Structural Basis for the Recognition of Mycolic Acid Precursors by KasA, a Condensing Enzyme and Drug Target from Mycobacterium Tuberculosis *

Background: Although induced fit binding is central to molecular recognition, in most cases, the structural basis for this process remains to be elucidated. Results: We determined structural snapshots along the substrate binding reaction coordinate for the drug target Mycobacterium tuberculosis KasA. Conclusion: KasA binds acyl substrates via the induced fit concerted opening of a hydrophobic cavity. Significance: Improved understanding of induced fit binding is crucial for inhibitor design. The survival of Mycobacterium tuberculosis depends on mycolic acids, very long α-alkyl-β-hydroxy fatty acids comprising 60–90 carbon atoms. However, despite considerable efforts, little is known about how enzymes involved in mycolic acid biosynthesis recognize and bind their hydrophobic fatty acyl substrates. The condensing enzyme KasA is pivotal for the synthesis of very long (C38–42) fatty acids, the precursors of mycolic acids. To probe the mechanism of substrate and inhibitor recognition by KasA, we determined the structure of this protein in complex with a mycobacterial phospholipid and with several thiolactomycin derivatives that were designed as substrate analogs. Our structures provide consecutive snapshots along the reaction coordinate for the enzyme-catalyzed reaction and support an induced fit mechanism in which a wide cavity is established through the concerted opening of three gatekeeping residues and several α-helices. The stepwise characterization of the binding process provides mechanistic insights into the induced fit recognition in this system and serves as an excellent foundation for the development of high affinity KasA inhibitors.

Mycobacterium tuberculosis is dependent on mycolic acids, long-chain C 60 -90 ␣-alkyl-␤-hydroxy fatty acids that are essential components of the mycobacterial cell wall, which confers pathogenicity, intrinsic antibiotic resistance, and the ability to persist within the human macrophage (the structure of a specific mycolic acid, an ␣-mycolic acid, is shown as an example in the box at the bottom of Fig. 1) (1). The biosynthesis of mycolic acids requires the presence of two distinct fatty acid synthesis pathways, the mammalian-like type I (type I fatty acid synthesis (FAS-I)) 3 and the bacterial type II (FAS-II) systems (Fig. 1). The C 16 -18 acyl-CoA primers produced by the FAS-I system are elongated by the FAS-II pathway to C 50 -56 mycolic acid precursors, termed meromycolates (meromycolates are fatty acids that form the upper part of the mycolic acid depicted in the box at the bottom of Fig. 1) (2). In contrast to the FAS-I system, which consists of one large multifunctional dimer, the FAS-II system is constituted by monofunctional enzymes, which can be targeted selectively (3). After its activation within the mycobacterial cell, the first-line anti-tuberculosis drug isoniazid inhibits the enoyl-ACP reductase InhA, thereby validating the FAS-II pathway as a promising target for the development of novel antibiotics (Fig. 1) (4).
Like InhA, the KasA enzyme is also essential for mycobacterial survival and is thus an attractive drug target (5). This homodimeric ␤-ketoacyl-ACP synthase (KAS) catalyzes a Claisen condensation reaction between acyl-AcpM and malonyl-AcpM in each elongation cycle of the FAS-II pathway, during which the substrates are attached to Ser-41 of the small acidic acyl carrier protein (ACP; AcpM in M. tuberculosis) via a phosphopantetheine linker (Fig. 1) (6). The KasA catalytic triad consisting of Cys-171, His-311, and His-345 facilitates the formation of the new carbon-carbon bond via a three-step mecha-nism ( Fig. 1) (7)(8)(9)(10). During this process, the active site cysteine is acylated and subsequently attacked by the decarboxylated malonyl-AcpM to yield a ␤-ketoacyl-AcpM, which is extended by two carbon atoms. The acylated enzyme intermediate is mimicked by a C171Q KasA variant in which Gln-171 forms hydrogen bonds with the KasA oxyanion hole (11)(12)(13). This acylenzyme mimic is particularly important for future drug design studies as the natural products thiolactomycin (TLM) and platensimycin inhibit KAS enzymes by binding preferentially to the acyl-enzyme ( Fig. 1) (10,12,14). Despite its modest affinity, the antibiotic TLM has been shown to possess efficacy in Serratia marcescens and Klebsiella pneumoniae murine models of infection (15). Importantly, TLM inhibits both classes of condensing enzymes, the KAS I/II enzymes (FabB/F in Escherichia coli, KasA/B in M. tuberculosis) that participate in the fatty acid elongation cycle and the FAS-II-initiating KAS III enzymes (FabH) (16 -18). Antibacterials such as TLM, which interact with multiple targets, are thus ideal lead compounds because single-target mutations in the organism will not lead to full resistance (19).
For the development of improved TLM-based drug candidates, it is critical to understand the substrate recognition process, catalytic mechanism, and inhibition of KAS enzymes. Clearly, a certain amount of flexibility is an important requirement for FAS-II enzymes in order to bind and process their long chain hydrophobic substrates. This is supported by the obser- The FAS-I system, which also provides precursors for phospholipid biogenesis, generates C 16 -18 and C 24 -26 acyl-CoA substrates de novo (2). Subsequently, the FAS-II system elongates the C 16 -18 acyl-CoA primers to the very long C 50 -56 meromycolate chains (top part of the depicted mycolic acid; the terminal portion synthesized via FAS-I is highlighted in red), which are attached to the shorter C 24 -26 aliphatic ␣-branch originating from FAS-I (bottom part of the mycolic acid) to yield the mature mycolic acids (1). ␣-Mycolic acids are the most frequent mycolic acids in M. tuberculosis (2). Three ␤-ketoacyl-ACP synthases are involved in the FAS-II system, namely FabH, KasA, and KasB (highlighted in blue). They are inhibited by the natural products cerulenin, thiolactomycin, and platensimycin (shown in the bottom blue box) and catalyze a Claisen condensation reaction between acyl-AcpM (acyl-CoA for FabH) and malonyl-AcpM via the three-step mechanism depicted in the top box (7,10). The reduction of the generated ␤-keto group to a methylene group is achieved via the subsequent actions of MabA, HadAB/BC, and InhA (43,62).
vation that the FAS-II enoyl-ACP reductases are highly flexible proteins prior to substrate binding (20 -22). In the present work, we probed the KasA ligand binding mechanism using TLM analogs substituted at the 3-position, which were designed to mimic the two KasA substrates, acyl-AcpM and malonyl-AcpM. We revealed a unique process of substrate binding, including extensive structural changes that can be triggered by one of our most potent inhibitors (23), laying the foundation for the development of new anti-mycobacterial agents.

EXPERIMENTAL PROCEDURES
Expression and Purification-Following a previously described procedure (10,11), KasA and the C171Q KasA variant were expressed in Mycobacterium smegmatis mc 2 155 and subsequently purified using affinity, anion exchange, and size exclusion chromatography.
Structure Determination-Data were integrated with XDS (KasA-TLM5-I) (24), XDSAPP (C171Q KasA-TLM) (25), or Imosflm (26), respectively. Scaling was performed using Scala (27). For Phaser molecular replacement (28), we used our previously determined TLM-bound KasA and C171Q KasA structures (lacking all ligands) as search models (PDB entries 2WGE and 2WGG) (11). Alternate cycles of model building and maximum likelihood refinement (including preceding TLS refinement (29)) were performed with Coot (30) and Refmac 5 (31). Based on the electron density maps, inhibitor binding sites could be identified unambiguously. Similar to our previous study, inhibitor densities were weaker but sufficient for wildtype KasA compared with the C171Q KasA variant (11). Hence, we used a 200-fold molar inhibitor excess during the co-crystallization of the wild-type KasA-TLM5 complex. In addition, soaking (for 1 and 5 min; KasA-TLM5-I and -II structures, respectively) was performed to enhance inhibitor occupancy. As expected, the TLM5 electron density improved with longer soaking times, whereas the resolution dropped from 2.4 to 2.7 Å. Refinement statistics are given in supplemental Table S1.
To avoid model bias, omit maps were calculated prior to inclusion of the respective ligand. Distances and angles are given as mean values for both subunits and as mean values Ϯ S.D. if multiple structures were analyzed. All structural figures were prepared with PyMOL (32).
Lipidomics-Two different and equally successful protocols were used to extract bound ligands from the protein samples. 10 l of KasA (100 M), C171Q KasA (150 M), or buffer reference samples were incubated for 30 min at 20°C with 10 l of a papain solution (1 mg/ml in 50 mM PIPES, pH 7.0, 300 mM NaCl, and 10 mM DTT). Alternatively, the samples were treated for 5 min at 55°C with 10 l of a 10 M urea solution. After the addition of 30 l of isopropyl alcohol, 10 min of sonication, and centrifugation, the supernatants were analyzed by LC/MS.
To analyze the phospholipid content of the KasA expression host, liquid cultures of M. smegmatis mc 2 155 with induced C171Q KasA expression were prepared as described previously (11). 30 mg of wet cell pellet were treated with 1 ml of Folch solution (chloroform/methanol ϭ 2:1) and subjected to 10 ultrasonic pulses. After centrifugation and separation from the cell debris, the supernatant was evaporated, and the pellet was dissolved in methanol and analyzed by LC/MS. All solvents used for lipid analysis were LC/MS grade purchased from Biosolve (Valkenswaard, the Netherlands). Reversed-phase chromatographic separation of the extracted lipids was performed on an ACQUITY Ultra Performance LC system (UPLC, Waters, Milford, MA) using an ACQUITY UPLC BEH C 18 , 1.7-m particle size, 2.1 ϫ 100-mm analytical column (Waters, Eschborn, Germany). The column was maintained at 60°C, and the injected sample (5 l) was eluted using a linear binary solvent gradient of 30 -100% eluent B over 10 min at a flow rate of 0.3 ml/min. Eluent A consisted of 10 mM ammonium acetate in water/acetonitrile (60:40, v/v), and eluent B consisted of 10 mM ammonium acetate in 2-propanol/acetonitrile (90:10, v/v). The UPLC was coupled to a hybrid quadrupole orthogonal time-offlight mass spectrometer (SYNAPT G2 HDMS, Waters, Manchester, UK). The negative electrospray ionization mode was used with capillary voltage and cone voltage of 0.8 kV and 25 V, respectively. The flow rate of the desolvation gas (nitrogen) was 800 liters/h, and the desolvation temperature was kept at 350°C. The system was equipped with an integral LockSpray unit with its own reference sprayer, and leucine encephalin (m/z of 554.2615) was used as the internal reference. The quadrupole was operated in a wide band RF mode. Data acquisition took place over the mass range of 50 -1200 Da. Two discrete and independent interleaved acquisition functions were automatically created. The first function collected the low energy data where molecule ions were acquired, whereas the second function collected the fragments of the molecule ion (high energy data) by using a collision energy ramp from 15 to 35 eV. In both instances, argon gas was used for collision-induced dissociation.
Synthesis of 3-Substituted TLM Derivatives-Enantiomerically pure (5R)-TLM and TLM analogs (93-96% enantiomeric excess) were synthesized using a procedure similar to a reported protocol (23,33). The detailed syntheses will be reported in a separate publication.
Inhibition Kinetics-Binding of TLM18 to wild-type and mutant KasA was quantified by monitoring changes in the intrinsic tryptophan fluorescence of the enzyme at wavelengths of 280 nm for excitation and 337 nm for emission, as described previously (23). Briefly, the data were collected using a Quanta Master fluorometer (Photon Technology International). Inhibitors were dissolved in DMSO and titrated into 1 M solutions of KasA and C171Q KasA in buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM dithiothreitol, pH 8.5).

RESULTS
KasA Inhibition by 3-Substituted TLM Derivatives-TLM is a natural product that binds preferentially to the KasA acylenzyme with low micromolar affinity (10). To improve the affinity of TLM for KasA, elucidate the validity of KasA as a drug target, and develop chemical tools to investigate the substrate binding and catalytic mechanism of KasA, we designed and synthesized two classes of 3-substituted TLM derivatives ( Table 1) that display structural features reminiscent of either the acyl-or the malonyl-AcpM substrates for the enzyme (23). The linear, hydrophobic 3-substituents of TLM3, TLM4, and TLM18 were attached to the thiolactone scaffold to mimic the long acyl chain of acyl-AcpM ( Fig. 2A, top box). In contrast, the 1,3-dioxo group of malonyl-AcpM is more closely reproduced by TLM5 and TLM6 carrying 3-acetyl and 3-trifluoroacetyl moieties ( Fig. 2A, bottom box). We subsequently solved the structures of the C171Q KasA acyl-enzyme mimic in complex with these TLM derivatives with resolutions of up to 1.5 Å. Importantly, the basic TLM binding mode is conserved for all investigated inhibitors and, thus, permits a detailed analysis of the effects of the different substituents on the enzyme. In particular, the TLM isoprenoid chain intercalates between two peptide bonds, and the thiolactone carbonyl group forms hydrogen bonds with His-345 and His-311 (Fig. 2, D and E) (11).
To reduce steric clashes between the new 3-position substituents and the KasA binding site, small but significant shifts in the TLM binding mode can be observed, including rearrangements of water molecules and slight conformational changes mainly observed for Phe-237 and its adjacent residues Phe-210, Ala-215, Met-216, and Phe-239 (Fig. 2, D and E). In addition, TLM5 and -6 move slightly away from Phe-237, thereby increasing the hydrogen bonding distance between the thiolactone carbonyl and His-345 from 2.74 Ϯ 0.07 to 2.93 Ϯ 0.04 Å but simultaneously enabling the formation of a hydrogen bond between the acetyl-carbonyl oxygen and His-345 at a distance of 2.96 Ϯ 0.08 Å (Fig. 2E). The observation of this bifurcated three-centered hydrogen bond between the 1,3-dioxo group and His-345 explains the observed 2-4-fold gain in affinity compared with TLM (Table 1) (23) and validates these compounds as good malonyl-AcpM analogs. In accordance, the carboxyl group of platensimycin, which was also suggested to mimic the malonyl substrate, was found to be located at a comparable position in an E. coli FabF structure (PDB entry 2GFX) (12). The additional weak hydrogen bonds between one of the TLM6 fluorines and the hydroxyl groups of Thr-313 and Thr-315 might explain why this compound displayed the most potent K i * value (460 nM (23)) of the investigated inhibitor series ( Fig. 2E and Table 1).
In contrast to these malonyl-AcpM analogs, the inhibitors TLM3, -4, and -18 are characterized by a 4 -18-fold decrease in affinity with respect to TLM (Table 1) (23). However, the residence time, which was suggested to be a particularly important kinetic quantity for the prediction of in vivo drug efficacy (34,35), increased 4-fold for TLM3 with a value of 33 min for this C171Q KasA slow onset inhibitor. Consistently, our structures also emphasize that this subset of compounds behaves differently compared with TLM5 and -6. Slight changes in the central TLM binding mode enable the extension of the more hydrophobic 3-substituents into an orthogonal direction toward an aromatic box consisting of Phe-210, Phe-237, Phe-239, and His-345 (Fig. 2, B and D). In addition, the TLM18 azide group forms a hydrogen bond to Gln-171 and participates in astacking interaction with Phe-237 and His-345. Despite these interactions, the binding affinity of TLM18 is reduced compared with TLM (K i * ϭ 34 M compared with 1.9 M, respectively) ( Table 1) (23). However, the affinity increases 2-fold for the wild-type enzyme (K d ϭ 112 M compared with 226 M), indicating that the desolvation of Gln-171, necessary for the direct interaction with the additional azide ( Fig. 2D), may lead to the reduced affinity of TLM18 toward the C171Q variant.
We note that the geometry of the observed interactions for all investigated complexes suggests that the thiolactone ring might be present as an enolate with the negative charge delocalized between the thiolactone carbonyl oxygen and the C-4 hydroxyl group (Fig. 2C). The short distance of the thiolactone carbonyl oxygen to His-311 (2.85 Ϯ 0.07 Å) clearly indicates the presence of a hydrogen bond (Fig. 2D). However, the thiolactone carbonyl double bond is not optimally oriented for hydrogen bond formation given that the lone pairs of electrons are in the plane of the bond (36). Taken together with the observation of a short (2.58 Ϯ 0.15 Å) and therefore probably charge-assisted hydrogen bond between the TLM hydroxyl oxygen in the 4-position and a conserved water molecule bound below the thiolactone plane (Fig. 2D), it seems likely that TLM analogs bind in a delocalized and deprotonated fashion to KasA (structures 3 and 4 in Fig. 2C). This is consistent with the knowledge that TLM and its analogs are vinylogous acids that can be deprotonated at physiological pH ( Fig. 2C; according to the program Moka (37), the predicted pK a of TLM is 5.94) (38). In this regard, His-311, which is believed to act as a general base during catalysis (7), may facilitate deprotonation of the tautomeric form of the inhibitors where the hydroxyl group is located next to the thiolactone sulfur (tautomer 2 in Fig. 2C).
Identification of a Phospholipid Bound to KasA-The electron density maps of our C171Q KasA structures also indicated the presence of a long-chain molecule close to the active site (Fig. 3). Due to the need to add PEG during C171Q KasA crystallization, we initially interpreted this density as a long PEG chain (11). However, the observation that this molecule was later present in protein crystals obtained in the

Induced Fit Substrate Binding Mechanism of KasA
absence of PEG suggested that it was a host-derived lipid. Intriguingly, using mass spectrometry, we demonstrated that, in contrast to wild-type KasA, the C171Q variant protein samples contained significant amounts of phospholipids (Table 2 and supplemental Table S2 and Fig. S1). Co-purification and crystallization of these phospholipids with C171Q KasA suggested that they bound specifically to this KasA variant. Three main phospholipids are observed with respect to the identity of the headgroup ( Fig. 3; phosphatidylethanolamine, phosphatidylinositol, and phosphatidylinositol oligomannosides), whereas the acyl chains are primarily palmitic (16:0) and tuberculostearic acids (19:0) ( Table 2). Interestingly, the latter methyl-branched fatty acid is a lipid component of Actinomycetales, such as mycobacteria, and can be used for the diagnosis of tuberculosis meningitis (39,40). C171Q KasA preferably binds glycerophospholipids containing acyl chains with 16, 17, and 19 carbon atoms (Table 2), which probably were co-purified from the expression host M. smegmatis mc 2 155 and crystallized with the C171Q KasA protein. Importantly, the major 19:0,16:0 phospholipids perfectly fit the observed electron density, with 16:0 attached to the central glycerol hydroxyl group (sn-2) and 19:0 at the terminal position (sn-1), which is in agreement with the previous finding that mycobacterial phospholipids are mainly composed of fatty acids with 18 or 19 carbons at position sn-1 and with 16 carbon atoms at position sn-2 ( Fig. 3) (40,41). For model building, we used a 19:0,16:0 phospholipid without a headgroup because it was not visible in the electron density, presumably due to its exposed position at the C171Q KasA surface (Fig. 3).
Location of the KasA Substrate Binding Sites-Clearly, the identified phospholipid represents a very long acyl chain bound to the C171Q KasA variant and defines the long KasA acylbinding channel (Fig. 4, A and B) that is created by hydrophobic residues mainly located in the capping region, a region of the KAS enzymes that is formed by genetically variable insertions ( Fig. 5 and supplemental Fig. S2) (11). A similar location of the cavity has also been observed in the E. coli FabB (ecFabB) C 12acyl-enzyme structure (Fig. 4A) (7).
To characterize the phosphopantetheine-channel, which is additionally necessary for the binding of both AcpM-substrates (38), we superimposed the active site cysteines of KasA and a CoA-bound E. coli FabH structure (Fig. 4C) (42). Consistently, the CoA molecule points into the cavity that contains the TLM inhibitors, underlining their potential as substrate analogs. We recently validated this model by NMR experiments showing interligand NOEs between a pantetheine analog and TLM bound to KasA simultaneously (23). According to this model and in line with FabF docking studies (43), Thr-313 and Thr-  red arrows). C, possible TLM species. TLM and its derivatives are vinylogous acids (the vinylogous system is highlighted in red) and might, therefore, be present in a deprotonated form at physiological pH (38). The two tautomers of protonated TLM (1 and 2) and the resonance structures of deprotonated TLM (3 and 4) are shown. D, acyl-AcpM TLM analogs bound to C171Q KasA. As a reference, the TLM structure is depicted in gray. Significantly deviating binding site residue conformations are shown for the inhibitor structure with the largest variance. The color code is depicted in A. Conformational changes and the differences in the TLM-binding pose are indicated by red arrows. E, malonyl-AcpM TLM analogs bound to C171Q KasA. As a reference, the TLM structure is shown in gray. The inhibitor color code is as described in A.
315, which are strictly conserved for KAS I and II enzymes (supplemental Fig. S2) and were proposed to be involved in ACP-substrate binding (44), contact the terminal phosphopantetheine peptide carbonyl oxygen (Fig. 4C), which is represented by one of the TLM6 fluorines (Fig. 2E).
In a recent kinetic study, Borgaro et al. (45) proposed that the binding of the acidic ACP protein to one ecFabB monomer leads to the delivery of its substrate to the active site of the adjacent subunit. They identified several basic amino acids on the protein surface that are important for ACP binding. Similarly, the corresponding KasA arginines 74Ј, 78Ј, and 79Ј on ␣-helix 3Ј as well as Arg-135Ј, Lys-136Ј, and Arg-214 (the prime designates the second KasA monomer) are ideally positioned to interact with the AcpM molecule. In this orientation, the bound AcpM would then permit the entrance of the substrates via the phosphopantetheine tunnel opening (indicated by the green arrow in Fig. 4D) of the neighboring monomer. In two of our nine crystal structures, a negatively charged HEPES molecule is bound to this basic KasA patch, where the acidic AcpM is expected to bind (cyan stick model in Fig. 4D). Importantly, we exclude the possibility that one or both substrates are delivered through the acyl pocket opening (blue arrow in Fig. 4D) due to distance considerations to reach the active site cysteine. An entrance through the acyl pocket opening would need to bridge the large distance between the phospholipid headgroup and the active site cysteine of ϳ30 Å (corresponding to 25 atoms) by the phosphopantetheine linker, which has a maximal length of ϳ16 Å (14 atoms) (Figs. 1 and 4, A and B).
Structural Basis for the KasA Substrate Preference-The mycobacterial ␤-ketoacyl-ACP synthases KasA and KasB are striking examples of enzymes that produce very hydrophobic meromycolates ( Fig. 1) with up to 56 carbon atoms in length in a cytosolic (i.e. hydrophilic) environment. KasA was found to extend short C 16 -18 FAS-I primers to C 38 -42 monounsaturated acyl chains, whereas KasB can produce full-length meromycolates (46,47). The unambiguous identification of phospholipids bound to the acyl channel of C171Q KasA explains the observed substrate preference of this enzyme. Remarkably, the  number of carbon atoms between the mutated active site cysteine and the terminal end of the sn-2 phospholipid acyl chain coincides with the maximal acyl chain length of 42 carbons generated by KasA (Fig. 4A). We assume that the C 42 chain cannot be further elongated because the acyl channel is capped by the peptide backbone of residues 239 -241, thus explaining why KasA is not able to produce full-length meromycolic acids. Consistently, phospholipids mimicking the C 44 (Table 2). Furthermore, we observed that the phospholipid does not bind in an extended manner but rather contains several kinks as predefined by the geometry of the surrounding KasA cavity (Fig. 4B). This non-linear architecture of the acyl channel might facilitate the binding of unsaturated primers, as exemplified by the presence of phospholipids with unsaturated fatty acids (Table 2). Indeed, it has been proposed that the cis -19 double bond is already introduced at the C 22 stage of meromycolate synthesis (2), and the observed non-linear shape of the acyl cavity could thus facilitate the binding of kinked unsaturated substrates with a migration of the cis double bond by two carbon atoms along the channel in each new elongation round.
In contrast to KasA, its homolog KasB produces full meromycolates with up to 56 carbons (46). Unfortunately, the available KasB crystal structure was not solved in its "open" acyl channel conformation, as observed for C171Q KasA, and thus the preference for longer substrates cannot be directly deduced (3,48). However, most residues that create the open C171Q KasA acyl-binding pocket are conserved in KasB. Among the few differences between KasA and KasB, E124F, P201R, and I347V are particularly interesting. The first two residues are located at the opening of the acyl pocket and might elongate the cavity in KasB. In contrast, Ile-347 lies between the aromatic box containing the hydrophobic TLM 3-substituents (Fig. 2D) and the terminal part of the hypothetical C 42 -acyl chain (Fig.  (11,23,42). The CoA molecule, which resembles the phosphopantetheinylated AcpM, is depicted in gray along with an intersection of the C171Q KasA-TLM5 structure. The entrance of the phosphopantetheine channel is indicated by a green arrow. The putative interaction between the terminal phosphopantetheine peptide carbonyl oxygen and the hydroxyl groups of Thr-313 and Thr-315 is depicted in the inset. D, proposed AcpM binding site. Basic residues located on ␣3Ј and 3Ј probably bind to the acidic AcpM protein, which is believed to deliver its substrate to the adjacent subunit (45). A HEPES buffer molecule (shown in cyan) bound to the ␣3Ј basic patch was observed for the C171Q KasA and C171Q KasA-TLM structures. The flexible regions of KasA are highlighted by red arrows in one subunit (the C171Q KasA-TLM5 and KasA-TLM (PDB entry 2WGE) structures are shown in yellow and gray, respectively) and as orange surface in the other subunit (see also Figs. 5 and 6 and supplemental Fig. S2). Blue and green arrows indicate the openings toward the acyl and phosphopantetheine pockets, respectively. 4B), suggesting that the acyl chain could be extended toward the aromatic box and TLM binding site in KasB due to distinct sequence variations (I145Y, G200T, and I347V). The hydrophobic 3-substitutent of TLM18 might mimic this alternative acyl binding mode (Fig. 4B).
Acyl Channel Opening Mechanism-Our structures unambiguously establish the location of all substrate binding sites, which immediately leads to two important questions. How does KasA achieve the uptake of its hydrophobic substrates, given that neither the long acyl chain nor its lipophilic binding pocket should be exposed to solvent for too long, and why do phospholipids not bind equally well to the wild-type enzyme? To address these questions, we solved the structure of TLM5 bound to wild-type KasA. Intriguingly, this malonyl-AcpM analog induces conformational changes that define an intermediate state between the wild-type KasA-TLM structure (PDB entry 2WGE (11)), which is characterized by a "closed" acyl channel, and the open structure observed for the acyl-enzyme analog C171Q KasA, which binds phospholipids. A per residue r.m.s. deviation plot of the three structures reveals five regions in close vicinity to the substrate binding sites that differ in their conformation between the enzyme and acyl-enzyme states (Fig.   5). Thus, we have successfully trapped for the first time a partially open acyl channel, revealing that the acyl substrate interacts with KasA through an induced fit binding mechanism. Although adaptive substrate recognition is a well known phenomenon, structural information about the dynamics of such processes is rare (49).
We characterized this mechanism and the associated conformational changes in a stepwise manner using the four different structures: KasA-TLM (PDB entry 2WGE), apo-KasA (PDB entry 2WGD), KasA-TLM5-I (nomenclature of all structures according to supplemental Table S1), and C171Q KasA-TLM5, of which only the final open state permits the binding of long acyl chains. Similar to the KasA-TLM complex, the unliganded KasA structure is almost closed (Fig. 5). The conformational changes toward the open state can be triggered by either the C171Q mutation, which mimics acylation of the active site cysteine (C171Q KasA-TLM5), or by the binding of TLM5 (KasA-TLM5-I). Because the natural malonyl substrate preferably binds to the open acyl-enzyme state during catalysis (Fig. 1) (13), its close analog TLM5 seems to have the capability to select for or to actively induce a partially opened conformation in the wild-type enzyme. The relatively strong interaction between KasA and this inhibitor (23) might be the cause for these conformational changes, which enabled us to trap an intermediate state between the wild-type and acyl-enzyme forms. In contrast, similar rearrangements were not observed with TLM, which binds with lower affinity to KasA (11,23).
Compared with the closed state, the three subsequent structures (apo-KasA, KasA-TLM5-I, and C171Q KasA-TLM5) are characterized by a Phe-404 conformation, in which the phenyl ring plane is rotated by 5°, (Ϫ)19°, and 59°, respectively (henceforth, we will indicate the stepwise changes observed for these three structures compared with the closed KasA-TLM state using the following notation: 5°3 19°3 59°) (Fig. 6, A and C). The introduction of the C171Q mutation that mimics acylation of the active site cysteine as well as the binding of TLM5 in a slightly different orientation compared with TLM apparently induces the rotation of the Phe-404 phenyl ring. This conserved residue was suggested to play an important role as a gatekeeper between the active site cysteine and the acyl-binding cavity opening upon acylation (11).
The alteration in the Phe-404 conformation induces a cascade of structural changes (Fig. 6). Initially, the adjacent residues Phe-210, Met-213, Val-142Ј, and Met-146Ј are shifted due to the rotation of Phe-404 (Fig. 6, C, D, and F). Compared with the closed reference state, the C ␣ atom of Phe-210 moves by 0.4 3 0.5 3 1.8 Å, and the C ␣ atom of Val-142Ј moves by 0.1 3 0.6 3 1.6 Å. Together with Ile-347, these three residues comprise a binding cavity for Leu-116 in the initial KasA-TLM structure. Due to the subsequent shift and tightening of this pocket, Leu-116 is successively expelled from the acyl channel (Leu-116 C ␣ moves by 1.5 3 3.3 3 4.3 Å) (Fig. 6C), which creates space for the binding of an acyl chain, and, hence, Leu-116 can be regarded as a second gatekeeper residue. This observation validates experimentally a similar proposal by Schaefer et al. (38) based on a molecular dynamics simulation.
Simultaneously, the Phe-404 induced shifts of Phe-210 and Met-213 directly lead to a movement of an entire range of residues, including amino acids 202-239 (r.m.s. deviation change of 0.2 3 0.4 3 1.1 Å) and therefore helix ␣9 (residues 204 -211) (Figs. 5 and 6D and supplemental Fig. S2). In particular, the conformational change of residues 212-216 causes a widening of the phosphopantetheine pocket, which accounts for the enhanced TLM affinity toward the acyl-enzyme intermediate (11). Helix ␣9 is in close proximity to a region containing residues 46 -63 (including ␣2 and ␤3) and 122Ј-148Ј (including ␣5Ј, ␣6Ј, and the 3 10 -helix 3Ј), which consequently move in a similar direction supported by the spatial rearrangements of Val-142Ј and Met-146Ј, which in turn are triggered by the rotation of the Phe-404 phenyl ring plane (r.m.s. deviation values change by 0.1 3 0.5 3 1.0 Å and 0.4 3 0.9 3 2.1 Å, respectively) (Fig. 6F). The extensive alterations in helices ␣5Ј and ␣6Ј in turn induce a slight shift of helix ␣3Ј (r.m.s. deviation change of 0.2 3 0.4 3 0.6 Å) and, together with the movement of Leu205 located on helix ␣9, the extrusion of Tyr-126Ј from the acyl-binding channel, which is stabilized in this final conforma-FIGURE 6. The KasA acyl substrate recognition mechanism. A-F, the phospholipid and C 12 -acyl chain (Fig. 4) are shown in gray. Gatekeeper residues Leu-116, Phe-404, and Tyr-126Ј are labeled in blue. Red arrows indicate the stepwise conformational changes apparent from the structures KasA-TLM (lilac; PDB entry 2WGE), KasA (green; PDB entry 2WGD), KasA-TLM5-I (yellow), and C171Q KasA-TLM5 (orange). For the second subunit, pale colors were used. The phosphopantetheine channel opening is indicated by a green arrow (Fig. 4). E contains a detailed view of the KasA dimer interface along a 2-fold axis of symmetry. The conformational changes observed for Tyr-126 and Asp-127 lead to a short hydrogen bond in both subunits. NOVEMBER 22, 2013 • VOLUME 288 • NUMBER 47 tion by a hydrogen bond to Asp-127 (Fig. 6, E and F). During this process, the distance between the hydroxyl group of Tyr-126Ј and the Asp-127 carboxyl oxygen decreases from 11.5 Å via 10.9 and 5.6 Å to ultimately 2.6 Å. Related via a 2-fold axis of symmetry, the exact same modifications are observed between Tyr-126 and Asp-127Ј, which indicates that the proposed cascade of conformational changes and, therefore, the opening of the substrate binding pockets take place in the neighboring monomer at the same time.

Induced Fit Substrate Binding Mechanism of KasA
Importantly, Tyr-126Ј may be envisioned as a third gatekeeper that protects the hydrophobic acyl cavity from solvent exposure by its hydrophilic OH group. According to the proposed mechanism, all three gatekeepers (Phe-404, Leu-116, and Tyr-126Ј) exit concertedly from the acyl channel to permit the simultaneous uptake of an acyl chain, as exemplified by the C171Q KasA-bound phospholipid (Fig. 6B and supplemental Movies S1 and S2). This concerted process is facilitated by considerable shifts of the helices surrounding both acyl channels and would avoid excessive exposure of the hydrophobic channel to the solvent ( Fig. 5A and supplemental Movie S1). Interestingly, these helices are part of the capping region, a genetically variable part of the KAS enzymes that may have evolved specifically to enable the binding of meromycolate precursors to KasA (Fig. 5B and supplemental Fig. S2) (43). The vastly different but also flexible FabH capping domain, which is formed from insertions located at distinct sites in the primary sequence (supplemental Fig. S2), has similarly been suggested to be involved in substrate binding (22,50). Moreover, a sequence alignment of the KAS enzymes shows that the three gatekeeping residues are conserved in KasB, thus suggesting that a similar substrate recognition mechanism may be present in this enzyme (supplemental Fig. S2). Consistent with this proposal, it has been suggested that Leu-115 (corresponding to Leu-116 in KasA) has to undergo conformational changes upon binding of cerulenin, which contains a long (C 9 ) hydrophobic chain ( Fig. 1) (10,48).

DISCUSSION
To address the question of how the ␤-ketoacyl-ACP synthase KasA ensures the binding of very hydrophobic long-chain mycolic acid precursor molecules, we utilized analogs of the natural product TLM that recapitulate key aspects of the structure of the two KasA substrates, acyl-AcpM and malonyl-AcpM. Our structural data support an induced-fit substrate binding mechanism, which involves the concerted rearrangement of three gatekeeping residues. This mechanism includes the movement of several helices, thereby creating the acylbinding pocket and thus accounting for the ability of the C171Q KasA acyl-enzyme mimic to bind phospholipids that are similar in size to the C 42 acyl chain of the natural substrate (Table 2 and supplemental Movies S1 and S2).
Strikingly, these conformational changes take place in the KasA region, where AcpM is expected to bind (Fig. 4D) (45), and thus it is likely that AcpM binding initiates this process. Based on this hypothesis, we propose the following mechanism. The acidic AcpM protein interacts via helices ␣2 and ␣3 with the basic KasA surface region containing helices ␣9, ␣3Ј, ␣5Ј, 3Ј, and ␣6Ј and thereby delivers both substrates through the phosphopantetheine tunnel (green arrow in Fig. 4D) of the adjacent subunit. Because all of the KasA helices mentioned above undergo conformational changes during the acyl channel open-FIGURE 7. Mechanistic insights. A and B, geometric possibilities for malonyl-AcpM binding. Two different models of malonyl-AcpM binding to KasA via the phosphopantetheine tunnel are displayed above TLM5. In addition, A depicts six water molecules of the C171Q KasA apo-structure that are not present in the binary complex. Water molecules that are probably displaced by the malonyl substrate are colored in red, and water molecules that might still be present upon malonyl-AcpM binding are highlighted in cyan. The red arrow indicates the water that was suggested to attack the carboxylate group of malonyl-AcpM (55). C, mechanistic consequences of TLM5 being a malonyl-AcpM analog. His-345 binds to the 1,3-dioxo group of TLM5, which mimics the malonyl group of the natural substrate. The side chain amide carbonyl oxygen of Gln-171 is located in the oxyanion hole formed by residues 171 and 404 (the side chain of Phe-404 is omitted for clarity). The proposed attack vector geometry of the acetyl-carbanion forming via decarboxylation is depicted in cyan. D, potassium cation binding site. A potassium ion is bound close to His-311 in all obtained KasA structures. The octahedral K ϩ binding site (highlighted with cyan dashes) consists of the main chain carbonyl oxygens of Asn-309, Ala-310, and Asn-400 (distances of 2.76 Ϯ 0.02, 2.97 Ϯ 0.05, and 2.71 Ϯ 0.03 Å for all C171Q KasA structures) and the side chain oxygens of Asn-309, Glu-354, and Asn-399 (distances of 2.68 Ϯ 0.04, 2.58 Ϯ 0.04, and 2.59 Ϯ 0.04 Å). A water molecule and Lys-340 connect the K ϩ ion with the catalytic residue His-311 via Glu-354. Similar cation binding sites are present in all structurally characterized KAS I and II enzymes (44). We unambiguously identified a potassium cation by analysis of the cation binding site and coordination geometry (64 -66), crystal diffraction data, and results from inductively coupled plasma optical emission spectrometry (data not shown). An anomalous difference omit map ( ϭ 1.7 Å) is depicted at 5 as a gray mesh. E, hydrogen bonding network surrounding His-311. Positively (Lys-340; depicted in blue) and negatively (Asp-319, Glu-322, and Glu-354; depicted in red) charged amino acids, which were proposed to be important for decarboxylation (44), interact with His-311 via a complex hydrogen bonding network, which includes a potassium ion located in a conserved cation binding site (D). This environment probably leads to the modulation of the electronic properties of the His-311 ⑀-nitrogen (magenta) to a catalytically ideal state (54).
ing process, it is tempting to speculate that AcpM might be the trigger for the observed substrate binding mechanism and specifically could induce the opening of the three gatekeepers. The acyl substrate could thus be injected simultaneously to the opening of the acyl pocket upon AcpM binding, ensuring that neither the hydrophobic acyl chain nor the acyl cavity are exposed to the solvent. In support of this hypothesis, we observe that the acyl chain itself is not able to induce these conformational changes, as exemplified by the absence of the phospholipids in the wild-type KasA binding cavity. Furthermore, our model explains why acyl-CoA substrates are not efficiently processed by KasA (45). In contrast, the FabH enzyme harbors a fundamentally different capping region (supplemental Fig. S2), which probably accounts for the altered specificity toward acyl-CoA primers (50).
In conjunction with the proposed substrate binding mechanism, our C171Q KasA structures provide an excellent model for the investigation of the three-step catalytic mechanism. In particular, the decarboxylation step of KAS enzymes has not been fully elucidated due to the lack of structural information to establish the location of the malonyl group. The 1,3-dioxo moiety of TLM5 is an excellent mimic of the malonyl group and, consistent with previous studies, is hydrogen-bonded to His-345 in the active site (44). However, it is not possible to distinguish between the two oxo groups (i.e. which of the two represents the malonyl carboxylate), thus leading to the two geometric possibilities depicted in Fig. 7, A and B. These possibilities represent the two conflicting mechanistic models that have been proposed for the decarboxylation reaction; either His-311 or Phe-237, which are both strictly conserved among KAS I and II enzymes (supplemental Fig. S2), were suggested to trigger the decarboxylation of malonyl-ACP (10,14,44). In the orientation depicted in Fig. 7A, the ⑀-nitrogen of His-311 could trigger the decarboxylation, whereas this role is indicated for one of the Phe-237 -faces in the alternative case (Fig. 7B). For both models, decarboxylation leads to the generation of the carbanion at the carbon atom located between the two TLM5 carbonyl groups (Fig. 7C). Interestingly, the evolving carbanion could approach the acylated cysteine with a Buergi-Dunitz trajectory of 110° (Fig. 7C), which is within the expected range of 100 -110°for the transition state attack vector angle of a carbonyl group by a nucleophile (Nu 3 CϭO) (51,52). For similar stereoelectronic reasons, the carbanion should attack the Cys-171 thioester carbonyl with an obtuse angle as assumed for the related aldol reactions (53). Thus, we reason that the malonyl group of the AcpM substrate is positioned with the carboxylate pointing toward His-311 (enolate attack angle of 92°versus 54°; see also Fig. 7, A and C) and therefore exclude the possibility depicted in Fig. 7B. Strikingly, all residues that were proposed to play an important role during decarboxylation (the charged amino acids Asp-319, Glu-322, Lys-340, and Glu-354, which are conserved among KAS I/II proteins) (44) and an octahedrally bound potassium ion identified in this study (Fig. 7D) are involved in a complex hydrogen bonding network contacting FIGURE 8. Proposed KasA reaction mechanism. This mechanistic scheme is strongly supported by our structural work and builds on previous studies (13,44,54,55,57). Gray dashed lines indicate hydrogen bonds. It was proposed that the dipole moment of helix ␣8 acidifies Cys-171 (supplemental Fig. S2) (43), leading to a zwitterionic His-311H ϩ -Cys-171 Ϫ ion pair in the free enzyme (box 1), which activates Cys-171 for the nucleophilic attack of the acyl-AcpM (box 2) (54). The His-311 ⑀-nitrogen (depicted in magenta as in Fig. 7E) probably destabilizes the malonyl-AcpM carboxylate during decarboxylation, as highlighted by the magenta dashed line in box 4. In contrast to a previous hypothesis regarding the decarboxylation reaction (55), our KasA binary complex structures contain no appropriately positioned water molecule that could be activated by the His-311 ⑀-nitrogen to trigger the decarboxylation reaction directly (box 4). Consequently, we suggest that the initial product is CO 2 , which can then be hydrated after decarboxylation (between steps 4 and 5). The evolving enolate (box 5) is stabilized by the neutral His-345, which is protonated at its ⑀-nitrogen because the ␦-nitrogen is hydrogen-bonded to the backbone NH of Ile-347 (54). Geometric considerations based on our structures suggest that the nucleophilic attacks of Cys-171 and of the acetyl-carbanion occur from the Si-and Re-faces of the thioester carbonyl groups, respectively (boxes 2 and 5; see also Fig. 7).
His-311 (Fig. 7E) and might thus be responsible for tailoring the catalytically ideal His-311 electronic state (54).
Based on our structural results concerning the binding mode of the malonyl group and taking previous findings into account (13,44,54,55), we propose the mechanism depicted in Fig. 8. A mechanistic study of the mammalian fatty acid synthase ␤-ketoacyl synthase domain, which shares the Cys-His-His catalytic triad and other important active site residues with KAS I enzymes (56), revealed that bicarbonate is the product of the condensation reaction (13). Subsequently, a similar outcome has been proposed and confirmed for Streptococcus pneumoniae FabF (spFabF) (55). Thus, an activated water molecule either directly attacks the carboxylate of malonyl-ACP or hydrates CO 2 after decarboxylation (13). The identification of a structured water molecule bound to the ⑀-nitrogen of His-311 (according to the KasA numbering) in most apo-KAS I/II crystal structures has previously been used to support this mechanism (55,57). Indeed, our C171Q KasA apo-structure confirms the presence of such a water molecule (indicated by the red arrow in Fig. 7A). However, in the binary complex, it is displaced by the thiolactone oxygen of TLM5 mimicking one of the malonyl-AcpM carboxylate oxygens. Assuming a neutral His-311 (deprotonated at N ⑀ ) prior to decarboxylation, as proposed for ecFabB and spFabF (7,55,57), we suggest that the ⑀-nitrogen lone pair electrons directly destabilize the adjacent malonyl-AcpM carboxylate (Fig. 8, box 4, magenta dashed line), leading to the transfer of the negative charge to the thioester carbonyl, which in turn is stabilized by His-345. After decarboxylation, one of the water molecules surrounding the substrate might re-enter the binding site and could hydrate the CO 2 molecule facilitated by a His-311-mediated activation (Fig.  7A). In support, an ecFabB H298Q mutant was decarboxylation-deficient, whereas an H298E mutant displayed residual decarboxylation activity at pH values between 6 and 8, which highlights the necessity for an electron-rich lone pair at the position of the His-311 ⑀-nitrogen (corresponding to ecFabB His-298) enabling both decarboxylation and binding of a water molecule (44).
Most enzymes are inherently flexible systems tailored to accommodate substrates and undergo catalysis (58). In light of the fact that the free energy of the transition state along the reaction coordinate has to be lowered efficiently, it is not very surprising that the dynamics of substrate binding and catalysis are also reflected in the inhibition of enzymes (49,59,60). This is nicely illustrated by recent studies concerning the induced fit binding of peptide deformylase and enoyl-ACP reductase inhibitors (20,49,61). For instance, the picomolar affinity of diphenyl ether enoyl-ACP reductase inhibitors can be related to their capability to mimic the transition state of the catalyzed reaction and to trigger the correlated ordering of a flexible loop (20,61). In the present study, we reveal snapshots of the dynamic process that is involved in the recognition of very long chain fatty acyl substrates by the drug target KasA. This investigation of induced fit substrate recognition and turnover has important implications for the design of optimized KasA inhibitors. For instance, TLM 3-acyl substituents can trigger the opening of the phosphopantetheine and acyl channels in the wild-type KasA enzyme. This observation now suggests ways in which these inhibitors can be tailored in order to occupy the hydrophilic and hydrophobic cavities in the active site, thus leading to compounds with improved affinity for KasA and ultimately to the development of urgently needed drug candidates for the treatment of tuberculosis.