Structural and functional evidence that lipoprotein LpqN supports cell envelope biogenesis in Mycobacterium tuberculosis

The mycobacterial cell envelope is crucial to host–pathogen interactions as a barrier against antibiotics and the host immune response. In addition, cell envelope lipids are mycobacterial virulence factors. Cell envelope lipid biosynthesis is the target of a number of frontline tuberculosis treatments and has been the focus of much research. However, the transport mechanisms by which these lipids reach the mycomembrane remain poorly understood. Many envelope lipids are exported from the cytoplasm to the periplasmic space via the mycobacterial membrane protein large (MmpL) family of proteins. In other bacteria, lipoproteins can contribute to outer membrane biogenesis through direct binding of substrates and/or protein–protein associations with extracytoplasmic biosynthetic enzymes. In this report, we investigate whether the lipoprotein LpqN plays a similar role in mycobacteria. Using a genetic two-hybrid approach, we demonstrate that LpqN interacts with periplasmic loop domains of the MmpL3 and MmpL11 transporters that export mycolic acid–containing cell envelope lipids. We observe that LpqN also interacts with secreted cell envelope biosynthetic enzymes such as Ag85A via pulldown assays. The X-ray crystal structures of LpqN and LpqN bound to dodecyl-trehalose suggest that LpqN directly binds trehalose monomycolate, the MmpL3 and Ag85A substrate. Finally, we observe altered lipid profiles of the ΔlpqN mutant during biofilm maturation, pointing toward a possible physiological role for the protein. The results of this study suggest that LpqN may act as a membrane fusion protein, connecting MmpL transporters with periplasmic proteins, and provide general insight into the role of lipoproteins in Mycobacterium tuberculosis cell envelope biogenesis.

Mycobacterium tuberculosis (Mtb) 2 is the causative agent of the human disease tuberculosis (TB). TB is one of the most devastating human diseases, and it remains a major public health concern. In 2017, there were ϳ10 million new cases and 1.3 million deaths caused by TB (1). The cell envelope of Mtb is a waxy, lipid-rich structure that acts as a robust physical barrier to host antimicrobial defenses and antibiotics. This intrinsic resistance is largely conferred by mycolic acids, extremely long chain (C 60 -90 ) ␣-alkyl-␤-hydroxy fatty acids that are major constituents of the mycobacterial cell envelope (2). Mycolic acids and their derivatives form an outer layer analogous to the outer membrane of Gram-negative bacteria, termed the mycomembrane. The mycomembrane consists of an inner leaflet of mycolylated arabinogalactan anchored to peptidoglycan (the mAGP complex) and an outer leaflet of predominantly mycolylated free lipids (3,4). In addition to acting as a passive physical barrier, certain cell envelope components can directly interfere with host immune responses (5,6). Therefore, the Mtb cell envelope plays a crucial role at the host-pathogen interface to promote bacterial survival.
The biogenesis of the mycobacterial cell envelope is complex and incompletely described. One class of proteins involved in transport of cell envelope components is the mycobacterial membrane protein large (MmpL) family. The MmpLs are polytopic membrane proteins that mediate transport of large hydrophobic substrates from the cytoplasm of the bacterium to the periplasmic space, where they are subsequently incorporated into the cell envelope (7)(8)(9). The MmpLs are considered members of the resistance-nodulation-cell division (RND) superfamily of membrane transporters that typically couple the proton motive force to export of their substrates (10, 11). A subset of MmpLs (MmpL1/2/4/5) are accompanied in the genome by genes encoding mycobacterial membrane protein small (MmpS) proteins (12) that are predicted to contain one or two transmembrane segments and are likely involved in substrate transport by their cognate MmpL protein.
The substrates of a number of MmpL transporters have been identified. Of particular interest are those MmpL proteins that transport mycolic acid-containing lipids. Most importantly, MmpL3 (Rv0206c) transports trehalose monomycolate (TMM) into the periplasmic space (13). There, mycolic acid transfer catalyzed by members of the antigen 85 (Ag85) enzyme complex generates the free lipid trehalose dimycolate (TDM) and arabinogalactan-anchored mycolic acids, which together form the mycomembrane (13)(14)(15). MmpL3 is therefore an essential Mtb protein, because of the central role of mycolic acids in mycobacterial physiology (16,17). Another important MmpL transporter is MmpL11 (Rv0202c), which we showed exports species-specific mycolic acid-containing storage lipids such as the mycolate wax ester and long-chain triacylglycerols in Mtb and meromycolyl diacylglycerol in Mycobacterium smegmatis (18,19). These lipids are important for Mtb persistence, because MmpL11-deficient Mtb exhibit attenuated survival during in vitro and in vivo models of infection and in an in vitro model of dormancy (10, 19).
Many of the insights into MmpL transporter function have been gleaned from comparison with other protein members of the larger RND transporter superfamily. RND superfamily proteins are primarily associated with antibiotic resistance and stress response in Gram-negative bacteria (reviewed in Ref. 20). They typically function in concert with a membrane fusion protein (MFP) and an outer membrane factor to form a tripartite export apparatus that extrudes substrates through the periplasm directly to the extracellular environment (21)(22)(23)(24)(25). Although Mtb exhibits pseudo-Gram-negative membrane organization, it remains to be determined whether the MmpLs interact with periplasmic and outer membrane proteins to form a similar tripartite pump that would enable efficient translocation and proper localization of their substrates. Some secreted lipoproteins direct the proper localization of virulence-associated lipids within the cell envelope, suggesting that these proteins may be functioning as MFPs in Mtb. Notably, translocation of phthiocerol dimycocerosate (PDIM) to the mycomembrane depends on the lipoprotein LppX, presumably through direct binding of PDIM by LppX (26). Similarly, the lipoprotein LprG binds both the virulence-associated glycolipid lipoarabinomannan and triacylglycerides and is required for translocation of these substrates to the cell envelope (27). Furthermore, LprG interacts with the Ag85A mycolyltransferase, implicating this lipoprotein in the transmycolylation processes that generate the mycomembrane (28). The MmpS proteins are also candidate MFPs, given that MmpS4/5 interact with their cognate MmpL4/5 partners and are required for transport of their siderophore substrates (29). However, only a minority of MmpL transporters have cognate MmpS proteins (MmpS1/2/4/5), and MmpL-MmpS protein-protein interactions have not been demonstrated for MmpS1/2. This raises the possibility that there are unidentified Mtb proteins that function as quasi-MFPs to assist in delivering MmpL-transported lipids through the periplasm to their cell envelope destinations.
We sought to identify proteins that might function as MFPs with MmpL3 and MmpL11. Using genetic methods, we identified the secreted lipoprotein LpqN (Rv0583c) as a periplasmic interacting partner of both MmpL3 and MmpL11. Biochemical and genetic studies suggested that LpqN may also associate with members of the Ag85 enzyme complex and plays a role in cell envelope lipid changes during biofilm maturation. These results suggest that LpqN may function as an MFP of MmpL3/ 11. This model is supported by the crystal structures of apo-and lipid-bound LpqN reported herein.

LpqN interacts with D2 loops of mycolate lipid transporters MmpL3 and MmpL11
MmpL3 and MmpL11 are structurally similar and categorized into the hydrophobe/amphiphile efflux 3 (HAE3) subfamily of the larger RND transporter superfamily (29,30). They possess two periplasmic loop domains, termed D1 and D2, as well as a cytosolic C-terminal domain (31). The D1, D2, and C-terminal domains are therefore the most likely regions of interaction between MmpL11 and biosynthetic enzymes or other transport machinery. To identify periplasmic proteins that interact with MmpL11 TB , we used the mycobacterial protein fragment complementation (M-PFC) system (32). In this two-hybrid system, the "bait" and "prey" are expressed independently as fusion proteins with two fragments of the murine dihydrofolate reductase (DHFR [F1,2] and DHFR [F3]). Functional reconstitution by two interacting mycobacterial proteins when expressed in M. smegmatis results in trimethoprim resistance. Integrative plasmids were constructed containing D1 and D2 domains in-frame with the DHFR [F3] fragment and then transformed into M. smegmatis to generate a bait strain. An H37Rv Mtb prey library of over 10 6 independent clones was generated in pUAB300. Clones encoding putative proteins that interact with MmpL11 TB were obtained by transforming the library constructs into the M. smegmatis bait strain and selecting for trimethoprim resistance. Ultimately, no positive clones were identified using the D1 domain. Using the D2 construct, we isolated two independent clones that contained a DHFR [F1,2]-LpqN fusion. LpqN was the only positive clone identified using the MmpL11 D2 domain as bait. This suggests that LpqN (Rv0583c) interacts with the periplasmic D2 domain of MmpL11 TB . MmpL11 and MmpL3 are topologically similar conserved transporters of mycolic acid-containing lipids. Therefore, we theorized that LpqN might interact with the periplasmic domains of MmpL3 as well. To address this and validate our initial M-PFC screen, we assessed the interactions between the MmpL3 and MmpL11 periplasmic D1 and D2 loops with LpqN using a directly cloned (nonlibrary) LpqN prey construct, which contained the soluble domain of LpqN fused to DHFR [F1,2]. We found that the D2, but not the D1, loops of both MmpL3 and MmpL11 conferred trimethoprim resistance when co-expressed with LpqN (Table 1). These results suggest that LpqN interacts with the D2 periplasmic regions of both MmpL3 and MmpL11.

Characterization of Mtb LpqN
To verify the interaction between LpqN and MmpL11 in vivo, we heterologously expressed affinity-tagged Mtb protein in M. smegmatis. Although both proteins were successfully expressed, we were unfortunately unable to co-purify LpqN with MmpL11 in the absence or presence of cross-linking (Fig. S1).

LpqN interacts with cell envelope lipid biosynthetic enzymes
We hypothesized that LpqN might act as an adaptor protein in the periplasm to facilitate interactions between MmpL proteins and periplasmic proteins. To identify secreted interacting partners of LpqN, we performed a pulldown purification of affinity-tagged LpqN incubated with Mtb culture filtrate proteins (CFPs) in the presence/absence of formaldehyde as a protein cross-linking reagent. Interacting proteins were subsequently identified via MS and categorized as weak or strong LpqN-interacting partners based on whether or not protein crosslinking was required for co-purification ( Table 2). As a control, we analyzed the background binding of CFP to the metal affinity resin. Intriguingly, secreted cell envelope biosynthetic enzymes, such as the mycolyltransferase Ag85A and the mycocerosic acid synthase Mas, were enriched in the LpqN co-purified samples relative to background. Ag85B also copurified with LpqN, although to a lesser degree. Ag85 and Mas enzymes are responsible for the biosynthesis of the cell envelope lipids TDM and PDIM, respectively. These data suggest that LpqN interacts with periplasmic cell envelope biosynthetic enzymes.
The possible interaction between Ag85A and LpqN was interesting, given the key role that the Ag85 complex plays in the biogenesis of the mycobacterial cell envelope. Furthermore, the primary substrate of the Ag85 enzymes is TMM, which is exported by MmpL3. Attempts to demonstrate a direct interaction between LpqN and Ag85A using a heterologous dualexpression M. smegmatis co-immunopurification strategy were unsuccessful (Fig. S2). However, there is precedent for interactions between lipoproteins and members of the Ag85 complex, as evidenced by the recently reported LprG-Ag85A interaction (28). LprG (Rv1411c) binds phosphatidylinositolcontaining glycolipids rather than mycolic acids (33). Nevertheless, LprG appears to contribute to mycolylation of cell wall analogues (28). We performed head-to-head M-PFC assays to determine whether the MmpL3 or MmpL11 D2 domains are capable of interacting with other LpqN family members (LpqT, Mtc28) or LprG. LpqT and Mtc28 also interact with the MmpL11 D2 domain, but not the MmpL3 D2 domain, via M-PFC (Table S1). The structurally distinct LprG protein also interacted with the MmpL11, but not MmpL3, D2 domain via M-PFC. These results suggest the possibility that multiple lipoproteins act as adaptors for the MmpL transporters.

LpqN contributes to Mtb biofilm lipid composition
MmpL3 transports TMM, which is essential for replication, whereas MmpL11 transports cell envelope lipids that are important for biofilm formation. To investigate a possible role for LpqN in mycobacterial cell envelope biogenesis, we generated an lpqN-deletion mutant in H37Rv Mtb (⌬lpqN) via allelic exchange (Fig. S3, A and B). There was not a growth defect in either 7H9 or Sauton's medium (Fig. S3, C and D). We then characterized Mtb WT, ⌬lpqN, and complemented strain cell envelope lipids from bacteria grown planktonically and in biofilms by TLC. We did not observe any differences in TMM and TDM levels between the WT, mutant, or complemented mutant (Mtb ⌬lpqN::lpqN) lipid profiles using the chloroform:methanol:ammonium hydroxide (80:20:2) solvent system (Fig. 1, A and B). We also analyzed the mycolic acid methyl esters of mAGP, which did not differ significantly between strains (Fig. 1C). Consistent with these results, we also did not see significant differences in the LC/MS profiles among the lipid extracts, in particular, of TMM and TDM (Figs. S4 and S5). This was not surprising because the ⌬lpqN mutant did not possess a growth defect.
Our analyses of Mtb biofilms over time indicate that the lipid profiles shift as the biofilm matures. Although there were no differences between WT and ⌬lpqN mutant lipids at 3 weeks ( Fig. 1D), we noted the relative accumulation of an apolar lipid in the lpqN mutant samples that ran with a similar Rf to the wax ester standard in 4 weeks (Fig. 1E, asterisk). We speculated that the apolar lipid (Fig. 1E, asterisk) in the mutant was a wax ester. However, mass spectrometric analysis of the extract of this apolar TLC spot did not yield mass spectra that directly identified any known mycobacterial lipids. These results countered our assumption that the spot is wax ester and are consistent with HPLC/MS results that showed no difference in wax ester between WT and LpqN (Fig. S4E). The ⌬lpqN mutant biofilm also appeared to accumulate a less-apolar band that was similarly recalcitrant to identification ( Fig. 1E, asterisks). Together, these biofilm lipid profiles suggest that LpqN may contribute to cell envelope lipid turnover or remodeling during biofilm development.

The LpqN crystal structure suggests that LpqN may directly bind lipids
To gain additional insight into LpqN function, the crystal structure of apo-LpqN was determined at a resolution of 1.65 Å using single anomalous dispersion (Table 3 and Fig. 2). Because LpqN might play a role in cell envelope biogenesis, we also determined structures of LpqN in complexes with dodecyl trehalose (6LT) and trehalose 6-decanoate (T6D) to resolutions of 1.37 and 1.74 Å, respectively. We resolved these two ligandbound LpqN structures by using molecular replacement and utilizing the apo-LpqN structure as a search model (Table 3 and Fig. 2). Both 6LT and T6D are water-soluble compounds that are structurally similar to the insoluble lipid molecule TMM in that they contain a two-ring system of the trehalose moiety. Therefore, 6LT and T6D are ideal compounds to mimic TMM

Insert in pUAB200
Insert in pUAB300 LpqN 50 Superimposition of the structures of apo-LpqN and LpqN-6LT results in an overall root-mean-square deviation (RMSD) of 0.8 Å, suggesting that these two structures are similar in conformation. However, detailed inspection indicates that the two structures are quite different. The main difference between the two conformations is the shift in location of ␣3 and the rearrangement of the secondary structures of several random loops located at the 6LT-binding site. These loops, including residues 93-96, 158 -160, and 167-170, convert into ␤ strands after binding 6LT. These changes result in a larger internal cavity for LpqN-6LT compared with that of apo-LpqN. In addition, a specific deep binding pocket, which was not observed in apo-LpqN, was found in LpqN-6LT to coordinate with the binding of the carbon chain of the 6LT ligand. Because the secondary structure of LpqN-6LT is very different, we reassigned its structural features to facilitate the description of this ligand-bound protein. The secondary structure assignments of 6LT-bound LpqN from the N to C termini are: ␣1 In the structure of LpqN-6LT, the 6LT molecule is bound within the cavity surrounded by ␤4, ␤5, ␤7, and ␣3. The dodecyl tail of 6LT is completely buried inside the LpqN protein, leaving its two sugar rings of the trehalose head group partially exposed to solvent. Within 6 Å of the dodecyl group of 6LT, 12 amino acids contact and secure the hydrocarbon tail (Fig. 3). Although the trehalose moiety of 6LT only partially contacts the LpqN protein, the binding of this two-ring head group is extensive. Ser-173 of LpqN forms a hydrogen bond with the trehalose head group to anchor this ligand. Gly-153 also donates its backbone oxygen to make another hydrogen bond with the trehalose moiety. Ser-154 and Gly-171 also participate in binding this head group. In addition, Gln-152 is 3.7 Å from the trehalose head group and interacts with the ligand via electrostatic interaction.

Characterization of Mtb LpqN
The structure of LpqN-T6D resembles that of LpqN-6LT. Superimposition of these two structures leads to an RMSD of  Like the binding of 6LT, the decanoate tail of T6D is completely buried in the binding cavity formed by LpqN. Within 6 Å of this hydrocarbon chain, at least eight amino acids are actively involved in anchoring the bound T6D. The trehalose head group of T6D is partially exposed to solvent and is surrounded by six amino acids. Ser-173 also participates by forming two hydrogen bonds with the sugar-ring system of T6D to secure the ligand (Fig. S4). The crystal structures of LpqN indeed sup-

Characterization of Mtb LpqN
port the idea that this lipoprotein is capable of binding cell envelope lipids, such as TMM.

Discussion
Using genetic methods, we demonstrated that LpqN interacts with the periplasmic D2 loop domains of the mycolate transporters MmpL3 and MmpL11. As such, LpqN may act as an MFP to facilitate substrate transport by MmpL3 and MmpL11. The outcome of this interaction is likely proper maturation of lipid substrates and/or the correct localization of these lipids within the cell envelope. We hypothesize that LpqN accomplishes these functions through direct binding of MmpLtransported lipids, as our structural data suggest.
Several observations suggest that there is substantial redundancy in this system. The cell envelope lipid profile of planktonic and 3-week biofilm-grown Mtb ⌬lpqN mutant did not differ from the WT, implying that the absence of LpqN does not affect the cell envelope lipid composition during these growth phases. Moreover, Mtb ⌬lpqN does not exhibit attenuated growth in planktonic culture (Fig. S1C). Because mycolic acid biosynthesis and MmpL3 are essential, we would expect that perturbing mycolic acid incorporation within the cell envelope would impact Mtb growth or viability. Given that lpqN-deficient Mtb grows at a normal rate and has a normal cell envelope lipid profile, LpqN is unlikely to be the sole facilitator of periplasmic mycolic acid transport in Mtb. In support of this, we note that the protein family to which LpqN belongs (PF10738) contains other secreted Mtb proteins such as the lipoprotein LpqT (Rv1016c) and the secreted proline-rich Mtc28 (Rv0040c) (31).
Lipid binding may be a shared feature of the LpqN protein family. The published structure of Mtc28 exhibits structural homology to eukaryotic lipid-binding proteins such as CERT and STARD13, as assessed by the DALI server (34 -37). Both LpqT and Mtc28, as well as the unrelated lipoprotein LprG (Rv1411c), interact with the MmpL11 D2 domain, but not the MmpL3 D2 domain, via M-PFC (Table S1). Whether LpqT and Mtc28 are capable of binding lipids is unknown, but LprG binds phosphatidylinositol-containing glycolipids and interacts with Ag85A (28,33). Thus, there are presumably multiple interconnected and redundant pathways of cell envelope biogenesis operating in the periplasm that are mediated by lipoproteins and other secreted proteins. Some periplasmic lipid mediators, like LpqN, appear capable of interacting with both MmpL3 and MmpL11. Other potential mediators, like LpqT, Mtc28, and LprG, appear to preferentially interact with MmpL11. Although MmpL3 and MmpL11 both export mycolic acidcontaining lipids, MmpL3 is essential and therefore more crucial to cell envelope physiology. Because lpqN-deficient Mtb is physiologically unimpaired, there are likely additional MmpL3interacting periplasmic lipid mediators that remain to be identified. The recent characterization of the MmpL3-associated protein TtfA in M. smegmatis supports the hypothesis that MmpLs function in concert with accessory proteins to transport their lipid substrates (38). LpqN and LpqN family proteins ͉R free was calculated as for R work but using 5% of the data that were excluded from the refinement calculation.

Characterization of Mtb LpqN
are therefore likely redundant periplasmic mediators of lipid transport. Our work raises a number of questions regarding the function of LpqN. An M-PFC screen performed in M. smegmatis initially identified LpqN as an interacting partner of MmpL11. We also used this system to show LpqN interaction with MmpL3. However, we were unable to demonstrate a direct biochemical interaction between LpqN and MmpL11 when heterologously co-expressed in M. smegmatis, suggesting a weak interaction (Fig. S1). We also cannot exclude the possibility that the LpqN-MmpL3/11 interaction is indirect and that some putative mycobacterial adaptor protein or molecule is required for LpqN to associate with these transporters. LpqN also appeared to preferentially interact with Ag85A when incubated with purified Mtb culture filtrate proteins. Similarly, we could not successfully co-purify these two proteins when heterologously co-expressed in M. smegmatis (Fig. S2). Again, this may reflect an indirect interaction with a species specific adaptor protein. The observation that Mtb ⌬lpqN biofilms appear to mature differently than WT Mtb biofilms potentially suggests temporal modulation of LpqN function and is the focus of ongoing investigation.
Recent work demonstrated that LpqN contributes to Mtb virulence in a manner apart from its role in cell envelope biogenesis. LpqN appears to interact with the host ubiquitin ligase Cbl that down-regulates antibacterial immunity and potentiates Mtb growth during ex vivo macrophage infection (39). Although this capacity of LpqN to function as a virulence factor is ostensibly unrelated to its apparent role in cell envelope biogenesis, this finding underscores the critical impor-

Characterization of Mtb LpqN
tance of the mycobacterial cell envelope and its components to pathogenesis.

Bacterial strains and growth conditions
The Mtb WT strain H37Rv was obtained from the ATCC. Mycobacterial strains were routinely maintained in Middlebrook 7H9 liquid medium (Difco) with 0.05% Tween 80 or on Middlebrook 7H10 agar (Difco), both supplemented with albumin dextrose salts (ADSs) containing 8.1 mg/ml NaCl, 50 mg/ml BSA, and 20 mg/ml dextrose. Glycerol was added to liquid 7H9 to a final concentration of 0.5%. Kanamycin (25 g/ml) and hygromycin (50 g/ml) were used to maintain bacterial selection when required.
The Mtb ⌬lpqN mutant was created via allelic exchange. Upstream and downstream regions of lpqN homology were amplified by PCR using ⌬lpqN 5Ј/3Ј Forward/Reverse primers (Table S2) and cloned to flank the hygromycin resistance gene in pYUB854-rpsL (40). The resulting linear hygromycin product flanked by regions of lpqN homology was amplified by PCR, purified, concentrated, and used to transform electrocompetent H37Rv/pJV53 (41). Transformants were selected on 7H10 agar containing ADS and hygromycin (75 g/ml). Individual colonies were transferred to 7H10 agar containing hygromycin (50 g/ml). Deletion of lpqN was confirmed via PCR using flanking primers. Loss of the pJV53 recombineering plasmid was accomplished through iterative culturing on 7H10 agar in the absence of kanamycin and confirmed by susceptibility to kanamycin (25 g/ml). For complementation, lpqN ϩ ϳ1000-bp upstream sequence was amplified via PCR with lpqN Ϫ865/ ϩ1637 primers and cloned into the integrative vector pMV306 (42). The resulting complementation plasmid was transformed into the Mtb ⌬lpqN mutant and transformants selected on 7H10 agar containing ADS and kanamycin (25 g/ml) and confirmed via PCR (Fig. S1, A and B).

Lipid isolation and analyses
Mtb lipids were harvested by collecting and centrifuging planktonic or biofilm cultures in 50-ml conical vials, washing once with PBS, and then moving to glass test tubes containing 3-mm glass beads. Surface lipids were extracted by shaking with hexanes (ϳ5 ml) for 2 min, followed by pelleting (1000 ϫ g/10 min). Supernatants were removed into fresh glass tubes and treated with HPLC-grade CHCl 3 :MeOH (2:1, v/v) prior to removal from biosafety containment. For total lipid extraction, bacterial pellets were autoclaved prior to removal from biosafety containment, then extracted with ϳ5 ml of CHCl 3 : MeOH (2:1, v/v) and 1 ml of HPLC-grade H 2 O by shaking and bath sonication. The organic layer was separated by centrifugation (1000 ϫ g/10 min), transferred to preweighed vial, and dried under inert N 2 gas. mAGP was isolated as previously described (43). Mycolic acid methyl esters (MAMEs) were obtained by incubating mAGP in 15% tetrabutylammonium hydroxide (overnight, 100°C) followed by iodomethane (4 h, room temperature) and extraction with dichloromethane. Extracted lipids were resuspended in CHCl 3 or dichloromethane (MAMEs) and loaded onto aluminum-backed silica plates (EMD Millipore) for TLC analysis. Solvent systems used were chloroform:methanol:ammonium hydroxide (80:20:2, v/v/v), hexanes:ethyl acetate (99:5, v/v, three developments), and toluene:acetone (99:1, v/v) (Fisher). Lipids were visualized via spraying with 12-molybdophosphoric acid (Alfa Aesar; 10%, w/v in EtOH) and charring. The extracted lipids were also subjected to LC/MS analysis using Agilent 6550A QTOF system in conjunction with Agilent 1290 HPLC. HPLC separation and MS data acquisition were performed as described previously (52).

Mycobacterial protein fragment complementation assay
The M-PFC screen was previously described and relies on the reconstitution of two domains of the DHFR protein, which confers resistance to trimethoprim (32). The MmpL3 and MmpL11 D1 and D2 periplasmic domains were amplified by PCR, and the products were subcloned into pGEM-T Easy (Promega). Once confirmed by sequencing, the pUAB200 integrative plasmids were constructed such that the D1 and D2 domains were in-frame with the DHFR [F3] fragment. The pUAB200 was then transformed into M. smegmatis mc 2 155 generating bait strains. An H37Rv Mtb prey library of over 10 6 independent clones was generated in pUAB300. Briefly, H37Rv genomic DNA was partially digested with HpaI, and fragments between 0.5 and 2 kb were cloned into the ClaI site of pUAB300. Library DNA was generated by a MaxiPrep (Qiagen). To screen the library, electrocompetent bait strains were transformed with the Mtb library DNA and plated to Kan 25 /Hyg 50 /Trim 50 plates. Screens were considered successful if they resulted in 10 6 kan R hyg R transformants. Clones that grew on trimethoprim were screened by PCR to eliminate those that grew because of the presence of pUAB300 plasmids containing the Mtb dfrA (Rv2763c) gene. Sequencing of the prey Mtb genomic library plasmid identified the genes fused to DHFR [F1,2]. The MmpL11 D2-LpqN interaction was confirmed by direct cloning of LpqN lacking its signal sequence into pUAB300 to generate in frame fusions with the DHFR [F1,2] fragment. The MmpL D1 and D2-LpqN interactions were measured by determining the trimethoprim minimal inhibitory concentration. Similarly, LpqT, LprG, and Mtc28 were cloned without their signal sequences into pUAB300 to assess their interaction with MmpL3/11 D2 domains. A pUAB300 plasmid containing the Mtb dfrA (Rv2763c) gene was used as a positive control. Strains were serially diluted onto 7H11 agar containing 2-fold dilutions of trimethoprim from 6.25 to 200 g/ml.

Protein purification
The Escherichia coli LpqN expression construct was created by amplifying lpqN lacking its signal sequence via PCR with primers lpqN.61/687 followed by subcloning into pGEM-T Easy. Upon sequence confirmation, the product was cloned into pET-15b (Novagen) in frame with an N-terminal sixhistidine tag. This construct was transformed into E. coli

Characterization of Mtb LpqN
BL21DE3, and protein production was induced at A 600 ϭ ϳ0.5 with 0.4 mM IPTG (ϳ1 h/37°C). Bacteria were harvested via centrifugation and lysed via sonication. Protein was affinitypurified over HisPur Cobalt Superflow agarose resin (Thermo) and confirmed by Western blotting analysis.

LpqN-CFP pulldowns
Recombinantly expressed LpqN and CFPs (BEI Resources, NR-14825, lot no. 610560541) (500 g each) were incubated overnight at 4°C in buffer A (300 mM NaCl, 50 mM NaPO 4 ). Cross-linking was performed using 1% formaldehyde for 30 min at room temperature, followed by quenching with 225 mM glycine for 5 min. Cross-linked samples were buffer-exchanged back into buffer A prior to LpqN purification on HisPur Cobalt Superflow agarose resin (Thermo). LpqN from the non-crosslinked sample was similarly purified. Background CFP binding was assessed by flowing 500 g of CFP alone over the cobalt resin. All resin-immobilized samples were washed (twice in buffer A and once in buffer A ϩ 10 mM imidazole) and eluted in buffer A ϩ 200 mM imidazole. LpqN-interacting proteins and CFP background were identified by MS at the Oregon Health & Science University Proteomics Shared Resource.
Protein mass spectrometry 0.5 ml of each eluate was concentrated and buffer-exchanged with Amicon Ultra 3-kDa molecular mass cutoff filter. Samples were solubilized in 0.05% ProteaseMAX, reduced with DTT, alkylated with iodoacetamide, digested with trypsin overnight, and dried. Each sample was dissolved in 20 l of 5% formic acid, and 20 l/sample was injected into Orbitrap Fusion. Sample digests were loaded onto an Acclaim PepMap 0.1 ϫ 20 mm NanoViper C18 peptide trap (Thermo Scientific) for 5 min at a 5 l/min flow rate in a 2% acetonitrile, 0.1% formic acid mobile phase, and peptides separated using a PepMap RSLC C18, 2-m particle, 75-m ϫ 25-cm EasySpray column (Thermo Scientific) using a 7.5-30% acetonitrile gradient over 60 min in mobile phase containing 0.1% formic acid and a 300 nl/min flow rate using a Dionex NCS-3500RS UltiMate RSLC nano UPLC system. Tandem MS data were collected using an Orbitrap Fusion Tribrid mass spectrometer configured with an EasySpray NanoSource (Thermo Scientific). Survey scans were performed in the Orbitrap mass analyzer at 120,000 resolution, and data-dependent MS2 scans were performed in the linear ion trap using higher-energy collisional dissociation following isolation with the instrument's quadrupole. Comet (version 2015.01, revision 1) was used to search MS2 Spectra against an April 2017 version of the uniprot_Mycobacterium.tuberculosis_ h37rv_both.fasta (8352 entries) FASTA protein database, with concatenated sequence-reversed entries to estimate error thresholds and 179 common contaminant sequences and their reversed forms (44). The database processing was performed with Python scripts available at https://github.com/pwilmart/fasta_utilities. git. 3 Comet searches for all samples were performed with trypsin enzyme specificity. The average parent ion mass tolerance was 2.5 Da. Monoisotopic fragment ion mass tolerance was 1.005 Da. A static modification of ϩ57.02146 Da was added to all cysteine residues. A variable modification of ϩ15.99491 Da on methionine residues was also allowed.

LpqN expression and purification for crystallization studies
Briefly, the LpqN protein containing a His 6 tag at the C terminus was overproduced in E. coli BL21(DE3) cells possessing pET15b⍀lpqN. The cells were grown in 2 liters of LB medium with 100 g/ml ampicillin at 37°C. When the A 600 reached 0.5, the culture was treated with 0.2 mM IPTG to induce LpqN expression, and the cells were harvested within 3 h. The collected bacterial cells were suspended in 100 ml of ice-cold buffer containing 20 mM Na-HEPES (pH 7.2), 200 mM NaCl, 10 mM MgCl 2 , and 0.2 mg DNase I (Sigma-Aldrich). The cells were then lysed with a French pressure cell. Cell debris was removed by centrifugation for 45 min at 4°C and 20,000 rpm. The crude lysate was filtered through a 0.2-m membrane and was loaded onto a 5-ml Hi-Trap Ni 2ϩ -chelating column (GE Healthcare Biosciences, Pittsburgh, PA) pre-equilibrated with 20 mM Na-HEPES (pH 7.2) and 200 mM NaCl. To remove unbound proteins and impurities, the column was first washed with six column volumes of buffer containing 50 mM imidazole, 250 mM NaCl, and 20 mM Na-HEPES (pH 7.2). The LpqN protein was then eluted with four column volumes of buffer containing 300 mM imidazole, 250 mM NaCl, and 20 mM Na-HEPES (pH 7.2). The purity of the protein was judged using 12.5% SDS-PAGE stained with Coomassie Brilliant Blue. The purified protein was extensively dialyzed against buffer containing 20 mM Na-HEPES (pH 7.5), and concentrated to 15 mg/ml.
For the His 6 selenomethionyl (SeMet)-substituted LpqN protein expression, a 1-ml LB broth overnight culture containing E. coli BL21(DE3)/pET15b⍀lpqN cells was transferred into 20 ml of LB broth containing 100 g/ml ampicillin and grown at 37°C. When the A 600 value reached 1.2, the cells were harvested by centrifugation at 3000 ϫ g/min for 10 min and then washed two times with 10 ml of M9 minimal salt solution. The cells were resuspended in 20 ml of M9 medium and then transferred into a 2-liter prewarmed M9 solution containing 100 g/ml ampicillin. The cell culture was incubated at 37°C with shaking. When the A 600 reached 0.4, 100 mg/liter of lysine, phenylalanine, and threonine; 50 mg/liter isoleucine, leucine, and valine; and 60 mg/liter of L-selenomethionine were added. The culture was continued to incubate at 37°C with shaking for 15 min. Protein expression was the induced with 0.2 mM IPTG, and the cells were harvested within 3 h after induction. The procedures for purifying His 6 SeMet-LpqN were identical to those of native LpqN.

Crystallization of LpqN
All crystals of the His 6 LpqN protein were obtained using hanging-drop vapor diffusion. The LpqN crystals were grown at room temperature in 24-well plates with the following procedures. A 20-l protein solution containing 15 mg/ml LpqN protein in 20 mM Na-HEPES (pH 7.5) was mixed with a 2 l of reservoir solution containing 0.1 M sodium citrate (pH 6.0) and 1.9 M (NH 4 ) 2 SO 4 . The resultant mixture was equilibrated against 500 l of the reservoir solution. Crystals grew to a full size in the drops within 2 weeks. Cryoprotection was achieved by raising the glycerol concentration stepwise to 25% with a 5% increment in each step.
Crystals of SeMet-LpqN were prepared using similar procedures. The reservoir solution for crystallizing SeMet-LpqN is the same as that for native LpqN. These SeMet crystals grew to a full size in the drops within 2 weeks. Cryoprotection was achieved by raising the glycerol concentration stepwise to 25% with a 5% increment in each step.
The LpqN-6LT or LpqN-T6D crystals were prepare by incubating the purified LpqN protein (15 mg/ml) with 0.1% 6LT or 0.1% T6D for 2 h at 4°C before crystallization. Crystals of LpqN-6LT and LpqN-T6D were then prepared using similar procedures as above. Their crystallization conditions are identical to each other. The reservoir solution for these complex crystals is the same as that for apo-LpqN (0.1 M sodium citrate, pH 6.0, and 1.9 M (NH 4 ) 2 SO 4 ). Both crystals of LpqN-6LT and LpqN-T6D grew to a full size in the drops within 2 weeks. Cryoprotection was achieved by raising the glycerol concentration stepwise to 25% with a 5% increment in each step.
The LpqN protein contains two methionines (excluding the N-terminal methionine). Within the asymmetric unit of SeMet-LpqN, these two selenium sites were identified using SHELXC and SHELXD (46) as implemented in the HKL2MAP package (47). Single anomalous dispersion was employed to obtain experimental phases using the program Autosol (48) in PHENIX (49). The resulting phases were then subjected to density modification and NCS averaging using the program PAR-ROT (50) using the native structure factor amplitudes. The SeMet sites were also used to trace the molecule by anomalous difference Fourier maps, where we could ascertain the proper registry of SeMet residues. The initial model of LpqN was constructed manually using program Coot (51). Then the model was refined using PHENIX, leaving 5% of reflections in the Free-R set. Iterations of refinement were performed using PHENIX. Model buildings were done using Coot, which led to the current model ( Table 3).
The crystal structures of LpqN-6LT and LpqN-T6D were determined by molecular replacement, utilizing the final structure of LpqN as a search model. The procedures for structural refinement and model building were the same as those of LpqN (Table 3).

LpqN-MmpL11/Ag85A interaction
HA-tagged LpqN was amplified by PCR using lpqN Ϫ570.F/ ϩHA.R primers and the resulting product cloned into the pOLYG plasmid to generate pOLYG lpqN HA where LpqN is expressed from its native promoter. MmpL11 was amplified using mmpL11 TAP.F/R primers and ligated in-frame with the tandem affinity-tagged (TAP; His 6 and FLAG tag) tag into pOLYG-TAP plasmid to generate pOLYG mmpL11-TAP plasmid. MmpL11 expression in the pOLYG mmpL11 TAP is