High resolution crystallographic studies of α-hemolysin–phospholipid complexes define heptamer–lipid head group interactions: Implication for understanding protein–lipid interactions

Abstract

The α-hemolysin is an archetypal pore-forming protein that is secreted from Staphylococcus aureus as a water-soluble monomer. When the monomer binds to the membrane of a susceptible cell, the membrane-bound molecules assemble into the lytic heptamer. Although a bilayer or a bilayer-like environment are essential to toxin assembly, there is no high resolution information on toxin–phospholipid complexes. We have determined the structures of detergent-solubilized α-hemolysin heptamer bound to glycerophospho-choline or dipropanoyl glycerophosphocholine at 1.75–1.80 Å resolution and 110 K. The phosphocholine head group binds to each subunit in a crevice between the rim and the stem domains. The quaternary ammonium group interacts primarily with aromatic residues, whereas the phosphodiester moiety interacts with a conserved arginine residue. These structures provide a molecular basis for understanding why α-hemolysin preferentially assembles on membranes comprised of phosphocholine lipids.

Lipids play an important role in the structure, assembly, and function of membrane proteins. Self-assembling bacterial toxins provide model systems to study the processes of membrane protein assembly and bilayer insertion. As an archetype of this class of molecules, α-hemolysin (αHL) and its sequence-related relatives from Staphylococcus au-reus are particularly useful paradigms (Tomita and Kamio 1997; Gouaux 1998; Menestrina et al. 2001). Not only have high resolution structures been determined for the water-soluble monomer (Olson et al. 1999; Pédelacq et al. 1999) and the fully assembled heptamer (α₇) (Song et al. 1996), but an array of biochemical, biophysical and genetic studies have illuminated important elements of the assembly pathway (Bhakdi and Tranum-Jensen 1991; Gouaux 1998). An increasing number of related pore-forming toxins from pathogenic organisms such as Vibrio cholerae (Zitzer et al. 1995; Olson and Gouaux 2003), Aeromonas spp. (Rossjohn et al. 1998a,b), Bacillus anthracis (Petosa et al. 1997), and Clostridium perfingens (Hunter et al. 1993) have recently been discovered. Therefore, characterization of the mechanisms by which these toxins lyse cells is of significant biomedical importance. In addition, α₇ is a promising molecule for use in a host of biotechnological applications (Gu et al. 1999, 2001, Bayley and Cremer 2001).

The mechanism of αHL assembly involves binding of the water-soluble monomer (α₁) to the membrane, followed by conformational changes that result in the formation of a nonlytic heptameric prepore (α₇*), and ultimately the fully assembled heptameric pore (α₇; Walker et al. 1995^; Valeva et al. 1997a):

Susceptibility to trypsinolysis is a useful probe by which to monitor the conformational state of αHL. The water-soluble monomer is sensitive to trypsin at Lys-8 and Lys-131. After the monomer binds to an erythrocyte membrane or to phosphatidyl choline bilayers or micelles, the trypsin site at Lys-131, in the glycine-rich prestem, is refractory to cleavage (Walker et al. 1995). Throughout the α₁* and the α7* stages the amino latch (Lys-8) remains sensitive to trypsin and only with completion of assembly is it resistant to trypsin cleavage (Valeva et al. 1997b). To understand how the αHL monomer and heptamer interact with the membrane, and what roles lipids play in the assembly and structure of the toxin, we have undertaken a series of experiments to determine structures of the assembled heptamer in the presence of the head groups from phosphatidyl choline membranes: glycerophosphocholine (GPC) and dipropanoyl phosphatidyl choline (DiC₃PC).

The αHL permeabilizes liposomes composed of phosphatidylcholine or sphingomyelin and cholesterol, and the water-soluble head group phosphorylcholine blocks liposome lysis (Watanabe et al. 1987). Furthermore, liposomes composed of phosphatidylethanolamine, phosphatidylserine, phosphatidylglycerol, or phosphatidylinositol are resistant to toxin action (Watanabe et al. 1987). Model membranes made up of phosphatidylcholine also support the assembly of the related γ-hemolysin and leukocidin toxins (Ferreras et al. 1998). Because of the strict requirement for phosphatidylcholine head groups in the membrane binding and assembly of αHL, and because staphylococcal γ-hemolysin and leukocidin also demand phosphatidylcholine lipids for function, we set out to determine the binding sites for choline head groups by carrying out high resolution X-ray crystallographic studies of the αHL heptamer in complexes with GPC and DiC₃PC. In addition, we determined conditions for flash-cooling the crystals that in turn allowed us to measure data at higher resolution than was possible when carrying out data collection at room temperature. Here we describe and analyze the α₇–GPC and α₇–DiC₃PC structures at 1.75 Å and 1.8 Å resolution, respectively.

Results

Overall features of the structures

The structures of α₇ in complex with GPC (α₇–GPC) and DiC₃PC (α₇–DiC₃PC) were refined against diffraction data (Table 1), which extend to higher resolution in comparison to the data used in the refinement of the initial 1.9 Å resolution heptamer structure (Song et al. 1996). The higher resolution data and the low temperature at which the diffraction data were collected significantly improve the prominence and clarity of the electron density in regions that were poorly ordered in the room temperature structure, such as the stem domain. The protein assembly and the fold of α₇ in the GPC and DiC₃PC complexes are identical to the structure of α₇ determined at ambient temperature (α₇-rt) in the presence of β-octylglucoside. The α-carbon atoms of α₇-rt and those of the α₇–GPC and α₇–DiC₃PC complexes superimpose well; the root-mean-square (rms) deviations between α₇-rt and α₇–GPC and between α₇-rt and α₇–DiC₃PC are 0.39 Å and 0.37 Å, respectively.

Table 1.

Crystallographic statistics

	Data collectiona
Data set	Resolution (Å)	Observed reflections	Unique reflections	Mean redundancy	〈l/σ (I)〉	Completeness (%)	Rsym (%)
GPC	20.0–1.75	1,272,637	265,645	4.8	16.1 (1.8)	96.8 (84.2)	5.6 (37.0)
DiC3PC	20.0–1.80	1,048,080	244,757	4.3	14.6 (1.9)	91.6 (83.0)	6.4 (36.0)

	Refinementb
Data Set	Resolution (Å)	Reflections (N)	R/Rfree (%)	No. atoms	B values	Rms bonds (Å)	Rms angles (°)	Deviations B bonds	Deviations values angles
a Data sets were collected at beamline X4a at the National Synchrotron Light Source. The energy of the incident X-rays was 0.9669 eV, and the data were collected at 110 K using a R-Axis4 area detector. In the 〈l/σ (I)〉 Completeness, and Rsym columns the numbers in parentheses are for the highest resolution bins (1.83–1.75 Å and 1.88–1.80 Å for GPC and DiC3PC respectively). Rsym = ∑|I - 〈I〉|/∑ 〈I〉.
b R = ∑||Fo| - |Fc||/∑|Fo|. For the calculation of Rfree, 5% of the data was withheld from the refinement. In the No. Atoms column, the first, second, and third numbers indicate the number of protein, water, lipid, and chloride atoms, respectively. For the B values column, the set of five numbers, in order, gives the B values of the main chain, side chain, water, lipid, and chloride atoms.
GPC	20.0–1.75	256,100	23.0/26.9	16,415/1180 182/7	22.0/23.3 34.3/39.5 36.1	0.011	1.62	1.048	1.742
DiC3PC	20.0–1.80	223,515	21.0/25.1	16,415/1044 112/7	24.1/26.0 34.3/51.9 40.9	0.011	1.66	1.480	2.373

The α₇ oligomer is composed of the cap domain, the stem domain, and seven rim domains. The cap domain is made up of seven β-sandwiches and the amino latches of each protomer. Rim domains protrude from the underside of the heptamer, participate in only a few protomer–protomer interactions, and are in close proximity to the membrane bi-layer. The stem domain forms the transmembrane channel (Song et al. 1996). A crevice between the top of the stem domain and the rim domain defines the region involved in the interaction with the phospholipid head groups.

On the basis of strong electron density features (>6 σ) in |F_o|- |F_c| maps, there is one phospholipid head group binding site per protomer in both the α₇–GPC and α₇–DiC₃PC structures. The binding sites for GPC and DiC3PC are similar in location and occupancy and are positioned approximately across from residues Leu-116 and Ile-142 on the stem domain, as illustrated in Figure 1 ▶. Residues Tyr-118 and Val-140, which are two residues “down” the stem, bind the hydrophobic membrane-embedded portion of the stem (Valeva et al. 1996). The head group binding sites are located at the base of the rim domains in a crevice defined by Met-197, Lys-198, Thr-199, and Arg-200 on one side and by Trp-179 on the other.

Figure 1.

Ribbon drawings of the α-hemolysin heptamer complexed with lipids viewed perpendicular to the sevenfold axis (A) and parallel to the sevenfold axis (B). The lipid molecules are shown in CPK representation.

The α₇-rt, α₇–GPC, and α₇–DiC3PC structures, although similar in gross features, have significant differences. The largest deviations are localized to residues Trp-179 and Arg-200 in the rim domains, residues implicated in membrane binding (Walker and Bayley 1995). The indole ring of Trp-179 in the α₇-rt structure has relatively weak electron density and high temperature factors, whereas the same residue in the α₇–GPC and α₇–DiC₃PC structures is well ordered and forms one face of the phosphocholine binding site. Likewise, Arg-200 was disordered in the α₇-rt structure but is well defined in the GPC and DiC₃PC complexes.

GPC complex

All GPC molecules were easily modeled into density using a |F_o| - |F_c| omit-map contoured at 2σ. The GPC molecules are identified as L, M, N, O, P, Q, R, and the corresponding protein subunits are identified as A, B, C, D, E, F, and G, respectively.

The seven GPC molecules occupy identical subsites and superimpose with a rms deviation of 0.55 Å. The torsion angles in each of the seven bound lipid molecules are nearly identical as are the temperature factors (39.5 Ų). The temperature factors for the phosphocholine moiety are generally lower than those for the glycerol group, suggesting that the glycerol group exhibits greater spacial or temporal disorder.

The complimentarity between the binding sites and the phospholipid is substantial and illustrative of specific binding. The lipid head groups are mostly buried. Of the total accessible surface of the unbound GPC (239 Ų), 99 Ų or 41% remains accessible to solvent in the complex with α₇. Shown in Figure 2 ▶ are the main interactions in the lipid binding pocket and in Table 2A are listed the interactions of between the lipid and protein or solvent molecules. Formation of the binding site is modulated by the flipping of Trp-179 and the ordering of Arg-200. These conformational changes allow the protein to sequester the lipid head group in a basic and aromatic residue-rich pocket. The surrounding residues make numerous van der Waals (>4.0 Å) and polar interactions with the phospholipid (see Table 3 below). Surprisingly, however, none of the ester oxygen participate in hydrogen bonding interactions. In fact, there is only one direct hydrogen bond between the lipid and the protein. The O4 atom of GPC is hydrogen bonded to the main chain NH group of Arg 200 (Table 3). Water also plays an important role in the binding of GPC and one water molecule mediates interactions between the O4 atom of GPC, the main chain NH group of Arg-200 and Asn-201 and the OD1 atom of Asn-201. The same water molecule is also near the main chain NH group of Gly-202 (Fig. 2 ▶).

Figure 2.

Stereo ball-and-stick representation of the lipid-binding pocket. (A) GPC and (B) DiC₃PC. Hydrogen bonds are indicated as dashed lines.

Table 2A.

Interactions of GPC with the protein and water molecules

Lipid atom	Refined B-factora	Main and side chains located within 4.0 Å of lipid	Main and side chains located between 4.0 and 4.5 Å of lipid
C8	32.7	176, 179, 182, 194	176, 179, 194, 197
C7	32.6	176, 177, 179	176, 177, 179, 197, 198
C6	32.5	194, 197, 198	176, 194, 197, 198, 200
N1	33.1		176, 179, 182, 194
C5	34.3	179, 194, 200	179, 182, 200
C4	36.7	200	179, 200
O6	39.0		200
P1	39.5		199, 200
O5	41.3		198
O4	39.6	199, 200	199, 200
O3	39.7		200
C3	44.1	198	198
C2	47.3	177, 198	177, 198
C1	46.1	177	177, 198
O1	46.9	177
O2	48.8	177	177

^a B factors are averaged between the seven GPC molecules present in the complex.

Table 3.

Hydrogen bond interactions

Chain	Atoms involved in the hydrogen bonds	Distance (Å)
α7-GPC
L	O4–H2O 797	2.82
	O4–A 200 N	2.76
	H2O 797–A 200 N	3.20
	H2O 797–A 201 N	3.04
	H2O 797–A 201 OD1	3.71
	H2O 797–A 202 N	3.95
α7-Dic3PC
L	O4–H2O797	2.90
	O4–A200 N	2.76
	H2O 797–A 200 N	3.36
	H2O 797–A 201 N	3.14
	H2O 797–A 201 OD1	3.68

In the α₇–GPC complex there are 1044 ordered water molecules with B-values ranging from 26 Ų to 50 Ų. Most of solvent sites are identical to the sites in the α₇-rt structure. Larger differences can be found in the solvent structure in and around the phospholipid-binding site, with the most prominent difference being the key water molecule located in the lipid-binding cavity, participating in the interactions described in the previous paragraph; there is no such ordered water in the α₇-rt structure. In addition, water molecules, usually in groups of two or three, interact with the glycerol moiety and the protein in each of the seven binding sites.

In addition to the density for the lipid head group, we also observed a positive electron feature in |F_o| - |F_c| maps (>4σ) near the lipid on the opposite side of Trp-179. We have fit and refined a single chloride ion to these seven electron density features.

DiC₃PC complex

All of the atoms in the DiC₃PC molecules were modeled in density obtained from a |F_o| - |F_c| omit map. However, the density for glycerol and acyl chains was substantially weaker than that for the phosphocholine moiety. The seven DiC₃PC molecules occupy identical subsites and superimpose well with a rms deviation of 0.78 Å. As in the GPC complex, the DiC₃PC molecules are identified as L, M, N, O, P, Q, R and the corresponding protein subunits are identified as A, B, C, D, E, F, and G, respectively.

The torsion angles in each of the seven bound lipid molecules are nearly identical. The B factors for the DiC₃PC atoms are considerably higher (52.0 Ų) compared to the mean temperature factors for all protein atoms (25.1 Ų). The considerably higher B factor of the ligand compared to those of the protein could be caused by disorder of the phospholipid or by a subunitary occupancy. Similar to the α₇–GPC complex, the DiC₃PC ligands are partially buried. Of the total accessible surfaces of the unbound DiC₃PC (378 Ų), 210 Ų (55.6%) are not shielded with binding to α₇.

As with GPC, the access of DiC₃PC to the binding cleft of the protein is mediated by Trp-179 and Arg-200; these residues sequester the ligand in the head group pocket and involve the phosphocholine moiety in numerous van der Waals interactions (<4.0 Å) and solvent-mediated hydrogen bonds (Tables 2B, 3). Shown in Figure 2 ▶ are the primary interactions in the binding pocket and in Table 3 are the interactions among lipid, protein, and solvent molecules.

Table 2B.

Interactions of DiC₃PC with the protein and water molecules

Lipid atom	Refined B-factora	Main and side chain atoms located within 4.0 Å of lipid	Main and side chain atoms located between 4.0 and 4.5 Å of lipid
C8	32.7	176, 179, 182, 194	176, 179, 194, 197
C7	32.6	176, 177, 179	176, 177, 179, 197, 198
C6	32.5	194, 197, 198	176, 194, 197, 198, 200
N1	33.1		176, 179, 182, 194
C5	34.3	179, 194, 200	179, 182, 200
C4	36.7	200	179, 200
O6	39.0		200
P1	39.5		199, 200
O5	41.3		198
O4	39.6	199, 200	199, 200
O3	39.7		200
C3	44.1	198	198
C2	47.3	177, 198	177, 198

^a B factors are averaged between the seven GPC molecules present in the complex.

Clear nonprotein |F_o| - |F_c| electron density was found, near the lipid, on the opposite side of Trp-179. This electron density was observed in each protomer. Modeling the density as a chloride ion yielded the lowest R_work and R_free values.

Comparison of the GPC, DiC₃PC, and α₇-rt structures

Analysis of the two complexes allows us to assess (1) how the protein accommodates phospholipid head groups, (2) how the proteins accommodates DiC₃PC, a ligand that contains two additional fatty acid chains, and (3) how the organization of internal solvent is influenced by these differences in chain length and flexibility. Some differences are already evident from a qualitative analysis of the structures. In particular, the temperature factors for the phosphocholine head groups are lower than those for the glycerol moieties, suggesting that the head groups exhibit less spatial or temporal disorder in comparison to the glycerol portions of the lipids. Superimpositions of the main chain atoms of the two structures revealed an overall rms difference of 0.16 Å and we determine that not only are the protein conformations similar, but the positions and conformations of the lipid head groups are also similar.

When the relationships between the side chain atoms of all residues that interact with the bound lipids were compared, no apparent differences were noted in their positions. This includes Trp-179 and Arg-200 as well as all the other residues in the binding pocket. Both the α₇–GPC and α₇– DiC₃PC complexes have one ordered water molecule in their interior compared to α₇-rt. This ordered water molecule is located at hydrogen bond distance from both the phosphate and Arg-200. The corresponding region is disordered in the 1.9 Å α₇-rt structure. Thus, the presence of the short acyl chains does not substantially affect the head group position or the protein conformation.

Comparison of the two lipid complexes and the α₇-rt structure indicates some differences between the three structures (Fig. 3 ▶). Two substantial differences in side chain positions were noted, in particular Trp-179 and Arg-200. In the absence of contacts with the lipid head groups, their side chains are disordered. Binding of the lipids restrains these rotations and stabilizes their side chains in a single, well-ordered conformation, as indicated by their well-defined electron densities. Indeed, the mean B factors for the side chain atoms of Trp-179 and Arg-200 are 27.9 Ų and 29.9 Ų, respectively, and the side chains occupy similar positions in the α₇–GPC and α₇–DiC₃PC structures.

Figure 3.

Comparison between the native α-hemolysin structure (pink) and the complexes with GPC (green) and DiC₃PC (blue), obtained from the superimposition of the Cα atoms of the entire protein. In the stereo view are shown the interactions of the lipids with Arg-200 and Trp-179.

Discussion

The structures of α₇ in complex with GPC or DiC₃PC were solved to a higher resolution than the previously reported α₇-rt structure (Song et al. 1996). As expected, the increase in resolution and the use of low temperatures resulted in better resolved electron density for key regions of the heptamer, such as the transmembrane stem. Because we carried out the data collection at low temperature, and because GPC and DiC₃PC do not disorder the crystals, we were able to measure data to a much higher resolution in comparison to that previously collected from the α₇ in complex with diheptanoyl phosphatidylcholine, which was limited to Bragg spacings of 3 Å (Song et al. 1996). Most important, we carried out thorough refinement using the α₇–GPC and α₇– DiC₃PC data, whereas the α₇ complex with diheptanoyl phosphatidylcholine has not been subject to thorough refinement.

Interactions between the heptamer and the lipid head groups involve both the stem and the rim domains (Fig. 1 ▶). The rim–stem crevice is defined by the inner surface of the rim domain and the upper portion of the stem domain. The top of the crevice is marked by Tyr-182 and is about 8 Å above the putative membrane surface. Two important residues, Arg-200 and Trp-179, protrude their side chains into the crevice at positions estimated as ~5 Å above the outer leaflet of the lipid bilayer. The crevice constitutes an attractive pocket to accommodate the lipid head group, because it harbors basic and aromatic residues, key elements for interaction with the negatively charged and quaternary ammonium moieties on the head group. The phosphocholine molecules (Fig. 2 ▶) bind primarily through cation–π interactions between the quaternary ammonium group and the indole ring of Trp-179, and water-mediated hydrogen bonding between Arg-200 and the phosphate group. To a large degree, the cation–π interactions may provide a structural explanation for the reported choline-binding specificity (Ferreras et al. 1998).

Noteworthy is the comparison of the phosphocholine-binding pocket in α₇ and in the water-soluble LukF monomer (Olson et al. 1999). Both present a rim domain containing exposed aromatic residues with a tryptophan and an arginine residue (Trp-177, Arg-198 in LukF and Trp-179, Arg-200 in αHL) in the binding pocket. As in α₇, in LukF the binding of the phosphocholine head group orders the side chain of the tryptophan residue, which is disordered in the absence of lipid and causes a displacement of the arginine residue, due to the direct interaction with the lipid head group (Olson et al. 1999).

It is also relevant to compare αHL and LukF lipid head group binding sites and the lipid head group binding sites of other proteins, such as the antiphosphocholine antibody McPC603 (Satow et al. 1986) and the lipid-binding protein PDC-109 (Wah et al. 2002). In all four binding sites, a common feature is the presence of a tryptophan residue forming interactions with the quaternary ammonium group. In PDC-109, there is also hydrogen bonding between tyrosine hydroxyl groups and the phosphate group (Wah et al. 2002). The proximity of tryptophan residues to bound lipids is a recurring theme in membrane proteins and is evidence for the central role played by this residue. Because the indole side chain has both hydrophobic and hydrophilic characters, it partitions at the hydrophobic–hydrophilic interface in lipid bilayers (Jacobs and White 1989). As previously seen in membrane protein structures, tryptophans are frequently located near the bilayer surface and are oriented such that the indole N-H is directed toward the hydrophilic environment (Chattopadhyay and McNamee 1991); thus, tryptophans may play an important role in stabilizing membrane proteins through electrostatic interactions at the lipid bilayer surface.

Membrane proteins, although constituting about one-third of all proteins encoded by the genomes of living organisms, are still strongly under-represented in the database of three-dimensional protein structures. Furthermore, only a few membrane protein structures provide detailed information on how proteins interact with lipids (Fyfe et al. 2001). Some representative examples include bacteriorhodopsin (Luecke et al. 1999), the bacterial reaction center (Roth et al. 1991; McAuley et al. 1999), cytocrome c oxidase (Tsukihara and Yoshikawa 1998), ferric hydroxamate uptake receptor (Ferguson et al. 2000a,b), and the bacterial potassium channel KcsA (Valiyaveetil et al. 2002). Studies on bacteriorhodopsin (Luecke et al. 1999) documented the presence of 18 full or partial lipid acyl chains per protein subunit. The acyl chains occupy a large part of the contact surface between protomers in the trimer and they also mediate contacts between adjacent trimers. From a crystallographic study on the reaction center from Rhodobacter sphaeroides (Roth et al. 1991; McAuley et al. 1999), cardiolipin is located on the intramembrane surface and is engaged in hydrogen bond interactions with polar residues in the membrane interface region and hydrophobic interactions with the intramembrane surface of the protein. Key interactions involve direct contacts between the phosphate oxygen of the lipid head groups and basic amino acids or backbone amide groups, as well as contacts with lysines, tryptophan, arginine, and tyrosine residues, mediated by water molecules. Particularly interesting is the resolution of cardiolipin in the structure of the bovine cytochrome c oxidase (Tsukihara and Yoshikawa 1998). In fact, although in the bacterial reaction center there is no information on a specific role played by this lipid, here its removal causes a loss of enzyme activity. Cardiolipin is located at the interface of the subunits in the dimer and presumbably lipid removal disrupts important intersubunit interactions. The ferric hydroxamate uptake receptor (FhuA) from Escherichia coli (Ferguson et al. 2000a,b) belongs to the class of outer membrane proteins of Gram-negative bacteria that present a membrane-spanning region made of a β-barrel. FhuA interacts with lipopolysaccharide (LPS), a complex molecule that forms the outer leaflet of the bacterial outer membrane. The acyl chains of LPS are ordered on the protein surface and are approximately parallel to the axis of the β-barrel and make numerous van der Waals interactions with surface-exposed hydrophobic residues. The larger polar head groups of LPS make extensive interactions with charged and polar residues near the outer surface of the membrane. The structure of the bacterial potassium channel (KcsA) was solved in complex with a lipid molecule that Valiyaveetil et al. presume to be a phosphatidylglycerol (2002). Due to the partial disorder, the lipid was not well resolved and only the 1,2-sn-diacylglycerol portion of the lipid molecule was built into the model. The lipid is bound in a groove between adjacent subunits, with its head group near the extracellular surface and interacting with two arginine residues and the tail projecting into the outer membrane leaflet.

These crystal structures, together with our results (Fig. 1 ▶), provide detailed descriptions of interactions between integral membrane proteins and lipids. In general, lipid–protein interactions involve a combination of ionic interactions between the protein and the lipid head groups and van der Waals interactions between the lipid tails and the electro-neutral intramembrane surface of the protein. In the head group region, ionic interactions involve polar groups of a number of residues, the protein backbone and bound water molecules, whereas the lipid tails sit in largely hydrophobic grooves in the irregular surface of the protein. Thus, a common trend seems to be that the acyl chains occupy grooves in the protein surface, where they are engaged in hydrophobic interactions with apolar residues and backbone atoms. Moreover, our results together with data on other integral membrane proteins highlight the important role of lipids in mediating interactions between protomers in multimeric proteins.

The details of protein–lipid interactions are important to understand membrane protein folding, membrane adsorption, insertion and function in lipid bilayers (Fyfe et al. 2001). Both specific and nonspecific interactions with lipids may participate in protein folding and assembly (Roth et al. 1991; Tsukihara and Yoshikawa 1998; Luecke et al. 1999; McAuley et al. 1999; Ferguson et al. 2000a,b; Valiyaveetil et al. 2002). Moreover, specific protein–lipid interactions may contribute to cell binding and membrane insertion. Because bilayer destabilization is a necessary step in membrane insertion, it is interesting to hypothesize how proteins, which insert into lipid bilayers, perturb membranes. During the insertion and folding processes, membrane proteins must make a transition into the membrane, during which they fold into their native structure. One hypothesis is that the native structure is attained through sequencial stages of interfacial binding, secondary structure formation and insertion of secondary structure units into the membrane bilayer. These processes, presumably, require deformation of the equilibrium membrane structure. Moreover, a specific interaction with phospholipids may be closely related to the initial folding and assembly of membrane proteins. Unfortunately, the mechanism by which transmembrane segments are inserted into the lipid bilayers is still poorly understood and requires further studies. Many questions remain as to how membrane proteins integrate into the bilayer, how much they disturb local bilayer organization, and how many lipid molecules must be laterally displaced after the insertion and folding of transmembrane pores.

On the basis of our data, we suggest that αHL penetrates the target membrane with seven lipids specifically bound to the protein, thus perturbing the bilayer structure and producing localized defects in the bilayer (Fig. 1 ▶). Using the α₇ structure, we can calculate the cross-sectional area of the stem (~700 Ų). If we assume that the heptamer “sits” on the membrane with the sevenfold axis perpendicular to the membrane plane, and that one phosphocholine lipid molecule has a cross-section area of ~50 Ų(Gennis 1989), then insertion of the stem would displace ~14 lipids, or ~2 per subunit.

Materials and methods

Glycero-3-phosphocholine and DiC₃PC were obtained from Avanti Polar Lipids (Alabaster, AL); β-octyl glucoside (OG) was from Anatrace (Maumee, OH); PEG 5000 (MicroSelect grade) was from Fluka and all other reagents and chemicals were of analytical grade. The αHL monomer was expressed and purified, and α₇ was formed and purified as previously described (Bhakdi et al. 1981).

Protein crystallization

Crystals of α₇ were obtained by both the hanging drop and the sitting drop method, using a protein solution (ca. 6.7 mg/mL) in 10 mM Tris-HCl (at pH 8.0) and 25 mM OG. Drops containing 1.2 M ammonium sulfate, 0.05 M sodium cacodylate at pH 6.5, 0.125% (w/v) PEG 5000, 12.5 mM OG, and 3.35 mg/mL αHL were equilibrated against a reservoir solution containing 2.4 M ammonium sulfate, 0.1 M sodium cacodylate (at pH 6.5), and 0.25% (w/v) PEG 5000. Large single crystals of αHL were obtained after 2 weeks at 21°C.

Crystals containing GPC and DiC₃PC were prepared by soaking OG crystals for 1 week in three changes of solutions containing the same constituents as the OG crystals except that for the GPC soaked crystals, OG was replaced by 100 mM GPC. For the DiC₃PC soaked crystals, the soaking solutions contained both 25 mM DiC3PC and 25 mM OG. Soaking conditions were screened using small crystals, whereas the optimal conditions were refined using medium size crystals (0.2 × 0.2 × 0.2 mm³). The crystals were flash-cooled using the soaking solution supplemented with 15% (v/v) glycerol. After washing the crystal in the cryoprotectant solution for <30 sec, the crystal was quickly picked up with a loop and plunged into liquid nitrogen. Immediately before data collection, the loop was quickly transferred to a goniometer centered in a stream of gaseous N2 at ca. −160°C. One crystal for each complex was sufficient to collect complete data sets.

Data collection and processing

Diffraction data sets were collected from single crystals on beam-line X4A of the National Synchrotron Light Source using a Rigaku R-axis IV image plate detector. Diffracted intensities were processed using DENZO and SCALEPACK and programs from the CCP4 program suite. Statistics associated with the X-ray diffraction data are presented in Table 1. Both the α₇–GPC and the α₇–DiC₃PC crystals belong to the monoclinic space group C2 and have the following unit cell dimensions: α₇–GPC: a = 150.58 Å, b =134.94 Å, c =132.79 Å, and β =91.55°; α₇–DiC₃PC: a =150.93 Å, b =134.91 Å, c =133.05 Å, and β =91.22°.

Crystallographic refinement

The structures of the complexes were analyzed by difference Fourier techniques. The atomic model from the room temperature α₇ structure, minus the water molecules, was used as a starting model for the refinement of the α₇–GPC and the α₇–DiC₃PC structures. Initial rigid body refinement was followed by cycles of (1) Powell minimization, (2) overall and then individual B factor refinement, and (3) manual refitting. Rigid body refinement and Powell minimization were carried out using the X-PLOR program (Brünger et al. 1998). Electron density maps were calculated using programs from the CCP4 suite (Collaborative Computational Project 1994). Map analysis and modeling were carried out using the molecular graphics program O (Jones et al. 1991).

Due to slight changes in cell dimensions, the initial difference Fourier map was relatively noisy. The initial crystallographic R values, calculated using data between 20.0 Å and 1.8 Å and before refinement, were 0.415 for the α₇–GPC and 0.329 for the α₇– DiC₃PC data sets. Fourier maps, calculated with |F_o| - | F_c| and 2|F_o| - | F_c| coefficients revealed prominent electron density features in the region involved in the phospholipid interaction. After rigid body refinement and Powell minimization both the conventional R and R_free values dropped significantly (R/R_free were 25.9/29.6 for α₇–DiC₃PC and 27.9/29.6 for α₇–GPC). Water molecules were identified by peaks of difference electron density that were greater than 3.0σ in F_o - F_c maps and that were situated such that a corresponding water molecule could participate in a hydrogen bond network to the protein. A bulk solvent correction was applied and non-crystallographic symmetry restraints were placed on the main chain (25 Kcal mole⁻¹ Å⁻²) and side chain (4 Kcal mole⁻¹ Å⁻²) atoms of residues 120–137 because the density associated with this region of the structure was relatively weak. The molecular structures of DiC₃PC and GPC were fit straightforwardly to the density at the end of the refinement and by this point, the electron density for the lipid head groups was well defined. The crystallographic statistics are shown in Table 1. The quality of the final models was assessed using Procheck (Laskowski et al. 1993). In the Ramachandran plot 91% of the residues occupied the most favored regions and no residues were located in disallowed regions.

The relative accessible surface was calculated by dividing the accessible surface of the bound lipid in the structure (without surface water molecules) by the accessible surface of the unbound lipid in the same conformation (with a probe of 1.4).

Abbreviations

• αHL, α-hemolysin

• α₇, αHL heptamer

• GPC, glycero-phosphocholine

• DiC₃PC, dipropanoyl glycerophosphocholine

• rms, root- mean-square