Structure-based discovery of a small-molecule inhibitor of methicillin-resistant Staphylococcus aureus virulence

The rapid emergence and dissemination of methicillin-resistant Staphylococcus aureus (MRSA) strains poses a major threat to public health. MRSA possesses an arsenal of secreted host-damaging virulence factors that mediate pathogenicity and blunt immune defenses. Panton–Valentine leukocidin (PVL) and α-toxin are exotoxins that create lytic pores in the host cell membrane. They are recognized as being important for the development of invasive MRSA infections and are thus potential targets for antivirulence therapies. Here, we report the high-resolution X-ray crystal structures of both PVL and α-toxin in their soluble, monomeric, and oligomeric membrane-inserted pore states in complex with n-tetradecylphosphocholine (C14PC). The structures revealed two evolutionarily conserved phosphatidylcholine-binding mechanisms and their roles in modulating host cell attachment, oligomer assembly, and membrane perforation. Moreover, we demonstrate that the soluble C14PC compound protects primary human immune cells in vitro against cytolysis by PVL and α-toxin and hence may serve as the basis for the development of an antivirulence agent for managing MRSA infections.

Infection with Staphylococcus aureus can cause severe and devastating illness and is one of the leading causes of death by any infectious agent in the United States (1,2). S. aureus is notorious for its ability to acquire genetic determinants of antibiotic resistance and virulence that enhance fitness and pathogenicity (3,4). Methicillin-resistant S. aureus (MRSA) 2 now accounts for Ͼ60% of S. aureus isolates in United States intensive care units, severely restricting antibiotic treatment options (2). MRSA also spreads rapidly among healthy individuals in the community, causing predominantly skin and soft tissue infections and life-threatening infections, including bacteremia, endocarditis, osteomyelitis, and necrotizing pneumonia (2). Disturbingly, MRSA can live in the biofilm state (5,6), and it has long been recognized that biofilms increase resistance to antimicrobial agents and the host immune response (7). MRSA is currently treated with vancomycin, clindamycin, linezolid, and daptomycin (8), but resistance to these "last-resort" antibiotics has been reported (9 -13). For these reasons, the World Health Organization identifies MRSA as one of six "high-priority" pathogens that pose an enormous threat to public health (14). Thus, new therapeutics with novel mechanisms of action are desperately needed to combat this high-threat pathogen.
USA300 is the most prevalent strain of MRSA in the United States and represents a growing threat in both community and healthcare settings (15). Its heightened virulence and severity are related to the production of a mixture of cytolytic poreforming exotoxins that mediate virulence and impair host immune defenses (3,16). The pharmacological targeting of these cytotoxins has been recognized as a promising new therapeutic approach to reducing morbidity and mortality associated with MRSA infection. Leukocidins and ␣-toxin, secreted by S. aureus as water-soluble, monomeric polypeptides, constitute the ␣-hemolysin subfamily of ␤-barrel pore-forming toxins (17). Five different bipartite leukocidins have been described, including Panton-Valentine leukocidin (PVL), leukocidin ED (LukED), two ␥-hemolysins (HlgAB and HlgCB), and leukocidin AB (LukAB; also known as LukGH), each of which consists of two distinct polypeptides referred to as the S and F subunits (reviewed in Ref. 18). Their cellular tropism and species specificity are determined by the S subunits LukS-PV, LukE, HlgA, HlgC, and LukA (19 -21). The S and F subunits and singlecomponent ␣-toxin share a unique modular structure consisting of the amino latch and prestem regions and the ␤-sandwich and rim domains (see Fig. 1A) (22)(23)(24)(25)(26). The X-ray crystal structures of the membrane-inserted pore oligomer forms of ␣-toxin, HlgAB, and LukGH and of the membrane surfacebound prepore heterooctamer forms of HlgAB and HlgCB have been determined (27)(28)(29)(30). These structures, and supporting biochemical and genetic data (26,(31)(32)(33), suggest that members of this subfamily share a common mechanism of cytolytic action (reviewed in Refs. 34 and 35). The cytolytic process begins with the binding of soluble toxin monomers to a cell surface receptor (21,36). The membrane-bound monomers then associate to form a nonlytic, oligomeric prepore.
Finally, the translocation of the prestem regions across the membrane results in the bilayer-spanning ␤-barrel pore structure and consequent membrane permeabilization and cell lysis.
MRSA strains that harbor the phage-encoded PVL have been linked to highly virulent and severe community-acquired skin infections (37), as well as necrotizing pneumonia and lethal necrotizing fasciitis (38). The role of PVL production in the pathogenesis of MRSA was demonstrated in both a rabbit model of necrotizing pneumonia and humanized mouse models of skin infection and pneumonia (39 -41). PVL induces leukocyte destruction and tissue necrosis through interaction with the complement receptors C5aR and C5L2 (42)(43)(44)(45). PVL, in conjunction with HlgAB, contributes to MRSA biofilm-mediated killing of neutrophils (46). On the other hand, the chromosomally encoded ␣-toxin lyses epithelial and endothelial cells, red blood cells, lymphocytes, and monocytes by targeting its receptor, the metalloprotease ADAM10 (36,47). The elevated expression of ␣-toxin in the USA300 clone and in historic human epidemic strains correlates with increased pathogenicity in mouse models of skin and soft tissue infection, pneumonia, and sepsis (48,49). ␣-Toxin also plays a role in biofilm formation by clinical MRSA isolates (50). Moreover, LukED relies on the chemokine receptor CCR5 to kill T lymphocytes, macrophages, and dendritic cells, as well as CXCR1 and CXCR2 to kill leukocytes (19,51). Inhibition of the interaction between LukED and CCR5 has been shown to block cytotoxicity and attenuate S. aureus infection in mice (19). Together, these cytotoxins can modulate phagocytic cell functions via their specific receptors and contribute to MRSA immune evasion and disease pathogenesis. As such, the discovery and development of new antivirulence agents that protect from the combined immune cytolytic activities of this subfamily of pore-forming toxins is of the utmost importance.
There is considerable evidence pointing to the role of phosphatidylcholine (PC) in the mechanism of pore formation by these toxins. PC is an absolute requirement for pore formation by ␣-toxin, HlgAB, and HlgCB and has been shown to inhibit their cytolytic effects (52)(53)(54)(55)(56). Particularly, crystallographic studies revealed the presence of single, highly conserved phosphocholine (PCho) binding sites on the rim domains of the monomeric F subunit HlgB and the ␣-toxin protomer in the heptameric pore complex (22,57). These binding sites have been shown by mutational analysis to be required for membrane targeting and cytolytic function of the two toxins (32,58). It is generally accepted that ␣-toxin and the F subunits LukD, LukF-PV, LukB, and HlgB also function in cell attachment through the engagement of their rim domains with the PC headgroup in the plasma membrane of target cells (52,54,57). In this report, we demonstrate that the soluble, monomeric, and oligomeric pore forms of both PVL and ␣-toxin deploy two distinct modes to recognize and bind the PC-containing membrane and suggest a novel molecular mechanism for PC-dependent pore formation by members of this subfamily. Furthermore, we find that the soluble compound n-tetradecylphosphocholine (C 14 PC) effectively inhibits the cytolysis of primary human immune cells by PVL, ␣-toxin, and LukED in vitro, thus demonstrating the potential utility of this antivirulence agent alone or in combination with antibiotics against MRSA.

C 14 PC binds to the rim domain of LukD at two adjacent but distinct sites
To better understand the molecular basis for the recognition of PCho by the leukocidin F subunits, we determined the crystal structures of LukD with and without C 14 PC at 1.5 and 1.75 Å resolution, respectively (Table 1). C 14 PC was selected in the present study as a PC mimic for its high micellization efficiency due to low critical micelle concentration. The two protein structures are closely similar, with an r.m.s.d. for C ␣ atoms of 0.69 Å. The rim domain forms an antiparallel, three-stranded open-face ␤-sandwich toppled by two surface-exposed consecutive ⍀ loops (residues 180 -194, ⍀1 and 195-202, ⍀2) (Fig.  1A). Two PCho moieties that bind to opposite sides of the ⍀2 loop were unexpectedly discovered upon examination of the difference electron density map in the C 14 PC-bound structure (Fig. 1B). The average B-factor for these two moieties is 22 Å 2 , and that for the surrounding solvent molecules and protein atoms is 21 Å 2 . The two binding sites are ϳ16 Å apart (Fig. 1C). The PCho moiety at the first binding site (site 1) is lodged into a concave pocket similar to one in HlgB (PDB code 3LKF). This pocket is formed by two extended segments (residues 171-173 and 176 -179, respectively) and the ⍀1-⍀2 junction (residues 191-197) (Fig. 1, A and D). The quaternary ammonium group of the PCho moiety engages in a cation-interaction with Trp 176 while forming a salt bridge to Glu 191 (3.79 Å) (Fig. 1D). Its N-methyl and methylene groups are in van der Waal contact (Ͻ4.0 Å) with the main-chain atoms of Asn 173 , Glu 191 , Leu 194 , and Gly 195 and with the side chains of Asn 173 , Trp 176 , Tyr 179 , and Glu 191 . Furthermore, the phosphate group is hydrogenbonded through its O2 oxygen to the main-chain amide of Arg 197 (2.85 Å) on one side of the pocket opening, and the side-chain of this residue also wraps around the three other oxygens (Fig. 1D). In addition, three water molecules form hydrogen bonds to the O2 and O3 oxygens.
Immediately adjacent to site 1 is a novel second binding site (site 2), where the PCho moiety occupies a shallow surface pocket that is framed by the C-terminal half of the ⍀2 loop (residues 198 -202) and the ␤14 -␤15 loop (residues 257-260) and flanked by the side chains of Tyr 71 , Asn 72 , Trp 256 , and Trp 261 (Fig. 1, A and D). The quaternary ammonium group is sandwiched between the aromatic rings of Tyr 71 and Trp 256 through cation-interactions, and the two indole rings of the latter residue and Trp 261 interact with each other in an edgeto-face fashion to engage the N-methyl and methylene groups, which also make contacts with the main-chain atoms of Ser 199 , Ser 200 , and Ser 201 and with the side chain of Asn 72 (Fig. 1D). The phosphate group is secured by a water-mediated hydrogen bonding interaction with the main-chain carbonyl of Ser 200 (O 2 -H 2 O ϭ 2.53 Å and H 2 O-O ϭ 2.76 Å), whose C ␣ and C ␤ atoms pack against the O1, O2, and O4 oxygens (Fig. 1D).
The highly complementary interactions between the two adjacent binding sites and the PCho moieties are ostensibly important for specific recognition and binding. The buried solvent-accessible surface area of PCho is 262 Å 2 at site 1 and 231 Å 2 at site 2, which correspond to ϳ77 and 69% of the unbound PCho surface area, respectively. The side chains of the con-

Discovery of a small-molecule inhibitor of MRSA virulence
served Trp 176 -Arg 197 and Ser 200 -Trp 256 -Trp 261 residues, seen below, that define site 1 and site 2, respectively, become more ordered upon binding to C 14 PC. This side-chain flexibility could allow these two adjacent, largely preformed pockets to efficiently accommodate the PCho moieties that have distinct binding poses and residue interactions (Fig. 1, C and D). Consistent with this argument, in differential scanning calorimetry (DSC) experiments, LukD (10 M) was found to unfold in a single cooperative transition, with a midpoint melting temperature (T m ) of 51.0°C, whereas this T m value was shifted to 52.8°C in the presence of PCho (4 mM), representing the enhanced thermal stability that accompanies complex formation (Fig. 1E). Thus, our results suggest a revised mode of PC recognition and membrane targeting by the rim domain loops.

Binding mode of C 14 PC to the rim domain of LukF-PV
To validate this binding mode, we co-crystallized LukF-PV with C 14 PC and solved its structure at 1.78 Å resolution (Table  1). In effect, PCho moieties engage the aforementioned two adjacent binding pockets on the rim domain surface (Fig. 2, A and B). At site 1, the quaternary ammonium group of the PCho moiety forms both a cation-interaction with Trp 176 and a salt bridge to Glu 191 (3.84 Å); its N-methyl and methylene groups interact with both the main-chain atoms of Leu 194 and Gly 195 and the side chains of Asn 173 , Trp 176 , Tyr 179 , Glu 191 , and Arg 197 , and the phosphate group is held in place by a hydrogen bond between its O2 oxygen and the main-chain amide of Arg 197 (2.72 Å), along with the side chain of this residue lying against the O2 and O3 oxygens (Fig. 2B). At site 2, the quaternary ammonium group participates in a cation-interaction with Trp 256 (Fig. 2B). Further contacts are made between the N-methyl and methylene groups and both the main-chain atoms of Ser 199 , Asn 200 , and Leu 201 and the side chains of Asn 200 , Trp 256 , and Trp 261 . Polar interactions are also observed between the phosphate and both the main-chain atom of Asn 200 and the side chain of Asn 202 (Fig. 2B).
The solvent-accessible surface area of PCho buried by the LukF-PV interaction comprises 264 Å 2 (79%) at site 1 and 214 Å 2 (63%) at site 2. DSC measurements reveal that the T m of

Discovery of a small-molecule inhibitor of MRSA virulence
LukF-PV increased from 50.3 to 52.3°C when it was bound to PCho. We note that the PCho moiety at site 2 has considerably higher average B-factor and poorer electron density than that at site 1 (70 Å 2 as compared with 31 Å 2 ), suggesting that the former moiety is less tightly bound and exhibits greater spatial or temporal disorder. In LukD, the aromatic side chain of Tyr 71 contributes to the cation-binding interaction at site 2 (see Fig. 1D), whereas the corresponding residue in LukF-PV (Thr 71 ) cannot make this interaction ( Fig. 2C), likely accounting for the lower-affinity binding site. The critical functional role of this affinity difference is highlighted by the observation that replacement of Thr 71 with a tyrosine endows LukF-PV with the ability to bind human erythrocytes and acquire hemolytic activity when combined with the S subunit of HlgAB (33). Therefore, the elaborate structural features of the two distinct, adjacent PCho-binding sites on the leukocidin F subunits may be explained by a selective pressure for membrane PC itself acting as their cell surface receptor.

C 14 PC binding by monomeric ␣-toxin H35A
To discern the mechanism in the attachment of ␣-hemolysin subfamily members to host cells, we determined the 2.80 Å crystal structure of C 14 PC in complex with the oligomerization-defective His 35 3 Ala mutant of ␣-toxin (␣-toxin H35A ) (59) ( Table 1). The asymmetric unit contains two nearly identical protein monomers (r.m.s.d. for C ␣ atoms of 0.44 Å), each bound to two PCho moieties (Fig. 3A). These moieties occupy the two adjacent binding pockets described above (Fig. 3B). At site 1, which is similar to that on the ␣-toxin protomer in the heptameric pore complex (57), the quaternary ammonium group of the PCho moiety makes a cation-interaction with Trp 179 (Fig. 3C). Its N-methyl and methylene groups are surrounded by the main-chain atoms of Met 197 and Lys 198 and by the side chains of Asn 176 , Gln 177 , Trp 179 , Tyr 182 , Gln 194 , Met 197 , and Arg 200 . Importantly, the O2 oxygen of the phosphate group establishes a strong hydrogen bond to the main-chain amide of Arg 200 (2.64 Å) that also makes side-chain contacts with the O2 and O4 oxygens (Fig. 3C). At site 2, the quaternary ammonium group forms a cation-interaction with Trp 260 , and the N-methyl and methylene groups interact with the main-chain atoms of Gly 202 , Ser 203 , and Met 204 and with the side chains of Ala 73 , Asn 74 , and Trp 265 (Fig. 3B). The phosphate group is clearly visible in the electron density map, although the fine detail of the oxygens is not clear. There are contacts of 3.19 Å between the phosphate and Ser 203 and of 3.62 Å between the phosphate and Trp 260 (Fig. 3C). Upon binding to ␣-toxin H35A , PCho buries 268 Å 2 (79%) and 203 Å 2 (61%) of its solventaccessible surface area at site 1 and site 2, respectively. DSC analysis shows that the addition of PCho increased the T m of ␣-toxin H35A from 50.8 to 52.4°C. We also observed that the average B-factor for the PCho moiety at site 2 is significantly higher than that at site 1 (112 Å 2 as compared with 75 Å 2 ). As discussed in the preceding section, the decreased affinity of site 2 for PCho may arise from the presence of an alanine at position 73 (corresponding to LukD Tyr 71 ) (Fig. 2C).

Discovery of a small-molecule inhibitor of MRSA virulence
Closer examination of the positions and conformations of the two PCho moieties in the superimposed cocrystal structures of C 14 PC with ␣-toxin H35A , LukD, and LukF-PV revealed remarkable similarities. There are few differences in the positions of the five key binding site amino acid side chains (Trp 179 , Arg 200 , Ser 203 , Trp 260 , and Trp 265 in ␣-toxin; equivalent to Trp 176 , Arg 197 , Ser/Asn 200 , Trp 256 , and Trp 261 in LukD and LukF-PV) in these structures. The three Trp side chains provide two important anchor points for locating the PCho moieties in the two adjacent binding sites, and the Arg and Ser/Asn residues are critical determinants in the binding of the two phosphate groups. Evidently, PC recognition specificity is achieved by a combination of stacking and hydrogen-bonding interactions and van der Waals contacts. Our study shows that membrane PC serves as the common receptor for ␣-toxin and the leukocidin F subunits, in agreement with previous observations (52,54,57). The presence of the two adjacent PC-binding sites on the toxin monomer is consistent with the estimated crosssectional areas of the PC-bound rim domain (ϳ150 Å 2 ) and one PC molecule (ϳ70 Å 2 ) (60).
Intermolecular contacts between the above two ␣-toxin H35A monomers comprising the crystal asymmetric unit are formed by residues in the ␤-sandwich domain (Fig. 3D). Comparison of the conformation of these contact residues with their interprotomeric equivalents in the unliganded and C 14 PC-bound heptamers of WT ␣-toxin (PDB code 7AHL; see Fig. 5) reveals no local conformational changes involving the main-chain or side-chain atoms. Superposition of the ␣-toxin H35A dimer onto two adjacent promoters in the above two WT toxin heptamers yields overall C ␣ r.m.s.d. values of 0.99 and 0.95 Å, respectively, indicating their structural similarity. Dimer interfaces have similar buried surface area values, from 2,061 to 2,171 Å 2 . It is also important to note that the crystal structure of unliganded ␣-toxin H35A (PDB code 4YHD) lacks the aforementioned intermolecular contacts between six independent monomers in the asymmetric unit. In this structure, both the amino latch and prestem regions have well-defined density with the exception of the six-residue prestem loop and pack against the ␤-sandwich core of the protein. By contrast, these two regions are apparently disordered in the C 14 PC-bound structure. Our results suggest that ␣-toxin H35A may be trapped in a PC-bound dimeric state, which may represent an on-pathway intermediate in the assembly of the heptameric pore complex.
Given their expected importance in membrane targeting, the five key PC-binding site residues are highly conserved or invariant in both ␣-toxin and the leukocidin F subunits but are absent in the S subunits, with the exception of a histidine at position 176 in LukB (Fig. 2C). Of particular importance, LukB exists as a soluble heterodimeric complex with LukA (61). This finding is consistent with the central role of the conserved Trp 176 of the three other F subunits in their binding to the PC bilayer (22,52,54) (this study). We therefore propose that the binding of the F subunit to the PC-rich membrane is allosterically coupled to heterodimerization with its S subunit counterpart. Likewise, membrane binding by ␣-toxin, mediated by PC and/or ADAM10, irrevocably commits the monomers to dimerization.

Discovery of a small-molecule inhibitor of MRSA virulence
The remarkable high degree of conservation of the two adjacent PC-binding sites among ␣-toxin and the F subunits reflects a strong selective pressure on the ability of these two sites to help anchor toxin monomers to the cell surface and to form intermolecular contacts that prime the ensuing formation of the oligomeric, membrane-inserted pore complex. In summary, the bivalent rim domain interaction with PC provides a mechanism by which soluble toxin monomers can recognize and target the PC-containing membrane, thereby promoting dimer-nucleated pore assembly. The relatively low affinity of PC-mediated binding may facilitate subsequent establishment of the final geometry of the oligomeric pore complex, which we discuss below. ␣-Toxin and the leukocidin S subunits also bind their cognate proteinaceous receptors (19 -21, 47), and these interactions likely work in concert with the PC-targeting mechanism to modulate toxin binding, pore formation, and cytotoxicity. Finally, and most importantly, struc-tural elucidation of the two conserved, adjacent PC-binding pockets on ␣-toxin and the leukocidin F subunits will guide the rational development of PC analogs as decoy receptors that prevent the cytotoxin from binding to susceptible cells.

Structure of the C 14 PC-bound PVL heterooctamer
In light of previous studies suggesting that PC plays a crucial role in the assembly and function of the ␣-toxin heptamer (54, 57), we co-crystallized the LukS-PV and LukF-PV proteins with C 14 PC in the presence of n-octyl-␤-glucoside. The structure of the complex was solved at 2.04 Å resolution by molecular replacement ( Table 1). The asymmetric unit contains one LukF-PV/LukS-PV heterodimer and a single LukS-PV molecule. The heterodimer interacts with three crystallographic 4-fold symmetry-related copies of itself to generate a heterooctamer (Fig. 4, A and B). In this ␤-barrel pore complex, four LukF-PV protomers (denoted A, C, E, and G) and four LukS-PV

Discovery of a small-molecule inhibitor of MRSA virulence
protomers (B, D, F, and H) are arranged in an alternating fashion around the central axis of the pore, in which the stem domain folds into an antiparallel ␤-barrel composed of 16 ␤-strands. We could not discern electron density corresponding to the bottom third of the stem domain in our structure. Two distinct interfaces between neighboring protomers involve residues that are distributed among the amino latch region and the ␤-sandwich and stem domains and bury 2,644 and 1,902 Å 2 of solvent-accessible surface area, respectively. The electron density map revealed clearly the presence of PCho moieties at three distinct binding sites on each of the four protomeric units of the PVL heterooctamer (Fig. 4, C and D).
The two adjacent binding sites are essentially the same as those on the above-described toxin monomer, whereas the other, novel site lies at the interface between the rim domain of a LukF-PV protomer (e.g. protomer A) and the proximal stem domain regions of protomers G and H. The average B-factor for the three PCho moieties is significantly higher than that for the surrounding residues (60 Å 2 as compared with 31 Å 2 ), possibly due to greater disorder and/or subunitary occupancy. Superposition of the PVL heterooctamer bound to C 14 PC onto the unliganded HlgAB (PDB code 3B07) and LukGH (PDB code 4TW1) heterooctamers yields C ␣ r.m.s.d. values of 0.67 and 1.14 Å, respectively, suggesting that the PVL pore does not undergo large conformational changes upon binding to C 14 PC. The three PCho-binding sites on a single protomeric unit are contained within a water-accessible crevice between the inner surface of the rim domain and the upper portion of the stem domain (Fig. 4, A and B). As noted above, the two adjacent sites correspond to those on the rim domain of monomeric LukF-PV ( Fig. 4D; see Fig. 2B), differing only in the presence of more stabilizing molecular contacts at site 2 on the heterooctamer. Specifically, the quaternary ammonium group of the PCho moiety makes a cationinteraction with Trp 256 , and the indole ring of this residue establishes an edge-to-face interaction with the indole ring of Trp 261 to pack against the N-methyl and methylene groups, which are also in contact with the mainchain atoms of Thr 71 , Ser 199 , Asn 200 , and Leu 201 and with the side chain of Ile 72 (Fig. 4E). At site 3, the aromatic ring of Tyr 137 of protomer H forms a cation-interaction with the quaternary ammonium group and stacks against the N-methyl and methylene groups that are also lined with the side chain of Ile 135 of protomer H (Fig. 4E). Furthermore, the O3 oxygen of the phosphate group hydrogen-bonds to the main-chain amide of Gly 175 of protomer A (2.65 Å), and the O1 and O3 oxygens engage both the main-chain atoms of Asn 174 and Gly 175 of protomer A and the side chains of Met 172 of protomer A and Gln 112 of protomer G (Fig. 4E). The solvent-accessible surface area of PCho buried upon complex formation is 258 Å 2 (77%) at site 1, 189 Å 2 (57%) at site 2, and 224 Å 2 (65%) at site 3.
Our results suggest that multivalent binding of the PVL heterooctamer to PC on the membrane surface leads to localized alterations in the lipid bilayer and thus promotes the insertion of amphipathic ␤-hairpins to produce the ␤-barrel piercing the bilayer. Critical residues Tyr 137 of LukS-PV and Gly 175 of LukF-PV at site 3 are invariant in the leukocidin S and F subunits, respectively (Fig. 2C), underscoring their functional importance. Furthermore, three similar PC-binding pockets also exist in protomers of the C 14 PC-bound ␣-toxin heptamer described below.

Binding mode of C 14 PC to the ␣-toxin heptamer
To evaluate the binding of the ␣-toxin heptamer to the PC headgroup in a membrane-mimicking environment, we determined the crystal structure of its complex with C 14 PC at 2.35 Å resolution (Table 1). In this structure, three PCho moieties are bound to each of the seven protomeric units in the water-accessible crevice between the rim and stem domains (Fig. 5, A  and B). The indole ring of Trp 179 mediates three-way interactions with these three moieties (Fig. 5C). Their conformations are clearly defined in three partially overlapping but distinct binding pockets of the crevice (Fig. 5D). One pocket corresponds to site 1 on the toxin monomer described above, whereas the other two are novel heptamer-specific binding sites (see below). The average B-factor for the three PCho moieties is 60 Å 2 , and for surrounding protein atoms, it is 33 Å 2 . The structure of the C 14 PC-bound heptamer is very similar to that of the unliganded heptamer (PDB code 7AHL; r.m.s.d. for C ␣ atoms of 0.48 Å), with only minor changes in the positions of side chains involved in direct contact with C 14 PC. The pairwise r.m.s.d. values among protomers A-G in the heptamer span a range from 0.13 to 0.17 Å for C ␣ atoms. The PCho moieties at each of the three binding sites have essentially identical conformation and orientation in each of the seven protomeric units, with average r.m.s.d. values of 0.34 Å for the first pocket, 0.32 Å for the second pocket, and 0.41 Å for the third pocket. For this reason, the following structural analysis of these binding pockets applies to all of the protomeric units.
The first pocket, defined by Trp 179 and Arg 200 , is the same as that on monomeric ␣-toxin H35A (see Fig. 3), albeit the hydrogen bond between the phosphate group of the PCho moiety and the main-chain amide of Arg 200 is considerably longer and weaker in the latter (Fig. 5E). The second pocket lined by all four residues on strand ␤12 of the rim domain snugly accommodates the PCho moiety (Fig. 5, D and E). It mediates a network of van der Waals contacts involving both the main-chain atoms of Gly 180 and Pro 181 and the aromatic rings of Trp 179 and Tyr 182 , forming hydrogen bonds via its hydroxyl group toward the O3 oxygen of the phosphate group (2.69 Å) and via its O2 oxygen with the main-chain amide of Gly 180 (2.84 Å) while in the cis rotamer.
The third pocket is located at the interface between the rim domain of protomer A and the proximal stem domain regions of protomers E and F (Fig. 5E), in contrast to the other pockets that are constituted solely by residues from the rim domain. The third pocket is formed by residues Asn 178 and Trp 179 from the rim domain of protomer A; by Leu 116 and Tyr 118 from the stem domain of protomer E; and by Tyr 112 , Ser 114 , Ile 142 , Gly 143 , and His 144 from the stem domain of protomer F (Fig. 5E). The indole ring of Trp 179 is situated to produce a cation-interaction with the quaternary ammonium group of the PCho moiety (Fig. 5E). The N-methyl and methylene groups participate in extensive contacts with the main-chain atoms of Gly 143 and Asn 178 and with the side chains of Tyr 112 , Ser 114 , and Ile 142 . The PCho moiety is further stabilized by a hydrogen bond between the O3 oxygen of the phosphate group and the ND1 atom of
These results strengthen the hypothesis that multivalent binding of the PC bilayer by the ␣-toxin heptamer may help overcome the energetic barrier to deformation of the membrane during assembly of the ␤-barrel pore lining, thereby driving the conversion of the prepore to the transmembrane pore

Discovery of a small-molecule inhibitor of MRSA virulence
complex. Indeed, replacement of Trp 179 and Arg 200 with alanines in ␣-toxin is known to lead to an arrested prepore state in which only the top half of the cytolytic ␤-barrel pore has formed (26). Together with analysis of intermediate stages of the ␣-toxin assembly process with engineered disulfide bonds (34), our study also suggests that the interaction between the ␣-toxin prepore and the PC headgroup may induce a large conformational change in the prestem region, which is essential for pore formation.

Structure of the ␣-toxin H35A heptamer in complex with C 14 PC
In the ␣-toxin pore structure, His 35 is located in the crucial interprotomeric contact region (27), and nonconservative replacements at this position (including H35A) have been shown to abolish heptamer formation and thus cytolytic activity and lethal toxicity (62)(63)(64). In light of our findings that the PC bilayer binding might promote both the oligomerization of ␣-toxin monomers and the structural rearrangements that accompany the prepore-to-pore conversion, we hypothesized that a high concentration of C 14 PC could facilitate the assembly of the ␣-toxin H35A pore complex. To directly test this hypothesis, we have determined the structure of the ␣-toxin H35A heptamer crystallized in the presence of 25 mM C 14 PC at 2.5 Å resolution (Table 1 and Fig. 5F); it is worth noting the use of 5 mM C 14 PC for the crystallization of the ␣-toxin H35A monomer (see Fig. 3 and "Experimental procedures"). In this mutant pore complex, PCho moieties bind in the first and second pockets described above on the rim domain of each protomer. In essence, the C 14 PC-bound structures of the ␣-toxin H35A and WT heptamers are nearly identical, with r.m.s.d. values of 0.04 -1.42 Å over 2,051 C ␣ atoms. The positions and conformations of the two PCho moieties are also similar. However, C 14 PC does not bind to the aforementioned interprotomer pocket on the ␣-toxin H35A pore, whereas B-factors for this mutant pore are considerably higher than those for the WT one (24 -201 Å 2 as compared with 13-73 Å 2 ), consistent with the pronounced effect of the H35A mutation on cytotoxicity (59). These results support our hypothesis that the PC-rich membrane acts as a critical effector of oligomerization and pore formation by ␣-toxin.
In summary, despite their different subunit composition and stoichiometry, ␣-toxin and the leukocidins likely follow an evolutionarily conserved PC-dependent pore assembly pathway, involving the initial membrane binding of toxin monomers and membrane-dependent dimerization and oligomerization, followed by the prepore-to-pore transition and membrane perforation. It should be stressed that our crystallographic results demonstrate that the interactions between PCho and the oligomeric pore forms of ␣-toxin and PVL differ considerably. Importantly, atomic-level insight of the toxin oligomer-PC interactions obtained here will facilitate the development of PC analogs that inhibit pore formation and thus block the immune cytolytic effects of this subfamily of proteins.

Inhibition of the cytotoxicity of LukED, PVL, and ␣-toxin by C 14 PC
The presence of the conserved PC binding sites in the leukocidins and ␣-toxin (see above) suggests that PC mimetic com-pounds may interfere with toxin-mediated killing of primary human immune cells. Therefore, flow cytometry experiments were conducted to first evaluate the ability of C 14 PC to diminish the cytolytic activity of LukED in Jurkat cells expressing CCR5. This Jurkat cell line has been shown to be susceptible to the toxin (19). LukED at a concentration of 2.5 g/ml resulted in ϳ80% lysis of Jurkat cells within 1 h at 37°C (Fig. 6A). We found that C 14 PC inhibited the lysis in a concentration-dependent manner, with an IC 50 value between 15 and 25 M (Fig.  6A). In sharp contrast, PCho did not show appreciable inhibitory activity up to 0.5 mM. We conclude that C 14 PC produces effective toxin inhibition by presenting multiple copies of the PC headgroup on its micellar surface, in accordance with previous observations (54).
To investigate the protective effects of C 14 PC on LukEDinduced lysis of primary human leukocytes expressing CCR5 and CXCR1 in vitro, LukED at concentrations of 2.5 and 5 g/ml was first preincubated with 50 M C 14 PC at 4°C and was subsequently added to peripheral blood mononuclear cells (PBMCs) labeled with specific cell surface markers. After 1-1.5 h at 37°C, the cells were stained with fixable viability dye eFluor 506 and analyzed by flow cytometry. Inhibition of LukED by C 14 PC was assessed by determining the relative abundance of viable cells after challenge with the toxin or medium. As expected, CD14 ϩ monocytes were significantly absent by 2.5 and 5 g/ml of LukED (Fig. 6D), whereas pretreatment with 50 M C 14 PC produced a 70 -90% protective effect against monocyte lysis (Fig. 6, B and D). Likewise, 50 M C 14 PC blocked the lysis of CD8 ϩ effector memory T cells by 50 -75% (Fig. 6, C and E) and of CD8 ϩ CCR5 ϩ T cells by 50 -95% (Fig. 6F). Moreover, 50 M C 14 PC also rescued 50 -85% of NK cells (Fig. 6G), which are highly susceptible to LukED due to their surface expression of CXCR1 (19). These results demonstrate that C 14 PC confers target cell protection by blocking the interaction between LukED and membrane PC.
We next sought to assess the ability of C 14 PC to abrogate the cytolytic activities of PVL and ␣-toxin using the in vitro cell viability assay described above. The addition of 2 and 10 ng/ml PVL led to 85-95% lysis of monocytes after 1.5 h of incubation at 37°C (Fig. 7, A and B). Pretreatment with 100 M C 14 PC suppressed the lysis by 90% (Fig. 7, A and B). Similarly, ␣-toxin at concentrations of 30 and 100 ng/ml caused 75-90% lysis of monocytes and ϳ50% lysis of CD3 ϩ T cells after incubation at 37°C for 24 h, whereas 100 M C 14 PC caused 75-90% reduction of the lytic activity (Fig. 7, C and D). We conclude that C 14 PC is a broad-spectrum small-molecule inhibitor of LukED, PVL, and ␣-toxin and that membrane PC contributes to the mechanism of their cytolytic action.

Implication for MRSA drug discovery
The high prevalence of highly pathogenic MRSA is creating a crisis in modern healthcare due to the limited therapeutic options available, the toll of severe disease and mortality it inflicts, and the enormous cost of inpatient care to which it contributes (3,4). The ability of MRSA to form biofilms on necrotic tissues and medical devices is also an important virulence mechanism that complicates infections (5, 6). As antibiotic resistance continues to emerge, disarming the major viru-

Discovery of a small-molecule inhibitor of MRSA virulence
lence mechanisms of MRSA strains has the potential to become an alternative therapeutic approach aimed at limiting host tissue damage while aiding immune clearance. The ␣-hemolysin subfamily of cytotoxins represents a prime target for antivirulence drug development, owing to their critical roles in inacti-vating host immune defenses, destroying tissue barriers, and modulating inflammatory responses (3,16). mAbs targeting ␣-toxin have been shown to prevent human lung cell injury in vitro and protect experimental animals against lethal S. aureus pneumonia (65). Several such mAbs are currently in clinical

Discovery of a small-molecule inhibitor of MRSA virulence
trials, including mAbs MEDI4893 and KBSA301 (Refs. 66 -69). Given the variability of MRSA immune evasion determinants, such single-target drugs are most likely to be inadequate to achieve a therapeutic effect (68). Our structural elucidation of the two conserved, adjacent PCho-binding pockets on the rim domains of ␣-toxin and the leukocidin F subunits will guide the rational development of PC analogs that prevent cytotoxin assembly and pore formation in the susceptible cell membrane, thereby blocking the cytolytic effects of this subfamily of proteins. Using a combined structural biology and pharmacological approach, we have been able to demonstrate that C 14 PC is a novel broad-spectrum inhibitor of PVL, LukED, and ␣-toxin in vitro. In light of the safety of miltefosine (hexadecylophosphocline, C 16 PC), an oral drug used for the treatment of leishmaniasis (70), we expect that C 14 PC will likewise be well-tolerated in humans. Considering its conserved mechanism of action and low production costs, C 14 PC may provide the basis for the development of prophylactic and therapeutic agents that reduce the virulence of MRSA infection.

Chemicals
All chemicals used were of analytical grade. Unless otherwise indicated, chemicals were purchased from Sigma-Aldrich. Detergents were from Anatrace.

Discovery of a small-molecule inhibitor of MRSA virulence
tion; suspended in 50 mM sodium acetate buffer, pH 5.4, 25% sucrose, 5 mM EDTA, and 5 mM DTT; and lysed at 4°C using an Avestin Emulsiflex C3 homogenizer. Inclusion bodies were isolated by centrifugation, washed twice with the same buffer, and subsequently incubated overnight at 4°C in 50 mM sodium acetate buffer, pH 5.4, 5 mM DTT, and 6 M guanidine HCl or 8 M urea. Insoluble material was removed by centrifugation, and the protein solution was then dialyzed for 2 days at 4°C against three changes of buffer A (50 mM sodium acetate, pH 5.4, and 1 mM EDTA). After removal of the insoluble material by centrifugation, the refolded recombinant toxin was loaded onto a CM-Sepharose CL-6B column equilibrated with buffer A and eluted using a linear gradient from 0 to 1 M NaCl. Fractions containing the recombinant toxin were pooled, dialyzed against buffer A, concentrated, and loaded onto a GE Mono S 5/50 GL equilibrated with buffer A, and the toxin was eluted using a linear gradient from 0 to 0.5 M NaCl. The toxin was further purified using size-exclusion chromatography on a GE Superdex 200 10/300 GL equilibrated with 50 mM sodium acetate, pH 5.4, 100 mM NaCl. Fractions containing the toxin were pooled, concentrated to ϳ20 mg/ml, and stored at Ϫ80°C until use. The concentration of the toxin in purified preparations was determined through UV absorbance measurements.

Crystallization
All crystallization experiments were performed at room temperature using the hanging-drop vapor diffusion method by mixing 1 l of protein solution with an equal volume of precipitant solution. Crystals of LukD were grown from protein at 12 mg/ml in 10 mM sodium acetate, pH 5.4, and precipitant solution (20% PEG MME 2000, 10 mM NiCl 2 , and 0.1 M Tris-HCl, pH 8.5). For data collection, the crystals were cryoprotected with 15% glycerol in the mother liquor and then flash-cooled in liquid nitrogen. The C 14 PC-LukD complex was crystallized from protein at 10 mg/ml in 10 mM sodium acetate, pH 5.4, 10 mM C 14 PC, and 30 mM n-octyl-␤-D-glucoside (␤OG) and precipitant solution (28% PEG 400, 0.2 M MgCl 2 , and 0.1 M HEPES, pH 7.5). The crystals were flash-cooled by plunging directly into liquid nitrogen. Crystals of LukF-PV complexed with C 14 PC were grown from protein at 10 mg/ml in 10 mM sodium acetate, pH 5.4, 10 mM C 14 PC, and 30 mM ␤OG and precipitant solution (2.6 M ammonium sulfate, 5% PEG 400, and 0.1 M HEPES, pH 8.5). The crystals were flash-cooled in liquid nitrogen. The C 14 PC-␣-toxin H35A complex was crystallized from protein at 10 mg/ml in 10 mM sodium acetate, pH 5.4, 5 mM C 14 PC, 40 mM ␤OG, and 0.4 mM deoxy-Big CHAP and precipitant solution (1.5 M ammonium sulfate, 0.25 M potassium sodium tartrate, and 0.1 M sodium citrate, pH 6.0). The crystals were transferred into stabilizing solution (2.25 M ammonium sulfate, 5% glycerol, 20 mM C 14 PC, and 0.1 M sodium citrate, pH 6.0) and then allowed to equilibrate against 3 M ammonium sulfate for 1 h at room temperature prior to flash-freezing in liquid nitrogen. The PVL heterooctamer in complex with C 14 PC was crystallized from LukF-PV at 6.7 mg/ml and LukS-PV at 6.3 mg/ml in 10 mM sodium acetate, pH 5.4, 15 mM C 14 PC, and 40 mM ␤OG and precipitant solution (0.16 M magnesium formate). The crystals were transferred into dehydrating solution (2.7 M ammonium sulfate, and 20 mM C 14 PC) and then allowed to equilibrate against 3 M ammonium sulfate for 3 h at room temperature prior to flash-freezing in liquid nitrogen. Crystals of the ␣-toxin heptamer-C 14 PC complex were grown from protein at 8 mg/ml in 10 mM sodium acetate, pH 5.4, 15 mM C 14 PC, and 30 mM ␤OG and precipitant solution (2 M ammonium sulfate, 0.2 M potassium sodium tartrate, and 0.1 M sodium citrate, pH 6.0). The crystals were flash-frozen in liquid nitrogen. The ␣-toxin H35A heptamer in complex with C 14 PC was crystallized from protein at 10 mg/ml in 10 mM sodium acetate, pH 5.4, 25 mM C 14 PC, and 40 mM ␤OG and precipitant solution (1.9 M ammonium sulfate, 0.25 M potassium sodium tartrate, and 0.1 M sodium citrate, pH 5.2). The crystals were flash-cooled in liquid nitrogen.

Data collection and structure determination
Diffraction data were collected at 100 K at beamline X4C at the National Synchrotron Light Source at Brookhaven National Laboratory, at the Cornell High Energy Synchrotron Source (CHESS) beamline F1, and at the Stanford Synchrotron Radiation Lightsource (SSRL) beamline 9-2. The diffraction data were processed with HKL-2000 (71). Initial phases were determined by molecular replacement using Phaser (72) with respective models of HlgB (PDB code 1LKF), LukF-PV (1PVL), ␣-toxin H35A (4YHD), the HlgAB heterooctamer (3B07), and the ␣-toxin heptamer (7AHL). Refinement was carried out in Refmac5 (73), alternating with manual rebuilding and adjustment in COOT (74). Coordinates for C 14 PC were generated using LibCheck (75). TLS refinement was performed in Refmac5 (76). Crystallographic data and refinement statistics are summarized in Table 1.

Structural analyses
Model quality was judged using the programs Rampage, Procheck, and Sfcheck (77)(78)(79). Protein-ligand contacts for the toxin-C 14 PC complex structures were analyzed using the program COOT (80). The r.m.s.d. values were calculated using the program SuperPose (81). Molecular and solvent-accessible surfaces were calculated with the AREAIMOL program (82) from the CCP4 suite (83). PyMOL (DeLano Scientific) was used to render structure figures.

Differential scanning calorimetry
Protein thermal stability was determined by differential scanning calorimetry (DSC) using a Nano-DSC model 602000 calorimeter (TA Instruments). Protein solutions in buffer A (20 mM sodium acetate, pH 5.8, and 50 mM NaCl) in the presence and absence of 4 mM PCho were subjected to a temperature increase of 1°C/min from 0 to 100°C under a pressure of 3 atm, and the evolution of heat was recorded as a differential power between reference (buffer A) and sample (10 M protein in buffer A) cells. The resulting thermograms (after buffer subtraction) were used to derive thermal transition midpoints (T m values). Fitting to the two-state scaled model provided in Nano-Analyze software was used to obtain a T m value. The experiments were repeated two times with consistent results.

Isolation of human peripheral blood mononuclear cells
Blood samples were obtained from healthy, consenting donors as buffy coats (New York Blood Center) and leukopaks

Discovery of a small-molecule inhibitor of MRSA virulence
(AllCells, Alameda, CA). Human PBMCs were isolated from peripheral blood by density gradient centrifugation using Ficoll-Paque Plus (GE Life Sciences).

Cytolysis inhibition assay
Flow cytometry was used to assay permeabilization of the plasma membrane (pore formation) by LukED, PVL, and ␣-toxin in Jurkat cells and primary human immune cells as described previously (84). Briefly, C 14 PC (6 -100 M) was preincubated individually with different concentrations of the LukD and LukF-PV F subunits and ␣-toxin in a V-bottom 96-well plate for 30 min at 4°C. These mixtures were then added to prestained PBMCs and incubated with the cognate LukE and LukS-PV S subunits for 1-1.5 h and with ␣-toxin for 24 h in a humidified 5% CO 2 incubator at 37°C. The cytotoxin-treated cells were stained with a viability dye and analyzed by FACS. IC 50 values were calculated using GraphPad Prism by fitting data to single-slope dose-response curves constrained to 0 and 100% values.

Data availability
All data are contained within the article. The atomic coordinates and structure factors (codes 6U33, 6U2S, 6U3F, 6U3T, 6U3Y, 6U49, and 6U4P) have been deposited in the Protein Data Bank.