose the Galleria mellonella moth larvae model due to its moderate costs and ethical considerations. G. mellonella has been used successfully to measure virulence of K. pneumoniae and many other bacterial pathogens 52 . Using this model, we found the ΔdkcA was partially compromised for virulence over the course of the experiment (Fig. 7A). Very few viable bacteria could be recovered from ΔdkcA infected larvae, compared to wild type infected larvae, where bacteria could be recovered at 12 and 16 h post infection (Fig. 7B). This suggests that mutations of dkcA Cells grown to OD 600 of 0.8 in MH2 media containing 50 µg/ml Apr, 0.1% arabinose were diluted ten-fold and incubated in the same media alone or exposed to 3 µg/ml colistin or 3 µg/ml colistin plus 25 µM CCCP for 45 min at 37 °C. After exposure, cells were washed with fresh MH2 medium twice, resuspended in MH2 and CFU/ml was determined by plating serial dilutions. Each experiment was repeated three times. Statistical significance was calculated by unpaired Student's t-test. **: p < 0.01. ***: p < 0.001.  www.nature.com/scientificreports/ and other genes 6 that lead to K. pneumoniae colistin sensitivity can also cause loss of virulence and these gene products may also be attractive targets for anti-virulence agents.

Discussion
Multidrug resistant bacterial infections pose an enormous public safety risk and are a challenge to modern medicine, made worse by a lack of new antimicrobial drugs 2,3 . The emergence of carbapenem-resistant Klebsiella pneumoniae is of particular concern and has forced the reintroduction of colistin therapy as a last resort treatment. Colistin is an antimicrobial peptide belonging to the polymyxin family that can cause numerous side effects including nephrotoxicity so dosage must be strictly monitored 53 . The recent emergence of mobilized colistin resistance genes that spread via horizontal gene transfer has further compromised the use of colistin clinically 54 . Helper drugs that can lower the effective dosage of colistin by targeting the "secondary resistome" would be useful therapeutics 6 . Drugs that alter colistin resistance may also result in sensitivity to the cationic  www.nature.com/scientificreports/ antimicrobial peptides of the innate immune system and thus act to target virulence even in the absence of antibiotic administration 4,5 . Potential targets may be Klebsiella pneumoniae DkcA or ArnT as each gene deletion strain displays a similar eight-fold reduction in MIC to colistin. Gram-negative bacteria often become resistant to colistin by activation of genes and pathways that lead to the covalent modification of anionic cell surface lipopolysaccharide with cationic substituents, causing loss of electrostatic binding by the cationic antibiotic 35,55 . In Klebsiella and other species, these pathways are controlled by the PhoPQ and PmrAB two-component signaling pathways which lead to modification of lipid A with Ara4N and phosphoethanolamine 55 . While K. pneumoniae does harbor a chromosomal homolog of EptA involved in phosphoethanolamine modification of lipid A 56 , our MS analysis suggests this species makes up a minor component of the total lipid A suggesting Ara4N modification is the main contributor to colistin resistance.
Less well characterized is the contribution of membrane potential to colistin sensitivity and resistance. Here, we show that the K. pneumoniae ΔdkcA is slightly but significantly hyperpolarized and tested the contribution of this hyperpolarization to the observed colistin sensitivity. First, we showed that brief pre-exposure of both wild type and ΔdkcA strains to the proton ionophore and depolarizing agent CCCP results in higher survival in the presence of a lethal dose of colistin. The augmented survival against colistin by brief pre-exposure to CCCP is probably due to cytoplasmic acidification by CCCP, consistent with our hypothesis 20 . Next, we showed that treatment with sub-lethal concentrations of the antibiotic spectinomycin causes hyperpolarization of K. pneumoniae consistent with what was shown for B. subtilis 43 . Importantly, this hyperpolarization was accompanied by an increase in sensitivity to colistin. Therefore, we could mimic the effect of the ΔdkcA mutation on both the membrane potential and colistin sensitivity using sub-lethal concentrations of spectinomycin. Relieving the hyperpolarization with magnesium also rescued the sensitivity to colistin. Finally, we showed that the effect of membrane potential on colistin sensitivity was independent of lipid A modification by demonstrating the same effect in a K. pneumoniae ΔarnT mutant. While ΔarnT is sensitive to colistin, consistent with a previous report 6 , exposure to spectinomycin increased this sensitivity while also hyperpolarizing the membrane. Collectively, these results suggest that membrane hyperpolarization can directly lead to colistin sensitivity independently of any changes to the cell surface LPS (Fig. 8).
The DedA family of membrane proteins is widely distributed in nature, found in all kingdoms. Until recently, most of what is known about the family came from studies in E. coli due to pleiotropic phenotypes of inactivating mutations. These phenotypes include defects in growth, cell division and hypersensitivity to alkaline pH, antibiotics and membrane penetrating dyes [10][11][12][13]15,16 . The correction of these phenotypes by growth in slightly acidic pH and the presence of membrane embedded charged amino acids suggest that members of the DedA family are proton-dependent transporters required for maintenance of optimal PMF 12,14-16 . Our recent studies in Burkholderia thailandensis and B. glumae also support this hypothesis [17][18][19][20] . DedA family proteins have an evolutionary relationship and, indeed may share structural similarity to LeuT, a bacterial homolog of the neurotransmitter:sodium symporter family 58,59 .
While there is no published structure of any DedA family member, modelling of the structure of Human TMEM41b has been carried out which is predicted to possess structural features found in Cl -/H + antiporters 60 . The function of human TMEM41B is unknown but it has been identified as being required for autophagosome formation [61][62][63] and as a host factor required for infection with flaviviruses as well as SARS-CoV-2 64,65 . While the eukaryotic proteins of the DedA family containing the so-called VTT domain (for VMP, TMEM41, Tvp38) 63 are distantly related to their bacterial counterparts, and a functional relationship has not been established, the presence of an absolutely conserved glycine residue suggests an evolutionary relationship 66 .
Invertebrate models of bacterial virulence are increasingly common due to lower cost than traditional rodent studies 52 . There are no ethical constraints and their short life span makes them ideal for large-scale studies. While insects do not possess an adaptive immune response, their innate immune response is quite similar to mammals 67 . Insects possess Toll-like receptors, CAMPs, and neutrophil-like cells called hemocytes 68 . There are currently more than 2000 articles published on PubMed on the wax moth Galleria mellonella infection model. Galleria mellonella has been used to study virulence of Klebsiella pneumoniae 40,52,69 and other bacterial pathogens 70,71 . We show here that Klebsiella pneumoniae DkcA is required for virulence in G. mellonella, which may be linked to the sensitivity to colistin. These studies are consistent with previous studies demonstrating the importance of optimal membrane potential 39 and cationic lipid A modifications 40 for virulence.
In summary, we have shown that the Klebsiella pneumoniae DkcA protein is required for colistin resistance in a lipid A-dependent and -independent manner. DkcA displays similarity to proton-dependent membrane transporters and is required for virulence in moth larvae. The bacterial DedA family is evolutionarily conserved and remains an intriguing mystery in terms of its physiological function and it potential for therapeutic targeting.

Material and methods
Bacterial growth conditions. E. coli cultures were grown on LB medium (1% tryptone, 0.5% yeast extract and 1% NaCl). Klebsiella pneumoniae was grown in LB or cation adjusted Mueller-Hinton broth 2 (MH2; Sigma-Aldrich). Antibiotics colistin, ampicillin (Amp; 100 µg/ml), apramycin (Apr; 50 µg/ml) and spectinomycin (at indicated concentrations) were purchased from Sigma-Aldrich or VWR. Cultures were grown at 37 °C unless otherwise indicated. For complementation analysis with pBAD-or pSCrhaB2-derived vectors, arabinose or rhamnose at indicated concentrations was included in growth media. Bacterial strains and plasmids are listed in Table 1.

Creation of Klebsiella pneumoniae
ΔarnT. The K. pneumoniae ST258 arnT gene was disrupted using lambda red recombination 50,72 . Bacteria were grown in low salt LB (5 g/L NaCl), pH 8.0 up to the arnT disruption step. pACBSR-Hyg R was introduced into ST258 by electroporation and grown at 30 °C. Knockout primers were www.nature.com/scientificreports/ designed in a way that arnT_KO_forward primer has 60 bp homology to the 5' upstream region and arnT_KO_ reverse primer has 60 bp homology to the 3' downstream region of the arnT gene (Table 2). An apramycin resistance cassette (Apr R ) was amplified from pIJ773 50 using these primers. The linear DNA fragment was purified using QIAquick Gel Extraction Kit (Qiagen). Purified linear DNA was treated with DpnI and repurified prior to introduction by electroporation into ST258/pACBSR-Hyg R . Following recovery in the absence of antibiotic, the cells were plated on 50 µg/mL Apr. Resistant colonies were screened for colistin sensitivity and gene disruption was verified using PCR primers P1F, P2R, P3F, P4R ( Site-directed mutagenesis. Site-specific mutants were created by using a previously published protocol 74 . The primers carrying the site-specific mutations ( Table 2) were used in a PCR reaction to amplify a vector containing the wild type gene. The PCR product was then digested with Dpn1 and used to transform competent  Membrane preparation and western blotting. Cell membranes was prepared from exponential phase cultures containing indicated plasmid DNA grown under inducing conditions (0.1% arabinose) using a previously published protocol 15,75 . Equal amounts of protein were resolved by SDS-PAGE and transferred to PVDF membrane. Western blotting was carried out using penta-His (Qiagen) primary antibody at 1:5000 dilution and goat-anti-mouse horseradish peroxidase (HRP) conjugated secondary antibody (Pierce) at 1:5000. Detection was performed using the ImmunStar HRP kit (Bio-Rad).

Susceptibility to colistin and other antibiotics.
For testing the susceptibility on solid medium, overnight cultures were freshly diluted 1:100 in MH2 media with appropriate antibiotics and additives, and grown to OD 600 ~ 0.6 at 37 °C in a shaking incubator. Five microliters of serially log 10 -diluted cells were spotted onto MH2 or LB agar plates containing various concentrations of antibiotic. Growth was analyzed after incubation for 24 h at 37 °C unless otherwise indicated. Antibiotic MIC's were measured using the broth microdilution method in 96 well plates or using colistin test gradient strips (Biomerieux). All experiments were repeated at least three times.
Mass spectrometry. For isolation of lipid A, K. pneumoniae cultures were grown at 37 °C to an OD 600 of ~ 1.0.
Lipid A chemical extraction was carried out after mild acidic hydrolysis of LPS as previously described 76,77 . For visualization of lipid A by mass spectrometry, lipids were analyzed using MALDI-TOF (ABI 4700 Proteomic Analyzer) in the negative-ion linear mode as previously described 78,79 . Briefly, lipid A samples were dissolved in a mixture of chloroform-methanol (4:1, vol/vol), and 1 µl of sample was mixed with 1 µl of matrix solution. The matrix consisted of 5-chloro-2-mercaptobenzothiazole (CMBT) (20 mg/mL) resuspended in chloroform-methanol-water (4:4:1, vol/vol/vol) mixed with saturated ammonium citrate (20:1, vol/vol). One µl of sample-matrix mixture was loaded on to MALDI target plate for final analysis.
Measurement of membrane potential. Measurement of membrane potential was performed using JC-1 dye 80 . Briefly, 2.5 × 10 7 cells from the overnight cultures were inoculated in 20 ml of fresh LB broth in a 250 ml flask and grown for about 1 h 40 min at 37 °C with shaking. ~ 6 × 10 8 cells were taken from each culture, washed with permeabilization buffer PB, pH 7.5 (10 mM Tris, 1 mM EDTA, 10 mM glucose) and resuspended in fresh PB. 3 µM of JC-1 dye was added and incubated in the dark at 30 °C for 30 min. Cells were washed and resuspended in PB buffer and fluorescence measurements were carried out using a JASCO FP-6300 spectrofluo- Table 2. Oligonucleotide primers used in this study. www.nature.com/scientificreports/ rometer. Relative membrane potential is expressed the ratio of red (595 nm) to green (530 nm) fluorescence with excitation of 488 nm.

Galleria mellonella model of virulence. Larvae purchased from Carolina Biological Supply Company
(item# 143,928) were injected with 5 × 10 5 CFU of bacteria resuspended in 5 µl PBS pH 7.4 or PBS alone. After the injection, larvae were incubated at 37 °C in the dark over a period of 96 h. Larvae were monitored for their survival every 24 h. Dead larvae were identified by the lack of motility or bodily movement against physical stimuli. To measure bacterial survival in G. mellonella, larvae were injected with 5 × 10 5 CFU in PBS pH 7.4 and incubated at 37 °C in the dark. Larvae were homogenized at indicated times and serial dilutions were plated in LB agar containing 100 µg/mL apramycin. No apramycin resistant bacteria were isolated from control larvae injected with PBS under these conditions.