Dimethylaminophenyl hydrazides as inhibitors of the lipid transport protein LprG in mycobacteria

Abstract

Assembly of the bacterial cell wall requires not only the biosynthesis of cell wall components, but also the transport of these metabolites to the cell exterior for assembly into polymers and membranes required for bacterial viability and virulence. LprG is a cell wall protein that is required for the virulence of Mycobacterium tuberculosis and is associated with lipid transport to the outer lipid layer or mycomembrane. Motivated by available co-crystal structures of LprG with lipids, we searched for potential inhibitors of LprG by performing a computational docking screen of ~250,000 commercially available small molecules. We identified several structurally related dimethylaminophenyl hydrazides that bind to LprG with moderate micromolar affinity and inhibit mycobacterial growth in a LprG-dependent manner. We found that mutation of F123 within the binding cavity of LprG conferred resistance to one of the most potent compounds. These findings provide evidence that the large hydrophobic substrate-binding pocket of LprG can be realistically and specifically targeted by small molecule inhibitors.

Graphical Abstract

Cell wall biosynthesis inhibitors are ubiquitous antibiotics used throughout our enduring fight against bacterial infections. Strategies to break down the cell wall have focused on inhibiting enzymes involved in the biosynthesis of cell wall components. However, as our understanding of how these components are transported to the cell wall has improved— particularly on the molecular and mechanistic level— and our appreciation of their contributions to bacterial survival and pathogenesis has grown, transport pathways have become realistic and attractive therapeutic targets.

In Gram-negative bacteria with two membranes, the assembly of the outer membrane presents particular challenges that demand correspondingly complex machinery: Crossing two lipid bilayers and an intervening aqueous periplasm requires a panoply of proteins that includes soluble chaperones, membrane proximal lipoproteins and integral membrane transporters. These diverse players afford distinct opportunities for small-molecule intervention, but also present myriad challenges for inhibitor discovery and characterization. Transport to the Gram-negative outer membrane comprises three major pathways: the Bam pathway for outer membrane proteins, the Lol pathway for lipoproteins, and the Lpt pathway for lipopolysaccharides. In addition, the Mla pathway performs retrograde transport of phospholipids to help maintain outer membrane bilayer asymmetry^1–3. Both phenotypic whole-cell- and target-based discovery approaches have yielded inhibitors (primarily of integral membrane transporters) in some of these pathways^4–8. While functional inhibition of the target is the primary mechanism of action, other intriguing modes have been uncovered. For example, the synergistic activity between novobiocin and polymyxin against Escherichia coli is explained by novobiocin’s ability to stimulate LptB activity and thereby increase lipopolysaccharide at the cell surface for targeting by polymyxin⁹. Even if transport is not strictly essential for survival, the potential to sensitize pathogens to other antibiotics makes transport pathways appealing targets.

Although mycobacteria are taxonomically classified as Gram-positive, they are surrounded by two distinct lipid layers that make them structurally similar to Gram-negative bacteria. Due to its unique structure and composition compared to the Gram-negative outer membrane, the outermost lipid layer of mycobacteria is commonly referred to as the mycomembrane. Knowledge of lipid transport pathways in mycobacteria lags behind that in Gram-negatives. Nevertheless, significant strides have recently been made in targeting MmpL3, an essential cytoplasmic membrane transporter for the cell wall precursor lipid, trehalose monomycolate. Inhibitors of MmpL3 were initially identified by their activity against the human pathogen Mycobacterium tuberculosis (Mtb) The function of MmpL3 as a transporter for mycolic acids in the cell wall was elucidated only after identifying mutations that confer resistance^10,11. The most recent seminal finding was the crystal structure of MmpL3, with the promise of enabling structure-based design^12,13. Even so, target-based screening would require a functional assay, a difficult requirement for an integral membrane transporter. Elegant studies in spheroplasts have examined MmpL3 function in the presence of several inhibitors^14,15; however, radioisotope detection and limited sensitivity of the spheroplast system restrict its use in higher-throughput screens.

In general, the difficulties of working with integral membrane proteins biochemically pose significant challenges for target-based inhibitor development. However, many outer membrane transport pathways also require soluble accessory proteins in the cell wall^16–18. These proteins are potentially more tractable for target-based inhibitor discovery because they are more likely amenable to in vitro screening and structure-based design. They vary in fold, but share a central hydrophobic cavity, consistent with their roles in binding to and thereby solubilizing hydrophobic substrates such as lipids within the aqueous periplasm.

The inhibition of lipid transfer proteins has precedent in mammalian systems. Microsomal triacylglyceride transport protein (MTP) and cholesterol ester transfer protein (CETP) function in the assembly of lipoproteins and have been targeted for the treatment of hypercholesterolemia and atherosclerosis^19,20. While no structural information is available for MTP with the FDA-approved inhibitor lomitapide (BMS201038), co-crystal structures of CETP with various inhibitors show that the compounds do not displace cholesterol ester from CETP, but instead bind proximally in the central channel through which lipids are thought to transit during transfer between lipoprotein particles²¹. In contrast with this noncompetitive binding, inhibitors of fatty acid binding proteins (FABPs), which traffic fatty acids within eukaryotic cells and have functions in diverse metabolic and immune responses, compete directly for the binding of substrates to a hydrophobic cleft. While FABPs are amenable to in vitro screening via fluorescence-based binding assays, computational docking has proved a fruitful initial step^22–26. For example, a computational docking screen of one million commercially available compounds with FABP7 led to the experimental testing of 48 compounds, yielding one hit with submicromolar affinity for the anandamide transport protein FABP5 and activity consistent with modulation of endocannabinoid trafficking in both cell culture and mice. These results highlight in silico screening as an inexpensive first step in obtaining high-affinity, biologically active compounds, at least in cases where the structure of the target protein is known.

Proteins within bacterial outer membrane transport pathways have analogous functions and structures to FABPs and also play important roles in bacterial physiology. However, there have been few documented efforts to identify inhibitors, computationally or otherwise. Among the few examples is a whole-cell vancomycin susceptibility screen that yielded MAC13243, a putative inhibitor of the cell wall protein LolA, which binds to the lipid modification on bacterial lipoproteins during their transport to the Gram-negative outer membrane²⁷. Within mycobacteria, members of the lipid-binding lipoprotein (Llp) family have been implicated in lipid export to the mycomembrane. The structures of all four Llp proteins (LppX, LprG, LprA, LprF) are highly similar, but have a range of cavity sizes that may correlate with the sizes of their respective substrates²⁸. LppX is associated with the virulence-associated lipid phthiocerol dimycocerosate²⁹, and LprG with triacylated lipids, including triacylglycerides (TAG) and lipoglycans such as lipoarabinomannan^30–32. Importantly, loss of function in LprG severely attenuates Mtb in the mouse model of infection³², indicating its potential as a target whose inhibition could be part of a multidrug regimen, which is the standard of care in tuberculosis therapy. In addition, crystal structures are available for LprG bound to TAG or to the lipoglycan core lipid, phosphatidyl inositol mannoside^30,32. These attributes make LprG an attractive candidate for computational docking. Towards validating LprG as a drug target and exploring the requirements for and consequences of small-molecule inhibition, we present here the results of a computational screen for LprG ligands and subsequent experimental characterization of hit compounds for their in vitro target binding and cellular activity.

RESULTS AND DISCUSSION

For the initial screening library, we selected a subset of the ZINC database^33,34 of commercially available compounds that combined the Maybridge and AldrichCPR collections. The Maybridge collection was chosen because a previous whole-cell screen of this collection led to the identification of MAC13243 as a ligand for the lipoprotein transporter LolA²⁷, which shares structural homology with LprG. The AldrichCPR collection was selected because it has a high proportion of compounds with molecular weights > 500 Da. We reasoned that larger compounds were more likely to make favorable interactions within the ligand-binding cavity of LprG, given the large size of the hydrophobic portion of ligands co-crystallized with LprG: acylated phosphaptidyl inositol dimannoside (Ac₁PIM₂, 1415 Da; PDB ID: 3MHA) and tripalmitoyl glyceride (807 Da; PDB ID: 4ZRA), both of which fully occupy the hydrophobic cavity of LprG. Following flexible docking of approximately 250,000 compounds, we identified hits by subjecting the scored compounds to a series of filters that incorporated the grid score (a proxy for binding energy) and the footprint similarity score (a measure of similarity of protein contacts versus a known ligand; in this case, TAG)³⁵. Cheminformatics analysis was further applied to optimize for predicted solubility and for anti-Mtb activity and acceptable Vero cell toxicity as predicted by a Bayesian model³⁶. Setting cutoffs for Bayesian score and predicted solubility yielded 60 compounds, including 12 pairs of cis/trans isomers. Since the compounds were available only as racemic mixtures, the resulting list comprised 48 unique compounds (Table S1). This final set did not include any compounds from the Maybridge collection such as the LolA ligand MAC13243.

We then tested compounds for binding to LprG in a fluorescence-based competitive assay using 12-N-methyl-(7-nitrobenz-2-oxa-1,3-diazo) aminostearic acid (NBD-SA) as a reporter ligand. As noted earlier, NBD-SA has been used to identify ligands for FABPs. NBD is a solvatochromatic fluorophore, with a higher quantum yield of fluorescence in environments with, for example, low polarity or high hydrophobicity. Thus, NBD-SA binding to FABP results in increased fluorescence; decreased fluorescence in the presence of a putative ligand or small molecule is thus interpreted as direct binding and displacement of NBD-SA from the binding cleft²³. Although NBD-SA is significantly smaller than native LprG ligands such as TAG, it is much more soluble and therefore tractable for in vitro binding experiments. For these assays, Mtb LprG was expressed without the N-terminal secretion signal or conserved lipobox motif and was therefore non-acylated and soluble, but retained an intact binding cavity^30,32. The dissociation constant (K_d) for NBD-SA and LprG was 20 μM (Figure 1A). To identify compounds with micromolar affinity for LprG, we estimated assay conditions that would yield a ~50% decrease in NBD fluorescence for a compound with an effective concentration at 50% occupancy (EC₅₀) of 10 μM. Of the 48 compounds, five were insoluble in assay buffer and were not pursued further. Of the remaining 43, several resulted in reduced NBD fluorescence in the presence of LprG. However, the majority of compounds unexpectedly elicited increased NBD fluorescence in the presence and/or absence of LprG (Figure S1). Many of the compounds affected NBD fluorescence in the absence of LprG, possibly by aggregation with NBD-SA. (Note that based on their structures, none of the compounds are expected to fluoresce under the excitation/emission conditions used to monitor NBD fluorescence; Table S1). In addition, due to the large size of the LprG cavity, increased fluorescence in the presence of LprG could arise from a further increase in the hydrophobicity of the environment around NBD-SA due to simultaneous binding of compound and NBD-SA. For this study, neither hypothesis regarding aggregation or simultaneous binding with NBD-SA was pursued since several compounds appeared to successfully compete with NBD-SA for binding with LprG. These compounds were prioritized for follow-up. Quantitative metrics and cutoffs were established to take into account the range of effects the compounds had on NBD fluorescence both with and without protein [F(prot)/F(+prot) < 0.15], and with and without compound [F(+cmpd)/F(-cmpd) < 1.5] (see also Materials and Methods). This filtering yielded 11 compounds that showed a low degree of interaction with NBD-SA alone (Figure 1B; white bars) and moderate effects on NBD-SA together with LprG, relative to other compounds screened (Figure 1B; black bars).

Figure 1. Compounds identified as LprG ligands by computational docking affect binding of a fluorescent fatty acid to LprG.

A) The fluorescent fatty acid NBD-SA (200 nM) was incubated with increasing concentrations of purified LprG. Data shown are from one experiment with three technical replicates. A second experiment yielded the same dissociation constant. B) NBD-SA (200 nm) alone (white bars) or with LprG (10 μM; black bars) was incubated with the indicated compounds at 20 μM and the normalized fluorescence in the presence and absence of compound [ F(+cmpd) / F(−cmpd) ] determined. Dotted line indicates F(+cmpd) / F(−cmpd) = 1. Data shown are the average of two experiments with three technical replicates each.

The EC₅₀ was then determined for each of the 6 compounds (LB01 through LB06) with F(+cmpd)/F(-cmpd) ≤ 1 as an indication of NBD-SA displacement, using the same fluorescence assay as above but in a compound concentration-dependent manner (Figure 2). In the initial assay with 20 μM compound, LB02 yielded the highest fluorescence compared to the other 5 compounds when incubated with NBD-SA alone [highest F(-prot)/F(+prot)] (Figure 1B, open bars). Consistent with this result, LB02 with NBD-SA alone (in the absence of LprG) showed a concentration-dependent increase in fluorescence that precluded accurate EC₅₀ determination (Figure 2B; open circles). LB02 was therefore not further characterized. Although the remaining compounds did not display this behavior, low solubility overall prevented exact measurement of the EC₅₀. The EC₅₀ are therefore upper estimates and were used only to prioritize compounds for further characterization. Nevertheless, the EC₅₀ were consistent with the initial assay results: LB01 and LB03 had the highest affinities for LprG (low micromolar EC₅₀), while the affinities of LB04, LB05, and LB06 were much weaker (EC₅₀ ~100 μM) (Figure 2A, C–F). Overall, these results showed that the computational docking screen yielded small molecule ligands for LprG.

Figure 2. Hit compounds bind to LprG with micromolar EC₅₀.

A-F) NBD-SA (200 nM) and LprG (10 μM) were incubated with increasing concentrations of the indicated compounds. Data shown are from one experiment. EC₅₀ are the result of between 1 and 3 independent experiments; where the S.D. is reported, n = 3. Otherwise, all outcomes of individual experiments are reported.

LprG is an attractive potential drug target because it is required for Mtb virulence in an animal model and we have proposed that Mtb lacking LprG grows slowly in vivo due to metabolic dysregulation in the host environment³². Indeed, we have shown that under nutrient-limited conditions in vitro, loss of function in LprG indeed leads to a growth defect in both Mtb and M. smegmatis^32,37. We took advantage of this phenotype to assess the cellular activity of hit compounds based on growth inhibition of fastgrowing, non-pathogenic M. smegmatis as an experimentally expedient proxy for Mtb. Because removing lprG can variably affect expression of the co-transcribed major facilitator superfamily (MFS) transporter gene rv1410c^38,39, we used a targeted ΔMSMEG_3070–3069 strain, which lacks the entire homologous operon in M. smegmatis⁴⁰. Where applicable, complement strains were constructed using stably integrated Mtb lprG and/or rv1410c (Table S3). While the sequence identity (53%) and similarity (68%) between Mtb LprG and MSMEG_3070 are moderate, many residues that line the lipid-binding pocket are conserved, as determined from a comparison of the Mtb LprG crystal structure and an MSMEG_3070 homology model (not shown). This conservation further supported testing our computational results from Mtb LprG in M. smegmatis.

To facilitate compound screening, we initially tested compound activity in an autoluminescent strain in which a bacterial luxABCDE operon was inserted into the chromosome at the L5 phage attachment site. This operon encodes the biosynthesis of the fatty acid aldehyde substrate for the bacterial luciferase, which also depends on reduced flavin mononucleotide to generate light. Stable integration allowed us to maintain the operon in the absence of antibiotics and thus remove antibiotic selection as a variable while testing the effects of our hit compounds on bacterial growth and viability. Analogous strains have been used to monitor survival and response to drug treatment in vitro, in macrophages, and in animal models for M. smegmatis and Mtb^{41– 43}. We first validated that the growth phenotype of ΔMSMEG_3070–3069 was recapitulated by autoluminescence (Figure S2A). This is consistent with our previous characterization of the growth phenotype in these strains by OD₆₀₀³⁷. LB01 was insoluble in the culture medium and LB03 and LB06 had similar activity against both the wild type and ΔMSMEG_3070–3069 in initial experiments, suggesting that their activity does not stem from inhibition of MSMEG_3070 or MSMEG_3069. Based on these results, these three compounds were not further characterized. LB04 and LB05 also displayed dose-dependent growth inhibition of wild-type M. smegmatis (Figure S2B,C) and, crucially, ΔMSMEG_3070–3069 was less sensitive to LB04 an LB05 than the wildtype, suggesting that growth inhibition arises at least in part from LprG inhibition.

Given that LB04 and LB05 have cellular activity and differ by only one methylene unit in the central alkyl chain linker, we pursued structure-activity relationships based on structurally similar commercially available compounds. Compounds LB04-I, -II, and –III have 3, 4- and 6-carbon linkers (vs. 7- and 8-carbon linkers for LB04 and LB05) between the two dimethylamino phenylmethylidene hydrazide end groups (Figure 3A). The five commercially available compounds and their 5-, 9-, and 10-carbon linker analogues (both cis and trans isomers) had similar energies in docking calculations (Table S2). In accordance with their similar grid scores, the EC₅₀ for the commercially available compounds were also similar (~100–200 μM) (Figure 2D,E; Figure 3C–D). Interpretation of their biological activity using autoluminescent M. smegmatis strains was complicated, however, by a pronounced increase in the autoluminescence of ΔMSMEG_3070–3069 at all concentrations tested (Figure S2). These observations suggested that LB04-II and LB04-III (and possibly also LB04) modulate the function of the luxABCDE operon, either directly or indirectly. We concluded that these results were sufficient to indicate dose-dependent activity for all the compounds, but that in general autoluminescent strains should be used with caution when testing compound activity due to effects on luminescence that are potentially unrelated to viability.

Figure 3. Compounds structurally related to LB04 and LB05 have similar EC₅₀.

A) Structures of LB04, LB05, and related compounds LB04-I, LB04-II, and LB04-III. B-D) NBD-SA (200 nM) and LprG (10 μM) were incubated with increasing concentrations of the indicated compounds. Data shown are from one experiment. Reported EC₅₀ are the result of 1 or 2 independent experiments. Across multiple experiments, data for LB04-I were insufficiently robust to fit an EC₅₀.

Nevertheless, the increase in autoluminescence was largely independent of compound concentration for LB04-II and LB04-III. Assuming that this effect is an additive constant at all concentrations, the sensitivity of M. smegmatis to the compounds increased as the linker length increased from 4 to 6 carbons from LB04-II to LB04 (Figures S2B–E). We therefore sought to identify potential molecular underpinnings of the apparent differences in cellular activity among compounds despite their similar in vitro affinities for Mtb LprG. A manual review of the 30 lowest-energy computationally docked poses for LB04-II, LB04-III, LB04 and LB05 revealed that the scaffold often adopts a bent conformation, with one end group buried in the hydrophobic cavity (including contacts at the bottom of the cavity) and the other bent towards the protein surface at the opening to the cavity (Figure 4A). A comparison of the average footprint similarity profiles for each compound and for TAG revealed that the compounds have similar profiles to each other and recapitulate many contacts made by TAG, including contacts to valines and alanines that line the ligand-binding cavity, as well as to several aromatic residues such as F104, F123, and Y130 (Figure S3). Notably, while the interaction energies at F104 and Y130 are similar for all compounds (S.D. across all compounds: 0.22 and 0.16 kcal/mol or 10% and 3% of the average energy value at each position), the energy at F123 varied more greatly (S.D. 0.28 kcal/mol or 28% of the average energy). Moreover, across the 30 lowest-energy poses for each compound, the energy at F123 was bimodal with values either > 2 kcal/mol or close to 0. For poses with energy > 2 kcal/mol, the phenyl ring was either approximately parallel or perpendicular to F123, supporting a pi-stacking interaction (Figure 4A). While F104 and Y130 are conserved at the corresponding positions within the LprG homologue MSMEG_3070, F123 is a tyrosine (Y125). Nevertheless, we would expect that MSMEG_3070 Y125 to be capable of engaging in similar pi-stacking interactions with LB04 and related compounds.

Figure 4. LprG F123A confers resistance to LB04.

(A) Representative pose of LB04 (orange) in the crystal structure of Mtb LprG (PDB ID: 4ZRA). Inset highlights F123 (cyan) in proximity to an aromatic heterocycle of LB04. (B) Colony forming units (CFU) were enumerated by plating on agar after first culturing M. smegmatis for 18.5 h. Data are the average ± SD of six biological replicates performed in two experiments. *p < 0.02, **** p < 0.0001 vs wild-type; ++++ p < 0.0001 vs ΔMSMEG_3070–3069::lprG-rv1410c. (C) Percent viability (+cmpd/−cmpd) as determined by CFU (dark purple bars) after treatment with 10 μM rifampicin for 20 h. Data are the average ± SD of three independent experiments. ** p < 0.005, *** p< 0.0005 vs. wild type, ++ p<0.005, +++ p<0.0005 vs. ΔMSMEG_3070–3069::lprG-rv1410c. (D) Percent viability as determined by CFU after treatment with 100 μM LB04-III (blue bars) or 100 μM LB04 (orange bars) for 18.5 h. Data are the average ± SD of six biological replicates performed in two independent experiments. * p < 0.05, ** p < 0.005 vs wild-type; + p < 0.05, ++ p < 0.005 vs ΔMSMEG_3070–3069::lprG-rv1410c. E) Anti-Mtb LprG immunoblot against wild-type, knockout, and complement strains. The antibody does not cross react with MSMEG_3070. GroEL was used as a loading control. Data shown are from one biological replicate; the reported normalized fluorescence signal F(LprG)/F(GroEL) is the average of two biological replicates. (F) CFU were enumerated by plating on agar after first culturing M. smegmatis for 18.5 h. Data are the average of three biological replicates ± SD performed in one experiment. * p < 0.05, ** p<0.005 vs. wild type. (G) Percent viability as determined by CFU after treatment with 100 μM LB04-III (blue bars) or 100 μM LB04 (orange bars) for 18.5 h. ** p < 0.005, *** p < 0.0005 vs wild-type; + p < 0.05, ++ p < 0.005, +++ p < 0.0005 vs ΔMSMEG_3070–3069::lprG-rv1410c; † p < 0.05 vs ΔMSMEG_3070–3069::rv1410c; Δp < 0.05 vs ΔMSMEG_3070–3069. Data are the average of three biological replicates ± SD performed in one experiment. All statistical analyses are one-way ANOVA with Tukey’s test for multiple comparisons, assuming normal distributions and equal variance for all populations sampled. All other pairwise comparisons were not significant.

Based on these analyses, we hypothesized that F123 is a key residue in determining the activity of LB04 and related compounds and that removing the potential for pi-stacking interactions at this position by mutating the phenylalanine to leucine or alanine would modulate compound activity. Given the limitations of the autoluminescence assay noted above, we first confirmed the growth phenotypes of the wild type, ΔMSMEG_3070–3069, and complement strains by enumerating colony forming units (CFU) (Figure 4B). Upon treatment with rifampicin, ΔMSMEG_3070–3069 and the complement with rv1410c (which still lacks LprG function) showed greater sensitivity to the antibiotic than either the wild-type or lprG-rv1410c operon complement (Figure 4C). This is similar to what was observed by Hohl et al. upon complementation of ΔMSMEG_3070–3069 with the M. smegmatis homologues and is consistent with another report that the Mtb operon can complement multiple phenotypes of ΔMSMEG_3070–3069, including a sliding motility defect and increased susceptibility to ethidium bromide^40,44. Our results further underscore the antibiotic sensitivity phenotype in the absence of LprG and/or Rv1410c function and also the conservation of function across homologues.

In contrast, ΔMSMEG_3070–3069::rv1410c was significantly more resistant than the wild type to LB04-III and LB04 and sensitivity to both compounds was restored in the operon complement (Figure 4D). ΔMSMEG_3070–3069 had an intermediate phenotype, although the difference in viability compared to all other strains was not significant under these treatment conditions. While toxicity remained high (~50%) in the mutant strain lacking LprG function, presumably due to off-target effects, these results still confirmed specificity of LB04-III and LB04 activity for LprG and allowed us to test our hypothesis regarding the importance of F123 for compound activity. Complementing ΔMSMEG_3070–3069 with lprG-rv1410c containing lprG encoding a F123A or F123L mutation resulted in LprG expression similar to that of the wild type (Figure 4E). Growth of the LprG F123 mutant strains as measured by CFU trended higher than the lprG null strain, ΔMSMEG_3070–3069::rv1410c, but this difference was not significant (Figure 4F). While the limited growth complementation complicated testing our hypothesis, these results suggested that despite the fractional contribution of F123 to the overall binding energy (Figure S3), this residues plays an important role in LprG for capturing native substrates and that substrate binding and associated functions of LprG are required for mycobacterial growth.

Although the lprG mutant complement strains were not significantly different than the lprG null strain in growth, both were more sensitive to 100 μM LB04-III (Figure 4G), providing evidence that LprG F123A and F123L are partially functional and still inhibited by LB04-III. In contrast, F123A was significantly more resistant to LB04 than both the wild type and the wild-type complement. As we predicted, removal of the phenylalanine sidechain at residue 123 modulated inhibitor activity, although with greater structural restrictions than anticipated: changes in activity were observed for F123A, but not the more conservative mutation F123L, and for the activity of LB04, but not LB04-III, which is shorter by one methylene unit.

In contrast to their activity against M. smegmatis, LB04 and LB04-III up to 100 μM did not significantly affect the growth of Escherichia coli K12, Staphylococcus saprophyticus, or Corynebacterium glutamicum as representative Gram-negative and Gram-positive bacteria (Figure 5). Notably, C. glutamicum has a similar cell envelope architecture to mycobacteria, but lacks an obvious homologue of LprG. These results provide additional support for LB04 and LB04-III as compounds that limit mycobacterial growth by inhibiting LprG function.

Figure 5. LB04-III and LB04 do not have significant activity against other representative Gram-negative and Gram-positive bacteria.

Percent viability (+cmpd/−cmpd) as measured by BacTiter-Glo after culturing for approximately 4–5 doubling times in the presence of LB04-III (black squares), LB04 (gray squares), rifampicin (RIF; open circles), or streptomycin (STR; open triangles) for A) E. coli K12, B) S. saprophyticus, and C) C. glutamicum. Data are the average ± S.D. of three independent experiments with three technical replicates each for LB04 and LB04-III. For rifampicin, data are the average ± S.D. of two (E. coli) or one (S. saprophyticus) independent experiment(s) with three technical replicates each. For streptomycin data are the average ± S.D. of four independent experiments with three technical replicates each.

CONCLUSION

Targeting lipid binding proteins with small molecules has established precedent in mammalian systems, with clinical and pre-clinical compounds that inhibit proteins responsible for trafficking fatty acids and transferring lipids between lipoproteins. The mechanism of action is most clearly delineated for fatty acid-binding proteins, in which inhibitor binding evidently competes with fatty acid substrates. However, analogous inhibitors for lipid binding proteins in bacterial systems have not been extensively explored.

Given the absence of a well-defined and specific phenotype for lipid transport function that is amenable to screening, as well as the success of computational docking approaches for the functionally and structurally analogous FABPs, we used computational docking to identify inhibitors of the mycobacterial lipid transportassociated lipoprotein LprG. Our target-based approach yielded several commercially available compounds with low- to mid-micromolar affinity for LprG. Two of these compounds, both bis-dimethylamino benzylidene hydrazides, inhibited mycobacterial growth in a LprG-dependent manner, but also showed significant activity in the absence of LprG. Surprisingly, the double mutant strain lacking LprG and Rv1410c was more sensitive to LB04 and LB04-III than when the strain was complemented with Mtb rv1410c. One possible explanation is LB04 and LB04-III inhibit LprG, but are in fact substrates for Rv1410c. The presence of Rv1410c in ΔMSMEG_3070–3069::rv1410c therefore decreases intracellular LB04/LB04-III concentrations and reduces inhibition of non-LprG targets in the cytosol, conferring greater resistance to ΔMSMEG_30703069::rv1410c vs. the double mutant ΔMSMEG_3070–3069. While Hohl et al. showed that Rv1410c is not a general efflux pump, it is possible that Rv1410c can promote the export of molecules that are more similar to its native substrates. Clearly, additional studies are needed to clarify the precise functions of LprG vs. Rv1410c and resultant effects on membrane permeability and antibiotic susceptibility.

Non-specific inhibition of growth by LB04 and structurally related compounds could be due to the detergent-like properties of the scaffold, in which a saturated carbon chain connects two moderately hydrophilic aromatic heterocycles (Figure 3A). Indeed, the lengths of LB04 and LB05 are comparable to that of intermediate chain fatty acids (e.g., C16 and C18), which are more toxic to Mtb than short or long chain fatty acids⁴⁵. Preliminary microscopy analysis did not reveal obvious morphological defects following treatment of either wild-type or null strains with LB04 (data not shown); visualizing membranes and cell wall components with appropriate labels may help reveal less obvious effects. Overall, defining the relationship between cellular uptake, membrane disruption, and LprG-dependent activity for these compounds will be an important future area of investigation.

While LprG is conserved across all mycobacteria, the other members of the lipid-binding lipoprotein (Llp) family (LppX, LprA, LprF) are restricted to pathogenic species such as Mtb. LppX also has a confirmed role in lipid transport and Mtb virulence and all Llps bind to lipids in hydrophobic pockets of various sizes. Binding studies with LprG and deacylated forms of the putative substrate lipoarabinomannan suggest that hydrophilic contacts may also contribute to binding affinity and substrate specificity³⁸, but there has been no experimental evidence to suggest that any factor beyond size and hydrophobicity dictates affinity to the cavity. Here we showed that removing a single aromatic sidechain F123 affected the ability of LprG to complement the growth defect and conferred resistance to an inhibitor. While this finding suggests that specific targeting of LprG is possible, one appealing notion is that compounds that bind within the cavity of LprG may also bind to other Llp proteins. Because LprG is the only Llp conserved in Msm, our studies allowed specific analysis of LprG as a target, but inhibitors of LprG may more broadly inhibit Llp proteins as a family in pathogenic mycobacteria, including Mtb. As noted above, other transport proteins that operate in the same pathways (such as Rv1410c for the LprG pathway and MmpL7 for the LppX pathway) and therefore share substrates may also be inhibited by compounds that bind to Llp proteins, suggesting a possible route to further compromising Mtb virulence and survival.

While LB04 in particular shows promising activity and specificity against M. smegmatis in culture, the hydrazide groups and long alkyl chain mark this scaffold for moderate risk for safety in mammals based on in silico ADME-toxicity filtering⁴⁶. Additional structure-activity relationship studies will help determine whether the hydrazide functionality and linker can be modified to mitigate mammalian cell toxicity while improving anti-mycobacterial activity, towards exploring the consequences of LprG and Llp inhibition in the context of Mtb infection.

EXPERIMENTAL SECTION

Bacterial strains, culture media, reagents, and DNA mutagenesis

All bacterial strains are reported in Table S3. Unless otherwise noted, Mycobacterium smegmatis was cultured at 37 °C with orbital shaking in 7H9 Middlebrook (BD Biosciences) with 1% casamino acids (Amresco), 0.2% glycerol, 0.2% glucose, and 0.05% Tween 80, or in a modified Sauton’s medium in which propionate is the primary carbon source (4 g of L-asparagine, 0.5 g KH₂PO₄, 0.05 g ferric ammonium citrate, 0.5 g MgSO₄•7H₂O, 0.01 g ZnSO₄, 2 g citric acid, 0.05% (v/v) Tyloxapol, pH 7 per 1L, plus a final concentration of 10 mM sodium propionate). Escherichia coli K12, Staphylococcus saprophyticus, and Corynebacterium glutamicum were cultured in LB medium at 37 °C with orbital shaking.

All compounds tested for binding to LprG and inhibition of bacterial growth were purchased from Sigma (Table S1) and used without further purification. Rifampicin and streptomycin were purchased from Fisher Scientific and Calbiochem. Compounds were dissolved in DMSO (with the exception of streptomycin, which was dissolved in sterile water), and the final concentration of vehicle in all assays was 1% (v/v) unless otherwise noted.

Site-directed mutagenesis to generate the mutations F123A (forward primer GGAGCGATGCCGGTCCCGCCGCCGACATC; reverse primer GCGGGACCGGCATCGCTCCACTGGTTGGGCGTC) and F123L (forward primer GGAGCGATTTGGGTCCCGCCGCCGACATC; reverse primer GCGGGACCCAAATCGCTCCACTGGTTGGGCG) in Mtb LprG (Table S2) was performed using Phusion DNA polymerase (NEB).

Computational screen for small-molecule ligands of LprG

A virtual screen was performed using DOCK 6 to computationally dock a library of commercially available small molecule compounds³⁵. The library comprised the AldrichCPR and the Maybridge collection subsets from the ZINC database³³, yielding approximately 250,000 compounds. The structure of Mtb LprG (PDB ID: 3MHA) was prepared for docking by removing the co-crystallized acylated Ac₁PIM₂ lipid ligand to create apo-LprG. Charge was added to the structure using Amber ff99SB⁴⁷ and AM1-BCC^48,49 in UCSF Chimera⁵⁰. An energy grid for the receptor was generated by selecting surface spheres in 3MHA within 10 Å from PIM binding site, then the grid box was extended 8 Å from the selected spheres. Each compound was flexibly docked to the grid using the standard FLX docking protocol⁵¹. The grid scores were calculated for 10³ orientations and the lowest-energy pose for each compound was preserved. All compounds were ranked by grid score and the top 1,500 compounds were selected for further cheminformatic analysis and filtering.

The footprint similarity (FPS) scores⁵² for the top 1500 compounds were calculated. For these simulations, a more recent co-crystal structure of LprG in complex with tripalmitoyl glyceride (PDB ID: 4ZRA) was used³². TAG is also a triacyl lipid, but lacks the extended sugar headgroup of Ac₁PIM₂ that makes many hydrophilic contacts exterior to the hydrophobic ligand cavity. TAG thus represents a minimal “core” structure appropriate for footprint similarity analysis, which is designed to help identify ligands that will successfully compete with substrate binding. TAG was docked into apoLprG (PDB ID: 3MHA) with a rigid docking protocol⁵¹ followed by Cartesian minimization. The FPS for TAG in the final pose for 3MHA and the original structure 4ZRA were nearly identical, validating TAG docked in 3MHA as a reference for FPS scoring with the docked small molecules. The lowest-energy pose of each docked small molecule (as determined above) was subjected to energy minimization using the same protocol as for TAG. The van der Waals and electrostatic FPS score of each docked molecule versus the reference ligand TAG were computed using DOCK 6 and all 1,500 compounds were ranked by FPS. The top 500 compounds by FPS were selected, among which the top 100 compounds by grid score were retained (Group A). Similarly, the top 500 compounds by grid score were selected, among which the top 100 compounds by FPS were retained (Group B). The union of groups A and B yielded 133 unique compounds (Group C).

For the top 1,500 compounds by grid score, solubility was predicted using a multiple linear regression (MLR) model based on a ~1000-compound data set. The compounds were also ranked using a dual-event Bayesian model for activity against Mycobacterium tuberculosis and acceptable Vero cell toxicity (selectivity index ≥ 10) based on a 100,000-compound library³⁶. Of the Group C compounds from above, the 60 compounds with MLR solubility > −9 and Bayesian score > −1 were retained. There were 12 pairs of cis/trains isomers among these 60 compounds. The remaining 48 unique compounds were purchased from Sigma (as racemic mixtures for those compounds with isomers).

Additional DOCK calculations were performed on LB04, LB05, and the structurally related compounds LB04-I, LB04-II, and LB04-III. In addition, related compounds with 5-, 9-, and 10-carbon linkers (which are not in the ZINC database) were built in Chimera, followed by bond order correction and energy minimization. Both the cis and trans isomers for each compound were simulated. For these simulations, 10³ orientations were again calculated via the standard FLX protocol and the 30 lowest-energy poses (as determined by the grid score; the energies of all poses were within 10 kcal/mol of each other) were retained. In addition, 10⁶ orientations were calculated for rigid docking of TAG to LprG (PDB ID: 3MHA), which was found to improve the overall grid score compared to 10⁴ iterations (−112 vs. −75 kcal/mol) in addition to reducing internal repulsive energy (84 vs. 114 kcal/mol). The RMSD of the final TAG orientation compared to the TAG pose within the crystal structure (PDB ID: 4ZRA) was less than 2 Å, which is considered successful pose reproduction³⁵. This final pose was used as the reference for FPS scores, which were calculated for each of the 30 poses for each compound and then averaged.

Competitive binding assay for small-molecule ligands to LprG

A N-terminally truncated, non-acylated construct of Mtb LprG (also known as Rv1411c) was expressed and purified with a C-terminal 6xHis tag as previously described³². The K_d of the reporter ligand, 12-N-methyl-(7-nitrobenz-2-oxa-1,3-diazo) aminostearic acid (NBD-SA, Avanti Polar Lipids), was determined by adding 200 nM NBD-SA (0.5% DMSO v/v) final concentration to NA-MtbLprG in assay buffer (50 mM Tris, 1 mM DTT, 100 mM NaCl, pH 8.0). Assays were performed in 96-well plates and the total volume in each well was 200 μL. Buffer alone and buffer containing 200 nM NBD-SA served as negative controls. Fluorescence was measured at 22 °C with a microplate reader (Filtermax F5, Molecular Devices) using an excitation filter centered at 485 nm and an emission filter centered at 535 nm. To determine the K_d, raw fluorescence data (F) were fitted to the equation F = B x [LprG]/(K_d+[LprG]) + C, where B is a scale factor for the maximum response and C is a constant offset (GraphPad Prism). For data fitting, measurements in the absence of protein ([LprG] = 0) were plotted on the semi-log plot at [LprG] = 0.01 μM (plotting the data at lower values of [LprG] did not affect the resulting K_d). Of the 48 compounds purchased, 5 were insoluble in assay buffer, so 43 in total were assayed.

To compare the affinities of compounds to LprG by displacement of NBD-SA, 200 nM NBD-SA, 10 μM NA-MtbLprG, and 20 μM compound were mixed in assay buffer. Buffer containing 200 nM NBD-SA served as a negative control. Buffer containing 200 nM NBD-SA and 10 μM NA-MtbLprG served as a positive control. Potential assay interference from interactions between NBD-SA and a given compound was taken into account by measuring fluorescence from 200 nM NBD-SA and 20 μM compound alone in assay buffer [ F(−prot) ]. The total volume in all cases was 200 μL. Fluorescence from NBD-SA was measured as above. Cutoffs for forwarding compounds for further analysis were as follows. An upper limit for the effect of compound on NBD-SA in the absence (-prot) vs. presence of LprG (+prot) was defined as F(-prot)/F(+prot) < 0.15. A lower limit for the apparent displacement of NBD-SA from LprG by the compound (+cmpd) vs. DMSO vehicle (-cmpd) was defined as F(+cmpd)/F(-cmpd) < 1.5.

Use of NBD-SA at > 200 nM (and thus closer to the anticipated micromolar dissociation constants) was precluded by observed increases in NBD fluorescence at higher concentrations. The restricted assay conditions with [NBD-SA] << K_d meant that for the competitive binding assays, the simplifying assumption that total and free concentrations of NBD-SA are approximately equal does not hold and only effective concentrations at 50% binding (EC₅₀) under the specified conditions could be determined for each compound. Data were fitted to the equation Y=(Y_maxY_min)/(1+10^(log([LprG])-log(EC₅₀)) + Y_min, where Y_min was fixed to the average fluorescence for NBD-SA and 100 μM compound, as representative of the value when NBD-SA is completely displaced from LprG.

Autoluminescence assay to monitor M. smegmatis viability

Overnight cultures of autoluminescent M. smegmatis in modified Sauton’s medium (final OD₆₀₀ ~0.8–1) were subcultured to OD₆₀₀ 0.1 and incubated with varying concentrations of rifampicin, LB04 (Sigma R748544), LB05 (Sigma S129461), LB04-III (Sigma R747580), LB04-II (Sigma R747645), or DMSO vehicle (1% v/v final) in a total volume of 202 μL in 96-well plates. Plates were incubated in a multimode plate reader (FilterMax F5, Molecular Devices) at 37 °C with 15 min of orbital shaking between each measurement for approximately 20 hours total. Each condition was assayed in technical triplicate in two independent experiments. The percent viability at t = ~18.5h (1108 min or 4–5 doubling times in modified Sauton’s medium) was calculated as RLU(+cmpd)/RLU(−cmpd) x 100.

Luminescence assay to determine bacterial viability

E. coli, S. saprophyticus, and C. glutamicum were grown to mid- to late-log phase and then subcultured to OD₆₀₀ 0.1 with varying concentrations of LB04-III and LB04, as for M. smegmatis above. After incubation times corresponding to 4–5 doubling times (~2, 2.5, and 11.5 hours total for E. coli, S. saprophyticus, and C. glutamicum, respectively), viability was determined using BacTiter-Glo (Promega). Briefly, 100 μL each of BacTiter-Glo reagent and bacterial culture were mixed and incubated at 22 °C for 5 min prior to measuring luminescence (FilterMax F5, Molecular Devices). Standard curves for OD₆₀₀ vs. relative luminescence units (RLU) were used to determine the linear range and final cultures were diluted as necessary prior to measurement. Percent viability was determined as above for M. smegmatis.

Viability by enumeration of colony forming units

Overnight cultures of M. smegmatis were grown as for the autoluminescence assays and subcultured to OD₆₀₀ 0.1 in 96-well plates in Modified Sauton’s medium containing 100 μM compound, 10 μM rifampicin, or DMSO vehicle. After 18.5 h incubation at 37 °C with shaking, serial dilutions (1 × 10⁻⁵, 2 × 10⁻⁶, and 4 × 10⁻⁷) were plated on 7H9 or 7H11 Middlebrook agar containing 10% ADC, 0.5% glycerol, and 0.05% Tween 80. As a t = 0 control, 100 μL of serially diluted starter culture (1 × 10⁻⁴, 2 × 10⁻⁵, and 4 × 10⁻⁶) was also plated. Colony forming units (CFU) were enumerated after three days incubation. Percent viability was calculated as CFU(+cmpd)/CFU(+vehicle) x 100 to quantify the effect of compound treatment. All statistical analyses were performed using GraphPad Prism Version 8.3.

Immunoblotting

M. smegmatis lysates (20 μg total protein per well) and 1 μg purified NA-LprG as a positive control were analyzed by immunoblotting using anti-GroEL (1:500; sc-58170, Santa Cruz Biotechnology), anti-LprG (undiluted culture supernatant isolated from the Clone B hybridoma cell line, kind gift of Karen Dobos; corresponds to NR-51133, BEI Resources), and IRDye 800CW goat anti-mouse (1:15,000; 926–32210, LI-COR) antibodies. Blots were imaged using an Odyssey Clx scanner (LI-COR) and fluorescence was quantified using Image Studio (Ver 5.2, LI-COR). Normalized integrated signal for LprG versus the loading control GroEL was calculated as F(LprG) / F(GroEL).

ABBREVIATIONS

Ac₁PIM₂

phosphaptidyl inositol dimannoside

CETP

cholesterol ester transfer protein

CFU

colony forming units

EC₅₀

effective concentration at 50% occupancy

FABP

fatty acid binding protein

FPS

footprint similarity

K_d

dissociation constant

Llp

lipid-binding lipoprotein

MLR

multiple linear regression

Mtb

Mycobacterium tuberculosis

MTP

microsomal triacylglyceride transport protein

NBD-SA

12-N-methyl-(7-nitrobenz-2-oxa-1,3-diazo) aminostearic acid

RLU

relative luminescence units

TAG

triacylglyceride