Staphylococcus aureus FabI: Inhibition, Substrate Recognition and Potential Implications for In Vivo Essentiality

Summary

Methicillin-resistant Staphylococcus aureus (MRSA) infections constitute a serious health threat worldwide, and novel antibiotics are therefore urgently needed. The enoyl-ACP reductase (saFabI) is essential for the S. aureus fatty acid biosynthesis and, hence, serves as an attractive drug target. We have obtained a series of snapshots of this enzyme which provide a mechanistic picture of ligand and inhibitor binding including a dimer-tetramer transition combined with extensive conformational changes. Significantly, our results reveal key differences in ligand binding and recognition compared to orthologous proteins. The remarkable observed protein flexibility rationalizes our finding that saFabI is capable of efficiently reducing branched-chain fatty acid precursors. Importantly, branched-chain fatty acids represent a major fraction of the S. aureus cell membrane and are crucial for its in vivo fitness. Our discovery thus addresses a long-standing controversy regarding the essentiality of the fatty acid biosynthesis pathway in S. aureus and validates saFabI as a drug target.

Introduction

Resistant strains of the Gram-positive human pathogen Staphylococcus aureus emerged only two years after the introduction of the antibiotic methicillin in 1959. Since then, methicillin-resistant S. aureus (MRSA) infections have become an increasing problem in hospitals worldwide. To treat severe MRSA infections, vancomycin is currently used as the drug of last resort (Pantosti and Venditti, 2009). Thus, the initial occurrence of a vancomycin-resistant S. aureus (VRSA) strain in June 2002 is especially alarming (Sievert et al., 2008). Unfortunately, antibiotic resistance has drastically outpaced new antibiotic discovery, which flourished between the 1940s and 1960s. Between 1962 and 2000 only a few novel scaffolds like the carbapenems were approved, and novel antibiotics with alternative modes of actions are therefore urgently needed. Accordingly, from 2000 onwards oxazolidinones (e.g. linezolid), the lipopeptide daptomycin and the mutilins were introduced (Fischbach and Walsh, 2009).

One strategy that targets the bacterial cell envelope involves the inhibition of the type II fatty acid biosynthesis pathway (FAS II), which is responsible for the de novo production of phospholipid precursor molecules (Payne et al., 2002). In contrast to the large multifunctional mammalian FAS I synthase, the FAS II system of bacteria, protozoa and plants consists of individual monofunctional enzymes, allowing selective inhibition (Payne et al., 2001). In this pathway, the fatty acid chain, which is attached to the acyl carrier protein (ACP), is elongated by two carbon atoms per cycle (Figure 1). The final reduction is catalyzed by the NAD(P)H-dependent trans-2-enoyl-ACP reductase (FabI), which is known to play a key regulatory role in this pathway (Heath and Rock, 1995; Xu et al., 2008). This enzyme is highly conserved across many pathogens, however, the S. aureus enoyl-ACP reductase (saFabI) is the only known FabI with a clear preference for NADPH (Heath et al., 2000; Priyadarshi et al., 2010; White et al., 2005; Xu et al., 2008). The clinical success of FabI inhibitors, such as the first-line tuberculosis prodrug isoniazid (Lu and Tonge, 2008) and several compounds currently in phase I clinical trials (Gerusz, 2010), validates this enzyme as an attractive drug target. Furthermore, the FabI diphenyl ether inhibitor triclosan (TCL) is recommended as topical antiseptic to reduce MRSA skin colonization (Bamber and Neal, 1999).

Figure 1. The bacterial type II fatty acid biosynthesis pathway as a drug target.

S. aureus enzymes are shown in red, while isoforms in other organisms are depicted in black (Payne et al., 2001). Initiation of fatty acid biosynthesis requires malonyl-ACP, formed from acetyl-CoA by acetyl-CoA carboxylase (Acc) and malonyl-CoA:ACP transacylase (FabD). The β-ketoacyl-ACP synthase FabH (KAS III) performs the first Claisen condensation yielding a β-ketoacyl-ACP from malonyl-ACP and an acyl-CoA primer. The two carbons originating from malonyl-ACP are shown in cyan and the first two primer carbons in red. Subsequently, the β-ketoacyl-ACP is converted to a saturated acyl-ACP by the actions of the NADPH-dependent reductase FabG, the dehydrase FabZ, and the mostly NADH-dependent enzyme FabI. The saturated acyl-ACP is a substrate for additional rounds of elongation in which FabF catalyzes the condensation reaction. The long-chain acyl-ACP products are subsequently partially transformed into membrane lipids (a typical S. aureus phosphatidylglycerol is shown (Parsons et al., 2011)). Most bacteria use acetyl-CoA as the primer which results in straight-chain fatty acids (C₁₄ and C₁₆) (Mendoza et al., 2002), however, S. aureus FabH prefers branched-chain acyl-CoAs which yield branched-chain fatty acids (Qiu et al., 2005). The branched-chain primers isobutyryl-CoA, isovaleryl-CoA and 2-methylbutyryl-CoA are derived from valine, leucine and isoleucine, respectively, through the actions of the branched-chain aminotransferase BAT (ilvE (Madsen et al., 2002)), and the branched-chain α-ketoacid dehydrogenase BKD (lpd, bkdA1, bkdA2 and bkdB (Singh et al., 2008)). The branched-chain acyl-CoAs yield iso-C₁₄ and iso-C₁₆ (isobutyryl), iso-C₁₅ and iso-C₁₇ (isovaleryl), and anteiso-C₁₅ and anteiso-C₁₇ (2-methylbutyryl) fatty acids (Mendoza et al., 2002). Exogenous fatty acids are converted into acyl-ACPs by the acyl-ACP synthetase (AAS) for incorporation into cell membranes. Typical fatty acids of the human blood plasma are palmitic (21.3%) and linoleic acid (23.7%) (Holman et al., 1995). For S. aureus, exogenous oleic acid was only incorporated in the 1-position of phosphatidylglycerol (Parsons et al., 2011). Inhibitors of the NADPH-dependent FabI from S. aureus (saFabI) used in this study include 5-chloro-2-(2,4-dichlorophenoxy)phenol (triclosan), 5-chloro-2-phenoxyphenol (CPP) and 5-ethyl-2-phenoxyphenol (EPP) (Xu et al., 2008). See also Figure S1.

Recently, however, Brinster et al. challenged the validity of FAS II inhibitors as Streptococcus agalactiae can utilize exogenous fatty acids from the host blood serum to survive inhibition of its FAS II machinery (Brinster et al., 2009). Whether this resistance mechanism extends to all Gram-positive bacteria including S. aureus is currently controversially discussed (Balemans et al., 2010; Parsons et al., 2011). Importantly, S. aureus remains sensitive to FabI inhibitors in vivo, as shown in several rodent models (Balemans et al., 2010; Escaich et al., 2011; Park et al., 2007; Payne et al., 2002).

To investigate the suitability of saFabI as a drug target, we have structurally and biochemically characterized this enzyme with respect to inhibitor binding, conformational flexibility, ligand binding mechanism, cofactor and substrate specificity, quaternary structure and cooperativity. We identified significant differences in the S. aureus and Bacillus FabIs compared to the classical FabI proteins from organisms such as Escherichia coli, Burkholderia pseudomallei and Francisella tularensis; and of our structures with respect to the recently determined S. aureus FabI structures in complex with triclosan and in its apostate (Priyadarshi et al., 2010). Importantly, our analysis revealed that the substrate specificity of saFabI displays an increased specificity for branched-chain (BCFA) relative to straight-chain fatty acid (SCFA) precursors (Figure S1), consistent with the high content of BCFAs in the membranes of Staphylococci and Bacilli (Kaneda, 1991). Since BCFAs, which are barely found in human serum (Holman et al., 1995), are likely required for S. aureus fitness in vivo, we conclude that saFabI is a suitable target for drug discovery.

Results

Overall structure

To provide a molecular basis for the rational design of compounds that target saFabI, we solved the structures of ternary FabI complexes with the cofactor NADP⁺ and the diphenyl ether inhibitors triclosan (TCL), 5-chloro-2-phenoxyphenol (CPP) or 5-ethyl-2-phenoxyphenol (EPP), respectively (Figures 1, S1 and S2). As a member of the short-chain dehydrogenase/reductase (SDR) superfamily, saFabI contains an extended Rossmann-fold comprised of a central seven-stranded parallel β-sheet surrounded by three α-helices on either side (Figure 2B) (Grimm et al., 2000; Lu and Tonge, 2008). Furthermore, the structures of our saFabI inhibitor complexes clearly show the presence of homo-tetramers (Figure 2A) as observed in other organisms (White et al., 2005). The closest relationship is found with the FabI structures from the Gram-positive organisms Bacillus anthracis and B. subtilis as well as with the incorrectly designated “FabL” from B. cereus which is 100% identical to B. anthracis FabI (Figure 3).

Figure 2. Overall structure and flexibility of saFabI.

(A) Homo-tetrameric structure of saFabI. Each subunit of the TCL-2 structure is shown in a different color. The three perpendicular 2 fold non-crystallographic symmetry axes P, Q and R are indicated. The interfaces spanned by the axes P and R (Q and R) are referred to as the PR- (QR-) interface. (B) Structure of the saFabI monomer. α-Helices are shown in red, β-strands in yellow, loops in green, cofactor and inhibitor as space-filling models in grey. The chair-like structure consists of the Rossmann-fold (seat, lower part) and the active site (back, upper part). (C) Flexibility of the saFabI active site in its ligand-free form. Multi-rmsd values were calculated for the C_α atoms of the apo-1 (one subunit), apo-2 (all subunits) and TCL-2 (one subunit) structures using LSQMAN (Kleywegt and Jones, 1996). These values varied from 0.2 to 20.5 Å and are represented by color (from blue to red) and width of the TCL-2 model (residues 2–256). Rmsd-values for the missing residues 199–201 were interpolated in a linear fashion based on the values for the two adjacent amino acids 198 and 202. See also Figure S2.

Figure 3. Secondary structure of saFabI and alignment with similar proteins.

Secondary structure matching (SSM) was performed for the TCL-2 structure using PDBe Fold (Krissinel and Henrick, 2004). The secondary structure of saFabI is shown in the first line according to the program DSSP (Kabsch and Sander, 1983). Secondary structure elements participating in the QR and PR-interfaces are depicted in red and blue, respectively. Amino acids highlighted in red are conserved. The alignment comprises the 7 best SSM hits (Q-scores are indicated behind the sequences) including bsFabL and saFabG. The first two letters of the protein names indicate the bacterial source (sa = S. aureus, ba = Bacillus anthracis, bc = B. cereus, bs = B. subtilis, ft = Francisella tularensis, tt = Thermus thermophilus, ec = Escherichia coli, aa = Aquifex aeolicus). PDB codes and subunits are given in parentheses. Blue, red and cyan stars mark residues interacting with the cofactor, inhibitor or cofactor and inhibitor, respectively. Disordered or rearranged regions in all four available saFabI apo structures (apo-1, apo-2, 3GNS, 3GNT) are marked with a continuous line, regions which are additionally disordered or rearranged for distinct apo structures are represented by a dashed line. This Figure was prepared with ESPript (Gouet et al., 1999).

In addition, we solved two apo saFabI structures which display unexpected flexibility in close proximity to the binding site (Figure 2C). In one of the two apo structures all four monomers (apo-2; nomenclature of all structures according to Tables S1 and S2) differ in their active site (pairwise rmsd of 0.84 Å) and display conformations that, to our knowledge, have never been observed before. Compared to the inhibitor-bound complexes, three regions - the active site and two substrate-binding loops (Figure 3) - are either disordered or visible in a rearranged architecture. In contrast, previously published apo structures of FabI proteins from Gram-negative bacteria (PDB codes: 2JJY, 2P91, 2WYU, 3EK2, 3GRK, 3K2E) show disorder in only one region, the substrate-binding loop.

SaFabI inhibition

In contrast to the previously described saFabI-NADP⁺-TCL structure (PDB code: 3GR6) (Priyadarshi et al., 2010), our high-resolution saFabI complex structures with the unsymmetrically substituted inhibitors EPP and CPP unequivocally reveal a similar binding mode of diphenyl ethers including triclosan to saFabI as reported for homologous FabI proteins (Lu and Tonge, 2008; Xu et al., 2008) which allows the formation of two key hydrogen bonds connecting the A-ring hydroxyl group to Tyr157 and to the 2′-OH of the nicotinamide ribose at 2.5 ± 0.1 and 2.6 ± 0.1 Å, respectively (Figure 4B). TCL, CPP and EPP bind to a pocket composed of NADP⁺ and several mainly hydrophobic residues with average distances of 3.8 ± 0.4 Å (Figures 4B and C). Notably, the carbonyl oxygen of Ala97 forms a geometrically favorable halogen bond (Bissantz et al., 2010) with the TCL B-ring chlorine at position 4 (3.2 ± 0.1 Å). At the same time the chlorine is weakly hydrogen bonded to the amide NH group of Ala97 (3.2 ± 0.1 Å, Figure 4B). Accordingly, TCL exhibits an almost 3-fold lower K_i*^,app value as compared to CPP, which lacks the two B-ring chlorines and is rotated by 11 ± 2° into the binding crevice (Table 1, Figure 4D and Supplemental Results).

Figure 4. SaFabI inhibition with diphenyl ethers.

(A) Dose-response curves of diphenyl ether analogues. The steady-state reaction velocities (v_s) at various inhibitor (CPP, EPP, triclosan) concentrations are plotted as a fraction of the uninhibited reaction velocity (v_u). Each plot is fitted to the standard binding isotherm. R² = 0.88, 0.75 and 0.73 for the best fit curves to the binding isotherm for CPP, EPP and triclosan, respectively. (B) Schematic representation of all residues interacting with triclosan (TCL in red). The interaction pattern was derived from the TCL-2 structure. Directional interactions are depicted explicitly (black residues), amino acids creating the hydrophobic pocket are included in 1-letter code (residues in blue boxes). R = adenosine diphosphate. (C) Fit of triclosan in its binding pocket. Triclosan is shown as space filling model in grey. Ala95, Phe96, Ala97, Leu102, Tyr147, Tyr157, Met160, Ile207 and NADP⁺ are shown in surface representation in blue. Additionally interacting residues Ser197, Ala198, Val201 and Phe204 are located within the substrate-binding loop (SBL). (D) Comparison of the TCL and CPP inhibitor binding modes. The TCL-2 structure is shown in grey and yellow, the CPP structure in blue. The cofactor and some protein regions are omitted for clarity. Arrows indicate differences between the two structures. The B-ring is rotated by ~11 ± 2° upon removal of the two chlorine atoms. Simultaneously, the C_α atoms of Ala97, Leu102, Ser197 and Ala198 move towards the inhibitor by about 0.4 ± 0.1 Å, 0.3 ± 0.1 Å, 0.3 ± 0.1 Å and 0.2 ± 0.1 Å, respectively, and the side chain of Met160 rotates away from the inhibitor by ~13 ± 3°.

Table 1.

Potency of saFabI inhibition by diphenyl ether analogues.

Inhibitor	Ki*,app (nM)
CPP	161.2 ± 8.4
EPP	106.6 ± 7.9
Triclosan	60.9 ± 5.6

Induced fit ligand binding

SaFabI undergoes several major conformational changes upon cofactor and inhibitor binding, which have, to our knowledge, never been observed previously for its Gram-negative FabI orthologs, thus allowing us to describe a fundamentally expanded mechanism for ligand binding. All residues interacting with the diphenyl ether inhibitors are located in three distinct regions (Figure 3), which exhibit substantial flexibility prior to cofactor and inhibitor binding. As described for numerous FabI orthologs and other SDR proteins, the substrate-binding loop (SBL, 194–204) is disordered or rearranged in our apo structures (Grimm et al., 2000; Tipparaju et al., 2008). In contrast to other FabI structures, however, the saFabI structures reveal flexibility of a second substrate-binding loop (SBL-2, 94–108) as well as the active site loop (ASL, 147–157).

Superposition of the apo-1 structure with TCL-2 shows that the active site is wide open prior to cofactor and inhibitor or substrate binding (Figure 5A). In addition to the three aforementioned loop regions, the C-terminal SBL extension α7 is disordered. The apo-2 structure provides informative snapshots of α7’s motion on its way to the ternary complex conformation as every subunit shows a different arrangement of this helix (Figure 5B-F and Movie S1). In one subunit, this helix is shifted ~13 Å away from the active site (Figure 5B). When helix α7 moves towards the protein core, it is still rotated about 105° relative to the ternary complex conformation (Figure 5C) and assumes its final conformation via rotation angles of 70° and 20° (Figure 5D–F). Intermittently, the adjacent helix α8 is elongated by up to two turns (Figure 5C–D). Finally, the N-terminally attached SBL closes upon actual cofactor and inhibitor binding.

Figure 5. Cofactor and inhibitor binding mechanism.

Panels A to F show consecutive conformations of saFabI proposed to be adopted during cofactor and inhibitor binding. The three regions undergoing the most extensive conformational changes are highlighted (SBL-2 in cyan, ASL in dark blue and the C-terminally extended SBL in red as in Figure 3). One subunit of TCL-2 is continuously shown as a reference in yellow. (A) Superposition of the apo-1 and TCL-2 structures. Apo-1 is shown in light blue. (B–E) Superposition of the apo-2 and TCL-2 structures. Subunits B, C, A and D of apo-2 are shown in light blue. (F) Final ternary complex conformation. The hydrogen bonded residues Arg194 and Asn205 can act as a hinge for SBL-1 closure and are displayed in a larger scale in the black box (Maity et al., 2011). (G) Peptide flip mechanism and SBL-2 closure observed upon superposition of TCL-2 (yellow) with apo-2 (light blue, subunit A). The carbonyl oxygen of Ala95 flips between Ser121 in the apo structure and a water molecule bound to Ser197 in the ternary complex (a rotated view is displayed in the insert). See also Figure S4 and Movie S1.

During this process, the ASL is visible inside the ligand-binding pocket (Figure 5D) and SBL-2 adopts an entirely α-helical conformation (Figure 5B). Subsequently, SBL-2 is split into the 3₁₀-helices η2 and η3 and adopts a more open conformation (Figure 5D,G). A comparison with the ternary complex structure TCL-2 shows that a backbone flip of Ala95 propagates the movements of adjacent residues and leads to the subsequent closure of SBL-2 (Figure 5G). This peptide flip may be induced by the ordering of Ser197 located on the SBL, which allows the formation of a water-mediated hydrogen bond interaction with the carbonyl oxygen of Ala95. Simultaneously, the movement of Phe96 creates space for the binding of the TCL Bring and establishes an elongated acyl-channel. During NADP⁺ binding, a similar peptide flip mechanism was observed for the NADPH-dependent β-ketoacyl-ACP reductase ecFabG, which also belongs to the SDR family of proteins (Price et al., 2004).

We speculate that the concerted closure of both SBLs, instead of just the classical SBL, is the conformational change corresponding to the slow step of the slow-binding inhibition (Lu and Tonge, 2008). Based on sequence similarities, this mechanism should mainly be feasible for Staphylococci, several Bacilli and some Listeria strains since they contain a serine at position 197, which connects the SBLs (Figure 5G). Most other Gram-negative and Gram-positive bacteria harbor an alanine at this position and, thus, lack the required hydrogen bond acceptor (Figure 3). Consistently, all structurally characterized FabIs from Gram-negative organisms contain just a single flexible SBL (PDB codes: 2JJY, 2P91, 2WYU, 3EK2, 3GRK, 3K2E), as opposed to Bacillus FabIs (3OIF, 3OJE) and the related proteins bsFabL (3OIC) and ecFabG (1I01) which all feature three disordered or rearranged regions in their apo structure.

Dimer - tetramer transition

The apo FabI structures from Gram-negative bacteria contain homo-tetramers, whereas bcFabI/baFabI, in the absence of cofactor and inhibitor, were reported to be dimers in the crystal and in solution (Kim et al., 2010). So far, reports regarding the oligomeric state of saFabI have been inconsistent. It was noted that the published apo structures of saFabI (3GNS, 3GNT) are comprised of dimers, though the enzyme was found to be a tetramer in solution (Kim et al., 2010; Priyadarshi et al., 2010). Therefore two questions arose: What is the oligomeric state of saFabI prior to ligand binding? If dimers are present, does tetramerization and active site formation occur in a concerted fashion with cofactor and inhibitor binding as hypothesized for bsFabL (Kim et al., 2011)?

Our results clearly show that NADP⁺ and triclosan binding to dimeric saFabI induces simultaneous active site formation and tetramerization. In contrast to a previous report but in accordance with the respective apo saFabI structures (3GNS, 3GNT) (Priyadarshi et al., 2010), we identified a dimer through analytical size exclusion chromatography (SEC) experiments at pH 8.0 (Figure S3A), whereas tetramers are present at pH 5.6 (see Experimental Procedures for details). We performed further analytical SEC experiments with the dimeric enzyme in its apo form or incubated with NADP⁺, NADPH or NADP⁺ and TCL. Our results unambiguously demonstrate a transition between dimers and tetramers upon both inhibitor and cofactor binding (Figure S3B), whereas binding of oxidized or reduced cofactor alone is not sufficient to induce tetramerization.

A structure and interface analysis of all available saFabI structures provides snapshots of the dimer - tetramer transition in the order 3GNT, 3GNS, apo-1, apo-2, TCL-2 (Figure 6 and Movie S2). During this process, the buried surface area of the QR-interface, about which the association occurs, increases significantly, as does the number of hydrogen bonds and salt bridges between interface residues (Figure 6C). The QR-interface is established via a 4-helix bundle comprising the long helices α4/α5 (kinked) and α6 (Figure 6, last column). Additionally, the corresponding N-terminal extensions η3 and η4 further stabilize this interface (Figure 3). Since the SBL-2 and ASL regions including these 3₁₀-helices are disordered in our apo structures, the QR-interface is destabilized and the dimers move slightly (~2 Å) away from one another. During the ordering of the ASL, the full tetramer interface and the catalytic active site are simultaneously established (Figure S4). The 3GNS structure clearly contains dimers due to the additionally disordered α4 region (Figure 3). This helix is therefore essential for the maintenance of the QR-interface. Whereas helix α6 of 3GNS is almost completely structured and just slightly shifted, the N-terminal portion of this helix is unstructured and partially disordered in 3GNT.

Figure 6. Dimer - tetramer transition.

From left to right the QR-interface in the saFabI crystal structures 3GNT, 3GNS (Priyadarshi et al., 2010), apo-1, apo-2 and TCL-2 is shown to illustrate tetramer association and dissociation (view as in Figure 2A). The two associating dimers are shown in grey and yellow with the respective QR-interface residues (according to the PDBe PISA server (Krissinel, 2010)) in cyan and red. (A) Front view of the dimer - tetramer transition. (B) Side view of the dimer - tetramer transition. (C) Close up view of the upper QR-interface half. Buried surface areas of the monomeric QR-interface (BSA_QR) as well as the numbers of the corresponding hydrogen bonds (N_HB) and salt bridges (N_SB) are given below according to PISA. See also Figures S3 and S5 as well as Movie S2.

Cooperativity

We hypothesize that the dimer - tetramer transition also applies to substrate binding, representing the molecular basis for the previously reported positive cooperative binding of NADPH to saFabI (Heath et al., 2000). Our data clearly show that NADPH binds cooperatively to saFabI with an approximate Hill coefficient h = 2 (Figure S5). Additionally, trans-2-octenoyl-CoA bound cooperatively to saFabI with a Hill coefficient h = 2.3. However, glutamate, which enhances the catalytic activity of the enzyme (Supplemental Discussion), was found to alter the cooperativity of binding to saFabI. In the presence of 1 M potassium glutamate, no cooperativity was observed for the binding of NADPH or trans-2-octenoyl-CoA to saFabI. Since cooperativity was still observed in the presence of 250 mM potassium glutamate, which is an upper limit estimate of the average S. aureus intracellular glutamate concentration (Björklind and Arvidson, 1978; Blanche et al., 1996), it is likely to be physiologically relevant. Mechanistically, the observed positive cooperativity of substrate binding could be transferred via the ligand-interacting SBL-2 and ASL regions and the attached long helices α4 and α6, which primarily constitute the QR-interface. The formation of such an interface would enhance cofactor and inhibitor/substrate binding due to the rearrangement of the two loop regions and vice versa.

Cooperativity has, to our knowledge, never been observed for NADH-dependent wild-type FabI enzymes, which already exist as tetramers in their apo forms (Heath et al., 2001). In contrast, a similar FabI tetramerization process might be possible for Bacillus species as α4 is additionally disordered in the dimeric bcFabI/baFabI apo structure (PDB code: 3OJE:A) (Kim et al., 2010). Similarly, the twisted tetramer architecture of the bsFabL apo structure (3OIC:A) comprises N-terminally shifted α4 and α6-helices and therefore links this protein to saFabI (Kim et al., 2011).

Cofactor specificity

The studies described above reveal fundamental differences between saFabI and classical FabI proteins. Hence, we also investigated cofactor specificity since, in contrast to all Gram-negative FabI homologues studied thus far, saFabI prefers NADPH instead of NADH as the reducing agent (Heath et al., 2000; White et al., 2005). Our structural analysis confirmed that the positively charged side chains of Arg40 and Lys41 are important for binding the additional 2′-phosphate of NADPH (Priyadarshi et al., 2010; Xu et al., 2008) (Figure 7A). However, we identified an RKXXS-motif that confers this unique specificity to saFabI. While the R40Q/K41N saFabI double mutant exhibited a drastically decreased k_cat/K_m value for NADPH, the specificity constant remained low for NADH (Xu et al., 2008). Our structures revealed Ser44 as a third crucial residue for NADPH specificity. The serine hydroxyl group forms a hydrogen bond to one of the phosphate oxygen atoms at a distance of 2.9 ± 0.1 Å, while the Ser44-bound phosphate oxygen forms additional hydrogen bonding interactions with the backbone amide nitrogens of Arg40 and Lys41, which are located in a tight loop between β2 and α2 (Figure 7A).

Figure 7. Structural basis of NADPH specificity.

(A) The 2′-phosphate binding motif. NADP⁺ is shown in grey, saFabI (CPP structure) in yellow. Additional parts of the protein are omitted for clarity. (B) Superposition of saFabI and ecFabI. SaFabI is shown in yellow, ecFabI (1QSG:A) in green. Asn41 is shifted 2.8 ± 0.3 Å towards the solvent compared to Lys41 preventing 2′-phosphate stabilization by an Asn41 backbone interaction.

Unlike saFabI, the enoyl-ACP reductase from E. coli (ecFabI) is NADH-dependent (Stewart et al., 1999). Superposition of saFabI with ecFabI shows that Leu44 and Gln40 clash with the 2′-phosphate of a hypothetically bound NADPH (shortest distances of 1.7 ± 0.1 and 1.5 ± 0.2 Å, Figure 7B). Bulky substituents at position 44 (numbers according to saFabI), such as Leu or Phe (Figure 3), appear to shrink the adenine ribose-binding pocket, thus decreasing the affinity towards NADPH.

To test the hypothesis that the RKXXS-motif is responsible for cofactor specificity, we generated a triple mutant with all three phosphate-interacting amino acids replaced by the respective E. coli residues (R40Q/K41N/S44L). Interestingly, compared to results reported in the absence of glutamate (Xu et al., 2008), there is only a 2.5-fold increase in the specificity constant of wild-type saFabI for NADPH relative to NADH in the presence of glutamate (Table 2). The glutamate-containing conditions also strongly affected the specificity constant of the R40Q/K41N mutant for NADH, which recovered to that of the wild-type enzyme for NADPH. Additionally, the specificity constant of the double mutant for NADH was 200-fold higher than for NADPH, compared to the 19-fold difference reported in the absence of glutamate. The triple mutant displayed an even more dramatic inversion of cofactor specificity with the specificity constant for NADPH nearly 10,000-fold lower than for NADH. This was driven mostly by a decrease in the ability to utilize NADPH, consistent with our prediction that Leu44 would clash with the 2′-phosphate. Interestingly, the K_m,NADPH of wild-type saFabI is higher than the K_m,NADH of the triple mutant (Tables 2 and S3) and other FabI homologues (Parikh et al., 1999; Sivaraman et al., 2003) which correlates well with the estimated higher microbial intracellular pool of NADPH compared to NADH (Bennett et al., 2009; Liebeke et al., 2010).

Table 2.

Kinetic parameters for saFabI wild-type, double and triple mutant utilizing NADPH (top) and NADH (bottom) at saturating trans-2-octenoyl-CoA concentrations (150 μM).

	Km,NADPH (μM)	kcat (min−1)	kcat/Km,NADPH (min−1 μM−1)
wt saFabI	189.4 ± 20.9	2171.2 ± 123.3	11.5 ± 1.42
saFabI R40Q/K41N	> 300		0.066 ± 0.0012
saFabI R40Q/K41N/S44L	> 300		0.0038 ± 0.00022

	Km,NADH (μM)	kcat (min−1)	kcat/Km,NADH (min−1 μM−1)
wt saFabI	> 300		4.7 ± 0.22
saFabI R40Q/K41N	139.8 ± 23.2	1852.1 ± 132.6	13.2 ± 2.4
saFabI R40Q/K41N/S44L	72.2 ± 26.4	1846.6 ± 266.3	25.6 ± 10.1

¹Agrees with the linear slope (11.1 ± 0.3) of the Micha elis-Menten plot at low [NADPH]

²Estimated based on linear slope of the Michaelis-Menten plot at low [NAD(P)H]

Using BLAST and PHI-BLAST searches (Altschul et al., 1990) with the newly identified RKXXS-motif, we found that this preference for NADPH links saFabI to all Staphylococcus and some Bacillus FabIs, as well as FabL and FabG, rather than to classical NADH-dependent FabIs.

Branched-chain substrate specificity

What is the underlying biological significance of the structural variations we observed for the FabIs from Staphylococci and Bacilli? Notably, the membranes of Staphylococcus and Bacillus genera contain branched-chain fatty acids (BCFA) as a major component (Kaneda, 1991). In contrast, most other bacteria (Kaneda, 1991) predominantly synthesize straight-chain fatty acids (SCFA). For the more common SCFA family members, membrane fluidity is controlled by unsaturated fatty acids (UFA), whereas, for BCFA family members, 12-methyltetradecanoic acid (anteiso-C₁₅) primarily fulfills this purpose (Kaneda, 1991). Thus, we speculated that the increased mobility of FabI from Staphylococci and Bacilli allows the preferential binding of branched-chain substrates.

In addition to different straight-chain substrates (see Supplemental Results), we analyzed two branched-chain substrate analogues - trans-5-methyl-2-hexenoyl-CoA and (±)-trans-4-methyl-2-hexenoyl-CoA, which contains the precursor analog to the anteiso fatty acids found predominantly in the S. aureus membrane (Parsons et al., 2011). Among the first-round substrates (Figure 1), the ratio of specificity constants is approximately 1:24:1 (straight = trans-2-butenoyl-CoA: iso: anteiso) (Table 3), although the substrate specificity of the anteiso-substrate may be higher depending on the stereospecificity of the enzyme. Considering that FabI is the rate-limiting enzyme in the FAS II pathway (Heath and Rock, 1995), this ratio is likely to be physiologically significant in determining the composition of fatty acids synthesized and incorporated into the cell membrane. Moreover, since the pool of first-round substrates entering the FAS II cycle is generated by the branched-chain specific S. aureus FabH enzyme, the synthesis of BCFAs should be highly favored in this pathway (see Supplemental Results for further details) (Qiu et al., 2005). To substantiate the relevance of this observation, we explored the branched-chain substrate specificity of the NADH-dependent FabI enzyme from the Gram-negative pathogen F. tularensis (ftFabI). The ratio of specificity constants (Table S3) for the shortest chain substrates is approximately 800:72:1 (straight: iso: anteiso) and hence fundamentally different from saFabI.

Table 3.

Kinetic parameters for wild-type saFabI utilizing various trans-2-enoyl-CoA substrates. See also Figure S1 and Table S3.

	Km,acyl-CoA (μM)1	kcat (min−1)1	kcat/Km,acyl-CoA (min−1 μM−1)
trans-2-butenoyl-CoA	1278.3 ± 120.2	40.1 ± 1.8	0.031 ± 0.003
trans-2-hexenoyl-CoA	138.6 ± 18.8	588.3 ± 25.3	4.2 ± 0.6
trans-2-octenoyl-CoA	20.7 ± 2.7	2645.4 ± 107.3	127.8 ± 17.5
trans-2-decenoyl-CoA	29.1 ± 7.5	323.7 ± 29.5	11.1 ± 3.0
trans-5-methyl-2-hexenoyl-CoA (iso-)	307.0 ± 41.5	172.9 ± 8.1	0.56 ± 0.08
(±)-trans-4-methyl-2-hexenoyl-CoA (anteiso-)	599.2 ± 119.1	13.8 ± 1.2	0.023 ± 0.005

¹K_m,acyl-CoA and k_cat values were determined at a fixed NADPH concentration (350 μM)

Discussion

Our combined results clearly indicate that saFabI possesses distinct characteristics that markedly differentiate it from classical FabI enzymes, while exhibiting similarities to Staphylococcus and Bacillus FabIs and FabL. We identified an RKXXS-motif, which is responsible for the preferential binding of NADPH to saFabI and most likely to FabI proteins of other Staphylococci and some Bacillus species. Our analysis also reveals a fundamental extension of the classical inhibitor binding mechanism including conformational changes in three regions and a dimer - tetramer transition upon cofactor and inhibitor binding that may also be applicable for FabI enzymes of related Staphylococcus and Bacillus species. The oligomeric transition is coupled to ligand binding and hence may provide the mechanistic underpinnings for the observed cooperativity upon substrate binding. Glutamate was shown to influence cofactor differentiation and to increase the activity of saFabI while reducing the cooperativity of substrate binding. Nevertheless, cooperativity was still observed at a physiologically relevant glutamate concentration, suggesting that glutamate may function as an intracellular metabolite that regulates enzyme activity (for further details see Supplemental Discussion and Figure S5).

In contrast to many other bacteria including Streptococci, the cell membranes of Staphylococcus and Bacillus species mainly comprise branched-chain fatty acids (Balemans et al., 2010; Brinster et al., 2009; Fozo and Quivey, 2004; Kaneda, 1991; O’Donnell et al., 1985; Parsons et al., 2011). We propose that S. aureus FabI, as well as Bacillus FabI and FabL proteins, have evolved differently compared to classical FabI proteins to enable the production of BCFAs. The additional flexibility observed in the saFabI structures and other characterized Bacillus FabIs may play a role in altering the substrate specificity to accommodate the more bulky branched-chain substrates. Accordingly, Ser197 in saFabI, which links the two flexible substrate binding loops after ligand binding (Figure 5G), is conserved in Staphylococci, several Bacilli and some Listeria. Furthermore, we found that FabI enzymes capable of utilizing NADPH belong mostly to the BCFA family of organisms, however, the rationale behind this observation is currently not clear. To generate a membrane containing mostly branched-chain fatty acids, S. aureus and B. subtilis FabH enzymes are known to prefer branched-chain acyl-CoA primers instead of acetyl-CoA (Figure 1, Supplemental Results) (Choi et al., 2000; Qiu et al., 2005). Our data clearly show that this preference is carried on in the FAS II cycle as the substrate specificity for the first-round branched-chain substrates relative to the first-round straight-chain substrate is markedly increased for saFabI compared to FabIs from organisms producing mainly SCFAs.

Since BCFAs are known to be important for the in vivo fitness of S. aureus and Listeria monocytogenes (Singh et al., 2008; Sun and O’Riordan, 2010), the requirement of some organisms for these fatty acids may explain the different susceptibility of Gram-positive bacteria to inhibitors of the fatty acid biosynthesis pathway. Importantly, only a minimal amount of BCFAs are present in the human and murine blood (1% of all plasma fatty acids in humans (Holman et al., 1995) and less than 1% in mice (Atshaves et al., 2005; Gloerich et al., 2005)). It is known that the modulation of membrane lipids enables bacteria to survive under distinct stress situations (Singh et al., 2008). Even though both types of fatty acids serve to increase the fluidity of the membrane bilayer, the structural and morphological membrane characteristics differ for BCFAs and UFAs (Legendre et al., 1980). Hence, compared to UFAs, supplementing with BCFAs did not confer the same level of fitness to E. coli UFA auxotrophs in response to cold stress (Silbert et al., 1973). Similarly, other evidence suggests that S. aureus and L. monocytogenes fitness is reduced in the absence of BCFAs. The survival of branched-chain α-ketoacid dehydrogenase (BKD) deficient mutant strains in murine animal models was significantly reduced (Singh et al., 2008; Sun and O’Riordan, 2010). Nevertheless, future experiments should clarify whether the defined composition of the human blood lipid pool, comprising mainly SCFAs and UFAs, can fulfill the in vivo survival and virulence requirements of Staphylococci and other BCFA family members (Brinster et al., 2009; Holman et al., 1995).

As the essentiality of the FAS II pathway for S. aureus infection in vivo is re-validated, a detailed structural and kinetic characterization of targets within the pathway becomes increasingly relevant. FabI is one such target, which has spurred many inhibitor discovery efforts. We have shown that the drug target FabI in S. aureus and closely related pathogens differs from its homologous proteins with respect to cofactor and substrate specificity, cofactor/inhibitor binding and quaternary structure. Knowledge of these differences, as well as the structures solved here and in similar studies, can aid in the development of antibiotics specifically targeting FabI enzymes from the important human pathogens S. aureus, S. epidermidis, B. anthracis and B. cereus, which all synthesize branched-chain fatty acids. Remarkably, three saFabI inhibitors are in phase I clinical trials since 2009, providing substantial hope for new MRSA drugs (Gerusz, 2010).

Experimental Procedures

Cloning, expression and purification

We optimized the saFabI purification protocol with respect to plasmid usage, E. coli expression strain, cultivation and buffer composition (see Supplemental Experimental Procedures for further details) based on previously described procedures (Priyadarshi et al., 2010; Xu et al., 2008). A detailed description of the protocols is provided in Table S4. FtFabI was prepared as described previously (Lu et al., 2009).

Site-directed mutagenesis, expression and purification of saFabI mutants

Site-directed mutagenesis was performed using the QuikChange Mutagenesis Kit from Stratagene to add a single point mutation (S44L) to the previously constructed saFabI R40Q/K41N mutant (primers: 5′-ACCAGAACGAACGTCTGCGTAAAGAGCTTGAA-3′ and 5′-TTCAAGCTCTTTACGCAGACGTTCGTTCTGGT-3′) (Xu et al., 2008). The sequence of saFabI R40Q/K41N/S44L was confirmed by DNA sequencing. Expression and purification of the double and triple mutant was performed as described in Table S4.

Crystallization, data collection and structure determination

Prior to co-crystallization utilizing the vapor diffusion method, purified saFabI was incubated for 2 h at 4 °C with a 10-fold molar excess of NADP⁺ (100 mg/ml stock solution in water) and a 20-fold molar excess of the respective inhibitor supplemented as solid powder or dissolved in DMSO (for the P1 structures, 100 mg/ml stock solutions), respectively. The complexes were crystallized by the hanging drop vapor diffusion method using different precipitants (see Supplemental Experimental Procedures). Diffraction data of the flash-frozen crystals were collected at BESSY II (Berlin) or the ESRF (Grenoble), respectively. All structures were solved by molecular replacement with baFabI (2QIO) as the initial search model (see Supplemental Experimental Procedures and Tables S1 and S2 for further details).

Analytical size exclusion chromatography

SaFabI was dialyzed into a buffer containing 20 mM Tris pH 8.0 and 200 mM NaCl. Equal sample volumes were applied to a calibrated 10/300 Superdex 200 GL column (GE Healthcare) pre-equilibrated with dialysis buffer. Prior to analytical SEC the protein (5 mg/ml) was supplemented with a 10-fold molar excess of NADP⁺, NADPH or an equal volume of water and a 20-fold molar excess of TCL (100 mg/ml stock solution in DMSO) or an equal volume of DMSO, respectively. All four samples (apo, NADP⁺, NADPH and NADP⁺/TCL) were incubated at 20 °C for 2 h. Experiments were performed in triplicate and the peak shift confirmed for a different protein batch at pH 8.0 (Figure S3B). Additionally, one saFabI sample was analyzed at pH 5.6 in 20 mM trisodium citrate pH 5.6, 500 mM NaCl and 100 mM EDTA (Figure S3A).

Synthesis of trans-2-enoyl-CoA substrates

Trans-5-methyl-2-hexenoic acid and (±)-trans-4-methyl-2-hexenoic acid were synthesized from isovaleraldehyde (Sigma-Aldrich) and (±)-2-methylbutyraldehyde (Sigma-Aldrich), respectively, via a Horner-Wadsworth-Emmons reaction with methyl (triphenylphosphoranylidene)acetate, as described previously (Ryzhkov, 1996). Trans-2-enoyl-CoA substrates were synthesized from their respective trans-2-enoic acid using the mixed anhydride method (Parikh et al., 1999). Products were characterized by electrospray ionization mass spectrometry. Trans-2-butenoyl-CoA was purchased from Sigma-Aldrich.

Steady-state kinetic assays

Kinetic experiments were performed on a Cary 100 spectrophotometer (Varian) at 20°C in 50 mM potassium phosphate, 150 mM NaCl and 1 M potassium glutamate pH 7.5 containing 8% glycerol (v/v) and 0.1 mg/mL bovine serum albumin (BSA). Reaction velocities were measured by monitoring the oxidation of NAD(P)H to NAD(P)⁺ at 340 nm (ε = 6220 M⁻¹ cm⁻¹). Kinetic parameters for saFabI were determined by measuring initial velocities at varying concentrations of one substrate, holding the other substrate concentration constant (for further details see Supplemental Experimental Procedures).

Inhibition kinetics

Progress curves were used to determine the steady-state inhibition of saFabI by the slow-onset inhibitors TCL, EPP and CPP. The final reaction mixture contained saFabI (100 nM), trans-2-butenoyl-CoA (1.5 mM), NADPH (350 μM), NADP⁺ (400 μM) and inhibitor (2% v/v DMSO). K_i*^,app values, corresponding to the final steady-state inhibition following slow isomerization of the initial enzyme-inhibitor complex, were determined by plotting the fractional steady-state velocities as a function of inhibitor concentration and fitting to the standard isotherm equation (Equation 1).

$$\frac{v_s}{v_u}=\frac{1}{1+\frac{[I]}{K_i^{\ast ,\mathit{app}}}}$$ (1)

where v_u is the initial velocity of the uninhibited reaction (for further details see Supplemental Experimental Procedures).

Highlights.

• Insights into saFabI ligand binding including a dimer-tetramer transition

• Identification of a loop motif that determines altered cofactor specificity

• Increased flexibility modulates substrate and inhibitor recognition

• Ability of saFabI to reduce branched-chain fatty acid precursor molecules