Phosphatidylcholine Cation—Tyrosine π Complexes: Motifs for Membrane Binding by a Bacterial Phospholipase C

Abstract

Phosphatidylinositol-specific phospholipase C (PI-PLC) enzymes are a virulence factor in many Gram-positive organisms. The specific activity of the Bacillus thuringiensis PI-PLC is significantly increased by adding phosphatidylcholine (PC) to vesicles composed of the substrate phosphatidylinositol, in part because the inclusion of PC reduces the apparent K_d for the vesicle binding by as much as 1000-fold when comparing PC-rich vesicles to PI vesicles. This review summarizes (i) the experimental work that localized a site on BtPI-PLC where PC is bound as a PC choline cation—Tyr-π complex and (ii) the computational work (including all-atom molecular dynamics simulations) that refined the original complex and found a second persistent PC cation—Tyr-π complex. Both complexes are critical for vesicle binding. These results have led to a model for PC functioning as an allosteric effector of the enzyme by altering the protein dynamics and stabilizing an ‘open’ active site conformation.

1. Introduction

Bacterial phosphatidylinositol-specific phospholipase C (PI-PLC) enzymes are virulence factors secreted by Gram-positive organisms. For Bacillus sp., their role is to downregulate the host immune response by releasing glycosylphosphatidylinositol (GPI) anchored proteins from the cell surface, generating diacylglycerol (DAG) along with the soluble glycosylated protein [1,2]. In terms of the structure and mechanisms, most of the studies of PI-PLC enzymes from Bacillus sp. have used phosphatidylinositol (PI) as the substrate, rather than a GPI-linked protein. A notable exception is the work by Lehto and Sharom on the kinetics of cleavage of a purified GPI-anchored protein that exhibits much of the kinetic behavior observed with PI as the substrate, confirming the use of PI as a substitute for GPI-anchored proteins [3,4,5]. The PI-PLC catalyzed reaction for the hydrolysis of PI to inositol-1-phosphate, shown in Figure 1A, occurs via a general acid–general base mechanism where the PI is first cleaved to membrane-soluble diacylglycerol (DAG) and myo-inositol 1,2-(cyclic)phosphate (cIP), the latter being too polar to partition into membranes. The cIP can also be hydrolyzed by the enzyme to myo-inositol-1-phosphate. However, cIP is a poor substrate with a low k_cat (12 s⁻¹) and a very high K_m (90 mM) [6]. The relatively small size of PI-PLC (34.8 kDa) and the ability to separate its activity toward interfacial (PI) and soluble (cIP) substrates make it a good system to study the detailed mechanism of the protein binding to different membranes and to explore how specific membrane components alter enzymatic activity.

Early work showed that PI presented in small unilamellar vesicles (SUVs) or solubilized in Triton X-100 micelles was a poor substrate for the B. thuringiensis PI-PLC (BtPI-PLC). With PI SUVs, k_cat and K_m were 73 s⁻¹ and 2.6 mM, respectively [7]. However, the BtPI-PLC phosphotransferase activity was significantly increased by including phosphatidylcholine (PC) in the vesicle or micelle along with PI [6,7]. Sphingomyelin, but not other phospholipids, also activated the enzyme, indicating that the phosphocholine group was critical for the activation of the enzyme. However, soluble PC molecules such as dibutyroyl-PC (diC₄PC) were not activators, indicating that PC activation requires an interface. With PC/PI SUVs, the magnitude of the increased enzymatic activity depended on the mole fraction of the PC (X_PC) as well as the total amount of phospholipids present. The data shown in Figure 1B used 10 mM PI SUVs, and additional POPC was added to increase the X_PC. For that concentration of pure PI, the enzyme would be ~75% saturated. The inclusion of PC in the vesicles increased the specific activity more than two-fold at its maximum. The apparent K_d, measured by fluorescence correlation spectroscopy [8], was comparable to the K_m, indicating that for PI SUVs, the rate-limiting step was binding to the SUVs. As shown in Figure 1C, the apparent K_d for BtPI-PLC binding to vesicles composed of dioleoylphosphatidylglycerol (DOPG) as the PI surrogate and 1-palmitoyl-2-oleoyl-PC (POPC) as the activator decreased ~1000-fold, with the tightest binding occurring in the PC-rich vesicles. The residence time of BtPI-PLC on a PC-rich tethered SUV vesicle, measured with single-molecule fluorescence microscopy, is 380 ± 50 ms [9], allowing many rounds of PI cleavage on the vesicle surface.

The second step of PI cleavage, cIP hydrolysis, was also increased with the addition of PC SUVs [7] or short-chain PC micelles [6]. Again, the addition of other phospholipids (PS, PA, and PG) in the micelles or vesicles did not activate the enzyme. The cIP is either in solution or bound to the protein and not partitioned into the membrane. Therefore, interfacial PC is an allosteric effector of BtPI-PLC. However, these kinetic and binding results do not differentiate between a distinct PC binding site on BtPI-PLC or the nonspecific membrane perturbation effects that alter the conformation or dynamics of PI-PLC. While these results provided data on the importance of PC or sphingomyelin for BtPI-PLC vesicle binding and activity, they did not elucidate the molecular mechanism for the enhanced binding and activity.

2. Experimental Results—Characterization of Specific PC Binding Site(s) on BtPI-PLC

2.1. B. thuringiensis PI-PLC, a Member of the TIM Barrel Superfamily

The crystal structures of Bacillus sp. PI-PLCs, the recombinant proteins, and the various mutants (see Table 1 in reference [10]) all show a distorted β-barrel structure. While a crystal structure for wild-type BtPI-PLC is not available, the B. cereus enzyme only differs by a few amino acids, and its structure [11,12] can be used to model BtPI-PLC. The structure of the enzyme, shown in Figure 2 with the key regions identified by mutating and assessing the loss of activity or membrane binding, is from molecular dynamics simulations of the enzyme docked on a dimyristoyl-PC (DMPC) bilayer [13]. The active site residues are in red. Trp47, in helix B, and Trp242, in the β7–αG rim loop, insert into the membranes and are key components of the interfacial binding site (IBS) of the protein.

2.2. Experimental Evidence for a Specific PC Binding Site on B. thuringiensis PI-PLC

While BtPI-PLC’s requirement for PC (or sphingomyelin) in an interface for optimal cleavage of PI (or optimal hydrolysis of cIP) is suggestive of a specific binding site for that zwitterionic phospholipid, identifying such a site is difficult. For many other amphitropic proteins that bind to membranes via specific phospholipid headgroups (e.g., PH domains that bind different phosphoinositides), a soluble polar group alone can often bind to the protein well enough to pinpoint the lipid binding site. However, soluble phosphocholine, glycerophosphocholine, and diC₄PC have very a low affinity for BtPI-PLC [6] (also see the Appendix for NMR data that show the very poor binding of diC₄PC to spin-labeled D205C). As an interface is needed for the PC activation of BtPI-PLC, other approaches were necessary.

The crosslinking agent 1-ethyl-3-[3-dimethylaminopropyl]carbodiimide hydrochloride (EDC) forms an amide bond between the nearby side chains of lysine and acidic residues (Asp, Glu). The somewhat naïve thought behind the experiment was that crosslinking the protein in the presence of an interface might trap it in an active form. BtPI-PLC was mixed with diheptanoyl-PC (diC₇PC) micelles rather than a bilayer PC interface since the crosslinked micelle/protein complex could be extensively dialyzed at pH 7 to remove all but the most tightly bound diC₇PC molecules. Crosslinked and dialyzed BtPI-PLC exhibited a mass increase of ~1 kDa (exptl. error ± 0.15 kDa), equivalent to two tightly bound diC₇PC molecules [14]. If the dialysis was performed at pH > 8, the mass increase was not detected, indicating that crosslinking stabilized a conformation of the protein where the affinity for the two diC₇PC molecules was very high around neutral pH, but not in basic conditions. In the absence of EDC, there was no excess mass observed after mixing the BtPI-PLC with (diC₇PC) and then dialyzing it. The crosslinked protein with the two lipid molecules bound was more than twice as active towards PI/diC₇PC as the uncrosslinked protein. Two surface tryptophan residues, Trp47 in helix B and Trp242 in the ←β7–αG loop, are important for vesicle binding [15,16]. If either one is removed, the crosslinked and then dialyzed protein has a mass for only a single tightly bound diC₇PC molecule.

With unaltered recombinant BtPI-PLC, one sees a substantial broadening of the diC₇PC ³¹P resonance when the enzyme is added [17]. This is due to the exchange of PC molecules between the monomers in solution, the micelles, and the potential BtPI-PLC binding sites, as well as the increases in diC₇PC micelle size upon BtPI-PC binding. Mixing the dialyzed crosslinked BtPI-PLC with diC₇PC micelles did not lead to line broadening, a result implying that the tightly bound diC₇PC molecules are not in fast exchange with the added diC₇PC micelles or monomers. These crosslinking experiments indicated that two PC molecules can be tightly associated with the BtPI-PLC but provided no information on where they were bound. There are nine Lys–Asp/Glu pairs within crosslinking distance, making identification of where the PC molecules were bound problematic.

The interaction of the protein with the phospholipid vesicles required a different approach to identify a PC binding site. While many experimental methods have been used to characterize the binding of proteins to membranes, most do not directly identify binding sites (see [10] for an extensive review of these methods). High resolution ³¹P relaxometry (also referred to as shuttle field-cycling ³¹P NMR relaxometry) is a very useful but not well known technique that can be used to identify and characterize specific protein interactions with phospholipids in small unilamellar vesicles or micelles. This type of relaxometry measures the spin-lattice/longitudinal relaxation rate (R₁ = 1/T₁) over a wide range of magnetic fields by rapidly shuttling the sample, excited at a high field, to different positions in the bore of the superconducting magnet for relaxation at a lower field (B_relax). The range of B_relax used for studying BtPI-PLC binding to small vesicles was 11.7 T down to 0.003 T [18,19]. After a time at B_relax that is varied, the sample is returned to the probe for a signal readout. Phospholipids in bilayers have many different motions covering a wide range of timescales [20] that alter the orientation or interactions of the phospholipid ³¹P–¹H dipoles. Different motions will give rise to nuclear magnetic relaxation dispersions (NMRDs) with correlation times related to the specific motion. The field dependence of ³¹P R₁ is the sum of all the individual NMRDs.

Figure 3 shows the ³¹P field cycling profile for POPC in POPC/d₃-DOPMe (1:1) SUVs as a function of B_relax. The dependence of R₁ on the Larmor angular frequency ω_P (where ω_P = γ_P B_relax) is also shown. The variation of the ³¹P R₁ with B_relax is characterized by three dipolar NMRDs, labeled R_D0, R_D1, and R_D2. R_D0, occurring at the lowest fields, has a correlation time, τ_D0, that reflects the overall tumbling of the aggregate (the bigger the particle, the longer the τ_D0) conflated with the translational movement of the phospholipids. For SUVs, τ_D0 is typically 0.5–1.5 µs. For R_D1, τ_D1 is in the 10–15 ns range; this NMRD arises from the axial/wobble motions of the phospholipid molecules. R_D2 is the result of fast dipolar motions that are very localized, e.g., changes in dihedral angles. These occur on sub-ns timescales. R_D2 is partially obscured by a fourth NMRD, R_CSA, which is from the fast motions associated with the chemical shift anisotropy of the ³¹P. The measurements of R₁ at the high fields of modern spectrometers primarily reflect the fast motions. The partitioning of a protein onto the bilayer can alter some or all of these ³¹P motions. For example, the differential changes in R_D0 and τ_D0 for POPC/DOPG SUVs when BtPI-PLC is added provide information on the translational diffusion of each phospholipid in the plane of the membrane [21].

The average distance of a phospholipid ³¹P from the specific regions of the transiently bound protein is provided by introducing a cysteine at a specific site on the protein and spin-labeling it. This provides an unpaired electron that is a much more potent relaxer than the -OCH_n- protons linked to ³¹P that normally dominate the dipolar relaxation of ¹³P. In these experiments, the ratio of each phospholipid in the outer monolayer to BtPI-PLC was typically between 230 and 260 (the range is because the SUVs are not a single size and but cover a range presenting different ratios of phospholipids in the outer surface [8]). At this ratio, there are few proteins on a given small vesicle; so, protein/protein interactions (a potential complication if both are spin-labeled) are unlikely to occur. Enhanced relaxation of the phospholipid ³¹P caused by a nearby spin label on the protein implies that a phospholipid must occupy that site for the correlation time of that NMRD. Therefore, we use changes in R_D0, which has the longest correlation time, to define a specific phospholipid/protein complex.

BtPI-PLC is an ideal candidate for this technique since it lacks cysteine residues. A cysteine can be introduced at many different sites on the protein, and as long as the enzymatic activity is not altered by the presence of the spin-labeled Cys, the increased ³¹P R₁ can provide an averaged distance of the ³¹P to the electron (r_P-e). This is obtained by subtracting the R₁ profile for the vesicles with the same amount of unlabeled protein from the profile of ³¹P R₁ for the vesicles with spin-labeled protein as a function of B_relax. The resultant ΔR_D0 NMRD provides the correlation time for the ³¹P–electron interaction, τ_P-e, and the maximum relaxation rate, R_P-e(0). As the ³¹P–¹H contribution to the relaxation has been subtracted (Equation (1)), the ΔR_D0 NMRD can be fitted with only the ³¹P spectral density function (Equation (1)) and a constant, c, equal to R_D1(P-e)(0)+R_D2(P-e)(0). The ratio τ_P-e/ΔR_P-e(0) and a correction for how much of the ligand is bound to the protein (estimated from the K_d) are used to obtain the distance of a given spin label to the phosphorus atom of each phospholipid (Equation (2)).

$$\Delta {\mathrm{R}}_{\mathrm{P}-\mathrm{e}}={\mathrm{R}}_{\mathrm{P}-\mathrm{e}}\left(0\right)/(1+{{\mathsf{\omega }}_{\mathrm{P}}}^2{{\mathsf{\tau }}_{\mathrm{P}-\mathrm{e}}}^2)+\mathrm{c}$$

(1)

$${{\mathrm{r}}_{\mathrm{P}-\mathrm{e}}}^6=(\left[\mathrm{protein}\ast \mathrm{ligand}\right]/\left[\left(2/3\right)\mathrm{total} \mathrm{ligand}\right] \left(\frac{{\mathsf{\tau }}_{\mathrm{P}-\mathrm{e}}}{\Delta {\mathrm{R}}_{\mathrm{P}-\mathrm{e}}\left(0\right)}\right){\left(\frac{\mathsf{\mu }}{4\mathsf{\pi }}\right)}^2{\left(\frac{\mathrm{h}}{2\mathsf{\pi }}\right)}^2 {{\mathsf{\gamma }}_{\mathrm{P}}}^2{\mathsf{\gamma }}_{\mathrm{e}}$$

(2)

Note that the 2/3 in Equation (2) accounts for the fact that the protein binds to the outer leaflet of the vesicle, which for our SUVs contains approximately two-thirds of the total phospholipids. A series of r_P-e for BtPI-PLC spin-labeled at different sites provides constraints for the ³¹P–electron interaction that, together with computer modeling, can localize specific phospholipid binding sites [22]. More details on the method are found in the Section 6.

Figure 4A provides a ³¹P spectrum for the dioleoylphosphatidylmethanol (DOPMe) and POPC in the same SUVs as well as a polar headgroup structure. Figure 4B illustrates the effect of the three spin labels picked to cover the different regions on the protein. Anionic DOPMe, the substrate surrogate, is a good inhibitor that is not hydrolyzed by the enzyme over the 24 h of the field cycling experiment. The DOPMe and POPC ³¹P resonances exhibit different R_D0 profiles, indicating different binding sites and proximities to the spin label (Figure 4B). Subtracting the control (vesicles mixed with protein lacking a spin label) and analyzing the R_D0 region with Equation (1) provides τ_P-e and ΔR_P-e(0). The ratio τ_P-e/ΔR_P-e(0) is related to r_P-e⁶ (Equation (2)). Figure 4C shows the r_P-e for POPC and DOPMe that has been extracted for the different spin-label positions in BtPI-PLC. Each Cys mutation is annotated as to the structural feature it is in or near.

For DOPMe, the strongest paramagnetic relaxation effect is with the spin label at H82C (near or in the active site). It is also relaxed by a spin label in helix B (W47C). The strongest relaxation effect for the PC is when the spin label is at the top of helix F (attached to D205C). This region of the protein is relatively far from the active site (Figure 4D). The nearby N-terminal end of helix G has an unusual composition with a string of four tyrosine residues, Tyr246, Tyr247, Tyr248, and Tyr251, whose mutation to Ser or Ala causes a large loss in binding affinity [13]. One or more of these could form a cation-π complex with the PC trimethylammonium moiety. Such a PC binding site in this region would be 15–17 Å from the active site.

If a single PC stayed in a cation-π site for the entire residence time of the protein on the vesicle (380 ms), it would be in a slow exchange with the bulk POPC in the SUV, and it would not be detected. As the relaxation of the ³¹P by the spin label is observed, there is a fast exchange between the enzyme-bound and the bulk phospholipid environments for both POPC and DOPMe. Again, this means that while the protein is anchored on the vesicle, the PC molecules are moving back and forth from the bulk bilayer to the enzyme binding site.

Further experimental evidence that the site identified for the PC was indeed a PC cation–Tyr complex was provided by engineering a site to mimic that of BtPI-PLC in the PI-PLC from Staphylococcus aureus. The enzyme from S. aureus (SaPI-PLC) has a similar structure to the BtPI-PLC (Figure 5A) but very poor affinity for PC-rich SUVs [23]. It lacks two of the four Tyr residues in helix G. Removing the remaining two Tyr (Y253S/Y255S) has little effect on the affinity of the enzyme for the PG/PC vesicles (Figure 5B). At X_PC = 0.7, there is only a 2-fold increase in K_d. In contrast, adding the two ‘missing’ Tyr (N254Y/H258Y) decreased the K_d for SaPI-PLC N254Y/H258Y dramatically, around 30-fold at X_PC = 0.8, and increased the PC ³¹P R₁ at low fields for POPC but not DOPMe (Figure 5C).

The final experimental evidence that vesicle binding was mediated by a PC cation–Tyr-π complex was the introduction of 3,5-difluorotyrosine (Y-F₂) into specific Tyr sites in SaPI-PLC [24]. If H258Y is forming a cation-π complex, then the K_d for N254Y/H258Y-F₂ should increase substantially for PC-rich SUVs because the negative charge in the aromatic ring is reduced by the attached fluorine atoms. If instead that tyrosine is inserted into the membrane, then the introduction of the two fluorine atoms in the Tyr ring will make the side chain more hydrophobic, which will decrease K_d. At X_PC = 0.8, the K_d for N254Y/H258Y-F₂ is 10-fold higher than for N254Y/H258Y, which is consistent with a cation-π complex. The experiments with Y-F₂ replacing specific Tyr in BtPI-PLC indicate that one or more PC cation–Tyr-π complexes are critical to the binding of that enzyme on the membranes.

3. Computational Results—Identification of PC Binding Sites on BtPI-PLC

3.1. Transient Opportunistic and Very Specific PC Cation–PI-PLC Tyr-π Interactions

The experimental data clearly supported the formation of a cation-π complex between PC and BtPI-PLC and localized a plausible PC binding site on the protein where one or more of a string of Tyr residues could form these complexes. For a more detailed view of the potential PC cation–Tyr-π sites, multiple 500 ns all-atom molecular dynamics simulations were run of the BtPI-PLC binding to dimyristoylphosphatidylcholine/dimyristoylphosphatidylglycerol (DMPC/DMPG) and DMPC bilayers [13,25].

One of the surprising results from these initial simulations was that many transient DMPC cation–Tyr-π complexes were formed during the simulation. However, two cation-π complexes, with Tyr88 and Tyr246, existed for more than 80% of the simulation time with the same phospholipid. Tyr88 is near helix B, and Tyr246 is in the N-terminal region of helix G, both of which are features known to be important for binding to membranes. Snapshots of two of these complexes are shown in Figure 6A,B. They also persist in mixed DMPC/DMPG bilayers at X_PC = 0.8 and 0.5 (Figure 6C) [25]. The DMPC complex with Tyr246 is consistent with the r_P-e obtained experimentally by high-resolution field cycling [22]. The Tyr88 complex was unexpected, but its contribution to the POPC R₁ does fit the shorter than expected r_P-e obtained for the POPC when the protein was spin-labeled on helix B or the active site residues (both regions fairly far from Tyr246 if that were the only cation-π complex). The multiplicity of the PC cation–Tyr-π complexes in the BtPI-PLC simulations also suggests why the decrease in K_d, for the engineered SaPI-PLC N254Y/H258Y binding to X_PC = 0.8 SUVs was to only 0.7 mM rather than the 2 µM for BtPI-PLC. There is no analogue of Tyr88 in the S. aureus enzyme.

Two other tyrosine residues, Tyr200 and Tyr251, also form transient cation-π complexes that are reasonably occupied. Tyr200 is in the BtPI-PLC active site where it has the role of stabilizing the bound PI inositol ring [11,12]. The DMPC forming a choline cation/Tyr200-π complex is halfway out of the bilayer and into the active site. BtPI-PLC enzymatic activity decreases at high X_PC [7], and the activator PC binding in the active site could be partially responsible for the activity decrease. It is worth noting that DMPG was not observed in the active site in any of the simulations. Kinetics studies with different phospholipid headgroups indicate that PG has a weaker affinity for BtPI-PLC than other anionic phospholipids with smaller headgroups [7,17]. The glycerol moiety is fairly flexible, unlike the inositol ring, which presents a face of axial protons able to interact with Tyr200. Anionic phospholipid inhibitors with small headgroups, such as phosphatidylmethanol or phosphatidic acid, are unlikely to interact directly with Tyr200, but they could interact with cationic Arg69 or protonated His32 or His82 in the active site to effectively inhibit the enzyme.

3.2. A Computational Approach to Estimating ΔG⁰ _bind Provides a Membrane Desorption Pathway for BtPI-PLC

Recent work to calculate the absolute membrane binding free energy for BtPI-PLC used a geometrical route and an atomistic force field to progressively detach the protein from the bilayer [26]. Along with a value for ΔG⁰ _bind (which agrees moderately well with the value from fluorescence correlation spectroscopy binding data for BtPI-PLC binding to PC SUVs), the method provides atomic-level details that describe the membrane-bound protein and how interactions change during the desorption process. The two persistent cation-π complexes, involving Tyr88 and Tyr246, show distinct interactions with other nearby amino acids in the membrane-bound form, and these change as the protein is extracted (Figure 7). In the membrane-bound state, the stability of the PC-cation–BtPI-PLC Tyr88-π complex is aided by a hydrogen bond between the Tyr–OH and the PC phosphate and three nearby residues interacting with the PC molecule: (1) Lys44 (in helix B) forms a salt bridge with the PC phosphate group, and (2) Gln40 and (3) Asn41 hydrogen bond with the DMPC (Figure 7A).

There is limited experimental evidence consistent with this network of interactions stabilizing the Tyr88 cation-π complex. However, the variants Y88A and Y88W lend support for the formation of a hydrogen bond between the hydroxyl group of Tyr88 and the DMPC phosphate (Figure 8). The Y88A protein has lost one of the cation-π interactions and its K_d increases significantly with an increase in the ΔΔG⁰ _bind of +2.5 kcal/mol for binding pure DMPC bilayers (Figure 8). Y88W does recover some of the binding energy lost by Y88A, but the protein still does not bind as tightly (or more tightly, as expected for a choline cation–Trp-π complex) as unmodified BtPI-PLC. The ΔΔG⁰ _bind for Y88W compared to wild-type PI-PLC is +1 kcal/mol, equivalent to the loss of a hydrogen bond. It is likely that a cation-π interaction is intact in Y88W, but the hydrogen bond between the aromatic indole ring and the DMPC phosphate is not present. The larger indole ring in this position and perhaps the misalignment of the N-H on the indole ring could preclude formation of the H-bond with the PC phosphate group.

The network for the other persistent cation-π adduct is much less complex. The membrane-bound DMPC cation–Tyr246-π complex with a PC molecule is stabilized by a hydrogen bond between the Ser244 and the DMPC phosphate (Figure 7B). However, when the protein is pulled more than 5 Å away from the bilayer center, the occupancy of that hydrogen bond of Ser with the DMPC starts to decline (Figure 7D). At r = 27 Å, a new interaction appears. The tyrosine -OH group forms a hydrogen bond with the DMPC phosphate. This bidentate Tyr246 complex with DMPC stays intact until the protein is close to completely detaching.

4. What Is the Allosteric Mechanism for PC Altering BtPI-PLC Enzymatic Activity?

It is clear that PC in a bilayer dramatically increases the affinity of BtPI-PLC for surfaces by forming PC cation–tyrosine-π complexes that anchor the enzyme on the bilayer. These complexes would be the first to form when the protein makes initial contact with the bilayer and the last interactions to break as the protein is released from the bilayer. Lowering the apparent K_d makes an important contribution to improving enzyme specific activity, particularly for low concentrations of vesicles. As the assay conditions for Figure 1B had a near saturated enzyme, the two-fold increase in activity could reflect an increase in k_cat. Rather than deal with complex models to sort out the interfacial kinetics, an easier approach is to use cIP, which has no affinity for phospholipid interfaces, as the substrate to test whether the PC/BtPI-PLC complex does in fact alter k_cat. For the cleavage of cIP in the absence of interfacial PC, the V_max and K_m for cIP are 20 µmol min^-1 m^-1 and 90 mM, respectively [6]. The presence of 8 mM diC₇PC increases k_cat ~7-fold and decreases the K_m 3-fold. This corresponds to an increase in enzyme efficiency from 130 to 2600 s⁻¹ M⁻¹. Simulations found little change in the BtPI-PLC structure in solution versus when bound to a bilayer [13]. A possible explanation for how PC, bound to the protein via two cation-π complexes, increases enzyme efficiency is that it alters the protein dynamics.

The crystal structure of the B. cereus PI-PLC had weak intensity in the rim of the active site. Helix B, as well as the loop residues 237–243, was particularly weakly defined [11,12]. Principal component analysis of the MD simulations (20–50 ns) of BtPI-PLC in aqueous solution [27] identified a clamshell-like motion with β-strands 1–5, along with associated loops and helices moving as one unit and strands 5b-8 moving as a second unit. This links the motions of helix B (residues 39–46) on one side of the clamshell with the loops of the N-terminal to helix F (residues 201–203) and helix G (residues 238–245). Both of these regions interact with the membranes, as assessed by mutagenesis and the MD simulations. The clamshell opens and closes over the active site and is observed in all simulations of the WT enzyme in solution. In the TIM barrel superfamily, a lid often controls access to the active site, and the loop between strand 7 and helix G is frequently associated with phosphate binding [28]. For BtPI-PLC in the absence of PC, this type of motion of loops and helices could certainly limit the active site access. This could be a strong barrier for a water-soluble substrate such as cIP binding in the active site.

Based on these simulations, two proline residues in helix G are important for the clamshell motion: Pro245, termed Pro(cap) because it is at the N-terminal end of helix G, and Pro254, termed Pro(kink) because it is where the G-helix bends. The opening and closing of the loop linking helix F and the helix G loop with helix B is above the active site [27]. The mutation of Pro(cap) to Gly or Tyr reduced specific activities towards both PI and PI/PC SUVs (Figure 9A) and increased K_d, the latter modestly (Figure 9B). Kinetic data for the soluble substrate cIP exhibited similar behavior. Pro(cap) is ~9 Å from the active site and adjacent to Tyr246; so, the substitution of that Pro could alter both the activity and the binding by altering the cation-π interaction. In contrast, the Pro(kink) mutants were not significantly affected in the absence or presence of PC. Simulations of both Pro mutants in solution showed that the clamshell motion was disrupted. The one anomaly in the simulations was that without Pro(kink), helix G was in an extended unkinked conformation that would allow access to the active site. This would yield specific activities and K_d values similar to those of the WT enzyme.

Unlike many other members of the TIM barrel superfamily [28], BtPI-PLC does not have a lid controlling access to its active site. Instead, the anticorrelated movement of the loops above the active site likely acts as a dynamic lid. PC molecules function as allosteric activators by anchoring the protein to the bilayer surface via formation of PC cation–BtPI-PLC Tyr-π complexes and stabilizing an ‘open’ form of the protein where the PI can efficiently diffuse into the active site. The principle is the same for cIP hydrolysis. PC, whether in vesicles or micelles, stabilizes the enzyme in a state where the active site is accessible to the water-soluble substrate. Without the PC interface, the probability of a cIP molecule colliding with the protein when the clamshell is open and the active site is accessible is small. Locking BtPI-PLC into an open conformation by binding PC leads to a large increase in k_cat for cIP. Consistent with this allosteric mechanism, soluble inhibitors based on the PI headgroup structure are poor inhibitors of the BtPI-PLC catalyzed hydrolysis of cIP, but the presence of a PC surface (to which the protein binds) makes them much more potent [29]. In that assay system, neither the substrate nor the inhibitor partitions into the interface. The binding of BtPI-PLC to PC and the formation of the two persistent cation-π complexes changes the dynamics of the protein to stabilize a conformation with an open active site.

The Dennis group has pioneered the concept that the membrane binding of several phospholipase A₂ enzymes allosterically leads to an open conformation of the active site [30,31]. In fact, their recent computational and experimental study of lipoprotein-associated phospholipase A₂ provides evidence that the membrane binding of the protein promotes a conformational change that opens access to the active site [32]. In contrast, in an aqueous environment, the positioning of a phospholipid in the active site is much less favorable.

BtPI-PLC is also allosterically affected by phospholipids. The initial binding of the protein does not require a large interfacial surface. Instead, two PC molecules form specific PC cation–Tyr-π complexes with the protein that then allow helix B and the β7-αG loop to insert into the bilayer. Rather than generating a significant conformational change in the protein, these complexes stabilize an open active site allowing processive catalysis of PI. The hydrolysis of soluble cIP is also enhanced because the active site is now open when the BtPI-PLC is bound to a PC-containing surface.

5. Conclusions

These studies of B. thuringiensis PI-PLC have provided a number of insights into how PC cation–protein Tyr-π complexes can contribute to anchoring a peripheral membrane protein on PC-containing interfaces, which in turn can aid in the insertion of hydrophobic residues. Enzymatic activity is enhanced by these complexes stabilizing open access to the active site. The two cation-π sites prevent a clamshell motion that obstructs the active site when the protein is in solution. So far, there are only a few other peripheral membrane proteins where PC-cation–aromatic Tyr/Trp/Phe complexes have been identified and shown to be important for membrane binding. These include a number of phospholipases in addition to BtPI-PLC: the cytosolic phospholipase A₂, whose C₂ domain forms a PC-cation–Tyr-π complex that is also stabilized by a Ca⁺² interacting with the lipid phosphate group [33]; phospholipase A₂ Naja naja atra, where ΔΔG data are available for replacing aromatic amino acids by Ala [34], and the PC specificity was shown to arise from cation-π complexes by simulations [26]; and a spider phospholipase D that uses Tyr cages around the bound PC cation [35]. The PC cation-π formation in the PLD appears conserved in many of the other members in the same clade, emphasizing these are distinct complexes that are useful for ensuring PC specificity. Simulations with PLD also showed that with a mixed PE/PC bilayer, no cation-π complexes were formed with PE. Neutrophil proteinase 3 [36], lung surfactant protein (SPA) [37], and equinatoxin, a soluble pore-forming toxin, [38] round out the group of interfacial enzymes with identified PC cation–aromatic amino acid π complexes. There is diversity in how many aromatic residues are involved in a PC cation–Tyr/Trp/Phe complex and in how many other residues contribute to stabilizing the complex.

Two of these proteins, BtPI-PLC and proteinase 3, both with verified PC cation–aromatic amino acid-π complexes, and the snake venom enzyme, Naja naja phospholipase A₂, which has ΔΔG values for aromatic to Ala replacements binding to PC vesicles, were further examined with free energy perturbation simulations to see the variation in complex location in the bilayer, number, and choice of aromatic residues and the energy lost in the alanine mutants [39].

Surprisingly, the interfacial aromatics mediating the cation-π interactions with choline-containing lipids can contribute as much to peripheral protein affinity for membranes as aromatics inserted below the phosphates. The π complexes show significantly higher free energy than the same amino acid in roughly the same location that is just partitioned in the membrane (compare F166 and F224, which have high occupancies for cation-π complexes, with F165 or Y88 and Y246 compared to Y247 in Figure 10). There is a cluster of Tyr or Phe cation-π complexes in the region of the phosphate and choline that have fairly similar ΔΔG values, but an energetically similar Trp cation-π complex (W61) can be significantly closer to the surface and still have a large ΔΔG values. Figure 10 emphasizes that aromatic amino acids are versatile in how and at what depth they can bind a PC headgroup and stabilize a peripheral membrane protein. Not only can the number and identity of aromatic amino acids vary, but the location of the complex can vary tremendously. The first region a protein will encounter on the way to the insertion of hydrophobic amino acids in a bilayer is that occupied by phospholipid headgroups. These studies strongly indicate that cation-π interactions of the protein with PC or other choline-containing lipids such as sphingomyelin are uniquely poised to help stabilize protein insertion at the upper region of the membrane interface.

6. Appendix: Details on Analysis of ³¹P NMRDs and Examples

The field dependence of the dipolar relaxation of ³¹P by protons, R_D, is described by the following equation:

$${\mathrm{R}}_{\mathrm{D}}=\left(\frac{{\mathrm{R}}_{\mathrm{D}}\left(0\right)}{2{\mathsf{\tau }}_{\mathrm{D}}}\right)(0.1\mathrm{J}\left({\mathsf{\omega }}_{\mathrm{H}}-{\mathsf{\omega }}_{\mathrm{P}}\right)+0.3\left({\mathsf{\omega }}_{\mathrm{P}}\right)+0.6\mathrm{J}\left({\mathsf{\omega }}_{\mathrm{H}}+{\mathsf{\omega }}_{\mathrm{P}}\right)$$

(3)

where $\mathrm{J}\left(\mathsf{\omega }\right)=\frac{2\mathsf{\tau }}{1+{\mathsf{\omega }}_2{\mathsf{\tau }}_2}$ $\mathrm{J}$(ω) is the spectral density. The ω_P and ω_H are the gyromagnetic ratios for ³¹P and ¹H, and τ_D and R_D(0) are the correlation time and maximum relaxation rate for the dipolar interaction responsible for generating the NMRD. Each dipolar NMRD identified in a field cycling profile will be characterized by a τ_D and R_D(0). Chemical shift anisotropy relaxation also contributes to the ³¹P relaxation, but only at high fields. The relaxation rate is described by Equation (4).

$${\mathrm{R}}_{\mathrm{CSA}}={\mathrm{k}}_{\mathrm{CSA}}\frac{{{\mathsf{\omega }}_{\mathrm{P}}}^2{{\mathsf{\tau }}_{\mathrm{CSA}}}^2}{1+{{\mathsf{\omega }}_{\mathrm{P}}}^2{{\mathsf{\tau }}_{\mathrm{CSA}}}^2}\cong {\mathrm{k}}_{\mathrm{CSA}}{{\mathsf{\omega }}_{\mathrm{P}}}^2{{\mathsf{\tau }}_{\mathrm{CSA}}}^2$$

(4)

If ω_P²τ_CSA² << 1, then the expression can be simplified to a square law dependence of R₁ on B_relax. The k_CSA is related to the CSA interaction size and asymmetry. The ³¹P R₁ data for dipalmitoylphosphatidylcholine in large unilamellar vesicles at 45 °C showed no deviation from a square law up to 21.3 T [40]; so, the approximation should be valid for the POPC/DOPMe SUVs used for the BtPI-PLC studies.

The total relaxation rate at each relaxation field, B_relax, or Larmor angular frequency, ω_P, is the sum of all the NMRD contributions. These small vesicles exhibit three dipolar and one CSA NMRD. The full R₁ profile from 0.003 to 11.7 T is the sum of these four NMRDs (Equation (5)). For a more detailed look at how different NMRDs are related to phospholipid motions see [21].

R₁ = R_D0 + R_D1 + R_D2 + R_CSA

(5)

Fitting the data is rarely performed for the entire field range. The region between 3 and 11.7 T is fitted as R_D2(0) + R_CSA; that sum is then is then subtracted at each field strength. What is left reflects R_D0 + R_D1. Since R_D0 NMRD is where the spin label will have the largest effect, we usually fit the R₁ data below 0.08 T as R_D0 plus a constant (which would be R_D1(0) + R_D2(0)). After subtracting the data for the sample with the unlabeled protein from the data with the spin-labeled BtPI-PLC in the low field region, the resultant ΔR_1(P-e) is fitted with a single spectral density term (Equation (6)) in order to extract the R_Do values for the ³¹P–electron dipolar interaction, τ_P-e and ΔR_P-e(0), and estimate r_P-e (Equation (7)).

$$\Delta {\mathrm{R}}_{\mathrm{P}-\mathrm{e}}=\Delta {\mathrm{R}}_{\mathrm{P}-\mathrm{e}}\left(0\right)/(1+{{\mathsf{\omega }}_{\mathrm{P}}}^2{{\mathsf{\tau }}_{\mathrm{P}-\mathrm{e}}}^2)+\mathrm{c}$$

(6)

$${{{{\mathrm{r}}_{\mathrm{P}}}_{-}}_{\mathrm{e}}}^6=([\mathrm{PLC}\bullet \mathrm{PL}]/[\mathrm{PL}{]}_{\mathrm{out}})\times \left({\mathrm{S}}^2{\mathsf{\tau }}_{\mathrm{P}-\mathrm{e}}/\Delta {\mathrm{R}}_{\mathrm{P}-\mathrm{e}}\left(0\right)\right){\left(\mathsf{\mu }/4\mathsf{\pi }\right)}^2 {\left(\mathrm{h}/2\mathsf{\pi }\right)}^2 {{\mathsf{\gamma }}_{\mathrm{P}}}^2{{\mathsf{\gamma }}_{\mathrm{e}}}^2$$

(7)

Equation (6) assumes that the ΔR_P-e represents the paramagnetic relaxation of a single bound POPC or DOPMe. Proximity of the spin-label to multiple PC binding sites (i.e., complexes with Tyr246 and Tyr88) can be done, but is more complex and the assumption of a single site is a good way to start. In Equation (7), μ₀ is the magnetic permeability in a classical vacuum, h is Planck’s constant, and γ_e is the gyromagnetic ratio for an electron. The order parameter S² is assumed to be 1 since faster motions that will only slightly change r_P-e on the µs timescale. Since BtPI-PLC binds to the external leaflet of the vesicle, the total concentration of each type of phospholipid is multiplied by 0.67 to 0.75 to estimate [PL]_out, the concentration of PC or PMe in the outer leaflet. For POPC/DOPMe (1:1) SUVs, the concentration of the phospholipids is sufficiently high compared to the apparent K_d so that all the enzyme will be partitioned on the SUVs. Therefore, [PLC∙PL] = [PLC]_o where [PLC]_o is the total concentration of enzyme added. While there are errors in τ_P-e and ΔR_P-e(0), the ratio of the two is unlikely to be off by more than a factor of two. More importantly, the dependence of r_P-e⁶ on τ_P-e/ΔR_P-e(0) means that the r_P-e extracted is fairly well defined.

For many of the spin-labeled BtPI-PLC cysteine variants, the low field dependence of R₁ on B_relax was examined with a different protein concentration as a check on the extrapolated ΔR_P-e(0). Figure 11 shows the R_D0 NMRD for DOPMe with spin-labeled H82C at 0.014 and 0.029 mM. As a quick check, the difference in R₁ between where the curves intersect the y-axis (it hasn’t reached R_D0 yet) and the R₁ at 0.1 T (where R₁ is the sum of R_D1(0) and R_D2(0)) should differ about 2-fold). The τ_D0 is similar for both samples and the ratio of R_D0(0) for the two samples (1.7) is fairly close be proportional to the ratio of the protein concentration used (2.1).

³¹P field cycling with spin-labeled BtPI-PLC has also been used to assess whether small molecules bind to the protein [41]. An example of this is shown in Figure 12. The protein was spin-labeled on D205C, near the proposed cation-π binding site, and on H82C, near the active site. The small molecule in solution has a field dependence of R₁ that exhibits R_CSA at high B_relax and then a constant R₁ that corresponds to the maximum R_D(0) for fast motions. If it binds to a macromolecule, there should be a new NMRD with a correlation time closer to the rotational correlation time of the protein. If the R₁ profile for the small molecule mixed with unlabeled protein in solution is subtracted, what is left represents the NMRDs for the effect of the spin-label unpaired electron attached to a protein Cys on the small molecule ³¹P. If there is no new NMRD, then the small molecule either does not bind or is too far away from the spin label. If the paramagnetic R₁ enhancement increases as the concentration of diC₄PC is increased, the site was initially not saturated. Figure 12 shows the effect of the two spin labels on the dC₄PC ³¹P R₁. There is virtually no effect of spin-labeled H82C on BtPI-PLC. As expected this small water-soluble PC has no affinity for the BtPI-PLC active site, nor does it bind to the Y88 cation-π site. However, there is a very small increase in R₁ with spin-labeled D205C. The NMRD responsible for this increase has a 5–10 ns correlation time and an ΔR_P-e(0) of 0.044 s⁻¹. No lower field increase in R₁ was observed, meaning any complexes that form do not persist more than ~10 ns. This indicates diC₄PC can bind to the protein, presumably in the PC cation-Tyr246 π site, but the binding and occupancy are quite low.