Structural basis for antibacterial peptide self‐immunity by the bacterial ABC transporter McjD

Abstract Certain pathogenic bacteria produce and release toxic peptides to ensure either nutrient availability or evasion from the immune system. These peptides are also toxic to the producing bacteria that utilize dedicated ABC transporters to provide self‐immunity. The ABC transporter McjD exports the antibacterial peptide MccJ25 in Escherichia coli. Our previously determined McjD structure provided some mechanistic insights into antibacterial peptide efflux. In this study, we have determined its structure in a novel conformation, apo inward‐occluded and a new nucleotide‐bound state, high‐energy outward‐occluded intermediate state, with a defined ligand binding cavity. Predictive cysteine cross‐linking in E. coli membranes and PELDOR measurements along the transport cycle indicate that McjD does not undergo major conformational changes as previously proposed for multi‐drug ABC exporters. Combined with transport assays and molecular dynamics simulations, we propose a novel mechanism for toxic peptide ABC exporters that only requires the transient opening of the cavity for release of the peptide. We propose that shielding of the cavity ensures that the transporter is available to export the newly synthesized peptides, preventing toxic‐level build‐up.


Introduction
ATP-binding cassette (ABC) transporters (exporters) are one of the largest superfamilies of membrane transporters and are found in both bacteria and eukaryotes (Holland et al, 2003); they are usually involved in efflux or detoxification pathways including multi-drug resistance in both bacterial and eukaryotic cells. Bacteria, under nutrient starvation, produce and release antibacterial peptides, microcins, which can kill microcin-sensitive cells and therefore provide more nutrients for the surviving bacteria (Duquesne et al, 2007a). The microcin J25 (MccJ25) is a plasmid-encoded, ribosomally synthesized and post-translationally modified antibacterial peptide consisting of 21 amino acids and adopting a lasso structure (Rosengren et al, 2003); its C-terminal tail threads through an Nterminal eight-residue macrolactam ring, where it is locked by bulky amino acid side chains, forming a compact interlocked structure. Four genes are required for the biosynthesis and export of MccJ25 (Duquesne et al, 2007b): mcjA encodes the linear precursor of MccJ25; mcjB and mcjC encode enzymes involved in the posttranslational modification of McjA to the lasso structure; and mcjD encodes an ABC exporter that is required for both MccJ25 secretion and self-immunity of the producing strain towards MccJ25 (Choudhury et al, 2014). Since the genes are under the same operon, production and export from the cell are very efficient and toxic levels of MccJ25 are prevented from building up. Inside the target cell (Runti et al, 2013;Mathavan et al, 2014), MccJ25 inhibits the bacterial RNA polymerase (Semenova et al, 2005).
We have previously functionally characterized and determined the high-resolution structure of McjD from Escherichia coli (Choudhury et al, 2014;Fig 1A). Its core architecture is composed of a dimeric transmembrane domain (TMD) of 12 transmembrane (TM) helices, which forms the translocation pathway across the membrane bilayer and contains the ligand binding site, and a dimeric nucleotide binding domain (NBD) where ATP binds and is hydrolysed. ABC exporters either use the alternating access mechanism or outward-only mechanism (Perez et al, 2015) to transport their substrates. ABC exporters that use the alternating access mechanism switch between inward-and outward-facing states, which exposes the ligand binding site alternatively to the inside or outside of the membrane, coupled to ATP binding and hydrolysis (Beis, 2015). The previous structure of McjD was determined in a nucleotide-bound outward-occluded conformation (occluded at both sides of the membrane), representing an intermediate state between the outward-and the inward-facing conformations (Choudhury et al, 2014).
In order to understand the detailed mechanism of toxic peptide export by bacterial cells, it is important to trap the transporter in different conformations. Here, we have determined the structure of McjD in a novel conformation, apo inward-occluded and an additional nucleotide-bound state, high-energy outward-occluded intermediate with bound ATP-vanadate (ADP-VO 4 ). We further characterized these new states in E. coli membranes using predictive cysteine cross-linking in inside-out vesicles (ISOVs). Using a spinlabelled McjD mutant and pulsed electron-electron double resonance (PELDOR), also known as DEER (double electron-electron resonance), the conformation of McjD was investigated in bicelles. In addition, we applied molecular dynamics (MD) simulations to study the dynamics of the outward-occluded conformation of McjD in the presence of ATP or ADP molecules bound at the NDBs. The new conformations in combination with the PELDOR data, transport assays in proteoliposomes and molecular dynamics simulations have significant mechanistic implications in understanding the detailed mechanism of the antibacterial peptide exporter McjD.

Structures of McjD in different conformations
In order to gain a detailed understanding of the conformations that McjD adopts during the transport cycle, we have determined its structure in the presence of the transition state analogue ADP-VO 4 (mimic of ATP hydrolysis) and in the absence of nucleotides, that is in the apo form, at 3.4 Å and 4.7 Å resolution, respectively (Fig 1B and C). We determined the structure of McjD in complex with ADP-VO 4 by molecular replacement using the McjD-AMPPNP (adenosine 5 0 -(b,c-imido)triphosphate) (PDB ID: 4PL0) structure as a search model (Choudhury et al, 2014). The structure was refined with an R work of 25.7% and R free of 26.0%. Clear electron density was observed for ADP-VO 4 and Mg 2+ (Fig EV1A-D). The presence of vanadate was verified by collecting the data close to the vanadium K-edge, 2.26 Å wavelength and calculating anomalous difference electron density maps (see Materials and Methods; Fig EV1C). At both NBDs, two strong electron density peaks of 14 and 20 sigma, respectively, confirmed the presence of two VO 4 groups. McjD-ADP-VO 4 is almost identical to the McjD-AMPPNP, as the two structures can be superimposed with a root-mean-square deviation (rmsd) of 0.7 Å over 560 Ca atoms (Fig EV2A and B). In contrast to the MsbA-ADP-VO 4 structure that adopts an outward-open conformation as a result of domain intertwining (Ward et al, 2007), the McjD-ADP-VO 4 structure is outward-occluded without domain intertwining (Fig 2A). We call it here high-energy intermediate outwardoccluded. The TMD dimer interface in McjD-ADP-VO 4 is formed between TM2 and TM5/TM6 from one subunit with the equivalent TMs from the opposite subunit similar to the McjD-AMPPNP structure. In the presence of ADP-VO 4 , the NBDs are dimerized with the P-loop and ABC signature motifs involved in the binding of the nucleotide. The McjD-ADP-VO 4 structure represents the transition state of a water molecule making a nucleophilic attack on the c-phosphate of ATP.
The McjD-apo structure was also determined by molecular replacement using the McjD-AMPPNP structure but only after splitting the McjD monomer into two domains, TMD and NBD (see Materials and Methods). The structure was refined to an R work of 31.3% and R free of 33.4%. Clear electron density could be observed for both the TMD and NBDs (Fig 3A and B). Inspection of the NBDs did not reveal any electron density for nucleotides that may have been co-purified with McjD ( Fig 3A). In the absence of nucleotides, ABC transporters adopt an inward conformation with their TMDs separated and NBDs disengaged (Ward et al, 2007;Fig 2B). A striking difference between McjD-apo and other apo ABC exporters, such as MsbA (Ward et al, 2007) and PglK (Perez et al, 2015), is that its TMD is in an occluded conformation, similar to the McjD-AMPPNP structure, whereas the NBDs have disengaged in the absence of nucleotides. This conformation is called here apo inward-occluded. Inward-facing ABC exporters display domain intertwining of TMs 1-3 and 6 from one subunit and TMs 4-5 from the opposite subunit, which results in opening of the TMD to the cytoplasmic side of the inner membrane (Ward et al, 2007;Perez et al, 2015). In the inward-occluded McjD, TMs 3-5 from one subunit have moved towards the equivalent TMs of the opposite subunit, resulting in the loss of intertwining and occlusion of the cytoplasmic opening (Figs 2B and EV2). The McjD-apo structure can be superimposed with the McjD-AMPPNP structure with an rmsd of 2.1 Å over 569 Ca atoms; their TMDs can be superimposed with an rmsd of 0.7 Å over 290 Ca atoms. The heterodimeric human sterol apo-ABCG5/8 (Lee et al, 2016) and Pseudomonas aeruginosa lipopolysaccharide apo-LptB 2 FG (Luo et al, 2017) do not display intertwining either.
In the absence of nucleotides, the NBDs of ABC exporters disengage (Ward et al, 2007;Perez et al, 2015). The NBDs of the McjDapo have separated by 7.9 Å (measured between S509 and S509') relative to the McjD-AMPPNP structure, a distance much shorter compared to MsbA-apo with a distance of 35 Å (Ward et al, 2007). The NBDs have moved in a "scissors-like" motion similar to the transition between inward and outward-open MsbA. A small degree of disengagement between the NBDs has also been reported for the heterodimeric ABC exporter TM287/288 (Hohl et al, 2012;Hohl et al, 2014), ABCG5/8 (Lee et al, 2016) and LptB 2 FG (Luo et al, 2017).

Cysteine cross-linking in ISOVs
The presence of apo inward-occluded and high-energy intermediate outward-occluded states have not been observed or characterized in other ABC exporters. We selected specific amino acids along the periplasmic, cytoplasmic and NBD sides to perform predictive cysteine cross-linking in ISOVs to further characterize them in their native lipid environment. We have previously shown that the occluded McjD is a result of the movement of TMs 1-2 and TMs 1 0 -2 0 and we characterized this conformation by cysteine cross-linking L53C in the presence and absence of nucleotides, ATP/Mg 2+ and AMPPNP (Choudhury et al, 2014). Since the McjD-ADP-VO 4 structure is also in an outward-occluded conformation, this mutant was further characterized in the presence of ADP-VO 4 . McjD-L53C is capable of forming cross-linked dimers in the presence of ADP-VO 4 , suggesting that the high-energy post-hydrolysis McjD also adopts an occluded conformation in the E. coli membrane similar to our crystal structure (Figs 4A and EV3A).
Since an inward-occluded conformation without subunit intertwining has only been observed for McjD and not for other ABC exporters, we also characterized this new conformation by predictive cysteine cross-linking in ISOVs. We designed the A122C mutant at the beginning of TM3 on the cytoplasmic side of McjD, with a Ca-Ca 0 distance of 7.3 Å . The McjD-A122C can form cross-linked dimers in ISOVs in the absence of nucleotides but less efficiently compared to the AMPPNP or ADP-VO 4 cross-linking experiments (Figs 4B and EV3B). This is the first time that an apo inward-occluded conformation has been reported in E. coli membranes. A possible explanation for the reduced A122C cross-linking in ISOVs in the absence of nucleotides is that McjD alternates between inward-open and inward-occluded conformations. Nevertheless, the transporter is capable of adopting this conformation within the membrane. In the presence of MccJ25, we observed small enhancement of A122C cross-linking (Figs 4B and EV3C), suggesting that binding of the substrate brings the TM helices closer.
In the absence of nucleotides, the NBDs of ABC exporters disengage. There are conflicting reports regarding the degree of NBDs disengagement in bacterial ABC exporters (Borbat et al, 2007;Zou et al, 2009;Zoghbi et al, 2016). We sought to characterize the degree of McjD-apo NBD disengagement by cysteine cross-linking of the S509C mutant. This mutant was capable of forming cross-linked dimers even in the absence of CuCl 2 , even after the ISOVs had been treated with 1 mM reducing agent, suggesting that the NBDs are also close within the E. coli membrane (Fig 4C). In the presence of AMPPNP and ADP-VO 4 , the McjD-S509C can also form cross-linked dimers that are in agreement with our crystal structures (Figs 4C and EV3D). The predictive cysteine cross-linking data both in the presence and in the absence of nucleotides indicate that McjD mostly exists in an occluded conformation with its NBDs closely engaged in the E. coli membrane. Pulsed electron-electron double resonance Since our crystal structures and predictive cross-linking experiments revealed that McjD exists mostly in an occluded conformation, we performed Q-band PELDOR distance measurements on spin-labelled McjD mutants. We aimed to investigate the conformational changes associated with the export of the antibacterial peptide MccJ25 from the cytoplasmic to the periplasmic leaflet of the inner membrane. The data discussed below were recorded on McjD reconstituted in bicelles (Ward et al, 2014). Based on our crystal structures, a pair of R1 spin labels was introduced at the periplasmic side of the TMD at the loop connecting TMs 1 and 2 (L52R1), and another at the NBDs at position C547 (C547R1) since they could provide us with distances along the transport cycle and reveal stable states ( Fig 5A). The ATPase activity of the labelled mutants was similar to the wild-type, unlabelled McjD. The The EMBO Journal McjD structure and mechanism Kiran Bountra et al mtsslWizard software (Hagelueken et al, 2012) was used to predict the interspin distances. The L52R1 mutant produced PELDOR time traces of good quality with clearly visible oscillations. In the apo state, McjD L52R1 produces a sharp interatomic distance peak at 28 Å and a broad distribution of smaller peaks in the range between 40 and 60 Å ( Fig 5B). The main peak at 28 Å is in close agreement with the distance predicted from the crystal structures using the mtsslWizard software (Hagelueken et al, 2012;Fig 5B, grey shade).
It was previously shown that in the presence of AMPPNP or ADP-VO 4 (Borbat et al, 2007). In contrast, our PELDOR data indicated that McjD did not undergo such large conformational changes upon nucleotide binding. The interatomic distance distributions in the presence of either AMPPNP or ADP-VO 4 were very similar to those of the apo protein ( Fig 5C and D). This is in very good agreement with both our AMPPNP-and ADP-VO 4bound crystal structures ( Fig 5C and D Fig 5D and E). A possible explanation for the apparent invariability of the interspin distances is that the outward-open conformation could be a short-lived and/or scarcely populated state, which would not be captured by PELDOR.
To analyse the movement of the NBDs of McjD during the transport cycle, we screened various cysteine mutants on the NBD surface. However, an active, well-behaved transporter could only be produced when the native cysteine at position 547 was used as an anchor for the spin label. The corresponding PELDOR time traces are shown in Fig 5F and G. The time traces did not show any clear oscillations, indicating a highly dynamic structure at the spin-labelled position. As a consequence, it was not possible to unambiguously determine the intermolecular background and the underlying (presumably very broad) distance distributions cannot be reliably determined.

Uptake of hoechst in proteoliposomes
Since we could not observe a stable outward-open McjD by PELDOR, we performed Hoechst transport assays in proteoliposomes using the L53C mutant to demonstrate that it transiently exists. We have previously shown that Hoechst 33342 (Hoechst) can be transported by McjD in proteoliposomes (Choudhury et al, 2014). In the absence of direct MccJ25 transport measurement in proteoliposomes, we reconstituted wild-type McjD in liposomes and The EMBO Journal McjD structure and mechanism Kiran Bountra et al performed competition transport assays with Hoechst and MccJ25 ( Fig 6A). In the absence of MccJ25 and presence of ATP, the proteoliposomes show reduced fluorescence for Hoechst, whereas in the absence of ATP, the fluorescence increased. In the presence of 10 molar excess MccJ25, we observe an increase in fluorescence relative to the ATP-only measurement, suggesting that MccJ25 competes with Hoechst for the McjD cavity and subsequent transport. L53C can be effectively cross-linked in the presence of CuCl 2 ( Fig 4A) and our transport assays were performed in the presence and absence of the oxidizing agent to probe the opening of the periplasmic side of McjD ( Fig 6B). In the absence of ATP, the proteoliposomes could not transport Hoechst, whereas the addition of ATP resulted in Hoechst uptake by McjD-L53C ( Fig 6B). Incubation of L53C with CuCl 2 , L53C ox and addition of ATP did not show any McjD-dependent Hoechst uptake ( Fig 6B); thus, disulphide locking of the periplasmic side of McjD is detrimental to its transport activity. After 10 min of L53C ox "transport activity", the proteoliposomes were treated with 1,4-dithiothreitol (DTT), to reduce the disulphide bridges (L53C re ). McjD-L53C re was capable to mediate the ATPdependent transport of Hoechst in proteoliposomes (Fig 6C and D) as a result of unlocking the periplasmic side of McjD, suggesting that McjD has to sample an outward-open conformation for the release of the substrate. L53C re shows a slightly reduced transport activity which could be due to incomplete reduction of all disulphides. This is our first evidence that McjD adopts an outward-open conformation that is mediated by both ATP and substrate.

McjD cavity accessibility by PEGylation in ISOVs
Both our structural and functional data show that McjD exists mostly in an occluded conformation in the absence of a ligand. We have previously identified residues within the cavity that are involved in ligand recognition using ligand-induced ATPase assays and mutagenesis (Choudhury et al, 2014). Here, we probed the accessibility of the cavity by the substrate using cavity-specific cysteine mutants modified with methoxypolyethylene glycol maleimide 10,000 Da (mPEG10k) in ISOVs; we selected mPEG10k despite being significantly larger than MccJ25 (2.1 kDa), as it would enable us to observe a distinct electrophoretic mobility shift in SDS-PAGE. The mutants, N134C, I138C, I287C and T298C, were designed to study the cavity from its cytoplasmic to periplasmic side ( Fig 7A).
In the presence of 1 mM mPEG10k and absence of the nucleotides AMPPNP, ATP or ADP-VO 4 , all mutants displayed higher molecular weight species ( Fig 7B); some higher molecular weight species were also observed in the presence of nucleotides but these could be attributed to apo McjD (AMPPNP and ADP-VO 4 can be consumed by other ATP-binding proteins in the ISOVs; therefore, not all of the McjD molecules were inhibited). These assays were performed in triplicate with different membrane batches and provided similar results. mPEG10k is not membrane permeable and can only bind to solvent-accessible cysteines, suggesting that the McjD cavity becomes exposed to the aqueous solvent environment. This is the first evidence that McjD adopts an inward-open conformation that allows ligand access to its cavity. Owing to the homodimeric nature of McjD, the mutants are present in both halves of the transporter. The N134C and I138C mutants display stronger interaction with mPEG10k compared to I287C and T298C since they are located closer to the cytoplasmic side of the cavity (Fig 7A and B). The I287C (I278C') and T298C (T298C') mutants are closer to the periplasmic side of the cavity and in closer proximity to each other; therefore, PEGylation of the first half transporter would prevent a second mPEK10k to bind to the equivalent site of the second half due to steric classes, which can explain the reduced PEGylation. The only evidence that MccJ25 is recognized by McjD is by ligand-induced ATPase assays (Choudhury et al, 2014) and ligand binding studies by non-denaturing mass spectrometry (Mehmood et al, 2016). From these studies, it is unclear how the hydrophobic peptide MccJ25 enters the McjD cavity. We sought to answer the following question: (i) does MccJ25 enter through a lateral opening from inside the membrane as suggested for other ABC transporters (Aller et al, 2009) McjD from the aqueous cytoplasmic side of the transporter rather than a lateral opening in the membrane. If McjD only existed in an occluded conformation and MccJ25 were to enter the cavity from the membrane, mPEG10k should not have been able to modify the N134C and T298C mutants since it is not membrane permeable and can only modify solvent-accessible cysteines. The sampling of an inward-open conformation is not affected by pH. The PEGylation is as effective at pH 9 as at pH 7.5 (Fig EV3F), suggesting that the pH does not induce an inward-occluded conformation.

Molecular dynamics simulations
We used MD simulations to investigate whether changes in the outward-occluded state of the McjD transporter can be induced by the presence of ATP or ADP in the nucleotide binding sites, in the absence of the MccJ25 substrate. McjD-AMPPNP was chosen as input structure, and embedded in a POPC lipid bilayer (see Materials and Methods). The distance between specific pairs of residues in the periplasmic loops, in the cytosolic side of the cavity (cavity bottom) and in the NBDs was monitored during the simulation time.
Periplasmic loops L53 lies at the periplasmic end of TM1 of each monomer (Fig 8A). In the outward-occluded state, the Ca distance between the two L53 residues is 6 Å . In a previously modelled outward-facing state, the Ca distance between the two L53 pairs increases to 35 Å , as a consequence of the opening of the periplasmic end of the TM helices (Gu et al, 2015). During the simulation time, the L53 distance remains close to that of the input structure, in the presence of either ATP or The EMBO Journal McjD structure and mechanism Kiran Bountra et al ADP, suggesting that no separation occurs between the TM1 helices ( Fig 8A). To further assess possible structural changes at the periplasmic end of the TM helices, we monitored the distance between D51 (TM1) and S282 (TM6), highly conserved among the transporters used to generate the outward-facing model of McjD (Gu et al, 2015). In such a state, these residues move apart, and their Ca distance increases from ca. 8 of the outward-occluded state to ca. 25 Å . During the simulations, no opening motions are detected for these pairs of residues (Figs 8A and EV4A). However, a smaller intensity peak can be noticed for the L53-L53 pair at distances ranging from 12 to 14 Å , for both the ADP-and ATP-bound simulations (Figs 8A and EV4A). The structures with this larger separation, taken after 600 ns, highlight a different arrangement for the L53-L53 pair (Fig EV4B). The opening between the two L53 residues is, however, still limited compared to the opening in the outward-facing state; thus overall, during the simulation time, we observe no transitions towards an outward-facing conformation of the periplasmic ends of the TMDs.

Cavity bottom
The Ca distance between the two A122 residues (Fig 8B) is 7 Å in the input structure. On average, this distance is ca. 6 Å for both the ADP-and the ATP-bound simulations (Fig 8B). For the ADP-bound systems, however, more simulations sample larger distances, up to The EMBO Journal ca. 12 Å (Fig EV4C), closer to the value of the outward-facing model (Gu et al, 2015).

NBDs
The distance between the two S509 residues targeted by crosslinking studies is 6 Å in the initial structure and 8 Å in the apo McjD structure. During the simulation time, we observe no separation at the S509 level for both the ADP-and the ATP-bound systems (Figs 8C and EV4D). Limited NBD dissociation was also confirmed by the distances between the two D512-H537 pairs, where D512 is the aspartate of the conserved D-loop and H537 is the histidine of the Switch motif (Figs 8C and EV4D).

Discussion
In this study, we have determined the McjD structure in a novel conformation, apo inward-occluded and an additional Since these new conformations have neither been previously reported nor observed for ABC transporters, we performed predictive cysteine cross-linking in E. coli ISOVs to verify their existence in the lipid environment. The cross-links covered the periplasmic side (L53), cytoplasmic side (A122) and NBDs (S509) of McjD, and they all formed effective cross-links along the transport cycle of McjD. To further study the occluded conformation of McjD, we carried out extensive MD simulations, starting from the AMPPNP-bound state, after replacing AMPPNP with either ADP or ATP molecules. These simulations, performed in the absence of the MccJ25 substrate, reveal no major opening of the transmembrane cavity at the periplasmic end of the TMDs (between the two TM1 helices based on the L53-L53 pair, and between TM1 and TM6 of each monomer, based on the D51-S282 pair), which would allow the transporter to switch towards an outward-facing state. The largest opening detected between the L53-L53 pair is of ca. 14 Å , significantly smaller than that of an outward-facing state or the minimum opening of ca. 20 Å (measured as distance between the centre of mass of the two L53 residues) required for allowing MccJ25 to be released on the periplasmic side of the membrane, as we have shown in previous simulation studies (Gu et al, 2015). The occlusion of the cavity resulted to be stable also at its cytoplasmic side (A122), with no evidence of a significant separation between the intracellular loops, in agreement with the cross-linking experiments. We acknowledge, however, that longer MD simulations might be required to explore the possibility that the larger L53-L53 distances detected in two simulations, together with the larger A122-A122 distances mainly sampled in some of the ADPbound simulations, might trigger a conformational change towards an outward-facing state. Interestingly, an occluded state of the cavity was also sampled in previous MD simulations studies carried out starting from the outward-facing structure of Sav1866 in the presence of ADP (St-Pierre et al, 2012), or in the absence of nucleotide (Becker et al, 2010). Combined, our experimental and theoretical work provides strong evidence that McjD exists mostly in an occluded state within the membrane in the absence of the substrate MccJ25. These data are in striking contrast to the previously published type I ABC transporter crystal structures and biochemical data that showed the existence of wide-open inward and outward conformations (Dawson & Locher, 2006;Borbat et al, 2007;Ward et al, 2007). In order to identify the mechanism that results in an outward-open McjD, we performed PELDOR measurements on McjD reconstituted in bicelles with spin-labelled L52C mutants, at the periplasmic side. Under all tested conditions, L52R1 revealed a distance distribution that fits the occluded conformation observed in the crystal structures. It is commonly thought that homodimeric multi-drug ABC transporters adopt an outward-open conformation upon nucleotide binding (Dawson & Locher, 2006;Ward et al, 2007). In contrast, heterodimeric ABC exporters assume this outward-open conformation upon ATP hydrolysis (in the form of ADP-VO 4 ; Mishra et al, 2014). McjD is the first report of a homodimeric ABC exporter that does not form a stable wide-open state on the periplasmic side in the presence of nucleotides. Interestingly, we could also not observe an opening of the periplasmic side in the presence of both MccJ25 and ATP. Hence, we propose that opening of McjD is transient and not long-lived enough to be observed under our experimental conditions. Our transport data in proteoliposomes provide indirect evidence that McjD adopts an outward-open conformation; using the L53C mutant and in the presence of CuCl 2 , McjD was not capable of transporting Hoechst compared to the untreated protein. Our data suggest that McjD does not need to adopt a stable or long-lived wide-open conformation to release its substrate. Indeed, our molecular simulations have previously shown that McjD requires an opening of ca. 20 Å to export MccJ25 from the cavity that reverts back to the occluded conformation soon after substrate release (Gu et al, 2015). Previous structural and biophysical data did not explain the mechanism of ATP-induced dimerization of the NBDs since they were found relatively far apart. Our cross-linking data show that the NBDs are in very close proximity to each other within the membrane; binding of ATP would be sufficient to allow dimerization of the NBDs without the need of large conformational changes. Previous MD studies carried out on the outward-facing state of Sav1866, where the NBDs are engaged in a tight dimer as in the outward-occluded state of McjD, showed different results, highlighting (i) partial NBD dissociation in the presence of ADP and inorganic phosphate (Oliveira et al, 2011); (ii) how the pattern of interactions between domains changes after hydrolysis of just one ATP molecule (Gyimesi et al, 2011) or after ATP removal (Becker et al, 2010); and (iii) the stability of this particular conformation even in the presence of ADP (St-Pierre et al, 2012). Here, our MD simulations, carried out at much larger timescales, reveal no dissociation at the NBD interface, independently by the presence of ADP or ATP. Our studies support the experimental findings that the binding of two nucleotide molecules at the NBDs (either ADP or ATP) is enough to establish a tight dimer coupled with the occluded conformation of the transmembrane cavity.
Finally, using PEGylation of the cavity and competition with MccJ25, we showed that MccJ25 is entering the cavity by the cytoplasmic side of the inward-open McjD rather than a lateral membrane opening as proposed for other ABC transporters (Aller et al, 2009). This is the first evidence that McjD adopts an inwardopen conformation for peptide binding.

Mechanism for antibacterial peptide export
ABC exporters alternate between inward-and outward-facing conformations to transport their substrates from the cytoplasm to the periplasm in an ATP-dependent manner (Beis, 2015). Structural and biochemical studies have shown that they adopt an inwardopen conformation in the absence of nucleotides (Ward et al, 2007). In the presence of nucleotides, AMPPNP or ADP-VO 4 , their NBDs dimerize and adopt an outward-open conformation. In both the inward-and outward-open states, ABC exporters display domain intertwining. The transition between the two states is not well understood. In our study, the nucleotide-bound occluded McjD did not display any domain intertwining and it was closed at both sides of the membrane that could explain this transition. McjD adopts conformations that are distinct from other structures and its mechanism cannot be explained with the current mechanistic model. Therefore, taken together the biochemical, structural and simulation data we obtained allows us to propose a new working The EMBO Journal hypothesis for the export of the antibacterial peptide MccJ25 (Fig 9), distinct from the mechanism established for other ABC exporters (Dawson & Locher, 2006;Ward et al, 2007 The presence of the occluded conformations along the transport cycle probably acts as (i) a shielding mechanism for preventing peptide re-uptake by McjD and (ii) a mechanism to prime McjD for MccJ25 binding, during its biosynthesis, without building up toxic levels of the peptide (Choudhury et al, 2014). The structures of the peptidase-containing ATP-binding cassette transporter (PCAT) PCAT1 from Clostridium thermocellum (Lin et al, 2015) and the lipid-linked oligosaccharide flippase PglK from Campylobacter jejuni (Perez et al, 2015) have also been determined in a nucleotide-bound outward-occluded conformation and without domain intertwining similar to McjD. Therefore, the mechanism we propose for McjD is probably relevant to other ABC transporters that are dedicated to specific substrate export, whose cavity is "immediately shielded" upon substrate release. In contrast, the structure of the ABC transporter associated with antigen processing (TAP) does not display an occluded conformation (Oldham et al, 2016) and it has been shown that it is trans-inhibited by external peptide concentration, thus providing a mechanism to prevent the build-up of high The EMBO Journal McjD structure and mechanism Kiran Bountra et al levels of peptides in the ER that can induce stress (Grossmann et al, 2014). The distinct McjD conformations and mechanism described here can probably explain how ABC transporters, such as MsbA and Sav1866, have acquired multi-drug resistance in bacteria. We propose that shielding of the cavity is important to provide substrate specificity.

McjD purification and crystallization
McjD was purified in 0.03% dodecyl-maltopyranoside (DDM) as previously described without modifications (Choudhury et al, 2014). McjD-ADP-VO 4 was prepared by incubating 15 mg/ml McjD (20 mM Tris pH 7.8, 150 mM NaCl, 0.03% DDM) with 2 mM ATP, 2 mM sodium orthovanadate and 5 mM MgCl 2 at room temperature for 1 h. Crystals were grown at 293 K using the vapour diffusion method by mixing protein and reservoir solution at 1:1. Crystals were grown from a precipitant solution containing 10% PEG 4000, 0.1 M ammonium sulphate, 100 mM HEPES pH 7.5 and 22% glycerol. Crystals appeared overnight and they reached maximum size after 4 days. The crystals were directly frozen into liquid nitrogen, and diffraction screening and data collection were performed at Diamond Light Source synchrotron. These crystals diffracted X-rays to a maximum resolution of 2.9 Å .
McjD-apo crystals, 20 mg/ml McjD (20 mM Tris pH 7.8, 150 mM NaCl, 0.03% DDM), were grown from a precipitant solution containing 5 mM MgCl 2 , 24% PEG400 and 100 mM glycine pH 9.0. Crystals appeared after a week and they reached maximum size after 2 weeks. Prior to freezing crystals, the cover slide was moved to a fresh reservoir solution containing 5 mM MgCl 2 , 28% PEG400 and 100 mM glycine pH 9.0 and incubated overnight. Crystals were directly frozen into liquid nitrogen. Diffraction screening and data collection were performed at Diamond Light Source synchrotron. These crystals diffracted X-rays to a maximum resolution of 4.5 Å .

Data collection
McjD-ADP-VO 4 diffraction data were collected at~60 K on I23, the long wavelength Beamline at the Diamond Light Source (Wagner et al, 2016), close to the vanadium K-edge, k = 2.26 Å . In total, 360°of data was recorded using the inverse-beam method (20°w edges). Images from a Pilatus 12 M detector were processed using autoPROC (Vonrhein et al, 2011) and STARANISO server (http://sta raniso.globalphasing.org/cgi-bin/staraniso.cgi). Further processing was performed using the CCP4 suite (Collaborative Computational Project, 1994). Due to severe anisotropy, the resolution limit was restricted to 3.4 Å in the refinement stage. The space group was determined to be C2 with two copies of McjD in the asymmetric unit.
McjD-apo diffraction data to 4.7 Å were collected on I04-1 at Diamond Light Source at a wavelength of 0.92 Å using an ADSC Q315 detector and processed using xia2 (Winter, 2010). Further processing was performed using the CCP4 suite (Collaborative Computational Project N, 1994). The resolution of the data and anisotropy analysis were evaluated by half-dataset correlation coefficient in Aimless (cut-off < 0.5; Evans, 2006). The space group was determined to be P4 2 22 with one copy of McjD in the asymmetric unit.
The data collection statistics for both the McjD-ADP-VO 4 and apo crystals are summarized in Table 1.

Structure solution and refinement
The McjD-ADP-VO 4 structure was determined by molecular replacement in Phaser (McCoy et al, 2007) using our previously published McjD structure (PDB ID: 4PL0; Choudhury et al, 2014) as search model. Initial refinement to 3.4 Å was carried out in REFMAC5 (Murshudov et al, 1997) and at later stages in Buster (Blanc et al, 2004). The structure was refined with restraints against the high-resolution structure 4PL0. After rigid body and restrained refinement, extra electron density corresponding to two ADP-VO 4 molecules and two MgCl 2 ions was identified ( Fig EV1); these molecules were built and refined. To verify the presence of ADP-VO 4 , anomalous difference maps were calculated in CCP4 using FFT. Strong electron density could be 1.23 0.7 a "spherical" looks at all data within a specific, spherical resolution range/bin, so this is the usual, well-known way of looking at data as a function of resolution. "Ellipsoidal" additionally requires that a data point be within the fitted ellipsoid in order to be considered.
observed up to 10 r at the vanadium position at both NBDs. Due to the medium resolution of the data, placement of the VO 4 was guided by the anomalous difference maps. The final model has an R work of 25.7% and an R free of 26.0%. The McjD-ADP-VO 4 structure has 93.3% of the residues in the favoured Ramachandran region and has eight outliers as calculated by MolProbity (Chen et al, 2010). Attempts to determine the structure of McjD-apo using the fulllength McjD monomer failed in both Phaser and MOLREP (Vagin & Teplyakov, 1997); therefore, the structure was split in two search domains, TMD and NBD. Phaser under Phenix (Adams et al, 2002) was able to locate the TMD and NBD with good packing scores. Similar solution could also be obtained using MOLREP in CCP4. Refinement to 4.7 Å was carried out in Buster (Blanc et al, 2004). Good electron density could be observed for both the TMD and NBDs including density for aromatic side chains (Fig 3). Continuous electron density could also be observed for the linker that connects the TMD with the NBD. Due to the low resolution, we calculated composite omit maps in CCP4 to remove any model bias from molecular replacement (Fig 3). Overall B-factors were refined with the final model having an R work of 31.4% and an R free of 33.4%. The McjD-apo structure has 89.5% of the residues in the favoured Ramachandran region and has seven outliers as calculated by MolProbity (Chen et al, 2010). All model building was performed in Coot (Emsley & Cowtan, 2004).
The refinement statistics for both structures are summarized in Table 1. The coordinates and structure factors of McjD-ADP-VO 4 and McjD-apo have been deposited to the Protein Data Bank with PDB ID codes 5OFR and 5OFP, respectively.

MccJ25 purification
MccJ25 was produced from cultures of E. coli K12 MC4100 harbouring the plasmid pTUC202 and purified as previously described (Zirah et al, 2011).

Site-directed mutagenesis
McjD mutants were generated using the QuickChange Site-Directed Mutagenesis Kit (Agilent Technologies). DNA sequencing of the resulting plasmids confirmed the presence of the mutants.

Cysteine cross-linking
Cysteine cross-linking experiments in ISOVs were performed as previously described without modifications (Choudhury et al, 2014), with the exception of S509C mutant that was pre-treated with 1 mM DTT for 30 min at room temperature to reduce preformed cysteine cross-links. The effect of the MccJ25 to induce cysteine cross-links of A122C was determined by adding 1 mM MccJ25 to A122C ISOVs for 10 min prior to CuCl 2 addition. Formation of cysteine cross-links was analysed by Western blotting. The gels were imaged using an ImageQuant LAS4000 (GE Healthcare). Densitometry analysis of the gel bands intensities was performed using the ImageQuant TL software (GE Healthcare); to quantify the extent of MccJ25-stimulated cross-linking of A122C, band intensities (automatic peak detection mode) corresponding to CLD formation were estimated for each gel lane.

Hoechst 33342 transport in proteoliposomes
McjD and McjD-L53C were reconstituted in proteoliposomes using the rapid dilution protocol. In brief, E. coli polar lipids were made to 20 mg/ml in water and were bath-sonicated for 15 min until the solution became less hazy. 100 ll of 0.1 mg/ml McjD or McjD-L53C was mixed with 20 ll lipid stock on ice for 10 min. The solution was further diluted into 4 ml of liposome buffer (50 mM Tris pH 8 and 50 mM KCl) and was incubated for further 5 min on ice. Proteoliposomes were pelleted by centrifugation at 45,000 g for 30 min. The supernatant was carefully removed and the pellet was resuspended in 100 ll liposome buffer. The proteoliposomes were immediately used in experiments. Uptake measurements were performed as before with minor modifications (Choudhury et al, 2014); the amount of protein reconstituted in liposomes was 0.4 mg/ml and the final Hoechst concentration was 0.1 lM. Data were recorded using a Cary Eclipse Fluorescence Spectrophotometer (Agilent Technologies). The mean value of two independent experiments was calculated and plotted in GraphPad Prism.

PEGylation
Inside-out vesicles of each cavity mutant were prepared as for the cysteine cross-linking experiments. The concentration of the ISOVs was adjusted to a total protein concentration of 10 lg/ll in 50 mM HEPES pH 7.5 and 150 mM NaCl. Where appropriate, ISOVs were incubated with 3 mM nucleotides and MgCl 2 at 37°C for 20 min. 1 mM mPEG10k was then added to the relevant conditions and incubated at room temperature for 1 h. PEGylation reactions were performed in triplicate. Competition assays were performed in duplicate at a final concentration of 1 mM MccJ25 (in DMSO) for 1 h with an equimolar ratio of mPEG10k. All reactions were quenched upon the addition of 2 mM N-ethylmaleimide and ran on nonreducing NuPAGE Bis-Tris gels. PEGylation results were analysed by Western blotting. The gels were imaged using an ImageQuant LAS4000 (GE Healthcare). Densitometry analysis of the gel band intensities was performed using the ImageQuant TL software (GE Healthcare); the degree of PEGylated species formed was quantified for the competition assays using the automatic peak detection setting.

Labelling of McjD mutants for PELDOR measurements
PELDOR McjD mutants were purified as the wild-type McjD with minor modifications to account for the labelling. In brief, McjD mutants in DDM after the reverse chromatography step (Choudhury et al, 2014) were concentrated to 500 ll using a 100-kDa MW cut-off concentrator. The protein was incubated with 10-fold molar excess of (1-oxyl-2,2,5,5-tetramethyl-Δ3-pyrroline-3-methyl) methanethiosulfonate dye (Toronto Research Chemicals) dissolved in DMSO. The solution was placed on an elliptical roller in dark for 3 hours. Excess label was removed by passing the protein through a Superdex S200 10/300 GL (GE Healthcare) equilibrated in 0.06% DDM, 300 mM NaCl and 100 mM TES pH 7.5, made in D 2 O. Fractions were monitored using an Ä KTA FPLC (GE Healthcare) at an absorbance of 280 nm. McjD fractions were concentrated using a 100-kDa MW cutoff concentrator, until a protein concentration of 18 mg/ml was reached. The sample was flash-frozen in liquid N 2 . cw-X-band EPR spectra of the samples were recorded and are shown in Fig EV5. The spectra show the typical broadening of the nitroxide spectrum that is due to the immobilization of the nitroxide on the protein. Note that small amounts of free label are visible in the spectra.

ATPase activity of purified McjD and mutants
The ATPase activity of the wild-type McjD-, McjD-L53C-and PELDOR-labelled McjD mutants was measured as previously described (Choudhury et al, 2014).

PELDOR measurements
For PELDOR measurements, the spin-labelled McjD samples were reconstituted into bicelles as previously described (Ward et al, 2014). If needed, the samples were supplemented with nucleotides or substrate and incubated at room temperature. The samples were supplemented with 50% deuterated ethylene glycol and transferred to a 3-mm quartz Q-band EPR tube and flash-cooled in liquid nitrogen. The PELDOR time traces were recorded on a Bruker ELEXSYS E580 pulsed Q-band EPR spectrometer, with an ER 5106QT-2 Qband resonator. The instrument was equipped with a continuous flow helium cryostat (CF935) and temperature control system (ITC 502), both from Oxford instruments. The second microwave frequency was coupled into the microwave bridge using a commercially available setup from Bruker. All pulses were amplified via a 150-W pulsed travelling wave tube (TWT) amplifier. PELDOR experiments were performed with the pulse sequence p/2(m A )-s 1 -p(m A )-(s 1 + t)-p(m B )-(s 2 Àt)-p(m A )-s 2 -echo. The detection pulses (m A ) were set to 12 ns for the p/2 and 24 ns for the p pulses and applied at a frequency 80 MHz lower than the resonance frequency of the resonator. The pulse amplitudes were chosen to optimize the refocused echo. The p/2-pulse was phase-cycled to eliminate receiver offsets. The pump pulse (m B ) was set at the resonance frequency of the resonator and its optimal length (typically 16 ns) was determined using a transient nutation experiment for each sample. The field was adjusted such that the pump pulse is applied to the maximum of the nitroxide spectrum. The pulse amplitude was optimized to maximize the inversion of a Hahn-echo at the pump frequency. All PELDOR spectra were recorded at 50 K with an experiment repetition time of 1 ms, a video amplifier bandwidth of 20 MHz and an amplifier gain of 42 dB. s 1 was set to 260 ns and the maximum of s 2 was set to values ranging from 2 to 6 ls. Deuterium modulation was suppressed by the addition of eight spectra of variable s 1 with a Ds 1 of 16 ns. The obtained time traces were divided by a monoexponential decay to eliminate intermolecular contributions and renormalized. Distance distributions were obtained from the background-corrected data by using the Tikhonov regularization as implemented in DeerAnalysis2016 developed by Gunnar Jeschke (Jeschke et al, 2006). The influence of different starting points for the background fitting was analysed with the evaluation feature of DeerAnalysis. The PyMOL (www.pymol.org) plugin mtsslWizard (Hagelueken et al, 2012) was used to predict distance distributions.

Molecular dynamics simulations
MD simulations were performed on the McjD-AMPPNP structure, obtained at a resolution of 2.7 Å (PDB ID: 4PL0; Choudhury et al, 2014). Two sets of simulations were built by replacing the AMPPNP molecule with either ATP or ADP, while the position of the Mg 2+ ions was kept as in the crystal structure. H537, the Switch motif, was set protonated for the ATP-bound simulations, and neutral for the ADP-bound ones. The protein in complex with either ATP or ADP was embedded in a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) bilayer using g_membed (Wolf et al, 2010), with a total of 500 lipids. The simulation box, including water and ions to neutralize the system, was 13 × 13 × 16 nm. Energy minimization, equilibration protocols and production runs were carried out with GROMACS 4.5.5 (Berendsen et al, 1995;Hess et al, 2008), using the ff54a7 version of the GROMOS 96 force field Schmid et al, 2011).
The ATP-bound system was energy-minimized with (1) position restraints on the protein heavy atoms (force constant of 1,000), ADP/ ATP molecules and Mg 2+ ions (force constant of 10,000 kJmol À1 nm À2 ) first; (2) then, with position restrains only on the backbone atoms (force constant of 10,000 kJmol À1 nm À2 ). For the ADP-bound system, we performed an additional intermediate step between (1) and (2), where we applied position restraints on the protein backbone atoms, ADP/ ATP molecules and Mg 2+ ions (force constant of 10,000 kJmol À1 nm À2 ), before the final energy minimization.
After energy minimization, the ATP-bound system was initially equilibrated for 10 ns, during which the protein heavy atoms were position-restrained with a force constant of 1,000 kJmol À1 nm À2 , while ATP/ADP molecules and Mg 2+ ions were restrained with a 10,000 kJmol À1 nm À2 force constant. 5 additional ns was performed with position restraints on the protein backbone atoms (force constant of 10,000 kJmol À1 nm À2 ). For the ADP-bound system, this step was preceded with 5 ns carried out with position restraints applied on the protein backbone atoms and on the ADP molecules and Mg 2+ ions as well (force constant of 10,000 kJmol À1 nm À2 ). A time step of 2 fs, the Berendsen barostat (Berendsen et al, 1984; with a relaxation time constant of 2 ps and a semi-isotropic reference pressure of 1 bar) and the V-rescale thermostat (Bussi et al, 2007; with a time constant of 0.1 ps and a reference temperature of 313.15 K) were used for the equilibration steps.
The two ADP-and ATP-bound McjD structures resulting from the equilibration protocol were used as starting structures for production runs. For the ADP-bound system, 10 independent runs, of 600 ns each, were generated. For the ATP-bound system, 10 independent runs were also performed, eight of 600 ns each and two more of 580 and 440 ns, respectively. A time step of 2 fs, the Parrinello-Rahman barostat (Parrinello & Rahman, 1981; with a relaxation time constant of 6 ps and a semi-isotropic reference pressure of 1 bar for coupling) and the V-rescale thermostat (Bussi et al, 2007; with a relaxation time constant of 2.0 ps and a reference temperature of 313.15 K) were used for the production runs.
Distances between the selected residues described in the Results were calculated as Ca distances. The probability density of the distances shown in the Results section is the average probability density obtained from the 10 simulations of each set (ADP-or ATPbound). Plots were generated with Matplotlib (Hunter, 2007), while protein figures, unless otherwise specified, were made with The PyMOL Molecular Graphics System, version 1.7 Schrödinger, LLC (DeLano, 2002).
Expanded View for this article is available online.