Structural basis for the broad substrate specificity of two acyl-CoA dehydrogenases FadE5 from mycobacteria

Significance

The lipid content accounts for approximately 60% of the dry weight of the cell wall of pathogenic mycobacteria with Mycobacterium tuberculosis having more than 250 genes involved in fatty acid metabolism. Previous studies have shown that acyl-CoA dehydrogenase (ACD), which introduces unsaturation into fatty acids, exhibits strict substrate specificity toward different CoA thioester groups. Here, we identified a unique ACD member in mycobacteria that exhibits broad substrate specificity. Furthermore, we determined crystal structures of the enzyme and enzyme–substrate complexes to explain the broad substrate recognition observed in this system. Given the importance of FadE5 in fatty acid metabolism, these new structures are excellent platforms for rational structure based antituberculosis drug discovery.

Abstract

FadE, an acyl-CoA dehydrogenase, introduces unsaturation to carbon chains in lipid metabolism pathways. Here, we report that FadE5 from Mycobacterium tuberculosis (MtbFadE5) and Mycobacterium smegmatis (MsFadE5) play roles in drug resistance and exhibit broad specificity for linear acyl-CoA substrates but have a preference for those with long carbon chains. Here, the structures of MsFadE5 and MtbFadE5, in the presence and absence of substrates, have been determined. These reveal the molecular basis for the broad substrate specificity of these enzymes. FadE5 interacts with the CoA region of the substrate through a large number of hydrogen bonds and an unusual π–π stacking interaction, allowing these enzymes to accept both short- and long-chain substrates. Residues in the substrate binding cavity reorient their side chains to accommodate substrates of various lengths. Longer carbon-chain substrates make more numerous hydrophobic interactions with the enzyme compared with the shorter-chain substrates, resulting in a preference for this type of substrate.

Tuberculosis (TB) is a leading cause of human fatalities, with new therapies urgently needed to help combat this disease. The cell wall biosynthesis pathway in Mycobacterium tuberculosis (Mtb) is a well-established source of molecular targets for drug development (1). Several commonly used anti-TB drugs are known to inhibit the biosynthesis of cell wall components. For example, isoniazid and ethionamide are inhibitors of mycolic acid synthesis, and ethambutol and cycloserine are inhibitors of arabinogalactan and peptidoglycan synthesis, respectively (2). Structures of several drug targets that are involved in the biosynthesis and transport of cell wall components, such as arabinosyltransferases and MmpL3, have been determined recently (3, 4). The arabinosyltransferases EmbA, EmbB, and EmbC are regarded as the targets for ethambutol. MmpL3 is responsible for the transport of mycolic acids. SQ109, a promising anti-TB drug, binds inside MmpL3. Structural information of drug targets would definitely promote the structure-guided drug design.

The M. tuberculosis cell wall is composed of a thick lipid coat, with the lipid content accounting for ∼60% of the dry weight of the cell wall. Its presence contributes to the virulence of M. tuberculosis (5, 6). Previous studies have shown that pathogenic mycobacteria can use fatty acids as their carbon source (7). Significantly, cholesterol alone can be used as the only carbon source (8, 9). In the M. tuberculosis genome, more than 250 genes are identified to be involved in lipid, fatty acid, and sterol metabolism (10). The pathway by which lipids are metabolized is therefore of great significance in understanding the growth and virulence of M. tuberculosis in the host.

In lipid metabolism pathways, acyl-CoA dehydrogenase (ACD) introduces unsaturation into fatty acids, with the cofactor acyl-CoA converting to enoyl-CoA (11). Although there are 35 annotated fadE genes in M. tuberculosis, not all are functional ACDs. Two FadE proteins encoded by adjacent fadE genes are heterotetrameric. In those examples, a heterotetramer is required to bind the FAD cofactor (12). The existence of six heterotetramers, fadE23-fadE24, fadE28-fadE29, fadE26-fadE27, fadE31-fadE32, fadE31-fadE33, and fadE17-fadE18, encoded by fadE genes have been reported. Most of these FadE enzymes characterized to date in M. tuberculosis function in cholesterol catabolism and play roles in the dehydrogenation of cholesterol substrates through β-oxidation (13). FadE28-FadE29 catalyzes the dehydrogenation of 3-oxo-4-pregnene-20-carboxyl-CoA (3-OPC-CoA) (14), while FadE26-FadE27 catalyzes the oxidation of 3-oxo-cholest-4-en-26-oyl CoA (13) and FadE34 is active toward steroid CoA esters containing five-carbon chains (15). FadE14, characterized as a novel acyl–acyl carrier protein dehydrogenase rather than an acyl-CoA dehydrogenase, is involved in the mycobactin biosynthesis pathway (16), and FadE9 is an ACD whose substrates are branched chain amino acids (17).

In a recent study, 84 drug-resistance–associated genes were identified through genome sequencing of 161 different M. tuberculosis isolates (18). Among these, several well-known drug-target genes (gyrA, ethA, rpoB, rpoC, embB, pncA, katG, thyA, and rpsL) were identified. However, other genes, including fadE5, were also identified. The fadE5 gene is located on a cluster identified as the fadE5-gap–like genomic region (19), which harbors fadE, pks, papA, fadD, pE, mmpL, and gap-like genes, and is broadly distributed across mycobacteria. In Mycobacterium smegmatis (Ms), these gene products are suggested to play roles in the biosynthesis of trehalose polyphleates, a family of trehalose-based lipids (19). MsFadE5, acting as an ACD, is suggested to introduce unsaturation into the fatty acyl chain produced from the FAS I (fatty acid synthase I) system, and the chain is then extended and modified by polyketide synthases (Pks) (20, 21). In M. tuberculosis, similar gene clusters have also been observed. Sulfolipids (SL) and polyacyltrehaloses (PAT) are both lipids in the cell wall of M. tuberculosis. Gene clusters containing pks, papA, fadD, and mmpL are also involved in the formation and export of SL and PAT (22, 23). FadE5 proteins are also suggested to play roles in the biosynthesis of the lipids in the cell wall.

ACDs are identified by the specific substrates they utilize. Many ACD members involved in the metabolism of lipids and amino acids have been characterized, including short-chain acyl-CoA dehydrogenase (SCAD), medium-chain acyl-CoA dehydrogenase (MCAD), long-chain acyl-CoA dehydrogenase (LCAD), very long-chain acyl-CoA dehydrogenase (VLCAD), isovaleryl-CoA dehydrogenase (IVD), isobutyryl-CoA dehydrogenase (IBD), branched short chain acyl-CoA dehydrogenase (BSCAD), and glutaryl-CoA dehydrogenase (GDH) (24). In eukaryotes, different ACD members exhibit strict substrate specificities toward different CoA thioester groups, which results from their distinct active-site architectures. In those ACDs that catalyze the dehydrogenation of linear acyl-CoA, the length of the substrate carbon chain that the enzyme can accept is limited by the length of the substrate binding cavity. The cavity lengths of SCAD, MCAD, and VLCAD are 8, 12, and 24 Å, and they have optimal specificities for fatty acyl-CoAs with chains that contain 4, 8, and 16 carbon atoms, respectively (25–27). Most ACDs form homotetramers composed of four ∼43-kDa subunits. Two exceptions are VLCAD, which is a homodimer consisting of two ∼67-kDa subunits (27, 28). The other is the α₂β₂ heterotetramer encoded by fadE genes in M. tuberculosis (13, 14).

Herein, we isolated FadE5 from M. tuberculosis (MtbFadE5) and M. smegmatis (MsFadE5). MsFadE5 has been shown to be involved in resisting drug uptake, most likely by affecting cell wall permeability. Both MtbFadE5 and MsFadE5 show broad selectivity for linear acyl-CoA substrates with chains varying in lengths from 4 to 22 carbons. Importantly, its substrate specificity is quite different from other homologs, most of which have a strict substrate requirement. To understand this unusual property, we determined the crystal structures of MtbFadE5 and MsFadE5 in complex with various substrates and MtbFadE5 in complex with C18CoA. These structures provide the molecular basis for the broad substrate selectivity of MtbFadE5 and MsFadE5.

Results

Drug Resistance Assays of MsFadE5.

To verify the roles of FadE5 in cell wall lipid biosynthesis, drug-resistance assays were performed (Fig. 1). In this experiment, the fadE5 gene in the M. smegmatis genome was disrupted by substitution of an internal segment with a hygromycin-resistance gene. Verification of gene disruption was performed by PCR followed by sequencing of the PCR products. The MsfadE5 gene was overexpressed in the fadE5-deleted cells for complementation studies. Fig. 1A shows the growth rate is the same for the fadE5-deleted cells and the wild-type cells. When treated with ethambutol (2 μg mL⁻¹) and streptomycin (0.1 μg mL⁻¹), first-line and second-line drugs, cells lacking fadE5 exhibited decreased drug-resistance when compared with wild-type. When the fadE5 gene was overexpressed in the fadE5-deleted cells, drug resistance to ethambutol was gained and drug resistance to streptomycin was even more evident (Fig. 1 B and C).

Fig. 1.

Drug-resistance assays in M. smegmatis. (A) Growth curves for wild-type M. smegmatis [Ms (WT)], MsfadE5-deleted strain [Ms (ΔfadE::hyg)], and MsfadE5-deleted strain with overexpression of MsfadE5 [Ms (pVV16::fadE5)]. (B) Growth curves of wild-type M. smegmatis, MsfadE5-deleted strain, and MsfadE5-deleted strain with overexpression of MsfadE5 in the presence or absence of 2 μg mL⁻¹ ethambutol. (C) Growth curves of wild-type M. smegmatis, MsfadE5-deleted strain and MsfadE5-deleted strain with overexpression of MsfadE5 in the presence or absence of 0.1 μg mL⁻¹ streptomycin. Each bar is the mean and SD of three measurements.

Substrate Selectivity of MtbFadE5 and MsFadE5.

Full-length MtbFadE5 and MsFadE5 were produced in E. coli for analysis. Gel filtration revealed that they both form homodimers in solution. This was further confirmed by sedimentation equilibrium analytical ultracentrifugation (AUC) (SI Appendix, Fig. S1 B and C). A sequence alignment between MtbFadE5, MsFadE5, and other ACDs is shown in SI Appendix, Fig. S1A. ACDs are classified according to the substrates they utilize. Most ACDs form homo- or heterotetramers composed of four ∼43-kDa protomers. Only VLCAD is a homodimer consisting of two ∼67-kDa protomers. FadE5 is also a homodimer consisting of two 66-kDa protomers. So, we inferred that FadE5 is a VLCAD at first. However, our assays showed that MtbFadE5 and MsFadE5 are both highly active toward substrates of varying lengths of linear acyl-CoA, including C4CoA, C6CoA, C10CoA, C12CoA, C16CoA, and C18CoA (Fig. 2 A and B). Isothermal titration calorimetry (ITC) was next performed to determine their binding affinities with the substrates. The catalytic base for most ACDs characterized to date is a glutamate residue. In SCAD it is E368 (29), while in MCAD it is E376 (30) and in HsVLCAD it is E422 (27). Sequence alignment of MsFadE5 and MtbFadE5 with other ACDs suggest that E447 is the catalytic base in MsFadE5 and MtbFadE5. To confirm this assumption, we expressed and purified E447A mutants for both enzymes and showed neither possessed activity (Fig. 2 A and B), confirming this residue is the catalytic base required for dehydrogenation. ITC was also performed using samples of MtbFadE5 E447A and MsFadE5 E447A. The results showed that the binding affinities increased with the length of the carbon chain (Fig. 2 C and D and SI Appendix, Fig. S2). C22CoA binds to MtbFadE5 with a K_d value of 0.16 μM, while C6CoA binds to MtbFadE5 with a K_d value of 196 μM. Similar values were obtained for the binding affinities of MsFadE5 with these substrates (Table 1). Thus, both MtbFadE5 and MsFadE5 show a broad substrate selectivity for acyl-CoAs but tend to prefer the longer rather than shorter substrates. The kinetic parameters (k_cat and K_m) for representative substrates were also determined (Table 2 and SI Appendix, Fig. S3). The free concentration of acyl-CoA esters is unlikely to exceed 200 nM under normal cellular conditions (31). Therefore, the effectiveness of FadE5 in cells using these substrates would be low. This may be a factor that contributes to the slow growth rate of M. tuberculosis and M. smegmatis.

Fig. 2.

Activity assays for MtbFadE5 and MsFadE5 and ITC studies. (A and C) Activity of MtbFadE5 with C4CoA, C6CoA, C10CoA, C12CoA, C16CoA, and C18CoA. The MtbFadE5 E447A mutant is inactive with the C18CoA substrate. ITC was used to determine interactions between MtbFadE5 C10CoA and C20CoA. (B and D) Activity and ITC data for MsFadE5.

Table 1.

The K_d values determined by ITC

	Substrate	Kd (μM)
MtbFadE5	C6-CoA	196 ± 78
	C10-CoA	67.1 ± 16.0
	C12-CoA	60 ± 9.2
	C16-CoA	4.4 ± 0.67
	C18-CoA	4.18 ± 0.82
	C20-CoA	0.19 ± 0.07
	C22-CoA	0.16 ± 0.06
MsFadE5	C6-CoA	316 ± 274
	C10-CoA	20.1 ± 1.3
	C12-CoA	24.0 ± 7.7
	C16-CoA	3.45 ± 0.31
	C18-CoA	0.72 ± 0.20
	C20-CoA	0.07 ± 0.05
	C22-CoA	0.19 ± 0.14

Table 2.

Steady-state kinetic parameters for MsFadE5 and MtbFadE5

Enzyme	Substrate	kcat (s−1)	Km (μM)	kcat/Km (M−1 s−1)
MsFadE5	C4CoA	0.53 ± 0.06	285.9 ± 53.1	(1.85 ± 0.13) × 103
MtbFadE5	C4CoA	1.07 ± 0.12	358.7 ± 71.8	(2.98 ± 0.26) × 103
MtbFadE5	C6CoA	0.80 ± 0.14	353.0 ± 105.6	(2.27 ± 0.28) × 103
MtbFadE5	C18CoA	0.61 ± 0.04	162.5 ± 28.8	(3.75 ± 0.42) × 103

Overview of the MsFadE5 Structure.

To investigate the reasons for the broad substrate selectivity of MtbFadE5 and MsFadE5, crystal structures of both enzymes were determined. Initially, we solved the structure of MsFadE5 (80% sequence identity with MtbFadE5). The crystal structure shows that MsFadE5 forms a homodimer (Fig. 3A). Virtually all residues could be fitted to the electron density map except for the region from 482 to 489. The N terminus of MsFadE5 consists of 467 residues and shares a canonical fold that is observed in other tetrameric ACDs. The additional residues, 467 to 611, form an α-helical domain at the C terminus that is not found in other tetrameric ACDs. The C-terminal helical domain exists specifically in VLCADs. In human VLCAD the C-terminal domain is responsible for binding to the inner mitochondrial membrane (32, 33). Residues 446 to 478 are disordered in that structure and are thought to be responsible for anchoring to the membrane. This is based on the observation that FadE5 was previously identified in a membrane fraction of M. tuberculosis (34). In FadE5, residues 482 to 489 are disordered and located in the surface of the C-terminal domain, suggesting that this region may be also involved in membrane anchoring and may only become ordered when interacting with the membrane.

Fig. 3.

Overall structure of apo-MsFadE5 and its active site. (A) Cartoon representation and Connolly surface (transparent) of the MsFadE5 dimer. The two subunits are shown in red and blue. FAD is shown in stick models with yellow carbon atoms. (B) 2F_o–F_c electron density map for FAD contoured at 1.0σ. (C) Surface representation (transparent) of the putative substrate binding cavity, with FAD shown as a stick model.

Active Site and Substrate Binding Cavity.

A FAD cofactor is present in each of the MsFadE5 active sites (Fig. 3B). These are located at the dimer interface. In all of the structurally characterized ACDs, there is a cavity close to the isoalloxazine of the bound FAD, which harbors the carbon chain of the substrate. The carbon chain is located proximal to the N5 atom of the isoalloxazine ring since it is involved in the α-, β-desaturation of the substrates. The length of the cavity limits the length of the substrate carbon chain. In SCAD (PDB ID code 1JQI), MCAD (PDB ID code 3MDD), and HsVLCAD (PDB ID code 2UXW), the length of their cavities are 8, 12, and 24 Å, respectively (27). However, the cavity length in MsFadE5 is approximately only 12.5 Å (Fig. 3C). The pocket therefore appears to be not long enough to accommodate substrates with long carbon chains. Thus, there must be conformational changes when the longer substrates bind.

Structural Basis for the Broad Substrate Selectivity of MsFadE5 and MtbFadE5.

The structures of the other reported ACDs in complex with substrates were solved by adding substrates to the native enzymes prior to crystallization. In our efforts to crystallize enzyme–substrate complexes, we followed the same strategy. However, we were only able to generate crystals where the product had been released. Superimposition of this structure and the apo-MsFadE5 structure reveals some differences (SI Appendix, Fig. S4B). In particular, the side chains of residues F294 and D295 exhibit conformational changes. The side chain of E447, the catalytic base, is reoriented to face to the isoalloxazine ring of FAD. R301, which electrostatically interacts with E447 in apo-MsFadE5, also undergoes a conformational change. In this structure, the electron density for Y446 shows that it adopts two different conformations: One conformation is the same as the apo form and the other conformation widens the binding cavity for substrates to bind (SI Appendix, Fig. S4A).

To prevent the release of the substrate, we obtained structures of MsFadE5 where E447 was mutated to alanine. Crystals could then be obtained in complex with the substrates C4CoA, C6CoA, C8CoA, C10CoA, C12CoA, C14CoA, C16CoA, C17CoA, C18CoA, C20CoA, and C22CoA (Fig. 4 and SI Appendix, Fig. S5). The structure of the E447A mutant of MsFadE5 in complex with C18CoA provides an example to explore the structural basis of the broad substrate specificity. In total, there are 11 hydrogen bonds between residues in the protein and the CoA portion of C18CoA (Fig. 5B). Conformational change in the side chain of R301 is needed to allow the formation of a hydrogen bond and the movement of F294 results in a π–π stacking interaction (3.6 Å) with the adenine of the CoA (Fig. 5C). This phenylalanine is not conserved among other ACDs, as indicated in the sequence alignment (SI Appendix, Fig. S1). The equivalent aromatic amino acid is only found in GDH (PDB ID code 3MPI). However, no similar π–π stacking interaction could be found in that protein. Thus, this is a feature specific to this enzyme.

Fig. 4.

Structures of the E447A mutants of MsFadE5 in complex with substrates. (Left) Cartoon representation (green) of the E447A mutant of MsFadE5 in complex with C4CoA (stick model with pink carbon atoms). (Right) 2F_o–F_c electron density maps (contour level = 1.0σ) for the substrates.

Fig. 5.

Interactions between C18CoA and residues in the E447A mutant of MsFadE5. (A) Cartoon representation of the structure. C18CoA is shown as a stick model with carbon atoms colored pink. (B) Hydrogen bonds between C18CoA and the residues in MsFadE5 are shown as dotted lines. Distances are given in Ångstroms. (C) The residues that interact with the protein with reoriented side chains. (D) Superimposition of the CoA binding sites in the HsVLCAD–C14CoA complex (yellow) and the E447A mutant of MsFadE5 in complex with C18CoA (green). C18CoA is shown as a stick model with carbon atoms colored pink.

To clarify the contribution of the interacting residues, mutational analysis was carried out and the K_d values measured by ITC (SI Appendix, Fig. S6 and Table S1). For the S171A, K225A, and R460A mutants using C20CoA, K_d values were several folds lower than for the wild-type enzyme. The D456A_R460A double mutant also showed a decreased affinity compared to the R460 single mutant. So, these residues are suggested contributed to the binding affinity of the protein and substrate. However, the R301A and F294A mutants both showed an increased binding affinity for C20CoA. We speculate that it would be easier for the ligand to enter into the pocket in the absence of these two bulky side chains. Notably, S171, K225, D456, and R460 could interact with the substrate without any conformational changes. However, R301 and F294 both undergo conformation changes upon substrate binding. The interactions between the two residues and the substrate are facilitated through the conformational changes. So, we infer that these two residues may not contributed to the original recognition of the substrate but the stabilization after substrate binding.

As suggested above, in the complexes, the cavity can be lengthened as longer substrates bind (Fig. 6). This is achieved by the residues along the cavity reorienting themselves upon substrate binding. In the MsFadE5–C4CoA structure, the side chain of Y446 rotates to allow binding (Fig. 6A). Thus, Y446 plays a type of gate keeper role since in both enzymes, this residue has to rotate to allow the substrate to gain access. In the MsFadE5–C10CoA model, the side chain of M134 changes conformation to allow C10CoA binding (Fig. 6B). However, in the MsFadE5–C12CoA and MsFadE5–C20CoA complexes, M134 undergoes a larger conformational change so that the cavity is completely open (Fig. 6 C and D). The entire length of the cavity is hydrophobic so that the longer carbon chains of the substrate make additional hydrophobic interactions, resulting in the preference of FadE5 for long-chain substrates. All in all, the side chains of the residues along the cavity wall guide the route and binding of the carbon chain. The side chain of M130 in MsFadE5 stretches into the cavity, forcing the carbon chain to take a turn (SI Appendix, Fig. S7). In MtbFadE5, this residue is substituted by a glycine. A structure of the E447A_M130G double mutant of MsFadE5 in complex with C16CoA shows that the carbon chain takes a different path compared to the single E447 mutant (SI Appendix, Fig. S7).

Fig. 6.

Conformational changes in the substrate binding cavities of the E447A mutant of MsFadE5 upon substrate binding. (A, Left) Cut-away surface representation of the substrate binding cavity in the C4CoA complex. C4CoA is shown as a stick model with carbon atoms colored pink. (Right) Conformational changes for the residues along the substrate binding cavity wall (green) compared with those in the apo structure (blue). (B) Conformational changes in the C10CoA complex. (C) Conformational changes in the C12CoA complex. (D) Conformational changes in the C20CoA structure.

We were unsuccessful in generating apo-MtbFadE5 crystals. However, we did make crystals of the E447A mutant of MtbFadE5 in complex with C18CoA. This complex is also a homodimer (Fig. 7A). The rmsd for all Cα atoms between the MtbFadE5 E447A–C18CoA and MsFadE5 E447A–C18CoA complexes after superimposition is only 0.327 Å, demonstrating the two structures are virtually identical (Fig. 7B). The electron density for the CoA portion of the substrate is strong. However, although the density for the carbon chain is poor, it could be reasonably fitted in the 2F_o–F_c map (Fig. 7 C and D).

Fig. 7.

Structures of the E447A mutant of MtbFadE5 in complex with C18CoA and its substrate binding cavity. (A) Cartoon representation of the dimer. Individual subunits are shown in yellow and red. (B) The 2F_o–F_c electron density map for C18CoA contoured at 1.0σ. (C) Equivalent 2F_o–F_c electron density map for C18CoA contoured at 0.6σ. (D) Superimposition of the MsFadE5 complex (green) and MtbFadE5 complex (red and yellow). (E) Superimposition of the substrate binding cavities in the MsFadE5 complex (green) and the MtbFadE5 complex (yellow). Residues along the substrate binding cavity walls that differ between the two structures are shown in stick models. The C18CoA in MsFadE5 is in pink, and C18CoA in MtbFadE5 is in blue.

The hydrogen bonds between the MtbFadE5 and the C18CoA are similar to those observed for MsFadE5 and its substrate (SI Appendix, Table S2). The π–π stacking between the F294 and the adenine of substrate is also present. The residues that have flexible side chains in the substrate binding cavity in the MsFadE5 complex adopt similar conformations in the MtbFadE5 complex, suggesting these may undergo similar conformational changes when the substrate enters the binding cavity. Thus, it is reasonable to suggest that the structural basis of the broad substrate selectivity of MtbFadE5 is the same as that of MsFadE5. However, there are several amino acid substitutions in the binding cavity in MtbFadE5 compared to MsFadE5. These are M134 and F126 in MsFadE5, which are substituted by F134 and W126 in MtbFadE5, respectively (Fig. 7E). SI Appendix, Fig. S8 shows how these changes impact on the binding mode of C18CoA. The size and orientation of these residue side chains are totally different, which might shape the binding pockets and decide the conformation of the carbon chain of C18CoA.

Discussion

To date, it has always been suggested that ACDs have strict substrate specificities and SCAD, MCAD, LCAD, and VLCAD are specifically active against fatty acyl-CoAs with 4, 8, 14, and 16 carbons in length, respectively (25–27). The size of the substrate binding pocket of ACDs determines the maximum length of the carbon chain that the protein can accommodate. The width and length of the cavity are restricted by the residues along the cavity wall. The length of the ACDH-11 binding cavity is 14 Å, which restricts the carbon length to 12. Kinetic assays also indicate that ACDH-11 has little activity against substrates with chains containing more than 12 carbon atoms. The cavity length is limited by the presence of Y344 and L159. The structure of the apo-form of ACDH-11 and the structure of ACDH-11 in complex with C11CoA display the same conformations for Y344 and L159. The temperature factors of these two residues are relatively low compared to all of the residues in ACDH-11 (28). These data revealed that the binding cavity of unbound ACDH-11 is no different to that when ACDH-11 is bound to the substrates. Similarly, in SCAD, I251 limits the length of the substrate binding cavity (29). In MCAD, Q95 and E99 limit the length of the substrate binding cavity (30). In the present MsFadE5 structures in complex with substrates, conformational changes of the residues limiting the binding cavity could occur to provide enough room for longer substrates to bind. Furthermore, in the MsFadE5 E447A–C22CoA structure, the cavity extends to the surface of the enzyme suggesting there is more room for substrates longer than C22CoA to bind.

HsVLCAD shows little activity for substrates with chain lengths of fewer than 12 carbons (26). In the HsVLCAD–C14CoA complex structure, only the electron density for the carbon chain portion of the substrate is observed clearly. The missing density of the CoA portion of the substrate is most likely caused by the high mobility of the CoA moiety. The absence of hydrogen bonds between HsVLCAD and CoA may result in the high mobility (27). When C11CoA in the ACDH-11–C11CoA complex structure is superimposed with the HsVLCAD–C14CoA structure, only one hydrogen bond is observed. SCAD and MCAD form five and four hydrogen bonds with the CoA of the substrate, respectively (28). In HsVLCAD, the binding affinity between the enzyme and the substrate is determined mainly by hydrophobic interactions between the enzyme and the long acyl chains. Therefore, HsVLCAD not only accepts acyl-CoAs with long acyl chains as a substrate but also shows a preference for them. In the MsFadE5–C18CoA complex, there are 11 hydrogen bonds and a rare π–π stacking interaction between F294 and the adenine of the substrate. These additional bonds seem to be the reason why FadE5 can accept substrates with short carbon chains. Although the hydrophobic interactions are not strong, the CoA portion of the shorter substrates can bind to the enzyme. To explore what precludes the analogous hydrogen bonds, the structures of MsFadE5 and HsVLCAD were compared. In HsVLCAD, the β-sheet containing strands 4 and 5 extends further away from the CoA binding site than that in MsFadE5, and in MsFadE5, the strands in the sheet are much shorter and face toward the CoA portion of the substrate so the residues can interact with CoA (Fig. 5D). Taken together, the apo-FadE5 and FadE5 complex structures reveal the structural basis for its broad substrate selectivity. This enzyme can accept short carbon chain acyl-CoA as a substrate since it can form several hydrogen bonds and also has a unique π–π stacking with the CoA portion of the substrate. In addition, FadE5 can accommodate longer-chain fatty-acyl CoAs since the residues that potentially limit the length of the substrate binding cavity can reorient their side chains.

FadE5 is a homodimeric ACD whose substrates are linear chain fatty acids. M. tuberculosis can make use of fatty acids of different lengths as a carbon source. These fatty acids are degraded through the β-oxidation pathway to provide the energy required for the growth of M. tuberculosis. In the first step of the β-oxidation pathway, ACD introduces unsaturation into the fatty acids, converting acyl-CoA into enoyl-CoA. FadE5 is probably involved in the degradation of fatty acids of varying lengths. In mycobacteria, the FAS I system produces C14-C26 acyl-CoAs through the de novo fatty acid biosynthesis, then unsaturation and modifications would be introduced into the acyl-CoAs. In some nontuberculosis mycobacteria, such as M. smegmatis, fadE5 locates in the gene clusters that are responsible for biosynthesis of polyketide lipids, so FadE5 might also be involved in introducing unsaturation into products from FAS I system, and the chain is then extended by Pks. Here, we report that FadE5s use C4CoA to C22CoA as substrates. In the FadE5–C22CoA complex structures, we have deduced that the substrate binding cavity could accommodate substrates longer than C22CoA. Similar findings have been observed for the M. tuberculosis β-ketoacyl-acyl carrier protein (ACP) synthase III (mtbFabH) (35). MtbFabH is a link between the FAS I and FAS II systems of M. tuberculosis and catalyze a decarboxylative condensation in the first step of FAS II processing. Escherichia coli FabH and other typical FabHs utilize C2-C6 acyl-CoA as substrates, while MtbFabH can tolerate a wide range of C12-C20 acyl-CoA substrates.

The substrate selectivity of FadE5 is in accordance with its role that we predict in introducing unsaturation into the products of FAS I. Moreover, the products can be incorporated as phospholipids into the membrane, or elongated by FAS II for the synthesis of mycolic acids, or elongated by specific PKSs to form other complex lipids like mycoketides, SLs, and phthiocerol dimycocerosate, or synthesize storage compounds such as triacylglycerides (36–39). As a result, it is suggested that FadE5 can play multiple roles in lipid metabolism in M. tuberculosis. The lipids in the cell wall of M. tuberculosis and M. smegmatis have been shown to be involved in formation of the cell wall permeability barrier. Here, drug-resistance assays were performed on wild-type M. smegmatis, fadE5-deleted M. smegmatis, and fadE5 overexpressed M. smegmatis. The fadE5-deleted M. smegmatis showed an obvious decreased resistance to ethambutol and streptomycin. Since we inferred that fadE5 plays an important role in the biosynthesis of the lipids in the cell wall, the decreased drug resistance of the fadE5-deleted strains may have resulted from the increased cell wall permeability.

In total, 35 FadE members have been identified in M. tuberculosis; such a large number of fadE genes in the genome reflects the necessity of M. tuberculosis to load a large variety of lipids. Thus, the development of small-molecule inhibitors against this and closely related enzymes have strong potential to be developed as new generation multitarget drugs. This study therefore advances the understanding of fatty acid metabolism in mycobacteria, and the availability of the structure and characterization information of FadE5 should facilitate the discovery and development of new antimycobacterial agents.

Materials and Methods

Protein Expression and Purification.

The genes encoding MtbFadE5 and MsFadE5 are Rv0244c and MSMEG_0406, and were PCR-amplified from M. tuberculosis H37Rv and M. smegmatis mc²155, respectively. The gene fragments were then fused into the pET28a expression vector with the EcoR I and Hind III sites by homologous recombination using the One Step Cloning Kit to produce an N-terminal His-tagged construct. Mutation was introduced into MtbFadE5 and MsFadE5 by the fast mutagenesis system from TransGen. The fused vectors were all transformed into E. coli BL21 Rosetta cells for expression. Protein expression in LB media supplemented with kanamycin (50 mg/mL) at 37 °C was induced by 0.5 mM isfopropyl-β-d-thiogalactopyranoside (IPTG) when the OD₆₀₀ reached 0.6. Cultures were then grown at 16 °C for 18 h. Cells were harvested by centrifugation and resuspended in MCAC-0 buffer (25 mM Tris⋅HCl, pH 8.0, 500 mM NaCl, and 10% glycerol), followed by lysis using sonication. The debris was removed by centrifugation at 18,000 rpm for 40 min. Supernatants containing the target proteins were loaded twice onto a Ni-NTA column (GE Healthcare) equilibrated with MCAC-0 buffer. The column was subsequently washed with MCAC-20 buffer (MCAC-0 supplemented with 20 mM imidazole). Then, the target proteins were then eluted with MCAC-500 buffer (MCAC-0 supplemented with 500 mM imidazole) and further purified using a HiTrap Q column (GE Healthcare) for anion exchange chromatography. For the final step, a Superdex G200 column (GE Healthcare) equilibrated with 25 mM Tris HCl and 150 mM NaCl, pH 8.0, was used to purify the target proteins by gel-filtration chromatography. The eluted proteins were evaluated by SDS/PAGE. Selenomethionine-substituted MsFadE5 protein was expressed in methionine auxotrophic E. coli strain B834 (DE3) in M9 medium, as previously described (40).

Drug-Resistance Assay.

M. smegmatis strains were cultured in 7H9 broth supplemented with 10% (vol/vol) ADC (1,000-mL solution contained 9 g NaCl, 50 g BSA, and 20 g glucose), 0.1% (vol/vol) Tween-80, 0.5% (vol/vol) glycerin, and 5 μg mL⁻¹ carbenicillin. Cells were grown at 37 °C to an OD₆₀₀ of 0.6∼0.8. Then the cells were diluted to an OD₆₀₀ of 0.1. The OD₆₀₀ was measured using 96 area plate (Corning) in the presence or absence of drugs. Data analyses were performed using GraphPad Prism 6.0.

Analytical Ultracentrifugation.

AUC was carried out using a Beckman Coulter Optima AUC equipped with an An60Ti rotor to determine the sedimentation velocities of MtbFadE5 and MsFadE5. The proteins were diluted in a solution containing 25 mM Tris, pH 8.0, and 150 mM NaCl. The samples were concentrated to 1 mg/mL. Sedimentation velocity data were collected at 42,000 rpm. A total of 99 scans were collected every 4 min using interference optics. All measurements were conducted at 4 °C. The data were analyzed using the program Sedfit with a continuous c(s) distribution model (41).

ITC.

The binding affinities of MtbFadE5 and MsFadE5 for C4CoA, C6CoA, C10CoA, C12CoA, C16CoA, C18CoA, C20CoA, and C22CoA were determined using an ITC-200 microcalorimeter (Microcal). All measurements were performed at 25 °C. The measurements were performed using the E447A mutants of MtbFadE5 and MsFadE5. Protein samples at 0.04∼0.10 mM in buffer containing 25 mM Tris (pH 8.0) and 150 mM NaCl were loaded in the reaction cell. For each binding assay, substrates at a concentration of 0.4∼0.90 mM in the same buffer were titrated into the protein samples with a syringe. The titration consisted of 19 injections of 2.0 μL every 120 s (except for the first injection of 0.4 μL).

Enzyme Assays.

The activities of MtbFadE5 and MsFadE5 were measured based on the reduction of ferrocenium as an electron acceptor at an absorbance at 300 nm (extinction coefficient of 2.75 mM⁻¹ cm⁻¹) at 25 °C within 1 min of the initiation of the assay (42, 43). The reaction mixture (100 μL) contained 25 mM Tris, pH 8.0, 150 mM NaCl, 1 μM protein, and 0.5 mM ferrocenium hexafluorophosphate. The assays were initiated by the addition of 0.4 mM of varying substrates including C4CoA, C6CoA, C10CoA, C12CoA, C16CoA, and C18CoA. The activities of the E447A mutants were measured in the same way. The kinetic parameters of FadE5 were determined by measuring the initial velocities at a pH of 8.0 and a temperature of 25 °C. The 100-µL reaction mixture was prepared by adding 1 mM ferrocenium hexafluorophosphate, diluting different substrates to the desired concentration. The reaction was initiated by adding enzyme to a final concentration of 0.8 µM. Initial velocities were examined at substrate concentrations from 50 to 500 µM. Michaelis–Menten plots were used to determine the kinetic parameters.

Crystallization and Data Collection.

Crystals were grown at 20 °C by vapor-diffusion mixing 1 μL of protein solution and 1 μL of reservoir solution. Selenomethionine-substituted MsFadE5 crystals were obtained in 0.19 M magnesium formate, 15.6% (wt/vol) PEG 3350, 2% (vol/vol) 2-propanol, 20 mM MES (pH 6.0), and 40 mM Ca(OAc)₂. A SAD (single-wavelength anomalous dispersion) dataset was collected at 100 K on beamline 17U of the Shanghai Synchrotron Radiation Facility (SSRF, China) from a cryoprotected selenomethionine-substituted MsFadE5 crystal. Crystals of native MsFadE5 were grown in 0.19 M magnesium formate, 15.6% (wt/vol) PEG 3350, 0.4 M (NH₄)₂SO₄, 0.4% (vol/vol) PEG 400, and 20 mM acetate pH 5.5. The dataset was collected on beamline BL41XU of the SPring-8 synchrotron radiation facility (Japan). Crystals of native MsFadE5 soaked with substrates were grown in 0.1 M Hepes sodium (pH 7.0), 2% vol/vol PEG 400, and 2 M (NH₄)₂SO₄. This dataset was collected on beamline 19U at the SSRF. Crystals of the E447A mutant of MsFadE5 were obtained in 0.1 M Bis-Tris (pH 6.5) and 2 M (NH₄)₂SO₄. The dataset was collected on beamline 19U at the SSRF. Crystals of the E447A mutant of MsFadE5 in complex with various substrates were obtained in 0.1 M Hepes-sodium (pH 7.0), 2% vol/vol PEG 400, 2 M (NH₄)₂SO₄ and 1 mM FAD by soaking with CoAs in the molar ratio of 1:5. These datasets were collected on beamline 18U and 19U at the SSRF. Crystals of the E447 AM130G double mutant of MsFadE5 in complex with C16CoA were obtained in 0.8 M NaH₂PO₄/1.2 M K₂HPO₄, 0.1 M acetate (pH 4.5), 1 mM FAD, and 1.2 mM C16CoA. The dataset was collected on beamline 19U at the SSRF. Crystals of the E447A mutant of MtbFadE5 in complex with C18CoA were obtained by cocrystallization with substrates at a molar ratio of 1:3. These crystals were grown in 20% (wt/vol) PEG 3350 and 2 M potassium nitrate. The dataset was collected on beamline 19U at the SSRF. The datasets were processed with the program HKL2000 (44) and XDS package (SI Appendix, Table S3).

Phasing, Model Building, and Refinement.

A SAD dataset from the selenomethionine-substituted MsFadE5 crystal was used to determine phases using the program Phenix AutoSol Wizard by the SAD method (45, 46). Twenty-three selenium atoms were located, and a figure of merit of 0.381 was obtained following the refinement of heavy atom parameters and phase calculations. Then, Phenix.autobuild was used for model building (47). Residues 482 to 489 could not be traced due to their poorly ordered electron density. The structure of apo-MsFadE5 was determined by molecular replacement with the Phaser (48) module in CCP4 (49) using the selenomethionine-substituted MsFadE5 model as a search template. The output model from molecular replacement was subsequently subjected to iterative cycles of manual model adjustment with Coot (50) and refinement with Phenix.refine (51). The structures of MsFadE5 mutants in complex with substrates were all determined by molecular replacement with Phaser using the apo-MsFadE5 structure as the search template. The structure of the MtbFadE5 mutant in complex with C18CoA was also determined using apo-MsFadE5 as a template. The output model from molecular replacement was then rebuilt with the program Phenix.autobuild. The models were further refined with Coot and Phenix.refine. All of the refined models were validated by MolProbity (52). The phasing and refinement statistics are summarized in SI Appendix, Table S3.

Data Availability.

The coordinates and structure factors have been deposited in the Protein Data Bank. The accession codes are summarized in SI Appendix, Table S3.